Report No. 24524-CO Colombia Poverty Report (In Two Volumes) Volume ll: Background Report November 1, 2002 Colombia Country Management Unit PREM Sector Management Unit Latin America and the Caribbean Region Document of the World Bank CURRENCY EQUIVALENTS (as of October 21, 2ID2) Currency Unit = IPeso (Coll$) Col$1 = US$ 0.000351 US$1 = Col$ 2876.400 WEIGHTS AND MEASURES Metric System IFISCAL YEAR January 1 to December 3 i ABBREVIATIONS AND ACRONYMS ARS Administradoras de Regimen Subsidiado CAIP Centros de Atenci6n Integral al Preescolar CASEN Caracterizaci6n Socioecon6mica Nacional CDF Comprehensive Development Framework CODHES Consultoria para los Derechos Humanos y el Desplazamiento DALYS Disability-Adjusted Life Years DHS Demographic and Health Survey DNP Departamento Nacional de Planeaci6n ECV Encuesta de Calidad de Vida ENH Encuesta Nacional de Hogares FGT Foster-Greer-Thorbecke index GDP Gross Domestic Product GNP Gross National Product HCB Hogares Comunitarios de Bienestar ICBF Instituto Colombiano de Bienestar Familiar ICETEX Instituto Colombiano de Credito Educativo y Estudios Tecnicos en el Exterior ICFES Instituto Colombiano para el Fomento de la Educaci6n Superior IDP Internally Displaced Populations IDB Inter-American Development Bank INURBE Instituto Nacional de Vivienda de Interes Social y Reforma Urbana ISS Instituto de Seguros Sociales LSMS Living Standards Measurements Study NHS National Household Surveys PACES Programa de Ampliaci6n de la Cobertura de Educaci6n Secundaria POS Plan Obligatorio de Salud POSS Plan Obligatorio de Salud Subsidiado PPP Purchasing Power Parity PSE Public Social Expenditure SCD SISBEN classification document SENA Servicio Nacional de Aprendizaje SISBEN Sistema de Selecci6n de Beneficiarios UNDCP United Nations Development Program for Drug Control and Crime Prevention UNDP United Nations Development Program UPAC Unidad de Poder Adquisitivo Constante VAT Value Added Tax UVR Unidad de Valor Real Vice President: David de Ferranti Director: Olivier Lafourcade Lead Economist: Marcelo Giugale Sector Manager: Norman Hicks Task Manager: Carlos Eduardo V6lez TABLE OF CONTENTS Part 1. How are the poor doing and why? Chapter 1. Two Decades of Economic and Social Development in Urban Colombia: A Mixed Outcome 1. Introduction ........................................................................2 * Adverse macro environment in the late nineties: recession and risk deterioration * Supply shocks and multiple structural changes * Objective and mainfindings * Opposing trends in social development and crime intensity indicators 2. Social Progress in Urban Colombia . .......................................................................6 2.1 Schooling and education: increasing trend with pro-cyclical fluctuations 2.2 Child labor: pro-cyclical with a decreasing trend 2.3 Child malnutrition and infant mortality: continuous improvement 2.4 Access to basic infrastructure: progressive gains with regional disparities 2.5 Life expectancy: advancing with gender bias 2.6 Violence: continuous deterioration since the 1970s 3. Economic Welfare, Income Distribution and Poverty: Are Urban Colombians Better-Off Than Two Decades Ago? .............................................. 13 3.1 Poverty Measures: substantial long-term progress with a recent set back 3.2 Average household income per capita 3.3 Welfare as average income corrected by inequality: The Gini coefficient and the Sen Welfare Index * The evolution of income inequality * Adjustingfor inequality along the lifecycle: the Paglin-Gini index * Average income corrected by inequality: the Sen Welfare Index 3.4 Welfare comparisons for any distributional weights: First Order and Generalized Lorenz Dominance 4. Poverty Profile: Faces of the Poor - Vulnerable Groups and Household Characteristics ....... ....... 28 4.1 Basic factors of income per capita generation: the poor versus the non- poor * From basic income generatingfactors to the faces of the poor 4.2 Faces of the poor: groups at risk and household characteristics 4.3 The poverty profile and the marginal effect of key household characteristics * Skill endowment * Demographics * Labor Market * Homeownership * Idiosyncratic shocks 4.4 Unemployment by skill, age and region STRUCTURE (F VOLUME R1R Part 1. 1H1ow are the poor doing and why? Chapter 1. Two Decades of Econonmc and Social Development in Urban CoRombia: A Mixed Outcome By Carlos Eduardo Velez, Mauricio Santamarfa, Natalia Millan and Benedicte de la Briere Chapter 2. Poverty and Welfare in Rura Colombia Duruing the Last Two Decades By Carlos Eduardo Velez, Benedicte de la Briere and Natalia Millan Chapter 3. The Reversal of Ilinequality Trends in CoRombin, a970-4995: A Combination of Persistent and IFluctuating Forces By Carlos Eduardo V6lez, Jose Leibovich, Adriana Kuigler, Cesar Bouill6n and Jairo Nuifiez Chapter 4. Who Bears the Burden of Crnime and Violence in Colombia? By Alejandro Gaviria Part Ifl. What is government doing for the poor? Chapter S. Public Social Expenditure in Colombia: lincidence and Sector Priorities in the 1990s By Carlos Eduardo Velez and Vivien Foster Chapteir 6 Subsidized Health JInsurance, SI[SBEN and Demand for HeaRth Ca¢re Among the Poor in Colombia By Giota Panopoulou 5. Labor Markets and Poverty Dynamics in Booms and Recession: .......................................... 42 5.1 The decomposition of poverty changes in terms of growth and inequality 5.2 The evolution of income per capita in terms of employment, wages, skill endowments, and dependency ratios (1978-99) * The dynamics of poverty in terms of basic income generating factors * Four types of labor market attachment: differences in poverty levels * Differences in income and poverty dynamics across types of labor market attachment and head skills 1978-99 5.3 The dynamics of poverty in the recession: lower wages or jobs lost? * Asymmetric dynamics of quantities and prices during the recession: job loss for wage earners versus salary reduction for the self-employed * The effects of labor market dynamics on households by heads' skill level and labor market occupational choices 6. Sources of Income Inequality: An Introduction ..................................... 51 6.1 Decomposition of inequality by household characteristics 6.2 Inequality and composition of income: labor income and other sources * Shorrocks decomposition 6.3 Labor Income Inequality: checking for hours of working week and lifecycle bias 7. Summary and Conclusions ...................................... 55 Chapter 2. Poverty and Welfare in Rural Colombia During the Last Two Decades 1. Introduction ........................... 62 * Macro-economic and sectoral environment * Concentration of Land and Credit * Displacement population and armed conflict in the rural areas * Illicit Crops and Rural Development * Main findings * Strong improvement in access to social services * Rural welfare improves until the mind nineties * Poverty determinants andfaces of the poor * Extreme poverty dynamic: growth is crucial * Regional comparisons: Central and Atlantic regions win, Pacific and Central loose 2. Social Progress in Rural Colombia ........................... 70 2.1 Schooling and education: persistent progress with regional convergence 2.2 Child Labor: a declining trend with some regional disparities over the business cycle 2.3 Basic infrastructure services (1995 and 1999): Access in rural cabeceras is comparable to urban centers 3. Economic Welfare, Poverty and Income Distribution: How Has Welfare in Rural Coombina Evolved During the Last Two Decades? ........................................................................ 74 3.1 Poverty Measures: clear progress in the 1980s plus smaller improvements, if not, deterioration, in the late 1990s. 3.2 Average household income per capita: strong improvement in the 80s, slowdown in the early 90s and some recovery after 1995. Regional discrepancies 3.3 Rural welfare as average income corrected by inequality: The Gini coefficient and the Sen welfare index o The Sen Welfare Index: rising inequality has diminished potential welfare gains 3.4 Welfare comparisons independent of distributional weights: First Order and Generalized Lorenz Dominance 4. Faces of the Poor: Vulnerable Groups and Household Characteristics ........... ......................... go 4.1 Basic factors of income per capita generation: the poor versus the non-poor 4.2 Faces of the poor: Typical characteristics of households with higher poverty risk o Vulnerable Groups: Children and recent migrants 4.3 The poverty profile and the marginal effect of key household characteristics o Idiosyncratic shocks 4.4 Unemployment by skill, age and region Unemployment peaks for the young (18 to 25) and those with intermediate skills 5. Sources of Extreme Poverty: Decomposition in Terms of Growth, Prices and ffmequality ....... ..... 102 6. Summary and Conclusions ..................................................................... 103 Chapter 3. The Reversal of ]Inequaty Trends in Colombia, 1978-19950 A Combination of Persistent and IlFuctuating IForces 1. ffIntroduction ..................................................................... 110 2. The Colombian Income Distribution: 1978, 1988 and 1995 .111 2.1 The Loss of the Inequality Gains of the 60s and 70s 2.2 Main forces driving the dynamics of income distribution 2.2.1 The Evolution of the socio-demographic structure of the working population o Higher and more egalitarian school attainment o Higher laborforce participation -particularly among women o Decreasing fertility rates 2.2.2 Macro events and changes in demand for labor o Satisfactory growth performance, but declining elasticity of employment to GDP o Increasing difficulties for low skill job creation in the 1990s o Rising costs of wage-earning job creation relative to self-employment o Negative shocks and adverse policies in early 1990s reduce agricultural output and encourage more labor-intensive crops o Rural credit and land concentration: two distinct phases 3. The Determinants of Household Income: 1978, 1988 and 1995 .116 3.1 Changes in the earnings equations o Increasing heterogeneity in returns to education and more convexity of the earnings function 3.2 Changes in participation and occupational choice behavior 4. Understanding Income Distribution Dynamics in Colombia: Factor Decomposing by Simulations, 1978-1988 and 1988-1995 ............................................. 123 4.1. Urban Areas: The reversal of the inequality trend by the combination of persistent and fluctuating forces 4.1.1 Individual Labor Earnings * Structural parameters of the earnings equation 4.1.2 Household income inequality and structural parameters: Discrepancies relative to individual income distribution * Participation and occupational choice 4.1.3 Endowment effects and non-labor income * Paradoxically, education endowment equalization deteriorates income inequality in urban areas, but not in rural areas * Family size is always equalizing * Non-labor income effect: always unequalizing and moderate 4.2 Rural Areas: The Reversal of Inequality Losses from 1988 to 1995 * 1988-1995: The Persistent Equalizing Effect of All Observable Factors Generates Lower Income Inequality 5. Conclusions and Discussion .......................................... 133 Chapter 4. Who Bears the Burden of Crime and Violence in Colombia? 1. Introduction .......................................... 142 2. Crime and Violence in Colombia: A Quick Overview .......................................... 142 3. Stylized Facts: Danger, Fear and Retreat .......................................... 146 4. Methodology and Estimation Results .......................................... 149 * Distribution of crime across victims o The fear of crime among the better-off o Domestic violence against women: the other side of the coin 5. Caveats .......................................... 155 6. Conclusions .......................................... 156 Part II What is government doing for the poor? Chapter 5. Public Social Expenditure in Colombia: Incidence and Sector Priorities in the 1990s 1. Introduction ........................................................................ 160 2. Public Social Expenditure in the 1990's: An Overview ........................................................ 160 3. The Nature of Social Programs ........................................................................ 163 3.1 Institutional Structures 3.2 Financing Mechanisms 3.3 Targeting Mechanisms 3.4 Value of Subsidies 3.5 Scope of social programs 4. Pro-Poor Coverage Dynamics in the 1990's .173 4.1 Education 4.2 Healthcare 4.3 Utilities 5. The Distributional Impact of Public Social Expenditure ....................................................... 180 6. Expansion Priorities in Social Programs ........................................................................ 184 7. Impact of Decentralization ......................................................................... 187 * Healthcare and education and the certification process 8. Summary and Conclusions ....................... ................................................. 191 Chapter 6 Subsidized Health Insurance, SISBEN and Demand for Health Care Among the Poor in Colombia 1. Introduction ........................................................................ 198 2. The Health Sector Reform ........................ ................................................ 198 2.1 Main Characteristics: The contributory and the subsidized regime 2.2 SISBEN classification in level 1 or 2 seemingly equates affiliation to the subsidized regime 3. Affiliation and Health Care Utilization Patterns ............................................................... 202 3.1 Affiliation Patterns for Households at SISBEN LEVELS 1,2 and 3 3.2 Health Care Utilization Patterns for Household Heads at SISBEN levels I and 2 4. Determinants of a SISBEN Classification Document Take-up .............................. 208 4.1 Theoretical Considerations 4.2 Methodology 4.3 Empirical Results 4.3.1 Estimates for the SISBEN level 1 and 2 populations 4.3.2 Estimates for the SISBEN level 3 population 5. The Effect of Holding a SISBEN Classification Document on the Utilization of Health Care Services and Out-of-Pocket Expenditure ................................................................ 216 5.1 Theoretical Considerations 5.2 Methodology 5.3 Empirical results 5.3.1 Health care utilization and the possession of SCD 5.3.2 Health care expenditure and SCD holding 5.3.3 Other explanatory variables 6. Conclusions ................................................................... 222 Appendixes to Chapter I l.A Data, Methodological Considerations l .B Statistical Appendix l.C Implications of Imputed Rent Adjustment for Poverty and Inequality in Colombia 1. Introduction. 2. Alternative models 3. Estimation Results 4. Poverty and Inequality Measures Under Alternative Imputed Rent Models 5. Summary and Conclusions to Chapter II 2.A. Statistical Appendix to Chapter III 3.A. Data, Methodological Considerations 3.B Analytical Framework Decomposing the Dynamics of Income Distribution 3.C Statistical Appendix to Chapter V 5.A Derivation of Unit Value of Subsidies for Public Service 5.B Statistical Appendix Acknowledgements This report was prepared by a team led by Carlos Eduardo Velez, Vivien Foster, Mauricio Santamaria, Natalia Millan and Benedicte de la Briere from the World Bank, Giota Panopoulus (Sussex University), Alejandro Gaviria (Fedesarrollo, Bogota). Research assistance was provided by Taizo Tekano and document preparation assistance by Tania Gomez and Anne Pillay. Preliminary versions of some papers were discussed in seminars at Banco de la Republica, Departamento Nacional de Planeaci6n, Fedesarrollo, and the Ministerio de Salud in Colombia. The Task Manager was Carlos Eduardo VYlez, Sector Manager Norman Hicks, Lead Economist Marcelo Giugale and Country Director Olivier Lafourcade. Chapter 3, 'The Reversal of Inequality Trends in Colombia, 1978-1995: A Combination of Persistent and Fluctuating Forces" by Carlos Eduardo Velez (World Bank), Cesar Bouillon (IDB), Jose Leibovich (Santander Investment, Bogota), Jairo Nuinez and Adriana Kugler (U. Pompeu Fabra), is part of the cross-country study, 7he Microeconomics of Income Distribution Dynamics in East Asia and Latin America, a joint IDB/World Bank research project directed by Frangois Bourguignon, Francisco Ferreira and Nora Lustig. The IDB started this initiative and provided research funding for the paper on urban Colombia. Concurrently, the World Bank (DEC) provided the financial support for the complementary rural background paper. The World Bank-LAC's Colombia Poverty Report Task supported the final stages of the urban paper and the merging of the urban and rural papers into a final product. CHAPTER I TWO DECADES OF ECONOMIC AND SOCIAL DEVELOPMENT IN URBAN COLOMBIA: A MIXED OUTCOME Carlos Eduardo Vdlez, Mauricio Santamarfa, Natalia Milldn and B6nddicte de la Brire World Bank (LAC/PREM) ABSTRACT This paper examines urban Colombia's socio-economic development from 1978 to 1999, including analysis of income inequality and welfare, and the evolution of poverty and its determinants. The evidence shows mixed results. First, social progress appears contradictory. Although most of the social indicators pertaining to education, health, and infrastructure show substantial long-term improvements during the last two decades, the simultaneous escalation of violence -mostly associated with the illegal drug trade originating in the seventies- has become a substantial social and economic burden deteriorating living conditions in urban Colombia. In addition, extensive economic welfare improvements during the eighties and early nineties were partially reversed in the late nineties when the economy entered into recession. We find that, from 1978 to 1995, extreme poverty fell by nearly two thirds -from 28 to 10 percent- and income per capita almost doubled. But the impact of the recent recessive period, with its adverse effects on both the level and distribution of income, pushed economic welfare measures back to late eighties levels (without failing to notice the escalation of unemployment to unprecedented levels). In studying the poverty profile, we find that the typical faces of the poor -e.g., children of all ages, young lower-to-middle-skilled household heads, recent migrants and non-homeowners- have not changed much in the last two decades, but have become more ostensible and polarized. Causes and determinants of poverty such as low education endowments and high dependency ratios are becoming more powerful in predicting poverty. We next study how the dynamics of poverty in urban Colombia are linked to economic growth, inequality and the evolution of basic income per-capita generating factors. We conclude that income per capita growth, rather than changes in the income distribution, explains most of urban poverty dynamics during the last two decades. For the representative household, the key sources of income growth have been the rise in education endowments and the reduction in the dependency ratios -via smaller family size. Simultaneously, changes in employment ratios or wages made a positive contribution up to 1995, but had a symmetric and detrimental effect during the recession that followed. However, low-skilled-headed households not only reap the most benefits from lower fertility and more education, but some from higher wages as well. During the economic recession, most of the poverty rise was generated by losses of wage-earning jobs, and the remainder by lower eamings for the self-employed. Nevertheless, some additional downwards adjustment in average wages was obtained via increasing labor market participation of low-skilled women in wage-earning jobs and a reduction in the gender wage gap. Cross-city comparisons of social and economic indicators disclose a pattern of convergence. Laggard cities like Barranquilla showed the highest progress, while Califaced the most significant losses. 1. IINTRODUCTION Based on GNP per capita -$2,200 in 1997- Colombia is considered a lower-middle income country by international standards. With a population of nearly 42 million Colombia is the third most populous country in Latin America, with an expanding urban sector that accounts for nearly 62 percent of the total population. Similarly to other Latin American countries, Colombian exhibits a high level of inequality and consequently, poverty incidence is also relatively large. This paper tries to measure and explain the evolution of poverty, inequality and welfare of urban Colombians during the last two decades. In that period Colombia underwent significant socio-demographic changes, while, the past decade saw major institutional and economic changes, including: (i) significant reduction in economic growth, with increasing fiscal imbalance and higher unemployment; (ii) increasing inequality and vanishing poverty gains during the second half of the 1990s, (iii) worsening country-risk indicators, and (iv) significant structural reforms in several key public policy areas. Adverse macro environment in the late 1990s: recession and risk deterioration Colombia is facing its most severe economic crisis since the 1930s. Since 1997, after two decades of positive and sustained growth, economic activity plummeted to the point of reaching negative figures in 1999. This has translated into increasing unemployment (20.2 percent in March 2000) and widening inequality and a reversal of the poverty reduction trend observed since the 1970s. Up until the early 1990s, prudent management of the Colombian economy allowed for low government debt levels which, together with low inflation rates by Latin American standards, led to steady -although moderate- growth rates. However, public spending in the 1990s entered an increasing path that pushed its share from 24 to 36 percent of GDP between 1990 and 1998. Fiscal imbalances led Colombia towards economic slowdown after 1996, and into recession in 1998-99. The fiscal position continued to deteriorate as economic slowdown adversely affected tax revenues. In 1998, the current account and fiscal deficits both ran at 5 percent of GDP. Table 1.1. Colombia at a glance 1900-1999 1980 1988 1995 1999 Population (millions) 28.4 33.6 38.6 41.5 Population growth gate 2.2% 2.0%o 2.0% 1.8% GDP growth rate 4.1% 4.1% 5.2% 4.3% Unemployment Urban 7.71%* 10.3% 8.7% 19.7% Rural 1.46%* 4.6% 5.0%o 10.9% Inflation Rate 26.4% 28.1% 19.5% 9.2% Govemment debt (% of GDP) -0.7% -1.4% -2.3% -5.8% Real effective exchange rate 146.68 87.34 100.00 102.68 Goods & Services (% of GDP) Exports 11.0% 12.0% 14.8% 18.2% Inports 12.3% 10.2% 21.3% 19.2% * 1978 Financial risk indicators for Colombia also worsened recently: (i) After being considered one of the most stable economies in the region, recent performance has shown greater volatility.' IRodrik (1999) estimates the probabilities of entering episodes of high volatility for various countries, and shows that Colombia faced a probability very close to zero for the last thirty years. 2 International risk ratings of the Colombian economy were recently downgraded, increasing the cost of external borrowing; (ii) Financial sector performance has deteriorated and become more fragile, with significant bailout costs (See Partow, 2001 and Acosta, 2000). The rule of law is under severe challenge and the judiciary system is stretched to its limits. Crime and violence -political, drug related and common- remain very high.2 The widespread practice of extortion and kidnapping -which pervades all strata of society- is increasingly weakening property rights over physical assets and thereby undermining the market economy. Moreover, de facto authoritarian regimes enforced by local warlords -paramilitaries or guerrillas- in effect rule some rural areas. Judicial uncertainty is leading to unpredictability of public mandate in key policy areas, thus reinforcing market uncertainty. A number of recent judicial rulings created increasing uncertainty about the "rules of the game" in several key economic markets such as labor, mortgage credit and the private provision of education. In addition, constitutional rulings on pensions rights have cast doubts about the contingent liabilities of the public sector.3 As a result, political mechanisms appear to be losing effectiveness in solving the fiscal crisis. For example, it is relatively uncertain that the reforms of the pension system eventually passed by the elected legislative branch -which are urgently needed to alleviate the unsustainable fiscal deficit- will become a public mandate, since they may be overturned by the Constitutional Court. Supply shocks and multiple structural changes During the last decade, the Colombian economy experienced supply shocks and several structural reforms that significantly modified the labor market environment. These include the Labor Reform of 1990 and the Pensions and Health Insurance Reform of 1993, as well as a string of trade liberalization measures -dismantling import quotas and reducing average import tariffs from 40 to 11 percent over the same three years-. Moreover, free trade agreements were signed with Mexico and Venezuela and the capital account was opened in 1993. At the same time, the economy suffered supply shocks caused by major discoveries of oil reserves, which led to expectations of exchange rate appreciation and a jump in export revenues during the second half of the 1 990s.4 Major changes in socio-demographic characteristics: fertility, educational attainment and female labor force participation. The socio-demographic characteristics of the working population continued to shift during the last two decades. Fertility rates decreased throughout the entire period. The urban population increased its educational attainment and labor market experience, and women augmented their labor force participation and are facing significantly lower gender earnings differentials. Objectives This paper's objectives are to understand the evolution of urban poverty, inequality and welfare during the last two decades, i.e., to describe clearly and explore their main underlying economic factors. We want to understand how the poor are doing in econoniic and social terms and what identifiable factors affected the evolution of poverty and welfare in urban Colombia for the period 1978-99. The evolution of the urban population's welfare is examined from two perspectives: social development and monetary income. Social development encompasses human capital, household infrastructure and violence. Analysis of income welfare is more extensive and covers 2 Homicide rates are roughly 59 per hundred thousand -among the highest in the world. 3 Below, see Box I about mortgage credit. 4Ex-post oil revenues were below expected levels due to lower oil prices in this period. Current oil prices are well above average historic levels. 3 several questions. The first is whether or not urban Colombians are better off after the last two decades, and to what extent the answer to this question is ambiguous. That is, whether ranking Colombian's welfare in one year versus another depends on value judgments about inequality and/or on distributional weights in some social welfare function. The second question is about the faces of the poor. Which are their typical characteristics and how have they evolved over the last two decades? Two perspectives are taken: first, identifying their faces or most common socio- demographic characteristics (children or elderly, single female household heads, disabled, etc.) and, second, measuring their relative disadvantages in terms of income per capita generation (be it low skills, or too many dependants or scarce employment opportunities for working age adults or low wages). The third question is how the dynamics of poverty in urban Colombia are linked to economic growth, inequality and, in particular how they are linked to the evolution of basic income per capita generating factors. Special attention is given to the recent recessive period in order to understand the impact that the dynamics of labor markets (both wages and job creation) have had on the dynamics of poverty. Finally, the last question is about the sources of the increasing income inequality trends. Which are the key household characteristics that explain inequality? Which sources of income -labor, property, pensions, etc.- are pulling inequality upwards? Are there any demographic or lifecycle effects that generate some noise into the inequality indicators? Household survey data. We use household survey data covering the period 1978-99 for the seven "grandes ciudades" in the years 1978, 1988, 1995 and 1999. In this fashion, we capture developments of poverty and inequality during three distinct sub-periods: pre-reform (1978 and 1988), post-reform (1995) and the current economnic recession (1999). Economic indicators for the years 1978, 1988 and 1995 are adequate for comparison, as in these three years, economic activity is almost at the peak of the business cycle. Growth is close to or higher than 4 percent and unemployment is low, between 8 and 10 percent. 1999 represents the most recent and deepest recessive episode. We also compare the current social situation of the country's main cities, characterizing regional disparities and areas of regional convergence. Main findings Opposing trends in social development and crime intensity indicators Although most of the social indicators in education, health, and infrastructure show substantial and persistent long term improvements during the last two decades, the simultaneous escalation of violence reveals a considerable deterioration of living conditions and has become a considerable economic burden in urban Colombia. In education, primary and secondary schooling show substantial improvements, especially in completion rates. Despite being pro-cyclical, child labor shows decreasing trends. Basic infrastructure in electricity, water and sewerage continues to show progressive extension of coverage. Life expectancy has improved by approximately one year every two calendar years; however, most of the gains benefited females, as male life expectancy is ruthlessly limited by violence. Infant mortality and malnutrition were approximately halved during the last two decades. Only recently -probably due to the economic recession- has education coverage declined. Income and poverty: substantial long-term progress with a recent set back Economic welfare -measured by income per capita- almost doubled from 1978 to 1995, but deteriorated thereafter. Following robust gains in all welfare measures during the 1980s until and the first half of the 1990s, the negative impact of the recessive period pushed welfare levels back to late eighties levels. Moreover, the adverse effect was stronger on the poor population and as a result income inequality increased, impinging upon what had been at least a decade of welfare improvements. 4 In summary, although violence continues to hamper development, and has increased significantly during the last two decades, both social and economic indicators performed very well during the eighties and the first half of the nineties. However, the recent economic downturn produced a severe deterioration of poverty and inequality indicators. Urban regional convergence and improvement in left behind cities. On the bright side, our study shows strong regional convergence during the last 20 years, mostly with catch-up growth from the cities that were very far behind average indicators in 1978, especially Barranquilla. We also find important improvements in average educational attainment of the urban population, from 6 years in 1978 to nearly 9 in 1999. Notwithstanding this point, this chapter and others of the poverty report, will show that these improvements still fall short of the country's needs. Faces of the poor: unchanged but more polarized Although the faces of the poor in urban Colombia have not been changing during the past two decades, they have been becoming more polarized. The poor are disproportionately represented by children of all ages, household headed by young low- and middle-skilled or female individuals, recent migrants and non-homeowners. These groups are clearly worse off than pensioners, the better educated, the elderly and non-recent migrants. Poor households in urban Colombia suffer increasingly from higher dependency ratios and lower skill endowments and employment rates. In order for a household to escape poverty, it is increasingly necessary that members other than the head be employed. While the presence of children increases the risk of poverty, that of the elderly produces the opposite effect. Likewise, older working-age household heads face increasingly lower poverty risks. Location -in this case, the city in which a household resides- also plays a role but it has been shifting over time. City poverty risks diverged up to 1995, but converged in the recessive period, with Bogota severely hit by the recession. Finally, household occupational choice combining both wage and self-employment provided some protection from poverty until the recession. Lastly, under the current economic downturn, it is the young high school graduates who have been most hit by unemployment. Poverty dynamics: GDP per-capita growth is key for poverty reduction Economic growth explains most of the gains and losses in urban poverty during the last two decades. In the last four years, poverty gains have resulted from to the combined effect of negative growth, increasing inequality and higher relative prices of the poor's basket of goods. Nevertheless, it is evident that growth's effect on poverty has increased over time, meaning that the welfare of the poor is now more dependent on macroeconomic performance. Dynamic decomposition of income per capita growth for the average urban household shows that the key sources of growth have been a rise in education endowments and a reduction in dependency ratios -via smaller family size-, but not changes in employment ratios or wages. But across skill levels and labor market attachments, household income growth factors reveal some heterogeneity. Lower fertility, and higher wages and education (but not higher employment ratios) have had the most impact on low-skilled-headed households' incomes. At the same time, the proportion of households with only-self-employed workers has risen; this is the only group suffering a significant loss in income per capita growth via wage reductions. More than wage reductions, job losses generate most of the rise in poverty during the recession Loss of wage-earning jobs and, to a lesser extent, lower earnings for self-employed, explain much of the poverty increase during 1995-99 which disproportionately affected low-skilled-headed households. However, some additional adjustment in wages was obtained via higher female labor market participation and a reduction in the gender wage gap - in both the self-employed and wage- 5 earning markets. In summary, households in which income is generated only in the wage sector faced increasing unemployment; some of them managed to enter the self-employment market and took a severe drop in labor earnings, while the rest were left without any labor income source. Increasing Inequality Education, be it of the household head or other household members of working age, is the only variable that consistently accounts for any sizeable share of inequality between homogeneous groups of households over the period of analysis. Its effect on inequality has also risen over time. Labor income is the driving force behind increases in inequality, and, especially wage income, has recently become more regressive. The components of "other income" do not have a homogeneous impact on income inequality: housing rents tend to be progressive, whereas pensions and interest earnings have played a greater role in rising inequality. Organization of the paper. The paper is divided into four sections after this introduction. The next section briefly presents the evolution of social indicators in education, child labor, child nutrition, infant mortality, access to basic infrastructure, life expectancy and violence. The third section examines how far has Colombian's welfare improved during the last two decades. We analyze the evolution welfare with three alternative social welfare measures: income per capita, income per capita corrected by inequality (the Sen welfare index) and poverty indexes; and full income distribution rankings (first and second order stochastic dominance). The fourth section focuses on the characteristics of vulnerable groups and the risk of poverty faced by such groups. Apart from establishing the faces of the poor in each period, we assess the marginal impact of different income generating factors on the probability of being poor. The fifth section tries to interpret the dynamics of poverty in terms of both macro and micro determinants factors. Here, we compare typical families according to the school attainment of their head or type of occupational choice and establish whether, human capital endowments, wages, dependency ratios or employment opportunities (by gender) have been key variables explaining progress or regression into poverty. The sixth section is an introduction to sources of income inequality. Since increasing inequality accounted for the substantial welfare losses during the last decade, we try to identify the main determinants of its evolution, either from the perspective of demographic characteristics or from the different types of income. When appropriate, an analysis by city is included in the each section. 2. SOCIAL PROGRESS IN URBAN COLOMBIA: SUBSTANTIAL LONG-TERNM EPROVEMENTS, EXCEPT IN VIOLENCE AND CRIME In this section, we examine the evolution of several of social outcomes that describe non-monetary dimensions of welfare for the urban population in Colombia (see Tables 1.2 and 1.3). It includes variables that affect living conditions such as illiteracy rate, school enrollment, educational attainment, grade completion rates, infant mortality, child malnutrition and labor, life expectancy, access to basic infrastructure services, and finally, violence. Tables A.1A-G report selected social indicators for seven major cities: Bogota, Barranquilla, Bucaramanga, Cali, Manizalez, Medellin and Pasto. (Distributional issues on access to social services and incidence of public social expenditure are discussed in Chapter 5.5) 5 Vlez and Foster (2001). 6 Table 1.2. Social indicators, Urban Colombia' 1978-1999 1978 1988 1995 1999 Average education> I8 years 6.2 7.7 8.4 8.9 Illiteracy rate2 5.3% 3.3% 2.8% 2.6% School enrollment Ages 7 to II 91.8% 94.8% 96.5% 95.3% Ages 12 to 17 76.9% 80.5% 84.4% 82.2% Ages 18 to22 31.2% 35.8% 41.0% 36.3% Complete primary school (ages 12 to 17) 67.0% 78 7% 77.7% 89.8% Complete high school (ages 18 to 22) 21.6% 35.3% 48.7% 59.2% Child labor Ages 12 to 16 12.0% 11.5% 9.9% 9.5% Ages 12 to 14 5.8% 5.0% 5.2% 3.7% Child Malnutrition3 Stunting, low height for age 16.8% 10.1% 8.4% 6.7% Wasting, low weight for height 22.4% 16.6% 15.0% 13.5% Low weight for age 4.9% 2.9% 1.4% 0.8% Crime4 Homicides (Per 100,000 pop.) 26 62 65 59* Access to public utilities Electricity NA 99.3% 99.6% 99.4% Water NA 97.4% 97.7% 99.0% Telephone NA 62.2% 71.0% 84.2% Sewerage NA 94.8% 96 0% 97.3% 1. Urban Colombia represents 67% of Colombia urban area: Barranquilla, Bucaramanga, Bogota, Manizales, Medelln, Cali and Pasto. 2. For population 12 years old & older. 3. For population under 5 years old; represents national data for 1977, 1986, 1995, and 2000. 4. Levitt and Rubio, 2000. * i998 figure Sources: Authors' calculations based on Encuesta Nacional de Hogares; Profamitia; and Encuesta Nacional de Demografa y Salud. 2.1. Schooling and education: increasing trend with pro-cyclical fluctuations During the last two decades, average educational attainment increased by 2.7 years and illiteracy rates was nearly halved. The average educational attainment of the adult population increased from 6.2 to 8.9 years over the entire period. Illiteracy rates were nearly halved between 1978 and 1999 in urban Colombia -from 11 to 6 percent. For individuals above 12 years of age, illiteracy rates fell from 5 percent to 2 percent in the respective years. School enrollment: slow progress partially reversed during economic recession. School enrollment is an educational indicator more closely related to the economic cycle. We report enrollment for three different age groups of boys and girls (7-11 years, 12-17 and 18-22), capturing primary, secondary and higher education enrollments. For all three age categories, these rates grew very slowly between 1978 and 1995 but decreased in the latest four years. The recent drop in the university-age population's enrollment is especially worrisome: in 1978, 31 percent of 18-22 year- olds were enrolled in school and this proportion was 36 percent in 1999, having fallen from its peak of 41 percent in 1995. The 64 percent of young adults not enrolled in school belong mostly to the economically active population and face rates of unemployment twice as high as the rest of the adult population, a fact that reveals their diminishing chances of avoiding poverty in the long run. 7 But completion rates improved more rapidly for primary and secondary schooling. The share of individuals within the intermediate age group -12 to 17- who have completed primary school rose from 67 to 90 percent between 1978 and 1999, while among the older group -18 to 22- the proportion of high school graduates increased from 21 to 59 percent. Coverage increased for all education levels and was marginally progressive for secondary and tertiary education. Chapter 6 (Tables 1.8 and 1.9) shows that coverage in primary, secondary and tertiary education increased by 8, 12 and 6 percent respectively between 1992 and 1997. Coverage rates improve for all income quintiles and access to education became much more progressive in the case of secondary schooling and only moderately progressive in the case of tertiary. The concentration coefficient was minus 0.563 (!) for secondary education and 0.403 for tertiary (still slightly better than the previous 0.445). Regional Comparisons Regional disparities in illiteracy tend to diminish. The overall improvement in illiteracy hides important regional disparities in the rates of improvements. llliteracy reductions since 1978 were especially sharp in Barranquilla, Bucaramanga, Manizales and Medellin, even through the rates remain above the national 5.7 percent average in 1999. These four cities started far behind the national average with iliteracy rates of 14 percent or above in 1978, but by 1999, they displayed illiteracy rates no higher than 7.5 percent. The catch-up was especially rapid for the city of Bucaramanga, which fared worst in 1978 at 17.8 percent illiteracy. In all cities, except in Bogota and Manizales, improvements mostly slowed down or stagnated between 1995 and 1999. The evolution of average educational attainment across cities follows a pattern of strong convergence. In 1978, the only city that displayed above average schooling was BogotA (7 vs. 6.2 years), followed by Cali with 6 years. Medellin, Manizales and Bucaramanga showed very close mean educational attainment, around 5.5 years, while BarranquiUa was far behind with 4.6 years. Thus, the ratio between the highest and lowest cities was 1.5. In 1999, when the urban average reached 9.1 years, the cities that fared the worst were Bucaramanga and Medellin, with 8.1 and 8.3 years of average educational attainment, respectively. BogotA was still the highest, with 9.8 years, with Barranquilla, now the second highest, one year behind. Cali, Manizales and Pasto stayed close with an average educational attainment approaching 8.7 years. Thus, average schooling converged anong cities as the ratio between the highest and lowest cities decayed to 1.2, and most importantly, the "behind" city of 1978, Barranquilla, again showed the highest improvement in this indicator. Across cities, school enrollment for age groups 7-11 and 12-17 shows strong convergence up to 1995, with some heterogeneity in the recessive period. Up to 1995, while Medellin and Cali caught up with the rest of the urban areas, Manizales and Pasto lost some of their significant advantage. In the recessive period, the worst reductions in enrollment occurred in Barranquilla, Manizales, Pasto and Cali -especially in the latter for high-school-aged children, with a drop of 8 percentage points between 1995 and 1999. School enrollment for individuals above 18 years of age, shows some persistent heterogeneity across cities. After being far behind, Barranquilla and Bucaramanga caught up to the urban average: the former mostly in the 80s and the latter in the early 90s. In contrast, over the years, Cali and Medellin remained persistently behind -by more than 5 and 8 percentage points, respectively. Simultaneously, in 1995 both Manizales and Pasto lost their relative advantages, and subsequently 8 followed diverging paths: Manizales recovering to 4 percent above the urban average and Pasto falling together with the urban trend.6 2.2. Child labor: pro-cyclical with a decreasing trend Child labor tends to fall slowly and seems pro-cyclical. Concurrently with improvements in school enrollment for adolescents, urban child labor showed a decreasing trend during the last two decades. The proportion of children aged 12 to 16 years who work decreased from 12.0 percent in 1978 to 9.5 percent in 1999, a low 2.5 percentage points over 21 years. Child labor is sensitive to variations in the level of economic activity. As in other Latin American countries, the proportion of working children in urban Colombia seems to fall during recessive periods of the economic cycle.7 The rationale is that the opportunity cost of sending children to school is higher during economic booms. The reason is that although income shortages become higher during economic recessions, incentives for sending children to school may increase as opportunity costs decrease. For example, children in school receive a guaranteed meal per day along with day care, allowing parents to work or take care of home activities. In addition, the benefit of the alternative choice falls with the probability of children finding jobs during recessions, a fact we will confirm below when we present unemployment rates by age groups. Defined in stricter terms, as the proportion of children from 12 to 14 years of age participating in the labor market, child labor remained practically unchanged from 1978 to 1995 at around 5 to 6 percent, and fell abruptly with 1999's recession to 3.7 percent. Child labor differs dramatically across cities. Child labor was consistently higher than average in Bucaramanga, Cali and Pasto over the whole period. In Cali, sharp increases occurred between 1995 and 1999, leading to rates higher than those of 1978. This was accompanied by a large fall (8 percentage points) in high-school-aged children's school enrollment rates. In contrast, one should note that child labor for both age groups remained significantly below average in Barranquilla and Medellin over the period. 2.3. Child malnutrition and infant mortality: continuous improvement Child malnutrition was more than halved between 1977 and 2000. We use data from 1977 and 1986 surveys by the National Institute of Health as well as the 1995 and 2000 Demography and Health Surveys for Colombia. The proportion of children under 5 years of age who are underweight fell from 16.8 to 6.7 percent between 1977 and 2000. This still represents around 120,000 malnourished children, a high number considering that the effects of malnutrition will cause irreversible welfare losses lingering throughout their entire lives.8 The proportion of children displaying stunting (low height for age), an indicator of chronic malnutrition, remains at 13.5 percent in 2000, six tenths of the 1977 proportion, a more moderate improvement. The proportion of children facing acute malnutrition (low weight for height) dropped from 4.9 percent to 0.8 percent between 1977 and 2000. These rates remain below the Latin America and Caribbean average. 6 In fact, the spectacular decrease of university-age enrollment for Pasto between 1978 and 1995 makes us doubt the veracity of the data for 1978, when school enrollment among the 18-22 years of age group was twice as large as the national average. 7 Cunningham, W. and W. Maloney, 2000. "Child Labor and Schooling Decisions." Mimeo, The World Bank, Washington, D.C. 8 Malnourished children are less likely to perform well in school and more likely to become disabled and/or economically dependent and thus turn into an economic liability for their families and society as a whole. 9 Infant mortality has been faliHng steadily over the last 20 years. As shown in Table 1.3, rates decreased from 44.3 per thousand in 1981 to 33.9 in 1990-959 and fell afterwards. Data from the 1995 and 2000 DHS surveys confirm the continued decline, with another decrease of 7 per thousand. Table 1.3. National infant mortality (under 1 year of age, rates per 100= born) Source 1981-82 1983-84 1986-87 1989-90 1990-95 1995-2000 1993 census 44 41 37 37 34 NA DHS survey NA NA NA NA 28 21 Source: Profamilia and Encuesta Nacional de Demografia y Salud 2.4. Access to Basic Infrastructure: progressive gains with regional disparitiesl0 Coverage of public utilities has grown, but aqueduct coverage still has room for improvement. Next, we record access of the population to different public utilities (water, electricity, telephone and sewerage), as indicators of satisfied basic needs. Since coverage for electricity and water reaches 99 percent (Table 1.2), it means that only small pockets of poverty -approximately 150,000 people- have no access to these basic services. Sewerage was not available to 5.2 percent of the population in 1988, a proportion nearly halved in 1999 -to 2.7 percent-. While almost 38 percent of the urban population did not have access to a telephone in their homes in 1988, this figure was reduced to 16 percent in 1999. Nevertheless, some alternative evidence suggests that National lHousehold Survey -NIHIS- indicators for aqueduct coverage might be biased upwards in the sense that water services are actually highly heterogeneous in both continuity and quality. The 1998 official figures for the Cambio para Construir la Paz, based on the 1996 national inventory on safe water and basic hygiene, found that real coverage differs from nominal coverage, mainly because having aqueduct connection does not imply access to potable water. Coverage rates in urban areas were 83 and 79 percent for aqueduct and sewerage, respectively -about 17 percentage points lower than the NHS figures. Moreover, it was found that 60 percent of the water Colombians have access to could lead to "health crises" and that 70 percent lacks adequate treatment. Likewise, the piped water supply is often highly discontinuous (World Bank, 2000). The growth in coverage for public utilities was pDrogressive, although income disparities in telephone access are still a cause of concern. Progressive marginal connections to public utilities, between 1988 and 1999, helped the lower income quintiles to catch up with the middle and high- income groups. For example, sewerage coverage increased by 7.6 and 3.4 percentage points for the two poorest quintiles of the household income distribution, compared to 0.3 and -0.3 percentage points for the two richest ones. The improvement in access to telephone was particularly spectacular for poorer households, with increases of 33.2, 32.1, 23.2, 15.8 and 5.3 percentage points for income quintiles 1, 2, 3, 4 and 5, respectively. Access to water grew by 5 percentage points for the first quintile, which is reaching full coverage. The ratio of coverage of the richest quintile to the poorest one thus reduced from 1.07 to 1.01 for water, 1.15 to 1.06 for sewerage and 2.55 to 1.40 for telephone. Despite convergence, regional dspearities in access to public utilities subsist: mainly in telephone and sewerage. Overall, we observe regional convergence in access to basic public utilities between 1988 and 1999, as shown by the decreasing ratios of highest to lowest coverage among cities (e.g. 9 Estimation from 1993 Census (Departamento Nacional de Planeaci6n). 10 A more detailed account of the evolution and incidence of access to public utilities is given in Chapter 5, The Distributional Impact of Public Expenditure. 10 1.89 to 1.27 for telephones, 1.01 to 1.00 for electricity). However, lack of coverage is concentrated in regional pockets. Despite economic prosperity during the past decade, in Barranquilla, 17 percent of the population do not benefit from sewerage and 51 percent do not have access to a telephone in their homes. Access to aqueduct also lags slightly behind the national average. Except for telephone lines, Pasto displayed higher Levels than the national average but very slow improvements. Home telephone lines were still missing for 48 percent of the population in this city in 1999, representing approximately three fourths of the poor. Access to these four services was better in Bogota except for electricity, which still is lacking for 0.9 percent of the population -equivalent to 50,000 people-, a slightly higher proportion than in 1995. 2.5. Life expectancy: advancing with a gender bias Life expectancy has increased by nearly 21 years during the last four decades, with a strong advantage for women and considerable regional differences. As shown in Figure 1.1, below, women have benefited most from these gains: the female-male gap was 3.4 years in the I950s, started widening to 6.6 in the late 1970s and rose to 8.3 in 1995, before returning to 6.3 at present. During the last decade, the gender gap varied widely between departamentos, with the maximum in Antioquia (14.2 years, 162 percent the national average!) and the minimum in Nariiio (4.4 years). Regional Comparisons across departamentos show that life expectancy is much higher in those in the Atlantic region.'" Apparently, longer life expectancy seems more prevalent in the least prosperous areas. Finally, the departments with higher life expectancy also reveal lower gender gaps. Most of the gender difference is associated with the extremely high risk of being murdered among the young Colombian males. Colombian men between 15 and 35 years of age -are 15 times (!) more likely to be homicide victims than women of the same cohort and twice more likely than men above 45 years of age (see Figure 1.2). Figure 1.1. Life expectancy by gender, Figure 1.2. Homicide rates by age and gender, Colombia, 1950-2000 Colombia, 1999 8o 225 7520 70 175 70 - , , , , , , , , , , 0 -,* 65 8125- 1945 195 1955 196D 1965 1970 1975 I80 1985 1990 1995 2000 200 15 - 24 25 - 34 35-4 45 - 59 60 & over Source: Departamento Nadional de Planeacl6n Age g roup 2.6. Violence: continuous aggravation since the 1970s After tripling from 1970 to 1991, homicide rates decreased in the 1990s, while extortion, kidnapping, car theft and armed robbery rates kept growing.12 Homicide rates almost tripled in the 1970s and 1980s and then fell by more than 20 percent, after the 1991 peak, when almost one in a thousand Colombians was murdered. The rates remain three times as high as in Brazil or Mexico and "1 With the cxception of males in the department of la Guajira. 1 25 12.6 Vioene:it,S contin.uoubggaaio n 00 sinersthein Crm1 nCoobaan7htCn0sDn Aoti, F edesf oLevi, S.rkand M.pe Ruio 2000 "Unoderstombandn. rm nClmi n htCnb oeAoti, are only surpassed by those in El Salvador. While almost all of the decline can be attributed to a decrease in Bogota, Medellin and Cali, homicide rates in Medellin are maintained at more than double the national average. Violence spread over the country between 1990 and 1997, in the sense that homicide rates decreased in the most violent areas while they increased in areas considered less dangerous. Contrary to homicide rates, extortion, car theft and armed robbery rates kept growing during the nineties, leading Colombia to display a remarkable rate of victimization, at more than 35 percent of households in 1997. Nevertheless, although Colombia still has one of highest homicide rates in the world, it ranks 12h in victimization out of 18 Latin American countries for which the information is available. According to the best empirical evidence available, drug trade seems to be the main explanation behind the staggering homicide rates pirevalent in Colombia today. Multiple empirical studies have tried to explain the determinants of homicide and crime in Colombia.'3 Five alternative hypotheses have been tested (1) the illegal drug trade, (2) impunity, (3) the presence of extra- governmental groups (guerrillas and paramilitaries) that have taken over traditional governmental roles in parts of the country, (4) poverty and/or income inequality, and (5) the possibility that Colombia's decades of internal strife has created a populous that is simply more innately "violence prone."'4 Sanchez and Nnfiez (2000) perform a comprehensive test of the alternative hypotheses with the most complete data set available for crime across different municipalities in Colombia.'5 Their main findings indicate that socioeconomic variables such as inequality, poverty, political exclusion, and lack of education have a positive effect on the crime rate. However, jointly, they only explain 6 to 12 percent of the total variability of the homicide rate. The rest -nearly 90 percent- is mainly explained by the intensity of illegal drug trade activities and its interaction with the presence of illegal anned groups -guerrillas and paramilitaries-. According to Levitt and Rubio (2000), these findings are consistent with international evidence that indicate a positive covariance between homicide rates and periods of intense illegal drug trade. Violence generates incireasing social costs and demands substantial public resources. Apart from the 30,000 lives lost per year in the war against illicit drugs, Colombia has sacrificed not only police and militayr officers but also several presidential candidates, politicians, intellectuals, and journalists.' Estimates of the human capital loss due to homicide are at least 1 percent of GDP.17 Moreover, investment and educational and labor market opportunities become severely limited in an atmosphere of violence and insecurity.18 At the same time, the demand for public resources to fight crime has escalated. Public expenditure in justice and security has more than doubled its share of GDP during the last decade. While current public expenditure in security and justice is around 5 percent, in 1990 it was close to only 2 percent of GDP. Simultaneously, private expenditure in security seems to be increasing even more rapidly: The ratio of policemen to private guards decreased from 2.5 in 1980 to only 1.0 in 1995.'9 Additionally, the World Health Organization estimates that 13 Sanchez and Nuiiez (2000) mention at least ten studies: Comisi6n de Estudio de la violencia (1995), Montenegro y Posada (1995), Gaviria (1998), Echeverri y Partow (1998), Echandia (1999), Sarmiento (1999), Moser (1999), L6pez and Garcia (1999) and Rubio (1997, 1999) 4 Levitt and Rubio (2000). 15 As mentioned above, variability of homicide rates across municipalities is very high. 16 Approximately 600 lives per year are lost in massacres, half of them in the Department of Antioquia. DNP- UNDP (1999). '7 Capital-cost estimates are based on decreases in disability-adjusted life years (DALYs); Trujillo and Badel (1998) and Londonlo (1996) quoted by Levitt and Rubio (2000). Additional costs in terms of medical assistance are provided by Bonadilla et al. (1995). IS See section II in Levitt and Rubio (2000) for detailed estimates. 9 Comisi6n de Racionalizaci6n del Gasto y de las Finanzas Ptblicas (1996) quoted by Levitt and Rubio (2000). 12 violence's effects on health care costs in Colombia are in the magnitude of 5 percent of its GDP, an undoubtedly alarmng proportion.20 Our main findings are contradictory in the sense that the escalation of violence -with its perverse effects on life expectancy among males and increasing overall social cost- reveals a considerable deterioration of living conditions in urban Colombia, yet, simultaneously, most social indicators in education, health and infrastructure show quite positive developments during the last two decades.2' 3. ECONOMIC WELFARE, INCOM:E DISTRIBUTION AND POVERTY: ARE URBAN COLOMBIANS BETTER OFF THAN TWO DECADES AGO? Are urban Colombians better off than two decades ago? In this section, we assess the evolution of economic welfare in urban Colombia from 1978 to 1999 with a set of alternative indicators that place varying emphasis on the social weight of income groups within the Colombian population. We restrict the analysis to individual welfare measured by monthly household per capita income and attempt to quantify the levels of welfare by introducing different measures with increasing level of complexity: (i) average income per capita, (ii) income per capita corrected by inequality and lifecycle effects (Sen and Paglin indices), (iii) poverty measures, and (iv) full income distribution comparisons. Measures of social welfare and distributional weights. In principle, social welfare should be a function of the welfare of all individuals. The simplest measure of welfare is the average household income per capita, incorporating two basic principles: "more is better than less" and resources should be measured with respect to needs -as the ratio of household income to family size. However, by using average household income per capita, the social weights of households are independent of their ranking within the distribution of income: a one percent increase in GNP has the same social value, independent of who benefits from it. If social welfare is considered to improve as a result of a transfer from the richer to the poor -the "principle of transfers"22-, then a more equal distribution of income brings larger social welfare -given the same average income per capita level-. Hence, social welfare becomes an index combining average income per capita and some measure of inequality -for example, the Gini coefficient. In this fashion, we avoid the error of considering solely the degree of inequality as a measure of social welfare in itself. If the welfare of the poor is of outmost importance, social welfare should be represented by poverty measures. This is equivalent to concentrating all the social weights of the social welfare function in the lower tail of the distribution of income and creates a welfare measure that ignores income variations among the non-poor. Social welfare rankings independent of distributional weights. Somehow, all three measures of social welfare proposed above incorporate a specific vector of social weights to households, according to their relative position within the distribution of income. Therefore, our assessment of the evolution of welfare over time might produce contradictory statements, depending on the type of indicator -and the implicit distributional weights- chosen. If possible, it would be desirable to avoid this controversy based on value judgments by producing a ranking of distributions of individual welfare, independent of social weights. Fortunately, full income distribution comparisons yield such 20 World Report on Violence and Health, World Health Organization, 2002. 21 See DNP-UNDP (1999), Chapter 1 for estimates of Human Development Index incorporating life expectancy estimates 22 Without reversing their relative positions (Dalton, 1920). See Deaton (1997), chapter 3. 13 rankings of welfare, and require minimum assumptions about social welfare in relation to income.23 As we will show below, when first order stochastic doninance and/or generalized Lorenz dominance hold, unambiguous statements about social welfare comparisons over time can be pronounced. In addition, a powerful corollary: when these dominance measures stand, all three previous measures of welfare -average income per capita, income corrected by inequality and poverty measures- must necessarily move in the same direction. 3.1. ]Poverty Measures: substantial long-term progress with a recent set back. We first present basic poverty indicators, which we follow over time. In Tables 1.4 and 1.5, we show the evolution of the poverty headcount, the poverty gap and the P2 index. The poverty gap describes the average distance of the poor from the poverty line. By expressing it as a percentage of the poverty line, we are in fact approximating the value of the resources necessary to bring the poor to the poverty line. The P2 index captures the degree of income inequality among the poor. Table 1.4. Income inequality and poverty indicators, Urban Colombia 1978-1999 1978 1988 1995 1999 Poverty Poverty rate 70% 55% 48% 55% Poverty Gap 35% 23% 19% 26% FGT P(2)2 21% 13% 10% 15% Extreme poverty rate 27% 17% 10% 14% US$ 2 per day poverty3 23% 13% 8% 12% Mean income per capita' 157,080 235,163 294,522 277,469 Income inequality Gini 0.473 0.486 0.522 0.545 Entropy Measure EO 0.34 0.41 0.48 0.54 Entropy Measure El 0.43 0.47 0.62 0.60 Entropy Measure E2 0.77 1.04 2.15 1.50 P90/PlO 8.0 8.5 8.4 12.0 P75/P25 3.0 2.9 2.9 3.5 Shareq5/Shareql 11.3 11.8 13.3 17.2 1. 1999 pesos, based on monthly household income. 2. Foster-Greer-Thorbecke index. 3. Based on Purchasing Power Parity Convertors from WDI database. 23 Basically, social welfare needs to be a monotone non-decreasing concave function of individual income in the case of Lorenz dominance. This can be relaxed partially for First Order Stochastic Dominance. For detailed explanations, see Deaton (1997), chapter 3, Champernowne and Cowell (1998) chapters 4 and 5. 14 Table 1.5. Change in poverty and extreme poverty rates, Urban Colombia 1978-1988 1988-1995 1995-1999 Poverty Extreme poverty Poverty Extreme poverty Poverty Extreme poverty Urban Colombia -15% -10% -7% -8% 7% 5% Barranquilla -22% -20% -12% -10% 6% -3% Bucaramanga -11% -8% -15% -9% 16% 3% Bogota -14% -6% -5% -7% 6% 7% Manizales -7% 1% -3% -16% -3% 4% Medellin -19% -15% -7% -8% 12% 5% Cali -14% -11% -6% -6% 5% 4% Pasto -7% -9% -12% -12% 7% 4% 1. Level change in percentage points Definition of poverty and extreme poverty. Extreme poverty (for the purposes of this study) is defined as a situation in which a person lacks the income necessary to purchase a very basic basket of food products that provides minimum caloric intake. We consider an individual to be extremely poor if his/her household per capita income falls below the extreme poverty line as defined by DANE,24 based on the 1985 Survey of Expenses and Income. A person is considered to be poor if his/her household per capita income is below the poverty line, a multiple (between 2 and 2.5) of the extreme poverty line, to take into account other basic necessities. We use city-specific poverty lines, incorporating cost-of-living differentials. After a continuous and significant reduction of poverty from 1978 to 1995, economic recession has pushed poverty indicators back to 1988 levels. Poverty decreased at rates close to 1.5 percentage points per year between 1978 and 1995. Unfortunately, the 1999 rates are again close to the 1988 ones, even slightly higher for the poverty gap. The recession not only increased the number of poor people, but the poor also became poorer, an observation also confirmed by the increase of extreme poor people, as reported in Table 1.5. Economic recession hits the extreme poor more badly and erases the substantial gains of the 1978-95 period. Although extreme poverty had sharply decreased with income growth (the 1978-95 period showed a 70 percent decrease), half of these gains were lost in the last 4 years. Colombia's economic recession seems to more severely hurt the very poorest segments of the population. The behavior of the P2 measure provides one more confirmation: it rose by 22 percent between 1995 and 1999, indicating both deeper poverty and a higher level of income inequality among the poor. It is worth noting that all these results are very robust to the poverty line chosen.25 In number terms, the evolution of poverty rates implies that in 1978 almost 5.4 million persons lived in poverty and 2.3 millions in extreme poverty in the main seven cities of Colombia. In 1995, the poor numbered 5.7 million and the extreme poor 1.15 million. The figures for 1999 are respectively 8.0 and 2.1 million. With some disparities, from 1978 to 1995, significant poverty reductions were visible across all cities; however, economic recession reversed the tendency, except in Manizales. Despite their low average per capita income, the least poor cities are consistently Bucaramanga and Pasto (see Table A.2). While Barranquilla remained the poorest city, it is also the one which saw the highest decrease in the poverty headcount, from 90 percent in 1978 to 62 percent in 1999. During the 1978- 95 period, Barranquilla and Bucaramanga obtained the largest reduction in poverty counts -34 and 26 24 Departamento Administrativo Nacional de Estadistica, BogotA, Colombia. 25 We computed poverty and extreme poverty rates for US$1 and US$2 per day lines as well as a wide range of hypothetical poverty lines, with no change to the above-described behaviors. 15 percent respectively. However, at the same time, BogotA and Manizales obtained the smallest improvements in poverty. Not all cities were equally affected by the crisis; Bucaramanga and MedellIn suffered the most in terms of poverty. The share of the poor rose from 36 percent in 1995 to 52 percent in 1999 in Bucaramanga and from 49 to 62 percent in Medellin. In that period, the only city that managed a continued decrease in the poverty headcount is Manizales. All others experienced steep increases, between 5 and 16 percent, with rising inequality among the poor. Extreme poverty followed a similar U-turn pattern, except in Barranquilla. Table 1.5 reports the changes in extreme poverty. The progress was huge between 1978 and 1995, period in which every city reduced this indicator to -at most- one third of its original value. In 1978, the highest share was also found in Barranquilla (49.0 percent), while the lowest could be detected in BogotA (20.9 percent), closely followed by Bucaramanga (21.8 percent). In 1999, the highest proportion of extremely poor was observed in Pasto (20.9 percent), and the lowest one in Bucaramanga (8.6 percent).26 Barranquilla and Medellin obtain the highest reductions in extreme poverty -31 and 23 percent respectively-. In the ]ast four years, on the other hand, all cities but Barranquilla saw extreme poverty rising between 3 and 7 percent, with the highest increase in BogotA and Medellin. Again, this pattern indicates that in Bogota the crisis hit more powerfully those at the very bottom of the distribution. With a 1999 extreme poverty rate of 15.5 percent, the capital city has lost most of its relative advantage of the late seventies, and ranks near the bottom together with Barranquilla and Pasto. The intensity of poverty fNlls dramatically untfil 1995, but in 19W it deepens and becomes more unequal. The evolution of the poverty gap -average distance of the poor to the poverty line- follows closely that of the extreme poor indicator. The poverty gap decreased substantially over the 1978-95 period (more than 40 percent in every city), and it rose in the latest four years in all cities. The growth of the poverty gap ratio varied between 2 and 10 percentage points, it was especially steep in Medellin (from 18.7 to 28.7 percent) and Bucaramanga (from 11.4 to 20.5 percent). It is important to note that in three large cities, (Bogota, Medellifn, and Barranquilla) this ratio ended up above 25 percent in 1999, while the three smallest cities showed better figures, albeit above 20 percent. Finally, the P2 measure displayed huge increases between 1995 and 1999, in all cities, especially in Bucaramanga where it more than doubled. This reveals not only that the depth of poverty increased, but also that inequality among the poor spread. 3.2. Average household income per capita Average household income per capita doubled between 1978 and 1995, but the recession erased at least 6 years of economic progress. A simple way to assess welfare changes is to consider the evolution of average household per capita income. Average household income per capita almost doubled: it went from C$157,080" (US$250) in 1978 to C$277,469 (US$441) in 1999. This overall increase hides very different patterns among the sub-periods. Average income per capita grew at an average yearly rate of 4.1 percent during 1978-88, 2.9 percent between 1988-95 but fell by 1.2 percent between 1995 and 1999, reflecting the severity of the crisis. The economic downturn has erased at least 6 or 7 years of economic progress for the average urban household. 26 Somewhat paradoxically, Pasto also exhibited the lowest poverty count in 1999. This suggests an atypical shape of the lower tail of its income distribution. 27 All monetary figures are in 1999 C$. The PPP adjusted exchange rate is US$ I = C$ 629. 16 Figure 1.3. Monthly mean household income per capita for the seven major Colombian cities, Urban Colombia, 1978, 1988, 1995, 1999 vo 400,000 0 a> 350,000 Xa 300,000 = 250,000 E 0 200,000- ° 150,000 0~~~~~~~~~~~~~00 o -C 100,000- 50,000 1975 1988 1995 1999 |--*Bannqa. -O *Bueannanp- -Bogta 11Manizaies I-Mefilh, --Cali _-4'Pasto Across cities, there is a pattern of convergence in average income per capita, with the exception of Bogota. The average trend in income changes is reflected to varying degrees among the 7 cities of the sample. In Figure 1.3, each line depicts the evolution of the city's average household per capita income. In all cities, the 1978-88 decade registered important welfare gains, although they were smaller for Manizales and Pasto. Over the next 7 years, Bogota, Bucaramanga, Manizales and Barranquilla experienced high rates of growth, while in Pasto, Medellin and Cali incomes grew at a slower pace. In the interval 1995-99, all cities except Pasto lost average income and as a result, a pattern of convergence became clearly visible for all urban areas, with the exception of BogotA. While average per capita income almost doubled, the maximum differences in income per capita, which were nearly C$80,000 in 1978, narrowed down to less than C$50,000 in 1999. Barranquilla grew most during the whole period; Manizales lost most. Each city followed its own trajectory. First, income gains were truly spectacular in Barranquilla, where mean household per capita income was multiplied by 2.65 over the entire period, with the highest growth achieved between 1978-95. Bogota started at higher welfare levels and saw the second highest gains -income nearly doubled over the period (1.9). Manizales experienced the lowest overall improvement at 1.39, with slow growth between 1978 and 1995 and a 10 percent decrease in the latest sub-period. In Bucaramanga, income was multiplied by 1.45 with a strong 93 percent gain (the third highest) between 1978 and 1995, before sharply decreasing by 25 percent in the next four years, the worst drop. Medellin and Cali would be the "average" cities, with growth slowing down progressively, for a total multiplication of income of 1.54 and 1.52 respectively. On the other hand, while Pasto also saw a total multiplication of 1.54, the evolution was quite different: it registered only 25 percent growth between 1978 and 1988, and was the only city still growing, albeit very little, between 1995 and 1999. 17 Figure 1.4. Deviation from average mean income per capita, adjusted by local cost of living. Urban Colombia 197g, 1988, 1995, 1999 30%/6 20% 10% _ 0 0/ ° 0% -10% - u D -20% 0L -30% -50% 1978 1988 1995 1999 Year 1-arranqulla - 0 Bucaramanga OBogotd -Manizales -- BMedelfn OCaI. -4 Pasto When income is adjusted by differentials in local cost of living, convergence is even stronger. In order to compare relative welfare among the cities, Figure 1.4 reports deviations from average urban income adjusting for differential cost of living. The price correction always brings substantial improvements in relative income in Bucaramanga, Manizales and Pasto, while reducing it in Bogota and Cali, except in 1999, when this correction improves Cali's relative standing -indicating some substantial reductions in local cost of living during the latest period. Despite these effects, the net losers from 1978 to 1999 are the cities of Bucaramanga, Pasto and Manizales, which drop more than 20 percentage points in relation to the urban average. Important losses are also visible in Medellin (12 percent), mostly concentrated in the late recessive period. After these adjustments, the major gains of Barranquilla are of similar size and Manizales still looses the most, but Bogota's relative advantage is reduced to 12 percentage points -half of the unadjusted figure. Although cost of living adjustments do not change the urban dispersion of average income in the starting period, they reveal a stronger convergence at the end of the period. Maximum to minimum differences went from 66 percent in 1978 to 31 percent in 1999, while correspondingly unadjusted figures were 63 and 53 percent. During the recessive period, cost of living changes produce surprising asymmetric effects on welfare in Cali and Bucaramanga. While nominal income losses from 1995 to 1999 in Cali are overcompensated by reductions in cost of living, the increase in cost of living exacerbated the substantial reductions in income sustained by Bucaramanga. Mean income per capita in this city (adjusted by cost of living) lost 24 percentage points relative to the urban average in a period of five years! 18 3.3. Welfare as average income corrected by inequality: The Gini coefficient and the Sen Welfare Index The evolution of income inequality: deteriorating mostly in the nineties Several authors have identified the mid 1960s as the breaking point in the regressive trend of income distribution during the first half of the XXth century.28 By the late 1970s, the Colombian economy had completed two decades of consistent reduction in income inequality and had improved its standing with respect to other Latin American countries. After the persistently raising inequality of the first half of the XXth century, substantial inequality reductions were observed during 1960s and 1970s as the economy grew. Colombia appeared as exemplary of Kuznets' well-known inverted U- shaped curve. However, during the late 1970s and the 1980s, inequality levels plateaued, and during the last decade took a clear "U-turn," erasing the equity gains of the two previous decades. Figure 1.5. Evolution of mean per capita income by quintile, Urban Colombia. (Index 1978=1) 2.2 2 , 2 //---~~~~~~--a 1 6 1 4 1 2 1978 1988 1995 1999 1-U.Ouint1le 1 *QuinUle 2 O-Quintile 3 -_-GuInIe 4 9 OuIndle 5 Quintile evolution. To start the analysis of inequality changes in the distribution of urban income, we examine the paths of average income by quintiles. Figure 1.5 presents the evolution of household mean per capita income by quintile, normalizing the average income of each quintile to one in 1978. As depicted, until at least 1995, income growth is an increasing function of the quintile. All quintiles displayed growth rates close to 50 percent over the period 1978-88. During the next 7 years, the highest quintile's average income grew by more than 30 percent, while for the other quintiles, this increase was close to 20 percent. Income in the bottom two quintiles grew faster than in quintiles 3 and 4, and thus the increase in inequality over this period was sharper in the central segments of the distribution. Again, we show that over the last period, on the other hand, quintiles 1, 2 and 3 experienced a sharp real drop in their average income, while households in the higher quintiles faced a much slower decrease, indicating an increase in income inequality. 28Urrutia (1984), Reyes (1987), Ocampo (1992), and Londofio (1995). Declining wage differentials of the 1970s fostered significant improvemer.ts in inequality (Misi6n de Empleo, 1986). According to Ocampo et al. (1998) the determinants behind these developments were: 1) the reduction of the rural labor force surplus, due to fast migration in the 1950s; 2) the robust pace of capital accumulation and modernization in the rural sector; and 3) the larger and well targeted investment in education and health delivered through the "Frente Nacional." On the progressivity of public social expenditure, see Selowsky (1976) and Vdlez (1996). 