Report No. 20707-LV The Republic of Latvia Poverty Assessment Volume 2 June 14, 2000 Poverty Reduction and Economic Management Unit (ECSPE) Eastern Europe and Central Asia Region Document of the World Bank LATVIA-FISCAL YEAR January I-December 31 CURRENCY EQUIVALENTS (as of 20 April 1998) Currency Unit = Lat 1 Lat = US$1.7 WEIGHTS AND MEASURES Metric System ABREVIATIONS AND ACRONYMS CDF Cumulative Distribution Functions CPI Consumer Price Index CSB Central Statistical Bureau FSU Former Soviet Union HBS Household BudLget Survey HEIDE Household Expenditure and Income Data for Transition Economies HH Household ILO International Labor Organization LFS Labor Force Survey IMF International Monetary Fund LG Local Government LVL Lats MCB Minimum Crisis Basket ML Maximum Likelihood OLS Ordinary Least Squares PIT Personal Income Tax RHS Right-hand Side SA Social Assistance SAPC Social Assistance Per Capita UNDP United Nations Development Program US United States UEB Unemployment Benefit Vice President: Johannes Linn, ECAVP PREM Director: Pradeep Mitra, ECSPE Chief Economist: Marcelo Selowsky, ECAVP Country I)irector: Basil Kavalsky, ECCO9 Team Leader: Branco Milanovic, DECRG Latvia Poverty Assessment Volume 2 TABLE OF CONTENTS Chapter 1. Poverty in the context of macro developments .............................................3.....................3 Chapter 2. Poverty in Republic of Latvia in 1997/98 ......................................................... 6 2.1 Introduction ....................................................................6 2.2 Measuring living standards: Levels and distribution ....................................................................7 2.3 Measuring poverty ................................................................... 9 2.4 Poverty profile: Cross-Tabulations . ......................... .................................. 12 2.5 Poverty profile: Regression Analysis ................................................................... 20 2.6 Conclusions ................................................................... 24 Chapter 3. An Analysis of the Labor Market in the Republic of Latvia ........................ ........................ 25 3.1 Introduction .................2.......................... . ...................... 25 3.2 Literature review and methodology ................................................................... 26 3.3 Earnings structure ................................................................... 32 3.4 Labor force participation ................................................................... 36 3.5 Labor demand ................................................................... 38 3.6 Unemployment .................................................................... 41 3.7 Conclusions ................................................................... 55 Chapter 4. Social Transfers and Social Assistance: An Empirical Analysis Using Latvian Household Surveys ................................................................... 59 4.1 Incidence of social transfers ................................................................... 59 4.2 An empirical analysis of Latvia's social assistance system ............................................. ............. 66 4.3 Performance of Latvia's social assistance: comparison with other transition countries .......... ..... 71 4.4 Why some poor households do not receive social assistance? ...................................................... 76 4.5 Regional inequality in distribution of social assistance ................................................................ 81 4.6 Policy recommendations ................................................................... 86 Annex 1 The description of the Latvia Household Budget Survey ................................................................... 90 Annex 2 "Mapping" of Latvia' s local governments into the classification provided by the household budget survey .................................................................... 98 List of Tables in Appendices Appendix A: Composition of Poverty Table Al: Composition of poverty ................................................................... 100 Table A2: Composition of poverty ................................................................... 101 Table A3: Composition of poverty ............................................................ 102 Table A4: Composition of poverty ............................................................ 103 Appendix B: Additional tables Table B 1: Household consumption budget shares (Percent) ................................................................ 104 Table B2: Poverty rates by gender and age (Percent) ...................................................... ........... 104 Table B3: Poverty rates by gender and age of household head (Percent) .................... ........................ 105 Table B4: Poverty rates by gender and education level (Percent) ........................................................ 105 Table B5: Poverty rates by gender and socio-economic group (Percent) .................... ........................ 106 Table B6: Poverty rates by gender and labor force status (Percent) ............................. ....................... 106 Table B7: Selected housing characteristics by household poverty status .................... ........................ 107 Table B8: Consumption regressions, by locality ................................................................. 108 Table B9: Consumption regressions, main breadwinner is head . ........................................................ 109 Table B 10 Poverty regressions (probit) ................................................................. 110 Appendix C: Sensitivity Analysis Table Cl: Sensitivity analysis ................................................................. 111 Appendix D: Functional form for regressions: linear versus semi-log .......................... ......................... 112 Appendix E: Labor Force Surveys Latvia labor force survey (May 1998) LO Standard unemployment definition (UE 2) . ..................... 113 Latvia labor force survey (May 1998) Registered unemployment definition (UE5) . ........................... 114 Latvia labor force survey (MaLy 1996) ILO Standard unemployment definition (UE 2) . ..................... 115 Latvia labor force survey (May 1996) Registered unemployment definition (UE4) . ........................... 116 References ................................................................. 117 1. POVERTY IN THE CONTEXT OF MACRO DEVELOPMENTS 1.1. Latvia has experienced profound challenges and opportunities during its transition from a planned to a market economy. When Latvia regained independence in 1991, it was left with an inefficient economic structure dependent on shrinking FSU markets and unable to compete in world markets. Latvia launched a market based reform program that has so far reached substantial results. Price liberalization, voucher privatization, trade liberalization, legal reform, institutional development, and social safety net improvements either have been implemented or are under implementation. A vigorous stabilization program supported by the Bank and the IMF has underpinned policies in the real sector, which quickly and substantially reduced macroeconomic imbalances and brought inflation under control. 1.2 In 1995 a wave of second generation transition issues common to a number of transition economies disrupted the early stages of Latvia's economic recovery. Competitive pressures on the profitability of enterprises, together with lax banking supervision and regulation, produced a serious banking crisis and resurfacing of fiscal imbalances. Supported by a Structural Adjustment Loan from the World Bank, the authorities embarked on a broad ranging reform program. The program sought to steer the Latvian economy toward a rapid and robust growth path, as a means to permanently reduce poverty. 1.3. The most recent Latvia CAS (177706-LV) was discussed by the Board in April 1998. The proposed program of lending and non-lending services in the CAS reflects the need to complete the structural reform program by putting the main emphasis on redefining and reshaping the public sector in Latvia. The CAS identified five areas for priority action by the Government. The first was strengthening the macro-economy andfinancial sector. Measures in this area included the strengthening of bank and non-bank financial supervision and reforms to ensure the medium-term sustainability of fiscal expenditures. The second was accelerating development of the private sector with emphasis on pushing ahead with the privatization of remaining large enterprises, along with the development of the regulatory framework needed for infrastructure privatization. Reform of the 'overburdened and poorly functioning judicial system' and the need for better corporate governance were noted as particular issues in this regard. The third priority was reshaping the role of the state with emphasis on pay reform and dealing with the growing problem of extra-budgetary funding of public sector agencies. A fourth area identified was promoting sub-national Government capacity to cope with environmentally sustainable regional development and local Government reform. The CAS noted the problems of the fragmentation of sub-national entities and their lack of administrative capacity and financing. The final area was strengthening the financing and delivery of social services. Among the issues were the affordability of the pension system and the need for ensuring that social assistance reached the neediest groups given the problems at the sub-national Government-level referred to above. 3 1.4. As part of the Bank's assistance to Latvia to define and implement an effective poverty reduction strategy, the Bank has completed in FY99 a social assessment of poverty in Latvia in participation with the UNDIP and the Latvian Institute of Philosophy and Sociology, called 'Listening to the Poor.' This study has obtained valuable qualitative information on living conditions and gender, and its implication for poverty. It was essential to follow-up the social assessment with a quantitative oriented poverty assessment to provide the Government with quantitative information through which it can address its priorities in the areas of poverty reduction. In addition, the Government of Latvia is working together with the UNDP, ILO and the World Bank to come to a strategy that will alleviate poverty in Latvia. Separate reports have been written on various poverty related topics and this poverty assessment feeds directly into this ongoing discussion of the interdepartmental working group of the Government of Latvia. 1.5 Latvia has encountered severe contraction of GDP per capita in the early years of transition but has since 1994 shown a strong recovery (see figure 1.1). As Latvia's overarching goal is to enter the European Union, it will need to continue not only with a strong macro economic perfonnance as to stay broadly within the Maastricht criteria, but also needs to post successive years of high GDP growth to bridge the income gap with the European Union. In 1997 Latvia's GDP per capita in purchasing parity terms was 27 percent of the EU average. To fully bridge the income gap with the EU, Latvia's economy will need to grow at about 4 percent per capita per annum for a consecutive period of 32 years. Obviously not an easy task to accomplish. 1.6 To be able to return and sustain this much needed rapid growth Latvia will need to increase not only investment levels in physical and human capital, but also improve the functioning of its govemment institutions that interact with the market place.' Robust growth is needed to alleviate poverty and to close the income gap with the EU. Sustained high level growth will require, among others, a better functionning labor market (see Chapter 3), a decrease of the level of bureaucracy, a need to redirect expenditures away from unproductive sectors, e.g. untargeted social safety net expenditures (see Chapter 4), a transparent regulatory framework and a continued emphasis on fiscal and monetary policies that promote a stable macro-economic environment (see Havrylyshyn, Izvorski and van Rooden, 1998). 1.7. Latvia's underlying characteristics predict that the economy should be able to grow at approximately 4 percent in the medium to long term, with short-term growth rates determined by recent macro economic developments (see Fischer, Sahay, and Vegh, 1998).. This level of growth would enable Latvia to improve the poverty situation substantiallly over the next decade. Ten years of robust economic growth would reduce the poverty headcount from 19.4 percent today to 5.5 percent." Although, Latvia's underlying economic fundamentals are l For a few years now, the Government of Latvia has been working with the World Bank on a comprehensive public sector reform agenda. This agenda will be supported by a programmatic structural adjustment loan (PSAL) operation that the Government has requested and will be presented to the Board in March 2000. 2Note that this simple calculation does not correct for changes in household composition and even more importantly it keeps the income distribution constant. Widening income inequality, if it persists, will reduce the anti-poverty impact of a given growth rate. 4 strong, it will be difficult to reach these growth rates till the Russian economy fully recovers. Therefore, it is important to have a poverty reduction strategy that actively pursues to increase the welfare of the poor and that creates opportunities for the current poor to lift themselves out of its current economic and social situation. Figure 1.1: Developments in GDP per Capita in Purchasing Power Parity and its underlying causes (1990-1998). A contraction in GDP during the early face of transition, countered by a declining population... Real GDP Developments Population Developments 150. 120 1990 1991 1992 1993 1994 1995 1996 1997 1998 5.0 11 0 r ;’fr.w> r S 00 - 0 °° a . 3- XXJX 102 i .Annuei % gw -Index (1990 =100,0) |Annu % che -Index (1990=100,0) ...together with an adjusthng inflathon and PPP exchange rate resulted initially in a sharp reduction in GDP per capita in PPP terms, but has turned around since 1994. PPP Conweralon factor GDP per capIta, PPP (cond nut International S) 1000 90 150 9400 ~~~~~~~~~~~~~~~~~~~~so 2 0.0 934S W 4450 WX 9 0'_4 <.50~~~~~~~~~~~~~~~~~7 ss, 1990 1991 1992 1993 1994 1995 1996 19719 1990 1991 1992 1993 1994 1995 1996 1997 199B |Annual% *e +4-Index (1990= 100,0) l |Anual% chang +Index (1990 .100,0) 5 2. POVERTY IN REPUBLIC OF LATVIA IN 1997/98 2.1 INTRODUCTION 2.1 In this chapter, an analysis of poverty in the Republic of Latvia using data from the 1997 and 1998 Latvia Household Budget Surveys (HBS) is presented. The HBS is a continuous multi- functional household survey prepared and implemented by the Division of Living Conditions and Household Budget Statistics of the Social Statistics Department of the Central Statistical Bureau of Latvia (CSB).' The data frorn the HBS were collected in the last quarter of 1997 and first three quarters of 1998, and the sample used in the analysis contains 7,690 households and 21,693 individuals. Household income and expenditure variables were adjusted using the CPI to the prices of October 1997. A detailed poverty profile for Latvia was recently constructed by Gassmann (1998) using the 1996 HBS data. The aim of the present chapter is to update the poverty profile of Gassmann (1998) and to see if the conclusions from the 1996 data are robust to differences in methodology with regards to the construction of the household welfare measure and the potential presence of economies of scale in consumption. Another aim of the present chapter is to focus on aspects of the profile that may have changed in recent years and also the aspects of poverty which are of most concern in the policy debate (for example, regional aspects of poverty and the poverty experience of young families). 2.2 The overall objective of poverty analysis is to identify which households and individuals are most at risk of poverty and to provide input into the design of policy measures that help to reduce poverty and improve living conditions for the poor. The first step of poverty analysis is defining a measure of living standards that can be used to rank households from least well-off to most well-off in a consistent fashion. The measure of household welfare used in this chapter is expenditure, although non-monetary indicators of welfare such as health status and access to basic services are also often used to rank households. The second step in poverty analysis is determining a rule that is used to distinguish the poor from the non-poor; this involves the setting of the poverty line against which household consumption is compared. The third step involves using the living standards indicator and poverty line to construct the poverty profile, which identifies the salient characteristics of the poor and thus provides information on the causes of poverty. 2.3 The structure of the chapter is as follows. In Section 2.2, there is a discussion of the measure of living standards used in the analysis (total household consumption) and a brief description of the levels and distrilbution of this measure. The poverty line used in the analysis is described in Section 2.3, as well as the index numbers of poverty (describing incidence, depth l For further details regarding the survey see Annex 1 which provides the official description of the survey kindly supplied by the Central Statistical Bureau of Latvia. 6 and severity) which are used to describe the poverty situation in Latvia. In Section 2.4, the poverty profile (which shows how the poverty rate differs across characteristics of individuals and households) is presented. A regression-based poverty profile is presented in Section 2.5. Section 2.6 concludes the chapter. 2.2 MEASURING LIVING STANDARDS: LEVELS AND DISTRIBUTION 2.4 It is common practice in poverty analysis to use household consumption as the measure of living standards. In former Soviet Union (FSU) countries, consumption is preferred to money income as a measure of household living standards because there is a tendency for income to be under-reported (this is particularly true of high-income earners) and also because many households rely on food grown in garden plots. The expenditure data were constructed so as to be comparable to data in the HEIDE (Household Expenditure and Income Data for Transition Economies) data set described in Braithwaite, Grootaert and Milanovic (1998). Full details of their construction are given in a data appendix. Briefly, household consumption consists of 'regular' cash and in-kind expenditures (such as food, clothing, rent and utilities, health care, transportation, entertainment and education), plus the self-reported value of consumption of home production2, plus cash and in-kind gifts given and received, plus goods and services received free from either a place of study or employment or through humanitarian or social assistance. 2.5 The household consumption measure differs in several ways to that used by Gassmann (1998). In particular, expenditures on consumer durables are included in consumption. Another difference is that while Gassmann used regression analysis to impute the value of utility expenditures for households not reporting this information, such imputation was not carried out for the present chapter. Finally, while Gassmann adjusted household consumption for regional differences in prices, unadjusted consumption is used in the present study. While these differences in the construction of household welfare are likely to make comparisons of poverty levels between 1996 and 1997/98 difficult, it is to be expected that the profiles of poverty will remain reasonably comparable between the two studies.3 2.6 The main measure of living standards used in the present study is monthly per capita household consumption (total household consumption, divided by household size). A problem with using per capita measures of living standards is that they do not reflect the demographic composition of the household (in general, there is reason to expect that young children and the elderly may have lower living costs than non-elderly adults) or the potential presence of economies of size (the presence of certain types of fixed expenditures, for example heating and rent, may imply that an additional household member does not cost as much as preceding members). However, in this chapter, sensitivity analysis is used to determine the extent to which 2 Note that the value of home consumption is estimated by the household itself. 3Lanjouw and Lanjouw (1997) provide evidence for Ecuador that while poverty levels can vary markedly with different definitions of consumption, poverty profiles (which show the relationship between poverty and household characteristics) tend to be much less sensitive. 7 economies of scale4 may influence the profile of poverty. The focus of most of the analysis is on the individual as opposed to the hiousehold. In the absence of information on the intra-household allocation of consumption, the s,ame per capita household consumption is attributed to each household member. All results presented in this chapter are weighted so as to be representative of the Latvian population. 2.7 Table 2.1 shows the distribution of monthly per capita household consumption expressed in October 1997 LVL. Average household consumption was around 41 percent higher in Riga compared with rural areas. For Latvia as a whole, average household consumption in the top 20 percent of the distribution of welfare was 5.7 times higher than average spending in the bottom quintile, and accounted for 47.7 percent of total consumption (while the lowest quintile accounted for 5 percent of total consumption). Table 2.1: Distribulion of monthly per capita household consumption Riga Other Urban Rural Latvia Mean Mean Mean Mean (LVL) % of total (LVL) % of total (LVL) % of total (LVL) % of total Lowest quintile 26.0 5.5 20.9 5.5 17.5 5.0 20.6 5.0 2nd 42.1 10.9 34.4 9.9 28.6 9.3 34.4 9.6 3rd 55.6 15.4 46.1 15.6 38.5 14.1 46.6 15.3 4th 73.7 21.8 61.2 22.2 52.3 23.3 62.7 22.3 Highest quintile 139.6 46.4 113.0 46.8 94.0 48.2 117.3 47.7 All 72.7 100.0 60.0 100.0 51.6 100.0 61.9 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Quintiles are defined in terms of persons rather than households. 