19 Under almost any measure, income inequality has been deteriorating during the last two decades. Inequality changes did not exactly follow the same patterns as average income per capita. The Gini coefficient steadily increased over the entire period, with acceleration in the latest 1995-99 sub-period. This has been driven in part by the increasing income accruing to the wealthiest segments of the population and the simultaneous decreasing share for the poor. This pattern of inequality dynamics is also shown by the stability of the P90/P10 and P751P25 ratios until 1995 -around 8.4 and 2.9 respectively- before they shot up to 12.0 and 3.5 in 1999. All three entropy measures also increased continuously until 1995. Thereafter, Eo continued to increase (from 0.48 to 0.54), confirming once more the increasing inequality among the poor. Tlhe increase in inequality - measured by the GiDi coefricient- was mainly concentrated in the 1988-95 period. The Gini coefficient increased by almost three points during this period, capturing the sharp inequality rise within the lower tail of the distribution. The Gini coefficient is very sensitive to changes near the mode of the distribution, at C$ 40,000 in 1995, on roughly the 15th percentile of the distribution. Thus, inequality increases in the lowest 15 percent of the distribution are captured by the Gini. Other measures of inequality reveal a similar deterioration of the income distribution during the 1995-99 period. In particular, the differential between the 90h and 10" percentiles grew by more than 40 percent. In addition, inequality between the 25th and 75h percentiles of the distribution had been remarkably constant over the 1978-95 period, before deteriorating sharply in the latest period. Finally, and this indicator portrays the magnitude of the concentration over these 20 years, the ratio of the share of income held by the richest quintile to that held by the poorest increased from 11.3 in 1978 to 17.2 in 1999. Section 5 will examine the mechanisms underlying these changes in inequality over the past two decades. Regional comparisons Throughout the entire period, inequality rose sharply in all cities, but Manizales. Various measures of household per capita income inequality for each city are presented in Table A.2. As mentioned earlier, the worst increase in inequality overall occurred between 1988 and 1995. This was driven by the worsening inequality situation in Bogota and Medellin. Other cities followed slightly different paths, with the sharpest increases in inequality occurring between 1978 and 1988 in Barranquilla, Bucaramanga, Manizales and Pasto. In Cali, the steepest rise took place in the 1995-99 interval. In contrast, Bucaramanga, and Manizales did not endure inequality growth during 1995-99. Although the dynamics of inequality are heterogeneous across cities, a pattern of convergence at a higher level is apparent over these 21 years. First, Barranquilla lost its most equal city place in 1978 (Gini index of 36.5) to Bucaramanga (Gini 44.5) in 1999. Second, in Manizales, the Gini index has been declining since 1988 (from 52.8 to 49.0), although it remained 1.1 percent higher than in 1978, when it reached 47.9. The remaining cities can be divided among those with two-digit inequality growth, and those with one-digit increases. Among the first group, apart from Barranquilla, we find Bogota (21 percent growth), Cali (11.7 percent growth), and Bucaramanga (10.6 percent growth). Within the second group of moderate inequality growth, apart from Manizales, fall Medellin (4 percent growth) and Pasto (8 percent growth). Thus, if the most unequal city in 1978 was Medellin, this "distinction" now belongs to BogotA in 1999. The ratio of the Gini indices of the most to least unequal city fell from 1.32 in 1978 to 1.05 in 1999. Hence, regional convergence indeed took place, but at higher levels of inequality. Paglin-Gini Index: After adjusting for differentials along the lifecycle, inequality trends look flatter The higher the income differences along the Difecycle the lower the Pagin-Gini index of income inequality. Even in the ideal case of a society characterized by complete lifetime equality, some 20 degree of income inequality will always arise because of the evolution of individual earnings along the lifecycle. For example, mature individuals (i.e., those in the latter portion of adulthood) usually figure within the richer segment of the population. If in addition to higher income levels, the elderly keep gaining demographic weight, total inequality will increase even if within cohort inequality is absolutely unchanged. To disentangle this "age effect" from pure income inequality, we use the methodology proposed by Paglin (1975), and compute the "Paglin Gini", free from this unavoidable component of inequality2 . Both indices are plotted in Figure 1.6. Figure 1.6. Gini and Paglin-Gini coefficients, Urban Colombia, 1978,1988,1995,1999 0.56 054 - 0.52 0 50 - 0.48 0.40~~~~~~~ - GNI _PAGUN GINI | 0.46- I GIN 044 042 040 0.38 1978 1988 1995 1999 Year Once rid of its lifecycle component, income inequality follows a much flaner curve than the standard Gini, except during the first sub-period 1978-88. The modest increase of the "Paglin Gini" points to the presence of increasing inequality associated with lifecycle effects, potentially because of higher earnings of the households headed by old individuals, ownership of real assets, smaller number of dependants.30 The gap between the two curves has widened since 1988, showing that the inequality increase may be associated in part with an increased dispersion of individual earnings along the lifecycle, in addition to changing returns to skills and other individual endowments. The unavoidable component of total inequality was at its lowest in 1988, at 12.4 percent but rose to 19.9 percent in 1999. Average income corrected by inequality: the Sen Welfare Index The rising trend of inequality reduced the potential welfare gains up to 1995 and aggravated the welfare losses during the economic recession. In order to adjust mean income per capita to 29 Paglin's argument is basically that the equality line against which the Lorenz curve is compared, is unattainable in reality and that it should thus be relaxed to account for the presence of inequality arising from the lifecycle. Once this is taken into account, the equality line turns into a curve, reducing the area between the Lorenz curve and the new equality curve. The area between the old equality line and the new equality curve is the inequality associated with .the lifecycle, or the "unavoidable" inequality. See Paglin (1975) and Champemowne and Cowell (1998), p. 80. 30 Morley, Robinson and Harris (1998) estimate transition matrixes for Colombia during the 1990s and find large degree of upward mobility during the lifecycle. 21 account for the changes in welfare due to income inequality, we use the Sen welfare index.3' Figure 1.7 depicts the evolution of the mean household per capita income along with the Sen welfare index. Average income increased by 88 percent between 1978 and 1995, and leveled off, if not decreased, during the second half of the 1990s. But with inequality increasing, the gap between average income and adjusted welfare has widened overtime. Welfare always increased at a slower rate than mean income 32, but in the latest period, it decreased by 2.6 percent, returning to levels similar to those of 1988, while income lost a yearly 1.4 percent. Figure 1.7. Average income per capita and the Sen welfare index, Urban Colombia, 1978, 1988, 1995, 1999 280,000- 230,000 20,00 Mean household 8 / ~~~~~~~~~~~~~Income per capita CL 0 San welfare index oo It80,000 - 130000 -w 130,000 ., 80,000 , 1975 1988 1995 1999 Year Hliad inequality remained constant, welfare would have increased with income: the increase in inequality was responsible for an 18 percent loss in welfare gains between 1978 and 1995, and an additional 5 percent loss in the last sub-period. Finally, to appreciate the magnitude of the inequality issue, the ratio of Sen index growth to mean income growth was 93 percent in 1978-88, 65 percent in 1995-98 but 177 percent in the latest time interval. Thus, during these last four years, the increase in inequality aggravated the welfare loss due to lack of growth by 77 percent. Regional comparisons Average income corrected by inequality shows a stronger pattern of convergence in social welfare. Figure 1.8 displays Sen welfare index deviations from the urban average for all seven cities.33 The adjustment by inequality modifies the distribution of social welfare across cities. Although the spreads of social welfare and incomes across cities are similar in 1978 -55 percentage points- the range of social welfare is much smaller in 1999 -23 percentage points-. The general pattern of large gains in income and welfare over the 1978-88 period, followed thereafter by slower welfare improvements, is generally present in all cities' profiles. 3' This index is defined as S=u(l-G), where u is sample mean income and G is the Gini coefficient. Sen (1973) elegantly shows that this index is an appropriate measure of welfare for a very wide class of welfare functions. In addition, it is a necessary condition for the stochastic dominance result using generalized Lorenz Curves. 32 Growth rates for mean income and the Sen welfare index were respectively, 5.0 and 4.6 percent in 1978-88, 3.6 and 2.2 percent in 1988-95. 33 After adjustments for differences in local cost of living. 22 Figure 1.8. Deviation from average urban corrected Sen welfare index, Urban Colombia 1978,1988,1995, 1999 40% 30% - - ' * 20% c ~10%- -30% ID -1 0% _ _ _ 0. -40% 1978 1988 1995 1999 Year --Barranqulla - a -Bucararnanga - BogotA W Manizales -( *Medellin _Cal Pasto Bucaramanga, Pasto and Manizales appear again as the main losers, but the magnitude of their income loss is attenuated by inequality adjustments. In Pasto and Manizales, serious inequality worsening aggravated the income loss of the 1978-88 period. On the other hand, after 1988, they fare better inequality-wise and are able to regain some welfare. For Bucaramanga, the fall in welfare in the last period was tempered by its relatively equal distribution of income. The relative position of all three cities is adjusted upwards and get closer to Bogota's level of welfare. The relative advantage of Bogota is adjusted downwards another 5 percentage points. However, the income loss is compensated by the lesser increase in inequality, which means that Bogota preserves it 11 percent advantage. In 1999, Bogota maintains the highest welfare index of all cities, yet Pasto, Manizales and Bucaramanga follow closely. The relative situation of Medellin is somewhat improved by inequality adjustinents. Total deviation losses from 1978 to 1999 become almost negligible (2.8 percent); however, they are quite substantial in the 1995-99 period (14 percent). Both income and inequality had improved in the 1978-88 decade. While the income losses started thereafter, the adjustments for inequality maintained the relative ranking of Medellin until 1995. However, the fall experienced during the recessive period sent Medellin to the lowest rank of all cities, after Barranquilla. Despite higher than average increase in inequality, Barranquilla obtained the largest gains in terms of income and welfare gains among the seven cities over the period 1978-99. Income gains were truly spectacular in Barranquilla, with the deviation from the average going from minus 43 to minus 15 percent, a 28 percent increase. Since inequality also rose steeply, the resulting increase in welfare was only 23 percent. Between 1988 and 1995, the growth in income was minor but the reduction in inequality improved the overall welfare. After 1995, income continued to grow but inequality worsened leading to a 4 percent loss in welfare. Despite their relative losses in terms of adjusted welfare (Sen index), in 1999 the three smallest cities, Pasto, Bucaramanga, and Manizales, stood close to Bogota and ranked far better than the three other "grandes ciudades" -Barranquilla, Cali and Medellin. This came through very different trajectories. In 1978, Bogota and Cali stood above average, but still far behind the three 23 smallest cities. This situation radically changed in 1988, when welfare in Manizales and Pasto fell below average, Cali and Medellin exhibited just above the average welfare levels. Bucaramanga remained the clear leader. In 1995, Pasto and Manizales regained some of the lost ground while Bucaramanga continued strengthening its relative position, at 28 percent above average. In 1999, this relative advantage decreased; Bucaramanga rejoined Manizales and Pasto, this time with BogotA. 3.4. Welfare comparisons for any distributional weights: First Order and Generalized Lorenz Dominance Until 1995, urban welfare levels imparoved substantially. As shown in Figure 1.9A, welfare levels in urban Colombia improved continuously and unambiguously for every percentile of the population between 1978 and 1995. The proportion of the population below the 1999 poverty line (at C$175,831) decreased from over 70 percent to approximately 50 percent during that interval.34 During the period 1978-88, improvements in earnings reached 47 percent for nearly all percentiles, except at both extremes of the distribution, where the improvements were slightly higher. In the next period 1988-95, earnings kept increasing by 17 percent for nearly the entire population. The result is that between 1978 and 1995, the 5h and 10h lowest percentiles saw their income increase by 75 percent while the 900 and 95h experienced 75 and 91 percent increases respectively. Figuare 1.9A. Cumulative income distribution, Urban Colombia 1978, 1988, 1995. 1 0.9 . o.~~~ 1995 Poverty > 0 line: - . 0.7 - 0 0.6 - 9X . 0.4 02 U o0.1 0 --1 0 50,000 100,000 150,000 200,000 250,000 300,000 350,000 Per capita monthly household Income, 1999 pesos 34The 1999 PPP adjusted exchange rate is US$1 = C$629. 24 Figure 1.91B. Cumulative income distribution, Urban Colombia 1988,1995, 1999. o 0.8 1999 Povcrty =oO 0°7 line: $175,831 0. - 0 0. 0.6- O 500 000 15000 20O200 000300 c 0.5 0.4- 0.3 -1999 0.2 2 0.1 0 0 50000 100000 150000 2000 250000 300000 350000 Per capita monthly household Income, 1999 pesos Figure 1.10A. Generalized Lorenz curves, Urban Colombia, 1995, 1999. 350,000 -S 300,000- 2 50,000- 0 E g! ~~-19991 9 200,000 - E E --1995 ? 150,000 100,000 50,000) O 0 5 10 15 20 25 3035 40 45 50 5560 65 7075 80 8590 95100 Cumulative percentage of population 25 Figure 1.10B. Generalized Lorenz curves, Urban Colombia, 1988, 1999. 300,000 - L_ I 5,0 2 250,000- E I~~~~~~~~~~~~~~~~~~~~~~~~~~ 200,000- h E 50,00018 o O0,0 -, , , . . . 0 0 5 101520253035404550556065707580859095100 Cumulative percentage of population In 1999, urban welfare unambiguously deteriorated with respect to 1995. The 1999 data present a starkly different picture; Figure 1.9B shows the cumulative distribution function (CDF) for years 1988, 1995, 1999. The cumulative distribution for 1999 lies to the left of the curve for 1995 for all monthly per capita household incomes below C$ 350,000. The poor lost more than the rich: the 5h and 10' percentile both lost 25 percent of income, while median income decreased by 11 percent; yet the 90th percentile income rose by 9 percent and the 95h gained one percent. Since first-order dominance does not hold, we tum to the generalized Lorenz curves of those two years in Figure 1.10A, which provide additional information on second-order stochastic dominance. Between 1995 and 1999, welfare again unambiguously deteriorated. Welfare losses due to economic recession are so massive that the welfare distribution of 1999 is not unambiguously superior to 1988. The cumulative distribution functions for 1988 and 1999, shown in Figure 1.9B, cut around the 40"' percentile. The generalized Lorenz curves, displayed in Figure 1. lOB, also intersect around the 60' percentile. This implies that social welfare measures with substantial weights on the poor tend to favor the 1988 distribution, and corresponds to the comparison of poverty measures in Table 1.4: the poverty count was very similar in both periods (55.0 percent in 1988 and 55.4 percent in 1999), but both the poverty gap (at 26 percent) and the P2 index (at 15.4 percent) were higher in 1999. Yet although the proportion of the population below the poverty line is increasing (along with poverty depth), the social welfare measures with more uniform distributional weights tend to favor the 1999 distribution when compared to 1988. For example, the average income per capita remains higher in 1999 at C$ 277,469 (C$ 235,163 in 1988), even when undertaking the Sen welfare index's correction for inequality, as shown in Figure 1.7. ln summary, are urban Colombians better off now than in 1978? They are, although not unambiguously better than in 1995 or even 1988. After continuous and significant reduction of poverty from 1978 to 1995, economic recession pushed poverty indicators back to 1988 levels. The extreme poor suffered the brunt, having had a good share of the substantial gains of the 1978-95 period erased and having underwent a deepening in the intensity of poverty in 1999. Income inequality had a rising trend over the whole period of analysis. The increase in inequality -measured by the Gini coefficient- was mainly concentrated in the 1988-95 period. This inequality trend reduced potential welfare gains from 1978 to 1995 and aggravated the welfare losses during the economic recession. However, once rid of its lifecycle component, income inequality follows a much 26 flatter curve -especially after 1988. Table 1.6 sums up this section's findings on alternative comparisons of social welfare in urban Colombia. Under any distributional weights, social welfare in 1995 urban Colombia was better than in any previous year -income per capita doubled with respect to 1978's level and all percentiles of the income distribution made comparable gains. However, a similar claim cannot be made about 1999. Although this year is unamnbiguously better than 1978, when social weights concentrate on the poor, social welfare in 1999 is clearly dominated by 1995, and even by 1988. Only when the welfare index is restricted to per capita income (adjusted or not for inequafiiy), does 1999 fare a bit more favorably than 1988 -the richer households are better off but the poorest 56 percent saw their welfare decrease in the 1988-99 intbrval. Economic recession sharply reversed~the improving welfare dynamics, observable until 1995. Table 1.6, Welfare Comparisons wuder alternative sodid welfare critera, Urban Colombia, 1978,1988, 199S and 1999 Social Welfare in year (colum) i 1978 1988 1995 1999 l~~~~~~~~~~~ - ! is greater than in Year (row) 1978 :l IiTIL t . . 1 . i-r 1 v) 1995 una nbigu ti-Jy unaxib'il. mu:m i Now:I. F.OD.: RM Oda Stdcanc 4munhtc, 2 OLD.: euai. z Lcu domin. _ . 3 YPCI_ A lncom per ca*1a ed. 4. povuty co=n P(O), poat eM F (1) and poveny intnmity P(2). F?inally, whichi Colombian cdty enjoys the highest welfare right now? The answer depends on the choice lot socal weights: by average Income, It is BogotA, but by poverty count, it I Pasto and by extreme poverty, it is Bucaramanga. Across cities, there is a pattern of convergence in average incomie per capita, with the exceptioni of Bogota, which got further ahead during the last two decades. In dynamic terms, Barrnquilla and Bucaananga underwent the largest gains and Manizales experienced the most losses. However, after adjusting averge income by differentials in local cost of living and inequality, welfare convergence is even stronger. BogotA loses all its relative advantage and in 1999, it bunches together with the three smallest cities of the sample, Pasto, Bucaranianga, and Manizales. They rnk far better than the three other "grandes ciudades- Barranquilch Cali and Medellfn. As in the previous caels, ranking cities by poverty also reveals a patten of convergence, but with quite different rankings. Despite their low average income per capita in 1999, the cities with the lowest poverty rates are Bucarananga and Pasto, and the poorest Baranquilla and Medellfn -with a difference of 10 perentage points. Nevertheless, Baanquilsa 27 was the city with the largest gains in poverty reduction. Moreover, city rankings are very sensitive to the poverty line used. If ranked by extreme poverty, Pasto becomes the poorest city, followed by Barranquilla, and surprisingly by Bogota. Bucaramanga maintains its ranking as the least poor city, followed by Cali and Medellfn. 4. POVERTY PROFILE: FACES OF THE POOR REMAIN THE SAME OVER TIM[E, BUT ARE BECOMING MORE OSTENSIBLE 4.1. Basic factors of income per capita generation: the poor versus the non-poor Basic factors of income per capita generation. Total household income is equivalent to the sum of labor and non-labor income received by its members. Consider a typical household h, with Nh members and Ah of them in working age, among whom Th individuals with Sh equivalent skill units are employed in the labor market for an average household wage per skill unit of Wh. The household labor income per capita can be written as the following identity:35 AhAN (1) Equation (1) describes the household labor income per capita as a product of average wage per skill unit, the household's average skills, the rate of employment of adult population (TIA) and proportion of working-age members (A/N). Therefore, if we ignore property income and transfers, one or all of the three following features should characterize poor households: low skill endowments, low employment rates and high children to adult ratios. The data show that poor households simultaneously suffer from lower skill endowments and higher dependency ratios. Table 1.7 compares poor and non-poor households. Several worrisome but not completely unexpected trends are highlighted in this table. First, skill endowments are lower in poor households. College graduate earners belong disproportionately to non-poor households: only 3.3 percent of poor households had at least one such member in 1999 while more than a third of non- poor households counted at least one. While schooling levels of household heads have increased, the poor only gained 1.8 additional years, while the non-poor gained 2.3 years, widening the gap. Non- poor households had 50 percent more years of schooling than the poor in 1999, and that ratio dropped by only 10 percent percentage points during the last 21 years. Since unemployment is lower the higher the schooling level, this indicator reflects the role of higher education in protecting access to jobs and ultimately, household income. Second, in 1999, household size and dependency ratios remained approximately 1.3 and 2.3 times as high in poor households, changing little during the last two decades. 35 We borrow this identity from Paes de Barros et aL for UNDP (2000). 28 Table 1.7. Income sources and needs: the poor versus the non-poor. Urban Colombia, 1978-1999 1978 1988 1995 1999 Poor Non-poor Poor Non-poor Poor Non-poor Poor Non-poor Needs Average number of people in the household. 5.5 4.3 4.9 3.9 4.7 3.7 4.5 3.4 Average number of children 12 yrs. & under 2.1 0.9 1.7 0.7 1.6 0.7 1.4 0.6 Educational endowment. Schooling, head of the household 4.7 8.2 5.6 9.1 6.2 9.2 6.5 10.5 Schooling, individuals older than 18yrs. 5.1 8.1 6.1 9.2 6.6 9.6 7.1 10.6 Households with at least I college graduate. 0.9% 17.8% 1.9% 24.5% 2.0% 25.5% 3.3% 34.7% Household employment and child labor Household head 83.7% 80.7% 80.3% 78.3% 81.3% 77.3% 74.5% 73.4% Other adults 17 & over 56.8% 68.0% 55.3% 68.8% 59.1% 71.1% 56.2% 68.9% Children 12-16 in labor force 11.0% 15.2% 11.2% 12.1% 9.8% 10.2% 10.4% 7.8% Social mobility School enrollment Ages 7 to II 90.8% 96.4% 93.3% 98.2% 95.3% 98.6% 93.9% 98.3% Ages 12 to 17 76.3% 78.5% 78.7% 83.7% 83.1% 86.1% 79.6% 87.1% Ages 18 to 22 25.9% 40.4% 26.7% 44.2% 30.6% 48.1% 24.9% 48.6% Poor households face lower overall employment rates, with lower employment of spouses and other household members of working age; however, employment of their heads does not explain the difference. Table 1.7 shows that household heads are more often employed than other household members. And the comparison between poor and non-poor households reveals that the proportion of poor heads employed is almost equal or only marginally above the non-poor's: from 2 to 4 percentage points during booms and down to 1 percent during the recessive year 1999. Employment of other household members (spouse and other adults) is consistently lower among poor households, by 11 to 13 percentage points over the 1978-99 period. Within non-poor households, the difference in employment rates between heads and other adult members is narrowing at a faster pace than within poor households. In 1978, the gap between heads and other adults was 12.7 percent in non-poor households. In 1999, it had fallen to a mere 4.5 percent. For the poor, the corresponding figures were 26.9 percent in 1978 and 18.3 percent in 1999. The gap in both groups closed by over 8 percent, but the poor seem always less successful at securing employment for other household adults than the non-poor. It is worth noting that, until the recession hit, the proportion of working children (from ages 12 to 16) was higher among non-poor households. This may suggest that some households had managed to escape poverty by raising labor force participation of all potential workers within the household. However, this strategy became apparently ineffective during the recession years. The continued lower school enrollment rates of the poor foretell mediocre prospects for the next generation's transition out of poverty. The differences in school enrollment rates between poor and non-poor households did marginally diminish until 1995, but the economic recession exacerbated 29 those differences back to maximum levels. As shown in Table 1.7, in 1995, when the gaps decreased most, the proportion of children of poor households aged 7 to 11, 12 to 17 and 18 to 22 in school was inferior to that of the non-poor by 3, 3 and 17 percentage points respectively. In 1999, those differentials jumped back up and reached 8 and 24 percentage points for the second and third age groups. Among young adults, the enrollment rate of the poor is only half the rate prevalent in non- poor households. This is a serious cause of concern since at that age, students usually complete their high school and start college, two levels of schooling that are key to escape poverty, as we will see below. The lower enrollment rates of poor students at the high school and college levels will weigh heavily on social mobility and the inter-generational transmission of poverty. In the regional context, social mobility (in education) is relatively low and shows a strong gender bias. According to DNP-UNDP (1999) and Florez et al. (1999) upward and downward social mobility -in education- in Colombia is relatively high, much better for females than for males. Despite improvements during the last decade, in the Latin American context Colombian urban social mobility remains behind Chile and Costa Rica but ahead of Brazil.36 Only in rural areas and for women rankings Colombia is ahead of Costa Rica. Moreover, when compared with the same set of countries, the mobility gap of Colombian male youngsters with respect to females is the second highest -only behind Brazil-. Social immobility is much higher at the extremes of the skill scale; therefore for extremely poor households, there seems to be a poverty trap associated with extremely low social mobility. Florez et al. (1999) show that in Bogotf in 1997, only 10 percent of the individuals in the first decile had the chance of attaining a school grade level above their parents' level (!). From basic income generating factors to the faces of the poor The factors in equation (1) characterize labor income per capita generation, but do not reveal the faces of the poor. In fact, their value is endogenously determined for each household by the interaction of the household's specific characteristics with the local socio-economic environment it faces. For example, the wage per equivalent skill unit W, depends on occupational choice -wage earner or self employed-, sector of economic activity, regional job market, etc. The average endowment of equivalent skill units Sh -human capital endowments- responds to the skill premium variation over time. The number of adults actually employed Th depends on national and regional unemployment rates, skill endowments and experience - especially for spouses-, number and age of children, and presence of other adults and their relative participation in the labor force or within the household productive and reproductive tasks. As, the number of adults, i.e., working age members, is a demographic characteristic, and the proportion of adults AAV is inversely related to the dependency ratio, defined as the number of children per working age individual (d = N-A /A). In the next sub-section, we use the previous framework to investigate the characteristics that identify households more vulnerable to poverty. 4.2. Faces of the poor: groups at risk and household characteristics Household characteristics and the risk of poverty. We now turn to the identification of groups vulnerable to poverty at each period. We approach this task by computing simple univariate relative risks of poverty for different sub-groups of the population. The results are presented in Table 1.8.7 The information in the table enables us: (i) to evaluate which variables explain most of the variability 3 CEPAL (1997) quoted by DNP-UNDP (1999) tables 70 and 71. See also Birchenall (1998) and Morley (2000) on Colombia. As one would expect, the U.S. shows higher social mobility than Colombia. 37 Each cell in this table is computed as risk = (Pg - PaU1 )/ Padi , where Pg is the poverty rate for the group in question and Pd, is the overall poverty rate. Thus, positive cells in the table refer to groups more likely than average to be poor, while the opposite is true for negative entries. 30 of the observed poverty risks among the different categories, and (ii) to assess the evolution of relative poverty likelihood for different population sub-groups . Throughout the entire period, the variables with the strongest relation to poverty were household size (Nh), the household employment rate (TI/Ah) and the skill endowment (Sh). Proxies for skill endowment include: head's schooling and average schooling attainment of all other household members older than 12. Occupation and sector of economic activity of the household head and his/her age, which capture some components of Ws, also proved to be somewhat important correlates of poverty rates. Consistent with the framework presented in equation (1), the groups of households which face a higher relative risk of poverty -above 20 percent- are (i) large households (Nh)--more than six members--, (ii) households with low adult employment rates T,/Ah -- less than 20 percent of the members older than 12 years employed--, (iii) attachment to low wage sectors (Wh) - for example, head employed in construction-, (iv) head employed as a blue-collar worker, or a domestic employee rWh), (v) low skills households (Sh) -- head high school dropout or less, and less than 6 years of average schooling for other members in working age--, (vii) other demographic characteristics: households with heads younger than 29 years of age, and (viii) households with the head living in consensual union. Marriage rates are lower among the poor than the non-poor. If those unions are less stable, divorces and/or separations will significantly affect the household's income-generating and wealth accumulation capacity. It is surprising that, apparently, households in which the head is inactive are less at risk of falling into poverty than the average household. However, this is more than compensated by employment of other household members. On the other hand, female headship does not appear to be associated with higher risks of poverty.35 The faces of the poor are not changing, although they are becoming more ostensible. Table A 1.7 in the appendix describes the change in risks for relatively protected and vulnerable households, throughout the entire period. Vulnerable households remain persistently so, from 1978 until 1999, except for households with a high school graduate head, whose relative risk increased by 29 percentage points over the entire period. Similarly, groups, who remained relatively protected, also started with lower relative risks. These observations lead us to conclude that poverty has not "changed faces" in urban Colombia in the last 2 decades. On the contrary, the groups most "at risk" have become increasingly vulnerable to poverty, while those who started out in safer positions have maintained or improved their situation. To fend off poverty, high school education does not suffice any longer; some college education and more labor force participation are becoming essential. While still in a somewhat safer position, households with high school graduate heads or other members have seen their relative advantage erode over the past 20 years. Their risk of poverty increased 28 percent percentage points from 1978 to 1999. On the other hand, as seen in Table 1.8, some college education and increased labor force participation -above 40 percent- within the household have become almost necessary "assets" to protect the household against poverty.39 Households with six members -or more- are becoming increasingly vulnerable. Large household size (Nh), as identified by equation (1), is an important risk factor. Households including six members or more have seen their relative risk multiply by a factor of 6 in the past two decades. This group comprised approximately 27 percent of the population in 1999. 38 However, we will show below that after controlling for other correlates of poverty, female headship increases the probability of poverty. 39 In -the framework of equation (1)- this corresponds to compensatory expansions in both skill endowments Sh and the employment ratio Th/Ah - 31 Table 1.8. Relative risk of being poor, Urban Colombia, 1978-99. Relative nsk Share of total 1978 1988 1995 1999 populaion In 1999 Household size one person -83% -76% -78% -76% 2% 2 to 5 persons -6% -18% -10% -8% 69% 6 to IO persons 4% 14% 22% 22% 27% Household employment' zero -1% 13% 5% 55% 8% O.75 -31% -40% -39% -37% 18% Head's sector of econonic activity Manufactunrng 0% 2% 2% 5% 14% Unilies 25% -15% -29% -42% 1% Construction 19% 22% 30% 32% 5% Comnmerce -5% 3% 5% 9% 18% Transportaton 8% 4% 0% 4% 9% Banling, insurance -21% -35% -22% -25% 7% Other services -1% 0% -1% -10% 19% Inacuve 0% -2% -8% 0% 25% Head's occupation Blue collar 24% 26% 27% 27% 16% White collar -20% -22% -18% -25% 24% Domestc 43% 46% 78% 27% 1% Self 0% 13% 10% 16% 28% Employer -47% -54% -58% -56% 5% Inacuve 0% -2% -8% 0% 25% Head's education' Uneducated 19% 46% 43% 54% 4% Pnmary 16% 26% 26% 29% 35% Hligh school dropout -3% 7% 17% 24% 23% Htgh school -46% -37% -26% -17% 20% College dropout -52% -68% -65% -59% 6% College graduate -85% -89% -90% -85% 12% Educationaverage household members older than 12 yrs Pnmary 24% 43% 53% 49% 17% High school dropout -17% -9% -34% 19% 54% High school -57% -61% -85% -39% 11% College dropout -79% -81% -95% -81% 15% College graduate -100% -93% -100% -89% 3% Head's age Betwveen20and29yrs 12% 14% 13% 23% 11% Between 30 and 39 yrs 5% 13% 20% 16% 28% Between 40 and 49 yrs. 3% -2% -1% 2% 26% Between 50 and 59 yrs. -9% -13% -16% -16% 18% Between 60 and 69 yrs -13% -17% -25%b -26% 11% Between 70 and 79 yms -25% -17% -18% -26% 5% Head's Mantal status Consensual 21% 43% 40% 33% 26% Married -2% -9% -15% -16% 46% Widower -11% -11% -15% -6% 9% Divorced 6% 9% 14% 5% 12% Single -14% -38% -31% -24% 7% Head's gender Male 0% 1% 0% 0% 76% Female -1% -2% 1% -1% 24% I e=# of employed/# of persons older than 12 years of age 2 Uneducated O years of education. primary 1-5 years of education high school dropout 6-10 years of educadon high school - 11 years of educaton college dropout 12-15 years of educadon college graduate 16 or more years of education 32 Facts versus common beliefs on vulnerability: Children of all ages, recent migrants and non- homeowners are clearly worse off than pensioners, the elderly and non-recent migrants. To complete the features of the faces of the poor, we now compute the incidence of poverty in some population groups, traditionally considered more vulnerable to poverty: migrants, women, children, non-homeowners, pensioners, the elderly, and the disabled. In Table 1.9, we report poverty rates within these groups, along with the overall urban poverty rate for comparison purposes. Popular beliefs are confirmed in relation to children of all ages and non-homeowners. Children under 18 consistently present higher poverty rates than the entire population, with an increasing proportion of infants and preschoolers in poverty. Homeownership clearly provides protection against poverty. Migrants used to fare better or similar to the overall population until 1995, but recent migrants are more likely to fall into poverty in the recession years. This must be partially associated with a shift from voluntary to forced migration -"internally displaced peoples "- due to the armed conflict in rural areas. On the other hand, incidence rates are nearly identical among men and women over the entire period. The disabled remain consistently poorer than the rest of the urban population since 1988, although incidence rates are decreasing for this group. Pensioners and the elderly do far better than the rest of the population and their relative standing seems to be improving over time. Tablel.9. Poverty count for different population sub-groups, Urban Colombia 1978-1999. Population 1978 1988 1995 1999 share in 1999 Urban Colombia 70% 55% 48% 55% 100% Households with unemployed heads 91% 80% 72% 78% 8% Children under 2 yrs. 80% 71% 63% 72% 4% From 2 to 6 yrs 81% 70% 63% 69% 10% From 7 to 13 yrs 80% 70% 62% 69% 13% From 14 to 17 yrs. 73% 61% 55% 64% 8% Migrants/just moved' NA 50% 50% 64% 1 % Non-homeowners 77% 64% 57% 63% 51% Disabled 69% 68% 60% 60% 1 % Migrants<5%2 NA 51% 46% 60% 1% Women 69% 55% 48% 55% 53% Migrants<10%3 NA 49% 43% 54% 2% Homeowners 62% 46% 40% 47% 49% Migrants<25%4 NA 50% 42% 44% 4% Over 65 years old 52% 42% 35% 37% 5% Pensioners 37% 32% 20% 24% 2% 1. Migrants/justmoved: refers to people who have lived less than 1% of their lives in the current city. 2. Migrants <5%: refers to people who have lived less than 5% of their lives in the current city. 3. Migrants <10%: refers to people who have lived less than 10% of their lives in the current city. 4. Migrants <25%: refers to people who have lived less than 25% of their lives in the current city. 4.3. The poverty proffle and the marginal effect of key household characteristics Poverty profile. Our measures of poverty indices and relative risks for different demographic groups allow for a clear picture of the faces of poverty, but they do not help us determine the characteristics 33 that cause a household to be poor. For instance, the data show that female household heads are less poor than the average population, but perhaps households headed by women tend to have other characteristics which protect them from poverty such that if we control for these characteristics, they no longer appear less vulnerable to poverty. Thus, we estimate the marginal effect of each key household characteristic on model the probability of being poor, including all variables that proved strong correlates of poverty in the previous descriptive statistics. This enables us to construct a poverty profile, based on a logit model (see Table A1.6 in the appendix). Our profile includes five main types of correlates: (i) skill endowment variables Sh (education and potential experience of the head, and average education of all other household members older than 12 years);40 (ii) demographic variables Nh (age, gender, migration and marital status of the head, number of members of working age4' Ah, number of dependents -children and elderly42); (iii) household labor market characteristics (Wh, Th), activity of the head, employment rate of other household members 12 and older, and a dummy variable indicating household members' employment as wage earners, self-employed or both and regional variables -6 dummies for the cities-); (iv) physical assets as proxied by homeownership; and finally (v) idiosyncratic shocks such as job loss, disability, divorce or widowhood. Skill endowment Human capital is key, but high school education is losing some of its protective power. Tables 1.10A-E present the effect of human capital on the probability of being poor. We proxy human capital with the educational level of the household head43, his/her potential experience44, and the average educational attainment of other household members older than 12. The table highlights various facts. First, as foreseen in the descriptive statistics, while a high school graduate head provided a significant advantage in 1988, this effect has been reduced dramatically over time, returning to the 1978 levels. This fact is surely related to the poor performance of the earnings of high school graduates and dropouts since 1988 (see the section on inequality). On the other hand, having a college-educated head is a considerable advantage. Second, the human capital endowment of the household -proxied by its average educational attainment of other members in working age- provides alternative protection against poverty. In 1995, a six-year raise in average schooling - equivalent to moving from complete primary to high school graduate- produced a 22 percentage points drop in the probability of poverty; however, after the recession this effect diminished to 14 percent. Third, while all educational categories are affected by the economic crisis, households with uneducated heads became increasingly vulnerable: after the relative alleviation brought by the construction boom of the first half of the 1990s, their probability of poverty doubled back above 1988 levels. 40 Potential experience is defined as min[(age-education-7), (age-12)]. 41 Ages 12 to 65. 42 The categories are: less than 7, 7-12, 12-65 and above 65 years of age. 43 The dummies are no schooling, primary school (1-5 years) - the base category --, high school dropout (6-10 years), high school graduate (11 years), college dropout (12-15 years) and college or more (16 or more years). 4 A variable capturing the interaction between education and gender of the head, but we do not report its marginal effect. 34 Table 1.10A. Marginal effects of human capital variables on the probability of being poor, Urban Colombia. 1978 1988 1995 1999 Head's education. ' Uneducated 0.5% 12.5% 7.5% 13.3% High school dropout -4.8% -15.1% -7.2% -5.6% High school -15.1% -32.4% -19.7% -17.9% College dropout - 15.5% -52.8% -38.8% -29.1% College graduate -30.4% -78.0% -65.1% -45.9% Head's experience.2 0.7% -0.4% 0.3% 0.2% Household's average education.3 -1.9% -5.5% -3.7% -2.3% i. Base category is heads with primary education. 2. Experience = min[(age-education-7), (age-12)] 3. Average taken over all household members other than the head who are older than 11 yrs. * Significant at 10% level or less ** Significant at 5% level or less In Tables 8 A-E, marginal effects are computed as [exp(x'ibeta)*beta* k] / [I+exp(x'ibeta)]2 Demographics The presence of children increases the risk of poverty, the elderly produce the opposite. Table 1.lOB displays the marginal effect of most of the demographic variables listed above. Households with one more young child than average face a 10 to 13 percent higher poverty risk in 1995 and again in 1999. The presence of people older than 65 years of age slightly decreases the probability of being poor -2.4 percent in 1995, 3.6 percent in 1999 -, probably through the effect of wealth accumulation and/or secure pension income, as shown in Tables 1.8A and 1.8B. The protective role of fixed income sources may be enhanced in the recessive period. Older working-age household heads face lower poverty risk and this lifecycle effect is becoming even stronger during recent years. Since 1988 poverty risk decreases with the age of the head and becomes minimal for the 58-67 year-old males. In 1999, these lifecycle effects become much stronger for male heads, and reach 21 percent for the youngest heads.45 Female heads face somewhat different risks: those younger than 47 years old already faced a 11 to 16 percent additional risk in 1978 and this shot up to 31 percent higher than 58-67-year-old males in 1995. These risks decreased somewhat in 1999 and older female heads were comparatively less at risk than their younger counterparts, while remaining more vulnerable than males of the same age. Female-headed households do consistently worse than male-headed ones. Contrary to the descriptive statistics, when controlling for other household characteristics, the multivariate analysis shows that female-headed households are more vulnerable than male-headed ones. In 1988 and 1995, all female-headed households were facing higher risks than their male counterparts and this risk had been increasing, nearly doubling from 18 to 31 percent for the youngest women. In 1999, the risks decreased, but still remained higher than 1978 risks. This evolution is presumably associated with the gains on gender wage differentials over the 1995-99 interval (see section on inequality). The fact that unconditional risk of poverty is smaller than conditional risk reveals some compensatory effect in terms of income generating factors (either larger education endowments, lower dependency ratio and/or larger employment ratio). 45 These findings are consistent with the evolution of the Paglin-Gini Index in section 3. 35 Table 1.I0B. Marginal effects off selected demographic variables on the probability of being poor, Urban Colombia Demographic Variables. 1978 1988 1995 1999 Head's marital status1 Consensual union 2.5% 7.3% 6.1% 4.8% Single 9.0% 4.0% -9.5% -0.5% Number of individuals by age category Younger than 7 yrs. 6.9% 23.1% 10.0% 10.8% From age 7 to 11 7.4% 22.2% 13.0% 10.9% From age 12 to 65 1.5% 5.5% 0.9% 2.8% Older than 65 -4.3% -2.3% -2.4% -3.6% Head's age and gender2 Males Younger than 28 yrs. 9.8% 10.3% 5.8% 21.4% From 28-35 yrs. 0.0% 4.4% 8.3% 17.3% From 36-42 yrs. -3.8% 2.0% 8.7% 12.8% From 43-47 yrs. -4.0% 0.9% 5.9% 11.1% From 48-57 yrs. -3.7% 2.8% 3.0% 6.3% Older than 67 -1.9% 3.3% 0.0% 7.6% Females Younger than 36 yrs. 15.5% 17.7% 30.6% 23.7% From 36-47 yrs. 11.2% 17.1% 23.5% 20.0% From 48-57 yrs. 4.4% 14.0% 25.4% 16.0% Older than 57 3.0% 8.4% 17.1% 12.7% 1. Married is the base category 2. Males 58 to 67 years old is the base category. * Significant at 10% level or less ** Significant at 5% level or less Labor Market When household members other than the head work, the household is much less likely to be poor. Table 1.1OC presents the marginal effects of some labor market indicators. The last row of the table confirms that a higher employment rate within the household -excluding the household head- remains the most effective insurance against poverty, with poverty risks decreasing by 13 percent for each additional household member employed.46 However, this effect diminished during the recession, probably because as households send more members to the job market, these obtain lower individual earnings.4' Simultaneous, household participation in both wage and self-employment sectors made positive a difference. Households whose active members are all self-employed have been harshly hit by the crisis, and face nearly 10 percent more poverty risk than those with all wage earners. Surprisingly, mixed labor market attachment (having both wage and self-employment income) provided more poverty protection than exclusive wage-earning -nearly 5 percent in 1995-, but the recession wiped out this effect and these households now face the same poverty risk as those whose working members 46 The average number of people 12 and older (excluding the household head) per household is 2, while the average employment rate is 40 percent. Thus, an additional person becoming employed raises the employment rate to 90 percent, which in turn, decreases poverty risk by 13 percent. 47 This phenomenon is confirmed by the data analyzed in Fiess et al. (2000). 36 are all wage-earners. The difference stems from the facts that self-employment income is on average falling and is more unequally distributed than wage income.'48 Meanwhile, households with only non- labor income have always fared better than any other category. In section V, below, we will carefully examine the impact of labor attachment type on household poverty dynamics. Table 1.10C. Marginal effects of selected labor market variables on the probability of being poor, Urban Colombia 1978 1988 1995 1999 Household labor market characteristics1 Both wage earners and self employed -2.2% -4.1% -4.9% 1.6% Only self employed 2.7% 3.9% -1.2% 9.8% Only non-labor income 2.3% -7.2% -15.1% -7.5% Employment rate2 -18.1% -60.5% -41.9% -27.2% Head unemployed first time 20.2% 11.6% 26.4% 3.7% Head unemployed but worked before 18.5% 54.8% 36.3% 25.0% Regional effects 3 Bucaramanga -15.1% -28.7% -36.4% -14.1% Bogota -5.6% -7.1 % -13.6% -0.1% Manizales -6.9% -13.0% -3.8% -7.5% Medellifn -8.1% -16.4% -12.4% -2.3% Cali -7.7% -12.0% -12.3% -6.5% Pasto -28.1% -15.4% -18.7% -9.2% 1. Households with only wage earners is the base category. 2. Employment rate = number of people employed other than the head l number of people 12 years old & older other 3. Barranquilla is the base category. * Significant at 10% level or less ** Significant at 5% level or less City risks diverged up to 1995, converged in 1999, with Bogot*i severely hit by the recession. Table 1. IOC displays marginal effects for the city of residence, using Barranquilla, the poorest city, as the reference point. Even after controlling for other characteristics, Barranquilla households persistently faced a higher risk of poverty. On the other hand, households in Bucaramanga were up to 36.4 percent less likely to be poor in 1995, but this difference has been starkly reduced to 14.1 percent in 1999. Surprisingly, over the past twenty years -except in 1995-, Bogota was always the city where the probability of being poor was second highest and in 1999, Bogotanos faced the same risk as Barranquilleros. Despite living in the richest city in the country, after controlling for household characteristics, households in Bogota seem to be worse off than nearly all other major cities during the recessive period. The crisis also produced a remarkable convergence of risks in different cities in 1999. Homeownership Homeownership provides good protection against poverty. In Table 1.10D, we present the marginal effects of different home tenancy status. Ownership of physical assets -like housing- seems to act as "insurance" against poverty, by providing some flexibility to the household in reacting to the unemployment episodes. Defacto occupants and users are the most vulnerable groups. Renters and mortgage payers are 13-15 percent more likely than owners to be poor but this risk has kept on 48 Table 18 shows a Gini of 0.46 for wage income and 0.57 for self-employment 37 decreasing, despite the crisis. Of concern is the face that, during the recessive period, homeownership has plummeted -by 10 percent-, back to its 1988 levels and, it affected mostly the second and third quintiles -poor or nearly poor households-. This drop creates a major policy challenge since homeownership is a major protective factor against poverty (see Box below). Table 11.1011D. Marginal effects of tenancy status on the probability of being poor, Urban Colombia Percent of population in Homeownership 1988 1995 1999 1999 Owners but paying mortgage 25% 21% 15% 9% Renters 31% 19% 13% 39% Usufruct 39% 28% 23% 5% De facto 50% 24% 26% 0.2% 1. Base category is homeowners. 2. In 1978, all non-owners were aggregated and the overall marginal effect was 5.1%. * Significant at 10% level or less ** Significant at 5% level or less 38 Box 1.1. The reversal of two decades of progressive homeownership Homeownership - a poverty prevention and self-insurance factor - in Urban Colombia had been increasing for all income groups since the late 1970s; however, the recent recession plus two other major market distortions and institutional changes reversed those rates back to 1988 levels. After increasing consistently until 1995, housing ownership rates plummeted by nearly 10 percentage points and fell significantly for all income groups, especially among the poor and nearly poor -second and third quintiles-. Figure B4.1. Index of building activity, Colombia Table B4.1. Urban homneownership by quintile.l990-2000 Change Is B - 1978 * 1988 1995 1999 1995-99 ^ Full ownership S Quintile 1 27%o 30o 37% 30% -6% 0 0 - Ouintile 2 45% 42% 46% 44% -2% 14 ° Quintile 3 46% 53% 55% 54% -1% 12 0 Quintile 4 52% 56% 60% 581% -2% C\ Ouinible 5 57%o 59% 63% 58% -5% 100 ° Average 48% 52% 49sh -3% e 8 0 -X With mortgage 9% 7°% 9O __ Totaf 45% 57% 600% 58% -1o/% X Source: DANE. Encuesta Nacional de Hogares, autors' calculations 40 inctudes both partlat and fult ownership 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 During the second half of the 1990s, the market for housing suffered three majors set backs. One is simply the economic recession that has reduced demand for housing and the other two have been through the credit market. First, via a temporary distortion of the credit market that pushed the interest rates well above long-run equilibrium and second via the restnction of supply of mortgage credit derived from the Colombian Constitutional Court rulings that ignored the potential adverse market response. As a sub-product of real exchange rate targeting derived from anti-inflationary policy objectives of the Central Bank, the real interest rate reached excessive levels in 1998 - of 20% (!), well above the long-run level for 1990s. Consequently, under fixed mortgage payments, liabilities of mortgage holders to the financial sectors increased at such a pace that in many cases they became "impagables" and the chances of maintaining homeownership were severely reduced for many homeowners. At some point, real interest rates were even above the reservation level for some homeowners who asked for cancellation of mortgage credits, but were prevented by the mortgage banks willing to hold on to these profitable but inefficient contracts. With the intention of finding a remedy to this problem, the Constitutional Court ruled that the government should pay excessive interest payments to mortgage holders -a transfer of at least I percent of GDP, finally funded by a specific tax on the financial sector according to Ley 945 of 1999-. In addition, the Court declared the UPAC mortgage framework law used since the early 70's invalid and ordered Congress to vote a new legislation following two main guidelines: the financial sector should give the lowest market rate to mortgage credit and financial cost of mortgage credit can not be capitalized. That meant cross subsidies from other forms of credit to housing and the elimination of flexible mortgage payments adjusted to lifecycle earning profiles. As a result of the uncertainty about the rules of the game in terms of mortgage credit and the mandated cross-subsidies, the availability of credit for housing dried up, and the valuable possibility of re-financing mortgage credit in the middle of a recessive period was severely limited. Both the high -above equilibrium- interest rates and the Constitutional Court intervention did not allow the market to operate efficiently and produced undesirable effects for homeownership and wealth transfers from households to mortgage banks. However, available evidence is insufficient to identify which of the two had the greatest impact. Although, interest rates have come down -thanks to policy changes in Central Bank policy- and a new law of mortgage credit is in place (Ley 945 of 1999), they have not produced any significant reactivation of building activity and ownership of housing keeps falling. * In summary, the poor got hit in two ways: diminished employment for low skilled workers in the construction sector and, lower equity and homeownership rates -a poverty-preventive and self-insurance factor-. 39 Idiosyncratic shocks UnempBoyed and disabled household heads were most affected by the recession. Table 1. I0E describes the effect of some shocks faced by the household, such as separation, widowhood, disability and unemployment, as well as some other labor market characteristics of the household head. Disability of the household head is a major risk factor of poverty, although the effect decreased sharply from 60.5 percent in 1988 to 17.7 percent in 1999. When controlling for other factors, households with single heads as a result of divorce, separation or widowhood do not seem to be more at risk than those with married heads. Unemployment of the household head is catastrophic for poverty risks, especially if the head used to work -25 percentage points more!-. As seen before, on the other hand, households with an inactive head are better off, especially if the non-labor income comes from rents. Since these earnings are somewhat less responsive to economic cycle swings than labor earnings, they act as a protection device to both economy-wide and idiosyncratic shocks in the labor market. Table 1.1OE. Marginal effects of exogenous shocks on the probability of being poor, Urban Colombia Exogenous shocks faced by the household 1978 1988 1995 1999 Head's marital status' Widower 2.5% 4.6% -7.6% -0.2% Separated / divorced 1.8% -0.3% -1.0% 0.3% Disabilities Disabled head 32.4% 60.5% 23.6% 17.7% Other disabled members -0.1% 3.3% 5.6% 2.5% Labor market variables for household head 2 Household work 12.2% 24.7% 5.5% 9.1% Rentier3 -6.1% -18.3% -22.3% -11.5% Pensioner -1.8% 3.5% -3.4% 0.6% 1. Status compared to married. 2. Compared to employed household heads. 3. Rent on property, dividends, etc. * Significant at 10% level or less ** Significant at 5% level or less 4.4. Unemployment by skill, age and region Unemployment peaks for the young with intermediate skills. Given the tight relationship between unemployment and poverty, we present unemployment rates for various demographic groups in Table 11. Recall from Table 1 that overall unemployment more than doubled between 1995 (8.7 percent) and 1999 (19.7 percent). In March 2000, it reached a historically maximum level of 20.2 percent in urban areas. College graduates face the lowest rates of unemployment (9.1 percent), while college dropouts face more than twice this rate. Individuals with intermediate education (high school dropouts and graduates) display the highest unemployment rates, at 23 percent -6 percentage points above unskilled workers. All groups, but college graduates, saw their rates worsen by 10 percentage points between 1988 and 1999. Unemployment rates decrease with age, although the effect seems to taper off after 35 years of age. More than one third of young adults face unemployment, calling for specific employment policies targeted at them. 40 Table 1.11. Unemployment for various demographic groups, Urban Colombia 1978 1988 1995 1999 Urban Colombia 7.7% 10.3% 8.7% 19.7% By sex Male 7.2% 7.7% 6.4% 16.7% Female 8.7% 14.2% 11.7% 23.0% By education Uneducated 3.3% 5.8% 4.6% 15.7% Primary 6.4% 8.5% 6.8% 17.3% High school dropout 9.6% 13.6% 10.9% 23.7% High school 9.0% 11.1% 10.0% 23.1% College dropout 15.5% 9.7% 10.2% 20.9% College 2.7% 6.0% 4.2% 9.1% By age 12 to 17 14.8% 24.0% 22.1% 44.4% 18 to 25 13.4% 18.2% 16.4% 34.6% 26 to 36 5.1% 8.5% 7.1% 16.5% 36 to 50 3.3% 4.4% 4.5% 12.4% 51 to65 4.0% 3.7% 4.0% 11.3% Over 65 4.9% 3.8% 4.1% 9.9% By quintile 1 12.2% 20.8% 18.0% 36.1% 2 12.4% 15.2% 11.9% 26.4% 3 7.7% 11.1% 8.8% 20.0% 4 6.4% 7.4% 5.9% 14.6% 5 3.8% 3.6% 3.6% 7.8% 1. Population 12 years old & older. 2. Uneducated: 0 years of education. Primary: 1-5 years of education. High school dropout: 6-10 years of education. High school: 11 years of education. College dropout: 12-15 years of education. College:16 years of education. 3. Quintiles are taken on household per capita income. Regional patterns of unemployment Cali's job market deteriorated the most during the last two decades; Bogota was severely hit by the recession. We now proceed to briefly analyze unemployment rates by cities (see Figure 11). Unemployment rates steadily increased between 1978 and 1995 and doubled in every city between 1995 and 1999, with the exception of Barranquilla, where the increase was "only" 47 percent; Barranquilla is thus now the city with the lowest employment rate in Colombia. Cali started with average rates in 1978 but its unemployment doubled from 1995 levels and is now the highest. BogotA suffered the most from the economic recession. Its unemployment rate more than tripled, but its level was exceptionally low in 1995 (5.8 percent, while all cities showed rates very close to 11 percent). Despite this jump, Bogota keeps the second lowest unemployment rate. 41 Figure 1.11. Unesmploymnnent by city, UJrban Coloumbia 25% 20% -~>Barranqullla ~ 15%- t c, //w}Buc lucanga E r" ogt 0r / 4ManizaIes E 10% M IIIn a) W7;-;C< -0 Medlif D -. / Cafi 5% -4- ~~~~~~~~~~~~~~Pasto 5% - ' 01% 1978 1988 1995 1999 Year Low skilled workers suffer less unemployment in IPasto and Barranquilhla -two frontier cities with strong links to international trade-. Analyzing unemployment rates for various sub-groups of the population in Tables A.4A4G in the statistical appendix, we find the national pattern of higher unemployment among the intermediately educated to hold in each city, with rates above 27 percent for college dropouts in Cali and Pasto. However, in Pasto and Barranquilla, unemployment for individuals with no schooling is much lower than anywhere else, at 4-5 percent, in contrast with Bogota where it has shot up from 2.5 to 17.9 percent between 1995 and 1999.49 Younger workers face the worst employment prospects in Call. Unemployment rates among the young are extremely high in Cali. Indeed, more than 50 percent of the individuals between 12 and 17 years of age were unemployed in that city in 1999, while that figure reached almost 40 percent for the individuals in the 18-25 years group.50 5. POVERTY DYNAMICS IN BOOMS AND RECESSION: GROWTH, INEQuALIry AND LABOR MARKETS In this section, we examine the dynamics of poverty from two perspectives: first from the perspective of aggregate effects of growth and inequality. Secondly, given the importance of average income per capita growth for poverty reduction, we decompose its aggregate variations in terms of the evolution of skill endowments, family size and dependency ratios, labor market participation and wage levels. We explore this decomposition at the aggregate level, based on several household types according to their insertion into the labor market -self-employment, wage-earning or both or none- or the skill level of their heads. 5.1. The decomposition of aggregate poverty changes in teirms of growth and inequality We start by decomposing changes in the headcount of the poor over time between three components (i) changes in real value of the poverty line (relative prices), (ii) economic growth (proxied by mean 49 Mainly induced by the end of the real estate boom in Bogota. 50 Recall that Cali also had unusually high labor participation rates for children aged 12 to 16. 42 income per capita growth), and (iii) changes in inequality (as measured by the share of each household's income distribution percentile in total income)51. Results are displayed in Table 1.12. Income per capita growth is the dominant factor behind the gains and losses in urban poverty from 1978 to 1995. While growth contributed positively and substantially to poverty reduction until 1995 (22 percentage points), higher inequality counterbalanced this effect by nearly 7 percentage points. Economic growth contributed to poverty reduction by 15 percentage points between 1978 and 1988, and 7 points between 1988 and 1995. Inequality, on the other hand, increased poverty by 2.5 percentage points first, and 4.8 points subsequently. Finally, more expensive food items raised the poverty line and increased poverty by 2.0 percentage points over the 1978-95 period. Table 1.12. Decomposition of poverty changes, Urban Colombia Actual change in Poverty Income poverty Growth Distribution Line Residual change 1978-1988 -15.2 -17.0 2.5 1.7 -2.3 50% Contribution 100% 112% -16% -11% 15% 1988-1995 -7.1 -11.7 4.2 0.3 0.1 25% Contribution 100% 164% -59% -4% -1% 1995-1999 7.5 3.3 2.2 3.3 -1.3 -6% Contribution 100% 44% 29% 44% -18% During the recession, poverty increases due to the combined effect of negative growth, increasing inequality and higher relative prices of food items. Poverty increased by 7.5 percentage points during the crisis period, wiping out the 1988-95 gain. This resulted from contributions of all three components in similar proportions: high prices and recession contributed 44 percent, while inequality contributed an additional 29 percent. The elasticity of poverty to mean household income has increased over time, reaching its highest level, -0.57, during the recessive period. This implies that proportional effects on poverty rates have been larger during the economic crisis than during the previous years. Thus, a one percent decrease in GDP is now associated with a larger change (increase) in poverty than the change (decrease) in poverty that a one percent rise in GDP once produced. A boost in the elasticity of poverty to growth means more poverty alleviation can be achieved through growth, but the consequences can be more severe if economic slowdown is revisited. If the poor are disproportionately vulnerable during times of crisis, as noted above, an economic slowdown also causes further deepening of poverty. 51 Formally, poverty rates in dates 1 and 2 can be expressed as functions of the level of the poverty line or price index of the poor's basket of goods (p), mean income (u) and the level of inequality in those dates (q), thus: P, = P(p1 p,, q, ) and P2 = P(P2 fP 2, q2) and our decomposition computes these three components i) Prices: P (P2,/12, q2) - P(P 1,2, q2) ii) Growth: P (P2 ,U2, q2)-P(P2 I UI Iq2) iii) Inequality: P (P2 1,4 2) - P(P2 '/42 q1) 43 5.2. The evolution of income per capita in ternns of skilD endowments, wages, dependency ratios and employment (1978-99) The dynamics ofpoverty in terms of basic income generation factors We have just seen that poverty dynamics follow the evolution of income per capita in relation to the poverty line. For any specific group of households, the poverty count will increase if its income per capitafalls and vice versa. In addition, any increase in the value of the poverty line (the price of the basket of goods consumed by the poor) will produce proportional effects, but this time in the opposite direction. Given identity (1), it follows that the change in real income per capita (y) for the household equals the sum of the percentage changes in average schooling, average wages per skill unit, adult to family size ratio and adult employment rate, or: y _w+s+ a+0 (2) where lower case letters y, w, s, a and 6 represent, respectively, the change of income Y, average wages W, average schooling S, adult to family size ratio (AIN) and rate of employment (TIA). Four types of labor market attachmentt: diffierences in poverty levels Before we explain the decomposition according to equation (2) we present in Table 1.13 the average schooling, family size indicators, employment rate and poverty rate for four different households grouped according to their labor market attachment (LMA). The four groups comprise those households whose working members (i) are all wage earners (type 1); (ii) are all self-employed (type II); (iii) are both wage earner(s) and self-employed (type In); and (iv) whose members only rely on non-labor income (type IV). We next identify the relative significance of each of the income generating factors in explaining income and poverty differences. IHlouseholds with mixed labor market atachment -type 1l1H. are less exposed to poverty than those linked exclusively to self-employment -type in-. In 1995 and 1999, the proportion of poor households is highest among type HI households (self-employed only) and their employment rate, at 29 and 30 percent in those two years, is the lowest among households that depend on labor income. On the other hand, type II households, which place their members in both self-employment positions and in the wage sector, have consistently faced the lowest poverty rate since 1988, at 36 percent then, falling to 31 percent in 1995 and increasing to 35 percent in 1999. Compared to type HII this type of household has more than twice the employment ratio -close to 67 percent-, better skill endowment - one more year of schooling- and marginally older heads with less children. Similarly, the advantages of type I -exclusive wage earners- relative to type m households are higher a skill endowment and employment ratio. A sequence of welfare improving labor market attachment types appear correRated to the stages of the household lifecycle. Households appear to follow a "1-3-2-4" sequence: the youngest households are in type (I) -40 to 42 years average head age-, the second youngest in type (I) -43-45 years-, the third in type (II) 46-48 years- and finally type (IV) has the oldest household heads -55-58 years-. In addition, type (11) households have the largest share of women employed -46 percent much larger than that of types (1) and (III) 41 and 32 percent, respectively. This suggests that as the household head ages and children require less parent presence at home, female labor participation increases and labor market attachment becomes mixed. At the latest stage in life, households' income earnings are generated out of non-labor income. 44 Table 1.13. Average household descriptive statistics by labor market attachment, Urban Colombia 1978-1999 No. of Average children Poverty Labor market Attachment educationI under7yrs Head'sage Employment2 Rate Type I. Only wage earners 1978 6.0 0.9 40 29% 66% 1988 7.6 0.7 41 36% 50% 1995 8.5 0.6 42 40% 44% 1999 9.4 0.6 42 40% 39% Type II. Wage earners and self-employed 1978 6.2 0.8 46 57% 56% 1988 7.9 0.6 48 64% 36% 1995 8.6 0.6 47 67% 31% 1999 9 1 0.6 46 67% 35% Type Im. Only self-employed 1978 5.1 0.8 45 24% 72% 1988 6.8 0.7 43 24% 59% 1995 7.4 0.6 44 29% 51% 1999 8.1 0.6 45 31% 53% Type IV. Only non-labor income 1978 5.4 0.4 59 0% 50% 1988 6.4 0.3 55 1% 51% 1995 7.0 0.2 58 0% 39% 1999 8.1 0.2 54 2% 39% I. Only for members 12 and older, excluding the household head. 2. Number of employed in the household divided by number of people 12 and older, excluding the household head. 3. Percentage of non-head household members 12 and older participating in the labor force Differences in income and poverty dynamics across types of labor market attachment and head skills 1978-99 In order to understand the evolution of per capita income, we apply the decomposition in terms of income generating factors described in equation (2) to the all households, to four types of households classified by their labor market attachment and to four household groups based on the head's level of schooling (see Table 1.14). Wage indexes are calculated by comparing predicted wages of the average level of schooling in the household. For that purpose Mincerean equations are estimated in every year -1978, 1988, 1995 and 1999- for wage earners and self employed and differentiated by gender. Therefore, wage indexes will necessarily be biased as long as they ignore the heterogeneity of skill levels within each group of households. 45 Table 1.14. D'ecomnposition of income dynamics in terms of skill endowments, dependency ratio, employment rate, and wages. Urban Colombia 1978-1999. CHANGE: 1978-1999 Population Share Mean income per Skills Dep'cy Emp. Wages ___________Share capita ratio rate W Poverty % count* 1978 1999 change Observed Predicted s a e w % change All households 100% 100% 0.0% 76.6% 43.6% 32.4% 9.3% 0.8% -1.6% -14.8% By heads' education Uneducated & primary 60.3% 36.1% -24.1% 34.3% 50.6% 24.7% 12.4% -0.7% 14.2% -8.7% Some & complete high school 31.2% 42.8% 11.6% 31.9% 12.2% 17.1% 2.2% 0.2% -7.3% -2.0% Some college 3.4% 7.2% 3.9% 58.7% 9.6% 9.7% 6.0% 0.4% -6.5% -11.3% College or more 5.2% 13.6% 8.4% 40.9% 5.2% 16.6% -11.9% 3.4% -3.0% -3.6% By labor market attachment Only wage earners 61.8% 46.4% -15.4% 16.1% 17.6% 10.3% 2.9% 2.3% 2.1% -19.2% Wage earners & self- employed 17.4% 18.9% 1.5% 13.2% 12.4% 9.4% 2.2% 2.4% -1.6% -14.6% Only self-employed 16.2% 23.3% 7.1% 9.7% 4.5% 11.5% 3.1% 2.1% 12.1% -7.7% Only non-labor income 4.6% 11.5% 6.9% 12.7% na 8.9% 3.0% na na 2.4% *Level change in percentage points. For the average household, the key sources of income growth between 1978 and 1999 are an increase in schooling endowment and a reduction in the dependency ratio -via smaller family size. Observed and predicted incomes grow by 77 and 44 percent, respectively. Three quarters of the predicted rise in income are explained by growth of school endowments -32 percent- and one quarter by reductions in the adult to family size ratio -9.3 percent-. The other two income generating factors -the change in the employment ratio and the wage change- are negligible. In summary, nearly 11 percentage points of the total reduction in poverty between 1978 and 1999 -14.2 percent- are associated with the rise in education endowments of the households. The population is shifting towards households with better-educated heads and lower poverty risks. An alternative way of looking at the impact of education is the changes in population shares by education of household heads. After being 60 percent of total urban households in 1978, the population share of households with heads with primary education or less fell to nearly half that number in 1999. Nevertheless, this still represents close to a third of the urban Colombian population. Simultaneously, households headed by high school graduates or dropouts -which enjoy 60 per cent higher per capita income and lower poverty risks- became the mode, with 43 percent of the total population. Thanks to the fast growth of college educated people, households headed by college graduates -with five times the earnings of primary school graduates- tripled their number vis- A-vis 1978 and represented one out of every seven in 1999. Low-skiDled-headed households reap the most benefits from lower fertity, higher wages and more education, but not from higher employment ratios. When households are grouped by heads' 46 schooling, income gains for low skill headed households are much larger -51 percent-. Their gains are nearly five times those of households with heads who have had some high school education or some college and ten times those of households headed by college graduates. Income gains derived from real wage gains and family size reduction are larger for less educated head households -14.2 and 12.4 percent respectively!-. The behavior of college-graduate-headed households produced a nearly asymmetric effect, with an income per capita loss of 12 percent due to increased fertility. On the other hand, households headed by college dropouts benefited by 6 percent owing to fertility reductions. Even within their own categories, gains in education endowments where the highest for the less- skilled-headed households -25 percent- and had a similar order of magnitude for the other groups. Wage index growth did represent a relative loss for all household groups except that with less educated heads. Households whose heads had some college or some secondary school education sustained the worst impact -minus 6 to minus 7 percent. The only group that received a gain form higher employment rates was that with college graduate heads with a 3 percent gain. Despite the relative success in income growth and poverty reduction of households with less-schooled heads, their poverty risk remains the highest among all groups -72 percent!-. Households with only-self-employed workers are becoming an increasing share of the population, and are the only group suffering a significant loss in income per capita growth via reduction in wages. Classifying households according to their LMA type shows quite a different picture. Income per capita gains are less heterogeneous, although the gains for the only-self-employed households are well below average (See Table 1.14). Income gains explained by higher schooling, adult to family size ratio and employment ratio are quite similar -average percent gains are, respectively, 10, 2.9 and 2.3-. The major sources of difference across groups are dynamics of wages and population shares. Households with exclusive provision of self-employed workers have lost 12 percent in terms of wage reduction, while simultaneously their share of the total urban population has increased by 7 percentage points. The clearly opposite case is found among only-wage-earner households; their wages have increased by 2 percent, but at the same time their population share has diminished by 15 percentage points. Nevertheless, they are still the dominant type of household, with 46 percent of the urban population. 5.3. The dynamics of poverty in the recession: lower wages or jobs lost? During the recession, the overall poverty and unemployment increases -of 7.5 and 11 percentage points, respectively- hid stark variations among households, depending on their labor market choices (see Table 1.15). All household types, except the type I "only-wage-earning" group, saw an increase in poverty of 10 or more percentage points. The worst hit were households with only self-employed workers or only non-labor income. Households whose members were all wage-earners saw the smallest variation during the recession: their poverty rate only increased by three percentage points from 44 to 47 percent between 1995 and 1999. Asymmetric dynamics of quantities and prices during the recession:job lossfor wage earners versus salary reduction for the self-employed Without any downward wage flexibility, wage earners faced large job losses during the recession. Table 1.15 displays changes in wages and changes in population shares for the respective groups. Households with some labor market attachment to wage earning jobs -type I and type 11- lost substantial share in the urban population. Their drop was equivalent to almost 6 percentage points: 3.2 percentage points for wage earners and 2.7 points for mixed labor market attachment. This reflects an annual rate of growth of wage earning jobs of 1.9 percent, well below the annual growth rate of the working population of 2.8 percent.52 Despite the labor demand reductions, individuals 52 According to calculations of L6pez (2001). 47 employed in the wage sector actually saw their wages rise by 3.7 percent. Hence, the formal wage sector, facing wage rigidities, responds to the economic recession by cutting jobs in an enormous scale. Table 1.15. Decomposition of income dynamicrs in terms of skill endowments, dependency ratio, empRoyment rate, and wages. Uirban Colombia 1995-1999 CHANGE: 1995-1999 Population Share Mean income per Skill, Dep 'cy Emp. Wages Population____ _ Shrecapita _____ ratio rate _ Poverty % ~~~~~~~~~~~~~~~~count 1995 1999 Observed Predicted s a e w change % change All households 100% 100% 0.0% -1.5% -1.0% 1.9% -0.1% -1.3% -1.5% 7.5% By heads' education Uneducated & primary 40.6% 36.1% -4.5% -5.1% -3.5% 0.7% -0.1% -1.9% -2.2% 11.6% Some or complete high school 43.3% 42.8% -0.6% -2.5% -3.0% 0.9% -0.2% -1.7% -2.0% 10.5% Some college 5.3% 7.2% 2.0% -1.9% -1.1% 0.7% 0.2% -1.9% -0.1% 6.0% College or more 10.3% 13.6% 3.4% -3.5% -0.9% 0.7% -1.3% 0.1% -0.4% 1.8% By labor market attachment )nly wage earners 49.6% 46.4% -3.2% 1.7% 6.3% 2.2% 0.7% -0.3% 3.7% 3.4% W'age earners & self-employed 21.6% 18.9% -2.7% -1.9% -0.5% 1.8% 0.2% -0.1% -2.4% 9.6% )nly self- employed 21.6% 23.3% 1.7% -7.1% -10.2% 2.1% 0.8% 0.0% -13.1% 11.5% )nly non-labor income 7.2% 11.5% 4.2% -1.1% na 2.9% 0.2% na na 14.5% *Level change in percentage points On the other hand, self-employed individuals maintain or increase thenir job share but face larger reductions in earnings. While wage earners' wages increased by 3.7 percent, self-employed individuals faced a 13.1 percent drop in salaries. The fall severely affected self-employed males, who lost 22 percent. Workers with only primary schooling lost 30 percent while college-educated employees saw their salaries decrease by 18 percent. In the wage sector, male employees faced a 5 percent average loss, spread from a 9 percent loss among primary school educated workers and stable wages for college graduates. It is then clear that the wage sector adjusted to the recession with a shrinking of posts while the self-employment sector adjusted through decreased incomes. Disparities in wage dynamics and employment by gender Disparities in wage dynamics by gender: the gender wage gnp continued to narrow during the recession -especially among less-skilled wage earners - as a result of an apparent substitution of males by females in the wage-earning sector. While male earnings decreased between 1978 and 1988 and again between 1995 and 1999, female earnings have been increasing steadily since 1978, 48 and the rise continued during the recession. Female wage earners actually saw their wage income increase by 15 percent (Table 1.16), while self-employed females managed to keep a 2 percent increase -somewhat higher for the low skilled-. Female workers with low schooling are faring better than college educated ones, with a 17 percent spread, which is exactly the opposite as is the case for male workers. Simultaneously, female workers are aggressively entering into all skill levels of self- employment jobs -increasing their share of employment between 4 and 5 percentage points (!)- (Figure 1.12). However, their entry into wage-earning jobs is mostly concentrated at the middle-low skill level. This suggests that part of the downward rigidity of the wage-earning sector -especially at the low-skill level- is being confronted by the substitution of male workers by female workers that are being attracted with relatively higher wages - given the gender gap-. Table 1.16. Changes in Wages and employment by skill level, Urban Colombia, 1995-1999 Change in female employment Change in Wages share Wage earners Self-employed Wage Self- Skill level Male Female Male Female eamers employed Total None & pnmary -9% 25% -30% 12% 4.8 5.1 4.7 Some & complete high school -8% 16% -21% 4% 2.5 4.4 2.79 Some college -4% 11% -18% 0% -0.2 3.8 0.36 Complete college 0% 9% -18% -3% 2.2 4.8 3.39 Figure 1.12. Change in female labor force share by occupational choice and skill Urban Colombia, 1995-1999 6.0 5.0 _ 4.0 ,3 3.0 2.0 - 1.0 -4--Self-employed 0.0 None & primary Some & complete Some ollege Complete college high school -1.0 Skill category Female entries were mostly by spouses who belonged more than proportionally to poor households. Male household heads and other males suffered heavy reductions in employment rates - minus 6 and minus 10 percent respectively- and they belonged more than proportionally to non-poor households (Concentration Coefficients -CC- for net labor market exit are 0.113 and 0.024 respectively).53 Other female workers (non-spouses) also reduce their employment rates by 3 percent 53 Similarly to the Gini coefficient, a higher value of the concentration coefficient means larger participation of the non-poor. More that proportional participation of the poor is indicated by negative values. 