2.8 Another measure of the equality of the distribution of consumption is the Gini coefficient, which is bounded between 0 (inequitable distribution) and 1 (equitable distribution). In 1997/98, the Gini (calculated over individuals) for per capita consumption was 0.34, which indicates that there has been only a marginal increase in inequality in Latvia since 1995.5 This apparent tapering off in the increase in inequality has been observed in some other transition economies such as Hungary and Poland. While consumption is equitably distributed in Latvia relative to what has been found for other transition economies, inequality is significantly higher than it was in 1989. Persistent inequality will have implications for the poverty reduction strategy for Latvia, since with greater inequality, a given increase in income leads to a smaller reduction in poverty. There was no significant difference in inequality across different localities in Latvia, with the Gini for each locality being equal to 0.33. However, as shown below, poverty rates vary significantly by locality. 4 Following Lanjouw, Milanovic and Paternostro (1998), economies of scale refers to the net effect on welfare of economies of size and equivalence scales (demographic composition). 5 The HBS was first conducted in 1995; prior to that year the Family Budget Surveys, which tended to underestimate inequality, were used in policy analysis in Latvia. 8 2.3 MEASURING POVERTY 2.9 Choice of the poverty line. There are two main approaches to constructing a poverty line. An absolute poverty line is constructed under the assumption that it is possible to define minimum standard of living based on physiological needs for food, water, clothing and shelter. In contrast, a relative poverty line is set so as to reflect a generally acceptable standard of living, which is specific to the country and the time of the study. Relative poverty lines are often set at a particular proportion of mean per capita consumption. Hence, they reflect the norms of the particular society being studied both through the decision as to what proportion of mean per capita consumption is considered appropriate for setting the poverty line and also the contents of the basket of goods and services used to measure consumption. 2.10 There is no official poverty line in Latvia. For the present poverty study, it was decided to choose a poverty line which is related to the social welfare policy in the country. When per capita consumption was used as the measure of household welfare, the poverty line was set at 28 LVL per person per month, which is 1/2 of the official crisis minimum basket (MCB), valued at October 1997 prices. This poverty line is also close to the 75 percent of the food component of the MCB (which would be 25 LVL); a line previously suggested by the World Bank. The poverty line of 28 LVL is also equal to 2/3 of the average pension in Latvia and is approximately equal to the minimum pension. Gassmann (1998) used three poverty lines, set at 24 LVL (the relative poverty line - 50 percent of 1996 mean per capita expenditures); 38 LVL (the official minimum wage of 1996); and 52 LVL (the 1996 crisis subsistence minimum as calculated by the Latvian Ministry of Welfare). It should be emphasized that if the poverty lines were set at different levels, this would obviously affect the conclusions about the level of poverty in Latvia (see Appendix C for a sensitivity analysis with regards to the setting of the poverty line). However, it is expected that the qualitative conclusions from the poverty profile would be reasonably robust to small variations in the level of poverty line. 2.11 As previously mentioned, much of the poverty analysis in this chapter is conducted using three measures of equivalent consumption, in order to assess the potential impact of economies of scale on the profile of poverty. The three measures are calculated using different values for 0 in the formula for the welfare measure: household welfare = total household consumptionl(household size)0. The thetas are arbitrarily set at levels which indicate different 'degrees' of economies of scale: 0=1 (representing zero economies of scale, and the welfare measure is per capita consumption); 0=0.8 (moderate economies of scale); and 0=0.6 (significant economies of scale).6 As mentioned above, the poverty line used with per capita consumption is 28 LVL per person per month. The poverty lines used for the other two measures of household welfare were set so as to hold constant the ratio between the per capita consumption level of the median person and the poverty line of 28 LVL. In particular, per capita consumption for the median person is 46.27 LVL per month, which gives a ratio of median consumption to the poverty line of 46.27/28=1.65. With 0=0.8, per equivalent expenditure for the median person is 56.23, and the appropriate poverty line is therefore 56.23/1.65=34 LVL per person per month. 6 The approach used here to model the impact of economies of scale on the profile of poverty differs to that presented in Gassmann (1998). Gassmann showed that for the 1996 HBS data, the ranking of households under differing assumptions of economies of scale was relatively stable (using the Spearman rank correlation coefficient) and thus argued that per capita expenditures were appropriate for the analysis of poverty. 9 Using the same method, the poverty line used with 0=0.6 is 42 LVL per person per month. By setting three different poverty lines for use with the three different measures of household welfare, the overall poverty heatdcounts calculated for each welfare measure are (as shown below) close to each other. This enables easier comparison of the specific poverty rates (for example, between the elderly and large families) calculated under the differing assumptions regarding economies of scale. 2.12 Measures of poverty: incidence, depth and severity. The incidence of poverty, or headcount index (P0), is defined as the percentage of individuals who are poor (i.e. who live in households that have monthly equivalent consumption below the poverty line). While the headcount index is the most widely-used measure of poverty, it does not give any information on the extent to which the welfare of individuals falls below the poverty line. The depth of poverty is measured by the poverty gap index (PI), which measures the average shortfall of equivalent consumption, expressed as a percentage of the poverty line.7 The poverty gap index is not sensitive to the distribution of welfare among poor households; if a household just below the poverty line were to make a transfer to a much poorer household, there would be no change in PI. The Foster-Greer-Thorbecke (1984) index (P2) measures the severity of poverty, and puts greater weight on the welfare levels of very poor households as compared with households with eq[uivalent consumption near to the poverty line.8 2.13 The incidence of poverty in Latvia (calculated over individuals) was 19.4 percent when per capita consumption is used as the wvelfare measure (Table 2.2).9 With an assumption of moderate economies of scale, the poverty rate is 17.8 percent, while with 0=0.6, the poverty rate is 18.2 percent. As is found in many countries, and as was the case in Latvia in 1996, poverty is much higher in rural compared with urban areas. With per capita consumption, the rural poverty rate was 28.5 percent, compared with only 10.8 percent of individuals being classed as poor in Riga and 19.5 percent in other cities. Allowing for economies of scale has no impact on the conclusion that poverty is higher in rural areas compared with Riga and other cities. Table 2.2: Poverty measures by location (percent) 0=1 0.8 6=0.6 Population P0 P; P2 Po P1 P2 P0 P1 P2 share Riga 10.8 3.2 1.4 10.5 2.8 1.2 11.3 2.9 1.2 32.6 OtherUrban 19.5 5.3 2.2 17.7 4.8 2.0 18.1 5.0 2.1 36.6 Rural 28.5 8.3 3.6 25.6 7.2 3.0 25.6 6.8 2.8 30.8 Latvia 19.4 5.5 2.4 17.8 4.9 2.0 18.2 4.9 2.0 100.0 7 Note that the average is calculated over all individuals (poor and non-poor), with non-poor individuals having a zero shortfall. s The three poverty measures can be calculated using the following formula: N I z pa where c= 0,1,2; q = number of poor individuals; N = number of individuals; z = poverty line; yi = consumption of i'th individual below the poverty line. 9 Since poor households tend to be larger, the household poverty rate (14.7 percent) was lower than the individual rate. 10 Source: Author estimates based on HBS, 1997, 1998. 2.14 Table 2.2 shows that the poverty gap for Latvia of 5.5 percent is quite small. A direct comparison of PI with other countries is problematic, as the poverty gap varies with the location of the poverty line. However, the average shortfall (as a percentage of the poverty line) of those individuals below the poverty line is 28.4 percent, when 0=1. While this indicates poverty in Latvia is deeper compared with Eastern European countries (recent estimates put the average shortfall for these countries at around 25 percent), it is not as deep as in Russia and Ukraine, where the average shortfall is over 40 percent (Milanovic, 1998). Poverty is deeper in rural areas, compared with Riga and other urban areas. However, the rural relative poverty ratio (defined as the rural headcount divided by the national headcount) declines from 1.47 to 1.41 as theta decreases from 1 to 0.6, while the relative poverty ratio increases for Riga and remains constant for other urban areas.1I 2.15 Latvia is a small country, however it is regionally very diverse. In Table 2.3, the poverty measures are presented on a regional basis, and they reflect marked regional differences in poverty. With per capita consumption as the welfare measure, the Riga region has a poverty rate of only 12.6 percent, while the poverty headcount for Latgale is 30 percent. These headcounts support the opinion of Latvians themselves that Latgale is the poorest region in Latvia (Gassmann, 1998) and CSB data which show that average per capita income in the Riga region is 40 percent higher than in Latgale. Kurzeme and Vidzeme have similar high poverty rates of approximately 24 percent, while the poverty rate in Zemgale is close to the national average. The ranking of regions by incidence of poverty is relatively stable with respect to differing assumptions about economies of scale. Table 2.3: Poverty measures by region (percent) 0=1 6=0.8 0=0.6 Population PO PL P2 Po PI P2 Po P1 P2 share Riga region 12.6 3.5 1.5 12.0 3.2 1.3 12.6 3.2 1.3 45.5 Kurzeme 24.5 8.0 3.8 22.7 7.2 3.3 23.9 7.2 3.2 13.8 Vidzeme 24.1 6.9 2.9 21.8 6.2 2.6 20.8 6.3 2.7 11.8 Zemgale 20.6 5.2 2.0 18.0 4.3 1.6 18.4 4.2 1.4 13.0 Latgale 30.0 8.4 3.6 27.1 7.4 3.1 26.8 7.1 3.0 16.0 Latvia 19.4 5.5 2.4 17.8 4.9 2.0 18.2 4.9 2.0 100.0 Source: Author estimates based on HBS, 1997, 1998. 2.16 The 1997/98 data reveal more marked regional differences in poverty than Gassmann (1998) concluded from the 1996 HBS data. This may reflect growing inter-regional differences in inequality and poverty", or else it may be because of differences in methodology. In particular, Gassmann (1998) adjusted household expenditures to reflect regional differences in prices, while this adjustment was not made in the present study. However, official regional price 10 This is because average household size in rural areas (2.5 persons) is larger than that found for urban households (2.3 persons). 1" As discussed in Goldman (1998), the current system of local funding of social assistance in Latvia may also be contributing to widening inter-regional differences in poverty. 11 indices are not available for Latvia. Furthermore, unofficial food price indices indicate only small regional differences, which are unlikely to have a large impact on poverty measurement. 2.17 Stochastic Dominance Tests. The incidence of poverty in Latvia is represented graphically in Figure 2.1, which shows the cumulative distribution functions (CDFs) of the log of monthly per capita consumption by different localities. The intersection of the poverty line with the CDF gives the incidence of poverty, which is the proportion of individuals with per capita consumption falling below the poverty line. Stochastic dominance techniques can be used to test the robustness of the conclusion that rural poverty is higher than urban poverty in Latvia. The theory of stochastic dominance shows that if the CDFs do not intersect at any point in the graph, poverty in the population represented by the curve lying everywhere above the other is greater, and this will be the case for whatever poverty line is chosen. This conclusion holds not only for the incidence of poverty, but also for measures of depth and severity of poverty. The CDFs in Figure 2.1 are typical of those found in many transition countries in that rural poverty dominates urban poverty over all possible poverty lines. Figure 2.1: C]DF of log monthly per capita consumption 0.9 0.8 CDF . Rural 0.7 0.6 Poverty line 0.5 CDF- Other Urban 0.4 0.3 __0.285 / // CDF - Riga 0.2 0.285 0.195 0.1 0.108 0O 0 1 2 3 4 5 6 7 Source: Author estimates based on HBS, 1997, 1998. 2.4 POVERTY PROFILE: CROSS-TABULATIONS 2.18 This section looks at how the poverty rate varies according to differing characteristics of individuals and households (tables describing the composition of poverty are in Appendix A). It is found that poverty in Latvia shares some of the general characteristics found in recent studies 12 of poverty in Eastern Europe and other FSU countries.12 In particular, the incidence of poverty in rural areas is higher than that found in the major cities, and children are particularly at risk of poverty, while poverty for pensioners is found to be lower than that of the general population. 2.19 Household size. Household composition is generally found to be a very important correlate of household poverty status and this is also true in Latvia. The headcounts in Table 2.4 suggest a strong positive correlation between household size and poverty when per capita consumption is the welfare measure. However, once economies of scale are allowed for, the relationship between household size and poverty becomes less marked. Of particular note is that with 0=1, single households have the lowest poverty rate, but with significant economies of scale, this is no longer the case. Table 2.4: Poverty rates by household size (percent) Population 0=1 0=0.8 0=0.6 share 1 7.4 12.1 23.2 14.1 2 12.0 13.8 16.7 23.5 3 16.8 15.7 15.6 23.6 4 21.9 17.9 15.7 22.7 5 39.5 30.2 22.8 10.2 6 42.1 32.3 24.0 3.5 7 46.8 29.5 21.8 1.3 8+ 50.9 43.7 25.2 1.0 All 19.4 17.8 18.2 100.0 Source: Author estimates based on HBS, 1997, 1998. 2.20 Presence of children. The impact of demographic composition on household welfare will depend on the extent to which different types of household members contribute to household resources and also the extent to which living costs vary between members. The contribution of children and the elderly to household resources will mainly depend on the availability and size of child- and age-related benefits. From Table 2.5, it is apparent that the presence and number of children aged 14 years and less is strongly correlated with poverty status. Individuals living in households with no children had a poverty rate of 12.2 percent, while this rate increased monotonically with additional children to 47.8 percent for individuals living in households with 3 or more children. These results support the findings elsewhere in the FSU and Eastern Europe that the presence of children is strongly related to poverty and hence should be considered as a candidate indicator for targeting. With moderate economies of scale (0=0.8), the impact of the number of children on poverty is reduced, however individuals living in households with 3 or more children still have poverty rates over twice the national average. 12 See, for example, Braithwaite, Grootaert and Milanovic (2000). 13 Table 2.5: Poverty rates by number of children <14 years (percent) Population 9=1 6=0.8 6=0.6 share 0 12.2 13.7 17.5 51.9 1 21.5 18.5 17.6 24.7 2 27.8 22.2 18.0 17.1 3+ 47.8 37.0 26.4 6.3 All 19.4 17.8 18.2 100.0 Source: Author estimates based on HBS, 1997, 1998. 2.21 Presence of elderly persons. From Table 2.6, it is apparent that households containing elderly persons do not have a higher risk of poverty. The poverty rate for persons living in households with no elderly persons or only one elderly person was close to the average poverty rate of 19.4 percent, while the poverty rate was 12.2 percent for those living with two or more elderly persons. In Russia, Ukraine and Armenia, it was also found that the presence of elderly persons in the household does not significantly increase the risk of poverty. The conclusion that households with elderly persons do not have a higher poverty rate is maintained even after allowing for the presence of economies of scale. Table 2.6: Poverty rates by number of elderly persons (percent) Population 9=1 6=0.8 0.6 share 0 20.5 18.3 17.8 72.9 1 18.0 18.4 22.1 19.9 2+ 12.2 11.5 11.6 7.3 All 19.4 17.8 18.2 100.0 Source: Author estimates based on HBS, 1997, 1998. 2.22 Table 2.7 gives further evidence to suggest that children contribute more to the dependency burden than do elderly persons. With per capita consumption, individuals from households with no children or elderly persons had a poverty rate of 12.5 percent. However, only 9.9 percent of persons from households with no children and two or more elderly members were poor. This qualitative result was robust to differing assumptions regarding economies of scale. It is interesting to note that the addition of elderly in small households reduces poverty headcounts, but in households with 1-2 children, the addition of elderly persons raises poverty rates. 2.23 Poverty by age groups. The above finding that the incidence of poverty is higher for households with children is reflected in the age distribution of poverty in Latvia. As shown in Figure 2.2, people under the age of 20 years have higher poverty rates than average, while people over the age of 40 years have lower poverty rates than average (this finding concurs with research on other FSU countries). With increasing econornies of scale, however, the age-poverty 14 profile is flattened considerably. Finally, there is a noticeable rise in the age-poverty profile for elderly women; with 0=0.6, this rise is quite marked and occurs earlier in the life cycle.13 2.24 Using per capita consumption as the welfare measure, individuals living in households with heads aged over 50 have a lower incidence of poverty compared to other groups (Figure 2.3), however this is less apparent when economies of scale are allowed for.14 Table 2.7: Poverty rates by number of children and elderly persons (percent) Population 0=1 0=0.8 0.6 share 0 children 14years 0 elderly persons 12.5 13.6 16.4 31.1 1 elderly person 12.7 15.5 22.5 14.7 2+ elderly persons 9.9 10.1 11.0 6.1 1 child 14 years 0 elderly persons 20.9 18.2 17.5 21.2 1 elderly person 27.4 23.1 20.2 2.9 2+ elderly persons 13.9 8.5 8.5 0.6 2 children <14years 0 elderly persons 26.3 20.9 17.4 15.1 1 elderly person 39.0 30.7 22.7 1.6 2+ elderly persons 35.8 35.8 24.0 0.4 3+ children 14 years 0 elderly persons 48.3 37.9 27.6 5.5 1 elderly person 43.7 32.7 18.5 0.6 2+ elderly persons 43.1 15.0 15.0 0.1 All 19.4 17.8 18.2 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: A person is defined as elderly if aged 65 years or over. 2.25 Poverty by gender. From Figure 2.2, it appears that males and females are equally at risk of poverty, which supports the findings in other FSU countries. However, as it is not possible with the HBS data to study the within-household distribution of consumption, it is therefore assumed that all household members receive an equal share. To the extent that this is not true, there will be a gender dimension to poverty, which is not being detected using the standard questionnaire format of the household survey. 2.26 While poverty incidence of individuals does not vary by their gender, this does not mean a gender dimension to poverty in Latvia cannot be discerned in the data. Overall poverty rates by gender of household head are identical (see, for example, Table B4 in the Appendix). 13 The rise in poverty rates for elderly women largely reflects the differential survival rates of women and men in FSU countries. 14 The household head is listed as the first person in the household roster (it is not clear from the survey documentation whether the household itself determines who the head is, or whether it is determined by some criterion of the CSB). Gassmann (1998) identified the household head as the main breadwinner in the family (see Table B9 in the Appendix to the present report, where this concept of headship is used in the consumption regressions). 15 However, when moderate or strong economies of scale are assumed, individuals living in households headed by women aiged 30 to 49 have higher rates of poverty (compared with those living in households headed by men in this age group) (Figure 2.3). The finding of a potential gender-dimension to poverty in Latvia supports what has been found in other FSU countries, and also supports Gassmann's (1998) conclusions based on the 1996 HBS data. Structural changes in the labor market and a fall in the affordability of day care would contribute to households headed by young females having a higher poverty risk. Figure 2.2: Poverty rates by gender and age 3 5 .0 _._._..........__ _____.._..... ... _____ . .. _. _.-------.... . __ . 30.0 25.0 ----------- ' -K- 20.0~~~ 15.0~~ 10.0 , -5 yrs 5-9 10-14 15-19 20-29 30-39 40-49 50-59 60-69 >=70 Ag. | 4K(M) tht-= I --- - - (M) tIetl-0.8 - K--(M) thbt1=0.6 - (F) th.ta= I -.- (F) th.ta=0.8 -_--(F) thetl=0.6| Source: Author estimates based on HBS,. 1997, 1998. For underlying numbers, see Table B2 in the Appendix. Figure 2.3: Poverty rates by gender and age of household head 30.0 . . -- --.. . . . .-.....,.-..- ._ ,. - ----. ---. 25.0 20.0__ _ _ _ °aL-~ ~~ --gt0> 15.0 10.0 <30 y..rs 30-39 40-49 50-59 60-69 >50 Ag. of hoo..hold ho.d 1ll----(M ) h.-l= I - - - (M ) -h,..=0.0 -4--(M 3 th.t.=0.6 - (F) th.t-= I - - - - (F) th.r.=0.a -0--8 F) rb.,.=0.6 Source: Author estimates based on HBS, 1997, 1998. For underlying numbers, see Table B3 in the Appendix. 16 2.27 Those living in households headed by elderly (Ž70 years) women have higher poverty rates (compared with those living in households headed by elderly men), regardless of presence or absence of economies of scale (in fact, as 0 decreases, the difference in poverty rates increases). With an assumption of strong economies of scale (0=0.6), this divergence in poverty rates is found to occur earlier in the life cycle, and is very marked; 27.4 percent of individuals living in households headed by elderly females are poor, compared with 16 percent of those living in households headed by elderly males. Thus, the HBS data indicate a sharp increase in the poverty risk of elderly women, most probably associated with women living longer than their partners. 2.28 Sensitivity of poverty rankings to different thetas. The impact of differing assumptions regarding economies of scale can be seen in Figure 2.4, which shows the relative incidence of poverty for various household types for a range of 0 between 1 and 0.5.15 For each value of 0, the poverty line is set so that the overall incidence of poverty (i.e. for all persons in Latvia) is 20 percent (for 0=1 the poverty line was set at 28.3 LVL, while for 0=0.5 it was set at 48.15 LVL). This approach then allows the comparison of the poverty rate for a given household type with the overall poverty rate, at differing levels of 0. The four household types studied are: households comprising only elderly; "high child ratio" households (defined as those households with a higher than average number of children); female-headed households; and households with a higher than average dependency ratio (defined as the number of non-working-age family members divided by household size). 2.29 For 0=1, the poverty rate of the elderly is very low and that of high child ratio households is very high. While the poverty rates of these two groups converge as 0 falls, the cross-over point (indicating a poverty re-ranking) does not occur until 0 is less than 0.65. This indicates that fairly significant economies of scale need to be present before one can conclude that the elderly are worse off than children in Latvia.16 As female- headed households tend to be on average smaller, the incidence of poverty for those living in such households rises as 0 declines from 1 to 0.5. For per capita consumption, female-headed households have a poverty rate slightly lower than the national average of 20 percent, however for a 0 slightly higher than 0.9, their poverty rate increases above that of the overall population (this cross-over point was similar to found for most of the countries in the Lanjouw, Milanovic and Paternostro [1998] report). Finally, for 0>0.8, poverty among the population residing in households with high dependency ratios is higher than the overall average, while for 0<0.8, such persons have relatively low poverty rates. This finding is in contrast with the results of Lanjouw, Milanovic and Paternostro (1998), who found that for all countries except Estonia, the poverty headcount for those living in high-dependency households was greater than the overall average, regardless of 0. 15 This figure is based on analysis which Lanjouw, Milanovic and Patemostro (1998) presented for several Eastern European and FSU countries (Latvia was not included). 16 Lanjouw, Milanovic and Patemostro (1998) found that a poverty re-ranking between the elderly and children occurred in Poland for 0 approximately equal to 0.5, while for all other countries in their study, this re-ranking occurred at 0=0.7 or higher. 17 2.30 Education. In theory, one would expect a relationship between an individual's ability to avoid poverty and his or her holding of assets in the form of human capital (e.g., education) and physical capital (e.g., land). However, the return to education is likely to be dependent on the stage of economic transition; the further advanced transition is, the higher the demand for well- educated workers who are able to adapt to newly emerging skill requirements. In Latvia there appears to be a reasonably strong relationship between education and household welfare. Table 2.8 shows a decreasing relationship between incidence of poverty and educational attainment of individuals aged 15 years and over; in May, the poverty rate for a person with higher education was 4.6 percent compared with Figure 2.4: Impact of theta on poverty rates for certain household types 30.0 3.................. .... .... ........ .. .. ... _.. .... . .. _.. ... .._ . .. .. ... ..._ . ...... ... 215.0 ...... ... .0.0 .5.0 25.0 A 0.0 1.0 0.9 0.8 0.7 0.6 0.5 Theta - Baseline poverty rate - 6 Elderly household - A High child ratio X Female-headed household -B High d ratio Source: Author estimates based on HBS, 1997, 1998. 23.3 percent for a person with primary schooling or no formal education. This phenomenon is also evident in the relationship between poverty status and education of household head. Individuals living in households headed by a person with higher education had a poverty rate of 6 percent, compared with 27.2 percent for individuals living in households headed by persons with primary or no formal educaition. It is interesting to note that there is really only a large 'payoff' to education (in terms of reduction of poverty rate) for the attainment of higher education qualifications; persons with vocational qualifications have higher than average poverty rates, while those with secondary education do not have poverty rates much lower than the average. 18 2.31 Socio-economic status. Table 2.9 shows poverty incidence by socio-economic status.17 Individuals living in wage-earning or self-employed households have poverty rates lower than the national average, while persons from households whose main income source is 'other' social benefits have a poverty rate close to 50 percent (the qualitative findings in Gassmann 1998 were similar). The lowest poverty rates are for persons living in self-employed households (however, only 5.8 percent of the population live in such households). As has been found in other FSU countries, individuals living in pensioner households do not have a higher than average incidence of poverty. Table 2.8: Poverty rates by education (percent) Povulation Riza Other Urban Rural Latvia share (Latvia) Education level, persons 215 years Primary or less 14.3 21.2 29.4 23.3 30.5 Vocational 20.0 24.8 32.7 26.0 3.8 Secondary 9.6 18.1 24.1 16.5 52.3 Higher 3.3 6.2 5.5 4.6 13.5 All 9.5 17.7 25.6 17.4 100.0 Education level of household head Primary or less 16.5 22.8 34.3 27.2 23.1 Vocational 23.2 28.4 33.8 29.6 3.6 Secondary 11.5 20.7 26.7 19.2 57.8 Higher 4.5 8.0 6.9 6.0 15.4 All 10.8 19.5 28.5 19.4 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. 2.32 Labor-force status. As expected, there is a strong correlation between labor-force status and poverty. The poverty rate of the employed (15.3 percent) is lower than the national average and that of the unemployed (38.5 percent) is much higher than the national rate for persons over 14 years of age (Table 2.10). The relationship between poverty and labor-force status of the household head displays similar patterns. Furthermore, in the multivariate analysis presented in Section 5, a significant link between the number of unemployed household members and household welfare is found. 2.33 Access to private plot. As has been found in other FSU countries, it appears that in Latvia, households are able to significantly improve their standard of living via consumption of home-grown produce (Table 2.11). This is particularly the case for households in rural areas where the presence of a food plot reduced the poverty rate from 30.4 percent to 22.6 percent. 17 The socio-economic group of a household is determined by what category accounts for largest share of total household income. 19 Table 2.9: Poverty rates by socio-economic group (percent) Population Riga Other Urban Rural Latvia share (Latvia) Wage earner 9.5 17.3 29.7 17.1 55.7 Self-employed 7.9 11.2 18.9 16.5 5.8 Pension 13.1 20.4 25.0 19.5 26.9 Other social benefit 19.4 44.0 68.6 47.1 1.7 Other income 12.0 32.4 34.2 29.2 10.0 All 10.8 19.5 28.5 19.4 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. Table 2.10: Poverty rates by labor force status (percent) Poaulation Riga Other Urban Rural Latvia share (Latvia) Labor force status, persons 215 years Employed 6.7 13.7 26.9 15.3 46.7 Unemployed 26.0 42.6 50.1 38.5 7.3 Not in labor force 9.8 16.4 22.3 16.1 45.9 All 9.5 17.7 25.6 17.4 100.0 Labor force status of household head Employed 8.7 15.8 29.3 17.7 56.6 Unemployed 25.3 45.7 56.2 40.9 6.1 Not in labor force 11.3 19.8 24.7 18.5 37.4 All 10.8 19.5 28.5 19.4 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. Table 2.11: Poverty rates-by type of settlement and presence of food plot (percent) Povulation Riga Other Urban Rural Latvia share (Latvia) No plot 11.3 20.9 30.4 20.5 74.6 Has plot 8.7 16.2 22.6 16.1 25.4 All 10.8 19.5 28.5 19.4 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. 2.5 POVERTY PROFILE: REGRESSION ANALYSIS 2.34 Regression analysis is often used to establish the existence of significant relationships between household welfare and the characteristics of those households. The results of the regressions can be used to establish whether a particular household characteristic is significantly 20 correlated with household welfare and thus can be used to predict household welfare as an input into social assistance targeting. 2.35 Table 2.12 presents the results from regressing the log of household consumption on different household characteristics.'8 The objective is to identify determinants of welfare and poverty which, in the short run, can be used as targeting variables. Hence it is strictly only valid to include those RHS variables which can be considered exogenous in the short-run. For most of the variables in Table 2.12, the assumption of exogeneity is not a heroic one. While household welfare determines the education that children receive and hence influences human capital accumulation, it is reasonable to expect that in a one-period model, the education variables are exogenous. Even the labor-market variables can be considered fixed in the short-run, since in transition economies such as Latvia, unemployment is high and largely structural, and often the supply of housing is not sufficiently flexible to permit easy migration to areas experiencing growth in labor demand (Braithwaite, Grootaert and Milanovic, 2000). 2.36 While the explanatory power of the models as indicated by the R2 statistics are quite low, many of the coefficients are significant. The following conclusions from the consumption regressions can be made.19 The human capital of a household is embodied in its members and hence their numbers (by age and sex) are included in the regression.20 The number of persons in all demographic groups is negatively correlated with welfare when per capita consumption is used as the welfare measure. However, with moderate economies of scale (0=0.8), this negative impact is decreased, and with 0=0.6, only the presence of children is associated with a fall in welfare. The implication is that even with the presence of economies of scale, households do not succeed in maintaining their welfare levels when the number of children increases. It is of interest that changes in theta only have a big impact on the estimated coefficients on the household composition variables; coefficients on the other variables in Table 2.12 do not vary significantly with the different measures of household welfare. 2.37 The human capital of the household is also proxied by the education of the household head (which arguably has a greater impact on welfare than the education of other household members). There appear to be strong returns to higher education in Latvia; the household head having a higher education degree is associated with per capita household consumption being 23.6 percent higher than that of the reference category (a household where the head has secondary 18 The semi-log functional form of the model was chosen for both theoretical reasons and because it passed a test of model specification (for details see Appendix D). 19 The comparisons to findings in Eastern Europe and other FSU countries refer to findings from Braithwaite, Grootaert and Milanovic (2000). 20 These variables also reflect the consumption needs of the household and thus capture the ability of a household to cope with a changing economic environment. 21 One would expect that the method used to determine household head may alter the conclusions in this section. However, consumption regressions which include the characteristics of the main breadwinner are presented in Table B9 in the Appendix, and the findings are qualitatively the same as those presented in Table 2.12. 21 education). It is of note that there does not appear to be any welfare gain associated with household heads having vocational training over pnmary or lower education. 2.38 The work experience of the household head is proxied by age. For all values of 0, the coefficients on household age and age-squared are not significantly different from zero. This result is at odds with what was found for the three Eastern European countries and the Kyrgyz Republic (where the expected inverted U-life-cycle patterns were in evidence). 2.39 The link with the labor market is captured by the inclusion of dummy variables representing whether the household head is unemployed or inactive, and also the number of unemployed household members. The strong distributional impact of unemployment is in evidence: over and above other household attributes, the presence of an unemployed household member reduces household welfare by approximately 25 percent. The head being inactive reduces welfare by around 17 percent and the head being unemployed reduces Table 2.12: Consumption regressions 0=1 0.8 0=0.6 Dependent variable: lop0fconsumptionl(household size) Of coefficient p-value coefficient p-value coefricient p-value Household composition v number of children 14 years -0.220 0.000 -0.151 0.000 -0.083 0.000 number of male adults -0.132 0.000 -0.047 0.000 0.038 0.000 number of female adults -0.127 0.000 -0.039 0.001 0.050 0.000 number of elderly persons -0.130 0.000 -0.037 0.009 0.055 0.000 Education of household head primary or less -0.166 0.000 -0.168 0.000 -0.169 0.000 vocational -0.126 0.000 -0.128 0.000 -0.131 0.000 higher education 0.236 0.000 0.236 0.000 0.236 0.000 Age of household head -0.003 0.246 -0.003 0.219 -0.003 0.196 (age of household head)x102 0.000 0.463 0.000 0.564 0.000 0.676 Head is female -0.071 0.000 -0.076 0.000 -0.082 0.000 Head is inactive -0.168 0.000 -0.169 0.000 -0.170 0.000 Head is unemployed -0.390 0.000 -0.391 0.000 -0.393 0.000 Number of unemployed in household72 -0.252 0.000 -0.249 0.000 -0.246 0.000 Household situated in Riga region 0.138 0.000 0.138 0.000 0.139 0.000 Household situated in Vidzeme 0.016 0.517 0.013 0.579 0.011 0.647 Household situated in Zemgale 0.083 0.000 0.085 0.000 0.087 0.000 Household situated in Latgale -0.054 0.014 -0.055 0.013 -0.055 0.012 Household situated in capital (Riga) 0.043 0.041 0.041 0.051 0.038 0.064 Household situated in rural area -0.073 0.000 -0.079 0.000 -0.085 0.000 Household has access to food plot 0.091 0.000 0.093 0.000 0.095 0.000 Constant 4.477 0.000 4.444 0.000 4.411 0.000 N 7690 7690 7690 R 2 .272 .215 .207 Percentage of non-poor households with correctly predicted poverty status 3/ 95.2 98.8 99.9 Percentage of poor households with correctly predicted poverty status 3' 28.6 8.8 1.7 Source: Author estimates based on HBS, 1997, 1998. 22 Notes: Omitted categories are head has secondary education and household situated in other urban area in Kurzeme region. P- values less than 0.05 (0.01) indicate that the coefficient is significant at the 5% (1%) level. 11A test of equality of the coefficients of the household composition variables was rejected and hence they are included separately, rather than including household size. 2 If the household head is unemployed then this is the number of additional unemployed household members. A household is predicted poor if predicted consumption (calculated using the estimated coefficients) is less than the poverty line. welfare by approximately 39 percent. The estimated impact of open unemployment on household welfare supports what has been found in Eastern Europe and other FSU countries. 2.40 Ownership of land is a key determinant of cash income and consumption of food, especially in rural areas. In most transition countries, ownership of land is not yet fully subject to household choice and hence it can be considered exogenous in the short tern. In Latvia, the presence of a food plot is associated with a 9.1 percent increase in per capita consumption. 2.41 Household welfare and poverty are also affected by the economic environment, which affects both income-earning opportunities and the level of social and economic infrastructure. Hence, the ability of a household to adjust to economic transition will be influenced by its location. The impact of location on household welfare is modeled by the inclusion of regional and locality dummy variables in the consumption regressions. Significant coefficients on these variables indicate that welfare differences across locations are not fully explained by the distribution of demographic and economic characteristics of households (i.e. there is a location- specific effect on welfare). In Latvia, it is apparent that households living in rural areas have welfare levels approximately 8 percent lower than comparable households residing in urban areas. The regional differences in poverty identified in Table 2.2 are also evident in the multivariate analysis; a household residing in Latgale has welfare approximately 5 percent lower than a comparable household in the Kurzeme region, while households in the Riga region and Zemgale have welfare levels between 8 and 14 percent higher. 2.42 Finally, a dummy representing the presence of a female household head was included in the regressions to test for the existence of a gender dimension to poverty in Latvia. Female- headed households have welfare levels which are 7-8 percent lower, which is consistent with the findings for other transition economies. 2.43 Despite the relatively low explanatory power of the regressions, they are reasonably accurate at predicting household poverty status. Using per capita household consumption as the welfare measure, 95.2 percent of non-poor households were predicted as non poor (using the estimated coefficients) but only 28.6 percent of poor households had their poverty status correctly predicted using the consumption regression. 2.44 Consumption regressions by locality. The per capita consumption regressions were run separately for each locality to see if the factors influencing household welfare differ depending on where the household is located (see Table B8 in the Appendix). While the findings are qualitatively the same for each locality, there are some noteworthy differences. The returns to education appear to be higher in rural areas (this is a surprising result), while the variables indicating the labor-market attachment of the household head and other household members are more important in determining household welfare in urban, compared with rural, areas. 23 2.45 Poverty regressions (probit). The possibility of systematic measurement error in consumption has led some authors (see for example, Braithwaite, Grootaert and Milanovic, 2000) to suggest the use of poverty equations where the dependent variable is binary (poor/non- poor). To the extent that the results from the probit model confirm the welfare-levels regression results, then the poverty equations are useful as robustness checks of the latter. The probability derivatives from the probit models reported in Table B10 in the Appendix give the predicted change in the probability of a household being poor, calculated at the mean of continuous variables and for a change from zero to one in the case of the dummy variables. Some of the conclusions about the impact of household characteristics on household welfare are different to those garnered from Table 2.12. For example, the household head having vocational education does not improve the probabiliity of the household avoiding poverty (relative to the head having primary education or less). With per capita consumption as the welfare measure, the head having vocational education is associated with an 11.1 percent increase in the probability of poverty (compared with the head having secondary education), while the head having primary education or less is associated with an 8.6 percent increase in the probability of poverty. The probits indicate that the relationship between the age of the household head and household poverty status forns an inverted-U shape (this shape was not present in the relationship between equivalent consumption and the age of the household head). The remainder of the results from the probits are qualitatively the same as was found in the welfare regressions. 2.6 CONCLUSIONS 2.46 In this chapter, consumption data from the 1997/98 Household Budget Survey (HBS) have been used to construct a profile of poverty in Latvia. The profile in this chapter updates that of Gassmann (1998), who used the 1996 HBS. Many of the characteristics of poverty in Latvia described by Gassmann (1996) are found to be present in the more recent data, despite certain methodological differences witlh regards to the construction of the household welfare measure. In particular, poverty is found to be more prevalent and deeper in rural areas, and, as has been found in other transition economnies, it affects the young more than the elderly. There appears to be a gender dimension to poverty in Latvia, with individuals living in households headed by young or elderly women having relatively high rates of poverty. As expected, education and labor-market attachment are very important in helping a household avoid poverty. Multivariate analysis (welfare-levels regressions and probit regressions) confirm the findings from the poverty cross-tabulations. 2.47 A major theme of this chapter was testing the extent to which the profile of poverty changes with differing assumptions regarding the presence of economies of scale. It was found that many of the conclusions about poverty based on per capita consumption (which implies zero economies of scale) are robust to the incorporation of economies of scale. For example, it was found that a level of 0 smaller than 0.7 (which implies fairly significant economies of scale) would be necessary to reverse the conclusion that individuals living in households with many children are more at risk of poverty than the elderly. 