49 while female household heads also increased their employment rate by 1 percent and, were almost neutral with respect to income distribution. However, female spouses continued to enter the labor market aggressively -increasing their employment rates by 4 percent- but they belonged mostly to poor households. The effects of labor market dynamics on households by heads' skill levels and labor market occupational choices Job losses explain more than half ot the poverty inrcrease during the recession. The recession was accompanied by a 7.5 percentage point increase in poverty (Table 1.15/17). Households with only wage earners decreased by 3.2 percentage points, while those with both wage and self-employed earners decreased by 2.7 percentage points. This decrease corresponded with a 4.2 percent increase in number of households with no-labor market income and a 1.7 percent increase in households which only access self-employment earnings. This latter group faces the highest poverty rates. At least half of the total poverty increase must be explained by this shift, which is associated with the 7 percent (!) increase in households with unemployed heads and 2.1 percent rise in households without employment for the head and other members (!). Larger income losses and the strongest poverty increase hEt low-skilled- headed households: jointly, wage reduction and job losses explain the outcome. Households whose heads have secondary education or less (including those who never attended school) faced the most severe income per capita loss: they lost 2.5 to 5 percent of income during the economic downturn and saw their poverty levels rise by 11-12 percentage points. Once again, this emphasizes how the crisis has hit those households that were in already in a vulnerable position. Income losses follow an inverted- U shape, with the next highest loss being that experienced by households with a college-educated head, at 3.5 percent. Households whose heads reached some primary school or less faced a 1.9 percent reduction in income due to job losses and a 2.2 percent loss due to decreasing wages. In addition, for households whose head completed some high school, the respective losses were of 1.7 percent for job loss and 2.0 percent for wage adjustment. Households headed by college dropouts did lose 1.9 percent in employment rates but very little via wage adjustments. Paradoxically, income losses for households with college-graduate heads were due almost exclusively to larger family sizes. During the recession, households whose heads completed college education lost 3.5 percent of their income, but faced almost no changes in their wage index or their employment ratio. However, they registered two major changes: first a significant reduction in their adult to family size ratio -an implicit increase in the number of dependent members- plus an unprecedented rise in their population share of 3.5 percentage points in just four years. Apparently, larger households -joining several family units- have reorganized around college-graduate heads. That could be a possible explanation for the shift in population share and the increase in dependency ratios for this type of household. lIn summary, households in which income is generated only thirough self-employment faced a drop in earnings while those that engage only nnn the wage sector faced job losses. The adjustment to the recession and the derived welfare losses seem to function according to separate mechanisms in both sectors. Households with only self-employed members took a severe drop of 13 percent in their wage index with a simultaneous increase in the number of people in that category of 1.7 percentage points. Meanwhile, the salaries of wage earners kept rising in real terms -3.7 percent- in the middle of the recession (!), but 4.2 percent of this group was expulsed into complete unemployment and 1.7 percent was left to try to generate income through self-employment only. Households whose members engaged in both markets faced both sets of adjustments: a reduction in earnings plus a reduction in population share. 50 6. SOURCES OF INCREASING INEQUALITY: AN INTRODUCTION In this part, we present preliminary results about the likely sources of inequality and their relative importance for different types of income. We also consider household behavior in terms of both lifecycle and adjustment of working hours. Chapter 3 presents a more thorough investigation of inequality changes via micro-simulation of structural parameters and endowments changes in a model of household income generation, labor force participation and occupational choice. 6.1. Decomposition of inequality by household characteristics Education accounts for a sizeable share of inequality. To explore the determinants of inequality changes, we decompose the entropy measures of dispersion into their "within" and "between" components, using the same kind of correlates applied to the poverty profile analysis in the previous section: head's gender, education, experience, city of residence and sector where employed, plus average education of other household members, dependency ratio and employment rate. The results of this exercise are presented in Table 1.17. Education, be it of the household head or the average of other household members above 12 years, is the only variable that consistently accounts for any sizeable share of inequality between homogeneous groups of households over the period of analysis. The occupation of the head also shows some minor explanatory power regarding the between component of inequality. We suspect, however, that since the occupational choice and labor force participation -other than the head- are so closely linked to educational attainment, they might also just be partially capturing the effect of education54. All other variables -head's gender, potential experience, age, city of residence and sector of economic activity- explain less than 0.03 points of the between component of inequality, irrespective of the entropy measure. Table 1.17. Inequality decomposition by household characteristics, Urban Colombia 1978-1999 1978 1988 1995 1999 Within Between Within Between Within Between Within Between Theil Index Head's gender 0.43 0.00 0 47 0.00 0.62 0.00 0.60 0.00 Head's educatLon 0.28 0.15 0.31 0.16 0.45 0.17 0.40 0.20 Head's experience 0.39 0.04 0.45 0.02 0.60 0.02 0.57 0 03 Head's age 0.41 0.02 0.46 0.01 0.60 0.02 0.58 0.02 City of Residence 0.41 0.02 0.45 0.01 0.60 0.02 0.57 0.02 Head's sector 0.40 0.02 0.45 0.02 0.61 0.02 0.54 0.06 Head's occupation 0.38 0.05 0.41 0.05 0.57 0.06 0.52 0.08 Adult tofalimily size ratio 0.10 0.08 0.10 0.10 Employment rate 0.06 0.07 0.06 0.08 Average education of the household 0.35 008 0.34 0.13 0.43 0 19 0.36 0.23 Fine2 0.18 0.24 0.20 0.26 0.36 0.26 0.28 0.32 1. Fine decomposition refers to 1764 groups according to certain caracteristics of the head ( 2 gender*6 education groups *3 expenence*7 occupation* 7 cities) Finer sample partitions of the sample -with education plus all other characteristics- do not produce a substantial raise of between inequality. Based on these results, we partition the samples in every year into 1,764 demographic/skill groups (2 genders, 6 education levels, 3 experience 54 See models of labor force participation and occupational choice in Chapter 3. The occupational categories included in the ENHs are: family worker, blue collar, white collar, domestic employee, self-employed and employer. Hence, aside from self-employed, the categories are basically given by the education of the individual 51 groups, 7 occupations, 7 cities), looking to obtain higher variability between these very finely defined groups. The results are presented at the bottom of Table 1.17 (under the heading of "fine"). At the most, all these observable characteristics explain 28 percent of income inequality between households (as measured by E(O)). Thus, 72 percent of the inequality between households remains unexplained, or driven by other factors. The discrhninatory power of human capita endowments has been increasing during the last two decades -especially other members'-. Not only is education a major component of the "between inequality", but its role has been increasing. For example, the average educational endowment of the household contributed 0.06 points in 1978 (measured by E(0)), and 0.21 in 1999, while the educational attainment of the head rose from 0.12 in 1978 to 0.19 in 1999. For both, most of the increase took place between 1995 and 1999, with the recession reinforcing the value of higher education. The within component of inequality increases, indicating widening differences inside each group. The final point concerns the within component of inequality. This component has increased for all partitions of the sample, meaning that not only did the gap between the "haves" and "have- nots" widen over these twenty years, but that the least well off inside each group fare relatively worse in 1999 than in 1978. 6.2. Composition of income and inequality: labor income and other sources Wage income has grown more unequal since 1988 and increased its share after 1995 Table 1.18 presents the evolution of shares and inequality for the different components of household income, namely wage income, self-employment earnings and other income. Wage earnings became more unequal, starting in 1988, for a total Gini increase from 0.38 in 1978 to 0.46 in 1999. Despite this rapid concentration, wages remain the least unequal component of income but their contribution had been sharply decreasing from 67 percent of total earnings in 1978 to 50 percent in 1995. Despite the recession, average wages continued to grow -by 13 percent- and this growth was indeed faster than at any other previous period, which accounts for the increasing share of wages from 50 to 57 percent between 1995 and 1999. Mean wage income grew by approximately 54 percent over the entire period. We note, however, that median wage income fell by 5 percent in the last four years, pointing to the relative loss in lower wages and gains in the higher part of the distribution. Table 1.18. Inequality by income sources, Urban Colombia 1978 1988 1995 1999 Wages Self-W ages Self- Other Wages Self- Other Wages Self- Other enzployment employment eOployraent employment Gini(%) 0.40 0.60 0.46 0.38 0.57 0.54 0.43 0.60 0.58 0.46 0.57 0.52 E(0) 0.28 0.70 0.48 0.24 0.61 0.54 0.32 0.67 0.66 0.35 0.58 0.49 E(1) 0.32 0.80 0.45 0.30 0.70 0.61 0.44 0.86 0.81 0.44 0.65 0.55 Mean income* 369 474 310 441 571 381 502 680 449 569 515 408 Median income* 247 232 297 296 277 236 337 375 225 320 250 240 Share of total personal income 67% 27% 6% 56% 28% 16% 50% 34% 16% 57% 30% 13% * In thousands of 1999 pesos In the recession, self-employment earnings lost share and became less unequal. Inequality showed no clear tendency and its Gini oscillates between 0.57 and 0.60, with a relative low at 0.57 in 52 1999 and 1988. In 1999, self-employment earnings remain the most unequal income source and their share decreased by 4 percentage points, reversing two thirds of the gains obtained during the 1988-95 period. This was caused by the severe drop of 24 percent in average income and 33 percent in median income, during the recession. The recession reduced both inequality and share of "other" income. Income from other sources concentrated until 1995 - with the Gini coefficient rising from 0.46 to 0.58-, but after the recession, the Gini fell back below its 1988 level, for a total increase of 6 percentage points. While other income came to represent approximately 16 percent of total income in 1995, a 10 point increase from 1978, its contribution decreased again to 13 percent in 1999. In an evolution similar to that of wages, the mean income increased by 18.0 percent but the median decreased by 4.9 percent, pointing again to losses in the bottom part of the distribution and higher gains in the upper tail. Shorrocks decomposition Up to 1995, self-employment income was the main source of inequality, with a 44 percent share. In Table 1.19, we use the methodology proposed by Shorrocks (1984) to measure the relative contributions of the different types of income to total income and inequality. While percentages varied, until 1995, the ranking was self-employment, wages, other income, and imputed rents. After the economic recession, wage income replaced self-employment as the major component of inequality. Consistent with the findings above, from 1995 to 1999, wages increased their contribution to household income inequality by 24 percentage points, up to 58 percent (!), while self- employment income contribution fell from 44 to 26 percent. On the other hand, the relative contribution of other earnings to inequality fell from 19 to 12 percent, while its contribution to income fell by one percent. Labor earnings (from wages and self-employment combined) made up 84 percent of total inequality in 1999, a proportion that exceeded 1995's share by 7 percentage points, and similar to that of 1978. Pensions and interest income contributed the most to inequality among other incomes, but the opposite occurred with housing rents. For 1999, the available informnation enables us to similarly decompose the "other" income in Table 1.19A. Again, all components contributed positively to the overall level of inequality, with the most important contributors being pensions (27 percent), interest income (25 percent), and imputed rents (16 percent). The concentration was highest among pensions and interest income. In contrast to pensions and interest income, imputed rents on housing property were much more progressively distributed and represented 40 percent of non-labor income in 1999. Even imputed rents on housing property seemed relatively neutral in 1999, even though homeownership suffered a severe and regressive reduction after 1995.55 Table 1.19. Shorrocks inequality decomposition by factor components of household income, Urban Colombia 1978-1999 1978 1988 1995 1999 Share of Share of Share of Share of Inequality Income Inequality Income Inequality Income Inequality Income Wages 36% 57% 24% 51% 34% 47% 58% 48% Self-employment 47% 24% 49% 25% 44% 30% 26% 28% Other 10% 9% 21% 14% 19% 14% 12% 13% Imputed rents 7% 10% 5% 10% 3% 9% 5% 10% Total 100% 100% 100% 100% 100% 100% 100% 100% 55 See section IV above. 53 Figure 1.13. The distribution of pension and non-pension earnings by household income percentile: Concentration curves, Urban CoRombia, 1999 100% 90% 90% _ Concentration Coeffidents: 80% Pension: 0 624 Non-pension- 0.539 70% Total: 0.543 60% - 50% -eo I -Non-pens:,o]n I / 40% - 30%- 20% 10% 0% - , -- - 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 Mean household income per capita percentile The poor do not benefit much from current pension benefits. Figure 1.13 shows concentration curves and concentration coefficients for total earnings and their pension and non-pension components, for 1999 urban Colombia.56 The poorest 50 percent of the population receive less that 10 percent of total pension benefits. Moreover, 65 percent of earnings via pensions go to the richest 20 percent of households. The concentration coefficients for pensions -measure of inequality comparable to the Gini coefficient- is 0.624 much higher that the Gini coefficient for income in 1999, 0.543.7 One caveat, however, is that the pension data do not distinguish between private and public sector pensions. Nevertheless, one can suspect that public sector pensions are even more concentrated among the upper percentiles of the population, since public employment is very regressively distributed.58 6.3. Labor Income Inequality: checking for hours of working week and lifecycle bias Restricting the analysis to full-time prime-age workers does not change the results. We first check the validity of the results obtained above regarding labor income inequality. It could be argued that the observed increase in wage income inequality is due for example, to a cohort effect along the 2 1-year period, or to differential changes in hours worked across various segments of the distribution. We thus computed inequality measures for monthly and hourly earnings59 on a sample of full-time 56 Once again, we provide further emphasis by depicting the shares of pension and non-pension income for mean household income percentiles in Figure 8B. 57 Even by Latin American standards this seems a very large level of inequality. Comparing with Wodon (2000), table 2.6, in Colombia the Gini elasticity for pensions (ratio of concentration coefficient to Gini) is higher than in all other countries, except in Brazil. 58 According to Vdlez and Millan (2001) the top quintile of the distribution of household income receives 80 percent of the public sector payroll and the Concentration Coefficient of public sector labor earnings is even more regressive 0.736. Which would be the highest in Latin America when compared to the figures reported by IDB in 1999 (IDB (1999) Amenca Latinafrente a la desigualdad, Figure 8.32). 59 To compute hourly wages we multiplied the variable "hours usually worked a week" by 4.29 (number of weeks in the average month) and divided monthly earnings by the resulting quantity. 54 workers6o aged 25 to 55 years. Results are presented in Table 1.20 for wage income and for self- employment earnings. Levels and changes in inequality have similar values to those in Table 1.18 but 1) the inequality of hourly wages is always larger than that of monthly wages; 2) there is slightly more inequality among full-time wage earners than was apparent earlier; 3) on the contrary, full-time self-employment earnings seem less unequal than earlier depicted but the difference is again very small. Table 1.20. Labor income inequality: hourly and monthly earnings, individuals aged 25 to 55 who work 40 or more hours a week, Urban Colombia 1978-1999 1978 1988 1995 1999 Monthly Hourly Monthly Hourly Monthly Hourly Monthly Hourly Wage Earners Gini (%) 41% 44% 38% 41% 45% 46% 47% 49% E(O) 0.28 0.32 0.24 0.27 0.33 0.35 0.37 0.41 E(1) 0.33 0.37 0.29 0.33 0.46 0.47 0.45 0.48 E(2) 0.52 0.59 0.48 0.54 1.35 1.26 0.82 0.84 p90/plO 5.22 6.25 3.92 4.76 4.62 5.04 5.51 7.48 p90/pSO 3.00 3.21 2.50 2.83 2.75 2.92 3.71 4.15 p75/p25 2.14 2.32 2.14 2.25 2.50 2.59 2.50 2.70 Self-employed Gini (%) 53% 55% 54% 55% 57% 58% 56% 58% E(O) 0.53 0.55 0.51 0.54 0.58 0.60 0.56 0.62 E(l) 0.56 0.58 0.63 0.66 0.78 0.78 060 0.66 E(2) 1.07 1.05 1.86 1.87 2.48 2.39 1.10 1.21 p90/plO 12.50 11.48 10.00 10.91 8.89 10.50 15.00 14.29 p90/p50 3.75 3.75 3.75 3.74 4.00 3.60 5.00 5.08 p75/p25 3.18 3.20 2.80 3.00 3.33 2.86 3 33 3.85 7. SUMMARY AND CONCLUSIONS Social and economic development: a mixed outcome After studying three development axes such as access to basic social services, violence indicators and economic welfare, the situation of urban Colombia does not reveal stable and collinear trends. First, social progress appears contradictory. Although most of the social indicators in education, health and infrastructure show substantial and persistent long-term improvements during the last two decades, the simultaneous escalation of violence reveals a considerable deterioration of living conditions and has become a significant economic burden in urban Colombia. Only recently -probably due to the economic recession- has education coverage declined. Urban Colombians are unambiguously better off now than in 1978, but not better than in 1988. Economic welfare -measured by income per capita- almost doubled from 1978 to 1995, but followed a clearly inverted "U" afterwards. Following the robust gains in all welfare measures during the 1980s until 1995, the negative impact of the recessive period pushed welfare levels back to late eighties levels. Moreover, the adverse effect was stronger on the poor population and as a result at least a decade of welfare improvements has been lost. Faces of the poor: the same but more polarized Poverty profile analysis shows that the typical faces of the poor have not changed much during the last two decades. However, those same faces are becoming more ostensible and polarized. Poor households in urban Colombia suffer increasingly from lower skill endowments, lower employment rates, and higher dependency ratios. High school education is losing some of its protective power, 6° 40 hours or more worked per week. 55 and some college education is becoming indispensable to avoid poverty. In employment rates, the disadvantage of the poor does not come from the difference in employment of their heads, but from the gap of employment of spouses and other household members of working age. Moreover, at the moment, unemployment peaks for the young and those with intermediate skills. Over the years, the presence of children is increasing its power in predicting poverty by increasing dependency and reducing labor force participation. The presence of older heads or other older members produces the opposite effect. Increasing risk of poverty comes also from disabled, recently unemployed or female heads, while recent migration -presumably displaced populations- is becoming an alarming signal in the late nineties. Homeowners and middle-to-high age-headed households are doing better. In summary, the faces of the poor are typically children of all ages, young low-middle skilled household heads, recent migrants and non-homeowners. These groups are clearly worse off than pensioners, the college educated, the elderly and non-recent migrants. Poverty dynamics: GDP per-capita growth is key for poverty reduction Decomposition of poverty dynamics shows that income per capita growth explains most of the gains and losses in urban poverty during the last two decades. However, during the recession, poverty increases due to the combined effect of negative growth and increasing inequality. In turn, our dynamic decomposition of income per capita growth for the average urban household shows that during the last two decades the key sources of growth are the rise in education endowments and the reduction in the dependency ratios -via smaller family size-, but not much from changes in employment ratios or wages. However, examination of household income growth factors across skill levels and labor market attachments reveals some heterogeneity. Low-skilled-headed households benefit the most from lower fertility, higher wages and more education, but not from higher employment ratios. At the same time, households with only-self-employed workers are becoming an increasing share of the population, and are the only group suffering a significant loss in income per capita growth via wage reductions. Job losses generate most of the poverty rise in the economic recession. A similar decomposition for the recessive period (1995-99) shows that most of the poverty increase - 7.5 percent- is explained by losses of wage-earning jobs and the remainder by lower wages for self- employed. Nevertheless, some additional adjustment in wages was obtained via higher female labor market participation and a reduction in the gender wage gap -in both self-employed and wage- earning markets-. Low-killed-headed households took the largest reductions in income and the strongest poverty increase, by simultaneous losses in wages and jobs. In summnary, households in which income is generated only in the wage sector faced increasing unemployment; some of them managed to enter the self-employment market and took a severe drop in labor earnings and the rest where left without any labor income source. Increasing Inequality Urban inequality has increased substantially during the nineties. Exploring the determinant factors of household income inequality shows that education, be it of the household head or other household members of working age, is the only variable that consistently accounts for any sizeable share of inequality between homogeneous groups of households over the period of analysis. In addition, the discriminatory power of human capital endowments has been increasing during the last two decades - especially other members'-. Analysis of income by sources shows that labor income is the driving force behind increases in inequality, with an increasing regressive effect, especially that of wage income, which, after the economic recession, replaced self-employment as the major component of inequality. The components of "other income" do not have a homogeneous impact on income 56 inequality: while pensions and interest income contribute most to inequality among other incomes, the opposite is true of housing rents, which are more progressive. Cross-city comparisons: convergence and improvement in left behind cities Cross-city comparisons of social indicators reveal a pattern of convergence, with strong catch-up by the laggard cities since 1978 -namely Barranquilla. Something similar happened with economic welfare indicators. After adjusting average income by differentials in local cost of living and inequality, welfare convergence is even stronger. Bogota loses all its relative advantage and in 1999, it bunches together with the three smallest cities of the sample, Pasto, Bucaramanga, and Manizales, which rank far better than the three other "grandes ciudades" -Barranquilla, Cali and Medellin-. Then which Colombian city enjoys the highest welfare right now at the turn of the century? The answer depends on the choice of social weights: by average income, it is Bogota, but by poverty count, it is Pasto and by extreme poverty, it is Bucaramanga. In dynamic terms, Barranquilla showed the highest progress -in all development indicators- while Cali faced the most significant losses.6' Figure 1.14. Colombia's main problems survey, Latinobarometro 2000 Violence_l Unemployment Education Corruption Other Low wages Housing &3 health 0% 10% 20% 30% 40% 50% Policy priorities: security, growth and education Available evidence shows (Figure 1.14) that Colombian citizens have public policy preferences that are clearly consistent with the main findings of this paper. Nearly four out of every five Colombians think that public policy priorities fall in either violence, unemployment, or education. According to the Latinobarometro surveys for the year 2000, 38 percent of Colombians see violence as the main problem facing the country, 25 percent think it is unemployment and nearly 15 percent believe it is education. Low wages and housing or health services are considered priorities by less than 5 percent of the population. Moreover, in relation to violence issues, policy priorities for the population appear to be mostly related with personal security and protection of property rights. Moser (1999) study on perceptions of the poor and violence in Colombia shows some similarities in the ranking of problems (violence, unemployment-poverty, education with 43, 14 and 6 percent respectively) and presents a very detailed presentation of specific violence problems which shows that 29 percentage points (!) are related to state presence in protection of personal security and the rights to life and property, namely, 61 See the Bank's 1999 special report on Cali Poverty in Cali, Colombia. 57 insecurity, robbery, intra-family violence, homicide, rape, fights, gangs, loitering (vagancia), threats, and guerrilla and paramilitary presence.62 Some results of this report help to identify the specific aspects of demand for education more clearly. First of all, many households that are facing severe reductions in income due to the recession are reducing their demand for private basic and tertiary education and are looking for alternative options or substitutes in the public sector, either by increasing their demand for public basic education - already congested- or by obtaining some form of publicly provided education credit. On the other hand, the demand for education appears to be related to quality as well. This should be the case if we consider that the disproportionate share of unemployment of high school graduates is related -at least in part- to a lack of acquisition of the skills demanded in the labor market.63 Finally, the priorities of public spending identified in Chapter 6 indicate that for the average poor household the first priority is the expansion of childcare and and the extension of public education to the preschool level. All of the above would require much greater flexibility of the education public finance framework in order to allow a responsive flow of public resources to match the popular demands for education (number of students for each specific modality) at the local level. Fortunately, in the past, Colombia has tried several innovative instruments with relative success, but some of them have been inexplicably abandoned or have had limited use. For example, 'Escuela Nueva', introduced in 1975 in the rural areas, performs very efficiently, yet its benefits have been generally underexploited, while the successful PACES scholarship program for secondary school that was introduced in the early 1990s was nevertheless abandoned despite its success.64 Similarly, the ICETEX public credit for higher education was introduced successfully in the 1960s, but its coverage is insufficient today. Recent initiatives include a Bolsa Scola type program -Subsidios Condicionados- and the expansion of public credit for higher education using the commercial bank network (two potential World Bank operations). Major changes are required in the legislative area. The Colombian government is preparing a proposal to reform the sub-national public finance regime for education -both as articles of the Constitution and Ley 60 in order to provide funds for education in proportion to the number of students served; the measures are expected to produce improvements in both efficiency and equity. With respect to unemployment, the obvious long-run remedy is to recover higher and sustainable levels of economic growth. However, remedial remedies in the area of social protection in Colombia are insufficient. A World Bank Report on social protection for Colombia establishes the priorities for the most important actions required in that area. In summary, political institutions in Colombia face top policy challenges in three main areas: 1) addressing governance issues related to violence concerns; 2) achieving macroeconomic and fiscal stability to recover sustainable growth rates in order to reduce unemployment; and finally, 3) the improvement of efficiency and equity in the provision of education, an area that is gaining increasing importance over the years and is clearly recognized as essential by Colombian citizens. 62 See Table 2.2 in Mosser (1999). This document provides much more detailed ranking of problem according to economic and socio-demographic variables. Since it is based on focus group methodology statistical significance has not been checked. 63 In fact, the national high school exam (ICFES) that monitors school performance is designed with the exclusive purpose of identifying and measuring school achievements in the areas of knowledge that are best aredictors ofs college performance. Not labor market performance of high school graduates. See evaluation as a randomized experiment by Angrist, et al., 2001. 58 References Angrist, Joshua D., Eric Bettinger, Erik Bloom, Elizabeth King, Michael Kremer. 2001. "Vouchers for private schooling in Colombia: evidence from a randomized natural experiment." MIT: mimeo. Birchenall 1998. "Capital Humano y Crecimiento Econ6mico," en Sanchez F. La Distribucion del ingreso en Colombia. Bogota: UJMACRO_DNP. Bourguignon, Francois. 1999. "Crime, violence and inequitable development': paper in progress. Washington, D.C. Comisi6n de Estudios sobre la Violencia. 1987. Colombia: violencia y democracia. Bogota: COLCIENCIAS, Universidad Nacional de Colombia. CEPAL. 1997. Panorama Social de America Latina. CEPAL. Santiago de Chile. DNP-UNDP. 1999. Informe de Desarrollo Humano para Colombia 1999. Bogota: Tercer Mundo Editores. Florez, C.E., F. Perez, et al. 1999. Riesgos y Oportunidades de las Familias Colombianas. mimeo. Bogota: Misi6n Social-DNP Gaviria, Alejandro 1998. "Increasing returns and the Economic Evolution of Violent Crime: The case of Colombia," Discussion paper, University of California, San Diego IDB. 1998. Colombia: Eficiencia del Gasto Publico en Educaci6n. Washington, Serie de Estudios Sectoriales y Econ6micos. Regi6n 3 RE3-98-008. Lederman, Daniel, Fajnzylber and Norman Loayza. 1998. "%,Qu6 causa el crimen violento?", in Corrupci6n, crimen y justicia Una perspectiva econ6mica, Bogota: TM Editores, LACEA, pp.53-95 Londofio, Juan Luis. 1998. "Epidemiologia econ6mica de la violencia". Ponencia ante la Asamblea del BID. Cartagena. and Cathy Mcllwaine. 1999. "La violencia en Colombia segun la percepci6n de comunidades urbanas pobres", articulo no publicado, LCSES, World Bank. Montenegro, Armando and Carlos Esteban Posada. 1995. "Criminalidad en Colombia", Coyuntura Economica Vol. XXV N° I Morley, Samuel. 2000. La Distribucion del Ingreso en America Latina y el Caribe. Fondo de Cultura Econ6mica, Comisi6n Econ6mica para America Latina y el Caribe, Santiago, Chile. Moser, Caroline. 1999. "La violencia en Colombia: C6mo construir una paz sostenible y fortalecer el capital social" en Andres Solimano, Felipe Saez, Caroline Moser y Cecilia L6pez (Editores), Ensayos sobre Paz y Desarrollo: El caso de Colombia y la experiencia internacional, Bogota, World Bank. Rubio Bonadilla, Jose Luis, Victor Cardenas, Bernardo Coutolenc, Rodrigo Guerrero and Maria Antonia Remenyi. 1995. "Medici6n de los costos de la violencia." OPS. Rubio, Mauricio. 1995. "Crimen y crecimiento en Colombia." Coyuntura Econ6mica, Vol. XXV N°l. 1999. Crimen e Impunidad Precisiones sobre la Violencia. Santafe de Bogota: Editorial TM. 1997d. "Los Costos de la Violencia en Colombia". Documento CEDE 97-07, Bogota. Steiner, Roberto. 1997. "La Economfa del Narcotrafico en Colombia" mimeo, Fedesarrollo, Bogota. 59 1998. "Colombia's Income from the Drug Trade", World Development, vol. 26, no.6, pp. 1013-1031. Trujillo, Edgar and Martha Badel. 1998. "Los costos econ6micos de la criminalidad y la violencia en Colombia: 1991-1996". Documento No 76, Archivos de Macroeconomia, Bogota: DNP. 1993. "Entrada de Capitales, diferencial de intereses y narcotrafico" in Garay, Luis Jorge, Macroeconomfa de losflujos de capital. Bogota: Tercer Mundo-Fedesarrollo-Fescol. World Bank. 1999. "Poverty in Cali, Colombia." Background Report for the Cali City Development Strategy and the World Development Report 2001. Jesko Hentschel, PRMPO. December 17, 1999- .2000. "Colombia Water and Sanitation Sector: Review and Strategy" Draft for Discussion, LCSFW, Country Management Unit 4, January 2000. 60 CHAPTER II POVERTY AND WELFARE IN RURAL COLOMBIA DURING THE LAST TWO DECADES Carlos Eduardo Velez, Benedicte de la Briere and Natalia Millan World Bank/ LAC/ PREM ABSTRACT Apart from the obvious impact of macroeconomic trends and trade reforms on development, rural Colombia has been exposed to two other major phenomena during the last two decades, both with increasingly detrimental effects: the internal arned conflict and the escalating influence of illicit drug trade. The former, aside from taking the lives of many, has induced the forced migration of nearly 10 percent of the rural population, putting an increasing number of vulnerable groups at risk of poverty. The drug trade, inextricably linked to the armed conflict, provides a key source of financial support for armed groups. Colombia is today the world's largest supplier of cocaine and coca has become its fifth most valuable agricultural activity. In many ways analogous to our study of urban Colombia (Chapter 1), this paper aims to understand social progress and the evolution of economic welfare, including poverty and inequality, in rural Colombia. We use Rural Household Surveys for 1978, 1988, 1995, and 1999. Despite the unfavorable circumstances mentioned above, the results shows that over time improvements in social indicators have exceeded the gains in economic welfare. Strong progress in access to social services -education, health care and infrastructure- are reducing the urban-rural gap. Poverty rates have also been remarkably abated; most of the staggering progress was made between 1978 and 1995, with a reduction in the headcount from 68 to 37 percent. In the late nineties, as economic growth slowed down , extreme poverty remained relatively stable, unemployment soared, and there was significant deterioration of the poverty gap and poverty intensity indicators (all despite moderate average income per capita growth). Furthermore, the latter growth in average income per capita, we do find that the bottom 70 percent of rural Colombians experienced a fall in income. Although the rural poor exhibit some similarities to those of urban areas -e.g., high dependency ratios and low skill endowments -, the rural poor are more difficult to profile with a set of household characteristics, since key determinants of rural output heterogeneity are missing -namely quality and quantity of land and local infrastructure. Our decomposition of poverty dynamics reveals that, contrary to in the urban case, economic growth is not a unique dominant factor. Growth played the main role in the steep poverty reduction between 1978 and 1988, but most of the 10 percent reduction from 1988 to 1995 was due to a fall in the relative price offood items and, in the subsequent period, increasing income inequality impinged upon alleviation of extreme poverty. Instead of the usual convergence in urban areas, social and economic indicators reveal crossing and deviating paths in regional development that are linked to international trends of key agricultural products -such as coffee-, trade protection policies, and urban development poles within each region. In this process, the clear winners are the Oriental and Atlantic regions; the losers are the Central and Pacific. 61 1. INTRODUCTION In this chapter, we analyze poverty, income inequality, and welfare in rural Colombia for the years 1978, 1988, 1995, and 1999. Rural Colombia, as defined by the National Household Surveys conducted by DANE, is comprised of dispersed zones of all municipalities, non-municipal cabeceras, and about 850 remaining municipal cabeceras classified as rural.65 DANE classifies its rural data into four regions: Atlantic, Oriental, Central, and Pacific, depicted in Figure 2.166; in 1999, they made up 25, 26, 28, and 21 percent of the total rural population, respectively. It is important to keep in mind that DANE's regional division was done purely for statistical purposes, and does not coincide with that used by CORPES67 for administration and planning purposes. Thus, the regions cover very diverse realities, in terms of spatial population distribution, socioeconomic status, production, and geography. Nevertheless, some generalizations can be made about the regions themselves. The Oriental region (26 percent cabecera) is mostly characterized by a minifindio agriculture with a recent increase in commercial agriculture; the coffee- growing Central region (28 percent cabecera) by highly-paid extended wage labor and better returns to land; the Atlantic (31 percent cabecera) by poor wages and less access to land; and the Pacific (21 percent cabecera) by a very high intemal heterogeneity, ranging from the relatively urbanized and developed department of Valle del Cauca, the bucolic (and minifundista) departments of Cauca and Nariiio, and the indigent and sparsely populated department of el Choc6 (L6pez et al., 2000). Throughout the last two decades, Colombia's rurral population, as a percentage of total population, has declined by more than 4.5 percentage points, mainly due to migration to urban centers. According to L6pez et al. (2000), rmigration from rural areas was responsible for 43 percent of urban population growth in the 1950s. Subsequently, this proportion declined to 37 percent in the 1960s, 27 percent in the 1970s, and 23 percent in the 1980s. In the 1990s, it increased back to 28 percent, a similar proportion to that of 2 decades earlier. Much of this increase may be the result of displacement due to the armed conflict that plagues rural areas (this phenomenon is discussed in more detail below). At the same time, the role of agriculture as the main source of income in rural areas has diminished, giving way to an increasing service and, to a lesser extent, mining sector." Nevertheless, agriculture continues to play a significant role in the economy: since 1990, it has represented nearly 13 percent of GDP, and has contributed to over a third of foreign exchange earnings and over 30 percent of employment (Jaramnillo, 1999). 65 The 1999 data allow for desegregation into cabeceras and dispersed zones. See Appendix for full description of rural data. 66 Note that the National Household Surveys do not cover the departments of Amazonas, Arauca, Casanare, Guainia, Guaviare, Putumayo, San Andrds, Vaupds, and Vichada, which together make up about 2.6 percent of the national population. 67 Consejo regional de planificaci6n econ6mico y social. 68 The service sector now makes up 45 percent of the rural economy (Leibovich er aL, 1997)). 62 Figure 2.1. Colombian rural regions as classified by DANE u E h D E R W , / ~~~~ARAUCA CASANA~~~~~~ASNRE/ s < YI~~~~~~~CHADA 9 PUTUZ YAU~~~~~~~~PES, Pacific Region AMAZONAS AtlanUc Region Central Region Odental Region 63 Organization of the paper. In general, our rural analysis is partially analogous to that of urban Colombia (Chapter 2) and is organized as follows. The introduction presents the macro-economic environment facing the rural sector in the past two decades and then briefly discusses the armed conflict and the related displacement and illicit crop cultivation. We then summarize our main findings. The first section discusses rural social progress. Next, we study the evolution of welfare, followed by a section on poverty dynamics. The fourth section examines the course of income inequality, and the fifth concludes. Where appropriate, we include analysis by each of the four regions mentioned above. Macro-economic and sectoral environment Even though Colombia successfully avoided the large mnacro-economic imbalances that plagued Latin America at the end of the 1970s, it suffered from the shortage of capital of the early 1980s. The previous prosperity resulted from careful management of the coffee revenues from the earlier decade. Thereafter, the eighties started with a strong economic slowdown; the growth rate fell from an average of 5.4 percent between 1975 and 1980 to only 1 percent in 1982. This was accompanied by a rising deficit in the balance of payments as well as in the public sector, and a financial crisis. Some banks collapsed and various financial intermediaries were nationalized. After the slowdown in the early 1980s, adjustment policies and the coffee bonanza of 1986 helped restore internal and external balances. To confront the crisis, the government reversed some of the liberal policies adopted previously, accelerated the devaluation, increased tariffs by 8 percent (average tariffs reached 34 percent in 1985), restricted public spending, instated a VAT tax, purged the financial sector and provided support for social housing to stimulate productivity. In 1986, the growth rate returned to over 5 percent, the current account balance was positive and the public spending deficit remained under control. lln the later part of the eighties, the economy diversiried away from coffee, unemployment fell and growth was robust. Exports rose from 15.6 to a high of 22.7 percent of the GDP between 1986 and 1991. Non-traditional exports increased from 32.1 to 49.1 percent of total exports during the same period, while oil, gas and other minerals also rose from 18.6 to 32.6 percent of total exports. Unemployment had peaked at 14.7 percent in 1986, and subsequently oscillated between 9 and 12 percent in the late 1980s (Nnfiez and Bernal, 1998). The growth rate also remained relatively high in the latter half of the eighties, averaging at an annual rate of 4.5 percent between 1986 and 1990. Tlhe early 1990s saw a whole array of liberalization policEes in trade and capital flows. Average tariffs fell from 44 percent in early 1990, to only 11.8 percent in March 1992 (Ocampo et al., 1998). The exchange rate quickly became overvalued, partly because of large inflows of capital -a mix of fresh investment in new oil fields, increased borrowing by firms and some repatriation of drug money- encouraged by the country's favorable investment grading. The strong peso and trade liberalization hurt farmers hard and pushed up prices of non-tradables -especially real estate assets. Moreover, the International Coffee Pact collapsed in 1989, leading to a substantial fall in coffee prices. At the same time, total public spending rose at a pace incompatible with sustainable fiscal policy. Up until the early 1990s, prudent management of the Colombian economy allowed for low government debt levels which, together with low inflation rates by Latin American standards, led to steady -although moderate- growth rates. However, public spending in the 1990s entered an increasing path that pushed its share from 24 to 36 percent of GDP between 1990 and 1998. This corresponds partly to constitutional reform on sub-national public finance -la descentralizaci6n-, which led to increasing transfers to local governments for social spending that were not accompanied 64 by corresponding cuts in the central government spending. In addition, public expenses in justice and security increased due to the escalation of crime and violence, and public sector pensions demanded increasing outlays -as they were excluded from the social security reform in 1993. The Government financed its budget deficit mainly through a combination of tax reforrns -including greater reliance on VAT and increased measures to reduce tax evasion- and financing through commercial bank credit. In addition, after 1995, Colombia increased its reliance on external financing through bond issues. Total gross external Central Government debt increased by 125 percent between 1994 and 1999, while internal debt increased by 260 percent in the same period. At this rate, the public deficit was heading toward unsustainability. Fiscal imbalances pushed Colombia towards slowdown after 1996, and into recession in 1998- 99. The fiscal position continued to deteriorate as econormic slowdown adversely affected tax revenues. In 1998, the current account and fiscal deficits both ran at 5 percent of GDP. The country lost its investment-grade rating, elevating its cost of credit even more. Government debt servicing costs rose from 2.2 to about 4.4 percent of GDP in 1994 and 2000, respectively. Moreover, when the credit bubble burst, property prices plummeted and several banks filed for bankruptcy; the Government was burdened with even more liabilities as it restructured the financial sector and provided mortgage debt relief. The January 1999 earthquake that affected Colombia's main coffee- growing region also led to unforeseen government spending. GDP in 1999 contracted by over 4 percent compared to 1998 and private investment fell to 4 percent of GDP, from 13 percent in 1995. Concentration of Land and Credit Land and credit: de-concentration started in the middle eighties. Ownership of land underwent a period of concentration from 1974 to 1984. Since then land concentration decreased, as the land- weighted Gini coefficient decreased from 0.61 to 0.59.69 In a period of subsidized interest rates and rationing of credit, credit was particularly concentrated among large-scale producers between 1974 and 1984, but has since then competitive interest rates tended to de-concentrate it (Leibovich and Ndfiez, 2000). Displaced population &nd armed conflict in the r-ural areas In the fifteen years spanning from 1985 to 1999, between 1.2 and 1.5 million Colombians have been forced to leave their homes due to the escalating civil war, with the numbers rising to 308,000 in 1998 and 288,00 in the first half of 1999.70 Most of them remain internally displaced people (hereafter IDP) in the medium and large cities of Colombia.7' The displaced include a majority of women and children, who fled as the threats to their safety became closer and closer, as well as an overwhelrning majority of rural households, whose main income came from agricultural activities. Forced displacement appears to be concentrated in a few departnents. Even though studies of IDP in Colombia are few and their results are not comparable, one can note that the main expulsion 69 Intemational comparisons show that in 1996 the concentration of land in Colombia was inferior to that of Czechoslovakia, Paraguay, Brazil, Argentina, and Panama, (Deininger and Squire, 1998, quoted by Leibovich and Nuifiez, 2000). 70 From CODHES Boletln no. 28, February 22, 2000, Bogota; quoted in Partridge, W.L. with J. Arboleda. 2001. "The Population Displaced by Armed Conflict in Colombia," mimeo, 24 pp. Colombia Country Unit, The World Bank. 71 On January 25, 1999, the Armenia earthquake in Quindfo affected this department, as well as Risaralda, Caldas, and Tolima in the Central region, and Valle del Cauca in the Pacific region, leaving 150 to 200,000 homeless and also causing some displacement. 65 areas are Antioquia (Central region); Santander and Meta (Oriental region); C6rdoba, Magdalena and Sucre (Atlantic region); and more recently Choc6 and Nariiio (Pacific region) as well as CaquetA (Central region) (Erazo et al., 2000). According to The Economist (2001), these departments are also the ones that receive the greatest amount of petroleum royalties and the ones in which drug traffickers acquired the most land during the 1980s.72 l[lDP either try to stay close to their land in the departmental "cabeceras" or they mnigrate to the safer anonymity of larger urban centers such as Bogota -which may receive up to 25 percent of the IDP-, Medellin, Barranquilla and Cartagena. These forced migrants, who are trying to escape the various armed groups, fuel urban growth. A survey of 200 LDP in Bogota, Medellfn and Cartagena suggests that households that opt to reach departmental urban or peri-urban centers may be undertaking "preventive displacement," as their environment becomes generally unsafe and they foresee an escalation of the conflict at the local level. Those who seek larger and more distant cities may be trying to escape direct threats to their lives, when fleeing is the sole option to remain alive. The latter are typically leaders in their community and/or depend on agriculture as their main source of income. They must often leave in a very quick fashion and the majority are forced to abandon their assets (Erazo et al., 2000). The increasing trend in the numbers of IDDP and tne limited absorption capacity of the urban sector, especially in the context of the recession, are puting an increasing number of vulnerable groups at rising risks. IDP face a high risk of unemployment, as agricultural skills have little value in urban labor markets, thus making their income generation opportunities quite limited. In addition, inadequate housing and poor access to public services characterize the urban slums or spontaneous settlements where they arrive. Major economic losses, tremendous psychological hardship, and the added strain on household structure - atomization of families, single headship, etc.-, puts them at risk of resorting to begging, prostitution and delinquency and falling into a vicious circle of extreme poverty. The well being of children of IDP households, present and future, is of special concern as the households' food security is endangered and the opportunities for school enrollment and attendance shrink. The fullh economic and social effects of considerable forced migration -approximately 110 percent of the rural population- are difficult to assess. One of them may be the concentration of landholdings in the areas of "expulsion" and the increasing weakness of property rights on any kind of real assets. In turn, the latter give rise to perverse efficiency effects on the remaining rural population, via its obvious discouragement of private investment. Another may be the creation of a new highly vulnerable group, composed of returnees or re-settled IDP, who may not be able to return to their original income-earning strategies, due to the loss of assets or household members. On the other hand, population loss may lead to local labor shortages in rural areas and thus, rising wages. Illcit Crops and Rural Development Colombia is now the world's largest producer of cocaine and has recently become a signicant supplier of opium poppy and heroin. Production of the former grew from 158 metric tons in 1995 to 326 in 1999'3, with area under cultivation increasing from 50,500 to 122,500 hectares in the same period (UNDCP, 2000). In addition, Colombia hosts most of the clandestine manufacturing 72 "A Survey of Colombia. Drugs, War and Democracy." The Economist. April 21, 2001. It is estimated that drug barons may have acquired over one million hectares in cattle farms and other estates at that time. 73 The figures were 34,00 has and 51 tons in 1988. 66 laboratories, which reportedly process 80 percent of the world's demand for cocaine.74 In 1994, coca became Colombia's fifth most valuable agricultural activity, behind cattle, poultry, coffee and sugar (Jaramillo, 1998). Production of opium reached 102 metric tons in 1998, placing Colombia as the world's fourth producer (3 percent of the world's production); the area of production increased from 5,200 hectares in 1995 to 7,400 in 1998. The value of the poppy harvest was similar to that of beans, cassava and sorghum in 1994. Colombia is also a minor producer of cannabis: it is the second largest Latin American producer after Mexico. Coca production tends to concentrate in (or near) the Oriental and the extreme south of the Central region. Typically the coca leaf production areas are located far away from urban centers and in non-traditional agricultural areas. Table 2.1 shows that most of the cultivated areas are in the departments of Putumayo, Caqueta, Guaviare, and Norte de Santander, and Bolivar, where production is rising. While drug eradication efforts have been successful in departments such as Caqueta and Guaviare, on a national level, coca plantation has not diminished. The coca-growing areas overlap quite well with the territories of the guerillas and paramilitaries. Table 2.1. Coca crop estimates by main crop areas (in hectares) Dept. 1991 1992 1993 1994 1995 1996 1997 1998 1999 Caqueta (Central) 8,600 8,400 9,300 11,700 15,600 21,600 31,500 24,000 6,800 Bolivar (Atlantic) 5,300 3,400 2,300 2,000 3,500 6,500 N. Santander (Oriental) 7,000 7,800 Guaviare 21,400 22,900 24,100 26,300 28,700 38,600 29,000 7,000 8,200 Putumayo 2,200 2,400 4,000 5,000 6,600 7,000 19,000 30,100 56,800 Others 6,600 17,400 TOTAL 37,500 37,100 39,700 45,000 50,900 67,200 79,500 78,200 103,500 Source: National Anti-narcotics Agency (2000) Apart from their link to violence, illicit drugs -mainly cocaine and its derivatives (basuco)- negatively affect health of the local population. Addiction rates in Colombia ranged from 5.6 percent of the population 12 and above for cannabis in 1998, 1.6 percent of the population 15 and above for cocaine and derivatives (basuco and crack) in 2000, and 0.3 percent of the population 15 and above for opiates in 2000 (IJNDCP, 2001). In South America, these rates are the second highest after Brazil for cannabis and after Bolivia for cocaine products, and the third highest for opiates, even if the latter rate is still moderate. All these rates are rising, partly because of the "dumping" policies practiced by traffickers on "local" markets in response to enforcement programs in consumer markets (I gram of cocaine cost US $214 in the United States, US $100 in Europe and only US $1 in Colombia in 1997). Addiction rates are linked to high HIV infection rates, as drug use leads to the spread of HIV through needle sharing, as well as a higher probability of engaging in unprotected sex. Illicit crop production has serious direct and indirect environmental consequences for farmers through slash-and-burn practices and chemical contamination. To escape detection, crop areas shift constantly and are responsible for serious slash-and-bum clearing of forests, especially in the Afro-Colombian and indigenous reserved territories. One hectare of coca bush requires clearing of 2.5 to 4 hectares of forest and generates two tons of waste by-products. Contamination may result from the chemicals used in the illicit processing labs, which are dumped in the waterways. Aerial spraying of coca fields has also been claimed to cause potential health and environmental problems - plus the destroying of legal crops-. 74 The UNDCP reports seizure of 470 labs in 1996 and 14 tons, 208 in 1997 for 41 tons of coca paste. For heroin, 81 labs were seized in 1997. 67 Although the split of profits in the illegal drug trade seems to favor the dealers in the consumer countries, repatriated drug money represents a substantiaR amount for the Colombian economy and contributes to undermnine the country's governance. The best estimates of repatriated drug earnings go from US$2.5 to 5 billion a year, equivalent to 2 to 4 percent of the GDP (Economist, 2001). In fact, drug retail prices in consuming countries are orders of magnitude higher than prices in the producing country: while Colombian 1999 farmgate prices were US$200 per kilogram for opium, and US$950 for coca base75, the average retail prices in the United States that year ranged from US$240 per gram for heroin, US$67 for cocaine to US$10 for herbal cannabis. Part of the repatriated drug trade profits might help to improve the balance of payments and aggregate demand, but a substantial share goes to black market activities. Most profits may go into financing armed groups - weapons trade-76 or corruption of government officials7, leading in any case to a severe weakening of the rule of law and to the intimidation of the justice system. As seen above, the strengthening of armed groups and intensification of rural violence, along with the thriving kidnapping business, are directly affecting farming families. Other investments out of drug profits include cattle ranching and luxury housing, with low impact on rural sector productivity. Poverty serves to fuel the illicit production. At the other end of the spectrum, farmers often grow coca as a way to try to escape poverty. Farmers are attracted by the high profit margins and the availability of working capital and inputs, generally supplied by buyers. Even then, coca-producing areas remain among the poorest rural areas (Wagner, 2001). In addition, there is little public investment but for law enforcement, as the armed conflict and the drug eradication efforts drain government resources. Alternative development strategies have lagged behind in the eradication components of Plan Colombia. A note of caution. The reader should be aware of some limitation of this study of rural poverty. First, it is not a study of rural development; hence our characterizations of rural poverty might be incomplete or biased. In a typical rural household that employs its labor force in agricultural activities, welfare not only depends on its socio-demographic characteristics (human capital endowment, demographic composition, etc.) but also -and very importantly- on the characteristics of the environment (quality of soil, availability of water, infrastructure facilities, etc.). Most of the latter set of characteristics is not observed in the household surveys available to monitor the evolution of rural poverty and this paper not attempt to establish its influence at the regional or sub-regional level. In particular, the household surveys also do not cover ownership of land and access to credit, which a have significant influence on income generation in rural areas. The concentration of land in Colombia -although it improved between 1985 and 1996- is higher than that of Czechoslovakia, Paraguay, Brazil, Argentina, and Panama. Access to credit was particularly concentrated among large-scale producers between 1974 and 1984, but has since then tended to de-concentrate (Leibovich and Nuiiez, 2000). These phenomena, as well as that of illicit crop cultivation, armed conflict and displacement, may drive or influence some of the results we obtain in the remainder of the analysis; however, the data used in this report do not allow to account for it. 75 Information about coca leaf farmgate prices is not available in the 2000 UNDCP report for Colombia but reported prices are US$ 2.69 in Bolivia and US$ 2.13 in Peru. 76 According to Colombian government, drug rents represent nearly half of total guerrilla earnings; the rest come mostly from kidnapping and extortion. 77 Colombia's Opacity Index (Price Waterhouse) for 2000 is 60, the mean among 35 countries and the 15'h overall ranking, with the rankings for the five areas of evaluation as: economic policies 24h', legal protection for business 21t, corruption 17'h, government regulation 13'h and corporate accounting I I'h. httqp:Hwww.opacityindex.com. In contrast, Transparency International ranked Colombia as the 5th most corrupt in Latin America in 1999 (after Honduras, Bolivia, Ecuador, Venezuela). http://www.gwdg.de/-uwvw/ 68 Organization of the paper. The paper is divided into three sections after this introduction. The next section briefly presents the evolution of social indicators in education, child labor and access to basic infrastructure. The third section finds out to what extend rural Colombian's welfare has improved during the last two decades. We examine the welfare evolution with three alternative social welfare measures: income per capita, income per capita corrected by inequality (the Sen welfare index) and poverty indexes, and full income distribution rankings (first and second order stochastic dominance). The fourth section focuses on the individual and household characteristics of vulnerable groups. Apart from establishing the faces of the poor in each period, we assess the marginal impact of different income generating factors on the probability of being poor. The fifth section decomposes the dynamics of poverty in terms of macro determinants: growth, inequality and relative prices of basic food items. MAIN FINDINGS Strong improvement in access to social services Substantial improvements in education are reducing the urban-rural gap. From 1978 to 1999 illiteracy was halved, and enrollment rates for children of primary and secondary had considerable progress and reduced their gap -by 80 and 50 percent, respectively- relative to urban areas. However, despite great improvements in school attainment, the urban-rural gap remains unchanged. Illiteracy and school enrollment rates converge by region, except for secondary education in the Atlantic region. Improvements in health coverage were satisfactory as reported in Chapter 6. Child labor participation rates, while remaining very high compared to urban levels, show a decreasing trend and are pro- cyclical. In contrast to urban areas, , basic infrastructure coverage decreased slightly from 1995 to 1999, especially in the Pacific region. (Infrastructure data are only available since 1995.) Utility coverage lags significantly behind that of urban areas, although in rural municipal cabeceras, it is similar to that found urban centers. Rural welfare improves until the mid nineties Average monthly household income per capita grew dramatically in the first sub-period, relatively slowly between 1988 and 1995 and continued rising during the economic downturn of the late nineties. By region, the clear winners are the Oriental and Atlantic; the losers are the Central and Pacific. Income inequality fluctuated during the last two decades and its recent increase diminished the welfare gains from average income. After strong improvements up to 1995, extreme poverty stabilized; nevertheless, the poverty gap and poverty intensity indicators have been simultaneously deteriorating. Surprisingly, poverty in the Atlantic region did decline during this sub-period. 1995 is a better year welfare wise than 1978 and 1988; it is better for any welfare measure, except income per capita. Poverty determinants and faces of the poor The household types that face the largest relative risks of poverty are large, linked to the agricultural sector, and have low skill levels and low employment rates. Differences in vulnerability are most staggering for levels of education of both household heads and other household members in working age. Vulnerability does not vary greatly for household heads of different ages. The minimum employment rate and schooling required to escape poverty are increasing over time. The most vulnerable groups in terms of poverty also include households with recent migrants or those that are female-headed. As in the urban areas, self-employment, recent job loss and head's disability is associated with poverty. Pensions provide effective protection from poverty. Homeownership has a moderate effect on poverty outcomes -although smaller than in the urban case. In spite of being lower 69 than in urban areas, unemployment rates escalated during the recent economic slowdown. Unemployment most strongly hit younger individuals with secondary education, as well as college dropouts. Regionally, the Pacific and Oriental regions suffered the highest increases in unemployment, closely followed by the Central region. The Atlantic region, which had the highest unemployment rate in 1988 and 1995, is now faring better. Extreme poverty dynamics: growth is crucial Decomposition of poverty dynamics shows that during the 1980s, as in the urban case for poverty, economic growth played a dominant role in the steep reduction in extreme poverty.' In the 1988-95 interval, the 10-percent decrease in extreme poverty came mostly from the fall in food prices and to a lesser extent from growth and better income distribution. In the late nineties, the lack of significant growth and the severe worsening of the income distribution made extreme poverty alleviation come to a near halt. Regional comparisons: Central and Atlantic regions win, Pacific and Central lose Social indicators reveal deviating paths in regional development. Primary school enrollment rates have converged; the Pacific region made the most progress throughout the whole period but during the economic downturn, primary school enrollment rates dropped in both the Pacific and Oriental regions. The Atlantic and Central regions were not negatively affected. Secondary school enrollment rates have also converged, with the exception of the Atlantic region, which has always enjoyed considerably higher rates. During the last four years, all regions, except for the Pacific, experienced large drops in child labor. Utility coverage in rural municipal cabeceras is comparable to that of large urban centers. Welfare indicators by region show divergent and crossing paths in rural regional development. Robust growth in the Oriental region reflects the dynamism of Bogota as an urban development pole. Once the richest of the four rural regions, the Central region now ranks third. The stagnation of the Central region between 1988 and 1999 corresponds relatively closely to the sluggish dynamics of the coffee sector in the last decade. While the Atlantic region caught up with the rest of the rural sector in the late nineties, the Pacific region is losing some of the substantial gains obtained up to 1995. In poverty measures the Pacific and Central fare worst, while the Atlantic shows continuous sharp decrease in the extreme poverty rate. In the late nineties, poverty measures of the Central and Pacific regions surpassed 1988 levels. Meanwhile, progress slowed down dramatically in the Oriental region. 2. SOCIL PROGRESS IN RURAL COLOMBIA 2.1. Schooling and education: persistent progress with regional convergence llliteracy was halved between 1978 and 1999, but recent progress has been slow. Table 2.2 reveals that in 21 years, the illiteracy rate for people 12 years and older fell by well over 50 percent (from 29.4 percent to 14.8 percent), with most of this progress occurring between 1978 and 1988, when illiteracy fell by 10.9 percentage points (versus 4.0 percentage points in the following 7 years). At first glance, it is worrisome that in the past four years, the illiteracy rate underwent a slight increase of 0.3 percentage points.78 However, such is not the case for illiteracy measured for the population aged 7 and to 19: this rate decreased by almost 4 percentage points, revealing a more 78 Given its size it could be due to measurement error alone. 70 sobering picture. Nonetheless, rural illiteracy still remains high, and at 14.8 percent, it is over five times as high as the urban rate -2.6 percent. Table 2.2. Social indicators, Rural Colombia 1978-1999 1978 1988 1995 1999 Illiteracy rate* 29.4% 18.5% 14.5% 14.8% Average education > 18 years 2.18 3.74 4.31 4.63 School enrollment Ages 7 to 11 66.2% 85.4% 90.1% 90.5% Ages 12 to 17 43.5% 57.2% 63.7% 66.0% Ages 18 to 22 9.0% 14.6% 19.2% 20.6% Incomplete primary school (ages 12 to 17) 79.3% 50.7% 37.3% 33.2% Incomplete high school (ages 18 to 22) 98.7% 90.9% 85.8% 78.9% Child labor Ages 10 to 16 25.5% 27.9% 22.3% 19.3% Ages lOto 14 19.1% 23.1% 15.6% 13.4% Access to public utilities Electricity NA NA 87.1% 84.0% Aqueduct NA NA 64.2% 61.7% Telephone NA NA NA 15.4% Sewerage NA NA 32.3% 31.5% * For population 12 years old & older. Average school attainment for adults more than doubled, from 2.2 years in 1978 to 4.6 in 1999, but the urban-rural gap remains unchanged (Table 2.2). Tantamount to improvements in illiteracy rates, most progress in adult average education levels was made between 1978 and 1988, when adults gained an average of 0.16 years of schooling per year. Progress since 1988 has remained stable, with an average increase of 0.08 years of schooling per year. Throughout the period studied, there has been a constant (little over) 4-year gap with urban education. In contrast with urban areas, school enrollment rates remained constant during the recessive period, if not rose, for all age groups. Despite the general slowdown in improvement rates, considering that urban areas suffered drops in enrollment rates for all age levels from 1995 to 1999, the economic recession does not appear to have affected rural schooling as severely. Nevertheless, all schooling rates continued to lag behind those in urban areas. Enrollment rates for children of primary and secondary school age registered their most significant gains between 1978 and 1988, followed by moderate gains between 1988 and 1995 and only a slight improvement in the 1995-99 period. The 1999 primary school enrollment rate - 90 percent- is now only 5 percentage points lower than its urban counterpart, an impressive feat, especially when compared to the 1978 urban-rural gap of 26 percentage points. Children of secondary school age are also increasingly attending school: 66.0 percent in 1999 versus 43.5 percent in 1978, and the urban-rural gap was halved, from 32 to 16 percentage points in 1978 and 1999, respectively. 71 The overall improvement in enrollment rates for the 18-to-22-age group is notable, as it has doubled in the past 21 years, up from 9.0 percent in 11978 (versus 31.2 urban) to 20.6 percent in 1999 (versus 36.3 urban). This enrollment rate progressed at similar rates between 1978 and 1988 and 1995, but then stagnated during the recessive period. In contrast, the urban rate suffered a decline of almost 5 percentage points between 1995 and 1999. Regional Comparisons All four regions depicted similar patterns of reduction in illiteracy rates, with the most advancement made between 1978 and 1988, moderate gains between 1988 and 1995, and relative stagnation (or worsening) in the last period studied (see Table A.4A in the appendix). Rates were comparable in the Central, Oriental, and Pacific regions throughout the years studied -except for the Pacific's rate in 1978, which was over 10 percentage points above than the rest. It is actually noteworthy that illiteracy rates in the Pacific region fell by well over 50 percent in the first 10-year sub-period -from 34.2 to 15.6 percent-. Atlantic rates always remained significantly higher, but the gap has been narrowing over time. ID¢lieracy and school enrollment rates converged by region, except fTor the Atlantic. The latter, while following roughly the same trends, remains behind the others. Nonetheless, progress made in the Atlantic region is still notable, as illiteracy fell to 20.9 percent in 1999, from 46.1 percent in 1978, although still almost 6 percentage points higher than the rural average. Despite this contrast with the other three regions, school enrollment levels have remained close to the rural average for primary education, and higher than average for secondary and tertiary education in all periods. Therefore, higher than average illiteracy rates may point to a relatively large stock of older, illiterate people in the Atlantic region. In fact, when illiteracy is measured for those aged 7 to 19, the Atlantic leads the average rate by less than 2 percentage points. Primary school enrollment rates have converged; the ratio between the highest and the lowest rate fell to 1.05 in 1999 from 1.37 in 1978. The Pacific region made the most progress throughout the whole period -after having had the lowest rate in 1978, its 1999 rate now marginally surpasses the other 3 regions. During the economic downturn, primary school enrollment rates dropped in both the Pacific and Oriental regions. The Atlantic and Central regions were not negatively affected: rates remained steady in the Atlantic region, and rose a full 4 percentage points in the Central region - where secondary and tertiary enrollment rates also increased-. Secondary school enrollment rates have also converged, with the exception of the Atlantic region, which has always enjoyed considerably higher rates, around 12 percentage points higher than the rural average. Interestingly, tertiary school enrollment rates were also always highest in the Atlantic region, yet primary school enrollment rates were the lowest out of the four regions in 1988, only surpassed the Central region in 1995, and were close to average in 1999. The education system in the Atlantic region thus seems somehow better able to retain students after they reach secondary school age. In 1999, enrrollment rates for those aged 18 to 22 still differed across regions, although they have converged relative to 1978; the ratio between the highest and the lowest rate fell from 2.64 to 1.43, in 1978 and 1999, respectively. The Central region always lagged behind the other three until 1995: while the other regions experienced increases of 5 percentage points or more between 1988 and 1995, rates only increased by 2.6 percentage points in the Central region. In 1999, the Central region finally began to catch up. On a positive note, only the Pacific region experienced a weak fall (2.1 percentage points, not significant at a 95 percent confidence level) in tertiary school enrollment rates during the recessive period. 72 2.2. Child labor: a declining trend with some regional disparities over the business cycle Just as in our urban findings, child labor participation rates peaked during the 1988 economic boom and have been falling thereafter. However, in contrast with urban areas, the decrease was most substantial between 1988 and 1995, and then slightly less so between 1995 and 1999. The 1988 boom was largely driven by the exceptionally high coffee prices and the good weather, which led to an increase in the demand for agricultural labor. Thereafter, 1995 saw the beginning of the construction crisis. Considering the large rise in adult unemployment that occurred during the economic downturn (see Table 2.11), it is surprising that child labor participation rates have not declined to a larger extent. Perhaps the fact, as will be seen later, that rural regions have not experienced a decline in mean household income per capita -except for the Pacific region- may explain why the recent fall has only been in the rragnitude of around 3 percentage points. Despite their decline, rural child labor levels remain very high when compared to levels in urban areas. Regardless of the 30 percent drop in child labor between 1988 and 1999, child labor rates remain well above urban areas, at 19 percent for those 10 to 16 years of age, versus 9.5 percent in urban areas, and 13 percent of the 10-14 year-old group employed vs. 3.7 percent in urban areas.'9 About 31 percent of children 15 to 16 participated in the labor force in 1999, a rate just under half of the adults, -65 percent of those 17 to 65 years old. During the last four years, all regions, except for the Pacific, experienced large drops in child labor. The Central and Oriental regions had similar evolutions from 1995 to 1999, down from (about) 28 to (about) 21 percent for the 10-to 1-6-age category and down from (about) 20 to (about) 14 percent for the 10-to-14 age-category. The Atlantic region's decline was relatively small, but the region already had especially low child labor rates in comparison to the rest; rates in 1999 were at only 11.6 and 7.6 percent for children 10 to 16 and 10 to 14, respectively. In contradistinction with the rest of the rural areas, child labor was counter-cyclical in the Pacific region: itfell during the economic boom in 1988 and rose during the economic downturn between 1995 and 1999.8 One fourth of children between 10 and 16 years of age were economically active in 1999, up from 18 percent in 1995. The increase for the 10-14 year-old group has been even sharper (19 percent in 1999 vs. I1 percent in 1995), a worrisome evolution when considered together with the decline in school enrollment for that age group. It is interesting to note that in Cali -situated in the Pacific region- child labor force participation also increased (dramatically) during the recessive period. While this seems, at first glance, to contradict the urban findings of Cunningham and Maloney (2000)8', these authors also find that "if the husband or wife opens a business, the child is more likely to work." The rural Pacific region has seen an increase in the proportion of household heads in agriculture during the recession, which could mean increased involvement in smnall famnily farms, a business where captive farnily child labor is highly valued (lower supervision cost, flexibility, etc.) when wage labor is not affordable. 79 Child labor participation rates in urban areas are for children 12 to 14 and 12 to 16. Thus, the urban-rural difference in child labor participation rates may be even larger. 80 As will be seen later, rural Colombia as a whole did not experience a fall in income per capita between 1995 and 1999, but the Pacific region did. 8' As stated in the urban chapter, Cunningham and Maloney (2000) explain that child labor is pro-cyclical, rising during economic booms and falling during recessions, due to, among other things, changes in the opportunity cost of child labor and employment prospects, especially when parents start up micro-enterprises. 73 2.3. Basic infrastructure services (1995 and 1999):82 Access in rural cabeceras is comparable to urban centers In contrast with urban areas, service coverage decreased slightly from 1995 to 1999. Coverage for electricity, aqueduct, and sewerage decreased between the two years for which data are available (1995 and 1999). The largest fall (3 percent) was in electricity, access to which only 84 percent of the rural population had in 1999. Access to water followed with a 2.5 percent drop. This should reflect the severe reduction in the pace of new housing construction in the late 1990s (See Box 1.1 in Chapter 1). Utility coverage in rural municipal cabeceras is comparrable to large urban centers; however, notable disparities remain with sparsely populated areas. Economies of scale in the provision basic public utilities in urban concentrations facilitate greater coverage with comparable unit costs. For 1999 -the year for which data are available-, in areas labeled as municipal cabeceras, electricity and aqueduct coverage rates -both at 97 percent- are comparable to those of urban areas, whereas the corresponding figures in sparsely populated areas are 19 and 49, respectively. A look at access to public utilities by region between 1995 and 1999 shows a mixed picture. The Pacific and Atlantic regions had similar evolutions, but the Atlantic region fared better. Sewerage coverage improved slightly in both regions, by 1.3 and 2.3 percentage points, respectively. TDhe Paciffic region experienced a sharp decline in both electricity and aqueduct access (6.7 and 9.0 percentage points, respectively), while in the Atlantic region, this drop was smaller, at 5.0 and 2.7 percentage points, respectively. Although there was a slight decline in electricity coverage in the Oriental region, aqueduct and sewerage coverage increased by 10 and 8 percent, respectively. Meanwhile, the Central region remained steady in terms of access to electricity, yet experienced a 10 percentage point drop in both aqueduct and sewerage coverage. 3. EcONOMiC WELFARE, POVERTY AND INCOME DISTRI[BUIYON How has welfare in rural ¢Colombia evolved during the last two decades? Are rural Colombians better off now than two decades ago? Much like we do in the urban chapter, we assess the evolution of economic and social welfare in rural Colombia from 1978 to 1999 using a set of alternative indicators that place varying emphasis on the social weight of income groups. The analysis is restricted to individual welfare measured by monthly household per capita income, and attempts to quantify the levels of welfare by introducing different measures with increasing level of complexity: (i) average (and median) income per capita;83 (ii) inequality corrected income per capita (Sen welfare index);84 (iii) poverty measures; and (iv) full income distribution comparisons. Measures of social welfare and distributional weights. In principle, social welfare should be a function of the welfare of all individuals. In its simplest form, social welfare can be measured through average household income per capita, providing that "more is better than less," and that resources are measure with respect to needs, i.e. in per capita terms. Because this measure of welfare is independent of the distribution of income, a society in which one household holds all the resources 82 Data on rural infrastructure services is only available for those two years. 83 Due to unavailability of local cost of living data, we were not able to compare relative welfare among regions, as was done for urban data. 84 Unlike in our urban analysis, we do not use the Paglin index here because of lack of a discernable income inequality pattern across age groups. 74 is just as well off as one in which the same amount of resources are equally distributed amongst all households. For this reason, we also examine the Sen welfare index, which combines average income per capita with a measure of inequality -in this case the Gini coefficient. Next, we measure welfare by poverty measures; this is equivalent to concentrating all the distributional weights of the social welfare function in the lower tail of the distribution of income and ignoring income variations among the non-poor. Social welfare rankings independent of distributional weights. The three measures of social welfare proposed above incorporate a specific vector of social weights to households, according to their relative position within the distribution of income. Therefore, our assessment of the evolution of welfare over time might produce contradictory statements, depending on the type of indicator -and the implicit distributional weights- chosen. It is, however, possible to make unambiguous statements about two welfare distributions, independently of value judgments. As in the urban case, we compare full income distributions requiring minimum assumptions about social welfare in relation to income,85 and when first order stochastic dominance and/or generalized Lorenz dominance hold, unambiguous statements about social welfare comparisons over time can be pronounced. In addition, a powerful corollary: when these dominance measures stand, all three previous measures of welfare -average income per capita, income corrected by inequality and poverty measures- must necessarily move in the same direction. 