24 3. AN ANALYSIS OF THE LABOR MARKET IN THE REPUBLIC OF LATVIA 3.1 INTRODUCTION 3.1 Over time economic transition from central planning to market orientation can improve individuals' welfare. However, as transition moves forward, institutions change, restructuring takes place, and some groups in society will suffer. As part of an investigation of those people whose welfare decreases during transition in Latvia, it is important to recognize labor markets' central role in determining employment opportunities, opportunities which have a large impact on individuals' well-being. Wage and employment prospects determine whether a person is at risk of entering poverty or of successfully moving through transition1. As part of a broader inquiry into the causes and dynamics of poverty in Latvia, this chapter will analyze how the labor market operates to offer wages, allocate jobs, and affect livelihoods. 3.2 Specifically, it will: * Consider the wage structure that labor markets offer as central to evaluating individuals' welfare and to addressing gender discrimination; * Focus on labor market participation as an insight into labor supply and people's decisions about whether or not to work; * Consider regional differences in labor demand as the factor that pulls workers into remunerative jobs; and * Bringing together labor supply and demand, evaluate unemployment as an indicator of labor market effectiveness and of the likelihood that an individual risks entering poverty. After considering these issues, it presents policy suggestions that can make the Latvian labor market more effective and minimize the human costs of economic transition. 3.3 The chapter proceeds as follows. Section 3.2 reviews the literature on studies of the labor market and poverty in Latvia and presents the study's methodological approach. Section 3.3 summarizes findings with regard to the structure of earnings and gender inequality. Section 3.4 considers labor supply through an analysis of labor force participation. Section 3.5 discusses labor demand. Section 3.6 discusses the risk factors for becoming unemployed. Section 3.7 concludes and offers general policy suggestions. See, for example, Chapter 2, paragraph 2.12. 25 3.2 LITERATURE REVIEW AND METHODOLOGY 3.4 Because labor market outcomes have so many diverse implications for people's welfare, there is an extensive literature that touches on these outcomes, particularly during post- communist transition. Several relevant studies relate specifically to Latvian labor markets. 3.5 Labor Market Analyses. The Government of the Republic of Latvia has analyzed labor market issues in several reports. The Latvian Government's "Social Report 1996 - 1997" summarizes the issues central to the functioning of the labor market. It presents summary statistics of labor force participation, employment, and both registered and unregistered unemployed, discussing changes in these indicators and characteristics of people in each category. As part of this summary, it contends that unemployment does not vary by ethnic group, though it is not particularly clear how it reaches that conclusion. The report describes the legal basis for the difference between registered and non-registered, arguing that focusing on those who register underestimates actual unemployment. It also outlines the labor market policies in effect, including the activities of the State Employment Services to organize job-seeker's clubs, job placement, retraining, and paid temporary public works. The study discusses the key issues in worker remuneration and labor relations. 3.6 In a broad study of economic prospects, the Ministry of Economy's report on Economic Development in Latvia considers living standards and employment. It provides an informative overview social service policies, private purchasing power, and employment prospects in each of the major categories of economric activity. Like the "Social Report" it also calls attention to the differences between an unemployment definition focusing on those registered as unemployed and that used by the International Labor Organization. 3.7 The chapter on "Labor Market Issues in Latvia" from the IMIF Staff Report (1998) also directly addresses several of this study's topics. Like the other reports, the IMF pointed out that the official unemployment definition, based on those who register as unemployed, underestimates quite severely the actual unemployment level. In 1997 unemployment rates based on international standard definitions were 20 percent while the official rate was 7 percent. The study also records that the labor market participation rate has been declining since 1990 as people either become discouraged or return to increase their education. The 1997 participation rate of 50.5 percent is quite low compared to other countries. In his report on "Poverty and the Labour Market in Latvia" written for the ILO, Keune (1998) uses Household Budget Survey and Labour Force Survey data to consider the relationships between household expenditures, economic activity and poverty. 3.8 Social Assessment. Supplementing these poverty profiles are social assessment studies that conduct structured interviews with poor households to understand their impressions about the sources of poverty. According to the World Bank/UNDP 1998 social assessment, the households most likely to be poor are those with unemployed members, particularly older adults. There is some question as to whether being of non-Latvian nationality hurts or hinders employment prospects, and interviewees noted gender and age discrimination in the labor market. Further, particular regional characteristics make finding a job more onerous, so that where Soviet-era plants were created (such as Daugavpils in the Latgale region, Jelgava in the Zemgale region, and Liepaja in the Kurzeme region), employment prospects are grim. The study 26 also discusses incentives to register for unemployment benefits, presenting why some people are ineligible. 3.9 Methodology Overview. The chapter will build upon this literature through an empirical analysis of the Latvian labor market, seeking to verify and extend upon these findings through empirical analysis of micro-economic data. * It will focus first on the earnings structure that the labor market generates, pointing to the relationships between different worker characteristics and their expected earnings. * As part of this analysis of earnings structure, it will compare men's and women's wages, carefully discussing gender discrimination in the Latvian economy as well as other sources of discrimination. * The chapter will then consider labor supply, focusing on the decision to participate in the labor market as a central time allocation decision. - The study will present information about labor demand and its variation across different regions of the country. * Finally, it will consider Latvian unemployment, paying particular attention to alternative unemployment definitions, the factors associated with high unemployment, differences between men's and women's unemployment, and unemployment changes over time. 3.10 Data. To carry out this empirical analysis, the study uses four sources of data. - The Household Budget Survey (HBS) from the first quarter of 19972 is used to analyze the structure of earnings and predict how much people would expect to receive given their personal characteristics and location. * When analyzing Latvian labor force participation and unemployment, the chapter focuses on Labor Force Survey micro-data from the first quarter of 1998 (LFS 98). * To offer some information about trends in these labor market outcomes, the chapter compares these 1998 participation and unemployment findings with similar Labor Force Survey data collected in the second quarter of 1996 (LFS 96). * An Enterprise Survey and administrative data from the State Employment Service provide data on labor demand by administrative district. Table 3.1 presents descriptive statistics about the primary individual-level variables used in the analysis. 2 1 would like to thank Mr. Robert Ackland for extracting a sub-set of the HBS data including information on wages and relevant independent variables. 27 3.11 Because the HBS collected different information than the two LFS sources, it is worth drawing attention to difficulties comparing the three data sets presented in Table 3.1. Because the HBS data is comprised of people reporting non-zero wages from a main job, the 933 women and 914 men are not a random sample of the Latvian population but are selected based on the fact that they are working. As a result, their individual characteristics are not comparable with the respondents from the LFS surveys3, which consist of random samples of the whole population. For example, among people in the HBS who work, the average age is 40 years. In the entire Latvian population, the average age is 47.7 years for women and 43.6 years for men.4 Educational attainment is higher in the BBS data (e.g., 25 percent of women have higher education in the HBS data, though only 12 percent do in the LFS 98 data). The HBS and LFS surveys also used different categories to describe educational attainment: the HBS data distinguishes five categories, while the LFS data contains eight. This poses a problem for using the characteristics of the HBS data to predict the earnings of those included in the LFS (the rationale for which is described below). However, for predicting earnings it is possible to collapse the LFS data into five categories comparable to those used in the HBS. Those LFS respondents who reported having "Secondary Specialized", "Technical Secondary", or "Comprehensive Secondary" education are grouped into the "Secondary Education" category when being compared with the HBS data. 3.12 Earnings Structure and Gender Discrimination. The chapter first analyzes the structure of earnings in the Latvian labo:r market. While it would be convenient for the analysis if the LFS survey included in-depth information about worker's earnings, like many other labor force surveys, this instrument did not ask working respondents detailed information about their income. Instead, the chapter relies upon the fBS data to discuss earnings. 3.13 Using HBS data, the chapter presents ordinary least squares regressions of the human capital characteristics associated with earnings differences. From this analysis, we gain insights into the returns to different types of education and how earnings are related to respondents' age. Because of their different earnings structures, men and women are treated separately. Through a careful decomposition of these gender differences, we consider the sources and degree of gender discrimination in the Latvian economy. These regressions also include respondents' nationality and region. From these elemrrents, the study provides information about other rigidities and discrimination in the Latvian labor market. Because of the difficulties of moving easily from one area of Latvia to another, earnings differ markedly by region. The regression analysis allows one to decompose how much such differences are due to the characteristics of people in each region and how much are inherent to the regional economic situation. Further, we can use this information to consider earnings differences according to nationality. 3.14 These earnings regressions serve a second function. With information from the HBS data about earnings structure and individual characteristics (education, age, gender, nationality, etc.), we assume that similar relationships hold in the LFS data sets. Though the LFS data include inadequate information about people's earnings as noted above, by looking at LFS respondent 3 When one selects from the LFS '98 and ]LFS '96 only those people who are currently working, the demographic characteristics are roughly comparable to the HBS sample. 4 While the rest of the LFS analysis uses sample weights to ensure that the information represents the Latvian population, these descriptive statistics are unweighted to show the characteristics of the sample itself. 28 characteristics, we can predict on average what their earnings should be in the Latvian labor market, regardless of whether they actually worked. These predicted earnings are useful for analyzing labor supply and unemployment. 3.15 Labor Force Participation. Central to any analysis of a labor market is labor supply. During transition, people must decide how much labor to offer for work. A crucial aspect of how people respond to changing incentives during transition is that many choose not to work and leave the labor force. As existing labor market analyses have shown, in Latvia participation rates are falling. Non-participation may be due to an interest in pursuing education, family responsibilities, age, or a realization that earnings do not make up for the next best alternative use of ones time. As the Latvian labor market moves through its transition from central planning to more market-based incentives, it is important to monitor who chooses to leave the labor force. 3.16 By investigating labor force participation, the study explores labor supply decisions leading to decreased participation. Relying on the econometric tool of probit analysis, it presents evidence about the characteristics of people who choose not to be part of the labor market. These probit findings summarize the marginal effects of gender, education, age, nationality and location on whether or not someone chooses to work. Further, because ones labor force decision is affected by the earnings that the labor force would offer, the chapter also includes information about the relationship between the earnings one expects to earn and the likelihood of participation. Finally, this labor supply analysis also looks to changes in participation across time, comparing the LFS 98 and LFS 96 data using similar methodology. 3.17 Labor Demand. Coupled with labor that people choose to supply, the quantity of labor that firms are looking to employ is a crucial determinant of employment possibilities. The larger the demand for labor, ceteris paribus, the better off will be those who participate in the labor force, for higher wages and lower unemployment will result. While firms demand labor of different types, distinguishing workers by their skills and experience, the total number of jobs which firms offer gives insights to overall labor demand. In addition, the vacancies firms seek to fill indicates how much labor they demand beyond that which they already employ. This study will use information about employment and vacancies, disaggregated by administrative district, to proxy labor demand in different areas of Latvia. 3.18 Unemployment. While there is some disagreement about the micro-economic causes of unemployment, the existence of people wanting but unable to work indicates that the labor market is not matching workers to jobs effectively. While different types of unemployment exist in all economies, in those transition economies adjusting to a new set of institutions and incentives, unemployment rates provide a useful barometer of the labor market's difficulties in equilibrating supply and demand. 29 Table 3.1: Descriptive statistics Household Budget Labor Force Labor Force Survey 1997:01 Survey 1998:01 Survey 1996:02 Women Men Women Men Women Men Earnings: main job (lats/mo.) 74.47 95.47 .. [Log Variance] [0.345] [0.387] HUMAN CAPITAL Higher Education 24.8% 19.5% . Secondary Education 66.5% 61.8% Vocational Education 2.4% 7.5% Primary Education 6.2% 10.4% Less than Primary 0.2% 0.8% Higher Education 12.2% 10.1% 13.0% 10.8% Secondary Specialized 19.7% 18.0% 22.6% 19.7% Technical Secondary 7.0% 12.2% 6.8% 13.3% Comprehensive Secondary , 23.6% 19.5% 22.0% 18.7% Vocational Education 1.9% 5.9% 1.8% 5.3% Basic Education 22.6% 25.0% 21.7% 23.3% Less than Basic 11.9% 8.9% 10.9% 8.3% No Formal Education 1.0% 0.4% 1.2% 0.6% Age 40.2 40.0 47.7 43.6 46.1 42.1 (Standard Deviation) (11.3) (12.8) (19.5) (17.9) (18.5) (17.1) NATIONALITY Latvian 59.3% 59.4% 61.9% 62.9% 58.8% 59.7% Russian 30.3% 30.7% 27.0% 26.0% 29.1% 28.3% Other Nationalities 10.3% 9.8% 11.2% 11.1% 12.1% 12.1% MARITAL STATUS Married 65.9% 80.1% 52.3% 63.7% 56.0% 65.4% Single 14.4% 16.0% 20.5% 26.7% 18.5% 25.8% Divorced 13.3% 3.2% 9.2% 5.4% 9.2% 5.6% Widowed 6.4% 0.8% 17.9% 4.2% 16.2% 3.2% REGION Riga City 35.4% 37.5% 23.6% 22.5% 29.1% 27.4% Riga Region 15.3% 13.8% 10.6% 10.8% 10.3% 10.3% Kurzene 12.2% 15.1% 16.4% 16.7% 14.0% 14.5% Vidzeme 10.3% 11.2% 16.6% 16.6% 16.6% 17.7% Zemgale 13.0% 11.3% 14.5% 13.8% 13.7% 13.5% Latgaie 13.8% 11.2% 18.4% 19.7% 16.3% 16.5% Urban 78.4% 75.6% 60.3% 57.7% 61.2% 59.1% Number of Observations 933 914 8304 6844 6221 5265 30 3.19 The LFS data allows one to generate different definitions of unemployment. According to the International Labor Organization (ILO), a person is unemployed if she has not worked in the last week and has looked for a job in the past two weeks. Using this ILO definition, one can establish who is working, who is unemployed, and who no longer participates in the labor force. As outlined in previous labor market analyses, the Republic of Latvia officially reports unemployment based on a definition of who registered with the State Employment Service. While this definition has the benefit of being easy to obtain from early transition years, it poses several problems. As discussed in Section 6, Latvian labor regulations preclude many who are out of work and looking for a job from registering. Ignorance of the unemployment registration process might deter others. Some who are currently working might also register as unemployed to get benefits. Because of these problems, if based on this "registered unemployed" definition, officially reported unemployment statistics will not appropriately reflect labor market operations. 3.20 Again using probit analysis, the study explores the individual characteristics associated with higher probability of -being unemployed. This allows one to identify carefully the risk factors for unemployment, including gender, education, nationality, and location. Further, it compares the risk factors for being unemployed according to the standard ILO definition with the factors indicating who registers as unemployed with the state employment service. The difference between these offers insight into how a non-standard definition of unemployed masks some aspects of the true unemployment situation. 3.21 To buttress this analysis, a separate probit sheds light on who of the unemployed according to the ILO definition was also able to register as unemployed with the State Employment Service. Further, to help understand how unemployment risks change over time, the chapter compares the LFS 98 data with the LFS 96 data. Finally, because the long-term unemployed face particular difficulties, the study uses an OLS regression on the log unemployment duration to highlight the characteristics that extend unemployment spells. 3.22 While the above analysis will provide an in-depth picture of the Latvian labor market, there are several crucial aspects of that market beyond this study's scope. For example, many observers point to the importance of the informal sector when discussing economies in transition. While the set of activities referred to as the informal sector undoubtedly influence Latvian workers, for definitional and practical reasons, this report will not seek to analyze them. Most importantly, the informal sector denotes many labor market behaviors, some of which are already captured in this data. If the informal sector means secondary jobs, self-employment or legal income earning activities of pensioners, then the HBS earnings data already captures this information. If it means illegal (black or "gray" market) behavior, then respondents are unlikely to report it to BBS or LFS enumerators. More qualitative methods would be more effective in capturing this phenomenon. If the informal sector refers strictly to that work for which no employer pays social employment tax, the analysis offers only limited, indirect insight5. 5 As will be discussed in the section outlining unemployment indicators below, one of the barriers to registering for unemployment benefits is the stipulation that ones employer pay social tax for nine of the 12 months prior to becoming unemployed. Differences in the probabilities of receiving unemployment might suggest who has most recently worked but not paid social tax. 31 3.3 EARNINGS STRUCTURE 3.23 As outlined in Section 3.2 above, the chapter analyzes the structure of earnings using ordinary least squares regressions of log earnings. These regressions present summary information about how the Latviian economy values individual characteristics such as education, age, nationality and the region in which a person lives. Consistent with an extensive literature on such earnings regressions, men's earnings structures are found to differ significantly from those of women, so that men and wormen are treated separately. Table 3.2 below presents the results of this analysis. 3.24 Payments to particular characteristics. Considering human capital characteristics, the Latvian labor market rewards those with more education. The benefit to being older increases among younger respondents then decreases after respondents have reached a prime earning age. However, the labor market values men's and women's human capital differently. 3.25 Returns to Education. For example, men with higher education earn 54 percent more than those with only primary education and women earn 46 percent more, controlling for all other characteristics. In an economy that is suddenly transforming itself and seeking people educated enough to respond to changes flexibly, it is not surprising that those with higher education will be doing relatively well. While it is valuable for both men and women to be highly educated, having secondary education is valuable only for men: men enjoy a 23 percent premium, while women with secondary education have earnings statistically indistinguishable from women with primary education. With regard to earnings, vocational education offers no statistically distinguishable benefit over primary education. The specific skills that one acquired from vocational training would quickly become out-of-date in an economy seeking to adjust to transition smoothly. 3.26 Age - Earnings Profiles. For both men and women, the profile of how earnings and age are related are concave. The positive significant coefficient on age (2.3 percent for men and 4.0 percent for women) and the negative significant coefficient on age-squared together suggest that, at young ages, earnings get higher as one gets older, peak at a certain age (34.5 years for men and 38.4 years for women), then begin to decrease with advancing age. As in other countries, this concave age-earnings profile is steeper for women than men, suggesting that the Latvian labor market does not pay well waomen much younger or older than the peak earning years. This age-earnings profile likely reflects that young women in their child-bearing years do not often gain as much work, so their earnings are pushed down. 32 Table 3.2: Determinants of (Log) Earnings Ordinary Least Squares Regressions (Absolute Values for T-Statistics in Parentheses) Men Women HUMAN CAPITAL (vs. Primary) Higher Education 0.54'* 0.46** (7.62) (5.68) Secondary Education 0.23k* 0.028 (3.69) (0.38) Vocational Education 0.11 0.15 (1.32) (1.10) Age 0.023** 0.040** (2.25) (3.42) Age Squared (x 100) -0.033** -0.052** (2.83) (3.76) NATIONALITY (vs. Latvian) Russian -0.066 -0.069* (1.53) (1.64) Other Nationality -0.030 -0.094 (0.47) (1.52) MARITAL STATUS (vs. Married) Single -0.29** -0.071 (4.48) (1.14) Divorced -0.17* -0.009 (1.64) (0.17) Widowed -0.15 0.12 (0.71) (1.53) REGION (vs. Kurzeme) Riga City 0.14** -0.045 (2.37) (0.73) Riga Region 0.065 -0.021 (0.95) (0.30) Vidzeme -0.21 *t -0.15** (2.91) (2.01) Zemgale -0.03 -0.077 (0.42) (1.08) Latgale -0.26** -0.26** (3.53) (3.71) Urban 0.27'* 0.15** (5.52) (3.01) Constant 3.70** 3.34* . (16.70) (13.25) N 914 933 R-squared 0.24 0.16 *Statistically significant at the .10 level; at .05 level Data Source: Household Budget Survey. 3.27 Earnings Discrimination against those of Non-Latvian nationality. The earnings analysis also suggests that after controlling for educational differences and regions, it is possible to discern statistically significant differences only for Russian women, not other non-Latvian groups. While ethnically Russian men and women both receive approximately 7 percent lower earnings than Latvians, ceteris paribus, only for women is that statistically significant at the 90 percent confidence interval. However, this rough measure of discrimination does not take into account the ways that human 33 capital might be valued differently for Latvian's and non-Russians. The section on discrimination below explores these differences in greater detail. 3.28 Marital Status. Marital status has a significant effect on men's earnings, while it seems to not influence womnen's earnings. Compared with those who are married of similar age and education, single men can expect to receive 29 percent lower earnings and divorced men 17 percent less. 3.29 Regional Effects. These earnings regressions provide significant evidence for differences in labor markets across Latvia's regions6. These regional effects are quite distinct between men and women. Compared to rural parts of the country and controlling for all other differences in individual characteristics, earnings in urban areas are 27 percent higher for men and 15 percent higher for women. Over and above this general urban benefit, men working in Riga receive an additional 14 percent wage premium. For women, the Riga labor market offers wages statistically indistinguishable from Latvia's other urban areas. Compared to the Kurzeme region, earnings in Vidzeme and Latgale are significantly less for both men and women. For example, men and women in Latgale receive earnings 26 percent less than in Kurzeme. This is likely due to the particularly difficult economic conditions; there. 3.30 Discrimination Analysis. Using this earnings analysis, we can consider the degree to which the Latvian labor market allows discrimination between men and women. On average7, men get paid 24.8 percent more than women. However, this could be due to differences between men's and women's characteristics (such as more experience or more relevant education). If those characteristics allow men to be more productive than women, then some of this earnings differential can be explained as payments for men's higher productivity. 3.31 Oaxaca Decomposition. The literature on discrimination addresses this situation using a Oaxaca decomposition.8 In this analysis, one assumes that some earnings structure represents payments to characteristics based entirely on worker productivity. Men's and women's actual earnings are compared to the earnings that each would receive if they were paid strictly according to this non-discriminatory measure of productivity. The difference between actual and predicted earnings is attributed to discrimination (in the event that actual earnings are below predicted earnings) or favoritism (in the event that actual earnings are above predicted earnings). Because a prime difficulty is to determine the non-discriminatory earnings structure, most studies treat the men's 6 The excluded geographic categories are "rural" and "Kurzeme". The "urban" dummy captures the general effect of living in any urban area versus living in rural areas, regardless of region. The regional dummnies ("Riga Region", "Vidzeme", "Zemgale" and "Latgale") reflect earnings differences between Kurzeme and these other regions. Finally, the "Riga City" variable is an interaction term between "urban" and "Riga Region", thus capturing the extra benefit to living in Riga, over and above the general differentials for urban areas and for the Riga Region. 7 Following the discrimination literature, this gross (unadjusted) logarithmic wage differential results from the calculation of: In(WMen 7women ) where W represents the geometric mean wage. a For the first application of this technique, see Oaxaca, Ronald, 1973, "Male-Female Wage Differentials in Urban Labor Markets", Intemational Economic Review 9, 693-709. 34 earnings structure as based strictly on productivity and then treat the women's earnings structure as such. Because the "true" productivity valuation is assumed to be bracketed between these two extremes, the true index of discrimination should fall between those generated by relying either on men's or women's earnings structures. 3.32 The results of the Oaxaca decomposition on the Latvia HBS data suggest that the gross unadjusted wage differentials under-estimate the degree of gender discrimination in the economy. If men's earnings are used as the standard, because women possess more of the characteristics that the Latvian economy values, men should be paid 2.4 percent less than women. Because they are paid 24.8 percent more, by this measure discrimination in the Latvian economy is 27.2 percent. If women's earnings are used as the standard, gender discrimination is shown to be even higher. Using women's earnings structures as a standard, men should be paid 4.1 percent less than women and gender discrimination is 28.9 percent. 3.33 For comparison, Oaxaca and Ransom (1994) report a gross logarithmic wage differential between US men and women of 30 percent.9 If the men's earnings coefficients are treated as the standard for non-discrimination, the discrimination indicator is 23 percent, while if women's earnings are used, the discrimination indicator is 28 percent. Oaxaca Decompostion 100 80 E 60 _ |EActu I 20 mBn's vorren's Figure 3.1 3.34 Another way to consider this same information is to identify average earnings for men if they were paid as women and the average earnings of women if they were paid as men. Figure 3.1 presents this information. Average men's earnings are 83.2 lats per month. If men were paid as if they were women, based on their observable characteristics, they would be paid an average of 62.3 lats per month. Average women's earnings are 64.9 lats per month. If women were paid as if they were men, their average wages would be 85.2 lats per month. 9 Using data from the 1988 Current Population Survey. 35 3.35 A similar decomposition sheds light on differences between Latvian and Russian earnings. On average'0 working Russians get paid 2.4 percent more than Latvians, so there does not seem to be earnings discrimination against this group. However, given their observable human capiital characteristics of age and education, if Russians were paid in the same way as Latvians they would receive 7.9 percent more than Latvians. As a result, Russians are paid 5.5 percent less than they should be paid if only their human capital characteristics were considered according to the earnings structure of Latvians. If the Russian earnings structure is used as the standard for appropriate payments to human capital, then Russians wou]d be paid 9.7 percent more than Latvians and the degree of discrimination against them would be 7.3 percent." 3.4 LABOR FORCE PARTICIPATION 3.36 Earnings structures in Latvia are the end result of labor supply and labor demand factors. We consider each of these separately. As transition proceeds, the incentives that people face change. As a result, people will alter their decisions about how much or whether to work. Because of the ease of measurement, a promising approach to analyzing labor supply is to consider which people choose to participate in the labor force. As noted above, overall participation levels have been dropping in Latvia. This section analyzes labor force participation as a function of people's characteristics and their expected earnings. 12 Labor Force Participation 100-I 80 ;;. 0.00000 0 0 60 .--- Women .S 4 0 . . _ . . Men .2 ' 20_ <10 '?O No~ 4' go go Age Figure 3.2 '° Again, these figures are based on th,e geometric mean. " The "human capital characteristics" on which this analysis is based does not include language ability, for that information was not in the data. Analysts in Latvia suggested that were differences in language included in the analysis, discrimnination against Russians would be diminished, because Russians' inability to speak Latvian hinders their economic productivity. 12 For the purposes of this section, labor force participation is defined as those people who are either employed or unemployed according to the ILO definition of unemployment. 36 3.37 To consider descriptive statistics first, 68 percent of men and 51 percent of women chose to participate in the labor force in 1998. Compared to 1996, fewer men and women are interested in working, for the rates were 72 percent and 54 percent at that time. The analysis below will explore the factors contributing to this decline. One of the prime characteristics determining who participates is individuals' age differences: at different times in ones life, various circumstances make it attractive or not to supply labor. 3.38 Figure 3.2 presents the age profile of participation for men and women in Latvia. Both men's and women's participation rates are low among those under age 25, as men and women are in school, having children, in military service, or have found it too discouraging to enter the labor force because of low wages and high unemployment rates. Participation rates then increase among those of prime working age between 25 and 55 years old, though women of age 25 to 35 enter the labor force in smaller proportions than men, often because they are having children. After 55 women's participation rates drop more rapidly than men's, perhaps because of different retirement ages for women, or because grandmothers leave the labor force to take care of their grandchildren so their children can continue working. Finally, participation rates drop as people retire after age 65. 3.39 Deterninants of Labor Force Participation. While these descriptive statistics provide some preliminary information about labor force participation, a more complete analysis must rely on a probit analysis of the factors that determine the likelihood that people will participate. Table 3.3 presents probit estimates for labor force participation in 1998 and 1996, separating the analysis for men and women. In general these probits suggest that as expected earnings increase, the likelihood that a person will participate in the labor force also increases, i.e. that labor supply elasticity is positive with respect to earnings. The structure of coefficients for men and women is roughly equivalent to that for earnings in Table 3.2. 3.40 Human Capital. Higher education leads to a greater likelihood that people will participate in the labor force. For example, compared to a man with only primary education, a person with higher education is 20 percent more likely to participate, while a woman is 34 percent more likely. These education effects appear to have remained roughly equivalent between 1996 and 1998, for the education coefficients are roughly equivalent for men and women in the two time periods. As Figure 3.2 suggests, the age participation profile is concave, increasing until approximately the age of 40 and then dropping at an increasing rate. 3.41 Nationality. Between 1996 and 1998, nationality has begun to have an effect on who chooses to participate. In 1996 after controlling for human capital and region, non- Latvians were no less likely to participate in the labor force than Latvians. By 1998, however, non-Latvian women are 9 percent less likely to participate and Russian men are 3 percent less likely to participate than their Latvian counterparts. Because there are few differences in other coefficients between 1996 and 1998, these changes in the effects of nationality are likely to drive the decreasing labor force participation rates observed in Latvia as non-Latvians decide to leave the labor force. 37 3.42 Geographic Differenices. For both men and women in both time periods, geographic location has very little effect on whether a person participates in the labor force, for nearly all the geographic indicators are statistically insignificant. The one exception is Latgale, where after correcting for all other characteristics in 1998, men are 11 percent less likely to participate in the labor force and women are 8 percent less likely. Given the degree of economic difficulty in this region, this negative participation component is likely to indicate a high proportion of discouraged workers. 3.43 Predicted Earnings. The central two columns of Table 3.3 also concern the participation rate using the 1998 LFS data. However, this specification includes predicted earnings as an explanatory variable, thus allowing some direct analysis of the earnings elasticity of labor supply. These earnings were predicted based on the regression coefficients presented in Table 3.2. As one would expect if higher earnings led to a greater likelihood of participation, the coefficient on log earnings for men is positive and significant, suggesting that a one percent increase in earnings leads to a 0.2 percentage-point increase in the probability that a man will participate in the labor force. However, the coefficient for women is not statistically significant. This suggests that women's participation decisions are not primarily driven by the earnings that they could potentially earn in the labor force, but rather by other factors. Those other factors might concern family obligations, such as children. 3.5 LABOR DEMAND 3.44 While understanding labor supply is central to labor market analysis, it is also important to consider labor demand. The number of jobs available to workers affects wages, unemployment, and people's expectations about their employment prospects. This is particularly true in economies undergoing post-communist transition, for economic restructuring has enormous implications for which firms remain viable and can maintain (or potentially increase) their demand for labor. However, it is difficult to combine a study of labor supply and demand, for one focuses on workers for the former and firms for the latter. wages, unemployment, and people's expectations about their employment prospects. 38 Table 3.3: Determinants of the Probability of Participating in the Labor Force Probit Estimates by Gender and Year (Z-scores in Parentheses) (a) Participation (b) Participation (c) Participation 1998 1998 1996 Men Women Men Women Men Women HUMAN CAPITAL (vs. Primary) Higher Education 0.20** 0.34** 0.11** 0.32** 0.16** 0.31'- (8.97) (14.62) (3.79) (9.32) (7.44) (11.86) Secondary Specialized 0.14*' 0.25** 0.11*' 0.25'' 0.13*' 0.24*' (7.21) (12.24) (5.22) (12.16) (7.10) (10.80) Technical Secondary 0.14** 0.26*' 0.10" 0.23" 0.15* 0.16^' (6.93) (9.77) (4.53) (8.76) (7.02) (5.24) Comprehensive Secondary 0.056*' 0.17** 0.006 0.16** 0.080'- 0.17*' (3.15) (9.10) (0.27) (8.36) (4.26) (7.62) Vocational Education 0.1 1* 0.19** 0.095* 0.18'' 0.10'' 0.15'' (3.98) (4.23) (3.4) (3.95) (3.59) (2.57) Age 0.081 '' 0.078** 0.083* 0.081 ** 0.058'' 0.078'' (30.13) (27.96) (32.21) (24.48) (21.16) (23.68) Age Squared (xlOO) -0.100 -0.098** -0.10'* -0.10** -0.076** -0.10** (33.63) (31.46) (33.45) (25.46) (24.81) (26.79) PREDICTED EARNINGS (log) 0.20** -0.020 (7.51) (0.37) NATIONALITY (vs. Latvian) Russian -0.032** -0.086*' -0.017 -0.012 (2.04) (5.66) (1.04) (0.67) Other Nationality -0.013 -0.089*- -0.034 0.018 (0.59) (4.19) (1.52) (0.75) MARITAL STATUS (vs. Married) Single -0.18'* -0.050'' -0.14'' 0.038 (9.15) (2.55) (6.33) (1.56) Divorced --0.17*' 0.077*- -0.15** 0.079** (6.25) (3.56) (4.96) (3.15) Widowed -0.160 -0.007 -0.160 0.058** (3.77) (0.29) (3.80) (2.12) REGION (vs. Kurzeme) Riga City 0.025 -0.029 0.00 0.01 (1.13) (1.28) (0.03) (0.51) Riga Region 0.005 -0.012 -0.017 0.011 (0.19) (0.45) (0.62) (0.36) Vidzeme -0.024 -0.014 -0.019 0.030 (0.94) (0.54) (0.70) (0.98) Zemgale 0.002 0.001 -0.047' 0.010 (0.08) (0.03) (1.70) (0.33) Latgaie -0.11'' -0.078' * -0.11 '^ -0.051' (4.30) (3.06) (3.77) (1.72) Urban -0.023 -0.092 0.015 0.011 (1.39) (5.37) (0.87) (0.57) Number of Observations 6844 8304 6844 8304 5265 6221 Log-Likelihood -2643.3 -3681.3 -2695.2 -3751.5 -2030.8 -2766.4 Observed Probability 0.681 0.514 0.682 0.514 0.721 0.535 Predicted Probability 0.722 0.437 0.716 0.436 0.763 0.463 *Statistically significant at the .10 level; '' statistically significant at .05 level Data Source: Labor Force Surveys. 39 3.45 Using a survey of Latvian firms, this study uses indicators of labor demand for each of Latvia's 33 administrative districts. The enterprise surveys'3 include infornation about the total number of workers which firms employ.14 With data on total jobs over time, we gain some insight into how labor demand has changed in each of the districts. To consider longer term changes in labor demand, the study evaluates the percentage change in jobs between 1989 and both 1998 and 1996. This indicator gives insight into long-term changes in employment and labor demand in each of Latvia's administrative districts. Yearly data also a]llows us to explore how labor demand has changed in the short term by comparing employment changes over the previous year. Using information from the State Employment: Bureau about vacancies in each district, we generate vacancy-employment ratios fcr each district. These indicate how many positions firms are seeking to fill beyond those already occupied. 3.46 Table 3.4 presents summary statistics for each of these labor demand measures. While each indicator is calculated for Latvia's 33 administrative districts, for clarity of presentation, the table includes information only about the country as a whole and its principal regional groupings. ln recent years firms have begun to employ more workers. In the short-term between 1997 and 1998, labor demand for all of Latvia increased 3.2 percent. However, much of that increase is concentrated in the Riga Region, where job growth was 5.6 percent. In Latvia's other regions, the growth in the year prior to 1998 was less than 2 percent. The employment change was not nearly as large two years earlier in 1996, when short-term labor demand grew 0.3 percent across the country. At that time, short-term employment was growing most at 1.2 percent in the Riga and Latgale regions and declined 2 percent over the previous year in Zemgale. 3.47 Considering long-term changes in the number of workers firms employed, the next two columns point to large reductions in labor demand between 1989 and the second half of the 1990s. For Latvia as a whole, there was a 45 percent drop in employment up to 1998, though some regions suffered more than others. For example, firm's demand for labor fell relatively less in the Riga Region (37 percent) than in Zemgale (54 percent). Declines in Riga City (35 percent) were less than outside of Riga (43 percent). 3.48 If changes in employment point to a brighter picture for Riga versus other of Latvia's regions, vacancy-employment ratios reinforce that image. Across the country, 13 1 would like to thank Mr. Ervins Rekke of the Central Statistics Bureau for providing me with yearly data about employment by administrative district. 14 Because the labor demand indicators used here are based on firm's records, they will not capture non-firm demand for workers, such as that arising from agricultural work or self-employment. In addition, the enterprise surveys do not cover very small firms, so that these measures are biased toward labor demand from larger employers. 40 Table 3.4: Labor Demand by Region Short-Term Short-Term Long-Term Long-Term Vacancy/ Vacancy/ Job Growth Job Growth Job Growth Job Growth Employment Employment '97-'98 '95-'96 '89-'98 '89-'96 1998 1996 Latvia 3.2% 0.3% -45.4% -47.8% 0.39% 0.28% Riga Region 5.6% 1.2% -36.5% -41.7% 0.57% 0.43% Riga City 5.3% 1.1% -35.2% -40.9% 0.65% 0.50% Riga Region w/o Riga 7.8% 1.6% -43.4% -46.2% 0.10% 0.06% Vidzeme Region 1.6% 0.0% -52.2% -53.2% 0.23% 0.19% Kurzeme Region -0.2% -1.0% -49.5% -50.0% 0.10% 0.10% Zerngale Region 1.2% -2.0% -53.9% -52.8% 0.27% 0.19% Latgale Region 0.9% 1.2% -51.2% -51.7% 0.23% 0.08% Source: Enterprise Surveys and SES Vacancy Summaries for every 1000 people employed, there were 4 vacancies in 1998 and 3 in 1996. Again, there is some variation from this country-wide average: in the city of Riga, there were 6.5 vacancies per thousand jobs, while in Kurzeme there was only 1 vacancy in 1998. For all regions but Kurzeme, vacancy-employment ratios increased between 1996 and 1998. 3.6 UNEMPLOYMENT 3.49 Unemployment, made up of people who want to work but cannot find a job, indicates labor market failures to match workers' supply and employers' demand. By analyzing unemployment rates and the composition of the unemployment pool, we gain insight into which worker characteristics are in greater supply than demand. Labor market policies can then seek to facilitate the matching of labor supply and labor demand, target people with those characteristics to decrease their labor supply, enhance demand for people with those characteristics, or alter the characteristics of those particularly likely to be unemployed. 