3.1. Poverty Measures: clear progress in the 1980s plus smaller improvements, if not, deterioration, in the late 1990s After a substantial reduction in the eighties, the poverty headcount ratio has remained relatively stable since 1988 (see Tables 2.3 and 2.4). Similarly to iurban areas, rural poverty fell sharply (14 percentage points) between 1978 and 1988.86 On the other hand, in contrast with urban areas, where poverty dropped by almost 4 percentage points between 1988 and 1995, rural rates only saw a decline of, 1 percentage point. (Recall that while urban incomes grew at an average annual rate of 2.9 percent in the 1988-95 period, this figure was only 0.7 percent in rural areas.) Nonetheless, during the economic downturn, rural areas were not as negatively affected, as mean income per capita continued to grow and poverty increased by only 0.2 percentage points, versus a 3 percentage point increase in urban areas. 85 Basically, social welfare must be a monotonic non-decreasing concave function of individual income in the case of Lorenz dominance. This can be partially relaxed for first order stochastic dominance. For detailed explanations, see Deaton (1997), chapter 3, Champernowne and Cowell (1998) chapters 4 and 5. 86 The urban poverty rate fell by 15 percentage points during this same period. 75 Table 2.3. Income inequality and poverty indicators, Rural Colombia, 1978-99 1978 1988 1995 1999 Poverty Poverty rate 94% 80% 79% 79% Poverty Gap 61% 43% 40% 44% P(2) 44% 29% 25% 29% Extreme poverty rate 68% 48% 37% 37% US$ 2 per day poverty2 73% 50% 4% 42% Household monthly income per capita I Median 36,847 62,764 67,692 65,000 Mean 51,980 89,613 94,851 101,630 Income inequality Gini 44.8% 46.5% 44.6% 50.2% Entropy Measure EO 0.36 0.40 0.36 0.46 Entropy Measure El 0.40 0.44 0.42 0.52 Entropy Measure E2 0.85 1.00 1.01 1.22 P90/PlO 7.4 8.8 7.2 9.7 P75/P25 3.0 3.0 2.7 3.0 Share q5/Share ql 10.2 12.3 10.2 14.4 1. 1999 pesos 2. Based on PPP Convertors from WDl data base. Table 2.4. Change in poverty and extreme poverty rates, Rural Colombia, 1978, 1988, 1995, 1999 1978-88 1988-95 1995-99 Poverty Extreme poverty Poverty Extreme poverty Poverty Extreme poverty Rural Colombia -14.0% -20.3% -1.0% -10.3% 0.2% -0.5% Atlantic Region -5.1% -5.5% -3.0% -11.7% -5.2% -11.9% Oriental Region -12.5% -20.4% -8.5% -20.3% -1.0% -3.1% Central Region -21.2% -24.0% 6.3% -4.6% 1.7% 4.9% Pacific Region -15.4% -29.1% 0.8% -3.4% 5.0% 6.7% 1. Level change in percentage points Not unlike in urban areas, extreme rural poverty declined much faster up until 1995. Both the extreme poverty rate and the US $2/day poverty rate declined much faster than the poverty rate during the first 2 sub-periods, a phenomenon not unlike that found in urban areas. Thereafter, these rates remained relatively stable, just as the poverty rate did. While the US$2/day poverty rate fell by an astonishing 50 percentage points between 1978 and 1988, extreme poverty rates fell by a 76 nevertheless substantial 20 percentage points.87 Between 1988 and 1995, when poverty was reduced by only I percentage point, extreme poverty and the US $2/day poverty rates declined by 10 and 7 percentage points, respectively. Both of these rates have recently stagnated, with the extreme poverty rate slightly decreasing, and the US $2/day poverty rate slightly increasing between 1995 and 1999. Relative stability of poverty counts hides recent deterioration of the poverty gap and poverty intensity (P2). As described for certain regions, the recent relative stability of the poverty count as measured by the three different poverty lines hides a widening of both the poverty gap and the P2 index in the last 4 years. These two indicators had progressed faster than the poverty count until 1988, but since then the poverty gap and P2 index have risen by 3.2 and 3.5 percentage points respectively (see Table 2.3). Given that most of the population lies below the poverty line, we can conclude that much of rural Colombia has suffered the negative impact of the recession. Most worrisome is the larger rise in the P2 index, implying that the brunt of the recession is being carried by the very poor. This is also indicated by the fact that the US$2/day poverty rate has increased slightly, although the extreme poverty rate has declined. Regional Comparisons: strong heterogeneity with problems concentrated in the Central and Pacific regions88 The Pacific and Central fare worst, while the Atlantic improves in all indicators The decrease in poverty rates of the 1978-95 sub-period was concentrated among the extreme poor, especially in the Oriental region, where extreme poverty rates went from 75.4 percent to 31.7 percent (Appendix Table A.5). Substantial improvements in poverty indicators came to a halt after 1988 for both the Pacific and Central regions. In fact, the Central region actually registered a 6.3 percentage point rise in the poverty rate between 1988 and 1995.89 During the 1995-99 period, only the Atlantic region registered a continued sharp decrease of the extreme poverty rate -an 11.1-percentage-point drop-. The poverty rate fell by 5 percentage points, enough to bring the Atlantic region down to the average rural poverty rate. (Except for in 1978, poverty rates in the Atlantic region had always been above the national rural average.) This region was also the only one to undergo a drop in the poverty count, poverty gap and P2 index, as depicted in Appendix Table A.5. The drop in the poverty gap was the largest, followed by the drop in the poverty count and the P2 index. Meanwhile, progress slowed down dramatically in the Oriental region. During the latest sub- period, the Oriental region has followed a pattern similar to that of rural Colombia as a whole: while the poverty count slightly decreased, the other two indicators -PI and P2- have risen, indicative of a deepening of poverty. 87 Such a large drop in the US$ 2/day poverty rate is partly due to changes in the purchasing power of US$ 2: in 1978, the extreme poverty line, which measures food item prices, was US$ 1.43, whereas in 1988, it rose to US$ 2.80. In 1978,51 percent of the population lived below US$ I a day, a proportion more comparable to the 1988 figure of 32 percent under US$ 2 a day. 88 The same poverty line, designated as rural, is used for all four regions. Thus, we do not differentiate for differences in cost of living between regions. 89 Recall that mean income per capita also declined in the Central region during this period. 77 IBn the late nineties, poverty measures of the Central and Pacific regions increase above 1988 levels. The Central and, especially, the Pacific regions fared the worst between 1995 and 1999 and have reached poverty indicators inferior to those of 1988, erasing some of the progress made between 1978 and 1988. Most of the deterioration is in the poverty gap and the P2 index, once again indicating a deepening of poverty. In the Central region, the poverty count increased between 1988 and 1995 and stabilized thereafter, while the poverty gap and poverty index have both risen by almost 7 percentage points, compared with a rise 2.4 and 0.9 percentage points, respectively in the previous period. The increase in poverty among the poorest of the poor is also visible in the extreme poverty rate, which had been falling until 1995, but increased back to 1988 levels thereafter. The Pacific region, which displayed very stable numbers during the 1988-95 period, is seeing a quick deterioration of all the poverty indicators. The worsening is more pronounced in the poverty gap and P2 index, both rising by about 9 percentage points, but the 5 and 7 percentage point rise in the poverty and extreme poverty counts, respectively, are also not negligible. 3.2 Average household income per capita: strong improvement in the 1980s, slowdown in the early 1990s and some recovery after 1995. Regional discrepancies. Mean household income grew dramatically in the first sub-period, followed by a relative slowdown between 1988 and 1995 (Table 2.3). The rise in mean household income per capita between 1978 and 1988 was of 72 percent, rising at an average annual growth rate of 5.6 percent per year, faster than the growth in urban areas (4.1 percent). 1988 was an exceptional year across all rural regions, with agriculture playing a major role: the devaluation helped agricultural exports, which were further supported by high tariffs for cereals, and coffee production and prices remained high. Later, mean income rose at the much slower rate of 0.7 percent a year between 1988 and 1995 (versus 2.9 percent in urban areas), but continued rising between 1995 and 1999, this time at the faster average annual rate of 1.4 percent (versus minus 1.2 percent in urban areas). This overall evolution hides striking differences between regions and income groups, as will be discussed later. Mean household income in rural Colombia did not suffer during the recessive period. As opposed to the fall experienced in urban areas, mean (monthly) household income per capita rose during the economic downturn; however, median household income per capita declined. After behaving (rising) much like average mean income, the median income fell at an average annual rate of 0.8 percent during the recessive period. The fall in the median income and the rise in inequality in 1999 indicate that portions of the population were indeed negatively affected by the recession. Over the years, the rural-urban gap fluctuates and falls in the late nineties. It is also notable that throughout the 21-year period studied, rural average income per capita remained at roughly one third of that in urban areas. However, urban-rural disparities in economic growth produced fluctuations in the relative income ratio (Table 2.3). Despite impressive rural growth between 1978 and 1988 that brought relative income to a qualified maximum of 38 percent, the better performance of the urban sector up to 1995 reduced rural Colombia's relative income to a minimum of 32 percent. This was then partially reversed by negative urban growth in the late nineties, bringing rural relative income to 37 percent in 1999. Regional discrepancies Divergent and crossing paths in rural regional development. Regionally, mean income per capita followed very different paths: trends and relative positions changed throughout the periods9o (see Table 2.4 and Figure 2.2). Different factors -coffee sector dynamics, trade liberalization and export 9' We are grateful to Carlos Felipe Jaramillo at the Colombian Trade Bureau, for sharing some of his insights with us. 78 crop shifts, urban-rural integration, and the armed conflict - played different roles across all regions, leading to starkly diverse pathways. Moreover, the differing economic dynamics of the main regional urban centers (see Chapter 1) and significant migration to urban centers have also affected rural Colombia. Figure 2.2. Mean household income per capita by region, Rural Colombia, 1978, 1988, 1995, 1999 120,000 - 110,000 - 100000 a. 90,000 o w| 80,000- K E~ 0. o 70,000 0 60,000- - -rCental 0 C wa 50,000 , , Pacific 40,000 1978 1988 1995 1999 Year Mean income per capita in the Oriental region grew 175 percent throughout the last 2 decades.9' The Oriental region, the poorest of the four in 1978, almost doubled both its mean and median income per capita between 1978 and 1988. In the next seven years, it also experienced remarkable growth, as mean income per capita rose by 24 percent, a staggering rate when compared to the other three regions during this sub-period.92 The Oriental region then grew by 15 percent between 1995 and 1999, making it the richest region in 1999. Robust growth in the Oriental region reflects the dynamism of Bogot&i as an urban development pole. The strong economic growth, and the very low unemployment of Bogota up to 1995 - the lowest among the 7 cities, then-, should have influenced growth in the Altiplano Cundi-boyacense, pulling up rural wages by induced migration of unskilled workers to the unprecedented construction sector boom.93 In addition, the demand for labor in the drug cultivation sector might have influenced the labor markets in Meta -located in the Oriental region-, and Arauca and Guaviare -not included in National Household Surveys. Wages in these departments have traditionally been higher and immigration to this sparsely populated area increased during the 1990s (Jaramillo, 1999). 91 Total growth between 1978 and 1999 was 84, 66, and 65 percent in the Atlantic, Pacific, and Central regions, respectively. 92 In contrast, the average rural mean income per capita rose by only 6 percent between 1988 and 1995. 93 Sanchez and Nuifiez' (2000) rigorous and comprehensive study on Colombian regional economic development highlights the strong influence of large local urban markets on regional development. They study Colombian municipalities for the 1973-95 period and measure the regional influence of large urban markets on income per capita growth. 79 Once the richest of the fouir rural regions, the Central region now ranks third. The Central region also experienced a large increase (though not as substantial as the Oriental or Pacific regions') in mean and median income per capita in the first ten years studied. However, mean income per capita in the Central region suffered between 1988 and 1995. Both mean and median income per capitafell by 8 percent in the next 7 years (1988-95). Subsequently, the median continued to fall (by 11 percent) and the mean rose by only 6 percent from 1995 to 1999. The stagnation of the Central region between 1988 and 1999 corresponds relatively closely to the dynamics of the coffee sector. As Ocampo (1989) points out, the coffee cycle is very closely tied to the macroeconomic performance of coffee-growing regions. The disintegration of the International Coffee Pact in 1989 led to plummeting coffee prices (see Table 2.5), and the exchange rate appreciation also contributed to the reduction of coffee growers' earnings.94 In addition, several drought years and the broca disease contributed to the agricultural crisis, as well as the January 1999 earthquake in the coffee-growing departments of Quindio, Caldas, and Risaralda, which damaged some of the local plantations and left thousands homeless.95 Table 2.5. Colombian coffee prices*, 1978-200l Internal Federation International price market price 1978 4,795 2,047 1988 4,121 1,642 1995 2,628 1,371 1999 2,033 1,198 2000 1,871 1,101 2001 1,433 916 * All prices are in 1999 pesos per pound; 1978-2000: average September price; 2001: average February price. Source: Federaci6n Nacional de Cafeteros de Colombia Additionally, Huila and Tolima are Central region departments where cotton, rice, sorghum and corn used to be important crops. Most of them, especially cotton and corn, saw their market collapse after the liberalization, consequently leading to a fall in labor demand. The Atlantic region caught up with the rest of the rural sector in the late nineties. The Atlantic region, on the other hand, grew relatively slowly during the first two periods studied, making it the region with the lowest average income per capita in 1988 and 1995 (in 1978, it ranked second, behind the Central region). Thereafter, the Atlantic region recovered to average rural levels after experiencing an astonishing 38 percent rise in mean income per capita between 1995 and 1999; this was the largest increase experienced by any region during this period. The performance of the rural Atlantic region is somewhat puzzling. Although robust growth in the main urban centers served as a development pole, agricultural activities faced considerable obstacles during this time. As depicted in the urban chapter, although poverty risks were high in Barranquilla, they were decreasing at a fast pace and salaries and unemployment were the lowest in that city during the recession period. This should have had a positive effect on the Atlantic region, as 94 The largest coffee harvests in Colombian history occurred in 1990 and 1993. Since then production has dropped. Average employment fell from 803,000 to 600,000 persons/year since 1996 (Jaramillo et al. 2000). 95 The 1999 earthquake also caused considerable damage in the department of Valle, in the Pacific region. 80 Barranquilla has been shown to act as a strong development pole in this region.96 Another influence may be the fact that cotton cultivation is decreasing to the benefit of that of sorghum, corn and livestock, which all have lower labor requirements, leaving many to migrate upon losing jobs. In Cesar, for example, when cotton disappeared, migration took place to Venezuela and Barranquilla, as well as to the department of Guaviare. In addition, with the Atlantic region has a high concentration of armed groups and thriving illicit crops; displacement to large urban areas such as Barranquilla, Cartagena and Monterfa has been quite high.97 In turn, some of the labor migrating from rural areas found employment in the region's urban centers. Thus, migration may have played a complementary role in explaining the Atlantic region's recent dramatic rise in mean income. The Pacific region is losing some of the substantial gains obtained up to 1995. Together with the Oriental region, the Pacific rural area experienced the largest gains in mean and median income per capita between 1978 and 1995. However, it was the only region to experience a drop in mean income per capita in the latest sub-period; mean income per capita fell by an alarming 26 percent (median income per capita declined by 18 percent). At only C$ 76,893 in 1999, the Pacific's mean income per capita was 24 percent lower than the rural national average. All other regions had average per capita incomes above C$ 100,000 - although still less than half the urban average per capita income. The Pacific region suffered from both the deterioration of the coffee sector and the effect of trade liberalization on medium-sized family agro-businesses, as well as the concomitant recession in Cali's construction sector. This region also had coffee plantations on the hillsides and suffered when coffee prices and production fell. After the liberalization and the removal of agricultural tariffs, sugar cane, with low labor input demand and under some protection in the Andean group, substituted the formner mix of corn, sorghum, cotton and soybean, linked to medium-sized family agro-business. This substitution of highly labor-inten'sive crops, together with the decreasing coffee production after the end of the International Coffee Pact, must have led to reductions in rural labor market demand. Although urban and regional markets are better connected in the Valle del Cauca area, regional urban markets did not help to absorb the available labor. In 1995, Cali underwent a deep recession in the construction sector, with increasing unemployment of unskilled workers.98 3.2. Rural welfare as average income corrected by inequality: The Gini coefficient and the Sen welfare index Fluctuating income inequality diminishes rural welfare gains from average income in the late nineties99 After an improvement in the 1988 to 1995 period, inequality deteriorated in 1999 to reach levels much higher than those registered in 1988 (see Table 2.3). In contrast with urban income inequality, which rose throughout all periods, rural inequality increased more from 1978 to 1988, but 96 See Jaramillo, 2000. Evidently, as mentioned above, the same argument of Sanchez and Nufiez (2000) applies. 9t C6rdoba, the department where Monterfa is located and Atldntico, where Cartagena and Barranquilla are located, have the highest percentage of reception of IDP in the region (6.15 and 5 percent of the total number of IDP in 1995, the 4th and 6' rank nationally, in Erazo et al. 2000 about IDPs, footnote 3). 98 Sanchez and Ndfiez' (2000) rigorous and comprehensive study on Colombian regional economic development highlights the strong influence of large local urban markets on regional development. 99 Chapter 4 presents a more thorough investigation of inequality changes via micro-simulation of structural parameters and endowment changes in a model of household income generation, labor force participation and occupational choice. 81 then decreased between 1988 and 1995 -the Gini fell 0.446, close to that of 1978.'°° However, since then, inequality has reached unprecedented levels for any inequality measure - the Gini coefficient reached 0.502, well above the previous highest, reached in 1988. Inequality from the quintile perspective: As in the urban areas, the evolution between 1978 and 1988 shows all quintiles rising, but the poorest lag behind. Figure 2.3 displays the increasing trend of household mean income per capita by quintile, normalized to one in 1978. It is important to note that this period of rural growth did not benefit all income levels equally, as the lowest quintile lags behind the other four by 25 percent on average, and behind the top quintile by almost 30 percent. Figure 2.3. Evolution of mean per capita income by quintile (1978=1) 2.2 2 0 --Quintbla 1 -lt Ouintile 2 ..6* -0QumUtfle 3 1 8 - - O,uintWe 4 P'7 1 8 COuintile5 1 6- 1.4 - 12 - 1.0 1978 1988 1995 1999 On a positive note, the deterioration of inequality in terms of quintile evolution reversed substantially between 1988 and 1995. Most of the benefits go to the lower middle part of the distribution. Here, the second and third quintiles lead in positions relative to 1978, quintile 4 remained relatively stable, and the first quintile caught up with quintile 5 after a near 40 percent gain in income. The divergence between the top and lower quintiles explains the inmcreasing inequality between 1995 and 1999. The picture turned sour as the bottom quintile lost almost all of its previous gains, whereas the top quintile saw its income rise by 40 percent. The second quintile lost 16 percent, to converge with the third and fourth at about an 80 percent gain relatiVe to 1978. Thus, inequality within the middle quintiles has remained relatively stable over time, while the first and fifth quintiles have diverged with a 60 percent gap in relative income gains between the two. Entropy measures reveal that the fall in inequality in the 1988 to 1995 period was due to less inequality within the lower tail of the income distribution, while the upper end remained stable. EO and El (mean log deviation and Theil index) fall, but E2 remains almost constant. This is consistent with our quintile analysis, as the fifth and fourth quintiles' relative income remained stable, while the first and second quintiles experienced a relative rise in income and the gap between the two 100 Recall that in urban areas, the largest increase in inequality, as measured by the Gini and entropy coefficient, occurred during this 1988-95 period. 82 decreased. Thereafter, in 1999, as the richest and poorest quintiles distanced themselves from each other, the worsening of inequality also occurred within the lower and higher tails of the income distribution. Other inequality measures are equally consistent. Again consistent with our analysis of income quintiles, P75/P25 remained stable over the three sub-periods, whereas P90/P10 and, especially, the income share of the upper quintile relative to the income share of the lower quintile, increased. These three measures of inequality were alleviated during the 1988-95 sub-period (as also shown by the Gini measure), but the reversal since then brought inequality figures, except for P75/P25, to levels worse than those of 1978. In particular, the gap between the richest and the poorest has worsened dramatically, as P90/Pl0 reached 9.7 in 1999, up from 7.2 in 1995, and the income share of the upper quintile relative to that of the lower quintile rose from 10.2 to 14.4 in 1995 and 1999, respectively. Regional Comparisons: Regional inequality measures diverged up to 1995 and then converged in 1999. The Central region was the only region that did not experience a rise in inequality during the first decade studied. With the exception of the Pacific, inequality fell in all regions in the second sub-period and increased in the latest. The Gini coefficients in the Atlantic and Oriental regions followed patterns similar to the overall rural average. The Atlantic region had improved substantially between 1988 and 1995, as the Gini fell to only 0.417 (down from 0.456 in 1988) and the gaps between the top and bottom deciles also improved. But during the recession, this improvement was erased and the Gini reached a peak of 0.504. The Oriental region also reached a peak in 1999, but this was not much higher than its 1988 Gini. Income inequality in this region was higher than the rural average, although, as seen earlier, the Oriental region had lower poverty rates (and higher income per capita), than overall rural Colombia. In the 1988-95 period, the improvement in the Gini coefficient for the Central region (down from 0.396 to 0.384) was relatively small, especially when compared to that of the Atlantic and Oriental regions. However, this came after a fall in the Gini coefficient during the 1978-88 sub- period, during which the other three regions had increases in income inequality. Furthermore, the Central region always enjoyed the lowest inequality measures in rural Colombia. Nevertheless, between 1995 and 1999, the Gini coefficient faced a dramatic rise to 0.488 in 1999, now converging with the other three regions. Unlike in the other regions, the Pacific region's Gini rose between 1988 and 1995 (reaching a peak of 0.518) and'then improved in 1999, making it the most equal region in that year.01 Looking at P90/P10, P75/P25 and the income share of the upper quintile relative to the income share of the lower quintile reveals a different story: just as in the other regions, the gaps between richer and poorer households actually fell in 1995 and then rose in 1999 to reach levels unprecedented since 1978. This corresponds with the simultaneous deterioration of the income distribution in Cali in that latest period. In terms of distribution changes, data for the Pacific region also show that the relatively wealthy were negatively hit-average income fell but the median did not fall as much and the Gini improved (see Appendix Table A.5). Agricultural rents fell and medium-sized family farmers, whose '°' Recall that the Pacific region was also the only one to experience a fall in mean income per capita during the 1995-99 sub-period. 83 livelihood depended almost exclusively on agriculture, saw their profits eroding. Poorer rural households might have been hit less severely by diversifying their employment to various sectors in addition to agriculture -mainly services-. The poorest were again hard hit, in particular by the three- fold rise in unemployment (see Table A.6D in the Appendix). The Sen Welfare Index: rising inequality has diminished potential welfare gains Although income has risen substantially since 1978, a concu¢rrrent rise in income inequality has reduced potential welfare gains by nearly 20 percent. Figure 2.4 depicts the evolution of welfare by two alternative measures: mean household income per capita and the Sen welfare index.'02 The increase in mean income between 1978 and 1999 was 95 percent, but the simultaneous increase in inequality resulted in a 19 percent loss in potential welfare. Had inequality remained at 1978 levels, the total welfare gain would have been in the magnitude of 95 percent, versus an actual gain of 76.4 percent. In the 1978-88 sub-period, the loss of welfare due to inequality was equivalent to only 3 percent. On a positive note, a more favorable distribution of income in 1995 led to welfare gains 4 percent greater than if inequality had remained the same. Figure 2.4. Welfare, Rural Colombia, 197$, 1988, 1995, 1999 105,000 95,000 85,000 75,000 / -M _ C* 65,000 / 55,000 - 45,000 - 35,000 - 25,000 1978 1988 1995 1999 However, in the recessive period average income gains weire more than compensated by the rise in inequality. This may have not resulted in reduced mean income per capita, but an increase in the Gini coefficient of 0.06 points led to a 4 percent drop in welfare between 1995 and 1999 -erasing the otherwise 7 percent gain in income-. By the Sen welfare index, even if mean income per capita did increase between 1995 and 1999, rural Colombians became worse off in 1999. Regional Comparisons'03 Since 1978, all regions have experienced a rise in the gap between average income and inequality-adjusted welfare (see Figure A2.1). 102 See urban section for a description of the Sen welfare index. 103 Rural data are not adjusted for local differentials in the cost of living. 84 As mentioned beforehand, inequality improved in all regions between 1988 and 1995, with the exception of the Pacific region. Consequently, although in 1995 the Pacific region enjoyed highest mean income per capita relative to the other regions, it only ranked third in terms of the Sen welfare index. Meanwhile, the Central region had the highest Sen welfare index, due to its exceptionally low Gini coefficient in this year -0.384-. Much like in rural Colombia as a whole, mean income and inequality-adjusted welfare in the Central region follow divergent paths during the 1995-99 sub-period. Although the region experienced a fall -12 percent- in inequality-adjusted welfare during the 1995-99 sub-period, it, too, experienced a rise in mean household income per capita. Had inequality remained constant between 1995 and 1999, the Central region could have benefited from a 6 percent rise in the Sen welfare index.'04 Nevertheless, income inequality was still low enough for the Central region to surpass the Atlantic region -where mean income per capita rose above the Central region's between 1995 and 1999- in terms of inequality-adjusted welfare. In the Pacific region, inequality reductions diminished the potential welfare losses from a fall in mean income. During the 1995-99 sub-period, while inequality decreased, income fell concurrently. At least the benefit of greater equality resulted in lower welfare losses than otherwise and the gap between average income and inequality adjusted welfare decreased. 3.4. Welfare comparisons independent of distributional weights: First Order and Generalized Lorenz Dominance Rural welfare is unambiguously better in 1988, 1995 and 1999 than in 1978. As shown by the cumulative distribution functions in Figure 2.5A Relative to 1978, the other three years studied, 1988, 1995, and 1999, reveal unambiguously large gains in welfare with first order stochastic dominance. The percentage of people under the 1999 poverty line (C$ 125,497) decreased from 95 percent to 79 percent in the 21-year period studied. Gains were even larger for those well below the 1999 poverty line: the proportion of people below the 1999 extreme poverty line (C$ 49,901) fell from 65 to 37 percent during the last two decades. '04 The Central region registered the highest increase in inequality for the 1995-1999 period; its Gini coefficient rose by 0.104 points, from 0.384 to 0.488, in 1995 and 1999, respectively. 85 Figure 2.SA. Cumulative income distributions, RuraD CoDombia 2978, 1198, 1995, and 1999 100% - CL ao 70%/ 17 0% - 1995lS9PVel o 6% 50% Percapitamoh oeld i 1999 p Loren domianc ovr 1Poverty 10%- $49,901 line: $125,497 10% e 0% 0 50,000 100,000 150,000 200,000 250,0W0 Per capita monthly household income, 1999 pesos The great majority of the rural population -except foir the top 7 percent- saw a welfare improvement from 1988 to 1995. The 1995 cumulative distribution of income is always to tight of 1988, except for after the 97n percentile, when two lines cross (Figure 2.5Ba). Interestingly, those at the bottom of the distribution benefited the most during this period, as we suspected earlier. Generalized Lorenz curves for 1995 and 1988. (Figure 2.6 does show 1995 as having generalized Lorenz dominance over 1988.) Figure 2.5s. Cumuative income distributions, Ruwral Colombia 1988, 1995, and 2999 c- 90% 8 9 0% M 70%/ PO%" Imnp: a $49,901 .~60% npr r-50%- .2 E) 30%/ -199988__ 0 0% I 0 25,000 50,000 75,000 100,000 125,000 150,000 Per capita monthly household income, 1999 pesos Most of the rural population was worse off in 1999 than in 1995. The cumulative distribution functions for 1995 and 1999 in Figure 2.513 reveal that welfare turned for the worse for the bottom 70 percent of the population during the recessive period, while the top 30 percent made a slight gain. The proportion of people whose household income per capita put them below the 1999 extreme poverty line increased from 34 to 37 percent in 1995 and 1999, respectively. The fact that most of the population experienced a decrease in welfare in the 1995-99 period, despite a boost in mean household income, is also revealed by a drop in the Sen welfare index, as shown in Figure 2.4. 86 Nevertheless, in 1999 rural welfare was above the 1988 level for 95 percent of the population. Relative to 1988, however, welfare did increase for Colombians in rural areas in 1999, with the exception of those at the bottom 5 percent of the income distribution. Therefore, relative to urban Colombia, rural areas sustained a less severe deterioration of welfare during the 1990s.'05 Figure 2.6. Generalized Lorenz curves, 1999 pesos, Rural Colombia, 1988 and 1999 C a 100,000- E o o.E 80,000 o60,000 - 199-9 Co 0 0 2 . 40,000 E20 ZE20,000 0 0 10 20 30 40 50 60 70 80 90 100 Cumulative percentage of people In summary, it is unambiguous, by any welfare measure, that rural Colombians were better off relative to 1978 in the other three years studied (see Table 2.6). Whether or not welfare underwent a relative improvement within the 1988-99 period is dependent upon the social welfare indicator examined. 1995 is clearly better than 1988 in terms of all poverty measures, income per capita, and the Sen welfare index. However, it does not exhibit first order dominance over 1988, although it does show generalized Lorenz dominance. In 1999, welfare in terms of income per capita expanded relative to 1988 and 1995. However, 1988 is superior to 1999 in terms of the poverty gap and P2 measure, and 1995 dominates 1999 in terms of all poverty measures and the Sen welfare index. Thus, it is evident that although mean income per capita may have increased during the last four years, by most welfare measures, rural Colombians are not better off in 1999. 105 In chapter 3 we saw that nearly the 50 percent urban poorest were worse off in 1999 than in 1988. 87 Table 2.6. Wlfseve R_rat E mb, 197 1 15 ad 19 Social Wet&re in year (column) 198 1 988 i 1995 1999 .- ^ t > . isnWerM ian year (row) 1.978 I la" 1988 w .-. ........ (L wo, . tjWt~S , , D i . ig , i, ; ,'| '.s , Ii' I - ix *2. t _nm~ _,._ .~A ,,4 , _ p7ot _- F%1)-r -~ p~t-aa__-_-r 4.FACES OF THM PoOMR VULNRttALE GROUP AMD HOUSJOWD CHARACERISTCS 4.1. Ba*I factors of income per topfl a geeratln: the poor versus the non-poor Identify baslc faetors d income per capita genrtI.o to ',¢'\2 the poor 'v wt the non- poor'. Household total inconm is identically equal to the sum of labor and non-labor incone received by its members. Consider a typical household h. with NA memben and At, of them in working age, among whom T, individuals with St, equivalent skill units am employed in the labor mrket for an average housebold wage per skill unit Of WA. The household kabor inwome per capita can be written as the following identity:1 . h a W-S, . , , , . . 6 , Equation (1) describes the household labor income per capita as a product of average wage per skill unit, the household's averape skills, the rate of enployment of adult population (TIA) and proportion of working-age members (A j). Therefore, if we ignore property income and transfers, poor 106We borrow thi identity from Paes de Barro es aL for UTNDP (2000). 88 households should be characterized by one or all of the three following features: low skill endowments, low employment rates and high children to adult ratios. Poor households simultaneously suffer from higher dependency ratios, lower skill endowments and lower employment of spouses and other adults (non-heads). Differences look less pronounced than in urban areas. Poor families are becoming smaller, but they reniain significantly larger and have larger proportion of children than non-poor. Table 2.7 shows that the average family size for poor households fell from 6.1 members in 1978 to 4.9 in 1999 whereas non-poor households only saw a drop in family size of 0.3 people during the same period. Nevertheless, the gap between poor and non-poor household size is still high at 1.5 -that found in urban areas is 1.1-. Consistently, poor rural households were larger than poor urban households, while non-poor households were smaller in rural areas. The decrease in household size among the poor is accompanied by a decrease in the average number of children aged 9 and under (from 2.2 in 1978 to 1.4 in 1999), while this number stays stable in non-poor households at around 0.5. Table 2.7. Income sources and needs: the poor versus the non-poor, Rural Colombia, 1978-99 1978 1988 1995 1999 Poor Non-poor Poor Non-poor Poor Non-poor Poor Non-poor Needs Average number of people in the household 6.1 3.6 5.2 3.8 4.9 3.4 4.8 3.3 Average number of children 12 yrs. & under 2.2 0.6 1.6 0.6 1.4 0.5 1.2 0.5 Educational endowment. Schooling, head of the household 1.8 2.9 2.8 4.5 3.2 5.1 3.4 5.8 Schooling, individuals older than 18 yrs. 2.1 3.3 3.3 5.1 3.8 5.7 4.0 6.3 Households with at least I college graduate. 0.0% 1.1% 0.3% 3.2% 0.3% 5.5% 0.9% 9.8% Household employment and child labor Household head 90.9% 89.0% 85.6% 84.8% 84.1% 83.7% 81.3% 79.9% Other adults 17 & over 58.0% 72.1% 58.6% 69.4% 59.6% 75.2% 58.3% 72.4% Children 10-16 in labor force 24.9% 38.6% 26.8% 33.9% 21.3% 27.4% 19.5% 18.5% Schooling of the head of household increased, albeit slowly during the last sub-periods among poor households and much more rapidly among the non-poor, where heads doubled their schooling in the past 21 years. A parallel evolution is taking place regarding the education of adult members of the households: while adult members of poor households had completed one less year of schooling than their counterparts in non-poor households in 1978, the gap was more than 2.3 years in 1999. (However, this gap is still in the magnitude of about 33 percent of non-poor educational levels). The percentage of households with at least one college-educated member stagnates among poor households, until finally reaching one percent in 1999, while 10 percent of non-poor households have a college-educated member. This disparity is not unlike that found in urban areas. Secondary school enrollment gaps switched during the economic recession. School enrollment for primary school-aged children among poor households is catching up with their counterparts in 89 non-poor households, reaching respectively 90 and 96 percent in 1999, (from 66 and 75 percent in 1978). Among high school aged teenagers, rates were actually higher among poor households until 1988, they equalized around 64 percent in 1995 and the tendency has been reversed in the latest 1999 survey, with 71 percent of teenagers in non-poor households enrolled in school, vs. 65 percent of poor teenagers. A gap between poor and non-poor enrollment in tertiary education has only recentfly become apparent, as university-age adults in poor households were just as likely to be enrolled in school until 1995.'°' The non-poor now have an advantage with 19.5 and 24 percent enrolled among the poor and non-poor, respectively. At least on a sobering note, the gap between the rich and poor has been the result of a larger increase in non-poor enrollment rates, rather than in a drop school attendance among the poor. This phenomenon may be explained by different behavior patterns of labor force participation rates among the poor and non-poor. Employment of spouses and other adults makes the difference. Analogously with our urban analysis, employment rates for spouses and other adults'08 are much lower for poor households, whereas those for household heads are only marginally higher. The gap in employment rates for spouses and other adults between the poor and the non-poor has remained steady at around 14 percentage points'09, highlighting the importance of overall household employment rates for escaping poverty. In urban areas, the gap stands at around 12 percentage points, but employment rates are consistently higher in rural areas, especially that of the household heads. In 1999, the proportion of both poor and non-poor employed household heads in rural areas was 7 percentage points higher than in urban areas; for other employed adults in the household, the rural proportion was 2 and 4 percentage points higher for the poor and non-poor, respectively. Child labor participation rates declined sharply for the non-poor. The proportion of working children 10 to 16 years old has been decreasing since 1978 among the non-poor and since 1988 among the poor. It was always higher among the non-poor, until 1999, when rates declined by almost 9 percentage points for the non-poor and only 2 percentage points for the poor, bringing them both to around 18 percent. As seen above, this decline in labor rates was accompanied by a rise in secondary school enrollments among the non-poor. With employment prospects on the decline, non-poor families are no longer putting children to work as a successful strategy to escape poverty, an evolution similar to the one in urban areas. 4.2. Faces of the poor: Typical characteristics of households with higheDr poverty risk From basic income generating factors to the faces of the poor. The determinant factors that characterize income generation in poor households do not always reveal the faces of the poor. In fact, the value of employment rates or dependency ratios or earnings per skill unit are all endogenously determined for each household by the interaction of the household's specific characteristics with the local socio-economic environment it faces. For example, the wage per equivalent skill unit W, depends on occupational choice -wage earner or self-employed-, sector of economic activity - agriculture or services- and, regional job market, even including the quality of the soil if the household is involved in agricultural activities. 107 The difference between poor and non-poor enrollment rates is not statistically significant at a 95 percent confidence level. 108 i.e. other adults other than the household head. 09 Except for in 1988, when the gap was 11 percentage points. 90 In this sub-section we 1) explore which characteristics of the household increase / decrease their risk of being poor and 2) to what degree socio-demographic groups that are traditionally considered vulnerable -e.g. children, the elderly, women, etc.- are exposed to poverty risk. Table 2.8. Relative risk of being poor, Rural Colombia, 1978-99 Relative risk Share of total 1978 1988 1995 1999 population in 1999 Household size one person -53.9% -51.5% -59.3% 49.0% 1.8% 2 to 5 persons -5.5% -8.7% -7.8% -7.6% 53.1% 6 to 10 persons 2.1% 8.0% 9.4% 10.1% 41.1% More than 11 persons 5.3% 8.4% 16.7% 18.7% 4.0% Household employment rate] zero 6.8% 3.6% 1.8% -0.3% 4.9% O.75 -12.2% -14.6% -27.2% -25.5% 13.5% Head's sector of economic activity Agriculture, fishing 0.9% 6.1% 5.9% 8.1% 54.4% Mines 5.2% 2.6% 14.8% 3.5% 0.9% Manufacturing -2.2% -7.3% -2.5% -8.8% 4.2% Utilities 6.8% -8.7% -34.2% -12.0% 0.4% Construction -8.0% -5.6% 1.3% -1.2% 2.4% Commerce -8.9% -10.8% -5.3% -10.7% 7.3% Transportation -5.4% -14.8% -7.7% -11.3% 3.1% Banking, insurance 6.8% -17.2% 48.5% -25.2% 0.5% Other services -9.5% -19.5% -19.8% -24.8% 9.5% Inactive -0.9% -1.1% -2.8% -2.0% 17.2% Head's occupation Family worker 6.8% 0.5% -6.7% 0.1% 0.1% Blue collar 2.6% 1.2% 7.4% 2.1% 25.2% White collar -6.5% -19.1% -24.2% -23.0% 12.7% Domestic NA -21.9% 7.5% -10.3% 0.9% Self 0.6% 9.4% 6.6% 8.5% 38.6% Employer -5.9% -7.0% -13.6% -8.5% 5.3% Inactive -0.9% -0.1% -2.8% -2.0% 17.2% Head's education2 Uneducated 2.1% 10.6% 11.6% 7.7% 21.6% Primary -0.6% 2.0% 2.3% 5.2% 59.2% High school dropout -9.5% -14.7% -10.6% -10.7% 10.3% High school -49.3% -35.3% -34.9% -31.5% 5.7% College dropout NA -76.2% -77.8% -50.9% 1.2% College graduate -8.8% -42.9% -51.4% -60.2% 2.0% Average education, household members 10 yrs. & older Uneducated 2.1% 10.8% 13.8% 3.1% 2.8%. Primary 0.5% 6.2% 7.6% 9.1% 64.8% 91 Table 2.8. Relative risk of being poor, Rural Colombia, 1978-99 (continued) Relative risk Share of total 1978 1988 1995 1999 population in 1999 High School Dropout -15.3% -23.0% -16.5% -11.1% 28.5% High School NA -61.2% -61.1% -58.4% 2.2% College dropout NA -75.8% -91.3% -86.1% 1.4% College graduate NA -76.7% -82.5% -98.5% 0.3% Head's age From 0 to 19 yrs. 4.6% -2.6% 0.2% 2.2% 0.4% From 20 to 29 yrs. -0.8% 2.5% 1.3% 4.8% 9.1% From 30 to 39 yrs. 1.3% 5.1% 4.3% 3.5% 24.2% From 40 to 49 yrs. 1.1% 0.7% 3.0% 0.6% 24.3% From 50 to 59 yrs. -1.0% -6.8% -1.6% -3.2% 18.9% From 60 to 69 yrs. -1.8% -1.3% -11.6% -2.0% 13.9% From 70 to 79 yrs. -2.0% -3.4% 0.3% -5.9% 7.0% Over 80 yrs. -9.8% -6.8% -1.5% -6.7% 2.1% Head's Marital status Consensual 0.7% 9.3% 7.8% 7.8% 35.8% Married 1.4% -1.9% -0.6% -1.2% 43.8% Widower 4.8% -5.1% -6.2% -7.5% 8.9% Separated/Divorced -5.6% -0.1% -3.1% -10.4% 7.6% Single -15.3% -18.7% -29.4% -20.9% 3.8% Head's gender Male 0.2% 0.7% 0.9% 1.3% 84.1% Female -2.0% 4.2% -5.7% -7.0% 15.9% 1. e=# of employed/# of persons 10 and older 2. Uneducated: 0 years of education. Primary: 1-5 years of education. High school dropout: 6-10 years of education. High school: 11 years of education. College dropout: 12-15 years of education. College graduate: 16 or more years of education. Households facing larger relative risks of poverty are typically large in size, have low skll levelLs, and suffer from low employment rates. Specific categories with higher risk include (Table 2.8): (i) households with more than 6 persons (ii) households whose employment rates are between 0 and 40 percent, (iii) households whose head is involved in agriculture and fishing, (iv) households in which the head is self-employed, (v) households whose head is uneducated or has only received primary schooling, (vi) households whose education average for members of working age"° is zero or at primary level. lIncreasing poverty risk for large households. Overall, persons living alone and households with more than six members have seen their poverty risk increase in the period 1995-99, although the former very small group is still better protected than most. Households with more than 6 members, representing 45 percent of the population, faced a risk of poverty at least 10 percent higher than 110 10 years old or older. 92 average in 1999.11" This places a much larger share of the population at risk than in the urban areas, where households of a similar size represent only 27 percent of the population. Over the years, higher household employment rates are necessary to escape poverty. Households with low employment rates are consistently more at risk, but a worrisome trend is that even households with very high employment rates (above 50 percent) are now facing higher poverty risks, as if the income that the working members raise were no longer enough anymore to sustain the households. Only households with employment rates greater than 75 percent have becomne less vulnerable since 1988. Those whose employment rates between 0 and 20 percent are particularly vulnerable, facing a risk of poverty that is almost 20 percent higher than average. Employment in agriculture offers decreasing opportunities to avoid poverty. Households in which the head engages in agriculture and fishing (51 percent of the rural population) are now facing a greater poverty risk, at 8.1 percent in 1999, compared with only 0.9 percent in 1978. The construction sector has also lost its protective power, as households with heads in this sector now face average poverty risks, possibly a lingering effect of the 1995-97 crisis in that sector. The best protection is provided by heads employed in banking, insurance or other service sectors (less than 11 percent of the rural population in 1999). Relative to 1988, this protection has increased over time, as has that of being a white-collar worker. Households with blue-collar and self-employed heads (more than half the 1999 rural population) consistently faced a higher risk of poverty. Secondary education of the head makes a larger difference in rural areas. Contrary to the urban areas, some high-school education still improves the household poverty prospects. Having an educated head clearly provides insurance against poverty, but this protection has gradually eroded over time. Differences in vulnerability are most staggering for certain levels of education. The effect kicks in as soon as the head reaches some level of high school (still only 10 percent of the rural population in 1999). However, there has always been a stark difference between those households with heads who are high school dropouts and those with heads with a secondary education or more, which represented only 9 percent of the rural population in 1999. The protective effect of education above primary has decreased since 1995, except for that of college graduates, whose relative risk is now at minus 60.2 percent, down from minus 51.4 percent. The relative risk of facing an uneducated or primary level educated head has increased relative to 1978, but household with uneducated heads actually saw their vulnerability decline during the recession. Education endowment of other household members in working age also provides a great degree of protection against poverty. The education of household members (non-head) 10 years of age and older has an especially strong premium for average levels equivalent to secondary schooling or higher. While protection against poverty for this group has decreased overtime -with the exception of the college graduate category- their premium has actually increased, as those in the high school dropout and primary categories have become more vulnerable. Very few households benefit from this protection, however, since less than five percent of the 1999 rural population had household education averages of high school or more. Vulnerability does not vary greatly for household heads of different ages. Protection from poverty kicks in when household heads reach the age of 50, but the effect has remained very small over the years, usually at less than 5 percent. Households with younger heads have become more vulnerable over time and those with heads aged from 20 to 29 years saw their vulnerability deteriorate "' Households with more than II persons faced the highest relative risk -almost 20 percent-, but this group only represented 4 percent of the population in 1999. 93 the most during the economic slowdown and now face the most risk of falling into poverty, at 5 percent. Consensual union households have been facing higher risks of poverty, while households with widowed, separated or divorced heads have experienced an increase in protection against poverty. Women-headed households have consistently faced less poverty risks than male-headed ones, and this difference has increased over time. While the relative poverty risk has increased for male-headed households, female-headed households are now 7 percent less likely to be poor. Compared to urban areas, differences in relative poverty risk across household characteristics appear less pronounced. While in urban areas, relative poverty risks generally fluctuated between 20 and 60 percent, they generally ranged from 5 to 30 percent in rural areas. Perhaps this comparable difficulty to identify the rural poor vs. the urban poor through the same set of characteristics is explained by missing information on key determinants of rural output heterogeneity. The obvious candidates are quality and quantity of land for agricultural use and the availability of infrastructure at the local level -transportation, telecommunications, etc.- Vulnerable Groups: Children and recent migrants Poverty threatens young children. Just as in urban areas, we find that children are consistently more likely to be poor than the population on average, especially those between 0 and 13 years of age. Since 1988, they have had poverty rates averaging 8 percentage points above the rural average. Children from 14 to 17 years old and women face poverty rates close to the rural average. Table 2.9. Poverty count for different sub-groups of the population, Rural Colombia 1978 1988 1995 1999 Rural Colombia 93.6% 79.6% 78.7% 78.8% Children under 2yrs. 97.3% 86.1% 87.7% 86.1% Ages 2 to 6 97.6% 87.8% 86.9% 87.4% Ages 7 to 13 97.3% 87.6% 86.9% 86.5% Ages 14 to 17 92.8% 80.8% 80.9% 79.8% Women 94.4% 80.4% 79.8% 79.5% Non homeowners 92.6% 80.7% 82.1% 79.3% Homeowners 94.2% 79.1% 76.9% 78.5% Disabled 93.3% NA 84.0% 78.0% Recent migrants' NA 79.0% 73.9% 79.2% Migrants<5%2 68.9% 79.5% 71.8% 75.9% Migrants<25%3 75.5% 74.1% 73.1% 75.1% Migrants<10%4 59.3% 77.9% 69.0% 73.3% Pensioned NA NA 32.8% 28.6% 1. Recent migrants: people who have lived less than 1% of their lives in the current area. 2. Migrants <5%: people who have lived less than 5% of their lives in the current area. 3. Migrants<25%: people who have lived less than 25% of their lives in the current area. 4. Migrants <10%: people who have lived less than 10% of their lives in the current area. In the late nineties, recent migrants -probably desplazados, or I[DP- are increasingly exposed to poverty. Migrants in general are slightly less poor than the overall population. The exceptions are those who just moved and might still be missing the right connections and insertion in their new place of residence; they face poverty rates marginally above average. However, all migrants, especially 94 recent migrants, have seen an increase in poverty rates in the 1995 to 1999 period. This deterioration of migration as protection against poverty -which we also observed in urban areas-, may be partially indicative of forced migration or Internally Displaced Population due to the armed conflict concentrated in rural areas. Contrary to urban areas, homeownership provides much weaker protection against poverty. Poverty rates of non-homeowners have been decreasing both in relative and absolute terms and are now only slightly higher than those of homeowners. In the rural sector, homeownership does not seem to protect against poverty; as the poor and non-poor exhibit similar home-ownership rates. Pensioners are better off. Just as in urban areas, retired people receiving pensions are much less poor than the average population, especially in 1999, when only 28.6 percent of them were poor. This category, however, only represented 0.4 percent of the rural population in 1999. 4.3. The poverty profile and the marginal effect of key household characteristics Poverty profile. In order to estimate the marginal effect of each key household characteristics, we model the probability of being poor, including all variables that proved strong correlates of poverty in the previous descriptive statistics. The poverty profile was constructed based on logit regression analysis, determining the marginal effects of key household characteristics related to skill endowments, demographics, labor market choice, regional effects, housing ownership and idiosyncratic shocks. The results are presented in Tables 2.8A-E. Human capital, as proxied by educational endowment, provides effective protection against poverty. However, the protective effects of a household head that has completed high school or more declined in 1999, after peaking in 1995. Households with uneducated heads are no longer facing a significant risk of falling into poverty relative to those with a primary education, and those households whose head has had some high school education are now slightly more protected against poverty. Average household education still provides some protection against poverty but this effect has also diminished over time. Experience (age) of the head does not shield the household from poverty. Table 2.10A. Marginal effects of selected human capital variables on the probability of being poor, Rural Colombia Human Capital Variables. 1978 1988 1995 1999 Head's education.2 Uneducated 1.4% *- 3.0% '- 4.5% 0.9% Some high school -2.1% -6.7% -7.1% -7.8% High school -6.8% -12.1% -16.5% -11.4% Some college NA -28.7% -32.9% -18.9% College graduate NA -32.7% -53.7% -32.6% Head's experience.3 0.0% -0.3% -0.2% 0.3% Household's average education4 -0.7% -2.3% -2.3% -1.7%' I Marginal effects are computed as:[exp(x'ibeta)*beta* k] / [I+exp(x'ibeta)]2 2 Base category is heads with primary education. 3. Experience=min[(age-education-7),(age-12)] 4. Average taken over all household members other than the head who are older than 12 yrs. * Significant at 10% level or less ** Significant at 5% level or less 95 Households with young children now face a higher probability of falling into poverty, while the presence of working-age individuals is becoming more inportant. The effect of an increase in the number of children under 7 years old went up to 4 percent in 1999, after being insignificant in the two previous years analyzed. The presence of children from ages 7 to 9, old enough to be enrolled in primary school but not to enter the labor market, also increases the risk of falling into poverty, but this variable fails to be significant in 1999. An increase in the working age population of a household, or of people aged 10 to 65, decreases risk by 3.5 percent; in 1995, this figure was much higher, at 8 percent. People over 65 had no effect on poverty in 1978, but became increasingly important until 1995, when the positive effect peaked at 13.1; since then the effect has gone down to only 4 percent. Table 2.10B. Marginal effects of selected demographic variables on the probability of being poor, Rural Colombia Demographic Variables. 1978 1988 1995 1999 Head's marital status' Consensual union -0.9% 0.7% -2.7% -0.3% Single -0.6% 4.5% -10.6% -12.4% Age categories2 Younger than 7 yrs. 1.1% 0.0% -0.5% 3.9% From age 7 to 9 1.7% 9.9% 11.1% 6.2% From age 10 to 65 -1.3% -7.4% -8.5% -3.5% Older than 65 1.1% -7.2% -13.1% -4.9% Head's age and gender? Males Younger than 28 yrs. 2.4% 4.6% 1.9% 10.7% From 28-35 yrs. 0.1% 0.6% 0.8% 6.7% From 36-42 yrs. 0.9% 0.7% 1.2% 3.8% From 4347 yrs. 1.9% -1.5% 3.1% 1.5% From 48-57 yrs. 0.5% -1.3% 5.5% 0.1% Older than 67 yrs. -4.0% 0.9% 6.2% -4.7% Females Younger than 48 yrs. 5.6% 4.5% 14.7% 19.9% Older than 47 yrs. 3.4% 1.9% 12.9% 7.9% 1. Status compared to married 2. Variables represent number of people per household in each age category. 3. Males between 58 and 67 years of age is the base category. * Significant at 10% level or less ** Significant at 5% level or less IFemale heads face an increasing risk of poverty, especiaUly if they are young or middle aged. Descriptive statistics showed that female-headed households were apparently becoming increasingly protected from poverty; however our regression analysis, controlling for other factors, proves otherwise: Ceteris paribus, young-female-headed households are generally more likely to fall into poverty, and this likelihood has increased over time (the marginal effects, displayed in Table 2. 1OB, are compared to households headed 58-to-67-year-old males). In fact, in 1978, differences in poverty likelihood were not too pronounced in terms of head age and gender; young-female-headed households were most likely to be poor, but this difference was only in the magnitude of 6 percent. 96 By 1999, a household with a -female head under 48 years old was 20 percentage points more likely to be poor than one with a 58-to-67-year-old male head. Next in line were households with a male head under 28 -11 percent more likely to be poor-, but households with female heads over 47 followed closely -their probability of being poor is 8 percent greater than that of the base category. As in the urban areas, self-employment is associated with poverty, as are low household employment rates. Self-employment's marginal effect on poverty decreased slightly in 1995 but has since then peaked at 18.5 percent. As expected, the household employment rate strongly mitigates poverty risks, and this effect has increased dramatically since 1978, rising almost 20 percentage points to reach 25 percent in 1999. This implies that on average, an additional household member becoming employed is associated with a 10 percent decrease in poverty risk.' 12 Household diversification in self-employment and wage sectors was a successful strategy for poverty alleviation in 1978, but has now become a factor that may increase poverty, a phenomenon similar to that observed in the urban areas. Table 2.10C. Marginal effects of selected labor market variables on the probability of being poor, Rural Colombia 1978 1988 1995 1999 Household labor market characteristics' Both wage earners and self-employed. -2.6% 3.1% -0.9% 2.4% Only self employed -0.7% 17.7% 14.9% 18.5% None 0.8% 5.4% 9.7% 5.0% Employment rate2 -5.7% -21.0% -25.5% -24.6% Regional effects Municipal cabecera NA NA NA -15.9% Oriental regiont 3.3% * -1.6% -6.1% ** -1.7% Central regiont -0.8% -12.6% -3.1% * 3.1% Pacific regiont 2.1% -5.5% 2.5% 9.2% ** 1. Households with only wage earners is the base category. 2. Employment rate = number of people employed other than the head I number of people 10 & older other than the head t Atlantic region is the base category. * Significant at 10% level or less ** Significant at 5% level or less Despite controlling for other covariates, households living in the Pacific region face a higher probability of being poor (9 percent). The Atlantic and Oriental regions were not significantly different from each other in 1999. It is notable that the Central region contributed to a 13 percent decrease in poverty relative to the Atlantic in 1988, but since then has become associated with an increase in poverty. Living in a municipality capital decreased the probability of poverty by 17 percent in 1999. 112 We arrive at this figure based on an average household size -excluding the household head- of 2.4 and an average employment rate of 35 percent. Hence, one additional person becoming employed would increase the employment rate, on average, by 40 percentage points, leading to a 10 percent increase in poverty risk. 97 IHiomneownership has a moderate effect on the probability of falling into poverty, surely because property is not as much a differentiating factor as in the urban areas, a fact confirmed by the data in Table 2.10D.1; the percentage of homeowners is around 65 percent for all quintiles, with no simple correlation with income, a somewhat different picture from the one emerging in the urban areas. Table 2.10D. Marginal effects of tenancy on the probability of being poor, Rural Colombia Home Percent ofpopulation ownership 1988 1995 1999 in 1999 Owners but paying mortgage 1.6% 5.2% 6.5% 17.4% Renters 3.5% 5.2% 2.0% 13.0% Usufruct 5.4% 12.6% 1.0% 21.6% Defacto 18.9% 6.8% NA 0.1% 1. Base category is owner. 2. In 1978 all the non owners are aggregated and thus we cannot show them on the table. 3. In 1978 the marginal effect of being a non-owner was -0.9% but not significant. * Significant at 10% level or less ** Significant at 5% level or less Table 2.101D1. IHomeownership by guintile, RuraD Colombia 1978 1988 1995 1999 Quintile 1 68.9% 68.5% 64.7% 67.8% Quintile 2 63.9% 64.1% 60.5% 62.8% Quintile 3 65.9% 68.7% 65.0% 61.7% Quintile 4 60.7% 64.8% 69.1% 61.3% Quintile 5 58.5% 69.2% 70.6% 64.1% Average 63.6% 67.1% 65.9% 63.5% Idiosyncratic shocks Job loss and head's disability increases poverty likelihood while pensions provide effective protection (Table 2.10E) First-time unemployment for the household head contributes to risk of poverty, but this effect has decreased over time, going down from a 20 percent increase in probability in 1988, to 10 percent in 1999. Quitting the job market increases the probability of poverty by slightly more than being unemployed for the first time. Non-labor income in the form of pensions or rent provides effective protection against poverty, decreasing the risk by 14 and 19 percent in 1999 respectively. Nevertheless, the protective effect gained by rents has gone down by 7 percentage points during the recessive period. This is similar to the evolution in urban areas. On the other hand, pensions there have lost their efficacy during the recession, while they remain highly successful as a shield in the rural areas. Disability plays a significant role in increasing poverty risk, but this is only the case when the household head falls into this category (at 10.3 percent), also similar to its role in urban areas. 98 Table 2.10E. Marginal effects of exogenous shocks on the probability of being poor, Rural Colombia Exogenous shocks faced by the household 1978 1988 1995 1999 Head's marital status' Widower -3.9% 4.7% -4.2% -7.4% Divorced -1.4% ----- NA -10.6% Disabled head 2.1% 24.1% 10.3% Other disabled members 0.4% ----- -1.0% 5.0% Labor market variablesfor head of household2 Unemployed first time NA 39.8% 4 20.3% 10.4% Unemployed but worked before 3.2% NA NA 11.7% Household work -0.4% 4.6% 7.8% 10.1% Rentier3 -5.3% ----- -21.4% -14.4% Pensioner -10.6% ----- -20.6% -19.3% 1. Status compared to married. 2. Compared to employed household heads. 3. Rent on property, dividends, etc. 4. Includes unemployed but worked before. * Significant at 10% level or less ** Significant at 5% level or less 4.4. Unemployment by skill, age and region Finally since unemployment is so closely associated with rural poverty we want to explore its characteristics by skill, age, gender and region. Although lower that in urban areas, unemployment rates escalated and more than doubled during the economic slowdown (Table 2.11). Between 1978 and 1988, unemployment more than doubled from 1.5 to 4.6 percent, remained constant thereafter, and more than doubled once again in 1999, reaching an alarming 10.9 percent, a magnitude once only associated with urban areas.", It is worth noting that just as unemployment rates were rising, so were labor force participation rates, mostly driven by females entering the job market. Between 1978 and 1988, the female participation rate increased dramatically, from 18 to 29 percent. Thereafter, it continued to climb, reaching 33 and 37 percent in 1995 and 1999, respectively.'"4 The overall labor force participation rate rose from 53 to 59 percent during the two decades studied. Yet as more women entered the job market, they consistently encountered much higher unemployment rates than men, as noted in Table 2.11. In 1999, the unemployment rate for rural 113 The urban-rural ratio for unemployment ranged from 1.8 to 2.2 in 1988, 1995, and 1999. In 1978, the ratio was 5.3. "4 Meanwhile, male participation rates fell in 1995 vis-a-vis 1988 -from 84 to 81 percent- and then by I percentage point in the next sub-period; nevertheless, overall participation rates increased. 99 women was 19.7 percent, similar to the overall rate in urban areas. The corresponding figure for rural men was almost one third this rate, at 7.1 percent. Table 2.11. Unemployment for various demographic groups. Ruiral Colombia. 1978 1988 1995 1999 Rural Colombia 1.5% 4.6% 5.0% 10.9% By sex Male 0.7% 3.0% 3.1% 7.1% Female 5.3% 9.6% 9.8% 19.7% By education Uneducated 1.0% 2.7% 1.9% 4.6% Primary 1.3% 3.5% 3.6% 8.4% High school dropout 5.5% 8.7% 8.8% 16.6% High school 1.2% 13.8% 12.5% 23.6% College dropout 12.6% 12.5% 8.0% 16.3% College 0.0% 0.0% 4.3% 6.1% By age 12 to 17 2.1% 7.1% 7.2% 14.8% 18 to 25 2.8% 9.3% 10.3% 20.0% 26 to 36 0.9% 3.6% 4.5% 11.4% 36 to 50 0.6% 1.5% 2.7% 6.9% 51 to65 0.3% 1.7% 1.3% 4.8% Over 65 2.7% 1.8% 0.9% 2.2% By quintile 1 0.5% 5.5% 6.4% 13.6% 2 1.2% 5.8% 6.0% 13.1% 3 2.0% 5.6% 5.8% 12.2% 4 1.6% 4.8% 5.5% 11.3% 5 1.8% 2.6% 2.6% 6.6% 1. Population 10 years old & older. 2. Uneducated: 0 years of education. Primary: 1-5 years of education. High school dropout: 6-10 years of education. High school: 11 years of education. College dropout: 12-15 years of education. College: 16 years of education. 3. Quintiles are taken on household per capita income. Unemployment peaks for the young (18 to 25) and those with intermediate skills By schooling level, unemployment increased most strongly for higlh school graoduates or drop outs, and college dropouts as well. High school graduates faced an alarming 24 percent unemployment rate in 1999, more than twice the rural average and higher than its urban counterpart of 23 percent."5 Apparently, the fact that individuals with a high school education are failing to find 115 In urban areas, unemployment was also highest among those with complete or incomplete high school and incomplete college. However, in contrast with rural areas, these rates were only a few percentage points higher than the urban average. 100 employment that fits their skill level is most prominent in rural areas. Meanwhile, unemployment is lowest among those without an education, followed by those with a college and a primary education. By age, older adults experienced the most severe rise in unemployment rates. Unemployment generally tends to decrease with age and has always affected young adults most severely' 16; the 18-to- 25-age group now face an alarming 20 percent unemployment rate, almost twice the average rural rate. This is still much lower than the 1999 urban the proportion, 34.6 percent. Unemployment rates for the 5 l-to-65-age category have actually risen five-fold, from I to 5 percent in the last four years. Unemployment is associated with poverty, but differences by income quintile are not as marked here as they are in urban areas. Households in the first four quintiles of the income distribution face about a 13 percent unemployment rate, compared to 7 percent for households in the highest quintile of the distribution. In urban areas, unemployment differences are much higher and the poorest quintile faces unemployment rates close to 37 percent. Regional Comparisons The Pacific and Oriental regions have suffered the highest increases in unemployment, closely followed by the Central region. This is consistent with our earlier findings pertaining to poverty and income. Unemployment rates nearly tripled in the Pacific region between 1995 and 1999, reaching 13.3 percent, the highest in the four regions. In the Central region, unemployment is high at 12.3 percent, but given that it was already increasing in the previous period, the recession did not have as much of an impact in this region as it did in the Oriental region. Here, unemployment rates have been multiplied by 2.5 times in the last four years, reaching 10.4 percent, after a plateau at 4 percent between 1988 and 1995. Figure 2.7. Unemployment by region, Rural Colombia, 1978, 1988, 1995, 1999 12% 10 C 8%b; Central / / _E aes -wacmc - ,r/ EC 6 E 4 % 2% 0% 1978 1988 1995 1999 Year The Atlantic region, which had the highest unemployment rate in the four regions in 1988 and 1995, has the lowest in 1999. Its unemployment rate, while still having increased in the last four years, is now the lowest of the four regions (at 8.2 percent). Unemployment only rose by 2.1 116 In contrast, children aged 12 to 17 faced the highest unemployment rates in urban areas. 101 percentage points, or 34 percent, a low increase in comparison to the other three regions'. Nevertheless, it is clear that all regions are now experiencing unprecedented unemployment rates, close to, if not in, the double digits. Across all regions, high school educated individuals and those in the 18-to-25-age range consistently faced the highest unemployment rates (Table A.6.A-B). Of note is the 1999 unemployment rate for people with a high school education in the Pacific region - a staggering 32.6 percent, almost double the rate for those with incomplete high school. In the Oriental region, 1999 unemployment rates for college-educated individuals were unusually high at 10.6 percent (this rate wavered between 2 and 5 percent in the other three regions). In all regions, unemployment rates were consistently highest for people in age categories under 36 -peaking for the 18-to-25-age category- just as seen in the overall rural average data. 5. SOURCES OF EXTREME POVERTY: IDECOMPOSITION IN TERMS OF GROWTH, PRICES AND INEQUALIrY Finally, we examine the dynamics of poverty from the perspective of aggregate effects of growth and inequality. However we examine the effects of growth, relative price of food items and inequality on extreme poverty counts, rather than on the poverty count, as was done in the urban chapter. We choose to do this because, at levels up to 93 percent, the poverty count was too high to allow us to clearly see the differing effects of the three components. The results are displayed in Table 2.12. Table 2.12. Decomposition of changes in extreme poverty count, Rural Colombia. Actual change in extreme poverty Growth Prices Distribution Residual 1978-1988 -20.3 -25.6 5.5 0.1 -0.4 Contribution 100% 126% -27% -1% 2% 1988-1995 -10.3 -2.6 -6.2 -0.8 -0.7 Contribution 100% 25% 60% 8% 4% 1995-1999 -0.5 -3.6 -3.3 5.4 1.0 Contribution 100% 703% 644% -1049% -5% Growth, as in the urban case for poverty, played a dominant role in the steep reduction in extreme poverty between 1978 and 1988, while the relative rise in food-item prices dampened the effect. The worsening of the income distribution during this period slightly counteracted the growth benefits: it led to a rise in extreme poverty of only 0.1 percentage points. In the 1988-95 interval, the 10-percent decrease in extreme poverty came mostly from the fall in food prices and to a lesser extent from growth and better income distribution. As seen in the introduction, the macro-economic environment slowly became increasingly unstable during that period. The largest contribution to the decrease in the extreme poverty count is the steep -11 percent- 102 fall in food-item prices, responsible for 60 percent of the reduction."7 Growth contributed to 25 percent of the total fall in extreme poverty, while more equally distributed income contributed to 8 percent. Between 1995 and 1999, the worsening of the income distribution makes extreme poverty alleviation come to near halt. Both growth and a fall in the extreme poverty line strongly contributed to the slight fall -0.5 percentage points- in the extreme poverty rate.'8 Were it not for the particularly aggravating effects of an increase in inequality, extreme poverty would have fallen by almost seven percentage points. The residual also increased, calling for new underlying phenomena at play during that period. The same evolution took place in the urban areas, at a larger scale. 6. SUMMARY AND CONCLUSIONS The period between 1978 and 1988 was characterized by significant progress in nearly all socio- economic indicators -with the exception of income inequality-. Thereafter, progress slowed down considerably, indicative of the agricultural crisis affecting Colombia in the early 1990s and the national recession of the late 1990s. Nevertheless, in the latest sub-period, during which the urban sector suffered significant economic losses, the rural sector -with the exception of the Pacific region- continued to grow, this time even faster than in the period beforehand. However, poverty deepened and inequality reached its highest level, indicating that the recession hit the lower strata of rural society. The drug economy is fueling an increasing challenge to Colombia's governance, especially in certain rural areas. The concentration of illicit crops in Colombian rural areas dominated by guerrillas and paramilitary groups weakens the legal economy and creates a permanent challenge to the State's presence. The most severe challenge appears to be in the protection of lives and basic property rights of all strata of the rural population, since the judicial and security institutions are already stretched to their limits by the staggering level of violence in the rural areas. Moreover, rural violence is mostly connected to the presence of the illegal drug trade and armed groups. The following is a summary of our main findings: Social Progress * Illiteracy was halved from 1978 to 1999, with a recent slowdown in improvement rates. Similarly, enrollment rates for children of primary and secondary school age registered their most significant gains between 1978 and 1988, moderate gains between 1988 and 1995 and only a slight improvement in the 1995-99 sub-period. * Illiteracy and school enrollment rates converged by region, except for those in the Atlantic. The latter, while following roughly the same trends, kept apart from the others, especially that for secondary school age children, who have always enjoyed considerably higher enrollment rates. * Child labor participation rates, while remaining very high compared to urban levels, followed the same dynamics: they peaked during the 1988 boom and have been falling thereafter. During the last four years, all regions, except for the Pacific, experienced large drops in child labor. 117 The relative food-price effect on poverty might be biased since the rural poverty line is the rural average -not local or regional. This reveals another limitation of information sources about rural poverty in Colombia: price series on poverty baskets are not available by rural regions or sub-regions. The national CPI relative is clearly biased vis-a-vis the true rural local index. 118 Food-item prices fell by 6 percent between 1995 and 1999. 103 o In contrast with urban areas, basic infrastructure coverage decreased slightly from 1995 to 1999, especially in the Pacific region. However, utility coverage varies widely between municipal cabeceras and sparsely populated areas as well as between regions. Rural Welfare o Average monthly household income per capita grew dramatically in the first sub-period, relatively slowly between 1988 and 1995 and continued rising during the economic downturn that reduced income in urban areas. However, median household income per capita did decline in the last sub-period. Despite their overall growth (in average terms), rural incomes remain at roughly one third those of urban areas. o Strong discrepancies in regional development trends are re-ranking rural regions. Regional mean income per capita follows separate paths, as different factors --coffee sector dynamics, trade liberalization and export crop substitution, urban-rural integration, illicit crops and the armed conflict - play differentially across regions. The clear winners appear to be the Oriental and Atlantic regions, while the losers are the Pacific region, as well as the Central region, which used to have the preeminent position, linked to the cultivation of coffee. o Income inequality fluctuated during the last two decades. After an improvement in the 1988 to 1995 period, inequality deteriorated in 1999 to reach levels much higher than those registered in 1988, with inequality increasing in the tails of the income distribution, much like it did in urban areas. Regionally, inequality diverged up to 1995 and then converged in 1999. The Central region was the only region that did not experience a rise in inequality during the first decade studied. With the exception of the Pacific, inequality fell in all regions in the second sub-period and increased in the latest. o Although income has risen dramatically since 1978, the concurrent rise in income inequality in the late nineties diminished the welfare gains from average income per capita. The increase in the Gini coefficient during the recessive period led to a 4 percent drop in welfare between 1995 and 1999 -erasing the otherwise 7 percent gain in income-. Since 1978, all regions have experienced a rise in the gap between average income and inequality-adjusted welfare. o The poverty headcount ratio has remained relatively stable since 1988, but the poverty gap and intensity have recently worsened. Both the extreme poverty rate and the US $2/day poverty rate declined much faster than the poverty rate during the first 2 sub-periods, as in urban areas. Thereafter, these rates also stabilize. o By region, poverty dynamics are not homogeneous. In the Pacific and Central regions, this recent relative stability of the poverty count hides a widening of both the poverty gap and the P2 index in the last 4 years. Both regions have reached poverty indicators inferior to those of 1988, erasing some of the progress made between 1978 and 1988. Meanwhile the Atlantic improves in all indicators and registers a continued sharp decrease of the extreme poverty rate, between 1995 and 1999. Progress slowed down dramatically in the Oriental region. o Not surprisingly, rural welfare is unambiguously worse in 1978 than in 1988, 1995, or 1999. However the welfare ranking of the three former years is dependent upon distributional weights. Welfare was greater in 1995 than in 1988 by almost all welfare measures; it does not exhibit first order dominance due to a decrease in welfare for the top 7 percent of the population. Although 1999 appears welfare enhancing due to gains in income per capita -with respect to 1988 and 1995- and gains in poverty and the Sen welfare index -with respect to 1988-, the two earlier years are both superior in terms of the poverty gap and the P2 index. In fact, 1999 only surpasses 1995 in terms of income per capita -all other measures show that welfare has turned for the worse. 104 Poverty determinants andfaces of the poor * Just as in the urban case, rural poor households simultaneously suffer from higher dependency ratios and lower skill endowments. Although they may have shrunken in size, poor families remain significantly more numerous and with more children than non-poor ones; dependency rations in rural households are well above those.of urban households. While schooling of poor household heads increased, albeit slowly during the last sub-periods, the increase has been more rapid among the non-poor. Consistent with our urban analysis, employment rates for spouses and other adults are much lower for poor households, whereas those for household heads are only marginally higher. Child labor participation rates declined sharply for the non-poor. * The household types that face the largest relative risks of poverty are large, linked to the agricultural sector, and have low skill levels and low employment rates. Households with more than six members have seen their poverty risk increase in the sub-period 1995-99. Households in which the head engages in agriculture and fishing (51 percent of the rural population) are now facing a greater poverty risk. Increasingly high household employment rates (above 75 percent) are becoming necessary to escape poverty. Differences in vulnerability are most staggering for levels of education of both household heads and other household members 10 years of age and older. Vulnerability does not vary greatly for household heads of different ages. * The most vulnerable groups in terms of poverty incidence are households with young children, recent migrants or those that are female-headed. In the late nineties recent migrants are starting to become poorer. Contrary to the findings in urban areas, homeownership does not offer significant protection from poverty. * The regression analysis confirms most of the following results. A high educational endowment provides effective protection against poverty. Female-headed households with young children now face a higher probability of falling into poverty. As in the urban areas, self-employment is associated with poverty, as are low household employment rates. Job loss and head's disability, not surprisingly, increase- poverty likelihood, while pensions provide effective protection. Homeownership has a moderate effect on poverty outcomes -although smaller than in the urban case. Finally, households living in the Pacific region face a higher probability of being poor. In summary, a tenant household headed by a middle-aged or young female and with 3 children under 7 should face about a 32 percent higher probability of being poor than the average rural household. * Although lower than in urban areas, unemployment rates escalated during the recent economic slowdown. Unemployment peaks for the young (aged 18 to 25) and those with intermediate skills. Unemployment has also most strongly hit individuals with some or complete secondary education, as well as college dropouts. As in urban areas, unemployment is clearly associated with poverty. Regionally, the Pacific and Oriental regions suffered the highest increases in unemployment, closely followed by the Central region. The Atlantic region, which had the highest unemployment rate in 1988 and 1995, is now faring better. Extreme Poverty Dynamics * Poverty dynamics decomposition shows that during the 1980s, as in the case for urban poverty, economic growth played a dominant role in the steep reduction in extreme poverty. In the 1988- 95 interval, the 10-percent decrease in extreme poverty came mostly from the fall in food prices and to a lesser extent from growth and better income distribution. In the late nineties, the lack of significant growth and the severe worsening of the income distribution made extreme poverty alleviation come to a near halt. 105 o Poverty line definition is a limitation of this analysis. The definition of the rural poverty line is subject to average national inflation of food-item prices, not that found at the local level. This may cause severe biases in the measurement of poverty if the dynamics of prices of food items differ substantially between urban and rural and across rural regions. Given the intensity of poverty in rural areas, some additional research effort should be devoted to this issue. 106 References "A Survey of Colombia. Drugs, War and Democracy." The Economist. April 21, 2001. Balcazar, A. 1998. "Las transformaciones agricolas en la decada de los n6venta", Centro de Estudios Ganaderos y Agricolas, CEGA, Preliminary Draft. Castanio L. M. 1999. "La distribuci6n de la tierra rural en Colombia y su relaci6n con el crecirniento y la violencia, 1985 y 1996". Tesis PEMAR, Uniandes. Cunningham, W. and W. Maloney. 2000. "Child Labor and Schooling Decisions." mimeo, The World Bank, Washington, D.C. Erazo, J.A., F.A. Galan S., A.M. Ibanez and S. Kirchhoff. 2000. "Diversas causas y Costos del Desplazamiento: ,Quien los Compensa?" mimeo, 194 pp. Final report for Departamento Nacional de Planeaci6n, Bogota, Colombia. GonzAlez, C., C. F. Jaramillo. 1993. "Competitividad sin Pobreza", FONADE, Tercer Mundo Editores. Jaramillo, C. F., C. A. Romero and 0. A. Nupia. 2000. "Integraci6n en le mercado laboral colombiano: 1945-1998". Banco de la Repuiblica Working Paper, Bogota. Jaramillo, C. F. 1998. Liberalization, Crisis and Change in Colombian Agriculture. Westview Press, Colorado. Jaramillo, C. F. 1999. "Macroeconomic Reforms and Agriculture: The Case of Colombia" Mimeo, November 29, 1999. Leibovich, J. and T. Magnac. 1996. "Demanda de educaci6n en las zonas rurales colombianas", (Junio) Coyuntura Econ6mica, Fedesarrollo, Bogota. Leibovich, J., L A. Rodriguez and 0. Nupia. 1997. "El empleo en el sector rural colombiano", Documento Cede 97-08, Universidad de los Andes, Bogota. Leibovich, J. and J. Nunez . 1999. "Los activos y recursos de la poblaci6n pobre en Colombia." Documento CEDE. Bogota: Universidad de los Andes, Centro de Estudios sobre Desarrollo Econ6mico. Leibovich, J. 1996. "El desempenio econ6mico de los migrantes." Revista de Planeacion y Desarrollo. 27 (4): 47-66. L6pez, H., A. Cardona, and J. Garcia 2000. Empleo y pobreza rural 1988-1997. Colombia: Tercer Mundo Editores. Lorente L, A. Salazar y A. Gallo. 1994. "Distribuci6n de la propiedad rural en Colombia", en Transformaciones en la estructura agraria, Minagricultura, 80 ahos. TM editores, Banco Ganadero, Caja Agraria, Vecol. Misi6n de Estudios del sector Agropecuarios. 1990. "El desarrollo Agropecuario en Colombia", Tomo I, Editorial Presencia, Bogota. National Anti-narcotics Agency, Colombia. 2000. "Colombia's war against drugs. Actions and results, 1999 - 2000." 2000. Nuinez, J. and R. Bemal. 1998. "El desempleo en Colombia: Tasa natural, desempleo ciclico y estructural y la duraci6n del desempleo." mimeo, I.D.B., Washington, D.C. Ocampo, J.A. 1989. "Ciclo cafetero y Comportamiento Macroecon6mico en Colombia, 1940-1987." ConyunturaEcon6mica. 21(3): 125-158. 107 Ocampo, J.A., M.J. Perez, C. Tovar, and F.J. Lasso. 1998. "Macroeconomia, ajuste estructural y equidad en Colombia: 1978-1996." In Politica Macroeconomica y Pobreza en America Latina y el Caribe. Ganuza, E., L. Taylor and S. Morley, eds. UNP, CEPAL and IDB. Partridge, W.L. with J. Arboleda. 2001. The Population Displaced by Armed Conflict in Colombia, mimeo, 24 pp. Colombia Country Unit, The World Bank. "Spraying Misery". The Economist, April 21, 2001 Torres, F. and J. Nuinez. 2000. Geography and Economic Development in Colombia: A municipal approach. IDB, Research Network Working Paper. R-408. United Nations Drug Control Program. 2000. Global Illicit Drug Trends 1999. United Nations Drug Control Program. 2001. Global Illicit Drug Trends 2000. Wagner, M.L. 2001. "Illegal Drugs and Development in Colombia and Bolivia." Draft, 27 pp. The World Bank, Washington, D.C. 108 CHAPTER III THE REVERSAL OF INEQUALITY TRENDS IN COLOMBIA, 1978-1995: A COMBINATION OF PERSISTENT AND FLUCTUATING FORCES Carlos Eduardo Vdlez, Jose Leibovich, Adriana Kugler, Cesar Bouill6n and Jairo Nuifiez ABSTRACT Between 1978 and 1995, Colombia made a U-tum in income inequality and underwent significant changes in key socio-demographic characteristics and labor market indicators. The dynamics of inequality were asymmetric. Rural and urban inequality both diminished until 1988. Afterwards, rural inequality kept falling but the opposite happened in urban areas with a predominant effect at the national level. This paper measures the specific contributions of determinant factors to that evolution. On the basis of a microeconometric model of household income changes in inequality are decomposed into variations due to changes in (i) the returns to human assets -education and experience and unobserved ability-; (ii) individual/household endowments in these assets; and (iii) in laborforce participation and occupational choice behavior. Ourfindings show that even periods with moderate changes in inequality conceal strong counterbalancing effects of equalizing and unequalizing forces. Some of these factors are persistent while others are less stable and strongly dependent on economic conditions. Persistent forces are related to demographics and labor supply, - family size, labor female participation, and educational endowments-. Moreover, the forces behind changes in the distribution of individual earnings have a relatively moderated impact on the distribution of household income. Two of our main findings are contrary to our expectations. The increasing participation of less skilled women generated asymmetric effects between household and individual wage distributions. Second, the much larger skill wage differentials prevalent in the urban areas, explains why education endowment equalization worsens income inequality in urban areas but improves it in rural areas. Finally, although the aggregate effect of fluctuating forces is greater in size -vis-ai-vis persistent ones-, the long-run inequality trends is best predicted by the aggregate effect of changes in family size, increasing labor participation of women and the equalizing expansion of education. JEL Codes: D63, J24, J31 anda 015 109 1. INTRODUCION By the late 1970s, the Colombian economy had completed two decades of consistent reduction in income inequality. For some time income inequality in Colombia was exemplary of Kuznets' well- known inverted U-shaped curve: after the growing inequality of the first half of the XXth century, substantial inequality reductions were observed during the 1960s and 1970s as the economy grew. Yet, the inequality improvements became marginal during the late 1970s and the 1980s, and inequality took a "U-turn" in the late 1980s, completely reversing the equity gains of the two previous decades. The rise in national inequality during the 1988-95 period in Colombia was driven by a large increase in inequality in the urban sector, as well as by the simultaneous increase in inequality between urban and rural areas. At the same time, Colombia experienced significant changes in the socio- demographic characteristics of the population. Between 1978 and 1995, the most significant changes in these respects were the following: i) higher educational attainment of the labor force -particularly among women - and greater work experience; ii) a drop in fertility leading to smaller family size; iii) a fall in the gender earnings gap; iv) pronounced fluctuations in the structure of wages by educational level; and v) increased female labor market participation. At the same time, the Colombian economy was subject to major structural reforms and macroeconomic changes that modified key labor market parameters and affected labor market performance through different channels. The structural reforms of the early 1990s covered several areas: trade liberalization and trade integration agreements with neighbor countries; liberalization of the capital account; and major changes in labor and social security legislation. The latter increased the relative cost of labor with respect to capital, and became a source of difficulty for job creation. In addition, the economy suffered supply shocks linked to major discoveries of oil reserves. Rural economic activities experienced a marginal shift from agriculture, stncto sensu, and industry to mining and services. In addition, during the late 1970s and early 1980s, agriculture was subject to a faster process of concentration of land and rural credit. Finally, that sector was hit by a set of negative shocks in the early 1990s: lower tariff protection, real exchange appreciation, lower international prices, drought and violence. The purpose of this paper is to decompose the dynamics of income inequality -urban and rural- so as to measure the specific contribution of some of the preceding factors to changes in income inequality. Within a micro-simulation framework based on a reduced form model of individual earnings and participation in the labor market, the following factors are evaluated."9 (i) The returns to observable human assets (education, experience) and individual characteristics like gender, location or occupational status; (ii) the changes in the distribution of these assets and individual characteristics in the population.; (iii) the changes in labor force participation and occupational choice behavior; and finally, (iv) the changes in the overall effect of unobservable earning determinants.. This technique is used to decompose the changes in inequality and measure the contribution of each of the preceding factors for the periods 1978-88, and 1988-95, and both for individual earnings and household income. Our findings show that periods with moderate changes in inequality conceal strong counterbalancing effects of equalizing and unequalizing forces. The strongest determinants of individual income distribution dynamics are returns to education, schooling endowments and the effects unobservables on earning inequality, in addition to family size and non-labor income for household income. Some of these factors are persistent while others are less stable and strongly dependent on economic 119 The methodology for the decomposition analysis is presented and discussed in chapter 2 of this book. 110 conditions. The analysis also shows that the forces behind changes in the distribution of individual earnings differ in intensity from those that determine change in the distribution of household income. A combination of persistent and fluctuating forces characterizes the dynamics of urban income inequality in Colombia between 1978 and 1995 and explain the reversal that took place in 1988. The persistent forces are linked to demographics and labor supply: the evolution of family behavior - smaller family size and increased labor participation of women- and the growth of educational endowments. The unstable or fluctuating factors tend to respond to changes in the labor demand function, namely to its labor skill profile. Although the aggregate effect of persistent factors is moderate in size -vis-a-vis fluctuating ones-, it is perhaps the best indicator of long-run inequality trends. Some of these effects are also present but of much less importance in the rural sector. Two of our main findings are contrary to our expectations. Intuitively, a greater and more egalitarian education endowment in both urban and rural areas is expected to reduce income inequality. However, according to our decomposition exercise, this was only true in rural areas. Paradoxically, education endowment equalization deteriorated the income distribution in urban areas in both periods -1978-88 and 1988-95. Some basic analysis shows this apparent contradiction is explained by the strong convexity of the earnings functions and by the larger inter-quintile differences in returns to education prevalent in urban areas -with respect to rural areas. Secondly, increasing female participation in the labor market generated asymmetric effects on income per capita distribution vis-a- vis individual labor earnings distribution -regressive for the former and progressive for the latter. As it is shown below, this surprising discrepancy is easily explained with a simple statistical line of reasoning. The paper is divided into four parts. In the first section, we examine the evolution of inequality and poverty indicators for three years: 1978, 1988 and 1995, as well as the changes in some labor market indicators and in the distribution of socio-demographic characteristics. We also briefly review the main structural reforms and macroeconomic developments that affected labor market performance. In the second section, we model the income generating process and provide estimates of parameters that describe the evolution of the structure of earnings and participation behavior. The following section discusses the outcome of the decomposition exercises, which measures the contribution of different factors to the total inequality change. Finally, we summarize and conclude. 2. THE COLOMBIAN INCOME DISTRIBUTION BETWEEN 1978 AND 1995 2.1 The recent U-turn in inequality Several authors have identified the mid 1960s as the breaking point in the regressive trend of income distribution during the first half of the XXth century.120 However, the evolution over the last two decades suggests that a break in the other direction took place. The reduction in inequality was steady from the mid 1960s until the late 1970s. Inequality then plateaued from 1978 to 1988 and then increased significantly from 1988 to 1995, practically erasing the equity gains of previous decades. 121 12'Berry and UrTutia (1976), Urrutia (1984), Reyes (1987), Ocampo (1992), Londono (1995) All these authors give evidence of significant distributional improvements dunng the 1970s. (Misi6n de Empleo, 1986). Ocampo et al. (1998) explain this evolution by: 1) the reduction of the rural labor force surplus, due to fast migration in the 1950s; 2) the fast pace of capital accumulation and modernization in the rural sector; and 3) larger and better targeted investments in education and health delivered through the "Frente Nacional" - see Selowsky (1976) and Velez (1996) for an in depth analysis of these expenditures. 121 The choice of years 1978, 1988 and 1995, which is justified by the availability and comparability of distribution data is also justified for long-run economic comparison. In those three years, economic activity is almost at the peak of the business cycle. Growth is close to or higher than 4% and unemployment is low, between 8% and 10%. 111 As may be seen in Table 3.1, household income inequality indexes for urban and rural areas are relatively stable from 1978 to 1988 but exhibit opposite tendencies during the 1988-95 period. In urban areas, the Gini coefficient is flat and the Theil index fell a little in the first period. Some reduction of inequality in the upper tail and some increase in the lower tail of the urban distribution are revealed by the simultaneous drop in the Transformed Coefficient of Variation and the increase in the mean log deviation index. Table 3.1. Decomposition of total inequality between ruiral and urban areas. Colombia, 1978, 1988 and 1995. 1978 1988 1995 Urban Rural Total Deconwositon Urban Rural Total Dcootsition Urban Rural Total Deonpoastion Between WVtlun Between Within Between Within Household Inequality Gm Coefficient 50 2 43.5 53.9 50 2 444 54 1 544 407 561 M4emalogdemiation.E(O) 380 338 447 8 36 42.5 37.3 496 9 40 50.5 300 558 13 42 Thell. E(1) 526 346 560 8 48 50.3 35.0 552 8 47 706 29.4 747 11 63 thx Coeffof vanauonE(2) 1536 60.3 1704 7 163 105.1 50.5 122.2 7 115 2827 458 331.5 10 321 Populanon Share 57 4 42.6 60.2 39 8 607 39.3 InconleSbhae 761 239 79.0 210 82.6 174 Relative Inome (to the tman) 1.3 0 6 1 3 0.5 14 0 4 UrbanRural Urban Urban Rural Urban Urban Rural Urban CGnt Coeffirtents Males Females Males Fentales Males Fenwles All tdnviduals 47 8 38.5 44 7 39.0 50.3 36 6 WageEarrer 42 1 327 39.5 343 450 39.1 Self-Employed 60 8 54.0 5365 590 59 4 57 4 Swine DANE. fxei NacnaMl d Ho1r.L Autm' catic After 1988, urban inequality deteriorates significantly as indicated by all summary inequality measures reported in Table 3.1.122 In rural areas the evolution is almost identical between 1978 and 1988: the Gini and the Theil deteriorate a little, and the lower and upper tail inequalities show the same rise and decline as in urban areas. From 1988 to 1995, however, rural inequality follows a different path. A clear improvement is noticeable in all inequality indices shown in Table 3.1. This improvement in the rural income distribution was not sufficient to prevent national inequality to rise under the pressure of the increase in the inequality of urban incomes, which represent approximately 80 per cent of national household income. It is true that, in addition, the urban-rural income gap increased since 1988, as urban income per capita nearly doubled between from 1978 to 1995 whereas rural income only increased by 50 percent. The latter evolution is of little importance to explain the overall worsening of the national distribution of household income, however. Most of the increase in national inequality after 1988 is explained by changes within urban areas, whereas the limited changes in the national income distribution during the previous decade reflect parallel distributional changes within both urban and rural areas. In view of that relative autonomy of the evolution of urban and rural inequality and their clear contribution to overall inequality, both sectors will be analyzed separately in the remaining of this paper. In urban and rural areas, the inequality of earnings among all employed persons follows a pattern somewhat similar to household inequality. Data from 1978 to 1988 reveal a pronounced fall in inequality for all individual urban workers (bottom of Table 3.1 ) and stability for rural workers. 122 Figures in Table 3.1 confirm previous studies that examined the changes in urban income inequality. For example, Ndfiez and Sanchez (1998a) find a decrease in the Gini coefficient from 0.47 in 1976 to 0.41 in 1982 and an increase to 0.48 in 1995. Similarly, Ndffez and Sanchez (1998b) find a decrease in the variance of the log wage from 0.65 in 1976 to 0.59 in 1986 and an increase to 0.64 in 1996. See Ocampo et al (1998) as well. 112 From 1988 to 1995, rural inequality of individual earning falls slightly whereas urban inequality increases quite significantly. To conclude this short review of distributional trend in Colombia since 1978, it is worth mentioning that, despite income inequality fluctuations, social welfare in urban Colombia improved substantially and unambiguously from 1978 to 1988 and from 1988 to 1995. This is because the doubling of income per capita compensated all changes in income distribution. In rural areas, welfare improvements are unambiguous between 1978 and 1988, but somewhat ambiguous between 1988 and 1995. V6lez, et al. (2001) find first order stochastic dominance in both periods in urban areas, and during the first period in rural areas as well. However, from 1988 to 1995 in rural areas, second order stochastic dominance is only satisfied up to the 90th percentile. 2.2 Main forces driving the dynamics of income distribution The purpose of this paper is to identify the forces that shaped the changes of income inequality within urban and rural areas during the 1980s and early 1990s. Before turning to a detailed analysis of them, we first review the social and demographic developments that may have affected the distribution of income directly, or through the supply of labor. In addition, we will sum up the simultaneous structural reforms and macroeconomic events that had a major impact on the demand side of the labor market. 2.2.1 The Evolution of the socio-demographic structure of the working population Higher and more egalitarian school attainment Urban education became higher and more equally distributed throughout the period. The proportion of urban workers with complete primary education or less fell by nearly 20 percentage points - see Table 3.2 - whereas their average number of years of schooling went up from 6.4 to 8.9 years. A more detailed analysis also shows that the increase in educational attainment was greater among women, specifically among younger women who either caught up or surpassed men. This general increase in education came with some equalizing of schooling attainment. For instance, the coefficient of variation of the number of years of schooling in the 1975 born cohort was half what it was four decades earlier. Progresses in educational attainments were also observed in the rural population - the average number of years of schooling went up from 2.1 to 3.9. Overall, however, it remained considerably behind the urban sector. As for urban population, the inequality of educational achievements fell substantially. Higher labor force participation -particularly among women. Changes in labor force participation have been substantial over the period, especially among women. Table 3.2 shows that the average employment rate for women increased from 37 to 51 percent in urban areas and from 20 to 30 per cent in rural areas. Interestingly enough, most of this gain in labor force participation was amongfemale household heads or spouses. Overall, the share of wage earners in the urban labor force remained relatively constant at around 44 percent. However, the proportion of men employed as wage earners decreased noticeably, thus suggesting that a higher proportion of women was employed as wage workers. This tendency was still clearer in rural areas where women entering the labor force tended to concentrate in wage work in commerce ands services (L6pez, 1998). 113 Table 3.2. Labor Market Endicators Urban 1978 1988 1995 Mate Femk Total Male Femle Total Male Female Total Labor Maret Statistics AveragnepiY-nt 88 9 370 62 4 88 6 43 3 644 90.4 51.0 692 Empbyed by pKhi 614 38 6 100 58 7 41 3 100 56.6 43 4 100 Unempblyntrat 69 103 82 78 13.9 10.3 6,8 114 8.8 Ubrbang Populaton by Gmups Percenta of wa ner 64 2 25 6 43 5 597 28 9 43 3 563 32 8 43 7 Pecenta ofsefemployed 240 100 165 283 12.6 199 33.4 163 242 hactive 119 644 401 12.0 585 36.8 10.3 509 321 Total 100 100 100 100 100 100 100 IUD 100 Ave rgms per nmth (ihCol 1995) 239 150 211 253 182 Z28 296 206 261 Avra wavspeh- r(ThCoS1995) 13 08 12 0.9 15 12 Rural 1978 1988 1995 Male Femle Total Male Female Total Male Female Total Labor Market Satastcs A Wma.npbyrientRate 768 196 49.1 790 26.5 530 761 296 531 Eniyeby m1er 814 186 1000 75.6 244 1000 72.5 2750 1000 Unernploment rawe 13 54 Z1 23 8.9 40 Z6 970 4.7 %ir*ng Populaton by Groups Pecentalaof waseamer 465 76 26.7 479 137 306 469 16.5 317 Percenta of self empbyed 264 82 171 27.2 10.1 18.5 26.0 12.5 193 biactive 271 842 56.2 249 76.2 50.9 271 71.1 491 Total 100 100 100 100 100 100 100 100 100 Avwammigper month(rCot1995) 106 68 99 118 86 111 115 86 107 Source DANE, Enojta Naciral de Hogues Audths ctam Table 3.3. Changes in Socio-Denmographic Characteristics, urban and rural areas 1978,1988,1995 Urban Rural 1978 1988 1995 1978 1988 1995 Age structure of the population in working age (Percentage) 12-24 34.9 28.4 23.7 47.2 44.6 40.5 25-34 27.4 32.7 32.7 18.4 20.7 22.1 35-44 18.5 20.9 24.3 15.0 15.8 16.8 45-65 19.1 18 19.2 19.4 18.9 20.6 Education structure (labor force) (Percentage) Illiteracy 4.2 2.1 2.1 37.9 22.1 19.8 Primary 43.6 32.8 26.8 54.3 60.5 57.8 Incomplete Secundary 28.9 28.8 27.4 6.6 13.0 15.8 Complete Secundary 11.2 19.8 24.9 1.0 3.6 5.3 Incomplete Superior 6.3 7.1 8 0.1 0.5 0.8 Complete Superior 5.8 9.5 10.8 0.1 0.3 0.6 Total 100 100 100 100 100 100 Average number of years of education 6.4 7.9 8.9 2.1 3.4 3.9 Household size 5.1 4.3 4.1 5.9 5.1 4.7 Source: DANE, Encuesta Nacional de Hogares. Authors' calculations. Decreasingfertility rates Table 3.3 shows that family size fell in urban areas from 5.1 persons in 1978 to 4.3 in 1988 and 4.1 in 1995. For the average household, this change in size produced, other things being equal, an increase in per capita income of 24 percent, which represents a fourth of the total gain in real earming per capita for the average Colombian household over the period. This evolution was even more 114 pronounced in rural areas. Overall, this reduction in family size affected all income groups, although in different proportions. It may be seen in figure 1 that, in urban areas, family size fell proportionally more for lower-middle-income households. Figure 3.1. Average household size by income decile. Urban Colombia 1978, 1988 and 1995. 8 7.5 o 6.5 - 0 Co 6 '- 3.5 ID 5.5 5~~~~~~~~~~~~~~~~~~~~~~~. <4.5 4 -1978 -- 1988--e1995 3.5 1 2 3 4 5 6 7 8 9 10 Income decile 2.2.2 Macro events and changes in demand for labor The growth performance of the Colombian economy was satisfactory between 1978 and 1995. GDP per capita grew at an average annual rate of 1.8 percent. But the growth rate was higher by one percentage point between 1988 and 1995.123 Labor demand was less dynamic, an evolution likely to have affected the evolution of the income distributions. Employment growth fell quite significantly after 1990. Several macro-events and structural reforms during the early 1990s explain this lack of dynamism of labor demand for less skilled workers. These are: (i) an exchange rate appreciation and labor legislation reforms in the early 1990s that increased the relative cost of labor relative to capital; (ii) a tendency of domestic industry to invest in more capital intensive technology, as exposure to international competition rose due to tariff reductions and regional trade integration; and (iii) a gradual re-composition of productive activities towards more capital intensive activities, as production shifted from agriculture and industry to mining and services. The substantial rise in payroll taxation in the 1990s also slowed down the demand for unskilled labor and the generation of wage-earning jobs'24, despite the labor reform of 1990 (Ley 50), which reduced labor costs by diminishing the expected value of the cost of dismissals -"cesantias". Only one factor helped to reinforce the demand for low-skilled labor: the five-fold increase in construction activity in the early 123 Cycles were not completely absent. The economy went through a moderate recession period during the first half of the 1980s In the second half of the 1980s, macroeconomic policy kept a competitive exchange rate and a moderate public deficit. Interest rates fell and some trade restrictions were lifted. Non-traditional exports grew at a high pace. After a low level of activity in 1991, possibly due to trade liberalization, the economy recovered in 1994 and 1995. 124Payroll contributions increased 13 percentage points (!) - up to 13.5 percent for pensions and 12 percent for health insurance. This was on top of preexisting payroll taxes of 9 percent, earmarked for labor training, and social welfare programs. In summary, including remunerated annual leave and 'semester premia' the reforms of 1993 lifted total payroll contributions to 59.4 percent for regular workers. Cardenas and Gutierrez (1996) show the increasing complementarity of skilled labor and capital in the Colombian manufacturing industry. 115 1990s, closely related to exchange rate appreciation, which derived from unprecedented capital inflows.'25 On the agricultural side, the first half of the 1990s was characterized by a set of negative circumstances and policy measures that produced a major reduction in output. The removal of import controls, the lowering of tariffs, exchange rate appreciation, low international prices, scarce credit, frequent drought and increasing violence all contributed to a substantial agricultural decline (Jaramillo, 1998). Changes in rural credit and land ownership should have had more direct effects on the distribution of income. The 1974-1984 decade witnessed an increase in the concentration of land ownership (Lorente et al., 1994. However, this trend reversed in the subsequent decade, when the Gini coefficient of land ownership went down from 0.61 to 0.59. The same egalitarian evolution occurred in the credit market. Until 1984, credit and interest rate subsidies were concentrated among large-scale producers. But a shift occurred between 1984 and 1993.The controls over interest rates gradually crumbled and credit tended to de-concentrate (Gutierrez, 1995). 3. THE DETERMINANTS OF HOUSEHOLD l[NCOM1E: 1978,1988 AND 1995 The explanation of the dynamics of income distribution relies on some representation of household income generating behavior in the various periods under analysis. Household income is modeled as the outcome of two interrelated process: (i) the determination of labor earnings as a function of observed and unobserved individual characteristics (ii) the individual decision to participate to the labor force as a wage worker or a self-employed and/or his/her probability of being employed.'26 This section presents the main results of the estimation of earning and occupational choice equations, and highlights the most prominent changes in underlying individual or market behavior that should bring about changes in the distribution of income during the 1978- 1995 period. Urban and rural earnings are modeled independently. In each case, four separate Mincerian earning equations are estimated for the logarithm of self-employed' and wage workers' earnings and for each gender. Explanatory variables are the number of years of schooling and potential labor experience, and location. Both schooling and experience include quadratic terms that control for heterogeneity in returns by levels of schooling or experience. For urban areas, equations for men are estimated by OLS, while a two-stage Heckman selection bias correction is used for women. For rural areas, the Heckman correction is applied to wage earners for both genders but OLS is used for the self- employed because the selection bias failed to be significant. Occupational choice behavior is estimated as a multinomial logit model with three possible status on the labor market: 1) self-employed, 2) wage earner and 3) inactive. This model is estimated separately for household heads, spouses and other members of the household - with gender dummies being included in each case. The same occupational model is used for all individuals at working age in rural areas. Explanatory variables include the variables likely to affect the potential individual earnings -schooling, experience, region and gender - variables that describe the earning and domestic production capacity of all other household members - i.e. household composition summarized by number of household members by gender and age group, average schooling, and average experience. 251n addition to the liberation of the capital account in 1993, major discoveries of oil reserves, which resulted in a jump in oil exports revenues contributed to expectations of exchange rate appreciation in the second half of the 1 990s. See Cardenas and Velez (1997) F261or detailed model specification, see Chapter 2. 116 3.1 Changes in the earnings equations The eight panels of Table 3.4 show the individual log earning regressions male and female wage earners and self-employed in urban and rural areas for the three years considered in this analysis. Table 3.4A. Earnings equations of wage and self-employed male and female urban workers: 1978, 1988, and 1995. Variable 1978 1988 1995 1978 1988 1995 Male Wage Earners Male Self-employed (OLS) (OLS) Constant 9.0234 9.5537 9.8234' 8.4609 8.9284' 9.361 1' Schooling 0.0474 0.0027 -0.0379 0.1232 * 0.0901 0.0321 Schooling2 0.0046 * 0.0055 0.0075 0.0007 0.0024 0.0051 Expenence 0.0727' 0.0541 0.0476 0.0867 0.0561 * 0.0536 Experience2 -0.0011 -0.0007 -0.0007 -0.0013' -0.0007 * -0.0007 Residual Vanance 0.5142 0.457 0.5211 0.885 0.7913 0.8156 No. of Observations 2,234 9762 8,534 834 4,635 5,059 R2 0.4774 0.4659 0.3983 0.2818 0.3216 0.3029 Female Wage Earners Female Self-employed (Heckman Correction) (Heckman Correction) Constant 9.2313 * 9.3672 9.4141 8.2978 8.3962 8.8958 Schooling 0.0267 t 0.0383 -0.0015 0.0361 0.0457 0.0254 Schooling2 0.0049 0.0034 0.0062 0.0068 0.0063 0.0061 Experience 0.0399 0.0416 0.0337 0.0342 0.0461 0.0448 Experience2 -0.0007 -0.0006 -0.0006 -0.0004 t -0.0006 -0.0006' Residual Variance 0.4587 0.458 0.4934 0.8905 0.9159 0.9127 No. of Observations 4046 18676 17,621 4,046 18,676 11,837 Chi(2) 774 3229 3,082 201 792' 844 Note: Regional Dummies are omitted from the table * Indicates significance at the 1% level or better, indicates significance at the 5% level, and t ndicates significance at the 10% level. For all years and for all occupational status, coefficients have the expected sign and are generally highly significant. The positive estimate of the quadratic term on education reveals that the marginal rate of return to schooling increases with schooling within all groups - except for male rural self- employed in 1995- the reverse being true of experience, as predicted by the Mincerian model. 117 Table 3.41B. Earnings equations of wage and seff-employed male and female rural workers: 1978, 1988, and 1995. Variable 1988 1995 1988 1995 Male Wage Earners Male Self-employed (Heckman Correction) (OLS) Constant 10 4208 10 7522 9 2593 9 2058 School 0 0221 -0 0050 0 0749 0 0738 School2' 0 0021 0 0042 0 0005 -0 0005 Age 0 0668 0 0474 0 0656 0 0730 Age2 -0 0008 -0 0005 -0.0006 -0 0007 Atlantic -0 3041 -0 2729 -0.0335 -0 0317 Onental -0 2324 -0.0454 .0 2765 -0 2297 Central -0 2345 -0 2016 0 0583 -0.2490 Model Chi2 1237 2 1970 0 Adj R2 _ _ 0 1243 0 1180 No observations 4438 4691 2515 2604 Female Wage Earners Female Self-employed (Heckman Correction) (OLS) Constant 9 8676 10 0758 10 5254 10 0828 School 0 0800 0 0527 0 0636 0 0647 School2 0 0015 0 0021 0 0014 0.0035 Age 0 0576 0 0508 0 0040 0 0186 Age2 -0 0005 -0 0005 0 0000 -0 0001 Atlantic -0.2306 -0 1884 -0 1274 0 1923 Oriental -0 1947 -0 0025 -0 5907 -0 0297 Central -0.1825 -0 1305 -0 1065 -00722 Model Chi2 1028 6 1081 3 Adj R2 0 4211 0.3877 No observations 1300 1645 965 1246 ' Significant at 5% level Source DANE. Encuesta Nacional de Hogares Authors' Figures 3.2A. and 3.21B. Change in income from changes of returns to education, relative to workers with complete secondary education. Male and female wage earners, Urban Colombia, from 1978 to 1988 and from 1988 to 1995 40°ho \ | ~--78-88 140% O 30%/o 30 °20°6 20%- 2 10% \ 16h \ C 1D °°/°-________0 -- o -10°h%~~ a 1° -20% -20% 1 2 3 4 5 6 7 8 9 101112131415161718 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 years 6f schooling years of schooling Figures 3.2.A to 3.2.D show how the changes in parameter estimates for schooling affected wage differentials across schooling levels for urban male and female wage and self-employed workers. 118 Figures 3.2.C. and 3.2D. Change in income from changes of returns to education, relative to workers with complete secondary education. Male and female self-employed, Urban Colombia, from 1978 to 1988 and from 1978 to 1995 40%40 3 1- - -78-88 88-95 1 / E 30 Yo l---78-88 88-95 O 20% _ __ / \ 1-0% -\ -= -- D10% \ 0-10% ~ ~ ~ ~ ~ ~ ~ ~ 20 ~~~~~-10% ~ ~ ~ ~ ~ ~ ~ ~~~~-0 -20Yo _ 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 -2D% 1 2 3 4 5 6 7 B 9 10 11 12 13 14 15 16 17 18 years of schoodng years of schooling Changes in schooling returns clearly contributed to flattening the earning-schooling profile of men between 1978 and 1988, and therefore to equalizing the earnings distribution. Indeed, it may be seen that the relative income of low educated workers increased much more than that of more educated. In comparison no change took place for self-employed women whereas middle educated wage female workers seemed to lose in comparison with other educational levels. The evolution was radically different between 1988 and 1995. For men, relative incomes increased both at the lower and the upper end of the distribution of schooling, with a priori ambiguous effects on inequality. The same was observed for female wageworkers, as in the previous period, whereas the evolution was unambiguously equalizing for self-employed women. This evolution of earning differential with respect to education is broadly consistent with the macroeconomic factors that affected the labor market since the early 1990s - capital deepening and complementary demand for skilled workers at the top of the distribution, construction boom and demand for unskilled workers at the bottom. Compared to the urban labor market, returns to education in rural areas behaved similarly, but showed more heterogeneity over time and across labor groups (Table 3.4B). Returns to education increased with school attainment, except for self-employed males, in 1988 and 1995. As in the urban case, the convexity of the earning equation with respect to schooling decreased from 1978 to 1988 and increased again after. The variance of the residuals of the earning equations represents the joint dispersion across earners of the rewards to unobserved skills, measurement error and transitory components of earnings.'27 Table 3.4A shows a reduction in that variance between 1978 and 1988 and an increase between 1988 and 1995 for all urban male earners, whereas changes are somewhat limited for urban women. Observed 127 This variance may also reflect differences in working time. In imperfectly competitive labor markets, it may also reflect heterogeneity on the demand side of the market. 119 changes in that variance seem large enough to affect the inequality of individual earnings and that of household incomes.'28 It is clear from Table 3.4 that shifts in earning differentials across gender and occupational groups depend on the characteristics of earners. Considering a person with 8 years of schooling and 10 years of experience in urban areas leads to a small increase in the male/female wage differential but a large drop in the differential between men (wage workers or self-employed) and self-employed women. Most of the resulting substantial drop in the male-female earning gap is actually taking place between 1988 and 1995. In the rural sector, the same exercise with a reference individual with 3 years of schooling leads to a continuous substantial drop in the earning differential between male self- employed and wage workers but an increasing gender wage differential in favor of men. Changes concerned with experience are of very limited amplitude. Regional differences declined for all groups between 1978 and 1988, but did the opposite during the 1990s.'29 3.2 Changes in Participation and Occupational Choice Behavior Occupational choices are modeled as a multinomial logit. Three choices are considered: inactivity, wage work and self-employment. Dependent variables include all characteristics of individuals as well as summary characteristics for the household they belong to. The estimation is made independently for household heads, spouses, other male and other female adult members. The main features of occupational choice behavior within these various groups of individuals and their evolution over time are summarized in what follows. Urban Labor force participation displays the usual features - see Table 3.5. Higher levels of education increase the probability of being employed, in particular for spouses.'30 Participation decreases with experience or age for household heads and spouses, but it tends to increase for other household members. Spouse participation is particularly sensitive to demographics and household potential income. It falls with the number of children in the household and with the average human capital endowment - education and experience - of other household members. The latter effect is quite strong.'31 From 1978 to 1988, changes in the average participation tate are insignificant among male household heads. It was very substantially positive for spouses and female household heads, and negative for other household members. All this is in full agreement with the aggregate evolution shown above in Table 3.3. More interestingly, this evolution was not neutral with respect to education, but the bias depends on the group being considered. Married women' participation increased more among the least educated - see Figure 3.3 - whereas participation declined relatively more for the least educated secondary male household members. From 1988 to 1995 participation kept increasing for all women, with the same bias towards the least educated. Other male household members also saw a tilt of participation in favor of the least skilled. As in the preceding period, changes in participation among household heads were negligible. 128 Less inequitable access to land and credit should be associated with lower residual variance of rural labor earnings. 12The difference between the largest regional premium and the largest regional penalty declined throughout the period for male and female wage earners and for self-employed men. For example, the difference fell from 10.9 and 16.3 percent between 1978 and 1988 to 4.4 and 2.7 percent from 1988 to 1995 for male and female wage earners. 30 As usual, the participation of household heads is uniformly high - above 98 percent. 131 A 17% drop in participation was associated in 1978 with a drop of average education of other household members from college to primary. 120 Table 3.5. Marginal Effect of Selected Variables on Occupational Choice among Wage Earners, Self- employed, and Inactive. For Urban Heads, Spouses and Other members of the household, and All Rural Workers: 1978, 1988, and 1995. Variable 1978 1988 1995 1978 1988 1995 Urban Household Heads Self-employed Inactive Constant -33.8% -23.5% -2.2% -16.9% -19.7% -20.4% Schooling -0.1% -0.2% -0.9% 0.2% -0.1% -0.2% Experience 1.2% 0.9% 0.7% 0.0% 0.1% 0.2% Gender (Female) -13.8% t -13.0% -14.7% 15.6% 12.8% 11.3% Cluldren <2 4.1% 0.8% -1.0% -3.7% -1.0% 0.0% Children2-5 1.7% 0.9% 1.1% 0.1% -0.6% 0.0% Children 6-13 0.9% 0.8% 2.6% -0.9% -1.1% -0.9% t No. Observations 2587 12657 12104 2587 12657 12104 Pseudo R2 0.1812 0.1418 0.1364 0.1812 0.1418 0.1364 Urban Spouses Self-employed Inactive Constant -13.1% -19.6% -22.4% 23.6% 28.2% 26.8% Schooling 0.3% 0.7% 0.7% -1.5% t -2.6% -3.2% Experience 0.2% 0.3% 0.3% 0.0% -0.3% -0.2% Children < 2 -3 2% -2.0% -2.8% 4.8% 7.9% 10.3% Children 2-5 -0.7% 0.4% -0.5% 2.4% 2.1% 4.2% Children 6-13 -0.3% 0.2% 1.8% 1.3% 0.8% -0.8% No. Observations 1931 9586 9233 1931 9586 9233 Pseudo R2 0.0909 0.0907 0.0898 0.0909 0.0907 0.0898 Urban Other Self-employed Inactive Constant -10.0% -15.1% -10.0% -3.2% 3.4% -7.3% Schooling 0.0% 0.2% 0.0% -3.2% t -3.0% -2.8% Experience 0.4% 0.5% 0.6% -0.1% -0.1% 0.0% Gender (Female) -9.8% t -11.9% -7.9% t 34.5% 28.5% 23.2% Female * Children <2 -2.8% -1.1% -1.4% 17.3% 15.1% 12.8% Female * Children2-5 6.1% -2 0% 0.5% -0.6% 5.2% 5.5% Female * Children 6-13 -0.2% 0.7% 0.4% 4.2% 1.6% 1.8% No. Observations 3009 12787 11437 3009 12787 11437 PseudoR2 01432 0.1185 0.1219 0.1432 0.1185 0.1219 All Rural Workers Wage Earner Inactive Schooling 0.9% t 0.4% 0.7% -1.6% t -0.9% -15% Age 1.4% 1.0% 0.8% -2.0% -1.9% -1.8% No. Observations 13084 18781 19992 13084 18781 19992 Pseudo R2 0.394 0.3419 0.3277 0.394 0.3419 0.3277 No:eTheexcludedcategones are wage-eamers for urban and self-empbloyed for rural Signrficancemdicators ofMultmonialLogrtEstumators at the 1% evel, at the 5% level, and tat the 10% level Somevasiables used m themDdelare not mcluded m thetable. Urban Average educ levelof the household,averageyrs ofemper,ence,no ofothernralesbetween 14&65yrs old,no ofotherfenales between 14&65yrs old,no ofotherrnues olderthan5.,no of otherfemialesolderthanM6 RuraLPop lessthanu2ytsold,pop betweenn3&5ys old,pop between6&9yrs old,isalebetween 18 & 65 yrs old, fenale between 18 & 66 yrs old, inal older than 65, female olderthan 65, average educ level of household, avetage age ofhousehold, 3regionalduttnes,fenrle,household head,and spouse "ttndicates both wtth and wihout fetaiedunms, fenab dunsmes were not included in the Spouse Logs tniodel Source. DANE, Encuesta Nactonal de Hogares Authors' cakulatons The negative impact of family size on female labor force participation shifted over time too. It ended up concentrating among spouses in households with very young kids but most of that evolution took place between 1978 and 1988 - see figure 3.3. With respect to the effect of the characteristics of other household members on spouse participation, figure 3.3 shows an interesting evolution. It would seem that.the increase in spouse participation tended to concentrate first in households with a relatively 121 higher potential income, as summarized by the average educational level of non-spouse members. But between 1988 and 1995, that increase concentrated more among less educated households. This feature will prove important below. Figure 3.3. Probability of being employed or a wage earner according to various individual or household characteristics: Urban Colombia, various groups of household members, 1978, 1988 and 1995 . Spouses Spouses 100% - 1006 >~90% gt 90% .2 E 80% oSfz E 80% - 370% /70% so 670%% - 1 --1978 60% 50% -- ~~~~~~~~~~-1988 50% Xo 40% --1995 40% . . . . . 30% 30% 5 6 7 8 9 10 11112 13 14 1516 1 2 3 4 5 6 7 8 9 1011 121314151617 18 Years of schooling Years d schooling of other household members Other females Spouses (complete high school) 100% - 00 CXE; 890% _D-S 80% G 70% - 70 - 0L 80% 380 70% - 50 2 40% 240% 30% - ........ ... 30% 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 Years of schooling Number of children (0-2) Male heads Other males (complete high school) 100% 100% 90%- ID ° 1 80% -- a E 70%- 0 0 60% - 0 co 50% - 50% - EL 40% - 40%- 30% - . . . . . . . . . . . . . . 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 Year of expenence Years of schooling Concerning the choice between wage work and self-employment, estimates conform with what is observed elsewhere. Wage work tends to be more frequent for younger and more educated individuals. The effect of education tends to be more pronounced among spouses and other household members.132 The education gradient for wage employment became positive and significant for household heads in 1995 too. Over time, two evolutions are noticeable. First, all males are moving 132 Ten years of schooling represent an increase of 10 percent in probability of being a wage earner, given that one is employed. 122 somewhat away from wage work. Female workers do the same, to a smaller extent, during the 1978- 88 but spouses and females heads go back into wage work in substantial numbers after 1988. Second, the choice between wage work and self-employment tended to become more sensitive to differentials in age and education. Rural The main features of occupational choice behavior in rural areas are similar to those observed on the urban side. Labor force participation tends to increase with education and age in every period. Wage work tends to increase with school attainment and decrease with age. Over time, the most noticeable changes are the increase in female participation with a bias towards wage work. However, female participation rates remained very limited. Another noticeable evolution is that, with time, the sensitivity of wage work to education tended to concentrate more in higher educational levels. In summary, the main noticeable features of the evolution of individual earnings determinants and occupational choice behavior are the following: o A reduction in the convexity of earnings with respect to schooling for most workers during the 1980s followed by increase in that convexity - especially for women - during the 1990s. * A fall in the dispersion of the rewards for unobserved skills or unobserved earning determinants among male workers between 1978 and 1988 and an increase in the dispersion of these rewards between 1988 and 1995 for all workers, except self-employed women. o A fall in earnings differentials by gender and between the wage and self-employment sectors during the 1980s, and a rise during the 1990s. o A continuous increase in female participation, more pronounced among spouses and female household heads and in the urban than in the rural sector. This evolution was differentiated by individual and household characteristics. Overall female participation increased more among urban less educated people. The shift from wage work toward self-employment among urban male workers was more pronounced among the least skilled. The same evolution took place among women between 1978 and 1988 but it reversed afterward. 4. UNDERSTANDING INCOME DISTRIBUTION DYNAMICS IN COLOMBIA: FACTOR DECOMPOSITION BY MICRO-SIMULATIONS, 1978-1988 AND 1988-1995 We now move one step further and decompose the evolution of inequality by isolating its dynamic response to the changes in skill prices, in structural parameters of participation and occupational choice, and in skill endowments over time. We follow the framework established in the second chapter of this book. This section presents the results of the decomposition proposed there of the changes in the distribution of both individual labor earnings and household per capita income for the periods 1978 to 1988 and 1988 to 1995. For reasons discussed at the end of this section, the analysis focuses mostly on urban areas. The decomposition methodology consists of various steps. First, the changes in the distribution of individual labor earnings is decomposed into changes due to: (i) changes in retums to observable human assets -education and experience- and regional location premia; (ii) changes in the residual variance of earnings equations, or retums to unobserved productive characteristics; and (iii) structural changes in labor force participation and occupational choice. Then, remaining unexplained changes may be attributed to changes in the socio-demographic structure of the population, i.e., in the 123 distribution of individual 'endowments'. Among them we single out the paradoxical effect of the equalization of the educational endowment. Finally, we apply the same methodology to observed changes in the inequality of household income per capita and explain the reason for divergences vis- a-vis the results obtained with individual earnings. In the household case, we also examine the role of changes in family size. 4.1 Urban Areas: persistent and fluctuating forces behind the reversal of the inequality trend 4.1.1 Individual Labor Earnings During the 1978-88 period, income inequality among individual workers fell markedly. The Gini coefficient lost 3.1 percentage points, from 47.8 to 44.7. The decomposition of the change in inequality shown in Table 3.6 allows us to identify four major equalizing forces partially counterbalanced by two unequalizing factors. The equalizing forces are: (i) changes in returns to education; (ii) a drop in the variance of the residual term in the male earnings equation; (iii) a squeeze in earning differentials between gender and type of occupation (self-employment versus wage work); and (iv) changes in returns to experience. The unequalizing factors are (i) the change in the distribution of educational endowments and (ii) the shift in labor force participation behavior. On the aggregate, equalizing forces dominate and the change in inequality is negative. 33 Table 3.6. Decomposition income distribution changes for households and individual workers. Changes in the Gini coefficient. Urban and Ruiral Colombia (1978 - 88, 1988 - 95) Returns Endowments Change Pamrcip Residual Non-Labor Gina at on variance Family Income Education Experience Regions Constant Total Education Size Urban 1978- 1988 All Individual Workers -3 1 -2.3 -0.7 0 0 -0 8 0.8 -2.4 2.3 3.0 - - Households 0.0 -1.9 0.1 0.1 -0.2 0.7 -1.5 2.4 2.3 -0.6 -0 9 1988-1995 All Individual Workers 5 5 0.0 -0.1 0.0 0.0 0 6 2.5 2.5 1.2 - - Households 4.2 -0.1 01 0.1 0.4 -0.4 1.1 29 0.8 -04 0.4 Persistetr (P) and Fluctuating (F) Factors F/P P F P/F F P P F Rural 1988-1995 All Individual Workers -2.4 -1.2 -1.3 -1.1 n.a -0.9 -1.8 -0.4 -1.3 - Households -3.7 -0 I 0 1 -0 2 n.a -04 -0.6 - -0.4 -0.7 rLa. This evolution reversed between 1988 and 1995. The Gini coefficient rises from 44.7 to 50.3,134 completely erasing the gains in inequality experienced during the previous decade. Some of the 133Note that this aggregate evolution hides divergent evolutions across gender groups. From 1978 to 1988 there was a reduction of inequality among male workers, but a sigmficant increase for female workers. 134 This decomposition at the individual level is also consistent with the profile established for the different groups of workers; in fact, during that period, both occupational male groups and female wage earners experience a significant rise in inequality (see Tables A.I.A-D in the Appendix). Simultaneously, however, self-employed females exhibit opposite 124 preceding forces played a similar role during this period. However, most equalizing forces remained inoperative or even reversed direction, as with the variance of the residual term, whereas unequalizing factors remained present (Table 3.6). Most of the increase in inequality during this period is explained by three regressive forces: (i) the larger variance of the residual term in the earnings equation; (ii) the expansion of education endowments in the working population; and (iii) the change in occupational choice behavior We now analyze in more detail all those factors which had some influence on the evolution of the earnings distribution during one period or the other. Returns to education As seen in the preceding section -see figure 3.2 and Table 3.4- changes in returns to education were strongly equalizing during the period 1978-1988, especially among male earners. By itself, this evolutions explains a reduction of 2.3 points in the Gini coefficient of individual earnings. From 1988 to 1995, changes in returns to education were without effect on the level of earning inequality, as measured by the Gini coefficient. However, this neutrality is only apparent. It hides important counterbalancing adjustments at both tails of the schooling range. Practically for all groups of urban earners a simultaneous rise in the relative earnings of both the least and the most skilled workers took place - with the only exception of self-employed women as shown in figure 3.2. These counterbalancing forces resulted in no change in the Gini coefficient for the universe of all earners. But, such a change in the distribution is essentially ambiguous rather than truly neutral. Some other measures of inequality could have shown a rise while others would have shown a drop in inequality. 135 Residual Variance The fluctuation in residual variance shown in Table 3.2, mostly for male wages, affected income inequality accordingly. It brought a reduction in the Gini coefficient of 2.4 points for all workers for the period 1978-1988 and the opposite evolution in the subsequent period. By definition it is impossible to identify what is behind this phenomenon. It might be an economic phenomenon, for instance an increase in the relative return of the specific talent of people at the bottom of the distribution. But it might also simply be noise in the data. In the latter case, the overall drop in the Gini coefficient between 1978 and 1988 could be thought to be largely artificial, because very much influenced by the residual variance. This would not be the case of the increase observed in the following period, though - as may be seen in Table 3.6. Experience The substantial drop in the returns to experience of male wage workers shown by earning equations - see Table 3.4- as well as the flattening of earning profiles with respect to age, contributed to a 0.7 points fall in the Gini coefficient. Returns to experience kept falling and earnings profiles kept flattening in the second period, but at a much slower pace. As a result their effect on inequality was very small. inequality tendencies. Thus, the consistent behavior among male and female wage earners and their larger mass within the labor force provides the dominant effect observed at the.aggregate level. 135 As an illustration of this, it may be noted that despite very similar evolutions of the returns to education across gender, the Gini coefficient increased for women and did not change for men. 125 Earning differentials across labor-market segments The preceding effects were evaluated by modifying the coefficients of the earning functions in Table 3.4 so as to leave mean earnings constant in the various labor-market segments - defined by gender and occupational status.'36 This indeed permits isolating the change in the slope of the earning function from changes in mean earnings. Now, modifying intercepts so as to obtain the new mean earning, without changing the slope coefficients, permits identifying the distributional effect of changes in earning differentials across labor-market segments. Between 1978 and 1988, the mean relative earnings of the various groups of earners became less differentiated. The (real) mean earning of self-employed males -the highest of all four groups- fell by 9 percent, while the two lowest mean earnings, those of female wage earners and self-employed, increased by 6 and 40 percent, respectively (see Table 3.7)). This caused a moderate drop -.8 points - in the Gini coefficient for all individual earnings. After 1988, the gender gap rose again while the occupational status gap kept going down. From the overall distribution point of view, these two movements canceled each other. Overall, it turned out that the equalizing effect of the increase in the relative earnings of self- employed women canceled the two other unequalizing evolutions. Table 3.7. Mean income. Impact of change in the constant of the earnings equation (%) Relative income, 1988-95 1978-88 1988 (*) Male Wage Earners 0% 3% 1.00 Male Self Employed 7% -9% 1.20 Female Wage Earners -13% 6% 0.80 Female Self Employed 61% 40% 0.68 (*) Relative to wage earners average. Source: Authors' simulations and calculations. Participation and occupational choice The participation and occupational choice effect is also a large unequalizing factor for individual earners in both periods. It explains 0.8 additional Gini points from 1978 to 1988, and 0.6 from 1988 to 1995. Structural changes in participation involve both changes from inactivity to activity, and changes in occupational status, from wage work to self-employment (and vice versa). In order to describe the distributional impact of these phenomena, it is convenient to consider the percentiles of the original distribution of earnings where entries and exits from a particular occupational group take place.' 7 From 1978 to 1988, the two most significant changes in participation and occupational choices (Table 3.8) are a 4 percent shift of the labor force -mostly males- from wage work into self employment -partially compensated by a 1 percent move in the opposite direction- and the increasing participation of women - by 2 percentage points when changes in socio-demographic characteristics are not taken into account, as it is the case here-. 136 This is done by changing both the coefficients of the schooling variables and the intercept, see Chapter 2, page xx. 17 Note, however, that there need not be a direct relationship between the earning level of those who exit a particular occupational group and those who enter it. Indeed, the micro-simulation methodology used in this book describes changes in cross-sections of individuals not the job history of particular individuals. 126 Table 3.8. Simulated changes in participation and occupational choice. Urban Colombia 1978- 88 and 1988-95 (percentage of the employed). 1978-88 1988-95 All Males Females All Males Females Occupational Choice Self Employment into Wage Earninl 0.9% 0.6% 0.3% 0.4% 0.2% 0.2% Wage Earning into Self Employmen 4.0% 3.2% 0.8% 3.8% 3.0% 0.8% Participation 1.7% -0.4% 2.1% 5.0% 0.8% 4.2% It may be seen in figure 3.4A that the shift from wage work to self-employment among male workers has an unequalizing effect on the distribution. Entries exceed exits at the two extremes of the distribution, thus producing a kind of mean preserving spread. Things are more ambiguous for women participation. If it were not for the left hand hump in the curve, the net entry of women in the labor force would be similar to a mean preserving squeeze in the distribution, entries being more numerous at the middle than at the extremes. However, the hump on the left side of the distribution contributes to an increase in the Gini coefficient. Overall, it turns out that the Gini is little affected by the increase in woman labor force participation. Figure 3.4A. Simulated occupational choice Figure 3.4B. Simulated participation changes in changes (from wage earning to self employment) in percentage points (net entries) by percentile of percentage points by percentile of earnings. Males, earnings. Females, urban Colombia, 1978-88. urban Colombia, 1978-88. 10% 7I_P/o 9%/6 6% 8% i 5%__ 7% 41- _ &3hNet entry 61/6 - ~~~~~~3% --- -______ 5% - 4_/ Entry Ext C 2% _,-_ S% _ ___ __, 1% \ ;2%/ - _ / _0% , X_1 0% - - - __ r- 10 o- r 37 in- t;5 rAA 7r-rF9 R7 10 20 30 40 50 60 70 80 90 100 -3% Percentble of indMduai earnigs The evolution between 1988 and 1995 period is very similar to the previous one for the shift from wage work to self-employment, whereas the inequality enhancing effect of the increase in female participation is more unequalizing than before (see Figures 3.4.C and 3.4.D). Altogether, however, these two effects result in an increase in the Gini coefficient for the whole earnings distribution slightly lower than in the preceding period. The key structural changes behind the evolution just described are rather evident. First, the inequality of the distribution of earnings of self-employed is much larger than that of wage earners. The shift from wage work to self-employment thus tends to increase inequality as shown in figures 3.4A and 3.4C. This effect is possibly reinforced by the selectivity of that shift, and in particular the increasing likelihood for older cohorts and the least educated - see figure 3.3 - to be self employed. Second, two phenomena are at play behind the distributional impact of change in female participation. On the one hand, increase in participation tended to be more pronounced among the least educated. On the other 127 hand, that increase was higher in well-educated households during the period 1978-1988 and among least educated households in 1988-1995. The latter phenomenon explains the difference in the shape of the curves shown in figures 3.4B and 3.4D. Altogether the effects of changes in returns and in participation/occupation behavior explain a drop of 5.4 points in the Gini coefficient between 1978 and 1988 and a rise by 3.1 points between 1988 and 1995. The difference with actual changes point to a strongly unequalizing effect of 'endowments', that is the change in the socio-demographic structure of the population, of 2.5 and 2.3 points of the Gini respectively. The nature of the phenomena behind this residual of the decomposition analysis will be taken up below in connection with household income inequality. Figure 3.4C. Simulated occupational choice Figure 3.4D. Simulated participation changes in changes (from wage earning to self employment) in percentage points (net entries) by percentile of percentage points by percentile of earnings. Males, earnings. Females, urban Colombia, 1988-95. urban Colombia, 1988-95 10% _ 12% 9% 8% _10°6 7% ~ ~ ~ ~ ~ ~ ~~~~~~4 6% - L \Entry____ ___ _ 6%-\ 2% 6 /-% - Net e nt 4% - -4_ 0%- l1/ l . . .l 0 . 10 20 30 40 50 60 70 80 90 l(0 10 20 30 40 60 70 80 90 100 -2% - --- Percentile of Indrdual earnings Percentile of indcidua earnings 4.1.2 Household income inequality: the role of structural parameters From 1978 to 1988, the dynamics of urban household income inequality and that of individual income inequality are quite dissimilar. Lower inequality of individual labor earnings coincides with unchanged household inequality (Tables 3.1 and 3.6). It may be seen in Table 3.6 that most of that difference is explained by the much lesser equalizing role of changes in earning equations - returns and residual variance - for household income than for individual labor earnings. During the 1988-95 period, both the distribution of household income and individual earnings become substantially more unequal, this evolution being slightly less pronounced for household income. Table 3.6 shows that the difference comes essentially from differences in the effect of changes in occupational choice behavior, residual variance, and earning differentials across labor market segments. These various differences between changes in household income and individual earnings distributions are now taken up in turn. Participation and occupational choice It turns out that female participation is what explains the discrepancy between household and individual distribution dynamics from 1988 to 1995. While the occupation and participation effect was unequalizing for individual earnings, the same changes in the coefficients of the occupational model is equalizing at the household level - minus 0.4 versus 0.6. Figure 3.5, which shows the mean 128 change in employment by percentiles of the household income distribution helps to understand what is going on. In both periods, it was seen that female participation increased more for the least educated - figure 3.3. This would explain that net gains in employment are higher for the bottom than for the top half of the distribution of household income. However, another feature of the change in participation behavior must be taken into account. It is that participation tended to increase more in richer households, as summarized by the average educational level of non-spouse adult members, between 1978 and 1988, the reverse being true afterwards. As a consequence, the equalizing effect of increased relative female participation among the least skilled is counterbalanced by the household income effect in 1978-1988 - i.e. the upward sloping right end of the curve in figure 3.5. On the contrary, it is reinforced by the household income effect in 1988-1995, leading to a drop in inequality. It has been seen above that the same phenomena increased low earning labor force participants and contributed to a increase in the Gini coefficient for individual earnings. Figure 3.5. Changes in employment rate by income percentile. Females, 1978-88 and 1988-95, Urban Colombia. 8% 7%- 1978-88 5% 4%. \ 3%. 2% - 1% 0 10 20 30 40 50 60 70 80 90 100 income percentile Earning differentials across labor market segments From 1978 to 1988, the shift in relative earnings associated with the constant term was much less equalizing at the household level (-0.2 versus -0.8 for individuals). Gender gap reductions benefit female wage earners who are in the upper lower half and at the middle of the individual earnings distribution but more than proportionally belong to middle-high income households. It is therefore the case that closing the gender gap tends to make the distribution of individual earnings more equal but the distribution of household income less unequal. This effect dominates for the distribution of household income on the period 1988-1995. Note on the other hand that the change in the differential of earnings between wage earners and self-employed is likely to have less impact at the household level because of some diversification of household members across occupational status. Experience During the 1978-88 interval, the equalizing effect of changes in returns to experience at the individual level are not visible at the household level. While the Gini coefficient for individuals fell by 0.7 points, the change is insignificant - it even increased by 0.1 points - for households. The explanation of this discrepancy is similar to the preceding one. Experienced workers tend to be in the upper half of the individual earning distribution, so that a drop in their relative earnings contributes to more 129 equality. However, they also happened to be fairly uniformly spread in the distribution of household income. Therefore their relative earning gains do not contribute to a reduction in inequality at the household level. Residual variance Finally, we note that the sign of the effect of the change in the residual term of earning functions is the same for household income as for individual earnings, but the absolute value is smaller. This simply reflects diversification within the households due to the presence of multiple earners. As it was assumed that the residual term were not correlated within households, increasing its variance leads to a larger increase in inequality at the individual than at the household level. 4.1.3 Endowment effect: education, family size and non-Rabor income We now come to the endowment effect. In Table 3.6 it is found essentially as the residual of the returns and the occupational choice effects with respect to actual changes in the Gini coefficient. However, it is also possible to simulate directly the impact on the distribution of individual earnings or household income of a change in the distribution of specific socio-demographic variables. This is done simply by importing from year 1988 to year 1978 the education level of individuals or the size of household, conditionally on the gender, the age and the region of residence of individuals or household heads. The same is done with 1988 and 1995 to evaluate the education and family size endowment effect in Table 6. 138 A somewhat unexpected result is that changes in educational endowments have a substantially regressive impact on both household and individual income inequality. The Gini coefficient would thus have increased by 2.3 points and 3.0 respectively if schooling levels had been distributed in 1978 like those observed in 1988, and by 1.2 and 0.8 respectively if 1988 schooling levels had been those of 1995 (Table 3.6). This effect is surprising because, as we saw in Section 1, the distribution of educational endowments among urban - as well as rural - workers becomes less unequal as the new cohorts entering the work force were on average more educated and less differentiated in terms of years of schooling. One would expect that more equality in education would lead to a reduction in income inequality. However, a simple argument shows that this is true only if the relative returns to an additional year of schooling are constant across schooling levels. As emphasized above, this is not the case in Colombia. In effect, earning equations show that the earning profile is convex with respect to education, the most educated workers benefiting proportionally more from an additional year of schooling than the others. This is true for wage earners and self-employed as well as across gender. Moreover, it was seen above that this convexity tended to increase over time. 139 To show that the convexity of the (log) earnings with respect to schooling is responsible for the rise in inequality attributed to the expansion of education, consider the following simplified framework. Two individuals (i = 1,2) have different marginal rates of return to education, r , this rate being higher for the most educated individual (rj < r2). -Also define inequality by the ratio of incomes Y2/Y,. If both individuals experience an increase in schooling, AS,, the proportional rise in labor income a, is by definition equal to 138 See detail of the procedure in Chapter 2. See equations in Tables A2 and A3. . 139 Fortunately, separate simulations of change in mean educational endowment and change in the distribution of education, reveal the equalizing effect of the latter and the dominant regressive effect of the former. Hence endowment equalization alone has an income equalizing effect -as expected-. 130 a, = r,ASi, i=1,2. For inequality to remain constant, changes in schooling should be such that both individuals have the same percentage change in income, a = a, = a2, that is r1 AS, = a = r2 AS2 Therefore, since individual 1 has a lower rate of return to education than individual 2, the increase in education for the latter must be smaller than for the former. That is, the ratio in changes in schooling should be inversely related to the ratio of rates of return: ASI = AS2 (r2/ r1) > AS2, since it was assumed that (rj/ r,J > 1. It follows that, if marginal rates of return are increasing with the level of schooling, changes in schooling will generally increase income inequality unless they are sufficiently progressive. - i.e smaller in absolute value for more educated people. Consider for instance a progressive change in educational endowments: AS2 < AS,, such that (r1/ r2) SU8l;w z ir o|..................... i;FL v s S NICARAGUA !!PVw%W4V1L A*,s COLCMB1A W*9q T / / -Childcare g 50°% g}v - Primary education CD 40°/ - Secondary education 308/6 - -Tertiary education 20% - 10% / 1 2 3 4 5 6 7 8 9 10 Deciles 175 Figure 5.6. Coverage of educatioinaD services over time 100% 90% 80% 70% z 60% - **-Pdmayl: 1992 X /-0-Primary: 1997 50% - Secondary: 1992 (3 40% - , " -'Secondary 1997 30°b - a-/-Tertiary: 1992 20%-- - TedtLary 1997 10%- 0%- 1 2 3 4 5 Quintles Figure 5.7. Market share of the public sector Tor different social programs, 1997 100% 90% - 28 80% 9 t 70% -rhWcare 60% Mni"~~~~~~- edcatton 50SO% \ Secodry edwatbucalko -Tert" yeducatlon 340%- ; 30% _/ o~~~~~~~~~~~~~~~~Tralntig X20% 10% 0% 1 2 3 4 5 6 7 8 9 10 Decile Figure 5.8. Market share of the public sector in childcare over time 100%/ _1 800° 35 70%o 60 60% D - 50%/a 400 30S 1=l~~~~~~992\[ 1 2 3 4 5 6 7 8 9 10 Deciles 176 Figure 5.9. Market share of the public sector in primary education over time 100% 90% -i i 50% _-1974 40 -1992 j 4%_ 1997 w 20% 1 2 3 4 5 6 7 8 9 10 DBcDes Figure 5.10. Market share of the public sector in secondary education over time 100% 80% U - 4 0% 50% ~ {30% 0% 1 2 3 4 5 6 7 B 9 10 40% D tO0% 7 10% 60% 10 0% £) 40% 30% 20% 2 10% -1974-1992 - 199 0% 1 2 3 4 5 6 7 8 9 10 Docile Fiur 51. are sar f hepblc eco i triay dcai177e tm 4.2. Healthcare Health insurance coverage almost doubled between 1992 and 1997. As mentioned above, the Colombian health care system underwent major reforms in 1993. A dramatic consequence of this was that coverage of health insurance increased from 31 to 58 percent. Furthermore, increases in health insurance coverage were concentrated in the lower income deciles. For example, insurance coverage in the first decile rose tenfold from 4 percent in 1992 to 41 percent in 1997, whereas in the tenth decile, coverage rose more modestly, from 65 to 80 percent. Not surprisingly, the concentration coefficient of the newly insured population is quite progressive, equal to -0.17. Three quarters of the newly insured entered the contributory regime. Of the 30 percent of the population that became newly insured by 1997, three quarters entered the contributory regime and the remaining quarter entered the subsidized regime. The concentration coefficient for those entering the contributory regime was 0.17, and -0.38 for the subsidized regime. The l[nstituto de Seguridad Socil -hereafter IISS- continues to play an important role in the reformed health care system and has tended to attract new subscribers from the middle-income deciles. A third of those entering the contributory regime became affiliated to ISS-the dominant EPS33 provider. As a result, ISS increased its share of the market from 15 to 21 percent over this period. The concentration coefficient of ISS affiliates fell from 0.37 in 1992 to 0.27 in 1997, indicating that the marginal incidence of new ISS affiliates was more progressive than the average incidence. Indeed, new ISS affiliates had a concentration coefficient of only 0.05. This is substantially lower than the concentration coefficient for new EPS affiliates, which was 0.20. The implication is that ISS attracted more of lower-income subscribers to the contributory regime than did the other EPS's. In fact, the proportion of new affiliates for ISS peaks in the middle-income deciles. The concentration coefficient for these new affiliates is 0.05. However, the other EPS's (private sector providers) have absorbed the bulk of the new business and appear to be capable of meeting the needs of the poorest. Some two thirds of new affiliates -a clear majority- opted for one of the other EPS's in preference to ISS. It is especially striking that in the first income decile, twice as many clients are affiliated to other EPS's rather than to ISS. Moreover, on average, affiliation to the other EPS's is more progressive than to ISS, with a concentration coefficient of 0.20 versus 0.27 for ISS. There are still deficiencies in the scope and progressivity of the subsidized regime. There is evidence that lower income households are more likely to be covered by the subsidized health insurance regime. However, even in the first decile there are three times as many people affiliated to the contributory regime as to the subsidized regime. Furthermore, the proportion of people enrolled in the subsidized regime does not decline very rapidly as income rises. This may well reflect the nature of the SISBEN targeting mechanism discussed above. Given that 20 percent of the population have eligibility under the SISBEN system, while only 7 percent are enrolled in the subsidized health insurance regime, there is clearly still a significant shortfall in coverage. Less than 15 percent of public health spending is allocated to the subsidized regime, with the remaining amount going to traditional supply-side subsidies to public healthcare providers (Vargas and Sarmiento, 1997). In addition, the average value of the demand-side subsidy falls well below the average cost of the obligatory healthcare package. Although insurance rates increased, treatment rates fell between 1992 and 1997. Ironically, although the proportion of the population covered by health insurance doubled between 1992-97, the proportion of 33 Empresa Promotora de Salud 178 those who were sick that reported receiving treatment fell by 8 percentage points. At the same time, the absolute number of people receiving health treatment rose by 13 percent. This could suggest that the lower treatment rates may in part be attributable to the larger demands that were being placed on the system. Alternatively, it may simply reflect differences in the design of the surveys across the two years. Notwithstanding, treatment rates remain higher and more progressive than insurance rates. While the overall coverage of health insurance is still only 58 percent, 74 percent of ECV97 respondents reported that they received healthcare when they became sick. The gap between coverage rates and treatment rates is even larger among lower income groups. For example, only 40 percent of those in the first decile have insurance coverage; however, 65 percent report receiving treatment when it was required. Meanwhile, in the tenth decile coverage rates and treatment rates were about equal, at 80 percent of the population. The existence of people who receive treatment without coverage provides some evidence about the group known as 'vinculados' who have access to public hospitals as long as they can cover 30 percent of the charges. Table 5.9. Beneficiaries growth and change in coverage rates from 1992 to 1997 Education Health Utilities Income 4umnule Chzldcare Pnmary Secondary Tertiary Insurance Treatment Electr Water Sewerage Change in Coverage (% points) Quintile I 9 13 3 -5 35 14 9 14 15 Quintile 2 7 7 2 -7 31 0 1 1 3 Quintile 3 9 14 5 -9 28 0 0 -2 -2 Quintile 4 5 12 6 -9 21 0 0 -1 I Quintile 5 12 13 17 -17 15 -3 1 1 -2 Total 8 12 6 -7 27 3 2 3 3 Growth rate of the number of beneficiaries (%) Quintile I -10 44 168 -19 428 36 41 55 79 Quntile2 10 18 113 7 199 27 34 34 39 Quintile 3 9 20 101 -5 128 21 33 30 30 Quwntile 4 11 6 93 2 74 12 34 33 35 Quintile 5 19 -12 96 2 36 5 33 33 29 Total 4 15 99 -4 116 19 35 36 38 Equity of coveraae Rrowth Concentration NA -0.563 0.403 NA -0.169 -0.531 -0.002 -0.013 0.000 Coefficient Note: A more negative concentration coefficient means that more than proportional benefits go to the poor. Figure 5.12. Coverage and composition of health insurance over time 60% 50%/ is ESS 40%/ 0) * EPS 01 ~ 300/6 o _ _ 0 Double insurance 20% _ 10% -/_ . Pnvate irsuarice 1_ 1l R~~~~~ Ofer public oo/¢ W_I , Rll ~~~~~lir6urarnce 00/. L3 1992 1997 E1 Year_ 179 Figure 5.13. Coverage rates for health insurance and treatment by decile, 1997 90% 80%Yo -Health irrsurance 70%/6 - (subsidized) 60%/ -Health insurance c' 50% (conibutory) 40' - Health 80° /minsurance 30°/6 - (oterall) 201%- -Health treatment 10% _ _ 0% 1 2 3 4 5 6 7 8 9 10 Deciles 5. THIE DISTRIBUTIONAL IMPACT: THE POOR BENEFIT SUBSTANTIALLY FROM OVERALL PSS The total value of social programs is estimated at C$5,000,000m. By combining data on participation in social programs from the ECV97 together with information on the unit value of subsidies, it is possible to study the value of the subsidies received by each income decile and hence the distributional incidence of these social programs. On this basis, the estimated overall value of subsidies distributed via social programs in 1997 was C$5,500,000m (Table 10), equivalent to 2 percent of GDP. This can be compared with the total value of PSS for 1997, which was C$15,200,000m. The discrepancy between the two figures indicates that the social programs covered in this study represent only about one third of total PSS. Data limitations preclude a more complete analysis. Subsidies are equivalent to a third of the income of households in the first quintile. The subsidies embodied in the various social programs are equivalent to 5 percent of household income at the national level. However, their importance as a proportion of household income varies substantially across quintiles. For the first quintile, the subsidies received via social programs represent 32 percent of household income, a fraction that falls to 2 percent for the fifth quintile. Thus, although in absolute terms subsidies are fairly evenly distributed across quintiles, their relative impact is still far greater on low- income households because they represent a higher proportion of their household income. Furthermore, the effect is somewhat greater in rural areas than in urban areas due to the lower baseline level of rural income. By far, the largest share of IPSS goes into state education. Of this total sum, the greatest part was spent on state education (80 percent) and public health insurance (13 percent). In rural areas, this pattern becomes even more accentuated, with 88 percent of subsidies going to state education and only 6 percent to health insurance. Information on utility subsidies is only available for water and sewerage in urban areas. However, these data show that the combined value of water and sewerage subsidies exceeds the value of health insurance subsidies, at least in urban areas.34 34 Taking into account the total value of energy subsidies in 1997, the share of educational subsidies would drop substantially. 