3.50 ILO Standard Unemployment Definition. As noted in Section 3.2, so that the unemployment rate adequately presents this concept of the difference between labor demand and labor supply, it is necessary to define unemployed people carefully. According to the International Labor Organization an unemployed person is not currently working but has actively looked for work in the last two weeks. In any survey it is difficult to sort all respondents into categories of the employed, the unemployed, and out of the labor force. The distinctions between those categories are often vague, and people involved in informal or illegal activities will naturally be reluctant to disclose their true work patterns. Despite these difficulties, because it provides the best indicator of how well a labor market allocates jobs for people who want them, this study will focus primarily on this international standard definition of unemployment. 15 In fact, there are several subtleties in the unemployment definition depending on how one defines looking for work, currently working and labor force participation. The United States Bureau of Labor Statistics defines several different degrees of unemployment ranging from a very restrictive definition (Ut) to a broader definition (U6). 41 3.51 The first table of Appendix E summarized the variables from the 1998 Labor Force Survey used to partition the sample of respondents into "working", "unemployed" and "out of the labor force". From that table describing the composition of the ILO standard unemployment definition, 7597 respondents are working and 1140 are unemployed, so the labor force sample is 8737." The third table of Appendix E includes the composition of unemployment according to the ILO standard using the 1996 LFS data. In this data, 5591 respondents work and :1516 are unemployed, so the labor force sample consists of 7107 people. 3.52 Definition Based on the Registered Unemployed. An alternative approach used officially by the Government of the Republic of Latvia defines the unemployed as those who have registered as unernployed with local employment offices. According to the Law on Employment, in Latvia those who are granted official unemployment status must: * Be citizens of the Republic of Latvia or residents who have a permanent residence permit and a stamp of the population register in their passports; - Be of working age; * Be able to work; - Not receive any salary or incomes of any kind of at least the size of the minimum wage; * Not undertake any business activities; * Be looking for a job; • Be registered with the state employment service associated with his or her place of permanent residence; and * At least once a month, visit the state employment service. 3.53 While many of its elements overlap with the ILO standard, under this definition, a person must register with the state employment service to gain unemployment status. These unemployment offices nrgister those who: * Have a stamp in their passports certifying that they live in the jurisdiction of the employment office's; * Have received back work- and tax-book from their last employer implying they no longer hold a job; and * Worked for an employer who paid social tax for employees for nine of the previous 12 months. 3.54 As a result of these regulations, there are several groups of people unable to register as unemployed, even though they meet the ILO standard unemployment definition. For example, this definition of registration would exclude: 16 Note that these unemployment figures do not weight observations according to their sampling probability, so that it is not representative of the population as a whole. In the analysis that follows, however, the data is weighted to represent the population. 42 * Those with disputed Latvian citizenship, a situation arising given ambiguity about the rights of ethnic non-Latvians or non-Latvian speakers; * Those who have moved to a region of Latvia other than where their passports say they live, such as those who leave their homes in search of work; * Those whose last employer faced financial difficulties before laying workers off, for those employers might be unable to pay social tax regularly in the 12 months prior to laying off workers; * Those whose last job was as a self-employed worker or in a small entrepreneurial firm that would not issue labor registration documents; * Those choosing not to visit the state employment service once a month, perhaps because its benefits or services are unattractive; or * Those unaware of the requirements necessary to register as unemployed. 3.55 Despite these drawbacks as a measure of labor market functioning, this registered unemployment definition is based on information easier to obtain than the ELO definition. Offices administering unemployment benefits programs are naturally repositories for registration data. Because of the prohibitive costs of conducting labor force surveys to measure unemployment according to ILO standards, many countries early in transition from central-planning adopted this registered unemployed approach. Having collected and reported such data for several years, they hesitate to switch standards, both because of the survey costs and the potential that the ILO definitions will present a more dire picture of the labor market. Despite having collected regular labor force survey data that would allow it to use this ILO standard definition and tracking its LO standard unemployment figures, the official statistics of the Republic of Latvia are based upon this registered unemployed definition. 3.56 Because the Republic of Latvia uses this registration-based definition, this study also partitions the sample of respondents into "working", "registered unemployed" and "out of the labor force" based on this administrative approach. This alternative definition leads to very different evidence. The second and fourth tables of Appendix I describe the composition of the registered unemployed definition. Using the 1998 LFS, 7612 respondents worked, and 561 did not work and registered as unemployed. The labor force sample thus consists of 8173 respondents and the unemployment rate is 7 percent. It is worth noting that among those reporting that they had a main job, 221 people also said that they had registered as unemployed, a fact which points to difficulties in this administrative definition. Using the 1996 LFS data, 5620 people reported themselves working (94 of these also registered as unemployed), and 442 registered as unemployed and were not working. The labor force sample thus is 6062 respondents and the unemployment rate is again 7 percent. 3.57 Descriptive Statistics. Cross-tabulations overview unemployment rates across groups. Table 3.5 presents unemployment rates for the ILO definition using the LFS 98 data, that based on those who registered as unemployed using the LFS 98 data, and the ILO definition using the 1996 data. 43 3.58 As noted above, there is a large difference between unemployment rates when measured with the ILO stancdard (14.5 percent) and when measured by unemployment registration (6.9 percent). According to these definitions, there was a large drop in the ILO standard unemployment rate between 1996 and 1998: in 1996 the overall unemployment rate was 22.3 ,percent. Given the decreases in labor supply and increases in labor demand discussed above, the fact that unemployment fell is not surprising, though the magnitude of its decline is impressive. 3.59 Further, these descriptive statistics suggest that men have higher unemployment rates than women (15.6 percent versus 13.3 percent), though according to the registration-based unemployment definition, women (7.6 percent) have higher unemployment rates than men (6.2 percent). If men are more likely to move to seek employment in other areas, to become affiliated with small entrepreneurial enterprises, or to receive fewer benefits from visiting state employment bureaus, they have more difficulty registering with the state employment service. As a result, we would expect men to have higher unemployment rates than the "registered" definition suggests. Table 3.5: Unemployment Rates by Gender, Unemployment Definition, and Year UE (ILO) 1998 UE (Reg.) 1998 UE (ILO) 1996 LATVIA 14.5% 6.9% 22.6% Men 15.6% 6.2% 23.0% Women 13.3% 7.6% 22.1% NATIONALITY Latvian 10.8% 6.0% 17.4% Russian 21.0% 8.5% 29.2% GEOGRAPHIC AREA Urban 17.5% 7.6% 25.1% Riga City 17.6% 5.3% 25.0% Riga Region 11.7% 3.3% 20.5% Kurzeme 11.3% 5.8% 19.9% Vidzeme 11.0% 7.0% 18.4% Zemgale 12.3% 6.6% 21.7% Latgale 18.5% 16.0% 26.1% AGE GROUP <25 26.5% 11.8% 33.7% 25-35 12.6% 6.2% 21.4% 35-45 12.9% 6.4% 18.6% 45-55 14.3% 7.3% 18.8% 55-65 10.9% 5.1% 24.6% Data Source: Labor Force Survey. 3.60 According to both definitions, Russians have a higher unemployment rate: 21.0 percent versus 10.8 percent for the ILO standard and 8.5 percent versus 6.0 percent for the registration-based definition. However, the Russian unemployment rate is proportionately larger when the ILO standard definition is used. If many Russians have 44 ambiguous citizenship status or previously worked in firms that did not pay social tax, we would expect the registration-based definition to understate their unemployment. 3.61 With regard to the distribution of unemployment, the ILO definition shows urban unemployment to be particularly high, 17.5 percent versus the national average of 14.5 percent. Interestingly, the city of Riga has a higher ILO unemployment rate in 1998, 17.6 percent, than the urban average; according to the registered unemployment definition, Riga's rate of 5.3 percent is lower than the national average of 6.9 percent. If many people are moving to Riga in hopes of finding jobs but are unable to register as residents of Riga, registered rates for Riga would be lower than the true unemployment rate. Seeking the higher labor demand that Riga offers, these economic migrants to Riga are not able to receive benefits and are hidden in the figures the Republic of Latvia reports. 3.62 Outside of Riga, each of the regions but Latgale have roughly similar unemployment rates. As a region, the unemployment rate for Latgale is particularly high according to both definitions, 18.5 percent if measured by the ILO standard and 16.0 percent if measured according to a registration-based definition. However, according to the ILO standard, the Latgale unemployment rate does not differ as much from other regions as it does according to the registered unemployment rate. One possible explanation for the high registration rate in Latgale is that firms there were able to pay their social taxes before laying workers off, so that many can register with the state employment service. 3.63 As in many different country contexts, youth unemployment rates are particularly high in Latvia, nearly twice as high as the rates for the society as a whole under both definitions (for example, unemployment is 26.5 percent among people under 25). As Figure 3.3 illustrates, the age profile of unemployment in 1998 is roughly equivalent for men and women, though among older men, the unemployment rates becomes notably higher. Among the 55-65 age group, women's unemployment is much lower than men's, which is likely a result of women being able to retire at a younger age than men, so they are leaving the labor force rather than becoming unemployed. This explanation is borne out in the descriptive statistics for labor force participation presented in Figure 3.2, where women's participation in the 55-65 age group is markedly less than men's. Unemployment Rates by Age a30 ,,25 - 20 ^Worren E 15 l n : <25 25-35 35-45 45-55 55-65 >65 Age Figure 3.3 45 3.64 Risk Factors for Unemployment. The descriptive statistics above indicate the risk factors for unemployment, for unemployment rates are higher among Russians, among the young, and among people living in certain regions (particularly Latgale). However, they do not allow us to decompose carefully the individual characteristics that lead one to be more likely to be unemployed. For example, the Latgale unemployment rate could be particularly high because that region contains a high percentage of Russians, or potentially more young people. To distinguish between these separate effects of individual characteristics, we again use probit analysis. This allows us to determine the marginal effects of different characteristics on the probability that one will be unemployed, controlling for all the other characteristics present for an individual. As such it isolates aspects that unemployment policies should focus on. 3.65 Table 3.6 presents probit analyses of the risk factors for unemployment. To allow comparison between the ILO standard definition and a definition based on who registered as unemployed, it includes siimilar probits for both definitions. To consider whether the risk factors for unemployment have changed across time, it also includes analysis based on the LFS 96 data. Though not reported here, after controlling for the factors in Table 6, overall unemployment rates for men do not differ greatly from those for women: men are found to be 2.3 percent more likely to be unemployed. Considering gender differences in the likelihood of registering fDr unemployment benefits, men are 1.4 percent less likely to register. However, while overall unemployment rates seem similar between men and women, the risk factors for men and women do differ significantly. Thus, Table 6 includes separate probit analyses for men and women. 3.66 Comparisons Across Samples. When comparing the marginal effects of different characteristics, the common practice with probit analysis is to present the amount that the probability of an outcome, becoming unemployed in this case, changes with a change in characteristics. However, these marginal effects need to be reported with respect to some starting-point probability. Again, convention suggests using the probability estimates for an "average" person as a starting point for these marginal effects, where an average person has characteristics equal to the sample mean. However, it is then difficult to compare marginal effects of characteristics across different samples. Because samples have different probabilities at their means, slopes are evaluated at different starting points on a non-linear cumulative density function and cannot be compared easily. To ease comparison of marginal probaLbilities, all the marginal effects across different samples in Table 3.6 are evaluated at the same probability. The standard comparison probability for all columns is that for men in the LFS 98 of 13.5 percent. 3.67 1998 ILO Standard Definition: Human Capital. In general, education reduces the risk of unemployment for both men and women, though the deterrent effect is larger for the former. Compared to those with primary education and controlling for age, nationality and region, in the LFS 98 data the probability of being unemployed is 14 percentage-points lower if a man has higher education, 12 points lower with secondary specialized education, 8 points lower with technical secondary education, 7 points lower with comprehensive secondary education, and 6 points lower with vocational education. 46 All of these differences are statistically significant. For women, only higher and secondary specialized education reduce the risk of being unemployed vis a vis primary education: higher education reduces the risk of unemployment 11 percentage-points and secondary specialized lowers it by 7 points. As in other countries, more highly educated people are less likely to become unemployed. 3.68 As men and women age, their probability of becoming unemployed decreases: coefficients on age are negative and significant for both sub-samples. For young men an extra year of age decreases the probability of becoming unemployed by 0.5 percentage- point; for young women it decreases the probability by 0.9 percentage-points. For women, the positive significant coefficient on age-squared indicates that the age- unemployment profile is convex: the effect of age on decreasing a woman's probability of becoming unemployed diminishes with advancing age, so that at age 70, age has no extra effect. 3.69 Nationality. Controlling for education and region, those of non-Latvian nationality are significantly more likely to be unemployed. On average, being Russian entails a 7 percentage-point increase in the risk of being unemployed for men and a 10 percentage- point increase for women. Those non-Latvian men of nationality other than Russian are 4 percentage-points more likely to be unemployed and non-Latvian women are 6 points more likely. 3.70 Labor Demand. Table 6 also includes data on labor demand by administrative district as described in the labor demand section above. The long-term job growth from 1989 to 1998 for each administrative district has a significant effect on men's unemployment. For men living in districts with larger labor demand -- or those districts where the long-term decrease in employment was smaller -- the risk of unemployment is smaller, as demonstrated by the negative significant coefficient on employment change. The vacancy-employment ratio for 1998 has no significant effect on the probability of becoming unemployed for either men or women, which is perhaps explained by measurement problems of true vacancies. 3.71 Region. After correcting for human capital, nationality and labor demand differences, regional unemployment rates are not particularly large. Men living in urban areas are 12 percent more likely to be unemployed than those in rural areas, while women are 9 percent more likely. Beyond this overall urban increase, the city of Riga has a statistically different unemployment rate: men living in Riga are 5 percentage-points more likely to be unemployed, while women face an 8 percentage-point higher unemployment risk living there. Latgale also has an unemployment rate significantly different than Kurzeme, where unemployment rates are 9 percentage-points higher for men and 5 points higher for women. 47 3.72 Predicted Wages. Though not reported in Table 6, a separate probit analysis was run including predicted wages that people would expect to receive.'7 From that analysis, the coefficient on predicted (log) wage is negative and significant for both men (-0.33) and women (-0.19). We can interpret these findings as showing that as potential earnings increase one percent, men's risk of being unemployed falls 0.32 percent and women's risk fall 0.19 percent. In general, this suggests that those with lower potential earnings are more likely to be unemployed. 3.73 Probability of RegisiLering for Unemployment Benefits. As mentioned above, unemployment rates as measured by who chooses to register are half as large as actual unemployment rates. In addition, this measure of unemployment identifies different characteristics as risks for becoming unemployed, which suggests it measures a different concept than how well the labor market matches supply and demand for particular characteristics. Men and women with higher education are less likely to register for unemployment benefits. For both men and women, the marginal effect of education on registering for unemployment is approximately equal to the marginal effects on actual unemployment. Thus, there is little education-based bias to the registration process. Unlike the ILO definition, there is no significant negative relationship between age and who registers, which indicates a flat age-registration profile, as opposed to the downward-sloping profile of unemployment and age. 3.74 Geographic differences in registration are more pronounced than for true (ILO) unemployment. Both men and women in urban areas are 9 and 6 percentage-points more likely to register, so the marginal effects of urban areas are less pronounced on registration. However, after controlling for all other variables for men registration is 10 percentage-points higher in Riga City than in other urban areas; for women, registration in Riga is no more likely tha[n in other urban areas. Thus, the low registration rates for Riga reported in Table 3.5 must result from the high concentration of well-educated men and the high labor demand in the capital city, factors corrected for in Table 3.6. Compared with Kurzeme, registration is significantly higher in Latgale and Vidzeme for men and women and higher in Zemgale for men. For example, men in Latgale are 23 percentage-points more likely to register than equivalent men in Kurzeme. 3.75 It is particularly interesting that nationality does not have as large an effect on registration as it does on unernployment. Russian and other non-Latvian men are no less likely to register for unemplo,yment benefits than Latvian men, even though, as seen from the left column, they are more likely to be unemployed. Russian women and women of other nationalities are more likely than Latvians to register after controlling for human capital, marital status, and region, though the marginal effects of nationality on registration are less than on unemployment. Using those who register for benefits as a measure of unemployment underestimates the true effect of nationality, particularly for men. 17 This probit needs to include only restricted independent variables. If a probit analysis were run including all the variables listed in Table 6 plus predicted wages, there would be an identification problem, for wages were predicted based on the independent variables included in Table 6. 48 3.76 Changes between 1998 and 1996. As mentioned above, the ILO standard unemployment rate measured in 1996 was much higher than in 1998, e.g., 23 versus 16 percent for men and 22 versus 13 percent for women. Summarized in the right-most columns of Table 3.6, different risk factors for unemployment generate this overall change in level. Again, those men and women who had more advanced education were much less likely to be unemployed, as evidenced by the negative and significant coefficients on all types of education other than vocational. For men, the marginal effects of education are roughly of the same magnitude across the two years. For women, education coefficients on education were higher in the former. For example, comprehensive secondary education led to a 5 percentage-point decrease in the risk of unemployment in was a much better deterrent to unemployment in 1996 than in 1998, for the negative 1996, though it had no effect in 1998. However, the magnitude of the deterrent effects of education differed between 1996 and 1998. As in 1998, non- Latvian's unemployment rates were significantly larger than Latvians, though the degree that ethnicity influences -unemployment is not very different between 1996 and 1998. 