180 Rural areas receive 5 percent more of PSS subsidies than their population share. Consequently, spending per capita in rural areas, at C$9.2m, is higher than that in urban areas -C$7. Im-. Thus, while for the average urban Colombian, PSS subsidies represent 4 percent of his income, in rural areas they represent nearly 8 percent. The national average is 5 percent. The overall distributional incidence of PSS is broadly neutral. The concentration coefficient for total PSS is -0.06, indicating that PSS is almost neutral with respect to the distribution of income (Table 5.10). Indeed, the figures for the share of total PSS by quintile show that each quintile receives close to 20 percent of the total. The first quintile receives slightly more, at 23 percent of the total, while the fifth quintile receives slightly less, at 16 percent of the total. However, it is worth noting that these estimates only consider the gross distributional incidence of subsidies, without netting out the incidence of tax contributions. PSS improves the distribution of income by 3 Gini points, with a much higher impact in rural areas. The overall effect of PSS on the distribution of income can be gauged by adding the value of subsidies to original household income and recalculating the Gini coefficient. The calculations show that the Gini coefficient at the national level falls by 2.6 percentage points as a result of PSS (Table 5.10). In proportional terms, the effect is slightly larger in rural areas than in urban areas: changes in the Gini Coefficients are, respectively, 3 and 1.9 percentage points. That is, the expected difference between any two rural Colombians as a proportion of mean income became three percentage points smaller after PSS subsidies were taken into account. This magnitude is proportional to the relative size of public subsidies to income and to the difference between the subsidy concentration coefficient of subsidies and the income Gini coefficient.35 As mentioned above, subsidies are less progressive within rural areas (-0.05 versus - 0.03), a fact that would make the change in the Gini coefficient smaller. However, as we already know, the relative size of PSS subsidies to income is twice as large in rural urban areas, more than enough to overcompensate the small differences in targeting. However, individual programs are either strongly progressive or regressive. The overall neutrality of PSS conceals a wide range of variation in the distributional incidence of specific social programs. This variation doubtlessly reflects the relative virtues of the different targeting devices adopted by each of the social programs. The three most progressive social programs are the HCB nurseries, primary education and the ARS subsidized health insurance scheme --all of which have concentration coefficients below negative 0.3-- while the three most regressive programs are housing subsidies (C=0.46), tertiary education (C=0.37) and the family subsidy (C=0. 11). Moreover, distributional incidence can vary significantly even within a particular social program. For example, in the ICBF childcare system, the HCB nurseries have a concentration coefficient of negative 0.35 but for the CAIP nurseries the value is much higher, at negative 0.09. Programns in rural areas are less progressive. Comparing program by program, the concentration coefficients for the rural areas are substantially higher than those obtained for urban areas, meaning that they are less progressive. A particularly salient difference occurs in the case of the public health insurance scheme ISS, which has a concentration coefficient of 0.02 in urban areas versus 0.41 in rural areas. Progressiveness is shown to be inversely proportional to the unit value of subsidies. The redistributive impact of any particular social program depends on two factors: the size of the subsidy and the progressivity of its incidence. The larger the subsidy, and the more progressive its incidence, the larger its effect on the overall distribution of income. An analysis of the social programs under consideration reveals that progressiveness is inversely related to the unit value of the subsidy (Figure 35 Adapting from Kakwani (1977), AGini = [subsidy concentration coefficient - Gini of income] x [subsidies / (income + subsidies)]. 181 5.14). Thus, the programs which distribute the largest subsidies in terms of unit value are also those which are most regressive in terms of incidence, and vice versa. Nevertheless, the total value of subsidies per sector tends to be directly correlated with its degree of progressivity. The programs which are most progressive tend to represent a relatively large overall subsidy, and vice versa. In particular, primary and secondary education c6me out as high-value, highly progressive subsidies. The reason is that, although the unit value of the subsidy is comparatively modest, both of these programs cover a large number of people. A key exception to this pattern is tertiary education, which is a highly regressive and has a high overall value. The general pattern suggests that targeting of marginal PSS is better targeted to the poor than average targeting, supporting the hypothesis of Ravallion and Lanjouw (1998) that by some political economy rationale, the initial benefits of PSS are captured by the upper middle-income classes and, that only when coverage increases sufficiently, do marginal services reach the poor more than proportionally. Table 5.10. Summary of results of analysis of distributional incidence of subsidies at the nadonla level, Colombia, 1997 Subsidy Redistributive Subsidy Size Subsidy Targeting Effect Gini coefficient RRE Value Share Share by quintile (%) Concent. reduction (*) (C$m) (%) 1 2 3 4 5 coeff. AGinix % 100 Childcare o HCB 140,000 2 35 31 22 9 3 -0.35 -0.10 4 1.63 o CAIP 95,000 3 18 30 20 19 13 -0.09 -0.05 2 1.10 o Total 240,000 4 28 31 21 13 7 -0.24 -0.15 6 1.42 Education o Primary 1,400,000 26 35 29 21 12 4 -0.31 -0.96 40 1.55 o Secondary 1,510,000 29 23 26 26 17 8 -0.16 -0.85 36 1.24 o Tertiary 1,500,000 28 5 7 18 31 38 0.36 -0.13 6 0.19 o Total 4,390,000 83 21 20 22 21 17 -0.03 -1.93 82 0.98 Training 1,200,000 2 12 17 24 26 20 0.11 -0.04 2 0.71 Family subsidy per 21,000 0 7 13 27 32 21 0.19 -0.01 0 0.54 child Health o ISS 240,000 5 9 28 31 8 24 0.04 -0.09 4 0.85 o Other EPS 170,000 3 19 16 31 30 5 -0.06 -0.08 3 1.04 o ESS 100,000 2 33 37 18 8 4 -0.35 -0.07 3 1.64 o Total 700,000 10 40 19 20 11 10 -0.27 -0.47 10 1.07 Total 5,500,000 100 23 20 21 19 16 -0.06 -2.60 100 1.00 Note: (*) RRE: Relative Redistributive Efficiency: ratio of share in Gini coefficient reduction to share in total amount of subsidies. Source: ECV 1997 Primary and secondary education make the largest contribution to the redistribution of income. It is interesting to examine the extent to which each of the social programs contributes to the overall change in the Gini coefficient (Table 5.10). The results show that primary and secondary education, respectively, contribute 40 and 35 percent of this overall change -their contribution to the change in the Gini coefficient exceeds their respective shares of the total value of subsidies. Tertiary education, on the other hand, absorbs nearly 30 percent of the total value of subsidies and contributes only 5 percent to the change in the Gini coefficient. The programs with the highest impact per unit spending are ARS and 182 HCB. Since they both represent very small shares of the total value of PSS, however, their impact on the change in the Gini coefficient is modest. Health spending through the ISS systems has become substantially more progressive since 1992. In a limited number of cases, it is possible to compare the incidence of spending with those obtained for 1992 (Velez, 1995). For the health system, comparisons are limited to the ISS system since the EPS and ARS systems were not introduced until 1993, whereas information for non-contributory public health care cannot be readily derived from the 1997 survey. Nevertheless, results for ISS suggest that health spending has become substantially more progressive, with the concentration coefficient falling from 0.22 in 1992 to 0.04 in 1997. On the other hand, there has been little change in the distributional incidence of aggregate educational spending, but evolution by education level has been somewhat heterogeneous. Overall education spending has become slightly less progressive since 1992, with the concentration coefficient rising from negative 0.08 to negative 0.03. Within this overall average, primary and tertiary education became somewhat more regressive, whereas secondary education became somewhat more progressive. Between 1992 and 1997, concentration coefficients for secondary education subsidies became more progressive and decreased from negative 0.12 to negative 0.16; however, the opposite occurred for primary and tertiary, whose concentration coefficients increased, respectively, from negative 0.35 and 0.33 to negative 0.31 and 0.36. ICBF childcare programs -HCB and CAIP- are still the most progressive in distributional terms; however, their targeting deteriorated from 1992 to 1997. In fact, both modalities of childcare programs of ICBF lost some of their distributive incidence in both urban and rural areas. The concentration coefficients of both CAIP and HCB programs increased from negative 0.22 and negative 0.44 in 1992 to negative 0.09 and negative 0.35 in 1997, respectively. On the other hand, the share of resources of the CAIP program within the ICBF - with higher unit cost and less targeted towards the poor- increased from 29 to 40 percent. PSS represented a similar magnitude relative to household income in 1997 and 1992; however, PSS subsidies became less progressive. In order to compare national incidence in 1997 with similar results 1992, we must take into account that, for 1997, we do not report subsidies on two major items -public utilities and special rural programs- which accounted for 29 and 4 percent of the total, respectively. With those corrections included, from 1992 to 1997 the relative size of PSS subsidies relative to income decreased marginally from 5.5 to 5.0 percent; in rural areas, it fell from 9 to 8 percent, whereas in urban areas, it remained at 4 percent. However, public subsidies became less progressive simultaneously as their concentration coefficients increased from -0.13 to -0.06. Other things being equal, the redistributive impact should be smaller in 1997. However, our calculations show that the redistributive impact is almost identical. This must be the case because income inequality is greater in 1997 and, that, in term increases the impact on the Gini. 183 Figure 5.14. Concentration coefficients against unit value of subsidy, by program, Colombia, 11997 0.6 iNURBE 0.4 0.2 Tertiary educatio aD ._ Income support MO0.2- o SENA .2 CAIP~~~2 o 0 - t ° Sectndary education 3 o -0.2 0 Primary education -0. FI1CR ESS health insurance -0.6 Unit value of subsidy (C$m pc pa) Table 5.111. Overall redistributive inmact of subsidies National Urban Rural Incidence (as a percentage of household income) o Quintile 1 3% 21% 32% o Quintile2 13% 10% 16% o Quintile 3 9% 8% 10% o Quintile 4 5% 5% 7% o Quintile 5 2% 2% 3% Total 5% 4% 8% Gini coefficient o Change24 .30 (percentage points) 2.4 1.8 3.0 *Water and sewerage subsidies are excluded to make comparison consistent between areas. 6. EXPANSION PRIORITIES IN SOCIAL PROGRAMS l[n this section, we explore the priorities of alternative social service programs. The aim is to use the consumption behavior of groups of middle-high-income households in their inter-sectoral choice of social services as an efficiency indicator for allocating additional public spending. These groups face much smaller financial constraints and are also more exposed to market forces in their decisions to select social services. The analysis performed allows us to derive the priority sectors for service expansion. Intuitively, service expansion should be larger in those sectors that register the greatest relative difference in access probability between middle-high and low-income groups.36 36 The analytical base of this argument is presented in Velez (1998). 184 Shortfalls in coverage: concentrated among the poor and variant across sectors There are still substantial shortfalls in the coverage of social programs. It has been shown that the current range of social programs achieves a significant degree of redistribution, reducing the Gini for the initial distribution of income by three points. However, the analysis of service coverage revealed that there are still significant segments of the population without access to essential public services. Overall, 66 percent of the corresponding age group lacks access to institutional childcare and 15 percent to primary school. Of those who complete primary school, 25 percent do not go on to secondary education, while of those completing secondary school, 57 percent do not go on to university. Furthermore, 41 percent of the population are still not covered by health insurance, and 57 percent lack connection to the sewerage network. The absolute size of the un-served population varies substantially across programs. Those who are not covered by health insurance represent the largest un-served group, amounting to 16.3m people. The next largest group is that living in households without a sewerage connection, which accounts for 12.8m people, while 6.4m people live in households without access to water. Coverage deficiencies are concentrated among the lower income deciles. It is no surprise to find that the population which lacks access to public services is disproportionately concentrated among the lower income deciles. The concentration coefficients for the population that lacks coverage are always substantially lower than the concentration coefficients for the covered population, and concentration coefficients of the non-covered population are positively correlated with its proportion relative to potential users (Figure 5.15). For example, in the case of water and primary education, the share of the non-covered and their concentration coefficient are close to 15 percent and between negative 0.26 and negative 0.34, respectively. The highest share and concentration coefficients for populations without coverage are found in tertiary education: 46 percent and 0.14, respectively. These patterns are clearly reflected in the proportion of the uncovered population which falls into each decile. For example, in the case of primary education, 19 percent of the unserved population falls in the first decile. An interesting exception is found in tertiary education, where the largest shortfalls in coverage are among the middle classes. Figure 5.15. Unserved potential users by sector: magnitude and targeting 0 20 - a Tertiary educ 010 - Potential 0 00 0 1 t0% 20% 30% 40% 50% 60% 70% 80% 90% 10 % OSecondary educ -0 10- -1 Health treatment m -0 20 -J Health insurance 0 Water OSewerage 0Chitdcare -0 30 ut Eol30 - te tricity -y * oPrimary educ -0 40 185 Erasing coverage gaps vis-a-vis higher income groups, rather than achieving universal coverage, is the appropriate coverage objective. The above measures of service shortfall assume that universal service is always the appropriate coverage objective, but even without financial constraint, it may not be so. In fact, in the tenth decile, where income is not such a binding constraint, coverage of social services is rarely universal (the only cases of universal coverage in the tenth decile are found in water and electricity). An alternative way of looking at the shortfall is to consider to what extent coverage in the lower deciles falls short of the coverage attained in the highest decile or any other income group selected as a point of reference. This can be expressed as a 'coverage gap', defined as the difference in coverage across the poor and higher income reference groups, and - analogously to the Poverty Gap concept - provides an indication of the extent of unsatisfied demand among the lower income deciles for each type of service (Table 5.12). This is equivalent to using Engel curves -in terms of probabilities of access- and comparing income elasticities for the poor across social services. For example, if coverage gaps are expressed as the difference between the first and tenth deciles, the largest gaps are found in tertiary education (54 percent), sewerage (43 percent) and health insurance (40 percent). However, if we pick another income group as the coverage objective, the sectors and their priority ranking may change. Prioritization for expansion across services should be in line with the size of the access probabibity gaps. Velez (1998) shows that the principle stated above -to allocate more resources to the service with the highest access probability gap and implicitly higher income elasticity- produces maximal welfare improvement on marginal spending on public social services.3' It also shows that the sectors of priority and their ranking may vary under different budget constraints. Since the size of the access probability gap varies with the chosen target decile, the highest priority service will not always remain constant. Table 5.12. Coverage of potential users and equity of access for various services Education C Healthcare Utilities Primary Secondary Tertiary Insurance Treatment Water Sewerage Electr. Coverage (%) o Decile 1 80 73 19 24 41 65 78 50 89 o Decile2 82 73 31 22 47 66 78 56 91 o Decile 3 81 77 29 20 50 68 77 58 91 o Decile4 87 75 13 27 55 73 81 64 91 o Decile5 87 76 36 23 59 76 83 66 92 o Decile 6 88 76 27 28 59 74 85 70 95 o Decile7 85 72 33 33 63 79 89 78 96 o Decile 8 92 84 38 35 69 80 91 83 97 o Decile9 92 70 61 38 73 79 95 87 98 o Decile 10 95 79 74 48 81 80 98 93 99 o Overall 85 75 43 34 59 74 86 71 94 Numbers (m) o Potential users 6.19 4.89 1.84 4.43 39.60 6.11 39.84 39.84 39.84 o Covered 5.29 3.69 0.79 1.20 23.26 4.51 33.43 27.00 37.20 o Not covered 0.90 1.20 1.06 3.24 16.34 1.60 6.41 12.84 2.65 Concentration Coefficient o Potential users -0.19 -0.05 0.28 -0.19 -0.05 -0.02 0.01 0.01 0.01 o Covered -0.16 -0.04 0.45 -0.07 0.05 0.02 0.05 0.11 0.03 o Not covered -0.34 -0.07 0.14 -0.23 -0.21 -0.12 -0.26 -0.25 -0.30 37 Following the basic idea of tax reform, Vdlez (1998) model assumes that assumes the status quo of current public and private provision and search for welfare Improvement at the margin. 186 Independently of the welfare target, childcare, sewerage and health insurance and care remain within the top four priorities. Some sectors show certain robustness in relation to the welfare objective. Childcare is almost always the highest priority service for expansion. In general, sewerage comes out as the second or third priority for expansion. Finally, health insurance has all the possible rankings between first and fourth. The prioritization of childcare and health insurance is consistent with results reported in a similar analysis in 1992 (Velez, 1998). However, in 1992, the other priority was secondary education. Yet patterns of growth in coverage since 1992 have only partially followed these prescriptions. Section 4 reported growth in coverage rates by type of service over the period 1992-97. The results showed that coverage of childcare services had actually fallen overall since 1992 across all deciles, while the largest coverage gains have been secured in health insurance and secondary education, which increased by 27 and 12 percentage points and with disproportionate benefits for the poor. To this extent, the public institutions showed some capacity to respond to the priorities derived from the gaps of consumption/human capital investment behavior of different social strata. Due to the high cost of substantial increases in tertiary education coverage, credit expansion should be considered instead. The results indicate that the costs of obtaining universal coverage in tertiary education is about twice as high as the total cost of reaching universal service in childcare, primary and secondary education, and health insurance put together. Even equalizing access probability across deciles with that of the current highest level would cost a sum equivalent to some 2.0 percent of GDP and would require PSS to increase by 40 percent! Therefore, given that tertiary education is not considered an essential public service in the same way that basic education and healthcare are, it will be excludedfrom the simulations below. Perhaps the aggressive expansion of college credit should be seriously considered, for the following reasons. First of all, tertiary education becomes a priority only when the consumption target of the highest deciles is incorporated. Second, given the excellent marginal returns to college education (close to 20 percent) that are still prevalent despite the duplication of tertiary education supply in the nineties, credits should be in high demand and should exhibit moderate risk. Attainment of universal coverage in four basic services would cost the equivalent of 2 percent of the Colombian GDP. The overall cost of obtaining universal coverage in the four basic services (childcare, primary and secondary education and healthcare) is C$2,011,709m, equivalent to 2 percent of GDP. Reaching this objective would imply a 37 percent increase in the total level of spending on PSS. The largest contribution to this cost comes from the childcare sector, which represents just over a third of the total. This is because of the relatively large size of the un-served population (3.2m) and the relatively large unit cost of secondary education (C$0.2m). On the other hand, atching the probability of access for all deciles with that of the highest observed probability would be substantially less costly. The cost of reaching the full Rawlsian objective of equalizing access probabilities across deciles (min=max) is less than half that associated with universal coverage. Overall, it represents just around 0.6 percent of GDP, equivalent to an 11 percent increase in current levels of PSS. The costs of meeting the intermediate decile targets are relatively modest - 0.5 and 0.17 percent of GDP for the eighth and sixth deciles. Indeed, costs rise non-linearly with the target decile, since, as the target decile increases, not only does the target coverage rate also increase, but so does the number of lower deciles whose coverage rates need to be raised. 7. IMPACT OF DECENTRALIZATION An important aspect of the reforms undertaken since 1990 has been the decentralization of health and education. This occurred in the context of a wider move towards the decentralization of the Colombian state, which originated in the 1980s and took root with the Constitution of 1991. As a result of these changes, the current political model in Colombia lies somewhere in between a unitary and a federal 187 state. The departments play a key role as intermediaries between the central government and the municipalities. They are also responsible for the provision of services which, due to economic considerations, are best provided on a regional scale, e.g., hospitals (Vargas and Sarmiento, 1997; IDB, 1998). Aside from political motivation, decentralization was also brought about with the hope that it would result in service improvements. Decentralization in Colombia is primarily motivated by political considerations, in particular the need to reconcile national unity with regional diversity. However, it is also expected that decentralization will lead to improvements in the efficiency and equity of public services. Figure 5.16. Long term trends in the decentralization of public spending 35%/ 25% 20%- n 15% 5% 0% The decentralization process has two facets: certiffication and increased local control of funds. The two are completely independent of each other, in the sense that funding levels are not affected by certification and vice versa. Certification is a formal procedure that transfers decision-making power to local authorities. Responsibilities can only be delegated to departments when they satisfy certain preconditions associated with administrative capacity. A second tier certification process follows in which departments delegate powers to the municipalities. In the case of health, the certification process has been slower than originally hoped; little more than half of all departments and less than 10 percent of all municipalities were certified during the first three years. One factor which has slowed down the process is the absence of incentives for certification, in the sense that certification does not increase the flow of fiscal resources to the government but only increases autonomy in decision-making. In education, results have been faster mainly because certification was made compulsory by 1997. Certification involves a two-tier delegadon procedure: from centraD government to department and from department to municipality. The decentralizatioli model involves a sharing of tasks between departments and municipalities. In the health sector, the municipalities take responsibility for primary health care, while secondary and tertiary care are provided by the departments, while both levels of government are active in the field of public health. In the education sector, the department has overall responsibility for providing, administering and investing in the service. However, these functions are supposed to be delegated to those municipalities with a population in excess of 100,000. The decentralized system also puts considerable emphasis on the managerial autonomy of schools. 188 The volume of funds allocated to local governments has increased substantially, although revenue collection remains under central control. The other important aspect of decentralization is the increasing flow of funds to local authorities. Between 1990 and 1997, local government funding increased by 80 percent and local government increased its share of total government spending from 35 to 40 percent. Although the legal framework for decentralization aims to expand opportunities for revenue collection by departments and municipalities, these remain substantially dependent on transfers from the general budget. These transfers accounted for about 80 percent of local government finance in 1997. The largest component of this transfer is the 'Situado Fiscal,' accounting for 44 percent of total finance. The legal framework, specifically Law 60 of 1993, lays down various rules which define the overall proportion of the central government budget to be allocated to the 'Situado Fiscal;' it amounted to about 26 percent in 1996. Law 60 also defines the way in which funds should be distributed between departments, and the way in which departments should allocate the spending between services such as health and education. The inter-departmental distribution rules put considerable weight on historical baselines, and as a result the per capita transfers for health and education differ dramatically between departments, ranging from US$11 to US$357 per capita in 1996. Current formulas to distribute funding across sub-national govermments are not equitable, promote inefficiency in tax collection, and impede transparency. Alesina et al. (2000) have shown that public funds transferred to local governments (Situado Fiscal and Participaciones Municipales) are defined through a complex set of formulas that end up generating unjustifiable inequities across regions in per capita spending on health and education.38 On the other hand, the efficiency cost of taxation is higher because transfers are completely independent of local collection of national taxes (income and VAT). A redefinition of regional distribution formulas that combines a transfer proportional to local tax collection plus a redistributive component to compensate the poorest regions would be a step in the right direction.9 Some evidence indicates that decentralization had a desirable impact on health and education; however, there is a need to assess how effective the certification process has been in improving efficiency and equity. In the previous section we saw that during the last decade decentralized sectors such as secondary education and health care seemed more responsive to social priorities than a centralized childcare program like ICBF. At the same time, Borjas and Acosta (2000) report convergence in educational outcomes and inputs -such as pupil to teacher ratios- in the main urban areas during the last decade. Moreover, they find that after the decentralization process "the allocation of public teachers was greatest in those areas that had the largest number of potential students per available teacher". To date there have not been any attempts to provide an empirical assessment of how effective certification might have been in improving the efficiency and equity with which health care and education services are provided at the department level (Vargas and Sarmiento, 1997). This section attempts to fill this gap by comparing improvements in efficiency and equity parameters between 1993 and 1997 across departments according to their year of certification. The Encuesta de Caracterizaci6n Socio-Econ6mica de la Poblaci6n Colombiana CASEN93 survey is used to compute efficiency and equity parameters at the baseline, while the ECV97 provides the basis for measuring outcomes at the end of the period. Both surveys are representative at the departmental level. Two methods of analysis are used. First, departments are grouped according to the year of certification, since this gives a measure of how far the decentralization process has gone in each case and thus comparisons of changes in performance parameters between departments belonging to each group can be made. Second, regression models are used to explain changes in perforrnance parameters in terms of years of decentralization and a range of other explanatory factors. 38 They show that "per capita spending in education in Bolivar, for example, is less than half of that observed in Cundinamarca and about one fourth of that observed in San Andres." 39 As proposed by Alesina et aL (2000) 189 Bealthcare and education and the certification process Simple descriptive statistics do not reveal any systematic relationship between improvements in the performance of health care and education and period of certification. For the health sector, the performance parameters considered are the increase in insurance coverage and treatment coverage rates achieved between 1993 and 1997 and the concentration coefficients for health insurance and coverage in 1997. For education, the performance parameters are the cumulative proportion of a particular age group that has reached a specific educational level, and the corresponding concentration coefficients. Departments are grouped according to when they were certified (before 1994 (only for health care sector), in 1995, in 1996, and in or after 1997). A comparison of these parameters indicates that all groups registered substantial improvements in health care coverage and equity during this period. Educational attainment registered little change at the primary level, but improved substantially at the secondary level for all three department groups. In general, educational attainment became more equitably distributed, particularly at the secondary and tertiary levels. However, it is difficult to find any connection between the duration of decentralization and the magnitude of the gain achieved in either health care or education. Nonetheless, it is noteworthy that the overall performance of the education system has impnroved substantially. Although it is hard to find performance differentials between departments with different dates of certification, it is nonetheless clear that the overall performance of the education system has improved significantly, both in terms of coverage rates, and in terms of promotion rates (i.e. the extent to which those entering a particular level of education manage to complete it). For example, in the 6-12 year-old age group, the completion rate for primary school has jumped from 13 percent of entrants to 27 percent, while among 13 to 19-year-olds, the completion rate has increased from 13 to 25 percent of entrants. These improvements may be partly attributable to "automatic" promotion and remedial and recovery courses that have recently become available. There is no evidence that certification has had a significantly positive impact on either health care or education sector perforRmance. The above simple comparisons fail to control for many factors that may explain differences in performance across departments. A more powerful approach is to estimate regression models for coverage rates and concentration coefficients. These models control for the historical starting point, as well as availability of public and private sector resources for health care. The date of certification is included as an additional explanatory variable in this model. If certification has had a positive impact on performance, this could be expected to show-up as a positive and statistically significant coefficient for the corresponding variable. The results indicate that in most cases the certification variable is not statistically significant. An interesting exception is the model for the insurance coverage rate, where certification is significant at the 10 percent level, but has a negative sign. Variables such as historic coverage rates, income per capita, the income Gini and health spending per capita are more important in explaining variations in performance across departments. By and large, the certification variable is not statistically significant in the regression models for education. The one exception is the model explaining the rate of entry into secondary school, where certification is positive and statistically significant. Results for the evaluation of the education and health care sectors should be treated with caution due to the relatively short time period elapsed since the decentralization process began. In both health and education, the decentralization process only began in 1995.4° This means that even for the earliest decentralizers, there are only two years of data available on which to judge the impact of this measure. Given the time lags involved in institutional reform, it is unlikely that the full impact of decentralization would have become apparent over such a short period. This represents a significant limitation of the current analysis and the results should therefore be treated with caution. 40 Note that beginning the certification process is not equivalent to starting decentralization. 190 Figure 5.17. Change in rates of educational attainment against year of certification 0.15 - -E- -Entered primary (6-12 years) r 0.05 - A\--- Completed primary (6-12 0.05 yas .0 *k Entered secondary (13-19 A 0 ~~~~~~~~~~~~~~years) 0 1995_ 1996 17 -s Completed secondary (13- ) 19 years) -0.05 - - Entered tertiary (20-25 years) -0.1I Figure 5.18. Change in concentration coefficient against year of certification 0. 15 a .-.---~~~~~~~~~~0Entered primary (6-12 0 005 years) c w1 -O--Com pleted prim ary (6- -O.0 5 1g6 1t97 12 years) -0.05 a Entered secondary .05 I-. (13-19 years) 8 ~ ~ ~ ~ ~ ~ ~ ~ ~~~~~a ----Cornpleted secondary = -0.156 (13-19 years) -0.2 - . \_-Entered tertiary (20-25 '7y e a rS U -0 25 - yas -0.3 8. SUMMARY AND CONCLUSIONS Public social spending trends and program characteristics The major changes in PSS in Colombia could produce major results. The 1990s were a period with substantial growth in the level of PSS, accompanied by profound structural reforms in the way in which services were organized and delivered. PSS in Colombia grew by 90 percent over the period 1990-97, increasing its share of overall public spending from 30 to 35 percent and its share of GDP from 8 to 15 percent. The fastest growth took place in the health sector, where spending tripled between 1990-97. As part of a wider process of decentralization, the proportion of public sector resources allocated to departmental and municipal governments rose from 35 to 40 percent. Changes of this magnitude could be expected to bring about very significant improvements in the coverage and incidence of public services. The analysis undertaken in this chapter provides the basis for a preliminary evaluation of these reforms, summarized as follows. 191 There is a great deal of diversity in the administrative structures of different social programs. For example, a variety of financing mechanisms are used, including general taxation, hypothecated payroll taxes (at an aggregate rate of 20 percent) and service-specific surcharges. There are also considerable differences in the targeting mechanisms used, with some being based directly on household income, others on housing characteristics, and others on a broader range of socioeconomic indicators. There are also a number of different institutional systems used to administer programs, with some depending on the central government, some on regional government, and others on industry based 'cajas de compensaci6n'. Coverage by income groups All social programs increased their coverage rates between 1992-97, with the exception of childcare and health treatment, where coverage fell somewhat. By far the largest expansions in service came in the health insurance and tertiary and secondary education sectors. In all cases, coverage growth disproportionately benefited the poor. Progress in health insurance coverage has been impressive, but some challenges remain. Perhaps the most dramatic area of change has been in health insurance, where a doubling of coverage was achieved over the period of analysis. Moreover, the expansion in coverage has been shown to be strongly progressive, particularly under the subsidized regime. These results provide a strong endorsement for the structural reforms undertaken in the health sector. Notwithstanding these substantial achievements, certain challenges remain. About 40 percent of the population are still without coverage, while only about a third of those eligible have been enrolled in the subsidized regime. Some doubts can also be raised about the accuracy of the variables used to target eligibility for the subsidized regime. A further concern is the decline in the proportion of the population reporting that they had received treatment for medical problems, in spite of the growth in insurance coverage. Completion rates in primary and secondary education have improved significantly. While the overall coverage of primary and secondary education has only improved modestly by about 10 percentage points, there have been notable gains in the proportion of students who succeed in completing each level of education. For example, in the 6-12 year-old age group, the completion rate for primary school has jumped from 13 to 27 percent of entrants , while for 13- to 19-year-olds, the completion rate has increased from 13 to 25 percent of entrants. Tertiary education has undergone a dramatic expansion, largely attributable to the private sector. There have been modest improvements in net enrollment ratios for primary and secondary education during the 1990s, of around ten percentage points in each case. However, the major changes have come in the tertiary sector, which doubled its enrollment between 1992-97. The public sector was only responsible for 20 percent of the increased enrollment, and hence this benefit cannot be entirely attributed to increases in PSS. Furthermore, it is very questionable why tertiary education, which has such a regressive incidence, should continue to account for about one third of educational spending. The childcare sector appears to have been largely neglected during the 1990s. Both in the 1992 study and in the present chapter, childcare was identified as one of the sectors with the highest levels of unsatisfied demand in terms of the relative access gap criterion. Moreover, the HCB program, administered by ICBF, was shown to be one of the most progressive social programs on offer in 1997, yet the number of places in childcare centers has shrunk by 4 percent and the coverage rate has fallen by 7 percentage points. Both of these findings suggest that greater emphasis should be put on childcare services in the future. Furthermore, the analysis shows quite clearly that the HCB nurseries are much more progressive in incidence than the CAIP nurseries, suggesting a case for reallocating resources across these programs on distributional grounds. 192 Coverage of public utility services has become more progressively distributed over time; however, use of service subsidies does not appear to be having the desired effect. There has been significant progress in expanding coverage of public utility services, with the number of household connections for electricity, water and sewerage rising by about a third between 1992-97. The electricity service has the highest level of coverage and the most progressive distribution of connections, while the sewerage sector has the lowest coverage and is least progressively distributed. Coverage has become increasingly egalitarian over time, although there are still a substantial number of poor households that lack access to sewerage. However, regarding use of service subsidies, the evidence suggests that the system of 'estratos' is not well targeted towards lower income households. The limited evidence on use of service subsidies suggests that, as a result of this problem, a substantial part of the benefits leaks to the non poor. The distributional impact of social spending The overall incidence of PSS is broadly neutral, but the poor benefit substantially. The total value of social programs is equivalent to 5 percent of household income, of which 80 percent is allocated to the state education system. The overall distributional incidence of PSS is almost neutral, with a concentration coefficient of negative 0.06. Although in absolute terms, subsidies are fairly evenly distributed across quintiles, their relative impact is still far greater on low-income households since they represent a higher proportion of household income -32 percent for the first quintile-. Rural areas receive a disproportionate amount of PSS; therefore, the redistributive impact increases proportionally. The most progressive social programs are HCB nurseries, primary education and the subsidized health insurance scheme, each with concentration coefficients of less than negative 0.3. The most regressive social programs are tertiary education and the housing subsidy, with concentration coefficients in excess of 0.3. In general, the programs that embody the highest subsidy per beneficiary are also the most regressive. However, those that account for the highest proportion of the total subsidy are the most progressive. By sector, the evolution of coverage across income groups is heterogeneous. For education, a sector in which inter-temporal comparisons can be made, the aggregate progressivity of PSS does not appear to have changed materially since 1992. However, by grade, the evolution has been somewhat heterogeneous. Overall education spending has become slightly less progressive since 1992, with the concentration coefficient rising from negative 0.08 to negative 0.03. Within this overall average, primary and tertiary education became somewhat more regressive, while secondary education become somewhat more progressive. Sector priorities and decentralization Independently of the welfare target, the highest priorities are childcare, sewerage and health care and insurance. Some sectors show certain robustness in relation to the welfare objective. Childcare is almost always the highest priority service for expansion and in general, sewerage comes out as the second or third priority for expansion. Finally, health insurance has all the possible rankings between first and fourth. Priority sectors under modest expansion objectives (equalizing access up to the sixth decile, are childcare, sewerage, health insurance and health care. Under a more ambitious reference income group such as the eighth decile, the sector priorities do not change that much; childcare continues to be the first, followed by tertiary education, sewerage and health insurance. The cost of substantial increases in the coverage of tertiary education is prohibitively high. For this reason, the aggressive expansion of college credit should be seriously considered, as well as since private returns are very attractive and the private sector has shown dynamic response to attend market demand. 193 ]During the ninetes, decentralized sectors seemed more responsive to social priorities. In 1992 (Velez, 1995) identified childcare secondary education and health as sector priorities. Growth in coverage rates by type of service over the period 1992-97 showed that coverage of childcare services had actually fallen; however, the largest coverage gains have been secured in health insurance and secondary education, which increased by 27 and 12 percentage points, respectively, and disproportionately benefited the poor. To this extent, these two decentralized sectors showed responsiveness to social needs, while the centralized sector -childcare- did not. However, decentralization -when seen as the certificadon process- does niot appear to have had any systematic effect on the coverage and progressiveness of health and education services. Comparisons of performance between departments with a longer and shorter history of certification indicated that observed differences have more to do with historical performance and the availability of resources than with decentralization itself. 194 References Acosta, 0. L. 2000. Gasto P6blico Social y Arquitectura Financiera: C6mo las Condiciones Fiscales Existentes y Esperadas Afectan la Provisi6n de Servicios Sociales en Colombia, Mimeo, World Bank Group, Washington DC. Alesina et al. 2000. "Decentralization in Colombia." Fedesarrollo Working Chapter Series, No. 15, August 2000. Bes, M., Hernandez, R. and C. Oliva. 1998. Descentralizaci6n en Colombia: Nuevos Desaffos, Mimeo, Inter-American Development Bank, Washington D.C. Borjas, G. and 0. L. Acosta. 2000. "Education Reform in Colombia." Fedesarrollo Working Chapter Series, No. 19, August 2000. Duarte Agudelo, J. H. and C L. Villa Arcila. 1996. Hacia un Nuevo Esquema de Financiaci6n de la Universidad Publica Colombiana, Mimeo, Santafe de Bogot. F16rez, C.E. y R. Mendez. 1993. Estudio de Incidencia del Gasto Pu'blico Social: Hogares Comunitarios de Bienestar: Quien se Beneficia? Misi6n de Apoyo a la Descentralizaci6n y la Focalizaci6n de los Servicios Sociales, Departamento Nacional de Planeaci6n, Santafe de Bogota. Gonzalez, J.I. and F. Perez-Calle. 199X. Salud para los pobres en Colombia: de la planeaci6n centralizada a la competencia estructurada, Coyuntura Social, Santafe de Bogota ICBF. 199X. Evaluacion de la Gestion del Instituto Colombiano de Bienestar Familiar, Malinowitz, S. 1998. Servicios Pu'blicos Domiciliarios, Vivienda y Distribuci6n de Ingreso, Misi6n Social, Departamento Nacional de Planeaci6n, Santafe de Bogota. Misi6n Social. 1998. Analisis de las Encuestas de Calidad de Vida para Evaluar el Imnpacto del Nuevo Sistema de Seguridad Social en Salud en Colombia 1993-97, Departamento Nacional de Planeaci6n, Santafd de Bogota. Molina, C.G., M. Alviar and D. Polania. 1993. Estudio de Incidencia del Gasto Puiblico Social: El Gasto Pdblico en Educaci6n y Distribuci6n de Subsidios en Colombia, Fedesarrollo, Santafd de BogotA. Molina, C.G., M.C. Rueda, M. Alviar and U. Giedion. 1993. Estudio de Incidencia del Gasto Pdblico Social: El Gasto Publico en Salud y Distribuci6n de Subsidios en Colombia, Fedesarrollo, Santafe de Bogoti. Mora, H., U. Ayala, C. Gutierrez and A. Velasco. 1999. Financiamiento de la Educacion por Medio de Subsidios a la Demanda: Evaluacion de la Viabilidad del Sistema de Capitaci6n, Fedesarrollo, Santafe de Bogota. Perez Calle, F. 1996. Cdlculo de la Unidad de Pago por Capitaci6n (UPC) de la Educaci6n Bdsica, Fondo de Cofinanciaci6n para la Inversi6n Social, Misi6n Social, Departamento Nacional de Planeaci6n, Santafe de Bogota. Perotti, R. 2000. "Public Spending on Social Protection in Colombia: Analysis and Proposals." Fedesarrollo Working Chapter Series, No. 18, August 2000. Ravallion and Lanjouw. 1998. Benefit Incidence and the Timing of Program Capture. Washington: The World Bank. Policy Research Working Chapter # 1956 SAnchez, F. and J. Nnfiez. 1998. Descentralizaci6n, pobreza y acceso a los servicios sociales. Quien se beneficio del gasto publico social en los noventa?, Coyuntura Social, Santafe de Bogota. Sarmiento A. and B.L. Caro. 1999. "La educaci6n en cifras" DNP, Bogota. 195 SENA. 2000a. El SENA y la Poblaci6n Beneficiaria de sus Servicios, Direcci6n General del SENA, Santafe de Bogota. SENA. 2000b. Costos de Formaci6n Profesional, Direcci6n General del SENA, Santaf6 de BogotA. SENA. 2000c. Evaluaci6n de Impacto de Cursos Largos, Direcci6n General del SENA, Santafe de BogotA. Velez, C.E. 1995. Gasto Social y Desigualdad: Logros y Extravios, Misi6n Social, Departamento Nacional de Planeaci6n, Santafe de BogotA. Velez, C.E. 1996. Los Subsidios por los Servicios Pu'blicos Domiciliarios de Energia y Acueducto y Alcantarillado: Por Que Difieren en su Impacto Redistributivo y Cudl es el Efecto Potencial de las Reformas de Tarifas, Banco de la Repdblica, Santafe de BogotA. Velez, C.E. 199X. Public Social Spending: Efficiency, Equity and Sectorial Restructuring, Mimeo, Banco de la Reptiblica, Santafe de BogotA. Thomas, M.R., C.E. Velez and V. Foster. 2000. Brazil: Selected Issues in Social Protection, Report No. XXX-BR, Latin America and Caribbean Region, The World Bank Group, Washington DC. Vargas GonzAlez, J.E. and A. Sarmiento G6mez. 1997. Descentralizacion de los Servicios de Educacion y Salud en Colombia, Mimeo, Departamento Nacional de Planeaci6n, Santafe de BogotA. Vergara, C.H. and M. Simpson. 1998. Andlisis del Servicio Educativo 1993-1997, Misi6n Social, Departamento Nacional de Planeaci6n, Santafe de BogotA. 196 CHAPTER VI SUBSIDIZED HEALTH INSURANCE, PROXY MEANS TESTING AND THE DEMAND FOR HEALTHCARE AMONG THE POOR IN COLOMBIA By Giota Panopoulu and Carlos Eduardo V6lez ABSTRACT In the new Colombian system of universal health insurance (following the 1993 reform), a proxy-means targeting mechanism, the SISBEN, assigns individuals to either a contributory or a subsidized regime. This chapter has two objectives:first, to understand household behavior in take-up of and affiliation with the subsidized regime, and second, to analyze the effect of take-up/affiliation on use of health services, in terms of household out-of-pocket expenditures, medical visits, medicines and hospitalization. The findings are based on the 1997 Encuesta Nacional de Calidad de Vida, a nationally representative, multi-topic household survey of living conditions. Results show that take-up of the subsidized regime increases both as a result of demand-side variables, such as deteriorated health status and poverty risk, and as a result of supply-side variables related to municipal policy environment, such as pre-reform availability of health service and available information. One keyfinding is that, after the reform, income seems to have ceased to be a constraint on access to healthcare for the poor; yet regardless, the subsidized regime does not stimulate any increase in use of healthcare services. Controlling for the potential endogeneity between take-up and healthcare seeking behaviors, we show that -for urban areas -beneficiaries of the subsidized regime consult doctors more often, but are less likely to face hospitalization, than non- affiliated individuals; rural households' behavior, on the other hand, is unaffected by affiliation. These findings should alleviate concerns that insurance may raise poor people's demand for healthcare. On the other hand, in the event of illness, households in the subsidized regime do face smallerfinancial burdens - that is, lower out-of-pocket expenditures -for medical consultations andfor medicine. However, the overall impact of the system is perhaps being limited by inefficiencies in the cumbersome targeting mechanism which lead both to relatively low take-up rates and to some inclusion of less poor households. 197 1. INTRODUCTION Historically, health insurance has been a privilege of the few in developing countries. Those covered by specific health insurance plans either have been those working in companies where part of their contributions are paid by their employers, or have been individuals who can afford and opt for private insurance. The rest of the population has resorted to the services of the national health system and received free or low-cost care in medical institutions financed by the government. However, many developing countries are currently considering the possibility of introducing compulsory health insurance schemes. One reason is to attract more resources to the healthcare sector. The introduction of compulsory insurance contributions for those who can afford it would allow the limited tax funds to be concentrated on providing healthcare services for fewer people and would thus improve coverage and raise standards. A second reason is to improve existing services where the quality of care is poor and resources are not used to the best advantage. In December 1993, Colombia introduced radical health sector reform through the passage of Ley 100 (Law 100). The new system provides universal health insurance involving cost-sharing between the employee and the employer on the financing side, and a pluralistic system of provision involving both private and public providers on the supply side, for both insurance and healthcare. One of the main tenets of Ley 100 is solidarity: every citizen, regardless of financial means, should have access to basic health services through an income-related contribution, with the low-income population being subsidized by a combination of government subsidies and contributions (in the form of premiums paid) by those with relatively higher incomes. As a result of the reform, health insurance coverage increased from 31 to 58 percent of the population between 1992 and 1997. These increases in coverage have been concentrated in the lower income deciles where, for example, insurance coverage in the first decile has risen tenfold from 4 percent in 1992 to 41 percent in 1997 (VYlez and Foster, 2000). However, the sustainability of health insurance financing has become a key issue. Critics of the reform fear that the introduction of the subsidized regime (rggimen subsidiado) for the low-income population will lead this group to greatly increase its demand for healthcare services. The increased expenditure in turn will cause financing problems which could eventually jeopardize the sustainability of the new system. The purpose of this chapter is three-fold. First, it analyzes the patterns and determinants of affiliation with the subsidized healthcare sector in Colombia; second, it estimates the effect of such affiliation on the utilization of healthcare services; and third, it discusses the policy implications of these findings and offers recommendations generally for the public health sector. The chapter paper is divided into six sections, beginning with this Introduction. Section II offers an overview of the health sector reform, as related to the subsidized regime. Section I describes affiliation and healthcare utilization patterns in the subsidized scheme. The following two sections move beyond the descriptive level to analyze the factors that influence affiliation with the subsidized sector, and the effects of such affiliation on both healthcare utilization and out-of-pocket expenditure. Finally, Section VI summarizes the chapter's findings and suggests policy implications. 2. THE HEALTH SECTOR REFORM Between 1992 and 1997, health insurance coverage almost doubled and became much more progressive, with increasing participation of the private sector. As mentioned above, the Colombian healthcare system 198 underwent major reforms in 1993. A dramatic consequence of this was that coverage of health insurance increased from 31 to 58 percent.41 Furthermore, these increases in coverage were concentrated in the lower income deciles. For example, insurance coverage in the first decile rose tenfold from 4 percent in 1992 to 41 percent in 1997, whereas in the tenth decile coverage rose more modestly, from 65 to 80 percent. In addition, twice as many individuals became insured by private providers as by the public provider, ISS.42 2.1 Main Characteristics: The contributory and the subsidized regime The Colombian healthcare system operates on two levels, contributory and subsidized. Those who have the ability to pay are enlisted in the contributory regime (re'gimen contributivo), while the poor and indigent are covered by the subsidized regime. Affiliation to the contributory scheme is contingent upon the payment of a monthly contribution equal to 12 percent of wage income, of which the employee pays 4 percent and the employer 8 percent. Self-employed individuals pay the total 12 percent of the contribution. In the subsidized sector, contributions to the health system are subsidized from fiscal and solidarity sources. The latter take the form of transfers from the contributory to the subsidized sector.43 Affiliates of the subsidized system have access to a basic benefit package known as Plan Obligatorio de Salud Subsidiado (Mandatory Subsidized Health Plan - POSS), which covers health promotion and education, primary healthcare, basic hospital services and treatment for a number of high-cost diseases. The POSS offers full coverage for maternity and child-care, including some secondary and tertiary care in hospitals. The POSS covers fewer services than the basic benefit package of the contributory regime known as Plan Obligatorio de Salud (Mandatory Heath Plan - POS); however, one of the goals of Ley 100 is the convergence of the two packages by 2001.44 The POSS, like the POS, offers family coverage. A family group includes spouses, or stable partners with a minimum two-year old relationship and any economically dependent children of either spouse. The latter include children under 18, full time students under 25 years of age, and disabled dependents of any age. 4' See Velez and Foster, 2001. 42 Ibid. The increase in private insurance rates is partly due to the fact that, as a consequence of the reform, all insurers, public and private, are obliged to offer insurance packages priced in accordance with the income of the contributor; as a result, insurance costs due not vary by provider. The fact that twice as many people are affiliated with private insurers may be indicative of the former being more efficient than the latter, or of private services simply being perceived to be of higher quality and thus being utilized more often. 43 One percent of the 12 percent wage contribution of the contributory regime is used to contribute to the subsidized regime. The Empresas Promotoras de Salud (Health Promoting Entities - EPSs), which are responsible for the collection of the monthly contributions, transfer these sources to a central fund called Fondo de Solidaridad y de Garantfa (Solidarity Fund - FOSYGA) which in turn is responsible for their distribution within the subsidised regime. 44 The POS aims to be a comprehensive healthcare package. Coverage excludes a limited number of activities, interventions and procedures. Such activities are defined in Acuerdo (Accord) 1938 of 1994 and include: a) cosmetic surgery, b) nutritional and varicose vein treatment for cosmetic reasons, c) infertility treatment, d) treatment against fatigue and insomnia, e) specific organ transplants, f) psychotherapy and psychoanalysis at an individual level or for a prolonged period, g) orthodontic and prosthetic treatment for dental care, h) elastic means of support, corsets, girdles, wheel chairs, insoles and orthopaedic shoes, i) treatments that are not recognised by the international medical community or treatments of experimental nature, j) treatments that use experimental medicines/substances or medicines/substances that are not included in the Manual de Medicamentos, Esenciales y Terapeutica, k) curative treatments for chronic and degeneraiive diseases, I) cancer and traumas that are in a terminal stage, m) educational and vocational activities that take place during the rehabilitation phase and are not strictly necessary in medical terms and n) activities that are not included in the the Manual de Actividades, Intervenciones y Procedimientos. 199 Affiliation to the subsidized sector is based on the proxy-means test, Sistema de Selecci6n de Beneficiarios para Programas Sociales (Beneficiaries Selection System for Social Programs), hereafter SISBEN, administered by local governments. SISBEN is a proxy-means test index, designed to provide local governments with a tool for targeting social subsidies, including health subsidies, to the poorest and most vulnerable segments of the population (Vdlez et al. 1998). The index is based on a household questionnaire - the Ficha de Caracterizaci6n Socioecon6mica (Socio-economic Classification Form - FCS). Using customized software, the SISBEN score is calculated for each household, determining its classification into one of six groups, with SISBEN level I being the poorest group. Households receive certification of their SISBEN score and/or level.45 The application, implementation and administration of SISBlN46 rests on municipalities, with the help of the departmental health authorities. The Consejo Nacional de Seguridad Social de Salud (National Council of Social Security in Health - CNSSS) has defined households that belong to SISBEN levels 1 and 2 as those whose members are eligible for the subsidized regime. In a second stage, members of households at SISBEN levels 1 and 2 have the right to subscribe to one of the insurance entities in the subsidized regime known as Administradoras de Regimen Subsidiado (Administrators of the Subsidized Regime - ARS). In turn, the ARS signs a contract with the relevant departmental health authority for each of its affiliates, after which, the individual is considered an official member of the subsidized regime. However, if either parent in the famnily is employed under contract, receives a pension or is self-employed and has a monthly salary higher than twice the monthly statutory minimum wage, the household must join in the contributory regime regardless of its SISBEN level. In theory, after complete coverage of the population at SISBEN levels I and 2, further beneficiaries from the SISBEN level 3 population may enter the subsidized regime, starting with the households of lower to higher SISBEN scores.47 Beneficiaries of the subsidized scheme may use public healthcare services covered by POSS, provided they issue a copago, or co-payment.48 The level of co-payment is contingent upon the individual's SISBEN level. For SISBEN level 1 individuals, the maximum co-payment is 5 percent of the cost of the service and the total cost of the service cannot exceed a quarter of the statutory monthly minimum salary. SISBEN level 2 individuals have a maximum co-payment of 10 percent of the cost of the service and the total cost of the service cannot be more than half of the statutory monthly minimum salary (see Figure 6. 1).49 Overall, health insurance coverage in Colombia disproportionately benefits the better-off, with the contributory regime being regressive (to the extent that formal employment is regressive), and the subsidized regime relatively progressive-especially in comparison to other social services.50 Figure2 depicts the progressive nature of the subsidized regime, yet undoubtedly shows a considerable need for improvement in coverage, illustrated by the nonetheless regressivity of overall health insurance coverage. Also of note is the disparity between the progressiveness of public health insurance coverage between 45 Certification can be in the form of a card; an official document, or a carnet (identity card) that ARSs provide their affiliates after their subscription (see next paragraph for details). A household usually receives a certification when it belongs to the lower SISBEN levels 1, 2 or 3. 46 In the case of indigenous communities, the application of SISBEN is not necessary and all community members can become affiliated to the subsidized sector. However, we are not able to analyze the behavior of this group since the ECV/97 survey does not include indigenous people. 47 SISBEN 4 and above are not eligible for the subsidized regime. 48 For most services, in accordance with Acuerdo (Accord) 30 of 1996. 49 In the case of affiliated households of SISBEN level 3, the copago is 30 percent of the value of the healthcare service but there is a loop hole in the legislation when the household has an SCD but is not affiliated. 50 See Chapter 5 (Velez and Foster, 2001) for discussion of the progressivity of education, childcare, and utilities, interalia. 200 rural and urban areas: the concentration coefficients are 0.02 and 0.41, in urban areas and rural areas, respectively. Figure 6.1. SISBEN classification document -SCD-: affiliation and healthcare co-payments in the subsidized regime Affiliated Copagos (1) SISBEN I - 5% With an SCD SISBEN 2- 10% SISBEN 3 - 30% Non-affiliated Cuotas de recuperaci6n Eligible (2) SISBEN 1 -5% Households SISBEN 2 - 10% Without an SCD _ Non-affiliated Full price (3) Three quarters of the newly insured entered the contributory regime - the rest are mostly poor and entered the subsidized regime. Of the 30 percent of the population that became newly insured between 1992 and 1997, three quarters entered the contributory regime and the remaining quarter entered the subsidized regime. The concentration coefficient for those entering the contributory regime was 0.12, and -0.38 for the subsidized regime.5' Figure 6.2. Coverage rates for health insurance and treatment by decile, 1997 100% 90%/ 80% i' 70% -+- Health insurance a 60% - (subsidized) E! Health insurance 50 - (contributory) o 40% - ,s Health insurance 30% - (overall) 20% - 10% - - 0% 1 2 3 4 5 6 7 8 9 10 Deciles 5' Ibid. 201 2.2 SISBEN classification in level 1 or 2 seemingly equates aiMliation to the subsidized regime Most potential beneficiaries take SISBEN classification into level 1 or 2 as an automatic entitlement to the subsidized regime. By law, classification in SISBEN level 1 or 2 ought to be a prerequisite for affiliation to the subsidized regime but does not necessarily imply affiliation per se. If a household completed the FCS and received a SISBEN classification document (SCD) but has not subscribed to an ARS, its members are not affiliated to the subsidized regime. Nevertheless, according to information provided in a series of meetings between the researchers and employees of Misi6n Social (Social Division) of the Departamento Nacional de Planeaci6n (National Planning Department - DNP), the Ministry of Health and other health related institutions, great confusion seems to surround this matter. This has led the majority of the Colombian population to believe that being classified as SISBEN level 1 or 2 is synonymous to affiliation to the subsidized regime. Health providers also often assess co-payment Levels only based on SISBEN level. According to the law, the main difference between affiliated and non-affiliated individuals with respect to healthcare charges is that the forner issue a copago, while the latter pay the full cost of the service. Nevertheless, hospitals have used the SCD as a proof of socio-economic status on which to base copagos not only for affiliated individuals but also the non-affiliated who have a classification document (Group 2 in Figure 6.1). The co-payments charged to the non-affiliated who have an SCD are called cuotas de recuperaci6n and are calculated on the same basis as the copagos charged to the affiliated. Therefore, for SISBEN level 1 and 2 households, the important factor for being charged lower healthcare prices is not affiliation, but possession of an SCD, as described in Figure 6.3. Figure 6.3. SCD, affilDiation and healthcare co-payments in the subsidized regime, moodified With an SCD "Affiliated" Copagos/ Cuotas de recuperaci6n (l) + (2) SISBEN 1 - 5% SISBEN 2 - 10% Eligible Households Without an SCD Non-affiliated Full price (3) 3. AFFiLIATION AND HEALTHCARE uILIZATION PATrERNS This section describes affiliation and healthcare utilization patterns in the subsidized regime using data from the Encuesta de Calidad de Vida, 1997.52 The survey provides nationally representative data about household living standards. It includes 9,120 households accounting for 38,516 individuals'3. The surveys aims at shedding light on the Colombian population's living standards, with emphasis on the effects of social policies regarding health, education and labor issues on living conditions. The ECV/97 contains a health module comprising 38 questions on possession of an SCD and individual's affiliation, health status, use of healthcare services and healthcare expenditure. It also provides detailed information 52 A nationally representative household survey conducted during the second quarter of 1997 by the Departamento Nacional de Estadistica (National Department of Statistics - DANE) with the support of Misi6n Social of DNP and the mninistnes of Agriculture, Health and Education. 53 The population of Colombia was approximately 39,840,000 people in 1997. 202 on durable goods, public services and a large range of income and demographic variables that allow the computation of the SISBEN index and the classification of households into SISBEN levels.54 Appendix A describes the SISBEN index and some issues related to its computation. We estimate 8.5 percent of the population to belong to SISBEN level 1 and 22 and 32 percent to levels 2 and 3, respectively. Table 6.1 presents the number of households and individuals by SISBEN level, as a result of our computation. The corresponding percentages at the household level are slightly higher, indicating that households of SISBEN levels 1, 2 and 3 are, on average, comprised of more members than are households of SISBEN levels 4, 5 and 6.5 As already mentioned, the unit of analysis for possession of the SCD is the household and, for reasons that will become clearer in Section V, we restrict analysis of healthcare utilization to the household head. Of the 5,068 households/household heads of SISBEN levels 1, 2 and 3, 560 (11 percent) belong to groups56 that forn separate social security networks from those mandated by Ley 100. In addition, 1,389 household heads (27.4 percent) appear to be eligible or affiliated to the contributory regime.57 Our analysis sample only includes household heads that are eligible for the subsidized regimne. This results in a sample size of 1,671 observations of SISBEN levels I and 2 and 1,448 observations of SISBEN level 3. Table 6.1. Number of households and individuals by SISBEN level Households Individuals Number % Number % SISBEN level 1 579 6.6 3,179 8.5 SISBEN level 2 1,729 19.8 8,272 22.2 SISBEN level 3 2,760 31.5 11,852 31.8 SISBEN level 4 1,992 22.8 7,941 21.3 SISBEN level 5 1,513 17.3 5,487 14.7 SISBEN level 6 180 2.1 533 1.4 Total 8,753 100.0 37,264 100.0 Note: The sample does not include 367 rural households (1,252 individuals) for which it was not possible to compute the SISBEN index. Source: ECV/97, own calculations. 3.1 Affiliation patterns for households at SISBEN levels 1, 2, and 3 The data confirm the equivalence of the possession of an SCD and affiliation. Although two different questions probe into affiliation status and possession of an SCD, the data show that all individuals with an SCD appear to be affiliated; which corresponds to Figure3. As can be seen, eligible households/heads can now be classified into two groups, according to only their SCD holding. We cannot distinguish affiliated and non-affiliated among SCD holders58 and will henceforth use SCD holding for affiliation. 54 The ECV/97 survey does not ask about the SISBEN level of the individual or the householdx 55 The tables in section III are calculated using unweighted data. Calculation using the ECV/97 weights gave very similar results to the ones reported here. The SISBEN index could not be computed due to missing information for 367 rural households (4 percent of the total sample) accounting for 1,252 individuals (3 percent); these households are therefore not included in the table . 56 Such groups include the armed forces, the police, teachers, employees of Ecopetrol - the major Colombian oil company - and affiliates of various charity programs e.g. Canitas, Plan Revivir etc 57 These are individuals who are employed under contract, receive a pension or are self-employed and have a monthly salary higher than twice the monthly statutory minimum salary. 58 Assuming that because these two groups face similar co-payments, they can be expected to behave similarly as far as healthcare utilization would be misleading. Individuals with an SCD that are affiliated should be better informed than those individuals with an SCD that are not affiliated. In addition, the former group can take advantage of the 203 In 1997, 41 percent of SISBEN level I and 2 households received an SCD, while the corresponding figure for households at SISBEN level 3 was much lower, at only 27 percent. Table 6.2 also shows that, only 35 percent of all households in the three lowest SISBEN levels have an SCD; this is due to the small number of SCD holders among households at SISBEN 3. Rural households are more likely to hold an SCD than urban ones, a result possibly due to the more efficient implementation of the SISBEN program in rural areas. Table 6.2. Number of households by S1LSBIEN level and SCDl) slatus in rural and uirban areas SISBEN levels 1 and 2 SISBEN level 3 SCD status . SCD status No Yes Total No Yes Total Urban 333 160 493 574 128 702 (67.5) (32.5) (81.7) (18.3) Rural 645 533 1,178 478 268 746 (54.7) (45.3) (64.1) (35.9) Total 978 693 1,671 1,052 396 1,448 (58.5) (41.5) (72.6) (27.4) Note: Figures in parentheses are percentages that add up to 100 percent horizontally, separately for urban and rural areas. Source: ECV/97, own calculations. Even though households have the right to demand an FCS questionnaire for completion --thus influencing the probability of obtaining an SCD--coverage is still relatively low among eligible SISBEN level 1 and 2 households. One reason for low coverage could be lack of information and/or interest. A second reason could be related to the stigma associated with the identification of being poor. Through the econometric analysis in Section IV, we will aim at establishing the demand-side factors that affect the probability of a household having an SCD and in such, try to understand why SCD take-up is low. Other contributing supply-side factors for the low levels of SCD possession include lack of municipal resources for carrying out the SISBEN program, lack of technical infrastructure, and insufficient workforce for the data collection. Such problems were reported by the majority of municipalities in an evaluation of the SISBEN implementation, carried out by the Misi6n Social of DNP in December 1996 (Misi6n Social, 1996). Some SISBEN level 3 households hold SISBEN documentation even when not all households of level 1 and 2 have one. That some SISBEN 3 households hold an SCD is not surprising in itself, since the legislation mandates that after full coverage of the SISBEN level 1 and 2 population, municipalities with resources can start covering the SISBEN level 3 population. Analysis at the municipal level nevertheless reveals, surprisingly, that for the majority of municipalities where SISBEN level 3 households hold an SCD, not all SISBEN level 1 and 2 households hold one. Household migration between municipalities could partly explain this finding. A second reason could be some inaccuracy in our calculation of the SISBEN index, which could wrongly classify as SISBEN level 3 a household that is in fact at SISBEN level 2.59 A third potential explanation is a certain degree of corruption among the municipal authorities. Separate econometric models for the probability of a household holding an SCD will be estimated for each group. This will take into account the different processes through which households at SISBEN levels 1 and 2 and those at level 3 acquire their SCD, in Section IV. We wish to test whether determinants differ for these two groups. network of providers in their ARS of affiliation and be accounted for by the health authorities that allocate resources to the subsidized regime. 59See also Appendix A. 204 3.2 Healthcare Utilization Patterns for Household Heads at SISBEN levels 1 and 2 We restrict the analysis of utilization of healthcare services to expenses related to hospitalization, medical visits and medications. Although the ECV/97 gathered information about a wide range of services, including utilization of and expenditure incurred on lab exams and x-rays, very few positive observations were recorded for the other healthcare services included in the survey. We also examine only the healthcare utilization patterns of household heads at SISBEN levels 1 and 2. This choice is justified on various grounds. First, as already mentioned, there is a gap in Colombian legislation with respect to the co-payments for individuals at SISBEN level 3 when they hold an SCD. This invalidates our hypothesis for the SISBEN levels 1 and 2 households that, irrespective of affiliation status, if an individual has an SCD, he or she will face the same co-payment. Second, as we will see in Section IV, the factors influencing the probability that a household at SISBEN level 3 possesses an SCD indeed differ from those for households at SISBEN levels I or 2. We therefore decide to examine the effect of an SCD on the utilization and expenditure of healthcare services only for the strictly eligible households at SISBEN levels I and 2. Before looking at the utilization and expenditure of healthcare services, we examine illness and treatment rates for the household heads at SISBEN levels 1 and 2, in Table 6.3.6° When compared to households heads of other regimes, households heads with an SCD (subsidized regime) have a higher illness rate (17.9 percent) than households heads of the contributory regime (16.3 percent) and a lower illness rate than household heads of one of the special regimes6' (See Table 6.4). Table 6.3. Illness incidence by SCD status. Number of household heads and illness rates Urban areas Rural areas SCD status SCD status No Yes No Yes Health problem 62 32 98 92 (18.6) (20.0) (15.2) (17.3) Notes. Figures in parentheses are calculated as the ratio of individuals that had a health problem in the last 30 days to the total number of individuals in the cell. In urban areas, 160 individuals have an SCD and 333 do not. In rural areas, 533 individuals have an SCD and 645 do not Source: ECV/97, own calculations Table 6.4. Illness rates of household heads for the subsidized, contributory and special regimes Subsidized regime Contributory Special (with SCD status regime (with insurance) Type of service No Yes insurance) Health problem 160 124 561 174 (16.3) (17.9) (16.3) (19.0) Note. Figures are calculated as the ratio of Individuals that had a health problem in the last 30 days to the total number of individuals in the cell Source ECV/97, own calculations. Table 6.5 shows that for the SISBEN level I and 2 household heads, treatment rates are higher for individuals without an SCD than for those with an SCD. This result is rather expected given the wide definition of treatment in the data of the ECV/97.62 Household heads affiliated to the contributory regime 60 An individual is considered ill if he or she has reported having a health problem (disease, accident, dental problem or other health problem) in the last past 30 days. 61 Armed Forces, police, teachers, Ecopetrol or affiliates to various charity programs. 62 "Treatment" includes: treatment by a professional or at a healthcare institution (hospital, clinic, health center) a health promoter, a nurse, a pharmacist or a pharmacist assistant, an empirical doctor, a witch doctor, a folk healer, a herbalist, a midwife, indigenous traditional medicine, alternative therapies (homeopathy, acupuncture, music therapy, floral essences), folk remedies and self-prescription. It should not be confused with "utilization," which, in this case, refers strictly to (professional) hospitalization, medical visits, and the use of medications. 