49 Table 3.6: Determinants of the Probability of Being Unemployed Probit Estimates by Gender, Unemployment Definition, and Year (Z-scores in Parentheses) UE (ILO) 1998 UE (Registered) 1998 UE (ILO) 1996 Men Women Men Women Men Women HUMAN CAPITAL (vs. 0 14,694 1,984 12,710 Income 1,100,363 131,203 969,160 Poverty gap 32,551 32,551 Social assistance as percentage of: Expenditures 0.29 0.81 0.24 Expenditure of those with SA>0 19.6 33.8 17.4 Income 0.26 0.51 0.23 Poverty gap 8.9 2.0 Social assistance per recipient 26.2 ($45) 25.4 ($44) 26.4 ($46) household (LVL/ $ p.m.) Expenditure per capita of throse with $81 $36 $98 social assistance ($ p.m.) a/ Memo: Expenditure per capita (overall $92 $39 $126 average $ p.m.) a/ Average HH size (overall avierage) 2.36 3.13 2.23 Notes: In October 1997 prices. Exchange rate: LVL 0.58=$1. p.m. = per month. SA=social assistance. HH=household. a/ Mean across households. amounts, the importance of social assistance for the recipient households was substantial: it covered one-third of expenditures of poor households and 17 percent of the non-poor (Table 4.4). 4.18 The big difference between the share of social assistance in overall expenditures (0.3 percent) and in the expenditures of recipients (almost 20 percent) indicates that social assistance was distributed in relatively large chunks and to a few people. And indeed, the average recipient household received almost $45 as against an average unemployment benefit of $60 pml3, or average wage of slightly over $200 pm. As we shall in the next Section, Latvia's social assistance can be considered "concentrated." 4.19 The poor who do not receive social assistance. 98 percent of the poor received no social assistance. The percentage of the excluded (the poor who do not receive social assistance) does not vary with welfare: as one moves toward the less poor the percentage of exclusion stays about the same (see Figure 4.6). 3 Both calculated from the Survey. There are 2.6 percent of all households who are receiving unemployment benefits. 70 Figure 4.4: Latvia: Social assistance received as percentage of the poverty gap by level of welfare (expenditure per capita) 10 0 0.~~~~~~~~~~~~~~ -J 4.6 o o/0 Q 0 0 0 0 10 15 acc. to tothhx_p Note: Mean calculated across households. 4.3 PERFORMANCE OF LATVIA'S SOCIAL ASSISTANCE: COMPARISON WITH OTHER TRANSITON COUNTRIES 4.20 Features of the system. Using the approach from Braithwaite, Grootaert and Milanovic (1999), we compare Latvia's social assistance to the social assistance systems of five transition countries (Bulgaria, Hungary, Estonia, Poland and Russia). We note first an exceptionally modest level of social assistance. Fewer households (1.5 percent) receive social assistance in Latvia than in any of the other five countries (Table 4.5). Social assistance finances less of household expenditures (0.29 percent) than in any country save Bulgaria. If we compare Estonia and Latvia, whose systems are, as we shall see below, similar, the percentage of households receiving assistance is almost two times as large in Estonia, and the importance of assistance in relation to population expenditures is greater. 4.21 But while Latvia's social assistance is extremely modest in its size, it is concentrated: those who receive social assistance, get in Latvia (in dollar terms) more than elsewhere, except in Poland. Further, social assistance covers almost 20 percent of recipient households expenditures, again a proportion higher than in any other country except Poland. 71 Figure 4.5: Latvia: Social assistance as percentage of expenditures by level of welfare C) 0 o 0O 0 0 C a 0 0 o 0~~~~~~~~~~~~~ 0 0~~~~~~~~~~ 30~~~~ '0 0 ,T0t,;.- . 0 o0 00 a 0300 C o- o o ° oo oD X ODO OM- C OD 0 15 100 acc. to tothhx_p Note: Mean calculated across households. 4.22 Performance of the system. How does Latvia's system perform compared to other countries? In order to make this comparison meaningful, we cannot base it on different poverty lines: the very fact that a country might have a low or a high poverty line (compared to its mean expenditures) will influence the calculated efficiency of the system. For example, if the poverty line is very low, the "eligible" population will be small, many poor may receive social assistance ("the error of exclusion" will also be small), and much of the poverty gap may be eliminated (thus showing high effectiveness too). The country may seem to perform very well but most of it may be due to a very austere poverty line which severely limits eligibility for assistance. If the poverty line were raised, the error of exclusion may increase and the coverage of the poverty gap may decline, but in reality the poor would be better off. Therefore, in order to compare different countries, we need to assume that the objective of the social assistance system in each country is the same. As in Braithwaite, Grootaert and Milanovic (1999), we assume that the poor in each country are the bottom ten percent of the population14 and that the objective of social assistance is to help them. The success of the social assistance system is then measured by how muchL of the (pre-assistance) poverty gap of the bottom decile is eliminated (effectiveness), and how much of disbursed social assistance is received by them (efficiency). 14 Ranked according to expenditures per capita. 72 Figure 4.6: Latvia: Failure to deliver social assistance (percentage of the poor who do not receive social assistance) 100- o 90 LJ 70 - 0 15 acc. to tothhx_p 4.23 Consider lines 6 and 7 (Table 4.5), and Table 4.6. Latvia's results are poor. Less than 15 percent of social assistance is received by the poorest decile, a proportion inferior to that of any country except Russia. Since social assistance is badly targeted, and total amount of spending is small, it is not surprising that social assistance covers only 2.9 percent of the poverty gap of the bottom decile the smallest proportion of all countries except for Bulgaria. 4.24 Table 4.6 complements these results with several additional statistics. We define relative effectiveness as the ratio between effectiveness, and social assistance shown as percentage of total expenditures. Here again, Latvia performs worse than all countries except Russia. The correlation between social assistance and household percentile, and the social assistance concentration coefficient, both of which we expect to be negative, are, on the contrary, positive, indicating an absence of a focus on the poor. Similar results obtain only in Russia, which according to all indicators of performance scores the worst. 4.25 One of the objectives in Braithwaite, Grootaert, Milanovic (1999) analysis was to determine the type of social assistance system exhibited by a country. It was done using three indicators: the level of the poverty line (compared to mean country expenditures), percentage of recipients of social assistance, and the importance of social assistance for the recipient households. The characteristics of Latvia's system are similar to those of Poland and Estonia: small percentage of recipients, but high importance of social assistance for those who get it (Table 4.7). This implies that Latvia's social assistance is concentrated, although its focus on the bottom decile is weak (Table 4.8). 73 Table 4.5: Characteristics and performance of social assistance systems Bulgaria Estonia Latvia Poland Russia Hungary System characteristics (1) % of HHs 2.55 2.7 1.5 3.7 13.0 24.4 receiving SA (2) SA as % of 0.11 0.38 0.29 0.74 0.45 1.1 expenditures (3) SA per 10 33 45 54 5 17 recipient HH ($ pm) (4) SA as % of 4.1 14.8 19.6 22.1 3.5 4.7 expend. Of recipients HHs (5) Eligibility 28 :39 53 77 65 55 threshold as % of mean per capita expenditures _ System performance (6) % of SA 22.3 34.7 14.8 20.5 8.2 27.2 received by the lowest decile __ (7) SA to the 1.3 7.0 2.9 9.4 3.3 28.8 bottom decile as % of the poverty gap a/ _ Overall expenditures and distribution (8) Poverty gap 1.9 2.1 1.5 1.6 1.0 1.1 of the lowest decile as % of all expenditures _ (9) Memo: 83 (67) 74 (71) 107 (113) 93 (99) 47 (32) 134 (128) Overall expenditure (income) per capita in $ pm b/ (10) Gini 28.6 (31.4) 30.7 (35.4) 34.1 (33.5) 27.4 (29.1) 40.1 (44.5) 22.8 (21.8) coefficient of expenditures (income) per capita (individual based) _ a/ Poverty gap of the lowest decile. The poverty gap is expenditure-based (after social assistance). b/ Household-weighted. Note: Countries ranked from left to right according to the percentage of households who are receiving social assistance. SA=social assistance. HH=household. Source: all countries but Latvia from Braithwaite, Grootaert and Milanovic (1999). 74 Table 4.6: Comparing the performance of the social assistance systems Hungary Estonia Poland Bulgaria Latvia Russia Efficiency: % of SA received by the 27.2 34.7 20.5 22.3 14.8 8.2 lowest decile Effectiveness: SA as % of the poverty 28.8 7.0 9.4 1.3 2.9 3.3 gap of the lowest decile Relative effectiveness 26.2 18.3 12.6 11.4 10.7 7.3 Correlation btw. SAPC and perc_h -0.13 -0.04 -0.06 -0.03 +0.01 +0.03 Concentration coefficient a/ -25.8 -16.2 -19.8 -13.8 +5.2 +8.2 Note: Relative effectiveness is calculated as the ratio between effectiveness, and social assistance as percentage of total expenditures. Countries are ranked from left to right according to relative effectiveness. SA=social assistance. SAPC=social assistance per capita. perc_h=percentiles of households formed according the household per capita expenditures. Table 4.7: Characteristics of the systems Poland Bulgaria Hungary Estonia Latvia Russia Poverty line High Low High Low High High Percentage of recipients Low Low High Low Low High Importance of SA for High Low Low High High Low recipients Type of system HLH LLL HHL LLH HLH HHL HI/L=level of poverty line: high/low (over/under 50 percent of average expenditures). HI/L=many or few receive SA (under/over 10 percent of the population). H/ILSocial assistance (SA) is important (high) or not (low) (under/over 10 percent of recipients' expenditures). Table 4.8: Taxonomy of social assistance: concentrated, dispersed, and irrelevant Importance of social assistance Number of recipients SA relatively important for SA relatively unimportant for recipients recipients Low number of recipients Poland Bulgaria Estonia Latvia [IRRELEVANT] [CONCENTRATED] High number of recipients Hungary Russia [DISPERSED] 75 4.26 In conclusion, Latvia's social assistance is: * very modest as the overall amounts disbursed and number of households who benefit from it are small; * however, for the recipients, social assistance represents an important source of income. The system is therefore concentrated, a feature it shares with social assistance in Poland and Estonia. * But while the system is concentrated, it is not focused on the poor, and its relative effectiveness is worse than in all countries considered here except Russia. 4.4 WHY SOME POOR HOUSEHOLDS DO NOT RECEIVE SOCIAL ASSISTANCE? 4.27 We have seen that the percentage of the poor who are not receiving social assistance ("error of exclusion") is about 98 percent. Can we explain who and why among the poor is "denied" social assistance? In other words, are there identifiable household characteristics that account for household's exclusion? Is it the fact that they live in rural areas, own durables (e.g. a car or a productive asset), have an able-bodied male living in the household, or have small families? Finding out what these characteristics are should give us a better grasp on the performance of the system. For example, if single mothers are systematically discriminated, that probably means that the system is operating worse than if households with able-bodied male (who might work informally) are systematically excluded. Also, it should allow us to look more carefully for the causes of exclusion. For example, if urban areas are systematically discriminated, is it because there are no social assistance offices in the cities or because the offices are understaffed, or perhaps because the allocation of central funds is biased against urban areas? 4.28 Methodology.'5 We waant to estimate econometrically what household characteristics are associated with errors of exclusion. We cannot estimate such regressions simply across all households because for the non-poor we cannot, by definition, observe errors of exclusion. We deal with a censored sample. Differently, to run the regressions across the poor households only would yield biased estimates because people are not poor or non-poor randomly. There are distinct characteristics (see Chapter 2) which are associated with poverty. If that is the case, then, running the regression across the sub sample of the poor would be tantamount to disregarding information from the entire sample, thus yieldingJ biased estimates. For example, we might find when running the regression across the poor only that the failure to deliver social assistance is strongly related to living in villages (peasants do not get much social assistance). But it could also be that living in a vil]lage is a strong determinant of poverty and once we take it into account, none of the discrimination against peasants per se remains. The same exogenous variable in our example (living in a village) explains both the poverty status and the error of exclusion. We need to distinguish between the two. To do so, we run a selection model where households first "select" to be in or out of poverty (the so-called "screening" equation). This is a probit regression because the dependent variable takes the value of either 1 or 0 depending on whether the household is respectively poor or '5 This section (Methodology) is reprinted fiom Braithwaite, Grootaert, Milanovic (1999; Chapter III). 76 non-poor. Then, in the second regression, we identify factors that --for the poor households explain their exclusion from social assistance controlling now for the factors that make people more likely to be poor. 4.29 We have, in essence, to face two important econometric problems: the use of limited dependent variable (binary variable in the first equation), and the selection bias (people "select" to be poor non-randomly). The first problem renders OLS estimators even asymptotically biased; the second problem also makes them biased. We address the selection issue by using the Heckman correction (or Heckman selection model); we address the limited dependent variable problem by applying the maximum likelihood (ML) estimation. We are thus able to obtain unbiased and asymptotically efficient estimators. 16 More formally, we observe an error of exclusion only if the household is poor, that is if 3lxi + ul > 0 where xl is a vector of household characteristics, Pl=a vector of coefficients (such as, for example, obtained in Chapter II) and ul=a normally-distributed random error term. At the same time, there is another equation explaining the exclusion error: FAILURE = 32x2 + au2 where x2 is a vector of household characteristics, P2=a vector of coefficients, u2=a normally-distributed random error term potentially correlated with the first error term (ul) if ct0. The two vectors of household characteristics (xl and x2) must have at least one different variable in order for the two equations to be identified. Our first ("selection" into poverty) regression is similar to the regression from Chapter 2: (1) DPOOR = fct (HHSIZE, DEDU1, DEDU2, DEDU3, AGE, AGE2, PRODUCA, DHOUSE, SHRWAGEY, DSEX, DLOC1, DLOC2, DLFS1, DLFS2) where binary (0-1) variables are prefixed by a D standing for dummy variable, and all variables are household-based, DPOOR = poverty status (poor=-1), HHSIZE = household size, DEDUI = dummy for primary education or less (of household head), DEDU2 = dummy for secondary (general) education of household head, DEDU3= dummy for secondary vocational or technical education of household head (omitted variable=university education), 16 Since we have a limited dependent variable OLS estimators would be biased. We thus need to use ML methods. This is an improvement over the usual, and until recently more common, Heckman two-stage estimation which solved the problem of selection bias but, by not using maximum likelihood estimation, still yielded inefficient (even if consistent) estimates. Until recently, using Heckman correction with ML methods was computationally prohibitive. 77 AGE = age of the household head, PRODUCA = ownership of productive assets, DHOUSE = dummy for tenancy status (vs. home ownership), SHRWAGEY = share of wage income in total household income (to proxy linkage with labor market), DSEX = dummy for female-headed household, DLOC 1 = dummy for other cities, DLOC2 = dummy for rural (omnitted variable=capital city), DLFS 1 = dummy if household head is unemployed, and DLFS2 = dummy if household head is inactive (ormitted variable=employed). Our second ("error of exclusion") regression is: (2) FAILURE = lfct (HHSIZE, DEDUl, DEDU2, DEDU3, AGE, AGE2, DURABLA, PRODUCA, DH[OUSE, DSEX, DLOC1, DLOC2, DLFS1, DLFS2) where all variables are the same except FAILURE = 1 if a household is poor and has received no social assistance. If household is poor and has received social assistance FAILURE =0; for all non-poor households, FAILURE is unobserved, and DURABLA = index of ownership of consumer durables (a new RHS variable), 4.30 While SHWAGEY is dropped for identification purposes. The rationale is that linkage with the fornal labor market (reflected in high value of SHWAGEY) might explain whether the household is poor or not poor, but not whether it is discriminated in the allocation of social assistance. DURABLA is a composite index of durables ownership. It is obtained by assigning to the ownership of each consumer durable good a value of 1 and then summing up the score (e.g. if a household owns a TV and a refrigerator it would score 2). 4.31 Due to the potentially important role that family composition and ownership of durables might have when deciding whether or not to deliver social assistance (as in means- testing), we experimenit with different formulations of the regressions. In one set, HHSIZE is replaced by the family composition variables: number of the unemployed in the household (UNEMPLN), number of children (CHILDN) and number of male adults (MADULTN). In the second set, ownership of specific durables, e.g. ownership of a car; black and white TV only, refrigerator, personal computer etc. are introduced in the equation instead of the cornposite durables index. The equation with household composition (instead of size) and ownership of individual durables, for example, will look like: (3) FAILURE = fct (UNEMPLN, CHILDN, MADULTN, DEDUl, DEDU2, DEDU3, AGE, AGE2, DCAR, DTV, DPC, DREFRIGERATOR, DMICRO, DSTEREO, DMOTOR, PF'ODUCA, DHOUSE, DSEX, DLOC1, DLOC2, LFS1, DLFS2) 78 where the variables in bold show the ownership of various consumer durables. 4.32 Finally, because of the difference in regional approach to the delivery of social assistance, we replace location variables (DLOC1 and DLOC2) with four regional variables dummy variables (DREG1=Riga region, DREG2=Kurzeme, DREG3=Vitzeme, DREG4=Zemgale, and omitted regional variable Latgale). 4.33 The results. Table 4.9 and Table 4.10 show the results of the errors-of-exclusions regression. Eight regressions are run combining the following three formulations: (i) number of household members, or household composition, (ii) location or region, and (iii) index of durables owned or individual durable goods.17 4.34 Note first the variables which are not significant. Level of education and sex of the household head, or his/her age are not found to make more or less likely the receipt of social assistance in any of the eight equations. Similarly, owning a house or being a tenant,18 or having own business do not seem to matter. 4.35 Moreover, the regional variables which we find significant both as determinants of poverty and unemployment are not significant here. Location, however, is. Table 4.9 and Table 4.10 show that urban households, both those living in Riga and outside of Riga, are -after controlling for all other characteristics-more likely to be excluded. The obverse of this is, of course, that rural households seem to be given preference in the allocation of social assistance. 19 4.36 Greater number of male adults (above one) is another characteristic correlated with likelihood of being denied social assistance. It seems that social assistance offices consider such families better able to find alternative means of sustenance. One might recall that until 1968 when the Supreme Court struck it out, a similar rule of "man in the house" was used by the US welfare offices to deny social assistance to households with able-bodied males (see Levitan, 1990, p. 51). 4.37 In one formulation, having unemployed head makes household less likely to receive social assistance. The result may be driven by the fact that the while very modest unemployment benefits may keep the family below the poverty line, the very receipt of the benefit renders the family de facto ineligible for social assistance. 20 Out of 974 households who have unemployed members (and out of which 302 are poor), only 21 are in receipt of social assistance. 17 As the "error of exclusion" equation is modified (e.g. by including household composition instead of household size), so is, in order to maintain the conditions for the exact identification, the first equation. 18 We cannot distinguish between rentors of public and private flats. 19 Using a poverty module attached to the 1998 HBS, the self-reported rejection rate (people who applied for social assistance but were refused) was 19 percent in urban, and 10 percent in rural areas (see Gassman and Neubourg 1999, p. 45). 20 Replacement rate of unemployment benefits ranges from 50 percent for those with 1-5 years of service, reaching 65 percent for those with more than 25 years of service. The average unemployment benefit received is some 30 percent of the average wage, while the per capita poverty line is about 1/4 of the average wage. Thus, a four-member household with one unemployed member, one member employed at less than 70 percent of the average wage, and two children will fall under the poverty line. 79 4.38 Ownership of durable goods, whether measured as an index, or as individual durables, does not appear to have an impact except for the ownership of refrigerator which makes the household more likely to benefit from social assistance. 