1 205 and to one of the special regimes have higher treatment rates than household heads with an SCD (subsidized regime) (See Table 6.6). Table 6.5. T1reatment incidence by SCD status. Number of household heads and treatment rates Urban areas Rural areas SCD status SCD status Treatment No Yes No Yes Yes 58 27 89 80 (93.5) (84.4) (90.8) (87.0) No 4 5 9 12 (6.5) (15.6) (9.2) (13.0) Total 62 32 98 92 Note: Figures in parentheses are percentages that add up to 100 percent vertically. Source: ECV/97, own calculations. Table 6.6. Treatment rates of household heads for the subsidized, contributory and special regimes Subsidized regime Contributory Special (with SCD status regime (with insurance) Treatment No Yes insurance) Yes 147 107 533 154 (91.9) (86.3) (95.0) (88.5) No 13 17 28 20 (8.1) (13.7) (5.0) (11.5) Total 160 124 561 174 Note: Figures in parentheses are percentages that add up to I 00 percent vertically. Source: ECV/97, own calculations. The data from the ECV/97 show that 6 percent of household heads at SISBEN levels 1 and 2 were hospitalized, 15 percent had general or specialized medical consultations and 29 percent consumed medicines of some kind. Of the individuals hospitalized, 44 percent resided in urban areas, while of those receiving medical consultations and medicines, 38 percent resided in urban areas. Table 6.7 shows that among household heads in the subsidized regime, those with an SCD display higher utilization rates for all types of services than those without an SCD. A possible explanation for this result is that individuals with an SCD face a lower price for healthcare services through co-payments which, in turn, increases their probability of using the service. This issue is further explored in Section V in the regression analysis, where the other possible explanation, adverse selection, is also examined.. Utilization rates of household heads in the contributory and other special regimes remain higher than those of household heads with an SCD (See Table 6.8). This could result from the higher level of benefits or number of services included in the POS or the special regimes, compared to the POSS. Additionally, people in the subsidized regime, due to being poor, could face higher transportation or time costs when seeking medical services and may also be generally less likely to seek healthcare in the first place. 206 Table 6.7. Users of health services by type and SCD status. Number of household heads and utilization rates Urban areas Rural areas SCD status SCD status Type of service No Yes No Yes Hospitalization 26 16 25 28 (7.8) (10.0) (3.9) (5.3) Medical 52 42 64 90 consultations (15.6) (26.3) (9.9) (16.9) Medicines 112 72 144 157 (33.6) (45.0) (22.3) (29.5) Notes, Figures in parentheses are calculated as the ratio of individuals that used the respective service to the total number of individuals in the cell. In urban areas, 160 individuals have an SCD and 333 do not In rural areas, 533 individuals have an SCD and 645 do not Source: ECV/97, own calculations Table 6.8. Utilization rates of household heads for the subsidized, contributory and special regimes Subsidized regime Contributory SCD status regime Special Type of service No Yes (with insurance) (with insurance) Hospitalization 5.2 9.1 9.2 9.4 Medical consultations 11.8 20.0 30.1 23.4 Medicines 26.2 31.7 41.7 37.9 Note. Figures are calculated as the ratio of individuals that used the respective service to the total number of individuals in the cell Source: ECV/97, own calculations. When spending on any health services in the subsidized regime, SCD holders face lower out-of-pocket expenditures than non-holders, both in urban and rural areas (See Table 9).63 This result is expected, since individuals who are SCD holders enjoy lower co-payments than individuals who are not. Similar to utilization rates, out-of-pocket expenditure for households heads in the other regimes are higher than those of SCD holders (See Table 10). Table 6.9. Out-of-pocket per capita expenditure for household heads' by SCD status and type of service (in 1997 Colombian pesos) Type of service Urban areas Rural areas SCD status SCD status No Yes No Yes Hospitalization 153,667 140,407 235,669 145,077 Medical consultations 12,621 1,323 8,450 3,810 Medicines 20,454 8,045 21,625 11,710 Notes: The figures reported are per capita expenditure only for household heads that have reported positive expenditure and not for the total number of household heads. Source: ECV/97, own calculations. 63 Hospitalization expenditure data do not include transportation costs. 207 TabRe 6.10. Out-of-pocket per capita expenditure for household heads for the subsidized, contributory and special regimes (in 1997 Colombian pesos) Subsidized regime Contributory Special (with SCD status regime (with insurance) Type of service No Yes insurance) Hospitalization 194,668 143,368 758,158 698,842 Medical consultations 10,304 3,073 16,309 8,770 Medicines 21,101 10,587 21,694 24,445 Note: The figures reported are per capita expenditure only for household heads that have reported positive expenditure and not for the total number of household heads. Source ECV/97, own calculations. Summary The principal findings of the descriptive analysis are the following: o The ECV/97 data reveal confusion between affiliation and possession of an SCD among the survey respondents. o 41 percent of households at SISBEN levels 1 and 2 hold an SCD, while the corresponding percentage for households at SISBEN level 3 is 27 percent. O Illness rates are higher for household heads with an SCD than those without an SCD in both urban and rural areas. Treatment rates, on the other hand, are higher for individuals without an SCD. 64 o Household heads at SISBEN levels 1 and 2 with an SCD display higher utilization rates for all types of services than household heads without an SCD. o Per capita out-of-pocket expenditure for hospitalization, medical consultations and medicines are lower for household heads at SISBEN levels 1 and 2 when they hold an SCD compared to those at the same SISBEN level without an SCD. 4. DETERMINANTS OF A SIISBEN CLASS1FICATION DOCUMENT TAKE-UP The aim of this section is to identify the determinants of take-up of an SCD in the subsidized regime. We follow the theoretical framework of the literature on take-up of means-tested benefits. This literature examines why some low-income individuals eligible for a welfare benefits do not in fact join welfare rolls. Separate models are estimated for SISBEN level I and 2 and SISBEN level 3 populations, with the assumption that the two groups follow different patterns. 4.1 Theoretical Considerations The literature on take-up of means-tested benefits was developed, mainly in the US and the UK, for a number of different types of benefit programs65, which include income supplementation for disability, housing, children-related expenses, etc. The majority, if not all, of these programs are associated with a 64 Treatment means receiving health service given that is ill. 65 References include Moffit (1983) for Aid to Farnilies with Dependent Children; Ashenfelter (1983) for the Seattle and Denver Income Maintenance Experiment; Halpem and Hausman (1986) for Disability Insurance; Duclos (1992) and Fry and Stark (1987) for the British Supplementary Benefit; Dorsett and Heady (1991) and Blundell et al. (1987) for the British Housing Benefit; Fry and Stark (1993) for the British Family Income Supplement 208 specific level of entitlement, mainly in the form of income, which the family receives if its claim is successful. Non-take-up of benefits is often attributed to lack of knowledge and misconceptions about the entitlement or to the opportunity costs of claiming it. In cases where the entitlement is believed to be small, the level of ignorance may be such that it is thought not worthwhile investigating the benefit. However, even in cases where the individuals are well aware of their entitlement, they may decide not to take it up because the costs of claiming in terms of time, hassle and social stigma outweigh the advantages of receiving the extra money (Cowell, 1986; Moffit, 1983). The case of SISBEN in Colombia is slightly different from the types of programs treated in the literature; thus we amend the basic framework to incorporate factors specific to this program. The related entitlement is a function of the expected distribution of ill health. Those who expect to need healthcare will enjoy higher entitlement levels than those who consider themselves relatively healthy. Therefore, those who expect to need healthcare may be more likely to obtain an SCD, as a result of adverse self- selection. Also, while the obtainment of an SCD depends on the initiative of the eligible individual or family, factors exogenous to the household such as the resources available to the municipality also play a key role. We model the decision between take-up of an SCD and non-take-up on the basis of the expected utilities in the two states. Any factor that increases the relative utility of possessing an SCD will increase the probability of take-up. This type of modeling will help us overcome problems arising from lack of data about ignorance/misconception and stigma/costs associated with claiming benefits. In particular, we can only address the issue of ignorance indirectly, by assuming it is correlated with individual and municipal characteristics recorded in our data. However, the analysis undertaken might allow us to comment on the second factor (stigma/costs) since the real costs and the stigma of claiming are more likely to outweigh the benefits if the level of entitlement is small. Thus, evidence showing a positive relationship between take-up and level of entitlement will provide support for the view that there may be significant costs associated with claiming (Blundell et al., 1988). In the case -of SISBEN, such costs may include burdensome paperwork. The relationship between take-up and income varies. In most empirical studies, income tends to have a negative and highly significant effect on the probability of take-up66 A plausible explanation is that individuals with higher incomes may feel a greater stigma of applying for a means-tested benefit. Thus the probability of take-up is likely to decrease, ceteris paribus, not only because the returns to claiming may be lower at higher incomes, but also because the subjective cost is higher. Cowell (1986), however, takes a slightly different approach, maintaining that stigma and the "hassle" factor may play a different role. If better-off people are more informed about benefits and better equipped to negotiate on their own behalf, they will find claiming less uncertain and less troublesome. Take-up might therefore increase in higher income groups. Economic position or wealth, independently of income, may also affect take-up, again because of stigma or subjective costs. In an attempt to estimate this effect, we include the score of the first three factors of the SISBEN index (housing, public services and human capital and social security) and the score of the fourth factor (demographic, income and occupational factor) minus the income points.67 Other individual characteristics such as gender, age, and attitude to health will affect the decision. Gender is expected to have a positive effect, given that women appear more concerned about health 66 See Craig (1991) for a review of such studies and Fry and Stark (1993). 67 For more details on the variables used for the construction of these factors, see Appendix A. 209 issues. In the same fashion, individuals who undertake more preventive care may thus reveal more attention to health issues. The effect of age on the probability of take-up cannot be easily defined a priori. On the one hand, older individuals may be more in need of health insurance as they face higher morbidity, which will increase their probability of take-up. On the other hand, old age is associated with infirmity that makes the complexities of the claiming procedures a greater hurdle, thus lowering the probability of take-up. We expect the first effect to dominate in the case of SISBEN. The duration for which the households expect to remain eligible for the benefit will positively affect take- up. Hence, we might expect those with fluctuating circumstances to be less likely to claim an SCD than those with a more stable socioeconomic environment. For example, an unemployed individual may be entitled to benefit, but may be expecting to get a job and therefore not qualify for the subsidized regime any longer. In contrast, individuals with a long tenure in poorly paid jobs may find it worth going through the process of claiming. The number of years in the same residence has a similar effect, as the more years an individual or family resides in an area, the more stable their environment. Single people may be more likely to claim than couples, if we assume that second adult in the family increases the household's ability to cope without claiming benefits.68 This should hold even if, for a given level of household income, additional family members add to needs and therefore raise the probability of take-up. Apart from household and household member characteristics, municipal factors may also have an impact on the probability of take-up given the importance of municipal authorities in the process of application, implementation and administration of the SISBEN program. We attempt to control for these factors using a number of municipal variables. Households in conservative municipalities might be less likely to take-up SISBEN. We would expect that conservative mayors would be less oriented towards social expenditure programs such as SISBEN than liberal ones, thus reducing the probability of take-up for the households in their municipality.69 We assume a positive relationship between the probability of take-up and the mayor's popularity, since given the preference of a mayor to protect the general interest of the community, it is expected that the higher his/her popularity, the more committed s/he will be to satisfying these interests. We use as a proxy for this factor the proportion of votes won by the current mayor of the households' municipality of residence. We also expect a positive relationship between the support of the council and the probability of take-up, as the more support a mayor enjoys among the members of the municipal council, the greater his or her ability to carry out his or her tasks and thus satisfy social needs. We proxy this factor by the number of council members belonging to the same political party as the mayor of the municipality in which the household resides. The higher the political participation in the community, the better informed households will be about their rights and hence, the more likely that eligible households will be holders of an SCD. We use as a proxy for political participation the ratio of the number of valid votes in the households' municipality of residence to the number of voters registered in the municipal census. All four electoral variables are for the year 1994. 68 See Fry and Stark (1987), Fry and Stark (1993), Dorset and Heady (1991), and Blundell et al. (1988). 69 An alternative hypothesis could be that conservative mayors might favor the SISBEN program as an alternative to subsidies. Nevertheless, given that hospitals of second and third level are under departmental authority, this hypothesis seems less strong. 210 We attempt to capture the effect of supply-side financing on the probability of take-up70 by including the average level of municipal health transfers7' for the period 1994-1997. We expect a positive relationship between the two variables, given that a wealthier municipality should be more able to supply all eligible households with an SCD. We also use the municipality 1993 Gini coefficient to control for potential distributional effects.72 The departmental take-up rate is expected to have a positive effect on the household take-up rate. We compute it as the ratio of individuals in a household with an SCD to the total population of the department using the data from the ECV/97. This variable should capture an information effect since the more individuals with an SCD in a department, the higher the dissemination of information among all eligible households. Finally, we include two proxies for-pre-existing departmental conditions - healthcare and socio-economic level--. We use the data from the CASEN/93 household survey on departmental hospitalization and health insurance rates for an equivalent SISBEN I and 2 population. Although the SISBEN program did not exist in 1993, we computed a SISBEN index along the procedures followed for the ECV/97, to obtain a CASEN/93 sample as similar as possible to our analysis sample. The 1993 departmental hospitalization rate serves as a proxy for levels of preexisting healthcare utilization, while the health insurance rate is used as a proxy for information and/or socio-economic status of the department. The lower the preexisting level of health supply, the higher the potential improvement of adopting a new subsidized healthcare system. We therefore estimated a model of take-up of an SCD with three groups of covariates: * household characteristics: the four factors of the SISBEN index, household income, the presence of household members reporting either bad health status or a chronic problem, number of years in the same residence and home ownership * household head characteristics: age, gender, marital status, employment status and attitude towards health prevention, and * municipal and departmental variables: At the municipal level, we use the mayor's political affiliation, his or her popularity, the support s/he receives from the council, the political participation of the population, health transfers and a Gini coefficient. At the departmental level, we use the SCD take-up rate for 1997 and the hospitalization and health insurance rates for 1993. Definitions of the variables included in the model of take-up are listed in Table A6.7 in Appendix B. 4.2 Methodology We model the probability that an eligible household holds an SCD with a probit. We denote this event by a dichotomous variable that takes the value of I when the household has an SCD and 0 otherwise. The probit model describes the probability as: 70 We would like to thank Fabio Sanchez, Profesor Investigador at the Universidad de los Andes for kindly providing the electoral and transfer data. The electoral data were collected by the Registraduria Nacional and the transfer data by the Unidad de Desarrollo Territorial. 71 Ley 60 (Decentralisation Law of August 1993) establishes transfers from the government to the municipalities of decentralised departments. Of these transfers, 25 percent must be spent on health and of these, 60 percent should go to the subsidized regime (targeted transfers). The other 40 percent are free health transfers. 72 We would like to thank the Misi6n Social of DNP for kindly providing the Gini coefficients. 211 Prob[y]= ID (X5) (1) where y is a Nxl column vector of observations on the dependent variable, possession of an SCD, X is a N x k matrix of the household and household head characteristics and municipal and departmental variables described above, X is a k x 1 column vector of unknown parameters that we want to estimate and 4Q(D) is the cumulative distribution function for the standard normal. The ECV/97 follows a stratified two-stage random sampling. The strata, or regions in the ECV/97, which are sampled separately, are statistically independent and can be analyzed as such. Models that do not take into account stratification might produce smaller standard errors. The primary sampling units are municipios (municipalities), while the secondary sampling units are segmentos (sub-clusters of 10 households within municipalities). In such a design, observations in the same cluster are not independent; for example households that are situated in the same municipality might have similar characteristics, etc. As Deaton (1997) points out, assuming independence of observations across clusters can also result in smaller standard errors - the difference can be a factor of 2 or more - yielding higher t-statistics. We therefore model affiliation in the contributory regime with a pseudo-maximum-likelihood probit model that takes into account the clustering and stratification of the survey design and accordingly corrects errors.73 We employ as a strata identifier the region variable and as a cluster identifier the municipality variable. 4.3 Empirical Results Separate regressions are estimated for urban and rural areas taking into account that the different institutional setting of the SISBEN index construction in these areas, both in terms of categories used and weights assigned. The sample size of urban households at SISBEN levels 1 and 2 is equal to 493 observations, while the rural sample includes 1,671 observations. In the case of SISBEN level 3 households, there are 702 and 746 observations in the urban and rural samples, respectively. 4.31 Estimates for the SISBEN level I and 2 populations Results of the estimation of the pseudo maximum likelihood probit (PMLP) for both urban and rural areas are reported in Table 6.11 for households in SISBEN level 1 and 2. The table includes the original PMLP estimates and their standard errors, as well as the estimated marginal or impact effects. The four factors of the SISBEN index have little effect in rural and urban areas. In the case of the urban model, none of these factors appear to be statistically significant at a conventional level. In the rural model, the demographic and occupational choice factor registers a statistically significant negative effect on the probability of take-up of an SCD. This factor may reflect better socio-economic status;74 the coefficient signs therefore provide weak support to the stigma hypothesis. Total household income75 - entered in log form76 - has a positive effect, significant only in the rural model. The computed marginal effect shows that a 10 percent increase in total household income increases the probability of having an SCD in rural areas by around 9 percentage points -in urban areas the effect is 7 percent and is only almost significant at the 10 percent level-. The positive coefficient of 73 For a detailed description of the model see Stata Manual 6.0. 74 A detailed description is included in Appendix A. 75 Definition of total household income is given in Section A.4 of Appendix A. Alternative definitions of income in which subsidies, assets and loans were excluded were also tested but proved to give very similar results to those reported here. 76 We also entered income linearly in both models; nevertheless, the logarithmic variable performed better. 212 income contrasts with the empirical results of the majority of the take-up studies, but validates the theoretical work of Cowell (1986) who supports that within the poor, the less poor are better informed and less constrained to search for and claim benefits. Table 6.11. Possession of an SCD: pseudo maximum likelihood probit of SISBEN 1 and 2 populations, selected variables Urban areas Rural areas Marginal Marginal Vanable Coefficient effects Coefficient effects Demographic and -0.0406 -0.0141 -0.0551 -0.0217 occupational factor 0.038 0.024 Household income (log) 0.2092 0.0728 0.2249 0.0888 0.129 0.131 Chronic problem 0.3550 0.1262 0.3025 0.1198 0.122 0.109 Gender 0.1344 0.0472 0.4905 0.1935 0.232 0.169 Unemployed -0.5063 -0.1554 -0.0338 -0.0133 0.261 0.596 Years in the same residence 0.0044 0.0015 0.0104 0.0041 0.003 0.003 Popularity 0.0124 0.0043 0.0127 0.0050 0.062 0.007 SDC rate 1997 0.0324 0.0112 0.0382 0.0151 0.007 0.007 Health insurance rate 1993 0.0208 0.0072 -0.0435 -0.0172 0.016 0.052 Hospitalization rate 1993 0.1058 0.0368 0.2567 0.1014 0.076 0.125 Constant -3.5894 -4.0294 1.244 1.207 No. of observations 493 1178 Note: ***, ** and * denote statistical significance at 1%, 5% and 10% respectively. Households with at least one member reporting a "chronic" health problem are more likely to hold an SCD than households where no member faces such a problem. This result is consistent with the entitlement hypothesis. In the case of SISBEN, entitlement is a function of the expected distribution of ill health; hence those who expect to need healthcare, such as the chronically ill, will enjoy higher entitlement levels than others. This result can also be interpreted as evidence of adverse selection. The other health-related variables, bad health status and health prevention, are poorly determined in both models. Gender affects are positive and well determined in the rural model. Age of the household head does not seem to affect the probability of having an SCD. Female household headship is correlated with SCD holding in the rural model. This result could be attributed to women's better information and greater concern about health-related issues. The finding that gender does not play an equally significant role in urban areas could be related to the change of male and female roles in the cities and women's double burden with household and labor tasks. An urban household with an unemployed head is 15 percentage points less likely to have an SCD compared to those with employed household heads. The lower take-up rates among the former type of 213 households may be related to the duration of eligibility. We expect that those who experience frequent changes of circumstance display lower take-up rates than those whose circumstances are more constant. Rural households with widowed or divorced heads are less likely to have an SCD than households in which the head is married. This may reflect the higher cost of claiming when there is no partner to share the burden of the claiming procedure. The alternative theory that the presence of a second adult in the family may improve the household's coping ability, without claiming benefits, thereby implying that single people may be more likely to claim than couples, is not supported by these data. Longer time in the same residence positively affects the probability of take-up in rural areas, although the magnitude of the coefficient is very small. This result is consistent with the hypothesis that those in more stable circumstances may find that the gain of a long period of benefit receipt outweighs the one-time claiming costs. Furthermore, the longer the tenure of a family, the more aware the local authorities may be of its socio-econornic status and the more they may encourage participation in the SISBEN program. The wider the popularity of the mayor, the greater the probability of take-up for households in urban areas. This finding is consistent with the hypothesis that a popular mayor will make a greater effort to satisfy his/her electorate demands. None of the rest of the municipal variables appears to have a significant effect on the dependent variable. The 1997 departmental take-up rate has a positive and statistically significant effect in both models. We interpret this variable as a proxy for economies of scale in information, where a large number of individuals with SCDs in a department is linked to better dissemination of information among all eligible households. Evidence shows that the higher the percentage of hospitalizations among the poor population of a department in 1993, the greater the probability of take-up of an SCD for a rural household residing in this department in 1997. A possible explanation of this finding stems from the theory on past utilization that individuals who have experienced health problems in the past and realize their cost on the household budget, are more likely to consequently acquire insurance in order to protect themselves from future expenditure. Although this variable is not on an individual level, we consider it the second best approximation of past utilization for the type of data available. A second explanation is the availability of local health services in the past, which increases the likelihood of receiving service once insured. 4.3.2 Estimates for the SISBEN/ level 3 population Results of the estimation of the pseudo maximum likelihood probit (PMLP) are reported in Table 6.12 for urban and rural households at SISBEN level 3.77 (Summary statistics are presented in Table A6.10 in Appendix B. For results of all variables, see Table A6. 11 in Appendix B). We compare these models to their counterparts for SISBEN levels 1 and 2 to establish if the determinants of take-up between the two samples are similar. Higher socio-economic status, as recorded through the housing factor and the demographic and occupational factor plays a negative role, perhaps because of the higher stigma associated with claiming a benefit. On the other hand, total household income has no effect. This stands in contrast to the case of the SISBEN level 1 and 2 model, where SISBEN factors were not so crucial and income played a positive role. 77 The results are reported analogously to those for SISBEN levels 1 and 2, in Table 11. 214 Table 6.12. Possession of an SCD: pseudo maximwn likelihood probit of SISBEN 3 population, selected variables Urban areas Rural areas Marginal Marginal Variable Coeff. effects Coeff. effects Housing factor -0.1185 -0.0287 *** -0.0418 -0.0153 * 0.029 0.023 Public service factors -0.0020 -0.0004 -0.0448 -0.0164 ** 0.036 0.020 Demographic and -0.0862 -0.0208 ** -0.0937 -0.0344 occupational factor 0.040 0.034 Bad health status 0.2114 0.0548 -0.3660 -0.1265 *** 0.162 0.135 Conservative 0.3196 0.0853 * 0.0714 0.0263 0.191 0.237 Liberal -0.0414 -0.0100 -0.2164 -0.0791 0.247 0.288 Gini -3.3551 -0.8113 * -0.0531 -0.0195 0.064 2.021 SDC rate 1997 0.0268 0.0064 *** 0.0292 0.0107 *** 0.006 0.005 Health insurance rate 1993 0.0277 0.0067 * 0.0239 0.0087 ** 0.012 0.011 Constant -1.3699 -1.9357 1.814 1.653 No. of observations 702 746 Notes. 1. ***, ** and * denote statistical signmficance at 1%, 5% and 10% respectively 2. Standard enors are reported in italics 3 Wald tests, distributed as x2, are used to test for linear restrictions. 4. Marginal effects for continuous variables are calculated at the mean Marginal effects for binary variables give the change in the probabihty of the dependent varable when the independent vanable changes from 0 to 1. 5 The variable "house ownership" is not included in the rural model because there are no observations talang the value I Age effects enter in a quadratic form, another difference with the SISBEN level I and 2 model. The positive linear and the negative quadratic coefficients suggest that younger individuals have a lower probability of take-up, but after approximately 50 years of age, the effect reverses. This result is consistent with the hypothesis that until a certain point, older age may indicate more healthcare needs, while thereafter it is more strongly associated with infirmity. Among political factors, it is interesting that, ceteris paribus, SISBEN level 3 households residing in an urban municipality. with a conservative mayor have a higher probability of holding an SCD. Recall the results of the SISBEN levels 1 and 2 model where the political party of the mayor had no significant effect. A possible reason for the difference with level 3 households is that conservative mayors may sympathize more with the interests of the less poor populations, thus favoring their demands. Although the popularity of the mayor does not feature as a statistically significant variable in the rural SISBEN level 1 and 2 model, it does have a significant positive effect in the rural SISBEN level 3 model. As for poorer households, the 1997 departmental take-up rate is positive and statistically significant in both models, again pointing to the existence of information externalities in departments with a higher number of SCD holders. Other well determined variables in the urban SISBEN level 3 model include the 1993 health insurance rate -positive - and the municipal Gini coefficient -negative-. As earlier, we associate the first with 215 greater awareness of health-related issues and increased likelihood of knowing about the SISBEN program in 1997. The negative sign of the Gini coefficient may indicate that the higher the inequality within a municipality - indicating most probably a greater number of poor people- the more difficult it might be for relatively better-off individuals to acquire an SCD. We now turn to the rural estimates for the SISBEN level 3 population and note the rather strong negative effect of not only the demographic and occupational factors but also the housing and the public service factors. They may suggest a stronger stigma effect for these populations. Similarly to the rural SISBEN level I and 2 model, chronic health problems exhibit a positive coefficient. On the other hand, the negative coefficient of the self-assessed variable "bad" health status is unexpected and indicates a negative relationship between entitlement and take-up that cannot be easily explained.75 Gender and marital status effects are similar to those of the SISBEN level 1 and 2 model, while the effect of number of years in the same residence is not well determined. Summary The analysis of this section has shown that the factors affecting the take-up of an SCD differ between households at SISBEN levels 1 and 2 and those at SISBEN level 3. This finding suggests that the two groups behave differently when it comes to obtaining an SCD. Within each income group, urban and rural estimates of take-up also differ. We have also shown that both demand-side and supply-side variables affect the probability of a household holding an SCD. On the demand side, the presence of a household member with a chronic problem implies a large increase in probability of acquiring a SISBEN card. Moreover, in rural areas female headed households and greater number of years at the same residence increase the probability of take-up. On the other hand, some household socio-economic status indicators, divorce and widowhood, and unemployment of the household head in urban areas have a negative impact on the dependent variable. On the supply side, the pre-reform level of access to health services and the intensity of take-up behavior in the region (department) increases the probability of SISBEN take-up, which suggests that previous availability of heath service supply and information externalities play a significant role regardless of household SISBEN level. Similarly, the more popular a mayor within his or her municipality is, the higher the probability that the households of this municipality possess an SCD. S. THE EFFECT OF HOLDING A SISBEN CLASSICATION DOCUMENT ON THE UTILIZATION OF HEALTHCARE SERVICES AND OUT-OF-POCKET EXPENDITURE This section presents the results of estimating the effect of the possession of an SCD on the utilization of and expenditures for healthcare services in the subsidized regime, controlling for the possible endogeneity of holding an SCD. We use data on hospitalization in the past 12 months, and medical consultations and medicines obtained in the past 30 days. 5.1 Theoretical Considerations We first briefly review the theory of demand for health and healthcare in relation to health insurance using an expected utility framework, where rational agents maximize a utility function defined over goods 78 The Spearman correlation coefficient between the variables of chronic problem and bad health status equals 0.269, indicating that including both in the model will not cause any multicollinearity problems. 216 and services and subject to a budget constraint. Following this approach, Grossman (1972) developed a model where the demand for healthcare is a derived demand; health is regarded as a fundamental commodity directly yielding utility, whereas healthcare is an intermediate one. Health status, as well as the effectiveness of medical treatment, is uncertain and in that sense, unpredictable and leading to irregularity in the consumption of healthcare. To limit the economic uncertainty characterizing health status after treatment, individuals seek health insurance. Protection will depend on different arrangements according to the nature of the uncertainty. Arrow (1963) distinguishes uncertainty in the incidence of illness and uncertainty in the effectiveness of the medical treatment. In the Colombian system, insurance offers protection against the latter in the form of income maintenance.79 By providing SCD holders with low-cost health services covered by the POSS. Such an insurance scheme may create incentives to increase consumption of healthcare, or moral hazard. This phenomenon arises when an insured individual alters his or her behavior in order to affect the probability of receiving a particular level of healthcare (Arrow, 1963; Pauly, 1968), since health insurance lowers the cost of healthcare borne by the individual. When patients are insured, physicians may also change their own behavior. They may prescribe more expensive and complex treatments than they would otherwise upon knowing that the patient does not carry the total financial burden of treatment.80 Insurance may also encourage individuals to spend less on preventive care and more on curative care (Pauly, 1986). Alternatively, in the knowledge that one is insured, one may pursue or continue health damaging activities, e.g., smoking (Besley, 1989). Holding an SCD may be endogenous to the decision of using healthcare services if both are influenced by a common variable, e.g., a preference for health. Assume that hospitalization and the possession of an SCD are both functions of observable variables X and Z, respectively. In addition, there is a variable v, say preference for health, which influences both hospitalization and the possession of an SCD. Hospitalization = f(X, SCD) + e(v), (2) SCD = g(Z) + u(v) (3) For example, an individual with stronger health preferences would be more likely to both seek healthcare and complete an FCS interview than an individual with less strong health preferences. The researcher can observe X and Z, but v is unobservable and thus cannot be included in the estimation of the equations of hospitalization and possession of an SCD. Endogeneity arises when there is correlation between a regressor and the error term. In this case, the SCD variable and E are both affected by the unobservable variable v and thus estimating hospitalization without taking into account the endogeneity of the SCD variable will lead to inconsistent estimates. Furthermore, the comparison groups (affiliated vs. non-affiliated) may not be randomly selected, as is often the case with non-experimental data, such as household surveys. Consumers that are expected to use (or not to use) health services choose to affiliate (or not affiliate). Ignoring endogeneity and/or self- selection will yield biased estimates of the effect of insurance/possession of an SCD on the utilization of healthcare services. Other individual factors such as health status, age and gender affect healthcare utilization. Since individuals most often use healthcare services when they are ill, health status is expected to be positively 79 The value of this protection is highlighted by the research of Perez et al. (2000) which shows how disease might bring a whole family into poverty. 80 Beyond the strictly "health" related costs, cost inflation may also be the result of an increase in the "non-health" care costs, such as provision of more luxurious surroundings, personal nurse attendance during the night etc. 217 and strongly correlated with utilization. Age and gender also enter as determinants of healthcare utilization since they determine morbidity. Pre-schoolers and the elderly face higher morbidity and thus, are expected to use more health services. Women use more healthcare services than men due, among other reasons, obstetrics related care and the fact that they tend to live longer. Education also features as a predictor of healthcare utilization, although its effect cannot be easily defined a priori. Better-educated people are better informed, which may lead to increased trust in a physician's advice. On the other hand, better-educated people tend to lead healthier life styles, thus decreasing the probability of utilization of healthcare services. Income and indirect costs of healthcare will affect demand. The ability to purchase services depends on income, thus indicating a positive relationship between the two variables. Indirect costs, such as time off from work, spent travelling or queuing may also be important in influencing an individual's decision whether or not to use healthcare services (see model by Acton, 1975). Family size and marital status could also affect service utilization. A higher number of family members may result in limited resources being shared among more individuals, including the total amount of care any individual receives, thus reducing the probability of an individual using a healthcare service. On the other hand, the presence of a second adult in the family may lead to better management of time within the household, and thus a higher probability of using a healthcare service when it is needed. In what follows, we explore the consumption of healthcare of Colombian households in SISBEN levels 1 and 2 with respect to utilization and out-of-pocket expenditures. The analysis focuses on the relationship between healthcare utilization/expenditure and the possession of an SCD. We expect that holding an SCD will lead to a higher probability of using a healthcare service mainly through lowering the price of the service to the user. On the other hand, we cannot a priori define the sign of the relationship between holding an SCD and healthcare expenditures. This will depend on the effect that the possession of an SCD has on the decision to seek care and the price elasticity of the demand for healthcare. 5.2 Methodology We estimate the two equations of utilization/expenditure and possession of an SCD for household heads. Given that the unit of analysis for the model of take-up of an SCD is the household, we decided to estimate the utilization regressions only for household heads. Variables other than the possession of an SCD that are included in the healthcare utilization and expenditure models are: o characteristics of the household head: health status, attitude towards health prevention, the presence of a chronic problem, the presence of a health problem in the last thirty days, age, gender, marital status, educational level, SISBEN level and employment status o household characteristics: household income and family size, and o municipal variables: existence of a health facility (health center or health post) in the municipality where the individual resides. The model describing the determination of utilization of healthcare services and SCD tenure is given by the following two equations: H, = Xi + yS, + F. (4) S,I = Z, + ul (5) 218 where i = 1., N, with N = number of individuals, H,* = the latent continuous dependent variable for the ith individual capturing the propensity to use healthcare services, S,I = the latent continuous dependent variable for the ith individual capturing the propensity of possession of an SCD, X, = a vector of characteristics that determines the ih individual's healthcare service utilization, 4 = a vector of characteristics that determines the ith individual's SCD holding status, £, and u, = error terms with H, = 1 if HI*> 0 H, = 0 if HI < 0 and S, = 1 if S, > 0 S, = 0 if S, < 0 where the dichotomous realization of the unobserved latent variable, H,* (resp. S,I) is provided by a dummy indicator variable, H,, (resp. S,,) which equals 1 if the observed individual receives the health service (resp. if the individual has an SCD) and 0 otherwise. We test for the exogeneity of S, using a test based on Smith and Blundell (1986). Given that the endogenous variable S, is a dichotomous variable, we estimate pseudo-probit residuals based on a method developed by Chesher and Irish (1987). The hypotheses are H.: £, is independent of S, and HA: £, is not independent of S, In the absence of endogeneity (the null cannot be rejected), we estimate the healthcare utilization equation using a pseudo maximum likelihood probit model. In the presence of endogeneity (the null can be rejected) we estimate the system using the Instrumental Variable (IV) approach.8' This approach aims at "purging" S, from its stochastic component and is implemented in two stages. In the first stage we regress S, on all pre-determined and exogenous variables, using a probit model. The predictions of this model, S,, express the endogenous regressor as the sum of two terms, the first being a combination of exogenous regressors (X, and 7) and the second a random error term. In the second step, we substitute the predictions into equation (1).82 Although A, is correlated with £, its instrument 5, is asymptotically uncorrelated with £,. The system is identified through variables that can affect possession of an SCD but not healthcare utilization and vice versa, such as the municipal and departmental variables in the SCD model. The model describing the determination of healthcare expenditure and the possession of an SCD follows a similar structure: E, = vM, + i S, + , (6) S, = oZ, + u, (7) where the same notations are used and in addition, F, is the latent continuous dependent variable for the ich individual, capturing the propensity to spend, M, is a vector of characteristics that determines the ith individual's healthcare service utilization, Ti, and u are error terms with E, equal to E, if F, > 0 and E, equalt to 0 otherwise. The main characteristic of equation (6) is that E, is censored at zero and ri; is 83 censored at -(vM,). We can estimate equation (6) using a tobit model. 81 The system was also estimated using a bivariate probit where the SCD variable entered the utilization equation as a function of all the variables determining it. However, the model failed to converge, possibly due to the small sample size. 82 The two-stage IV method produces large standard errors and thus may lead to the rejection of variables as non statistically significant when the contrary is true. Madalla (1983) reports the formula for constructing the appropriate variance-covariance matrix. However, an attempt to construct it produced evidence of a non-stable matrix; could be due to the small sample size. 83 For more details see Greene (2000). 219 We test for the exogeneity of Si using the same test as the one described above. That is, in the absence of endogeneity, we estimate the expenditure eauation using a tobit model. In the presence of endogeneity we estimate the system using the IV approach. 5.3 Empirical results As above, separate regression models are estimated for urban and rural areas. 5.3.1 Bealthcare utilization andthepossession of an SCD The empirical results reported in Tables A6.14 and A6.15 in Appendix B are based on the estimation of models of the utilization of hospital services, medical consultations and medicines for urban and rural areas, respectively. Each table includes the original estimates and their standard errors as well as the marginal and impact effects of the coefficients (Summary statistics are presented in Table A6.13 in Appendix B). Only in two cases do we reject the null hypothesis of the exogeneity of holding an SCD: hospitalization in urban areas and medical consultation in rural areas.85 We therefore estimate them with the IV method. In all other models, coefficients of the predicted residuals are not significant and thus we cannot reject the null and use a simple pseudo maximum likelihood probit model. In 4 out of 6 cases, the possession of an SCD does not seem to influence healthcare use. The central hypothesis to test is, whether, ceteris paribus, healthcare utilization is higher for individuals that face a lower net price of the healthcare service because they hold an SCD. This is the case only for the model of medical consultations in urban areas, where a household head with an SCD is 6 percentage points more likely to consult a doctor than a household head without an SCD. On the other hand, in the urban hospitalization model, the coefficient of the SCD variable is statistically significant but negative. Across all other models, the SISBEN coefficient is statistically insignificant indicating that the implementation of the program up to 1997 did not seem to affect the utilization patterns of the subsidized population for the services we study. A possible rationalization for the latter result is that it takes a while for poor people to increase their demand for healthcare as they come to realize the benefits to which they are entitled with their new insurance status. The subsidized regime first started in 1994 (Jaramillo, 1999), so it could be possible that the beneficiaries of the regime had not fully understood their rights by the time the ECV was administered, in the second semester of 1997.86 The ministry records reported considerable under- utilization of healthcare services among the poor after the beginning of the implementation of the reform. A second explanation could lie in the shortage of healthcare services in rural areas. The results suggest that the possession of an SCD affects utilization in urban but not in rural areas, possibly pointing to a supply constraint in rural areas. 84 An alternative approach is to estimate the two equations using a Heckman type "treatment effect" model. However, an attempt to estimate this model failed to converge, perhaps due to small sample size. For more details on the treatment effect model, see Limdep Manual, Version 7.0. 85 We use the regression estimates of Section IV. 86 This explanation was also put forward to the researcher in April 1999 in BogotA by the then vice-minister of health, Juan Pablo Uribe. 220 5.3.2 Healthcare expenditure and SCD holding Holding an SCD seems exogenous to healthcare expenditures for the services we study, in either urban or rural areas. The results of tobit expenditure models for hospital services (H), medical consultations (MC) and medicines (M) for urban and rural areas, respectively, show that in no case can we reject the null hypothesis. Our results show that the SCD does have a statistically significant and negative sign for the medical consultation and medicine models, in both urban and rural areas. This finding reflects the fact that holders of an SCD pay lower prices for healthcare services than non-holders do. The SCD variable has no effect in the model of expenditure for hospital services, possibly reflecting the fact that, for the majority of the population, hospital services are only partially covered by POSS. 5.3.3 Other explanatory variables Table 6.13 summnarizes the results for all twelve models of healthcare utilization and expenditure indicating which other variables appeared as statistically significant in each model and their sign. We only briefly comment on these results, given that our main concern is not the estimation of healthcare utilization models per se but the effect of the possession of an SCD on healthcare use. Table 6.13. Sign of statistical significant coefficients in the utilization and expenditure equations Variable Utilization Expenditure Urban Rural Urban Rural H MC M H MC M H MC M H MC M SCD - + Health: very - good Health: good - - Health: average Health + + + + + + + + + + prevention Chronic problem + + + + + + + + + + + Health problem + + + + + + + + Household + + + income Gender + + Age Age' + Free Union + Widow/er Divorced Single No education - Second. + + + + education Unemployed - Out of labor + + + + SISBEN 2 + Family size Health facility Notes hospital services (H), medical consultations (MC) and medicines (M). Only the signs of statistically significant coefficients are reported. 221 Self-reported health status variables are statistically more significant in explaining healthcare utilization and expenditure than socio-economic characteristicsu The parameter estimates are consistent with the view that short-term ailments (as captured by having had a health problem in the last thirty days) or chronic conditions are associated (relatively) more with the use of medical consultations and medicines. As far as the self-reported health status variables are concemed, results indicate that household heads reporting better health status face a lower probability of healthcare use and lower expenditures. Households heads who choose to visit the doctor at least once a year for preventive reasons, ceteris paribus, have significantly higher healthcare utilization and expenditure. The income parameter is found to be positive and statistically significant for utilization of medical consultations and expenditure for medical consultations and medicine in urban areas. This result is consistent with studies reporting positive income elasticities (see Grossman, 1972b; Acton, 1975; Coffey, 1983) and the finding that income is differentially important for different measures of utilization (Andersen and Benham, 1970; Hershley et al., 1975). Women are more likely to use hospital services and medical consultations in urban areas and spend more on medicine in rural areas, depicting the higher priority that women give to health issues compared to men. Age effects entering in the models in a quadratic form suggest that, in urban areas, the use of hospital services increases with age, while the age effect is negative on expenditure for medical consultations. Otherwise, the effect of age is statistically insignificant. Other factors affecting healthcare utilization and expenditure include education, employment status and SISBEN level. Better educated individuals tend to use more healthcare services and spend more on healthcare, especially in rural areas. Employment status, used as a proxy for the cost of time, suggests that individuals that are out of the labor force tend to have higher healthcare utilization and expenditure. Being in SISBEN level 2 is associated with an increase in the use of medical consultations, possibly reflecting the fact that individuals that must face higher co-payments prefer to use less costly healthcare services, such as medical consultations, as opposed to hospital visits. The presence of a health facility variable in the community significantly lowers expenditures for medical consultations and medicines in urban areas and hospitalization in rural areas perhaps because individuals tend to incur in smaller expenses when treated closer to their homes. Summarizing this section, we found evidence of endogeneity of the SCD variable only in the use of hospitals in urban areas and of medical consultations in rural areas, or 2 cases out of 12 use/expenditure cases. The results show that possession of an SCD in urban areas increases the utilization of medical consultations, while it decreases the probability of hospitalization. On the other hand, individuals who hold an SCD spend significantly less on medical consultations and medicine in both urban and rural areas. 6. CONCLUSIONS This paper has presented the factors that influence the take-up of an SCD in the subsidized healthcare sector in Colombia and the effect that the possession of an SCD has on the utilization and out-of pocket expenditure on healthcare services for the strictly eligible population of SISBEN levels 1 and 2. The findings shows that the take-up of an SCD in the subsidized regime is determined both by demand- side variables such as chronic conditions and household income -in rural areas-, as well as by supply-side variables such as the popularity of the mayor of the community, the intensity of services before the reform and the departmental take-up. Endogeneity is only important for hospitalization in the urban sample and for medical consultation in the rural sample. 222 Members of the subsidized scheme who are affiliated are more likely to undergo medical consultations and less likely to be hospitalized than non-affiliated individuals in urban areas. No such differences are apparent for rural areas. For both urban and rural samples, individuals with an SCD face lower expenditure for medical consultations and medicines than individuals without an SCD. (For a summary of all findings, see Table 6.14.) Table 6.14. Does the possession of an SCD affect the utilization of and expenditure for healthcare services? SCD Endogeneity Urban Rural Urban Rural Utilization Hospitalization - n.s. Yes No Medical Consultations + n.s. No Yes Medicines n.s. n.s. No No Expenditure Hospitalization n.s. n.s. No No Medical Consultations - - No No Medicines - - No No Note: n.s. denotes not statistically significant. The results of this study help in addressing two issues critical to the implementation of the reform health sector in Colombia. The first is related to the viability of the subsidized regime and the second to the welfare of its members. The evidence in this paper shows that expectations of higher healthcare use with provision of health insurance are not fulfilled, since individuals of the subsidized regime who are holders of an SCD are not more likely to use healthcare services than individuals who are non-holders. Critics of the reform claim that increased insurance coverage could lead to higher utilization of healthcare services, especially by the poor who have not enjoyed free insurance before. The increased expenditure, in return, would create financing problems in the system that could eventually result in its collapse. The hypothesis that healthcare use by beneficiaries of the new system has increased is not supported by the ECV/97 data. This finding suggests that the causes of the financial constraints faced by the Colombian healthcare system after the reform should be searched elsewhere and not within the utilization patterns of the poor. With respect to the welfare of insurance beneficiaries, our results show that affiliation and/or possession of an SCD have eased the financial constraints faced by the poor. The data show that individuals in households with an SCD have lower out-of-pocket health expenditures than those in households with no SCD. This result provides some evidence of the positive change that the health reform has brought to the lives of the poor in Colombia. Before the reform, lack of income was a constraint to access as reported in studies of healthcare utilization in Colombia (see Tono, 2000). In this paper, on the other hand, income features as a statistically significant variable only in three out of twelve healthcare models. This finding shows that the role of income has changed after the reform, possibly due to the subsidies, and that income lost its positive effect on the utilization of healthcare services, providing further support for the continuation of the policy of subsidizing healthcare services for the poor. Some inefficiencies nevertheless plague the subsidized sector. Three types of worrisome issues have emerged from our empirical analysis. The first disturbing finding is that affiliation and possession of an SCD appear to be synonymous. Although at first glance, this does not seem to create problems for the holders of an SCD who are not affiliated - since they face co-payments similar to the affiliated - in the long run it deprives them of the information and the network of providers that those affiliated to an ARS 223 enjoy. Furthermore, it creates problems for the system itself, since these individuals are receiving subsidies - through the supply subsidies to the hospitals - but are in fact unaccounted for in the healthcare system, which may contribute to the financial constraints mentioned above. This phenomenon suggests that information campaigns with respect to the nature and organization of this rather complicated health system would benefit the Colombian population and especially the poor, whose only insurance option is affiliation. In addition, an increase in the co-payments paid by individuals who hold an SCD but are not affiliated with respect to the co-payments of affiliated individuals will provide the former with a strong incentive to affiliate. The second worrisome finding is that in the majority of the municipalities where households of SISBEN level 3 hold an SCD, the condition that all households of SISBEN levels I and 2 should already have and SCD has not yet been fulfilled. This result suggests that it is not easy to control the procedure of households acquiring an SCD. In most cases, the mayor of the municipality has full responsibility for implementing and administrating the SISBEN program. This allows him or her, under specific circumstances, to act unilaterally. More community participation may improve the transparency and legitimacy of the whole process. Finally, the third perturbing finding is that, although the process of acquiring an SCD is free, claiming it may be costly. This conclusion is supported by the positive and statistically significant coefficient of the chronic problem variable in the take-up model. According to this result, households with chronically ill members are more likely to acquire an SCD because they expect higher entitlement levels (in the form of lower expenditure for healthcare services) than households with healthy members. This should cause officials to reconsider how the implementation of the SISBEN program takes place and find alternative and/or complementary ways of lowering the costs of claiming and identifying the majority, if not all, of the poor individuals in Colombia. 224 References Acton, J.P. 1975. "Non-monetary factors in the demand for medical services: some empirical evidence," Journal of Political Economy, vol. 83, pp. 595-614. Andersen, R. 0. and L. Benham. 1970. "Factors affecting the relationship between family income and medical care consumption," in Klarman, H. E. (ed.), Empirical Studies in Health Economics, Baltimore: John Hopkins Press. Arrow, K.J. 1963. "Uncertainty and the welfare economics of healthcare," American Economic Review, vol. 53, pp. 941-73. Ashenfelter, 0. 1983. "Determining participation in income-tested social programs," Journal of American Statistical Association, no. 78, pp.517-25. Besley, T. 1989. "The demand for healthcare and health insurance," Oxford Review of Economic Policy, vol. 5, no.1, pp.21-33. Blundell, R., V.C. Fry and I. Walker. 1988. "Modelling the take-up of means-tested benefits: the case of housing benefits in the United Kingdom," The Economic Journal, vol. 98, pp. 58-74. Castafio, E. and H. Moreno. 1994. Metodologia Estadistica del Modelo de Ponderaciones del Sistema de Selecci6n de Beneficiarios de Programas Sociales (SISBEN), Nota tdcnica no.], Misi6n Social, Departamento Nacional de Planeaci6n, Santafd de Bogota. Chesher, A. and M. Irish. 1987. "Residuals Analysis in the Grouped and Censored Linear Model," Journal of Econometrics, vol. 34, pp. 33-61. Coffey, R.M. 1983. 'The effect of time prices on the demand for medical services," Journal of human Resources, vol. 18, pp.407-24. Cowell, F.A. 1986. "Welfare benefits and the economics of take-up," London School of Economics, TIDI Discussion Paper, no. 98. Craig, P. 1991. 'Take-up of benefits: a survey," Journal of Social Policy, vol. 20, no. 4, pp. 537-65. Deaton, A. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy, John Hopkins University Press. Departamento Nacional de Planeaci6n. 1996. Evaluacion de la Etapa de Implementaci6n del Sistema de Seleccion de Beneficiarios de Programas Sociales (SISBEN), Reporte sobre Encuesta, Poryecto DNP - Programa de Naciones Unidas para el Desarrollo, Misi6n Social, Santafe de Bogota. Dorsett, R. and C. Heady. 1991. 'The take-up of means-tested benefits by working families with children," Fiscal Studies, vol. 11, no. 1, pp. 1-20. Duclos, J. 1992. 'The take-up of State Benefits: An application to Supplementary Benefits in Britain Using the FES," Discussion Paper no. WSP/71, Welfare State Programs, STICERD, London School of Economics. Fry, V.C. and G.K. Stark. 1987. "The take-up of supplementary benefit: gaps in the safety net," Fiscal Studies, vol. 8, no. 4, pp. 1-14. and . 1993. The take-up of means-tested benefits 1984-90, London: Institute of Fiscal Studies. Grossman, M. 1972a. The Demand for Health: A Theoretical and Empirical Investigation, New York, National Bureau of Economic Research. 225 1972b. "On the concept of health capital and the demand for health," Journal of Political Economy, vol. 80, pp. 223-55. Greene, W. H. 2000. Econometric Analysis, Fourth Edition, New Jersey: Prentice Hall International. Halpem, J. and J.A. Hausman. 1986. "Choice under uncertainty: a model of applications for the social security disability insurance program," Journal of Public Economics, vol. 31, pp.131-61. Hersley, J.C, H.S. Luft and G.M. Giannaris. 1975. "Making sense out of utilization data," Medical Care, vol.13, pp.838-54. Jaramillo, I. (1999), El Futuro de Salud en Colombia, Ley 100 de 1993, Cinco Afios Despues, Cuarta Edici6n, FESCOL, FES, FRB, Fundaci6n Corona, Santafe de BogotA. Madalla, G.S. (1983), Limited Dependent and Qualitative Variables in Econometrics, Cambridge University Press. Ministerio de Salud (1997), El ABC del Regimen Subsidiado de Seguridad Social en Salud, Red de Solidaridad Social, Santafe de BogotA. Moffit, R. 1983. "An economic model of welfare stigma," American Economic Review, vol. 73, no. 5, pp. 1023-35. Limder Manual, Version 7.0. Perez, F., C.E. Florez, E. Nina, L. Wartenberg, L.A. Rodriguez, H. Sanabria, E. Serrano, S. Grillo, S.E. Alonso and M.E. Andrade. 2000. 'Riesgos sociales y oportunidades de las familias colombianas'. Mimeo. Misi6n Social, Departamento Nacional de Planeaci6n, Santafe de BogotA. Pauly, M.V. 1968. "The economics of moral hazard: a comment," American Economic Review, vol. 57, pp. 231-7. Pauly, M.V. 1986. "Taxation, health insurance and market failure in the medical economy," Journal of Economic Literature, vol. 24, pp. 629-75. Reilly, B. 1988. Lecture Notes, Quantitative Methods for Masters in Development Economics and Masters in International Economics, Spring Term, University of Sussex. Smith R.J. and W. Blundell. 1986. "An Exogeneity Test for a Simultaneous Equation Tobit Model with an Application to Labor Supply," Econometrica, vol.54, no.3, pp.679-685. Tono T. 2000. Access to Healthcare in Colombia: The Effect of Income and Provider Availability on the Use of Medical Services in Seven Cities, unpublished PhD. Thesis, University of California, Los Angeles. STATA Reference Manual, Release 6, 1999. Velez, C. E. and V. Foster. 2000. 'Colombia Poverty Study. The Distributional Impact of Public Social Expenditure.' Mimeo. Poverty Unit, Poverty Reduction and Economic Management Division, Latin America and Caribbean Region, World Bank, Washington D.C. Velez, C.E., E. Castaiio and R. Deutch. 1998. 'An Economic Interpretation of Colombia's SISBEN: A Composite Welfare Index Derived from the Optimal Scaling Algorithm.' Mimeo. Poverty and Inequality Advisory Unit, Inter American Development Bank, Washington D.C 226 APPENDIXES TO CHAPTER I L.A Data, methodological considerations', and statistical appendix. Datasets. We use the "Encuesta Nacional de Hogares" (National Household Survey-ENH), carried out by the National Department of Statistics (DANE) and the National Department of Planning (DNP). The survey collects information on general attributes of the entire population (gender, age, education, etc.), and labor market variables for the "population of working age"2 (employment, occupation, income, etc.). We analyze the data from the surveys of June 1978, and September 1988, 1995 and 1999. Urban data contain information about Colombia's 13 largest cities, including the complete metropolitan area. Rural Colombia is comprised of the dispersed zones of all municipalities, non- municipal cabeceras, and about 850 remaining municipal cabeceras classified as rural based on population concentration, population of dwellings without adequate basic services, school attendance, value-added taxes, public and private service institutions, and the percentage of population employed by the agricultural sector. Rural municipal cabeceras normally have a population of less than 10,000. The rural data are classified into the following four regions: i. Atlantic, including the departments of Atlantico, Bolivar, Cesar, C6rdoba, la Guajira, Magdalena, and Sucre (31 percent cabeceras); ii. Oriental, including the departments of Boyaca, Cundinamarca, Meta, Norte de Santander, and Santander (26 percent cabeceras); iii. Central, including the departments of Antioquia, Caldas, Caqueta, Huila, Quindfo, Risaralda, and Tolima (28 percent cabeceras); and iv. Pacific, including the departments of Cauca, Choc6, Narifio, and Valle del Cauca (21 percent cabeceras).4 1999 data allow for desegregation into cabeceras and dispersed zones. Sample restrictions to guarantee comparability over time. To maintain comparability over time, we restricted the urban sample to the seven cities available in all surveys, which we loosely call "urban Colombia". They account for approximately two thirds of the urban population and are very heterogeneous in terms of location and socio-economic characteristics.5 Minimal income adjustments. We also decided to introduce as little "noise" to the data as possible. That is, we did not impute earnings for non-informants but removed observations with missing or zero values for household income, as well as households with at least one member employed at the time of the survey but with no income reported. To the extent that misreporting can be considered "quasi-random", the bias it causes is smaller than the one created by imputing ' Note: Methodology applies to Chapter 2, as well. 2 This includes individuals aged 12 or older. 3 L6pez Castafio, H., A. Cardona Arango, J. Garcia Zuluaga. 2000. Empleo y Pobreza Rural 1988-1997. CIDE, CEGA, IICA, TM Editores, Colombia. 4 The following departments, comprising about 2.6 percent of total national population, are not included in Colombia's National Household Surveys: Amazonas, Arauca, Casanare, Guainia, Guaviare, Putumayo, San Andres, Vaupes, and Vichada.. 5 The cities are Barranquilla, Bucaramanga, Bogota, Manizales, Medellin, Cali and Pasto. After the 1978 survey, the coding of metropolitan areas changed and thus we resorted to other methods to identify the seven cities. Although we are confident of the outcome of this exercise, some "leakage" may have occurred and observations from other cities may have been included, but this should not affect the analysis significantly. 227 values to the incorrect observations6. Nearly 20 percent of the original households records were removed from the analysis sample as a result of this adjustment. To account for this reduction in the number of observations, we re-scaled the urban sampling weights up, by dividing the sample into 42 city-strata cells (42=7 cities times 6 strata), and multiplying the original sampling weights by the ratio of the pre- to post-deletion number of weighted observations. The rural sampling weights were re-scaled in the same fashion, but only by dividing the sample into 4 regional cells for 1978, 1988, and 1995, and 8 region-cabecera cells for 1999. Top coding. Income top coding was a minor problem in the 1995 urban data when 0.08 percent of the sample of wage earners and 0.54 percent of self-employed displayed top-coded earnings. Because of these low proportions, we decided to ignore this problem. Simulations imputing higher incomes to these individuals did not alter results. Due to extreme outliers, rural data was top-coded using the following (nominal) cut-off points for average household income per capita, representing from 0.1 to 0.3 percent of the population: Table AIL..Top-coding cut-off points, rural data. 1978 26,444 1988 260,000 1995 1,376,000 1999 3,600,000 Rent adjustrments for homeowners. Due to their quantitative importance and their well-defined conceptual role, we made an exception and decided to impute rent payments to homeowners, who live in their property. These individuals extract a flow of goods and services from their property, which is not taken into account in their monetary income. Thus, the latter should be adjusted upwards to account for such in-kind earnings7. Otherwise, the comparison of income between renters and homeowner occupants is biased downwards for the homeowners. We develop a hedonic model of rent prices, which satisfies basic regression diagnostics and explains a large proportion of rent variability8. It also seems to avoid the problems of adjustments along the business cycle, present in the current Colombian Government methodology. We apply our model to the 1999 and 1988 data, the only years for which actual rent payments were reported. For 1995, we predicted rent payments using the 1998 equation coefficients, and deflated these predicted values using the DANE price index for rents. The 1978 data set contains no information on the characteristics of the dwelling unit. For urban data, we sorted the 1988 owners by their unadjusted household income in 20 percentiles, computed average imputed rents for these 20 groups, and applied them to the same percentiles of the 1978 distribution of non- adjusted household income of property owners. For rural data, this methodology consisted of computing deciles of unadjusted household income for owners in each region in 1988 and 1978, and then calculating mean rent imputed income in the formner year for each of these deciles. Then we computed the ratio between mean rent income and mean household income by income decile and household size in 1988. Finally, this ratio was applied to unadjusted income for the corresponding decile/region cell of the 1978 survey. 6 In addition, imputation with results from econometric predictions would lead to additional biases in the subsequent econometric analysis, due to correlation between error terms. 7 Another way to put it is that, while renters have to use a certain percentage of their monetary income to Fay for a place to live, owners who have fully paid for their property do not incur this expense. The exact specification of the model is available in appendix B. 228 Poverty lines and price indices by region. The poverty lines and price data were provided by DANE. DANE computes the extreme poverty line based on the minimum calorie and nutrient requirements of someone of average age and sex. The moderate poverty line is a multiple of the extreme one, which ranges from 2 to 2.5. Price data, were used to convert nominal incomes to real values expressed in 1999 pesos. Poverty lines and real values were computed separately for each city, therefore avoiding regional dynamic price effects. Separate poverty and extreme poverty lines were used for rural areas as a whole. Purchasing power parity converters from the World Bank's WDI were used to compute the US$ 2 per day poverty line. Table A1.2 Colombian CPI and PPP exchange rate ColombianC PPP Exchange PI rate' PPP Exchange rate2 1978 77.26 Col.$19.2 /US$ Col.$ 9.4/US$ 1988 9.24 Col.$ 104.61US$ Col.$ 85.4/US$ 1995 1.87 Col.$ 393.3/US$ Col.$ 393.3fUS$ 1999 1.00 Col.$ 650.6/US$ Col.$ 698.11US$ 1. Current US$ 2. 1995 US$ Source: DANE, WDI Table A1.3. Poverty, extreme poverty and US $2 a day lines in nonmnal Colombian pesos 1978 1988 1995 1999 Extreme Extreme Extreme Extreme Poverty poverty Poverty poverty Poverty poverty Poverty poverty Urban Barranquilla 1,922 871 16,474 8,087 80,508 37,501 63,061 163,581 Bucaramanga 1,776 803 15,226 7,461 75,234 33,028 56,935 151,782 Bogota 2,190 862 18,775 8,003 96,402 36,767 67,238 194,784 Manizales 1,692 792 14,504 7,357 75,062 31,454 57,405 145,765 Medellfn 2,048 852 17,557 7,917 81,334 32,066 55,460 170,549 Cali 2,095 853 17,957 7,927 90,789 34,729 58,524 158,775 Pasto 1,399 706 11,989 6,558 57,337 30,390 60,393 128,513 Rural 1,510 697 12,945 6,473 63,782 28,330 125,497 49,901 US $2 a day 573 5,198 23,923 42,469 Source: DANE, WDI data base. Welfare measured by income per capita. Our final measure of welfare is adjusted household per capita income, which is defined as the sum of incomes earned by each earner in the household, plus the imputed rent for the household, divided by the household size. For rural data, we excluded domestic employees and their offspring, as well as live-in-tenants when computing this figure. 229 LB Statistical Appendix: Table A1.4A. Social Indicators for Barranquilla, 1978-1999 1978 1988 1995 1999 Illiteracy rate* 7.3% 3.6% 3.3% 3.8% School enrollment Ages 7 to 11 90.4% 91.7% 96.7% 93.4% Ages 12 to 17 77.8% 84.6% 87.6% 85.0% Ages 18 to 22 19.1% 37.4% 43.5% 39.4% Child labor Ages 12 to 16 10.7% 4.7% 6.6% 4.8% Ages 12 to 14 4.7% 1.9% 2.6% 2.6% Access to public utilities Electricity NA 99.5% 100.0% 99.6% Aqueduct NA 92.2% 92.4% 98.6% Telephone NA 22.8% 22.8% 48.8% Sewerage NA 77.5% 75.1% 83.0% * For population 12 years old & older. Table A1.4B. Social Indicators for Bucaramanga, 1978-1999 1978 1988 1995 1999 Illiteracy rate* 8.8% 4.8% 3.9% 4.1% School enrollment Ages 7 to 11 89.4% 93.9% 95.6% 94.6% Ages 12 to 17 74.5% 72.7% 79.5% 81.4% Ages 18 to22 30.2% 29.8% 38.7% 39.4% Child labor Ages 12 to 16 15.4% 21.8% 17.9% 14.7% Ages 12 to 14 10.8% 13.7% 12.6% 7.3% Access to public utilities Electricity NA 98.4% 99.6% 100.0% Aqueduct NA 97.7% 99.6% 99.6% Telephone NA 44.8% 64.7% 90.2% Sewerage NA 98.6% 98.8% 99.5% * For population 12 years old & older. 230 Table A1.4C. Social Indicators for BogotA, 1978-1999 1978 1988 1995 1999 Illiteracy rate* 4.3% 2.8% 2.4% 1.9% School enrollment Ages 7 to 11 95.2% 95.7% 96.9% 95.9% Ages 12 to 17 79.1% 84.0% 85.2% 83.8% Ages 18 to22 37.4% 41.1% 45.7% 39.8% Child labor Ages 12 to 16 12.5% 11.7% 10.7% 7.3% Ages 12 to 14 5.6% 4.7% 5.5% 1.7% Access to public utilities Electricity NA 99.4% 99.4% 99.1% Aqueduct NA 99.2% 97.4% 99.1% Telephone NA 84.0% 88.9% 92.0% Sewerage NA 98.5% 98.0% 99.1% * For population 12 years old & older. Table A1.4D. Social Indicators for Manizales, 1978-1999 1978 1988 1995 1999 Illiteracy rate* 8.6% 5.7% 4.3% 2.5% School enrollment Ages 7 to 11 97.2% 93.0% 97.4% 94.8% Ages 12 to 17 87.4% 75.9% 85.7% 83.8% Ages 18 to 22 36.0% 35.2% 39.4% 39.9% Child labor Ages 12 to 16 7.5% 11.6% 8.3% 9.6% Ages 12 to 14 3.4% 7.9% 4.8% 5.9% Access to public utilities Electricity NA 98.7% 99.7% 99.7% Aqueduct NA 98.2% 99.9% 98.5% Telephone NA 38.6% 53.2% 79.5% Sewerage NA 98.8% 99.9% 99.2% * For population 12 years old & older. 231 Table AI14E. Social Indicators for Medellin, 1978-1999 1978 1988 1995 1999 Illiteracy rate* 6.4% 3.4% 3.1% 2.8% School enrollment Ages 7 to 11 88.9% 93.6% 95.9% 95.0% Ages 12 to 17 73.2% 77.7% 82.6% 82.6% Ages 18 to 22 23.1% 28.2% 31.7% 27.7% Child labor Ages 12 to 16 10.7% 10.3% 8.7% 6.8% Ages 12 to 14 5.6% 4.1% 4.8% 2.1% Access to public utilities Electricity NA 99.5% 99.5% 99.6% Aqueduct NA 97.7% 98.9% 98.4% Telephone NA 64.1% 76.6% 88.9% Sewerage NA 98.5% 98.9% 98.8% * For population 12 years old & older. Table A1.4F. Social Indicators for Cali, 1978-1999 1978 1988 1995 1999 Illiteracy rate* 4.6% 3.3% 2.4% 2.7% School enrollment Ages 7 to 11 87.1% 96.7% 95.5% 95.2% Ages 12 to 17 71.8% 75.6% 83.3% 75.3% Ages 18 to 22 25.7% 29.8% 36.1% 30.9% Child labor Ages 12 to 16 15.6% 12.9% 9.0% 17.6% Ages 12 to 14 8.1% 4.8% 3.5% 9.0% Access to public utilities Electricity NA 99.3% 100.0% 99.5% Aqueduct NA 95.5% 99.4% 99.6% Telephone NA 43.5% 58.0% 83.3% Sewerage NA 89.1% 99.5% 98.9% * For population 12 years old & older. 232 Table A1.4G. Social Indicators for Pasto, 1978-1999 1978 1988 1995 1999 Illiteracy rate* 4.3% 4.2% 3.2% 3.9% School enrollment Ages 7 to 11 98.3% 95.7% 99.3% 97.3% Ages 12 to 17 88.6% 80.4% 85.0% 80.4% Ages 18 to 22 60.9% 45.8% 41.3% 35.8% Child labor Ages 12 to 16 0.0% 15.6% 13.1% 16.4% Ages 12 to 14 0.0% 8.4% 9.1% 9.0% Access to public utilities Electricity NA 99.3% 99.9% 99.9% Aqueduct NA 98.7% 99.3% 99.5% Telephone NA 30.4% 33.2% 52.3% Sewerage NA 97.3% 97.6% 97.4% * For population 12 years old & older. 233 Table A1S. Income inequality and poverty indicators, 7 Cities 1978-1999 1978 1988 1995 1999 Barrnquila Mean mncorm per capital 82,225 177,428 223,410 217,833 Ginm 0 365 0 464 0 481 0 502 P90/PlO 51 7 3 79 93 P75/P25 2.7 27 2.8 3 0 Share q5/Share ql 5 9 95 110 13 5 Poverty rate 89 9% 68 3% 564% 62.2% Poverty Gap 49.6% 30 8% 22 9% 27.5% P(2) 32.0% 170% 122% 153% Extmre poverty rate 48 9% 28 5% 18 4% 15 9% Bucaramanga Mean incone per capita' 146,663 226,669 283,487 213,043 Gim 0 403 0 455 0464 0 445 P90/PlO 511 665 582 704 P75/P25 2 86 2 56 2 49 2 63 Share q5/Share ql 8 43 9.45 9 45 12 24 Poverty rate 62 2% 50 8% 360% 51.6% Extremepovertyrate 218% 139% 53% 86% PovertyGap 270% 183% 114% 20.5% P(2) 15.6% 9 0% 5.2% 106% Bogota Mean incomepercapital 182,017 281,512 371,922 345,385 Gim 0 468 0 490 0 548 0 569 P90/PlO 7.7 95 86 15 7 P75/P25 29 31 31 40 ShareqS/Shareql 10.8 124 15.2 19.4 Poverty rate 65 2% 51 4% 45.9% 52.3% Poveny Gap 314% 214% 180% 25.5% P(2) 185% 11 6% 94% 15.6% Extrerne poverty rate 209% 15.0% 8 4% 15 5% Manizaks Mean income percapital 157,629 195,705 243,601 219,105 Gin, 0 479 0 528 0 501 0 490 P90/PlO 11.1 94 68 105 P75/P25 3.2 3 0 26 3 5 Share qS/Share ql 122 14 4 114 13 9 Poverty rare 65.2% 58 4% 55.3% 52 6% Poverty Gap 29 9% 27.0% 21 5% 23.5% P(2) 17 7% 15 6% 11 2% 13 6% Extreme poverty rate 267% 27 3% 11 5% 15 2% MedeLin Meanincome percapital 144,420 225,868 249,148 218,821 Gim 0 484 0455 0 475 0505 P90/Plo 84 70 77 96 P75/P25 2 9 2 6 28 3 2 Shae q5/Share ql 11.6 97 10 9 15.4 Povertyrate 749% 561% 494% 619% Poverty Gap 38.8% 22 6% 18 7% 28 7% P(2) 242% 119% 96% 17.1% Extreme poverty rate 31.5% 16 5% 89% 140% CaU Mean income per capital 160,208 248,345 275,165 246,547 Gmi 0459 0475 0483 0513 P90/PlO 72 79 72 103 P75/P25 29 28 27 3 1 Share q5/Share ql 10 2 11 0 106 15 8 Poverty rate 68 5% 54 9% 490% 53 7% Poverty Gap 33 6% 22.7% 190% 23 3% P(2) 20.0% 12 2% 9 8% 13 3% Extrenie poverty rate 264% 151% 8 8% 12 3% Pasto Meanincomepercapital 128,708 161,233 191,404 198,454 Gin, 0466 0510 0483 0502 P90/plo 10 1 98 83 99 P75/P25 3 3 3 3 30 34 Shareq5/Shareql 83 140 11 1 133 Poverty rate 6377% 56.7% 444% 513% Poverty Gap 309% 264% 17 3% 23 2% P(2) 16 7% 15 4% 8 8% 13 3% 234 Table Al.6. Logit results: Marginal effects on the probability of being poor, Urban Colombia, 1978-1999. 1978 1988 1995 1999 Marginal Marginal Marginal Marginal Vanables coeff z coeff z coeff. z coeff z Number of household members Youngerthan7yrs 7% 986 23% 2352 10% 1481 11% 1983 Fromnage 12to65 1% 424 6% 1101 1% 2.20 3% 818- Fromage7toll 7% 975 22% 2198 13% 1602 11% 1650" Older than 65 -4% -2 36 -2% -0 97 -2% -1 22 -4% -2 75 Disabled 0% -0 06 3% 0 83 6% 172 2% 106 Male household head Youngerthan28yrs 10% 126 10% 214 6% 069 21% 401 From28-35 yrs 0% 000 4% 1 18 8% 131 17% 426 From 36-42 yrs -4% -0 82 2% 0 70 9% 131 13% 4 07 From 43-47 yrs -4% -104 1% 0 38 6% 1 31 11% 4 15 From 48-57 yrs -4% -1 30 3% 158 3% 101 6% 311 Older than 67 -2% -0 49 3% 1 37 0% 0 01 8% 2 93 Female household head Youngerthan36yrs 16% 189 18% 355 31% 377 24% 457 From36-47 yrs 11% 193 17% 476 24% 401 20% 543 From48-57yrs 4% 094 14% 464 25% 501 16% 511 Olderthan57 3% 064 8% 288 17% 359 13% 428 Head's education Uneducated 1% 0 21 13% 3 36 8% 2 37 13% 5 26 High school dropout -5% -310 -15% -814 -7% -411 -6% -482 High school -15% -5 84 -32% -11 76 -20% -8 01 -18% -11 60 College dropout -15% -3 97 -53% -11 53 -39% -9 40 -29% -12 44 College graduate -30% -610 -78% -15 30 -65% -13 06 -46% -15 89 Head's sex I head's education -2% -3 39 -2% -4 88 -2% 4 15 -1% -3 42 Employment rate -18% -9 71 -61% -28 56 -42% -2188 -27% -22 49 Household's average education -2% -7 29 -5% -19 83 -4% -15 90 -2% -15 37 Labor market vanablesfor household head Unemployed first ume 20% 0 80 12% 0 41 26% 158 4% 0 48 Unemployedbutworkedbefore 19% 3 99 55% 1157 36% 943 25% 14 83 Student 2% 012 6% 040 12% 1 73 Household work 12% 3 94 25% 7 39 6% 2 00 9% 4 83 Rentier3 -6% -I 26 -18% -221 -22% -233 -11% -I 99 Pensioner -2% -051 4% 106 -3% -1 18 1% 030 Disabled 32% 441 61% 683 24% 457 18% 384 Other 14% 325 21% 462 13% 327 15% 601 Head's expenence I % 165 0% -0 78 0% 0 64 0% 0 64 Head's expenence squared 0% -214 0% 1 18 0% -073 0% -138 Head's marital status Consensual union 2% 124 7% 3.87 6% 3 90 5% 4 91 Widower 3% 0 82 -5% -1.29 -8% -2 34 0% -010 Separated/divorced 2% 058 0% -008 -1% -032 0% 014 Singe 9% 2 91 4% -1 20 -10% -2 91 -1% -0 30 Household labor market characteristics Both wage earners and self employed -2% -1 40 -4% -2 22 -5% -3 02 2% 145 Only self employed 3% 173 4% 221 -1% -076 10% 1003 Nonlabor income only 2% 0 52 -7% -2 13 -15% -5 21 -7% -4 24 235 Regional effects Bucaramanga -15% 4.58 -29% -8.28 -36% -1147 -14% -7.31 Bogota -6% -2.56 -7% -3.02 -14% -6.77 0% -0 10 Manizales -7% -1.95 -13% -3.09 4% -1 09 -8% -2.93 Medellin -8% -3 39 -16% -6.38 -12% -5.68 -2% -1 56 Cali -8% -3.26 -12% 4.50 -12% -5.34 -7% 4 23 Pasto -28% -6.66 -15% -3.16 -19% 4.33 -9% -3.34 Household ownership Owners but paying mortgage 5% 2.83 25% 10.06 21% 9 08 15% 10.46 Renters 31% 19 13 19% 13.90 13% 14.39 Usufruct -- 39% 1167 28% 7.73 23% 11.72 De facto 50% 6 18 24% 2.21 26% 1.64 Constant 4% 0.45 33% 3.07 34% 3.41 19% 3.00 Number of observations 2375 12263 10115 10075 chi2 408.78 3086 61 3526.81 2626.84 Prob > chi2 0.00 0 00 0.00 0.00 Pseudo R2 040 0.42 0.33 0.43 * Significant at IO% level or less ** Significant at 5% level or less Marginal logit coefficients evaluated at means. 236 Table A1.7. Change in relative risk: Groups that fared well & poorly, Urban Colombia 1978-1999 Change in relative risk 1978-1988 1988-1995 1995-1999 Fared Poorly Household size 6 to 10 persons 10.2% 8.1% 0.1% Employment household zero 14.2% -8.8% 50.5% Head's sector of economic activity Construction 3.3% 8.4% 1.4% Head's occupation Self 12.5% -3.4% 6.1% Head's education Uneducated 27.7% -3.1% 10.8% High school dropout 10 1% 10.4% 6.7% High school 9.5% 11.1% 8.5% Head's Marital status Widower 0.2% -4.9% 9.6% Education average household members older than 12 yrs. High school dropout 7.5% -25.1% 53.3% High school -4.0% -24.5% 46.3% College graduate 7 2% -7.2% 10.6% Fared Well Employment household 0 F * 0.0000 Rsquared = 0.5153 Root MSE n .64027 Robust Rent Payment Coef. Std. rr. t P>t [95% Cord. lntervl Material-Well-Rerence: Brick IvivwO_2-Adobe -0 0035756 0 0017535 *2 039 4.10% -00070123 -00001388 IvrvOO3-Bahareque -0 1642386 0 0048402 -35 395 0 00% -01733332 -0 155144 ImWO_4-8ahareque buldo -0 2103338 0 0074821 -28111 000% -0 2249985 -0 195689 IvivOO 5-Wood 00378225 0 0078147 4 84 000% 0022506 0 0531391 IvivooB-Guadua -0.5184655 00175083 -29613 000% -05527811 -04841499 IVIVWC 7-Cloth (dropped) e terlal-Floor-Reference. Brick ImvvOl_-Sand 00081726 00046523 1757 790% -00009459 0017281 IvivOl_2-Cement -0232733 0W011687 -199137 000% -02350238 42304424 MvrvOl-3-Wood -0 0908822 0 D022524 -40 261 0 00% -0 0950987 -0 0882676 IvivOl-5-Luxury wood 0 2069032 0 002833 73 033 0 00% 02013506 0 2124558 tvivO 86Carpet 03195708 0W023201 137741 000% 03150235 03241181 Aqueduct vivO1 5 0 4938752 0 0076288 64 74 000% 0 4789233 0 508827 Electricity vwlv17 -00920338 00088711 -13394 000% -01055007 -00785665 Telephone vivOl8 01043443 00012803 81501 000% 0101835 01068536 Crtystmte-nReferece: Mecelltn, Strata 3 Ba'rrnquills Icue 11 -04888941 00136518 *35812 000% -05156511 -0462137 Iciue_12 -0321479 00058062 -55 368 000% -0332859 -03100991 Iciue_13 -0 1159459 0 0034782 -33 335 000% -0.1227631 -01091288 lcwue_ 4 01713456 0005651 30 321 0 00% 01602698 0.1824215 Iciuel15 07575978 0 0050038 151 405 0 00% 0 7477905 0767405 Iciue i6 1 209278 0 0057604 209.93 0 00% 1.197988 1 220568 Bucaramanga Iciues21 -0 2552538 0 0080824 -31.582 000% -0 271095 -0 2394127 IcLue.22 -03112936 00034991 -88 964 000% -03181517 -03044356 lCiue_23 -01329901 0 002585 -51 438 000% -01380577 -01279225 Icaue_24 02102275 00033374 62992 000% 02036863 0.2187688 ICLue_25 0 6711379 0 006462 103 859 000% 0 6584726 0 6838032 Icrue 28 0895032 0012342 72519 000% 08708422 09192219 Bogota Iciue31 -02511075 00061107 41093 0 00% -02630843 -02391307 litue232 -0 0471237 0 0018968 -24 844 000% -0 0508412 -0 0434061 IcLue_33 0 0916021 0 0017082 53628 0 O 0'/ O 0882542 0 0949501 Ilue_34 03148865 0 D028795 117 516 000% 0 3096347 03201382 lcute_35 05054518 00042923 117758 0 00% 04970391 05138845 clue 36 09278261 00036868 251 663 000% 09206001 09350521 iMarCitsee lcfue_41 0 9471387 0 0307751 -30 776 0 00% -1 007457 -0 8868206 ci ue_42 -0 3022119 0 0083803 -36 062 0 00% -0 3186369 -0 2857888 Icrue._43 -0 3848628 0 0043842 -87 784 0 00% -0 3934557 -0 3782699 ICLue44 00287731 0004485 5998 000% 00180218 00355244 lciue-45 08664098 00164077 40494 000% 06322513 06965683 Mtedellin Iclue 51 -02255819 0 0046238 48 787 000% -02346443 -02165194 lclue_52 -0 2058007 0 0021926 -93 881 000% -02100982 -02015033 Iclue_54 04608396 00024982 184465 000% 0 4559342 0465727 Iclue.55 08610935 0003to27 277534 000% 08550124 08671746 lclue 56 0 8387216 008188 102 433 0 00% 08226734 08547698 Call Iclue_61 -0 3784481 0 03585 -105 583 0 00% -0 3854748 -0 3714215 tcus 62 - 2288548 0 0023518 -97 312 0 00% -O 2334642 -0 2242454 Iciue_63 0 0040462 0 0020312 1 992 4 60% 0 0000651 0 0080273 IclueL84 0 3992509 0 0037779 105 68 0 00% 0 3918463 0 4066555 tciue 85 0 7401306 00038099 194265 00% 07326633 07475979 lcue866 1194178 00224054 53299 0 00% 1 150264 1238091 Pasto lcue_71 -05661634 00110215 -51369 000% -05877652 -05445616 lcaue_72 -0 5396845 0 D070542 -76 506 0 00% -0 5535105 -0 5258586 ciuej73 -0 2207738 0 D041479 -53 226 0 00% -0 2289035 -02126441 lciue_74 0,1494257 00077699 19231 000% 01341969 01846545 lcuet75 06546874 00109577 59747 0 00% 06332106 06761642 C or rooms hoW006 0 4465707 0 0019452 229 58 0 00% 04427582 04503831 Siantary-Reference: Sewage IhogOl_2-Septic Well -0 2183051 0007591 -28 495 0 00% -02311833 -0201427 lhogO1I3-Notconnected -0336869 00114838 -29.334 000% -03593769 -03143612 ihogOl_4-etdne -0 7547235 0 011O335 -66 584 000% -07 T69397 -0 7325074 lhogOl_5-banmar -01196819 00146902 -8147 000% -01484742 -00908895 IhoqO1_6-no sanilary servce -0 059059 0 0193628 -3 05 0 20% -0 0970089 -0 021109 knotl -00504894 00007179 -70299 000% -00518765 -00490623 knot2 00121194 00007887 15368 000% 00105735 00136852 knot3 -0 0303213 0 005248 -57 804 0 00% -0 0313494 -0 0292932 Constant cons 1044983 00086302 1210838 000% 1043291 1048674 245 1H1. Estiimation results The results of the estimation exercise are presented in Table Al. All the variables but two are significant at confidence levels under 0.01 percent. Additionally, the majority of them present the expected coefficient sign and their combined significance is very high. Table 9 reports the marginal effects of selected variables on rent payments; the results make intuitive sense.. The fit is close to 52 percent and the mean root square error is around 54 percent. Notwithstanding these facts, several adjustments were made to the initial specification. First, as it is very common in regressions involving nominal money figures, the initial equation displayed severe symptoms of heteroskedasticity (the Goldfeld-Quandt test was used to corroborate the presence of this problem), which compelled us to use a robust variance estimator, which has no effect on the estimated coefficients but produces consistent variance estimates. The estimator used was White(s "Sandwich" estimator (see Hamilton, 1992 pp 547-548). The plot of the fitted values versus the residuals (from the corrected regression) (Figure Al. in the appendix) shows no noticeable pattern that would make us suspect the presence of heteroskedasticity. Figure 1. Model Hl. Residuals vs. Fitted values of home rent payments. Urban Colombia, 1998 residuals vs fitted values 3.51154 0 0 0 0 0 0 0 ~ 0 0 Oo 0 I0~~~~ 0 0 0~~~~~~~ 0 ~~~0 0 0 -3 7897 I0 9.62949 13 8328 Fitted values We also tested for the presence of outliers (by using the square distance between each individual observation and the center of the probability mass), and found approximately ten of them, so we proceeded to eliminate them from the sample and obtained no change in the estimation results. Thus, they were reintroduced into the equation. Figure 2 in the appendix clearly indicates the possibility of either nonlinearities or discontinuities the variable "number of rooms" -hogOO6-. To cope with this problem we constructed a cubic spline from the original variable. This partitions the support of the variable into the appropriate number of segments (generally, at those points where the discontinuities or nonlinearities surface) and creates polynomials of order three that act as quasi-dummy variables in order to smooth the path of the conditional expected value of the variable. In this case, three additional variables were created and included in the estimation (the "knot" variables in Table 1), which considerably improved the goodness of fit. 246 Figure 2. Rent and the number of rooms 3 79098 0 0 0 ~~~0 0 0~~~~~~~~~~~~~~~~ 0. 0.~~~~~~~~~ 0.~~~~~~~~~~~~~~ 8 Co~~~~~ E 0~~~~~ 0)~~~~~~ 0 -3.00301 - 0 1 14 numero de cuartos o piezas We attempted to incorporate a variable to capture the different values that people place on the materials a house is built out of, depending on the climate of the region where the unit is located. More particularly, we included an interaction term between climate and materials, but this exercise yielded high collinearity in the estimation (too many dummies). Hence, it was dropped. Finally, we performed an informal but widely used test of misspecification consisting of regressing the dependent variable (rent payments) on the predicted values and square predicted values obtained from the application of the model. If the model is correctly specified, its predicted values should have no explanatory power when used as independent variables against the original dependent variable. This was the case in our test, since the squared predicted values were only significant at levels larger than 30 percent. Finally, to avoid underestimation of the variance of the predicted rent payments, we incorporate the error terms to the final rent payments. We carried out this task by randomnly sampling from a distribution with mean zero and variance equal to the one computed for the true disturbances. We repeated this sampling process 100 hundred times and thus ended up with 100 (slightly) different rent payment series. Then, the Gini coefficient for these 100 hundred series were calculated, and finally we chose the series which displayed the closest Gini to the average of the 100 indices computed in the previous step. Thus, our estimated rent payments are of the form exp(r, ), where r, = d + bx, + vi, Model Comparison Table 2 in the appendix compares the predicted imputed rent values for owners for both models. The top part corresponds to the estimations made by Model Y and the lower part corresponds to our estimations, Model H. It is important to note two things. First and most important, the standard deviation of our estimated values is significantly higher than model Y's (the former is approximately 150 percent of the latter), reflecting mainly two things. First, the exclusion of the error terms in model Y, which artificially reduces the dispersion (variance) of the predicted rent payments, and second, a point that will surface again below and is related to the fact that, by construction, model Y tends to underestimate the inequality of rent income. The other important 247 difference is related to distributional aspects. While the means are not that far apart (the ratio of model H/model Y is 1104 percent), the distributions are very different. This can be seen by comparing the cut-off points for the different percentiles presented in Table 2. For example, in our computations, the 10' and 90a percentiles are given by rent values of $68,101.6 and $496,587.4 respectively, while in model Y's estimation, the corresponding values are $127,765 and $372,112. Finally, both medians are practically the same, indicating less inequality in the distribution of model Y computation. This is important because, as mentioned earlier, inequality in Y's imputed rent values is a direct function of household income; so depending on what part in the distribution of the latter the renters and owners are located, by adding imputed rent values to household income, we are altering its distribution in a way that is not independent to current income. That is, the fact that inequality is significantly different between the two imputed rent values (Ginis of 0.27 and 0.43) implies that the measure of household income inequality taken on the adjusted income may be very different depending on what adjustment method is used, and the "double-counting" problem referred for model Y may introduce a spurious component in the final inequality measure, not to mention the artificial reduction implied by the exclusion of the disturbances mentioned above. Figure 3. o modely A modelh 362576 56225.3 - v _ _ , l _ 153530 4.3e+06 Ingreso sin ajustar Figure 3 confirms what was already known: model Y predicts rent payments that when plotted against unadjusted household income display exactly the same qualitative behavior than the latter, reflecting the basic features of the estimation methodology (i.e., household income determines rent payments). In model H, on the other hand, although rent payments tend to grow with household income in general, the slope of this relationship is flatter, especially along the highest deciles of unadjusted household income, than the one observed under model Y. We also note how the models predict very similar rent payments for low levels of household unadjusted income (along the first five to six percentiles on the latter's distribution), and model H's predictions are more dispersed than model Y's along this range. Finally, it is clear from the data presented in this figure that rent payments tend to be higher under model H than under model Y until the 9h decile of household income is reached, indicating an important underestimation of rent income along this interval if the latter model is used. For the highest income levels (101 decile) the situation is the opposite: model Y overestimates rent payments. 248 IV. Poverty and inequality measures under alternative imputed rent models Tables 4 to 7 present poverty, extreme poverty and inequality under three possible scenarios: unadjusted household income and adjusted earnings, distinguishing the two adjustment methodologies (model Y or model H). Independent of the method of adjustment, imputing rent income significantly reduces both poverty and extreme poverty. For the entire sample, the fall in the poverty headcount is of about 6 percentage points under model Y and 7 percentage points under model H. For model H, adjusted poverty represents approximately 86 percent of unadjusted poverty, while this figure reaches 88 percent if model Y is used. The corresponding reduction for the poverty gap is 5 percentage points under both imputation models, and adjusted extreme poverty represents 81 percent of the unadjusted figure. The proportional impact of the adjustment is larger for the poverty gap than for the headcount, showing that while a certain number of people are taken out from poverty due to adjustment, the ones that remain poor improve their situation. The implications of this exercise are more significant for extreme poverty. Imputing rents reduces the head count ratio of extreme poverty by nearly 5 percentage points with either methodology, which represents almost a third of total extreme poverty, reducing it from 16.2 percent to 11.7 percent. And for the extreme poverty gap, the rent imputation makes the gap fall until the new estimate is about 60 percent of its original value before adjustment. If we examine the figures for each of the seven cities in the sample, some interesting facts are revealed. First, the impact of imputing rents on the poverty headcount seems to be quite homogeneous across urban areas (i.e., the reduction in the poverty headcount is higher for all cities if model H is used instead of model Y). Second, the gap seems to be fairly independent of the choice of adjustment methodology, except for the case of Barranquilla (this is true for both poverty and extreme poverty). Third, in cities where poverty and inequality are very high (e.g., Barranquilla), the reduction in extreme poverty is much stronger under model Y. This highlights an interesting feature of model Y: by regressing rent payments on current household income, the mean of the predicted values may be artificially pushed upwards by observations with high income (in the case of Barranquilla this seems to be the case, since it is a city with one of the highest levels of inequality of unadjusted income). This fact, combined with the aforementioned compression of incomes around the mean, will in turn push up every household's imputed rent income, especially those with very low incomes. The other interesting revelation is that model Y fails to capture regional effects independent of household income and affecting the value property. Compared to model H, at the aggregate level, model Y tends to underestimate the reduction in poverty (Tables 3 and 7). However, this effect is more significant in the case of extreme poverty, for which it represents 0.4 percentage points that account for nearly 4 percent of the total number of extremely poor individuals. In addition, the alternative methodologies of imputing rent produce poverty household sets that not only differ in size, but also in composition. Table 3 presents the "poverty sets" under both methodologies. Taking model H as the comparison base model, 6,658,072 people are poor and of these, model Y classifies 222,975 as non-poor (Type I, exclusion error 3 percent). Model Y, on the other hand, finds that 6,744,788 are poor, while model H does not classify 309,691 of them as poor (Type II: 4.6 percent inclusion error). In summary, the two poverty sets differ in nearly 7.6 percent of their composition. The differences are much wider if we look at the extreme poverty set. Almost 9.1 percent of-the extreme poor, as grouped by model Y, are considered non-poor by model H (Type II error), while model Y excludes from the extreme poverty set more than 9.91 percent (type I error) of the 249 individuals classified as poor by model H. In this case total differences in composition add up to 19 percent. To conclude the poverty comments, we can refer once more to Tables 10 and 11. If we compare the proportions of rent payments to total household income by decile using unadjusted or adjusted (under model H) household income, we can gain a little more insight on why the effect of the adjustment is higher on extreme poverty than on moderate poverty. It suffices to notice how this ratio drops from almost 73 to 60 percent in the first decile, while for other nine the effect is only marginal. Finally, both methodologies reduce overall inequality but have differing impact on inequality measures. We computed the Gini coefficient for the entire sample with unadjusted income, using both adjustment methodologies. We find that, while for model Y, the reduction is nearly 1 percentage point -from 55.4 to 54.5-, model H reduces the Gini coefficient by almost two percentage points -from 55.4 to 53.8-. Although there is no theoretical or empirical presumption that inequality should fall after imputing rents, it is not completely surprising that inequality falls more under model H. Remember that Figure 3 showed that model Y tended to overestimate rent payments for higher income levels, hence adjusted incomes tend to be larger for the higher income deciles, with obvious consequences for inequality. This can be confirmed also by noting that the quasi-Ginis (Gini indices computed from the decilized information) show an important reduction under model H, when compared to model Y. V. Summary and Conclusions Rent payments are empirically relevant; they account for 26.5 percent of total household income of renting households. Moreover, this figure is significantly higher for low income households, and lower of high income ones. This not only means that a significant portion of income is generated through rents, but also carries important implications for poverty and inequality measurement. Ignoring them would result in a significant upward bias in the measure of poverty and an a-priori unknown one for the case of inequality. We find that without imputed rent adjustments inequality, poverty and extreme poverty are overestimated, and the error seems to be much larger in the case of extreme poverty. We also find that the methodology currently used in Colombia for imputing rent payments (model Y), overestimates poverty and inequality. Additionally, as a result of both inclusion and exclusion errors, model Y produces a set of poor and extremely poor households that differs from the one determined by the methodology presented in this paper (model H.). This latter methodology is based on the assumption that the price of living units depends directly on both its physical characteristics, and on its unobservable ones, such as location. Model Y, on the other hand, rests on the presumption that rent payments are a function of household current income. The consequences of this latter perspective were clearly identified in the paper, with the almost perfect correlation between household current income and predicted rent payments being the most important and making the final measure of income inequality highly unreliable. In terms of poverty measurement, although model Y overestimates it, the differences (both qualitative and quantitative) are of not so important magnitude. Lastly, our goal is to use this model back in time as long as Encuesta Nacional de Hogares data permit and to produce consistent series of poverty and inequality measures for the Colombian economy. We will carry out this task by applying the estimated coefficients for 1998 to the actual respective variable values for earlier years, and then deflating the imputed rent value by the regional rent price index, as published by DANE. 250 References Griliches, Zvi. 1971. Price Indexes and Quality Change. Cambridge, Mass.: Harvard University Press. Hamilton, L.C. 1992. Regression with Graphics. Pacific Grove, CA: Brooks/Cole Publishing Company. King, A. Thomas and Peter Mieszkowski. 1973. "Racial Discrimination, Segregation, and the Price of Housing." Journal of Political Economy v81, n3 (May-June): 590-606. Rodrik, D. 1999. "Why is there so much insecurity in Latin America?" World Bank Rosen, Sherwin. 1974. "Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition." Journal of Political Economy v82, n2 (Part II, March-April): S164-S169. 251 Table 2: Predicted Rent Payments, Comparison Percentiles Model Y Model H 1% $99,794.4 $26,978.6 5% $115,523.4 $48,825.3 Model Y Model H 10% $127,765.6 $68,101.6 Obs 1800672 1795589 25% $149,649.6 $109,067.7 Sum of Wgt. 1800672 1795589 50% $183,856.6 $185,743.0 Wean $230,191.9 $253,437.7 Std. Dev. 158971.9 238349.9 75% $250,984.3 $314,403.1 90% $372,112.8 $496,587.4 Variance 2.53E+10 5.68E+10 95% $488,523.0 $696,640.2 Skewness 4.793836 3.014552 99% $946,346.0 $1,222,384.0 Kurtosis 42.04919 16.44182 Gini: 0.277 0.423 Total: $414,500,126,720 $455,069,946,305 Table 3: Poverty Set under alternative models Poverty model y non-poor poor Total non-poor 7,135,395 222,975 7,358,370 96.97%_ 3.03% 100.0% 95.84%1 3.35%1 52.2% poor 309,691 6,435,097 6,744,788 r - - X 2 459%95.41% 100.0% 4.16% 96.65% 47.8% Total 7,445,086 6,658,072 14,103,158 52.79 47.21% 100.0% 100.00 100.00% 100.0% Extreme model y non-poor poor Total non-poor 12,386,05 155,212 12,541,269 98.76% 1.24% 100.0% 98.86% l 9.86% 88.9% poor 142,220 1,419,669 1,561,889 . 9:11 % 90.89% 100.0% 1.14% 90.14% 1-1;1% Total 12,528,277 1,574,881 14,103,158 88.83%9 _ 11.17% 100.0% 100.00% 100.00% 100.0% 252 Table 4: Povertv measures under alternative models Non-adjusted model Y model H Head Count Poverty Gap Head Count Poverty Gap Head Count Poverty Gap Sample Urban 55.0% 25.2% 48.7% 20.3% 47.2% 20.2% Barranquilla 63.0% 30.8% 54.1% 23.2% 54.7% 24.3% Bucaramanga 47.8% 18.8% 43.5% 15.8% 42.6% 15.3% Bogoti 50.8% 24.1% 45.3% 19.8% 44.8% 19.5% Manizales 55.9% 23.3% 48.5% 19.1% 47.9% 18.8% Medellin 56.3% 25.2% 48.7% 19.9% 48.2% 19.9% Cali 57.4% 25.5% 51.4% 20.6% 49.7% 20.3% Pasto 53.4% 21.7% 48.0% 18.4% 46.8% 18.2% Gmni 0.554 0.545 0.538 Table 5: Extreme poverty measures under alternative models Non-adjusted model y model h Head Count Poverty Gap Head Count Poverty Gap Head Count Poverty Gap Sample Urban 16.2% 5.1% 11.2% 3.0% 11.2% 3.0% Barranquilla 26.3% 8.9% 15.7% 4.2% 17.8% 5.2% Bucaramanga 10.1% 3.0% 7.4% 2.0% 7.0% 1.9% Bogoti 15.0% 4.8% 11.1% 3.0%/o 10.7% 2.8% Manizales 15.7% 4.9% 12.0% 3.2% 12.4% 3.2% Medellin 13.8% 4.1% 9.4% 2.2% 9.5% 2.4% Cali 15.9% 5.2% 10.7% 3.0% 10.8% 3.1% Pasto 20.1% 5.3% 15.6% 3.8% 15.8% 3.8% Table 6: Ratio Poverty models(h.yY)Dover t non-adiusted income model y model h hc pg hc pg Sample Urban 88.5% 80.8% 85.8% 80.4% Barranquilla 85.9% 75.4% 86.8% 78.9% Bucaramanga 90.9% 84.2% 89.1% 81.6% Bogota 89.3% 82.1% 88.1% 81.0% Manizales 86.7% 82.1% 85.7% 80.7% Medellin 86.5% 79.1% 85.7% 79.0% Cali 89.7% 81.0% 86.6% 79.6% Pasto 89.9% 85.0% 87.7% 84.0% 253 Table 7: Ratio Extrenm Poverty models(h,y ext.poverty non-adjusted incoma model y Me mE Ih Sample hc p h c pg Urban 69.3% 57.90/o 72.3% 60.00/o Barranquilla 59.7% 47.0% 71.70/o 59.2% Bucaramanga 72.7% 68.5% 69.6% 65.1% BogotA 74.2% 62.0% 75.1% 60.8% Manizales 76.3% 65.9% 77.4% 65.9% Medelifn 68.0% 53.20/o 68.70/o 56.8% Cali 67.0% 57.4% 69.1% 58.4% Pasto 77.6% 72.1% 78.4% 73.8% Table 9: Marginal Effects of Selected Variables on Rent Pafments Variable 11 Marginal Effect Material/Wall-Reference: Brick Adobe [ -0.4% Bahareque | -15.1% Wood [ 3.9% MaterlallFloor-Reference: Brick Cement [ -20.8% Carpet 1 37.7% Aqueduct If 63.9% Telephone Ii 11.0% City/Strata-Ret. Medellfn, Strata 3 Barranquilla/6 235% CaW6 230% Bogota/6 153% Medefifn/5 137% Cali/5 110% Medellin/4 59% BogotdI3 10% Cafi/3 0.4% Medellin/3 0% BogotaI2 -5% Medellfn/2 -19% Medellin/l -20% Bogota/i -22% Sanitary-Reference: Sewage Septic Well -19% Not Connected -29% # of rooms second 49% third 23% fourth 19% fifth 7 7% sixth 4% 254 Table-10: Descriptive Statistics for Actual Rent Payments RenVlncome Decile Non-renters Renters Total for Renters 1 206246 145781 352027 72.2% 58.6% 41.4% 100.0% 2 181262 176716 357978 45.3% 50.6% 49.4% 100.0% 3 187708 151612 339320 42.0% 55.3% 44.7% 100.0% 4 197628 151739 349367 34.9% 56.6% 43.4% 100.0% 5 234132 163830 397962 31.8% 58.8% 41.2% 100.0% 6 188474 113274 301748 27.8% 62.5% 37.5% 100.0% 7 244148 128117 372265 30.0% 65.6% 34.4% 100.0% 8 244042 103270 347312 22.0% 70.3% 29.7% 100.0% 9 242439 87381 329820 19.7% 73.5% 26.5% 100.0% 10 281353 68201 349554 12.6% 80.5% 19.5% 100.0%° Total 2207432 1289921 3497353 25.6% 63.1% 36.9% 100.0%° _ Gini 0.274 standard deviation 0.4199 255 Table 11: Descriptive Statistics for Actual Rent Payments (Adjusted Deciles) l lI| Rentlncome Decile Non-renters Renters I Total for Renters 1 121167 228766 349933 60.8% 34.63% 65.37% 100.00% 2 158878 190956 349834 43.3% 45.42% 54.58% 100.00% 3 183463 166068 349531 36.4% 52.49% 47.51% 100.00% 4 194131 155625 349756 32.9% 55.50% 44.50% 100.00% 5 230561 133340 363901 29.8% 63.36% 36.64% 100.00% 6 237025 99006 336031 27.0% 70.54% 29.46% 100.00% 7 249960 99389 349349 27.9% 71.55% 28.45% 100.00% 8 262097 87566 349663 22.2% 74.96% 25.04% 100.00% 9 277260 72731 349991 17.6% 79.22% 20.78% 100.00% 10 292890 56474 349364 12.4% __ _ 83.84% 16.16% 100.00% Total 1 2207432 1289921 3497353 63.12% 36.88% 100.00% 36.7% Standard Deviation 0.4199833 256 TO CHAPTER II 2.A. Statistical Appendix Table A2.1: Social indicators by region, Rural Colombia 197841999 1978 1988 1995 1999 Adantic Illiteracy rate* 46 1% 290% 22 3% 20.9% School enrollnment Ages 7 to 11 63 0% 81 4% 894% 89.7% Ages 12 to 17 57.6% 69.9% 761% 77 9% Ages 18 to 22 12 6% 169% 23.5% 24 6% Child labor force Ages 10 to 16 18 7% 17.5% 12 5% 116% Ages 10 to 14 14 3% 14 1% 8.6% 7.6% Access to public utilities Electncity NA NA 85.9% 80.9% Aqueduct NA NA 60.2% 57 5% Telephone NA NA NA 48% Sewerage NA NA 102% 12 5% Oniental Illiteracy rate* 229% 17 3% 13.3% 12 5% School enrollnent Ages7to 1 734% 898% 94.1% 916% Ages 12 to 17 47.5% 55 7% 597% 61.7% Ages ISto22 93% 159% 20.8% 22.1% Child labor force Ages 10 to 16 24.5% 36 2% 27 5% 20.5% Ages IO to 14 17.3% 30.5% 20.3% 13.6% Access to public utlitues Electicity NA NA 900% 87 9% Aqueduct NA NA 52 7% 62 4% Telephone NA NA NA 19.9% Sewerage NA NA 30.0% 38 0% Ce,aeral llihteracy rate* 19.8% 134% 11 2% 12 5% School enrollment Ages 7 to 11 70.8% 83.6% 84.7% 88 5% Ages 12 to 17 34 4% 49.8% 57 5% 63 2% Ages 18 to 22 48% 12 2% 14.8% 17.7% Child labor force Ages lOto 16 272% 320% 27.7% 20.9% AgeslOtol4 220% 278% 198%e 143% Access to public utilities Electncity NA NA 87 0% 87.8% Aqueduct NA NA 69 1% 59 5% Telephone NA NA NA 23 7% Sewerage NA NA 49.5% 39.2% Pafc Illhteracy rate* 34.2% 15 6% 12.1% 13.5% School enrollrnent Ages 7 to 11 53 5% 86.6% 94 5% 92.7% Ages 12 to 17 36.5% 55.1% 63 0% 60 5% Ages 18 to 22 10 7% 13 8% 19 3% 17.1% Child labor force Ages lOto 16 316% 22.9% 186% 251% Ages IO to 14 22 5% 16 5% 11 0% 19 0% Access to public utilhes Electrcity NA NA 84 8% 78 1% Aqueduct NA NA 77 8% 68 8% Telephone NA NA NA 11.4% Sewerage NA NA 35.1% 364% * For population 12 year old & older 257 Table A2.2: Income inequality and poverty indicators by region, Rural Colombia 1978-1999 1978 1988 1995 1999 AlLantic Household monthly income per capita' Median 38,630 48,949 54,924 64,745 Mean 56,549 69,996 75,447 103,952 Income inequality Gini 43.7% 45.6% 41.7% 50.4% P90/PlO 5.9 8.8 6.7 8.0 P75/P25 2.7 3.0 2.6 2.7 Share q5/Share ql 11.3 12.0 8.8 13.9 Poverty Poverty rate 93.3% 88.2% 85.2% 80.0% Poverty Gap 58.4% 52.1% 47.5% 43.6% P(2) 41.1% 35.8% 30.9% 28 0% Extreme poverty rate 65.4% . 59.9% 48 3% 36.4% Oriental Household monthly income per capita' Median 29,580 55,232 70,993 74,000 Mean 42,745 83,004 102,599 117,488 Income inequality Gini 45 3% 50.4% 46.8% 50.9% P90/PlO 7.6 11.2 9.3 10.6 P75/P25 3.1 3.3 3.0 3.2 Share q5/Share ql 9 4 14.3 11.4 15.6 Poverty Poverty rate 95.3% 82.8% 74.4% 73.3% Poverty Gap 66.4% 48.5% 38.6% 39.0% P(2) 50.7% 33.6% 24 9% 25.4% Extreme poverty rate 75.4% 55.1% 34 8% 31.7% Central Household monthly income per capita' Median 44,630 84,503 77,157 68,800 Mean 62,598 105,162 97,724 103,493 Income inequality Gini 42.3% 39.6% 38.4% 48 8% P90/PlO 5.6 6.4 57 8.7 P75/P25 28 2 7 2.4 2.9 Shareq5/Shareql 97 11 0 9.1 14.0 Poverty Poverty rate 91.5% 70 3% 76.6% 78.2% Poverty Gap 54.0% 32.9% 35.2% 42.0% P(2) 366% 19.6% 20.5% 27.2% Extreme poverty rate 57.3% 33.3% 28 6% 33.6% PacLi0c Household monthly income per capital Median 32,874 65,486 63,149 51,741 Mean 46,309 98,535 104,279 76,893 Income inequality Gini 45.7% 48.9% 51.8% 48.5% P90/PlO 8 5 7.7 7.2 10.6 P75/P25 3 3 3 0 2.8 3.2 Share q5/Share ql 10.0 12.6 12.5 13.1 Poverty Poverty rate 94.4% 79 0% 79.8% 84 9% Poverty Gap 64.1% 41 9% 42.3% 51.6% P(2) 48.2% 26.9% 26.9% 36.2% Extreme poverty rate 73 5% 44.4% 41.0% 47.7% 1. 1999 pesos, based on monthly household income 258 Table A23A: Unemployment for various demographic groups, Atlantic region. 1978 1988 1995 1999 Atlantic region 1.3% 5.5% 6.1% 8.2% By education Uneducated 0.9% 4.7% 1.5% 4.0% Primary 1.3% 2.8% 5.7% 6.2% High school dropout 5.5% 8.7% 7.9% 9.4% High school 0.0% 14.1% 16.0% 19.8% College dropout 0.0% 2.8% 11.1% 15.7% College 0.0% 0.0% 8.6% 5.1% By age 12to 17 2.9% 4.8% 11.1% 9.7% 18 to 25 1.8% 8.4% 14.3% 15.6% 26 to 36 1.9% 3.9% 4.6% 8.9% 36 to 50 0.1% 1.7% 2.5% 5.1% 51 to 65 0.6% 3.2% 1.2% 3.6% Over 65 0.0% 5.0% 2.0% 0.9% I.Population 10 & older. 2.Uneducated: 0 years of education. Primary: 1-5 years of education. High school dropout: 6-10 years of education. High school: 11 years of education. College dropout: 12-15 years of education. College: 16 years of education. Table A2.3B: Unemployment for various demographic groups, Oriental region. 1978 1988 1995 1999 Oriental Region 0.8% 4.1% 4.0% 9.8% By education Uneducated 0.4% 2.0% 0.4% 3.9% Primary 0.8% 3.2% 2.6% 6.9% High school dropout 2.3% 8.0% 9.5% 18.1% High school 0.0% 14.3% 8.4% 19.1% College dropout 0.0% 2.8% 12.2% 16.2% College NA 0.0% 2.3% 10.6% By age 12 to 17 1.3% 7.6% 6.5% 12.3% 18 to 25 2.2% 8.2% 7.5% 16.1% 26 to 36 0.0% 2.0% 4.0% 12.3% 36 to 50 0.0% 1.5% 2.5% 6.6% 51 to 65 0.0% 1.8% 1.2% 3.8% Over 65 2.1% 0.7% 0.2% 0.9% I.Population 10 & older. 2.Uneducated: 0 years of education. Primary 1-5 years of education. High school dropout: 6-10 years of education. High school: 11 years of education. College dropout: 12-15 years of education. College:16 years of education. 259 Table A23C: Unemployment for various demographic groups, Central region. 1978 1988 1995 1999 Central region 1.8% 4.6% 5.4% 12.3% By education Uneducated 1.6% 4.7% 4.3% 5.0% Primary 1.3% 2.8% 3.7% 8.9% High school dropout 7.1% 8.7% 9.7% 20.8% High school 5.7% 14.1% 13.0% 25.1% College dropout 23.4% 2.8% 3.5% 20.9% College 0.0% 0.0% 6.3% 2.2% By age 12 to 17 3.3% 4.8% 5.5% 16.8% 18 to 25 3.2% 8.4% 9.6% 23.7% 26 to 36 0.5% 3.9% 5.3% 11.6% 36 to 50 0.2% 1.7% 3.9% 7.6% 51 to 65 0.2% 3.2% 2.0% 6.4% Over 65 4.0% 5.0% 1.0% 2.2% I.Population 10 and older. 2.Uneducated: 0 years of education. Primary: 1-5 years of education. High school dropout: 6- 10 years of education. High school: 11 years of education. College dropout: 12-15 years of education. College: 16 years of education. Table A2.3D: Unemployment for various demographic groups, Padfic region. 1978 1988 1995 1999 Pacific region 2.0% 4.7% 4.5% 13.3% By education Uneducated 1.3% 1.6% 1.1% 6.3% Primary 2.0% 3.3% 3.2% 11.1% High school dropout 10.5% 10.3% 7.3% 17.7% High school 0.0% 14.8% 12.0% 32.6% College dropout NA 14.2% 8.2% 12.8% College NA 0.0% 0.0% 3.8% By age 12 to 17 1.3% 7.8% 8.7% 18.2% 18 to 25 3.6% 11.0% 10.2% 24.9% 26 to 36 1.4% 3.8% 4.1% 13.0% 36 to 50 2.0% 1.1% 1.6% 8.4% 51 to 65 0.5% 0.3% 0.4% 5.1% Over 65 4.0% 0.7% 0.7% 4.8% I.Population 10 years and older. 2.Uneducated: 0 years of education. primary: I-5 years of education. high school dropout: 6-10 years of education. high school: 11 years of education. college dropout: 12-15 years of education. college: 16 years of education. 260 Figure A2.1. Welfare by region, Rural Colombia 1978, 1988, 1995,1999 120,000 WAelfars Atlantic Region 105,000 90.000 60,000 45,000 30.000 15.000 1978 1988 1995 1999 Weifare: Oriental Region 120,000 175,000 o 90,000 02 (D 75,000 0. Oi 60,000 0) 45,000 30,000 19U000 1978 1988 1995 1999 1-lMoanincomel -4S-Sefllndex Welfare: Central Region 120.000 105,000 90.000 - 75.000 i 60,000 - 45,000 30,000 15,000 __ ,.... 1978 1988 1995 1999 Welfare: Pacific Region 120,000 t 105.000 UO 99000D 0) 75D000 0, 60.000 _ _ _ _ _ _ _ _ _ 45,000 30,000 15,000 1978 1988 1995 1999 261 TO CHAPTER M 3.A. Data Methodological Considerations Urban Data Surveys The household surveys used for this paper are EH19 from June 1978, EH 61 from September 1988 and EH 89 from September 1995, all from DANE, Colombia. To maintain comparability over the years of the study, we represent the urban area by the largest seven cities or what is called -"Urbano Tradicional"-. Thus we exclude medium and small sized urban areas -"Resto Urbano"- from the surveys of 1978 and 1995. The remaining group of cities is what we call "Urban Colombia". They account for 67 percent of the urban population (average over 1978-98) and are very heterogeneous in terms of location and socio-economic characteristics (they are Barranquilla, Bucaramanga, Bogota, Manizales, Medellin, Cali and Pasto)'3. Top Coding and Non-Informants We were concerned with the integrity of the data, which led us to decide to introduce as little "noise" as possible. That is, we did not impute earnings to non-informants, nor did we correct for the top-coding problem present in the ENHs. Regarding non-informants, our decision was to delete them. That is, we discarded all households who reported total income of zero or did not report at all (missing). Additionally, all households in which at least one of its members was employed at the time of the survey but did not report income were deleted. These adjustments implied the deletion of nearly 20 percent of the original number of households present in the sample. To account for this reduction in the number of observations, we re-scaled the sampling weights up, by dividing the sample into 42 city-strata cells (42 = 7 cities times 6 strata), and multiplying the original sampling weights by the ratio of the pre to post deletion number of weighted observations. Price Deflators All income sources in the three household surveys were deflated with Colombian CPI provided by DANE, Departamento Nacional de Estadistica. For each city we applied its own CPI, to take into account differences in regional prices. The base month and year is December 1988. Rural Data Rural household surveys for the years 1978, 1988 and 1995 were used. These surveys are carried out by the DANE.'4 The size of the surveyed samples were, respectively, 13,084, 18,781 and 19.992 (3,504, 5,603 and 6,020 households). These samples were taken in the four most populated regions in the country (Atlantic, Oriental, Central and Pacific). The relevant information was taken only from households whose income was obtained by labor activities, considering thus all the other income sources as exogenous. 131n the 1978 survey the coding of the metropolitan areas changed and thus we resorted to other methods -mainly demographic- to identify the seven cities. Although we are confident of the outcome of this exercise, it is not 100% exact. 14National Department of Statistics (Departamento Nacional de Estadisitca) 262 The following adjustments and corrections of the surveys were done: 1. The rurality criteria considered in the polls since 1988 were applied when analyzing the 1978 poll, thus being able to make comparisons. 2. In the three polls, households were removed from the analysis mainly for three reasons - the head of the household did not tell what was the labor income, in spite of being occupied, -The proportion of the "not informants" was more than 50% of the occupied population.3. People in charge of the domestic service who live in the households, their children and pensioners were removed from the analysis.4. Households with no income were removed.5. Outliers were removed. 6. The factor of expansion was calculated again taking into account the "Territorios Nacionales," as well as the households that were eliminated due criteria formerly outlined. 3.B. Analytical Framework Decomposing the Dynamics of Income Distribution C.E.Velez and C. Bouillon In this appendix we introduce the analytical framework to (i) understand the determinants of the distribution of individual and household income and, then (ii) to decompose the changes in inequality by isolating its dynamic responses to the changes in prices of skills, in skill endowments, and in structural parameters of participation and occupational choice, over time. We will follow the model and basic notation established in Bourguignon, Ferreira and Lustig (1999), the methodological paper that guides this cross-country project about dynamics of income inequality in Asia and Latin America. The proposed decomposition framework is a dynamic extension of the Blinder-Oaxaca decomposition. In addition to isolating changes in the prices and quantities of observable skills, this decomposition also incorporates joint changes in the prices and quantities of unobservable skills as in Almeida dos Reis and Paes de Barros (1991) and Juhn, Murphy and Pierce (1993) and, identifies the contribution of participation and occupational choice decisions on the changes in earnings inequality. We represent the distribution of income as the outcome of two interrelated processes: (i) the generation of earnings as a function of observed and unobserved characteristics of the individual and the corresponding market prices for those characteristics and (ii) the individual decision to participate in labor activities -wage-employment, self-employment- vis a vis labor inactivity -like household work or studying, etc.- as a function of the potential benefits of those alternative occupational choices, and in relation to characteristics of the household and the individual. 1. Income generation and the distribution of income Income of household i in period t is identically true to the sum of labor income of its members Y,Ij ( = 1, ...J) plus any other exogenous sources as transfers and property income Yo. y,, = £, y,g, + Yo,, Therefore, in order to understand household and individual income we most specify the determinants of the decision to be active in the labor markets and the determinants of their earnings. For simplicity, suppose for the moment that each household has only one member, an individual's income is given by a Mincerian earning equation, ln y,, = /3, x,, + El , where vector xi, represents the observable characteristics of the individual that determine his productivity level, l/, the prices or returns to those characteristics and e, the value of productivity determinants that are not observable for the individuals in the sample. The vector of characteristics, x,,, includes variables such as education level, potential labor experience, regional 263 location and gender. Thus, for individual i, income changes from one period to the next will depend on the observed changes of individual characteristics x,, , their prices 8, , and the distributional parameters of F, -the value of productive unobservables-. Nevertheless, this potential to generate income may not be realized if the individual in question does not participate in the labor market. We consider three possible labor statuses: 1) inactivity; 2) wage earner; and 3) self-employed. The individual will choose one state K -1, 2 or 3- if it provides him with the highest utility. That is, he will choose state K of (1,2,3) if UK is the greatest of all three options of behavior. Uk is defined as Uk, = 4, Z, + Ak k = 1, 2, 3 where, Z,, is a vector of observable characteristic of the individual and his household affecting the participation decision, , is the vector of corresponding parameters that measure the advantages or disadvantages of choosing each state k given each characteristic Z,, and, finally, Ak, are the benefits of unobservables characteristics of the individual and/or his household for each state k. Observable characteristics affecting labor participation Zi, must include two types of variables in addition to the ones already employed in the earning equation -education, experience, region of residence, gender-. First, variables that determine the value of the time in household activities vis a vis working in labor activities. For example, the presence of very young children at home may raise the value of staying at home for a member of the family or, altematively, the presence of another adult that could carry out child care could tip the occupational choice balance in favor of labor participation. Therefore, the demographic structure of the household of individual i should enter into the Z,, vector. Second, variables that determine the individual's reservation wage, i.e.: the marginal utility of consumption vis a vis the marginal disutility of work, such as earnings of other members of the family, or variables that indicate a capacity for such earnings -e.g.: education and labor experience, should also enter into the Z, vector. The probability of any individual i being in state k is given by Pk = exp (i4. Zd, / 2=J,Z3 exp (,, Z,d) Notice that, although occupational choices are observed, we do not observe the implicit utility levels for each occupational status, Uk, (k=l 2 3). Therefore, correspondingly, multinomial logit procedures are used to estimate A, (j=1,2,3), but inference about the distributional parameters of the errors Akt is not feasible. In summary the earnings of an individual i depend upon his observable and unobservable characteristics [x,,, ZIb 4,, At] and the parameters of the earnings and participation equation [A, i,] that hold for every individual in each period t. To make the notation more compact, we collapse all observable characteristics x,, , Z,, , into one larger vector and call it x,, . Thus the income of an active or inactive individual i at time t is represented as y,, = y I{l (X ,} , 4,,i,, i = 1, . .., n where {.} refers to the joint distribution of the corresponding variables over the set of individuals. Therefore, if the inequality index of the distribution of income on individuals is given by a function D [.] of all incomes, inequality in period t is given by 264 D, = D [{ylm Y2r,Y3,,--- 1] and in terms of characteristics and parameters is D, = D [y [ {X,,, E,,}, ,(,, A,] ] = H [X, ,,/, , Notice that if we allow more than one individual per household, we must expand the dimensions of vectors of the observable and unobservables variables in H[.] to accommodate this richer description of the household. Then, subject to these adjustments D, would represent the measure of inequality for income distribution among households. 2. Decomposition by determinant factors of income distribution changes We want to decompose changes of inequality D, from period TO to period Ti, Dr1 - D'M, for each of the arguments in the H [.1 function: the price parameters, ,8,, the participation parameters 4, and the population characteristics {x,, rJ. The Price, Participation and Population effects can be defined as: The Price effect BTo T, = H [xTro, Fro } . frir, Am] - D7-o where D7v represents the inequality level in period TO, DTo -=H [{IX,ro, ro }, Ao, The Participation effect LinT) = H [tx,To, rm ), 15r, AT,] - DTO The Population effect PTo T, = H [ {xiTT, ETI ) X, A rv] - DTO Notice that in principle this decomposition is not independent of the point of departure. That is, Birorj BrT,7. LronTl xLro , PVTOrI-PT17o And, therefore, decomposition of inequality changes is not exact DTI - DTO- BTO TI + LTO TI + PTo T, Within the Population effect we can calculate separately the Endowment effect, PXTO Tj, and the earning equation Error term effect, PETo rl. The former defined as PXTOTI = H [{x1T1, o}, /70,ATo] - Dro and the latter as PETOTI = H [{x,in, &TOo[or / cro] 1, fi3T, AT7] - DTO 265 Where ar, , oi refer to the corresponding standard deviations of the earning equation residuals in each period. Obviously, changing only the appropriate subset in the vector of observables (xi, ) or parameters (/J, it) we are able to simulate partial effects of prices, participation or population. For example, the effect of changes in returns to education can be obtained by replacing in the vector of prices fi the subset of parameters corresponding to years of education. Or it might be the case that we pretend to measure the impact of changes in the endowments of education x1,,,, then we must proceed to predict the level of education as a function for other exogenous variables in the endowment vector x.,,, and estimate Xilt = a;X-111 + 9b for t = TO and T1 to obtain the estimates of aqb aT), and the variances of the error terms a6a oakT. Predictions of xi,a for period Ti given characteristics of period TO, x.11m are equal to XE5TO>TI = CfTI X_,IT9 + Oro[CfeTI / dMVYS and simulated changes of inequality due to educational endowment change are given by PXEDUTOTI = H [{xElIv>T,,, xIIO, cTO} 1,0 o , A O] - DTO We also want to measure the effect of changes in household size x2,, on income inequality, thus we proceed in a similar fashion and estimate the number of children as a function of other exogenous variables in the endowment vector X.21, X2,t = 4X-2t + 4b for t = TO and Ti With the estimates of the parameters c6 Aq and the variances of the corresponding error terms o7s, oTr we are able to predict for every household the family size in period Ti given the household characteristics of period TO, XE 2,TO-T = d T1 X-2I7O + DIO [SEkTr/ ° Xr] Therefore, the simulated change in inequality associated to change in family size is PXFSTOT1 = H [{X 2170>TJ,, X-2.7, CM70, /6 . A21 - D 3. Econometric specification. In order to perform the simulations just described we must know the parameters in the earning and participation equation, and the properties of the residuals in the earning equation. Therefore, for every period of the study econometric procedures are required to obtain the estimates of the parameters f,, i, and the variance of the error term in the earning equation C2,. '5 Analogously, we can obtain the predictions of Xl,, for period TO given the characteristics of period TI, X 1,T2 266 In the earnings and participation model some informational constraints should be taken into account. One such constraint is the fact that income is only observed for individuals employed in the labor market, but not for labor inactives. Hence, the expected value of their first and second moments - mean and variance- will deviate from their true values and appropriate estimation procedures should be used to correct for the potential estimation bias. From the several options available to solve this problem, we use the Two Stage Heckman correction procedure in order to obtain consistent estimates of the parameters in the earning equation (please see Technical appendix below for details) and test for the alternative OLS procedure. Four separate earning equations are estimated for self employed and wage earners workers divided by gender. The participation model has three possible states of labor status: 1) self-employed, 2) wage earner and 3) labor inactive. This occupational choice model is estimated by a multinomial logit procedure for three different groups, according to their relative position within the family: 1) household heads, 2) spouses and 3) other members of the household. The estimation of the multinomial-logit model is a maximum likelihood procedure. Remember that utility levels are not observables and thus error terms luik, cannot be estimated. However, error terms can be consistently assigned to each individual according to his observed occupational behavior and his expected probabilities of being in each of the three states. Details on these procedures will be provided in the Technical Appendix. 4. Simulations of Some Specific Decompositions Endowment effect as a residual A proxy of the endowment effect PXTO T can also be obtained as a residual. It is identically true that DTI - DTo-H [{x,TI, rTo[ou / croJ }, I, ATlTJ]-D-m therefore DTI - DTO [H [{XIT,, CTh[ 0T1 / r] 1,1/rI, 2TI] -H [{x,To, TO[7 I /[or] }, |3Ir, A TI] + [ H [{x.T& ,- T[ oTI / To]o I}, 3TI , A TI] - DTO] Notice that, the difference between the first two terms of the right hand side is an approximation to the Endowment effect, that is PXTO Ti H [ {XT1i, c[ [ai / O~o Ar }3+I, ATI] -H [{Tx,m ErOYtar, / O]o }, /3T, ATl] then by substitution and rearranging we find that PXTO Tl = DTI -D7 - [ H [ {xiTo, T I( },/3Tr , i,] Tl - DTO I or PXTOTl DT, - H [I x,TO, 4TO[ 1J /"fO] },/, ,2Th] 267 Hence, the proxy of the endowment effect can be obtained as the difference of the final year distributional measure DT, and the simulated distribution after changing -from TO to Ti- all parameters except the endowment vector, {x,7v }. Occupational Choice Simulations and Unobservable Error Terms The behavioral choice rule is that any individual in period TO will choose occupational state K of {1,2,31 if the corresponding utility level is maximal Ugco > Uk7v for every k -K, k = 1,2,3. In any other period TI the simulated occupational choice of any individual i will be the one that provides the highest simulated utility level UKTo>TJ for his characteristics ZiTO under new behavioral parameters k*T,, k = 1,2,3. That is, the selected state is K = Argmax (Ukr->T, = A*T) Z.TO + A,kTO, k = 1,2,3) Since, residuals AkTo are not available, the solution given by Bourguignon et al (1998) is to draw the residuals from a double exponential distribution conditional on observed choices. A are independent random variables with a cumulative density function equal to F(w) = exp [ exp (w)] It may happen that in period TO the probability of the observed occupational choice for individual i is smaller than in other alternative states, that is, 2KTO ZIO is not the maximum of {f47v ZITO, k = 1,2,31. Therefore the residual for the selected state A,m,c must be large enough to override those differences, in order to provide consistent description of observed behavior. The following rules to draw residuals that satisfy those conditions, for each individual. For the residual of the selected state K Am= - I [-P, . In (9)] notice that the smaller PLKTh the larger AKT in order to make UKI the maximum for all k, and K the preferred state. However an additional condition is necessary for the other states to obtain the consistency between the errors, the estimated parameters and the observed behavior. AkTO = In [-exp(AKM) . PkrO/PKicO - In (a)] or exp (AkTO) = In [-exp(AKM) * PkTO/PA70 - In (9)] where 9 - Uniform (0,1) and the probabilities of each state are P,k, = exp (,A* Z,d / g=1,2,3 exp (,, Z,j), k = 1, 2, 3. Education Endowment and Family Size effect We mentioned above that endowment effects could be simulated for some subset of the endowment vector like education levels of individuals in the working population or the number of 268 members in the household. And in order to predict educational level for each individual we must estimate by OLS x,,, = ,x ,,, + ti, for t = TO and Ti and the exogenous variables included in x ,, are gender, region and a polynomial of fifth order on the age of the individual. Then estimates of the vectors aTO, aT, and the variances of the error terms ao, av,9 become the required inputs to run the simulations of the Education Endowment effect, PXEDUTOTI, defined in Section 2. Analogously we predict the number of household members with 13 years of age or less, using the OLS estimates of X2t = t4X.2,t + 4 for t = TO and Ti where the exogenous variables included in x.2,, are education of both head and spouse, age of the head, region and a third degree polynomial on the spouses age. Estimates of the vectors i, aTI and the variances of the error terms ao, 0rTj allow for the calculation of the family size predictions and to simulate its effect on income distribution, PXFSTOTI, defined in section 2.16 Other Income Change in the size and distribution of non-labor income Yo,, may be significant from one period to the next. Therefore a full characterization of income distribution given in section 2 should include this component of income, i.e.: D, = H [{xlb r, Y0u, }. A, Thus the changes in the distribution due to changes in the level and distribution of non-labor income from period TO to period Ti is given by NLY70>TI = H [I{x,,, £F YO,TO>TI } , Ar .] -DDTo Where Yo,7>TI represents the predicted level of non-labor income according to the changes in the mean and variance from period TO to period Ti. 16 We did not estimate the number of members of the household in working age or older that 65 years old. The simulated family size effect only includes the direct effect on income per capita, but did not incorporate its effect on labor force participation, or the crossed effect from predicted education into family size. 269 3.C Statistical Appendix Table A3.1A Decomposition of Within-Group Income Distribution Changes (1978-1988, 1988- Wage Urban Males, Colombia Contribution to change in inequality 1978 to 19S8 1988 to 1995 GINI E(O) E(1) E(2) GINI E(O) E(1) E(2) Observed Inequality 1978 42.1 29.8 34.2 55.2 Observed Inequality 1988 39.5 26.1 32.0 55.1 39.5 26.1 32.0 55.1 Observed Inequality 1995 45 0 34.2 48.9 151.4 Total change in inequality -2.6 -3.7 -2.1 -0.1 5.5 8.1 16.9 96.2 Average Contribution (%) (%) (%) (%) (%) (%) (%) (%) 11 Retums(**) -4.5 173 -6.1 165 -7.3 340 -14.5 15737 -0.1 -I -0.1 -1 0.4 2 2.7 3 Education -4.2 162 -5.7 156 -7 3 340 -15.4 16796 0.2 4 0.3 4 0.8 4 3.3 3 Experience -0.7 29 -1.1 29 -1.0 44 -1.7 1840 -0.3 -6 -0.5 -6 -0.5 -3 -1.2 -I Regions (**) 0.2 -8 0.3 -9 0.5 -24 1.4 -1508 0.1 1 0.1 1 0.1 1 0.6 1 Interaction term 0.0 0 -0.1 3 -0.2 9 -0.6 619 0.0 0 0.0 0 0.0 0 0.0 0 HI Endowments (*) (**) 4.3 -165 5.4 -147 8.6 -401 20.9 -22795 2.1 38 3.1 38 8.4 50 58.5 61 Education 0.8 -32 1.2 -33 1.4 -64 2.8 -3011 -0.1 -2 -0.1 -2 -0.2 -1 -0.8 -I Rest 3.5 -133 4.2 -114 7.2 -337 18.2 -19784 2.2 40 3 2 40 8.6 51 59.3 62 IV Errors -2.5 95 -3.5 96 -4.5 208 -10.9 11907 3.5 64 5.2 63 8.1 48 35.8 37 V Interaction term 1.1 -41 1.0 -28 0.7 -33 0.0 -54 -2.3 -42 -2.4 -30 -6.8 -40 -98.5 -102 i includes only simulations with parameter esunmates of the 1988 model, for the period 1988-1995 *Imputed as a residual Source Table %, in the Appendix 270 Table A3.1B Decomposition of within-group income distribution changes (1978-1988,1988-1995) Self Emaloved Urban Males, Colombia Contribution to change in inequality 1978 to 1988 1988 to 1995 GINI E(O) E(1) E(2) GINI E(O) E(1) E(2) Observed Inequality 1978 60.8 68.3 83.0 255.1 Observed Inequality 1988 53.5 51.8 56 6 108.9 53.5 51.8 56.6 108.9 Observed Inequality 1995 59.4 64.2 87.2 288.2 Total change in inequality -7.3 -16.5 -26.5 -146.2 5.9 12.4 30.7 179.3 Average Contribution (%) (%) (%) (%) (%) (%) (%) (%) II Returns(**) -1.4 19 -3.7 22 -2.9 11 -0.3 0 -0.4 -7 -1.1 -9 -1.1 -4 -7.1 -4 Education -0.1 2 -0.5 3 0.1 -1 4.3 -3 -0.7 -11 -1.7 -14 -1.5 -5 -5.3 -3 Experience -0.8 10 -2.3 14 -1.8 7 -8.8 6 0.1 2 0.2 2 -0.1 0 -2.1 -1 Regions(**) -0.5 7 -0.6 4 -1.0 4 11.4 -8 0.2 4 05 4 0.5 1 0.1 0 Interaction term 0.1 -1 0.1 0 0.0 0 -3.1 2 0.0 -1 .0.0 0 0.0 0 0.3 0 III Endowments (*) (**) -1.7 23 -2.5 15 -10 4 39 -78.5 54 5.7 95 11.8 96 29.5 96 174.5 97 Education 1.3 -17 3.2 -19 3.8 -15 18.1 -12 0.1 2 0.2 2 0.1 0 -0.6 0 Rest -3.0 40 -5.7 35 -14.2 54 -966 66 5.6 94 11.6 94 29.4 96 175.1 98 IV Errors -3.9 53 -9.7 59 -11.7 44 -51.1 35 0 7 11 1.6 13 2.3 7 11.4 6 V Interaction term 4.2 -57 9 9 -60 18.5 -70 263 3 -180 -3 4 -56 -7.2 -58 -22 1 -72 -357.2 -199 4* includes only simulations with parameter estimates of trhe 1988 model, for the penod 1988-1995 * Imputed Source Table %, in the Appendix 271 Table A3.1C. Decomposition of Within-Group Income Distribution Changes (1978-19S8, 1988-1995) Wage Urban Females, Colombia Contribution to change in inequality 1978 to 1988 1988 to 1995 GINI E(0) E(1) E(2) GINI E(O) E(1) E(2) Observed Inequality 1978 32.7 18.7 19.7 27.2 Observed Inequality 1988 34.3 20.4 23.1 35.4 34.3 20.4 23.1 35.4 Observed Inequality 1995 39.1 25.7 32.0 70.2 Total change in inequality 1.5 1.7 3.4 8.1 4.8 5.3 8.9 34.9 Average Contribution (%) (%) (%) (%) (%) (%) (%) (%) 11 Retums(**) -1.9 -123 -2.0 -115 -2.3 -67 -3.8 -47 3.0 62 3.5 67 4.6 52 10.0 29 Education -2.6 -167 -2.8 -163 -3.5 -102 -6.3 -78 3.0 62 3.5 66 4.7 53 11.0 31 Experience -0.1 -9 -0.1 -8 -0.1 -3 0.0 0 0.1 1 0.1 1 -0.1 -I -1.0 -3 Regions (**) 0 2 12 0.1 7 0.3 8 0.1 2 0.0 -I 0.0 -I 0.0 -1 -0.2 -1 Interaction term 0.0 1 0.0 -2 -0.1 -2 -0.2 -3 0.0 0 0.0 0 0.0 0 0.2 1 MEndowments (*)(**) 3.3 214 3.6 205 5.5 161 11.6 142 0.3 6 -0.3 -5 1.5 17 15.6 45 Education 0.5 32 0.5 31 0.6 18 1.1 13 0.0 -1 0.0 -I 0.0 -I -0.1 0 Rest 2.8 181 3.0 174 4.9 143 10.5 130 0.3 6 -0.2 -5 1.5 17 15.8 45 IV Errors 0.2 10 0.2 11 0.2 7 0.5 6 1.5 32 2.0 39 2.8 32 9.5 27 V Interaction term -0.5 -34 -0.3 -20 -0.7 -21 -2.6 -31 -1.8 -37 -1.2 -23 -2.4 -27 -18.1 -52 * Imputed ** includes only similations with parameter esumates of the 1988 model, for the penod 1988-1995 Source Table %, in the Appendix 272 Table A3.1D Decomposition of within-group income distribution changes (1978-1988, 1988-1995) (Heckman2S) Self Urban Females, Colombia Contribution to change in inequality 1978 to 1988 1988 to 1995 GINI E(O) E(1) E(2) GINI E(O) E(1) E(2) Observed Inequality 1978 54.0 52.9 57.6 109.1 Observed Inequahty 1988 59.0 66.6 72.9 171.6 59.0 66.6 72.9 171.6 Observed Inequality 1995 57.4 62.6 78.1 308.3 Total change in inequality 5.0 13.7 15.2 62.5 -1.6 -4.0 5.2 136.7 Contribution(%) (%) (%) (%) (%) (%) (%) (%) (%) II Returns** 1.0 20 2.1 15 2.4 16 7.9 13 -2.7 164 -6.6 167 -6.9 -131 -29.7 -22 Education 0.1 3 0.3 2 0.3 2 0.2 0 -2.6 160 -6.3 159 -63 -120 -14.5 -11 Experience -0.2 -4 -0.4 -3 -0.9 -6 -3.8 -6 0.2 -15 0.6 -15 0.2 4 -4.9 -4 Regions ** 1.0 20 2.1 15 3.2 21 14.8 24 -0.4 24 ;1.0 26 -1 0 -20 -5.4 -4 Interaction term 0.1 2 0.2 1 0.3 2 1.1 2 0.0 0 -0.1 2 -0.2 -4 -3 8 -3 III Endowments*** 3.0 60 9.1 66 10.3 68 47.2 76 0.8 -50 2.0 -51 11.1 212 157.3 115 Education 0.4 8 0.9 7 0.9 6 2.0 3 0.0 -1 0.1 -2 0.0 0 -0 3 0 Rest 2 6 52 8.2 60 9.4 61 45.2 72 0.8 -49 1.9 -49 11.1 212 157.6 115 IV Errors 1.0 20 2.6 19 2.8 18 9.8 16 0.2 -13 0.6 -14 0.8 15 5.7 4 V Interaction term -2.8 -57 -8.2 -60 -9.9 -65 -88.2 -141 1.0 -58 2.6 -65 -4.0 -75 -328.2 -240 includes only simulations with parameter estimates of the 1988 model, for the penod 1988-1995 * Imputed Source Table %,in the Appendix 273 Table A3.ME Decompgosition of changes in individual income inequality, 1978-4988 and 1988-1995. Urban Colombia. Contribution to change in inequality 1978 to 1988 1988 to 1995 GIN E(O) E(1) E(2) GIMNJ E(O) E(1) E(2) Inequality 1978 47.8 40.6 49.6 136 Inequality 1988 44.7 35.7 41.3 79.8 44.7 35.7 41.3 79.8 Inequality 1995 50.3 44.8 62.9 221.4 Inequality change -3.1 -4.8 -8.3 -55.8 5.5 9.0 21.5 141.6 Contribution (%) (%) (%) (%) (%) (%) (%) (%) (%) I. Participation 0.8 -26 1.5 -31 1.6 -20 3.7 -7 0.6 10 1.2 13 1.4 6 6.7 5 Males 0.6 -21 1.1 -23 1.3 -16 2.6 -5 0.4 7 0.7 8 1.0 5 4.4 3 Females 0.1 -5 0.4 -8 0.3 -4 1.1 -2 0.2 3 0.5 5 0.4 2 2.3 2 II. Returns -3.6 119 -6.0 123 -7.1 86 -20.6 37 0.2 3 -1.1 -12 1.7 8 10.6 7 Education -2.3 75 -3.5 71 -3.8 46 4.2 8 0.0 0 -0.2 -2 -0.3 -1 -3.9 -3 Experience -0.7 22 -1.1 24 -1.2 15 -5.4 10 -0.1 -1 -0.1 -1 -0.2 -1 -1.6 -1 Regions 0.0 1 0.2 -4 -0.2 2 2.7 -5 0.0 1 0.0 0 0.3 1 1.3 1 Constant -0.8 28 -1.8 37 -2.3 28 -14.1 25 0.0 0 -1.0 -11 1.5 7 15.3 11 Interaction 0.2 -7 0.3 5 0.5 -6 1.1 -2 -0.1 -1 -0.1 -1 -0.1 -1 -0.1 0 III. Endowments 2.3 74 3.9 -82 3.4 -41 -11.9 21 2.5 46 5.0 55 13.7 63 105.7 75 ... Education 3.0 -100 4.8 -99 6.1 -74 18.4 -33 1.2 22 1.8 20 2.4 11 6.6 5 Earnings effect 3.3 -108 5.3 -109 6.7 -81 20.2 -36 1.4 25 2.2 24 2.8 13 7.2 5 Participation induced effeGO.3 9 -0.5 10 -0.6 8 -1.9 3 -0.2 -4 -0.4 -4 -0.4 -2 -0.5 0 IV. Error term -2.4 80 -4.1 85 -6.3 77 -29 53 2.5 45 4.3 47 7.3 34 26 19 V. Interaction 0.1 -3 0.1 -3 0.2 -2 1.0 -2 0.3 6 0.6 7 0.4 2 -0.2 0 274 Table A3.1F Decomposition of changes in household income inequality, 1978-1988 and 1988-1995. Urban Colombia. Contribution to change in inequality 1978 to 1988 1988 to 1995 GINI E(O) E(1) E(2) GINI E(O) E(1) E(2) Inequahty 1978 50 2 38 0 52 6 153 6 Inequality 1988 50.2 425 503 105.1 502 425 503 105.1 Inequality 1995 54 4 50 5 70 6 282 7 Inequality change 0 0 4 4 -2.3 -48 5 4 2 8.1 20.3 177 6 Contribution (%) (%) (%) (%) (%) (%) (%) (%) (%) I Participation 07 -1856 23 52 0.0 2 40 -8 -04 -91 00 05 -07 -33 -0.8 -0.5 II Retums -1.9 4870 -30 -68 -38 168 -229 47 06 134 0.6 69 21 105 107 6.0 Education -1 9 4923 -3.4 -77 -3 4 150 -3 4 7 -0.1 -1 2 -0.2 -3.0 -0 I -0 7 -2.8 -1 6 Experience 0 1 -387 0 4 8 0 4 -18 6 0 -12 0 1 3 4 0.4 5.0 0 3 1.3 -0 8 -0.5 Regions 01 -373 04 8 0.0 1 -18.6 38 0.1 2.4 0.2 20 04 2.0 38 21 Constant -0.2 572 -0.4 -9 -0 8 34 -6.6 14 0 4 9 1 0 4 5 4 1.8 8.8 12 2 6.9 Interaction III. Endowments Education 2.3 -5988 4 3 98 4.6 -202 10 8 -22 0 8 19.9 1 7 21 6 1 5 7.4 1 1 0.6 Earningseffect 23 -5968 41 92 4.9 -218 144 -30 09 217 17 216 1.9 9.4 4.9 2.8 Participation induced effect 0 0 -69 0.3 7 -0 3 13 -3 3 7 -0.1 -1 7 0 0 0 2 -0 4 -2 0 -3.9 -2 2 Interaction Regions 00 61 -0.1 -3 -0.1 5 -1 1 2 00 08 01 09 0.1 0.6 08 0.5 Fanily size change -0 6 1571 -1 2 -28 -1 2 52 -6 6 14 -0 4 -8 9 -0 9 -11.2 -0.7 -3 4 -2 1 -1.2 Rest IV. Errorterm -1.5 3850 -2.8 -64 -40 176 -21 6 45 1 1 25 1 2 1 263 30 15 0 17 5 9.9 V. Interaction VI Non labor income -3 0 7667 -6 1 -139 -12 0 529 -107 2 221 0.9 22 3 1.8 22 9 3.1 15.2 19 2 10.8 275 Table A3.2A Decomposition of Changes in Individual Income Inequality, 1978-1988 and 1988-1995. Rural Colombia. Contribution to change in 1978 to 1988 1988 tX 1995 G1NII E(O) E(1D E(2) GINI E(O) E( E(2) Inequality 1978 38.5 28.8 26.5 36.7 Inequality 1988 39.0 31.6 26.6 33.0 39.0 31.6 -130.0 33.0 Inequality 1995 36.6 28.7 23.5 28.0 Inequality change 0.5 2.8 0.1 -3.7 -2.4 -2.9 153.5 -4.9 Contribution (%) (%) (%) (%) (%) (%) (%) (%) I. Participation -1.7 -337 3.7 131 -2.9 -5830 -2.9 78 -0.9 39 0.9 -31 -1.3 42 -1.0 21 II. Returns -3.0 -609 1.7 62 -5.4 -10760 -8.7 233 -1.7 71 0.0 1 -2.0 65 -1.3 26 Education -2.8 -561 2.3 81 -5.0 -9920 -8.0 213 -1.2 50 0.6 -22 -1.4 47 -0.8 16 Age -1.6 -311 3.9 140 -2.8 -5540 -3.1 82 -1.3 55 0.5 -17 -1.7 56 -1.5 31 Regions -2.0 -409 2.9 104 -3.4 -6750 -2.9 78 -1.1 44 0.6 -22 -1.4 45 -1.0 20 IH. Endowments 7.7 1548 -5.0 -179 12.5 25010 13.2 -352 -0.4 -88 3.5 126 0.0 -94 1.1 -30 Education -2.8 -554 4.2 149 -5.1 -10140 -7.1 189 -1.3 53 1.5 -51 -1.6 53 -0.1 3 Earnings -0.5 -106 -0.8 -27 -1.1 -2210 -2.9 78 -0.2 6 -0.4 15 0.0 -I 1.0 -20 Participation induced effect -2.2 -448 4.9 176.0 -4.0 -7930 -4.2 111.0 -1.1 47.0 1.9 -660 -1.7 54.0 -1.1 23.0 IV. Error -2.5 -502 2.4 85 -4.2 -8320 5.3 141 -1.8 78 -0.1 4 -2.7 87 -4.0 82 276 Table A3.2B Decomposition of Changes in Household Income Inequality, 1978-1988 and 1988-1995. RURAL Colombia. Contribution to change in inequality 1978 to 1988 1988 to 195 GINI E(O) E(1) E(2) GINI E(O) E(1) E(2) Inequality 1978 43.5 33 8 34 6 60.3 Inequality 1988 44 4 37 3 35 0 50.5 44 4 37 3 35 0 50 5 Inequahty 1995 40 7 30 0 29 4 45.8 Inequality change I 0 3.5 0.4 -9.8 -3 7 -7.3 -5 6 -4.7 Contribution (%) (%) (%) (%) (%) (W) (%) (%) (%) I Participation 1- 0 -108 -0.3 -8 -1 7 -452 -26 26 -04 12 -0.6 8 -07 12 -0.9 19 11 Returns -II -109 -2.0 -58 -1 9 499 -3 6 37 -0 2 6 -0 6 8 -0 3 5 0.0 1 Educaton -0 9 -88 -1.3 -38 -1 6 -416 -3 5 35 -0 1 4 -0.2 3 -0 1 2 0 0 -1 Age -0 1 -14 -0.3 -10 -0 3 -76 -0 9 9 0 1 -4 0 1 -2 0.2 -4 0 6 -13 Regions -0 1 -6 -0.4 -10 0 0 10 1 1 -12 -0 2 6 -0.4 6 -0.4 7 -0 6 14 IE Endowments 3.6 372 68 196 48 1267 -2.6 26 06 66 42 121 04 115 -84 85 Education -05 -55 03 9 -1.3 -331 -47 48 -04 11 -22 30 -05 9 -1.2 26 Earrmngseffect 00 -3 0.1 3 -0.2 -55 -1 I II -02 6 -0.7 9 -0.5 8 -1.5 33 Participation induced effect -0 5 -52 0 0 2 6 2 -1 1 -275 9 -3 6 37 0 -0 2 5 2 -1 6 21 3 0 0 0 8 0 3 -6 7 Fanuly size change -1 0 -100 -1 7 -48 -0 4 -99 78 1 -795 -0 7 20 -01 8 01 -1 -10 8 231 Rest 5.1 527 8.1 234 6.5 1697 -76.0 - 773 0 3 35 2 9 83 0 4 107 16 9 -173 IV Errorterm -05 -55 -1.0 -30 -08 -217 -10 10 -06 16 -1.0 -37 -10 -32 -2.1 -5 Table A3.3 Mean Income. Impact of change in the constant of the earnings equation (%) Relative income, 1988-95 1978-88 1988 (*) Male Wage 0% 3% 1.00 Male Self 7% -9% 1.20 Female Wage -13% 6% 0.80 Female Self 61% 40% 0.68 (*) Relative to wage earners Source: Authors' simulations and 277 Table A3.4A Modeling the Number of Years of Schooling Urban Rural 1978 1988 1995 1978 1988 1995 Age 3.99021 3.86210 2.72026 1.91356 2.07900 2.06230 Age2 -0. 17786 -0.16236 -0.11020 -0.09320 -0.09290 -0.08992 Age3 0.00365 0.00318 0.00210 0.00199 0.00182 0.00173 Age4 -0.00004 -0.00003 -0.00002 0.00002 -0.00002 -0.00002 Age5 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 Male 0.58770 0.50286 0.29735 0.05858 -0.09250 -0.31938 Barranquilla -1.28280 -1.24493 -0.35636 Bucaramanga -1.46570 - 1.09083 -1.24457 Manizales -0.96875 - 1.12754 -0.87689 Medellin - 1.40963 -1.13693 - 1.05497 Cali -0.63724 -0.73971 -0.83780 Pasto -0.63325 -0.66495 -0.38674 Cartagena -1.55282 -0.62283 -0.76620 AtlAntica -0.55480 -0.81550 -0.52160 Oriental 0.37080 0.36411 0.36420 Central -0.22894 0.16526 0.33530 Constant -25.07898 -25.06618 -15.20269 -10.15570 -10.64050 -10.24930 R-squaredt 0.11950 0.10920 0.10660 0.1336 0.1756 0.1835 No. observations 7527 35030 32774 13084 18781 19992 * Significant at 95% level tFor Rural: Adjusted R-squared Source: Encuestas de Hogares del DANE 279 Table A3.41B Modelling the nu]imlber of children per househoild Urban Rural 1978 1988 1995 1978 1988 1995 School Head -0.04716 -0.03464 -0.02352 -0.06260 -0.0306 -0.0221 School Spouse -0.18464 -0.14001 -0.08446 -0.06320 -0.0478 -0.0389 School2 Spouse 0.00880 0.00616 0.00317 Age Head -0.00752 -0.00733 -0.00857 -0.00111 -0.0031 -0.0021 Age Spouse 0.11060 0.14188 0.12405 0.17190 0.1534 0.1383 Age2 Spouse -0.00132 -0.00364 -0.00338 -0.00459 -0.0046 -0.0042 Age3 Spouse 0.00000 0.00002 0.00002 0.00003 0.00003 0.00003 Barranquilla 0.45701 0.40819 0.39450 Bucaramanga 0.33831 0.04006 0.07555 Manizales 0.14978 0.07072 0.00196 Medellin 0.35287 0.10241 0.05669 Cali 0.05298 0.01649 0.08177 Pasto -0.04702 0.23353 0.20298 Cartagena 0.82505 0.53157 0.28898 Atlantica 0.23211 0.1618 0.2385 Oriental 0.06420 -0.0066 0.0353 Central 0.06307 -0.1102 0.0373 Constant 1.42839 1.26377 1.21682 1.01386 1.000625 0.80718 No. observations 2847 14247 13706 3504 5603 6020 * Significant at 95% level Source: Encuestas de Hogares del DANE 280 TO CHAPTER V 5.A. Derivation of Unit Value of Subsidies for Public Services This section describes in greater detail the procedures that were used to derive the unit values of subsidies for different public services. These unit values provide the basis for the analysis of distributional impact. Childcare The unit costs for the two childcare services HCB and CAIP were inferred from official sources (ICBF, 199X). The corresponding unit costs are C$269,561 for HCB and C$470,663 for CAIP. Table A5.1: Derivation of Unit Value of Subsidies for HCB and CAIP HCB CAIP Percentage of total budget (%) 42% 13% Value of total budget (C$m) 248,000 747 Number of beneficiaries (#) 920,113 158,727 Implicit unit cost (C$) 269,561 470,663 Education For primary and secondary education, unit labour costs are obtained directly from the Colombian literature on 'Unidad de Pago por Capitaci6n'. Perez (1996) reports unit costs based on an extensive empirical analysis of costs per teacher and pupils for teacher from around Colombia. Mora et al. (1999) update this work to take into account the substantial real pay increases of 13.6 percent agreed for teachers between 1996 and 1998. Adjusting these figures to 1997 prices yields the unit labor costs reported in the table. To these must be added a mark-up to take into account expenditure on materials, buildings and so forth. The mark-up was determined by analyzing the structure of expenditures on education. Educational financing in Colombia comes from two sources: the Situado Fiscal (from the departmental government) which in 1997 financed 80 percent of total expenditure all of it related to labor costs, and the Participaci6n de Ingresos Corrientes (from the municipal government) which accounted for the remaining 20 percent and covered a mixture of labor and non-labor costs. By analyzing the structure of expenditure of the Participaci6n de Ingresos Corrientes it was possible to estimate the overall percentage of non-labor costs at 7.5 percent. Table A5.2: Derivation of Unit Value of Subsidies for Primary and Secondary Education Unit Labor Costs Unit Non-Labor Costs Unit Total Costs Primary * Urban 300,225 22,517 322,742 * Rural 351,550 26,366 377,916 Secondary 508,980 38,174 547,154 For tertiary education, data on unit costs for each university is taken from Duarte and Villa (1996). The same authors also provide information on the revenue structure of each public 281 university identifying what proportion of the corresponding budget is covered by grants from central and departmental governments. This proportion is applied to the total costs of education of each university, to estimate the total value of the pubic subsidy. An average value of the subsidy per student is then taken for the entire sector. Updated to 1997 prices this becomes C$3,856,400. Table A5.3: Derivation of Unit Value of Subsidies for Tertiary Education University Cost per Number of Percentage covered by Total value of public student students public subsidy subsidy (C$m) (C$m) Nacional 6.2 25,552 78% 122,772.2 Cauca 4.6 5,101 78% 18,183.0 Pedagogica Nacional 4.1 4,230 80% 13,908.2 Caldas 0.5 3,653 83% 1,455.4 Pedagogica y Tecnica 3.5 7,641 89% 23,527.0 Cordoba 8.1 2,580 93% 19,435.1 Tec. del Choco 4.6 1,599 87% 6,413.1 Tec. de los Llanos Orientales 4.0 1,353 76% 4,061.7 Tec. de Pereira 4.6 3,530 95% 15,493.2 Popular del Cesar 1.2 2,910 90% 3,195.2 Surcolombiana de Neiva 3.0 3,226 77% 7,526.6 Amazonia 3.5 1,049 83% 3,082.2 Antioquia 3.5 18,108 74% 47,435.7 Atlantico * 2.2 9,570 95% 20,365.0 Cartagena 3.3 4,638 78% 12,082.9 LaGuajira 2.2 1,181 69% 1,768.3 Magdalena 3.2 2,093 77% 5,205.5 Narifio 2.6 4,692 81% 9,767.3 Fco del P. Santander 1.3 5,756 56% 4,093.7 Pamplona 1.8 3,305 70% 4,071.7 Quindlo 1.8 7,745 64% 8,823.1 Industrial de Santander 1.7 9,726 92% 14,764.0 Sucre 3.0 854 69% 1,750.1 Tolima 2.6 4,573 77% 9,190.4 Valle 4.6 14,255 46% 30,425.8 Distrital Fco Jose de Caldas 3.2 7,128 78% 17,513.5 Cundinamarca 2.5 2,252 61% 3,406.8 Health The most recent available source of information on the unit costs of health care services comes from the Instituto de Seguridad Social (ISS) (1993). The information was updated using the health care component of the Colombian Consumer Price Index, which registered cumulative inflation of 231 percent between 1993 and 1997. The results are reported in the table. In the absence of any specific information on the Empresas Solidarias de Salud (ESS), it was assumed that the same unit costs are applicable in both cases. Table A5.4: Derivation of Unit Value of Subsidies for Health Care Services (C$) Consultation Dentistry Matemity Surgery Hospitalization ISS (1993) 7,521 12,377 138,695 333,244 273,255 Updated 17,143 28,591 320,385 769,794 631,219 282 Training The unit costs of vocational training under the government financed SENA program were derived directly from the published budget. For 1997 the total budget was C$398,225m and it was estimated that approximately 70 percent of this expenditure went on training (SENA, 2000). Since 1,000,609 persons were reported to have been trained in 1997, this yields a unit cost of C$229,992. Family subsidy In the case of the family subsidy ('subsidio familiar'), the survey respondents were asked to report the value of the subsidy directly taking into account both cash and in-kind benefits. Hence no external information was needed to derive unit costs. The average value of the subsidy to a recipient household was found to be C$197,156. Housing subsidy The questionnaire contained information on the total value of the subsidy received for the construction or purchase of housing. Since this subsidy takes the form of a grant for an item of capital expenditure, it was necessary to find some way of converting the subsidy into an annual flow for the purposes of comparison with other subsidies. This was done by calculating the proportion of the total purchase cost of the house that had been financed by the subsidy, and applying this percentage to the current rental value of the home (as estimated by the respondent). The average value of the subsidy to a recipient household was found to be C$1,183,764. Utilities The Departamento Nacional de Planeaci6n supplied information about water and sewerage subsidies in urban areas, but was not able to supply similar information about electricity subsidies. The information provided was the subsidy (or surcharge) per cubic meter by 'estrato' in each water company area together with the corresponding average monthly consumption. The subsidy per household was calculated by applying the unit subsidy to the smaller of the average monthly consumption or the subsistence consumption threshold of 20 cubic meters. Taking the average subsidy over water companies and over 'estratos' 1-3 yields an average value of C$102,261 for the water service and C$64,704 for the sewerage service. 283 s.B. Statistical appendix Table B.5.1: Year of decentralizatioun of health and education services by department Departamento Education Health Amazonas 1996 1996 Antioquia 1995 1994 Arauca 1997 Atlantico 1995 1994 Banranquilla 1995 1994 Bogota 1995 1993 Bolivar 1995 1998 Boyaca 1995 1996 Caldas 1996 1998 Caqueta 1995 Cartagena 1995 1994 Casanare 1997 Cauca 1996 Cesar 1997 1996 Choco 1997 1994 Cordoba 1997 Cundinamarca 1997 Guajira 1997 1994 Huila 1997 1994 Magdalena 1997 1996 Meta 1996 1994 Narino 1997 Norte de Santander 1996 Putumayo 1996 Quindio 1995 1998 Risaralda 1995 1994 San Andres 1997 1995 Santander 1997 1996 Sucre 1996 1994 Tolima 1996 1996 Valle 1995 1991 284 Table AS.2: Reconciliation of ECV 97 data with official statistics No. of Beneficiaries Overall Budget (C$m) Official data Estimate ECV97 Official data Estimate ECV97 Childcare (ICBF) * CAIP 163,342 249,736 74,707 69,000 * HCB 1,111,959 471,685 248,027 229,080 * Total 1,275,301 721,421 313,014 298,080 Education * Primary 3,593,994 4,007,823 1,447,288 Na. * Secondary 1,998,115 2,770,839 970,187 Na. * Tertiary 158,300 388,187 1,060,356 Na. * Total 5,750,409 7,166,849 3,477,831 Na Training (SENA) 1,000,609 509,011 222,973 195,000 Family subsidy 1,397,527 21,000 Health * EPS 12,536,405 15,289,645 3,020,676 * ESS 5,129,596 2,890,633 16,839 * Total 17,666,001 18,180,278 3,037,515 Utilities * Electricity _ 30,508,966# _ * Water _ 23,646,556_ * Sewerage 20,837,124# Only 1996 data was available. -ECV97 only covers the issue of housing subsidies for a sub-sample of the population. #fNumber of people living in households in 'I-3' with service connections. 285 Table A5.3: Cost implications of meeting alternaadve coverage targets Overall Education Childcare Healthcare Utilities (excl. Primary Secondary Tertiary Insurance Treatment Water Sewerage Electricity tertiary) Shortfall (m) o Decile4 0.16 0.05 0.00 0 12 1.20 0.12 0.10 0.25 0.02 o Decile 6 0.20 0.08 0.02 0.13 1.85 0.14 0.24 0.53 0.18 o Decile 8 0.44 0.39 0.09 0.38 4.66 0.35 0.61 1.29 0.29 o Min=Max 0.44 0.18 0.57 0 91 8.96 0.36 1.20 2.12 0.54 o Universal 0.90 1.20 1.06 3.23 16.34 1.60 1.36 2.78 0.58 Cost (C$m) o Decile 4 133,250 54,477 28,319 0 27,024 23,431 - - - - o Decile 6 180,276 70,475 44,623 71,726 29,182 35,997 - - - - o Decile 8 545,385 154,699 214,748 352,248 85,134 90,805 - - - - o Min=Max 629,705 154,914 96,840 2,215,200 203,277 174,674 - - - - o Universal 2,011,709 315,726 656,264 4,075,740 721,062 318,656 - - - - Cost % PSE o Decile 4 2.42 0.99 0.51 0.00 0.49 0.43 - - - - o Decile 6 3.28 1.28 0.81 1.30 0.53 0.65 - - - - o Decile 8 9.92 2.81 3.90 6.40 1.55 1.65 - - - o Min=Max 11.45 2.82 1.76 40.28 3.70 3.18 - - - o Universal 36.58 5.74 11.93 74.10 13.11 5.79 - - - - Final coverage o Decile 4 87 75 13 27 55 73 81 64 91 o Decile 6 88 76 27 28 59 74 85 70 95 o Decile 8 92 84 38 35 69 80 91 83 97 o Min=Max 95 84 74 48 81 80 98 93 99 o Universal 100 100 100 100 100 100 100 100 100 286 Table A5.4: Results of regressions to explain differences in access and equity to health insurance across departments Coverage rate 1997 Concentration coefficient 1997 Coeff. Std.err. T-stat. Coeff. Std.err. T-stat. Constant 0.437 0.153 2.855 -0.179 0.164 -1.090 Coverage rate 1993 0.629 0.351 1.790 0.922 0.377 2.447 Concentration coefficient 1993 0.163 0.388 0.420 -0.037 0.417 -0 088 Income pc 1993 0.019 0.026 0.737 0.022 0.027 0.826 Income Gini 1993 -0.012 0.205 -0.058 -0.190 0.220 -0.865 Health expenditure pc 1996 -0.148 0.118 -0.125 2.205 1.267 1.741 Years of decentralization 0.046 0.025 -1.870 0.007 0.026 0 259 No. of observations 28 28 Adjusted R squared 0 152 0.286 F statistic 1.839 2.873 Table A5.5: Results of regressions to explain differences in access and equity to health treatment across departments Coverage rate 1997 Concentration coefficient 1997 Coeff. Std.err. T-stat. Coeff Std.err. T-stat. Constant 0.679 0.212 3.206 -0 656 0.322 -2.034 Treatment rate 1993 0.299 0.255 1.172 0.834 0.389 2.145 Concentration coefficient 1993 -0.120 0.142 -0.845 0.212 0.216 0.984 Income pc 1993 0.042 0.019 2.190 0.022 0 029 0.750 Income Gini 1993 -0.263 0.156 -1.684 -0.083 0.237 -0.347 Health expenditure pc 1996 0.028 0.090 0.310 0.070 0.137 0.510 Years of decentralization -0.018 0.018 -1.014 0.022 0.027 0.797 No. of observations 28 28 Adjusted R squared 0.196 0.304 F statistic 1.540 3.039 287 Table A5.6: Results of regressions to explain changes in education coverage rates across departments Completing primary school Entering secondary school Completing secondary school Entenng tertiary education (6-12 year olds) (13-19 year olds) (13-19 year olds) (20-25 year olds) Coeff. Std.err. T-stat. Coeff. Std.err. T-stat. Coeff. Std.err. I T-stat. Coeff. Std.err. T-stat. Coverage 1993 0.499 0.446 1.118 0.299 0.254 1.174 1.015 0.718 1.414 0.411 0.410 1.002 Concentration coeff 1993 -0.009 0.111 -0.085 0.035 0.200 0.175 -0.154 0.088 -1.755 0.035 0.141 0.247 Income pc 1993 0.011 0.012 0.943 0.054 0.023 2.338 0.020 0.011 1.735 0.021 0.017 1.232 Income Gini 1993 0.038 0.088 0.433 -0.289 0.172 -1.676 -0.010 0.083 -0.116 -0.001 0.122 -0.011 Expenditure pc 1996 0.325 0.256 1.267 0.815 0.471 1.731 0.308 0.248 1.243 0.521 0.360 1.446 Years of decentralization 0.012 0.018 0.645 0.024 0.032 0.749 0.040 0.019 2.055 0.020 0.027 0.761 Constant 0.103 0.073 1.414 0.483 0.191 2.531 0.030 0.049 0.614 0.004 0.077 0.047 No. of observations 29 29 29 28 F statistic 1.35 2.62 1.98 1.42 Adjusted R squared 0.07 0.28 0.17 0.09 Table A5.7: Results of regressions to explain changes in education concentration coefficients across deparbtents Completing primary school Entering secondary school Completing secondary school Entering tertiary education (6-12 year olds) (13-19 year olds) (13-19 year olds) (20-25 year olds) Coeff. Std.err. T-stat. Coeff. Std.err. T-stat. Coeff. Std err. T-stat. Coeff. Std.err. T-stat. Coverage 1993 0.016 1.296 0.012 0.592 0.303 1.953 -0.988 2.257 -0.438 -0.276 1.278 -0.216 Concentration coeff 1993 0.736 0.299 2.464 0.137 0.239 0.572 0.396 0.276 1.435 0.245 0.448 0.547 Income pc 1993 0.052 0.032 1.636 0.019 0.028 0.691 0.002 0.035 0.070 0.050 0.050 1.014 Income Gini 1993 -0.558 0.236 -2.368 -0.104 0.205 -0.505 0.020 0.266 0.074 0.126 0.376 0.335 Expenditure pc 1996 0.967 0.689 1.403 0.645 0.561 1.148 -0.004 0.766 -0.005 1.240 1.060 1.170 Years of decentralization -0.017 0.049 -0.351 -0.022 0.038 -0.566 -0.015 0.060 -0.251 0.074 0.084 0.887 Constant 0.062 0.196 0.318 -0.453 0.227 -1.990 0.050 0.150 0.335 -0.092 0.227 -0.407 No. of observations 29 29 28 26 F statistic 2.75 2.20 0.48 1.04 Adjusted R squared 0.27 0.20 -0.13 [???] 0.01 288 Report No.: 24524 CO Type: SR