21 It is unclear why this should be the case. Table 4.9: Explaining error of exclusion: Regressions with the index of durables With three areas With five regions With HH size With HH composition With HH size With HH composition Poor who are "discriminated" in favor Poor who are Urban outside of Riga* Male adults* Large households Male adults "discrinminated" Urban outside Riga Unemployed head* against _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ A significant No No Note: The first column under each country gives the results for the regression which uses household size as explanatory variable; the second column gives the results using household composition. HH=household. All coefficients significant at 1% level unless otherwise * noted. Table 4.10: Explaining error of exclusion: Regressions with the individual durables With three areas With five regions With HH size With HI composition With HH size With HH composition Poor who are Ownership of Ownership of "discriminated" refrigerator* refrigerator* in favor Poor who are Riga* Male adults* Large households Male adults* "discriminated" Urban outside of Riga Riga* against Urban outside of Riga X significant l Note: The first column under each country gives the results for the regression which uses household size as explanatory variable; the second column gives the resualts using household composition. HH=household. All coefficients significant at 1% level, unless otherwise * noted. 4.39 In conclusion, we find that disbursement of social assistance displays a bias in favor of rural areas, although no regional bias could have been detected. Social assistance offices also tend to deny assistance to households with more than one adult male or headed by an unemployed person. Other than for the rural inhabitants, no other types of households, such as ithose headed by females, or by the elderly, or by the more educated, are found to be "discriminated in favor." 21 Of all consumer durables, refrigerator is by far the most commonly owned: 91 percent of all households, and 81 percent of poor households, own it. 80 4.5 REGIONAL INEQUALITY IN DISTRIBUTION OF SOCIAL ASSISTANCE 4.40 Regional inequality in the allocation of social assistance has repeatedly been raised as an issue (see Goldman 1998; World Bank, 1995). This is also the problem of which the Government has been aware. Regional inequities essentially stem from the way that the system of social assistance is organized and funded. Like in most countries, social assistance in Latvia is administered at the local level. But, in addition, the funding of social assistance is to some extent localized. Local governments receive block grants from the Center through the Equalization Fund and also raise their own funds. They are free to allocate both centrally-provided ands local funds for any use, depending on what they see as being a priority. Spending on education, health and public services thus competes with spending on social assistance. In principle, this approach is reasonable. First, Equalization fund should ensure that poorer local governments receive more funds than the rich, thus ensuring regional equity. Second, because local governments should know best what are local needs, the freedom to allocate money (that is, not to have earmarked uses) may be desirable. However, both points can also be questioned. First, Equalization fund may not ensure horizontal (that is, regional) equity. Second, even if Equalization fund achieved this objective with respect to total social spending, it may not achieve it with respect to social assistance. The poor often lack political power to "force" local governments to spend more on social assistance. Thus, reliance, in part, on local funding for social spending, plus lack of poor's political clout in the allocation of spending, implies that there are serious dangers of horizontal inequity: individuals with the same characteristics (e.g. low income level) may be treated differently depending on what part of the country they live in. 4.41 We shall try to check this hypothesis using two approaches. In the first, we use Household Budget data to obtain an estimate of territorial distribution of the need for social assistance (approximated by the number of the poor) and its actual distribution. In the second approach, we use very desaggregated data on social assistance from more than 500 local governments to contrast them with some demographic characteristics of the population. This is, of course, far from perfect because demographic characteristics do not imply "need". Unfortunately, we cannot contrast the allocation of social assistance at the local level with poverty at the same level-as ideally we would like because HBS data are not representative at that level (and indeed are not even presented), and poverty headcount cannot be calculated. 4.42 By combining five regional (Riga, Kurzeme, Vidzeme, Zemgale, and Latgale) and three local (large cities, small cities, rural areas) classifications from Household Budget Surveys, plus Riga city, we obtain sixteen regional units. One of them is empty (Vitgale large city), so we are left with 15 regional RBS units (see Table 4.11). For each of them we calculate from HBS, poverty headcount, and the disbursed poverty assistance per capita (Table 4.11). One can then obtain disbursed social assistance per poor person, that is social assistance per unit of "needs." Latvia-wide, social assistance spending was about 1 lat per month per poor person. However, eight regions are severely "under- provisioned": social assistance per poor person is less than 50 percent of Latvia-wide 22 Note that this is an average of social assistance disbursed and number of the poor. Since not all social assistance is disbursed to the poor only, a poor person will not receive on average I lat per month. 81 Table 4.11: Poverty headcounts and allocation of social assistance by regions (based on HBS data and HBS regional units) (1) (2) (3) Poverty headcounts Social assistance per Social assistance per (in %) person (lats per poor person (lats per month) month) (2): (1) Riga city 10.8 0.066 0.61 LargeciyJu r , ma'l) 13.1 t '0.',0 e 00 U i0 l$.'; e00! Rural areas 16.3 1.211 7.45 Kurzeme Large city (Liepaja) 14.7 0.240 1.64 Zemgale Large city (Jelgava) 15.3 0.077 0.50 Small cities 19.0 0.289 1.52 Rural areas 24.1 0.545 2.26 Latgale Large cities (Daugavpils, 23.4 0.121 0.52 Rezerne) Small cities ~~~25.5 4~0.030 0.12 Rualaras 38.4 00 t000:0000A000i000000tjj;0,X4l;0.82 900liiUf;0000Q.47X:00j00f Total Latvia 19.4 0.191 0.98 level. The "underprovisioned regions" are: in the Riga area, Jurmala and small cities; in the Kurzeme, Vidzeme and Latgale areas, small cities and rural areas. On the other hand, rural areas around Riga, rural areas in Zemgale, or the city of Liepaja receive between 1½ and 7 times as much as Latvia-wide average. We can conclude not only that the distribution seems to be uneven but also that a generalization based on location (rural vs. urban areas) is not a strong predictor of what areas do not receive sufficient social assistance. While, for example, small cities and rural areas in Kurzeme, Vidzeme and Latgale disburse inadequate amounts of social assistance, rural areas in Zemgale and around Riga receive far more than their "needs" seem to be. 4.43 The geographic map of Latvia enclosed here shows that the underprovisioned areas include all Eastern districts (rajons), and several (four) in the West of the country. The region in the Center of the country, around Riga and in the South, is better off in terms of 23 received social assistance. 23 White-shaded districts receive social assistance per poor person that is less than 50% of country-wide average. The districts where the major (republican) cities are, are "allocated" to the group to which the city belongs (e.g. spending levels in Liepaja, see Table 4.11, determine the shading of the Liepaja district). The "unallocated" districts straddle two 82 Latvia: Map of regional distribution of social assistance per poor person Social Assistance by Region (Lats per poor person per year) .Adlministrative Dhvisions . r:.--------E; . _ Iemational bNdaety- . . E STON -A E]less&anO.S more than 1.0 N S P injonsl caplbl r - ; 15< g . , . EB unallocated 30 2VU/5 1 U RJO 2 SA R 6:''j. OAGVL AOS15 LUOZS RAJONS t\' 24 ' VAK RA JON. t Sea j>, crt l = gRiga BsRUSSI :%2~~~~~~~~~~~~~~~~~~~~~~~~~~~~% 7 I AlZKSRAUK£5hONS 16 MAtIOVAS RAJONS 25 VALMIERS RAJONS .- . 2 UKSENES RAJONS 17 OGRES RAJONS 26 VENS1MS RAJONS .. 9 S BALVULSRAJONS 18 PREULdSSN 2 tRAJONS . ;..... }:.. . 4 BAUSKASRAJONS 13 LIEP>ASRS 22 TtSUNS - .. :----. .e -. 0$ao4MItoJ 1., socaltasisaceeevd 83~~~~~~~~~~~~~~~~~~~3 1- 5AJZRLSU RAJONS 14 JELAVA RAMOS 13 RTUKNES RFONS --i ..6 0AUGAVPILS RAJONS 15 LUDZAS RWJONS 24 VALKAS RAJONS..-.;'- 7 DO0ELES RAJONS 16 MADONAS RAJONS 25 VAtLMIERAS RJNS -,...;.-, : 8 GULSENES RAJN 17 OGRES RAS 26 VENiTSPILSRAJONS.-. .. .-- 9 JEKAsPMLs RAJONS la PREIJU RAJONS . : -. regions with different importance of social assistance; we thus could not precisely deteffrnine their correct level of social assistance received. 83 4.44 The HBS regions need to be mapped into the local government (LG) level at which social assistance is administered and funded. This presents somewhat of a problem because in a few cases, a given district can "belong" to two traditional regions used in the Household survey. At the level of towns and rural parishes (pagasts), however, there are no such problems: each pagast can be mapped into one of 15 HBS regions. Annex 2 shows the detailed "mapping" with cells belonging to the under-provisioned areas shaded. In total, we find that only one large city (Jurmala) receives insufficient social assistance; 70 cities and towins, and 338 pagasts (see Table 4.13 top panel). Total population living in the underprovisioned areas amounts to 934,000 people or 38 percent of Latvia's population. Out of these people, 876 thousands live in towns and rural areas (72 percent of all population living in towns and villages), and 60,000 in Jurmala (the sole underprovisioned republican city). 4.45 The conclusion regarding what are the underprovisioned regions according to the HBS data can be contrasted with what are the underprovisioned regions using information obtained from the Ministry of Social Welfare. The Ministry has provided the Bank mission with the very detailed data on the 1998 total allocation of social assistance by almost 560 cities, towns and pagasts. Social assistance is defined more broadly than in the HBS (see Table 4.2). Etroadly defined social assistance includes (1) cash and in- kind transfers disbursed in accordance with the Law on Social Assistance, namely general low-income support, cash and in-kind payment for rents and utilities, for wood and coal, for the care of children and the aged, and funeral allowance, and (2) other social care benefits like free food, free medical help (hospitals, drugs etc.), and support for children and family (free textbooks, kindergartens, school transport etc.). In Table 4.12, we use two definitions of social assistance: spending based on the Law on Social Assistance (point 1 above) and total spending (sum of points 1 and 2), and express it both per capita and per person of non-working age. 4.46 First, total social assistance (broad concept) is twice as large as narrow social assistance (disbursed in accordance with the Social Assistance Law): average per capita spending on broad social assistance was Lats 4.7 per year vs. about Lat 2.3 for the narrow concept. Social assistance transfers thus represent about 0.6 percent of population income calculated from HBS (see Table 2.1 in Chapter 2). The narrow concept of social assistance accounts for 0.3 percent of total income, a percentage which is exactly the same as obtained from Household surveys (see Table 4.1a above). 4.47 Second, inequality in distribution of social assistance is substantial. Whatever concept of social assistance or recipient used, the Gini coefficients is high. It ranges between 48 and 57. A note of caution is in order here. Even under the theoretical hypothesis of perfect targeting, inequality would still be present, and possibly high, because the poor are not evenly distributed across the country. Moreover, we do not know if the existing high Gini is high because the poor areas are well targeted or because most of the money is disbursed to the rich areas. Therefore, the Gini coefficient simply shows high inequality in the allocation. It says nothing whether that inequality is "justified" or not. 84 Table 4.12: Distribution of social assistance across local governments in 1998 Social assistance (narrow concept) Social assistance (broad concept) Per capita Per non- Per capita Per non- working person working person Mean (lats p.a.) 2.29 4.86 4.70 9.94 Standard deviation (lats 4.57 9.23 7.58 15.41 p.a.) Coefficient of variation 1.99 1.90 1.61 1.55 Gini coefficient 56.8 53.4 51.7 48.3 Local govt's with lowest Balgales Balgales Balgales Balgales disbursements Blontu Blontu Rojas Rojas Pavilostas Pavilosta Kraslava Kraslava Remtes Remtes Dobele Dobele Rojas Rojas Aizkraukle Berzaunes Local govt's with highest Valmieras Dobeles Balvu Balvu disbursements Dobeles Valmieras Dobeles Dobeles Aizkraukles Aizkraukles Aizkraukles Aizkraukles Balvu Balvu Valmieras Valmieras Kraslavas Kraslavas Kraslavas Kraslavas Source: Data provided by the Ministry of Welfare. Note: Total of 553 local govemments. 4.48 Third, inequality in distribution of social assistance decreases as we use a broader concept of social assistance and move from per capita to per non-working person approach. As can be seen in Table 4.12, the Gini coefficient for broad social assistance per non-working person is 48.3, but for narrow social assistance per capita it is almost 57. The same regularity is observable for the coefficient of variation. 4.49 Ideally, if we had HBS-derived data on poverty headcounts by 553 cities and pagasts we could compare spending per poor person across all 553 local governments. But, as explained above, the most detailed picture of poverty that we can obtain from BBS is at the level of 15 HBS regions. We thus have to resort to a palliative solution. We compute the per capita spending of (broadly defined) social assistance across all local governments, and define as underprovisioned the local governments that spend less than 50 percent of the country-wide per capita average. Ideally, such underprovisioned areas should correspond to the underprovisioned areas obtained from the HBS, and discussed in para 4.42-4.45. Table 4.13 shows the correspondence between the two classifications. The calculations based on the Ministry of Welfare data show that only 131 rather than 408 local governments can be considered underprovisioned. About 80 percent of these 131 local governments (104 to be exact), however, are also underprovisioned according to the BBS data. It seems that the use of the Ministry data gives us the "hard core" of the 85 underprovisioned areas. In terms of the population living in the underprovisioned areas, the calculations based on the Ministry of Welfare data give some 337,000 people or about 28 percent of total population living in towns and rural areas. This is much less than 876,000 people based on the HBS results. However, again, more 80 percent of people defined as underprovisioned according to the Ministry of Welfare are also underprovisioned according to the HBS data. Table 4.13: Comparison of underprovisioned areas according to the HBS data and Ministry of Welfare Number of LG's Accordin to HBs | Underprovisioned Satisfactory | Total Acc. to Underprov. 104 27 131 Ministry Satisfact. 304 118 422 of Welfare Total 408 145 _ 553 Population According to HBs | Underprovisioned Satisfactory Total Acc. to Underprov 281062 56099 337161 Ministry Satisfact. 594983 289816 884799 of Welfare Total 876045 345915 1221960 Note: "Undemprovisioned" areas according to HBS are defined as all areas where social assistance disbursed per estimated poor person is less than 50% of country-wide average. "Underprovisioned" areas according to the Ministry of Welfare are defined as all areas where "broad" social assistance per capita is less than 50% of country-wide average. 4.50 We conclude that the use of a relatively rough indicator of regional allocation of social assistance-broad concept of social assistance divided by the number of inhabitants-shows that (1) there is a great diversity between the local governments with the Gini coefficient only slightly below 50, (2) about one-fourth of LG's (131 out of 553) comprising 27 percent of the population living in towns and rural areas are underprovisioned, and (3) thus identified underprovisioned areas represent the lower bound, or the "hard core" of underprovisioned areas. The implication is that the use of the more readily available Ministry data will allow us to avoid Type II error (we are unlikely to misclassify a rich area as underprovisioned), but will not protect us from the Type I of error-a fair number of underprovisioned areas may be missed out. 4.6 POLICY RECOMMENDATIONS 4.51 The current system of financing of social assistance in Latvia imnplicitly treats poverty as an issue to be dealt with entirely by local governments. Differently put, it appears that poverty alleviation is of a limited concern to the national government. This is a problem because some local governments may truly be unable to cope with the task of such a size (for example, because fiscally poor pagasts are likely to have the largest percentage of the poor people) while some LG's may neglect social assistance. While it may be argued that such careless attitude will, in due course, lead to their electoral defeat, 86 the effects resulting from such an attitude (increased poverty) will become a national problem much before "election check" comes into force. Thus, the national government may be placed in the position where it needs to ensure some support for the poor regardless of the situation in a given pagast. 4.52 This proposal, however, is only a short-term palliative. The externality argument suggest strongly that sources of finances for the anti-poverty programs must be, to a significant even if not full, degree guaranteed at the central level. LG's are, of course, the best places to deliver social assistance, but they are not the best places to finance it or to decide whether or not to deliver it in preference to another social program. One obvious reason is that LG's where the poor are relatively most numerous are also the poorest in terms of tax revenues. The variability of social assistance spending per poor person is, as we have seen, fairly large. In 17 districts (see Map above) social assistance per poor person is less than 50 percent of country-wide average. According to the HBS data, almost 1 million people live in such "underprovisioned" areas. Even according to the more conservative Ministry of Welfare data, more than 300,000 people live in underprovisioned areas (see Table 4.13). Equalization fund can improve their position, but it is clearly doing it in an insufficient manner. 4.53 There are different ways in which the fact that poverty in any part of the country is a national issue can be translated into financing arrangements. The center can allocate grants to local governments, solely for purposes of poverty alleviation, approximately in relation to the extent of poverty in each of them. However, the detailed data on poverty headcounts -at the level of local government- required by such an approach cannot be derived from the Household budget surveys. But, as we have shown above, relatively reliable data on poverty headcounts can be computed for HBS 15 areas (see Table 4.11). All local governments within each of these 15 areas can be "assigned" the poverty headcount of "their" FBS area. Clearly, with each new HBS, the poverty map of Latvia may change and correspondingly would allocations of central funds change. The process can be repeated annually. The process would effectively introduce an official poverty line; that poverty line would be the same across all regions of Latvia (ensuring horizontal equity: a poor person in any place in Latvia is treated equally); and the financing for purposes of poverty alleviation would be guaranteed centrally. The system would therefore introduce explicit national standards. 4.54 There is, however, an inherent danger in this process. If local authorities could be fully relied to follow central rules regarding the eligibility threshold (poverty line), and additional conditions that the government may wish to impose (e.g. multi-children households, sole parent, long-term unemployed household head), the process could function. Even then there would be little incentive for the LG's to raise funds on their own-they would be entirely passive distributors of central funds. This is not desirable. Thus, a combination between centrally-guaranteed funds and incentive to raise some funds locally may be preferred. 4.55 The central government should encourage LG's to raise their own sources of revenues. The need for a balance between central's concern for equity and the need for local role may be fulfilled by conditional grants. Under one possible scenario, the central government would allocate 80 percent of the FBS-estimated LG's social assistance needs 87 as an earmarked block grant transfer. Then, for the next 40 percent of the social assistance expenditures, central government would provide only matching funds in the ratio of 1 to 1, that is, central government would spend one lat for each lat spent by LG's on social assistance.24 After that point, each additional spending would be either matched by the central government using a sliding scale (giving less than 1 lat for each lat spent by a local government), or would be entirely financed by LG's. (In order to prevent the abuse of the matching grants system, the government may set the ceilings to the level of local spending that it will match.) The proposal does not imply any additional fund disbursement from the central government. The expenditures from the Center are a "choice-variable": they can be made greater, smaller or equal to the current level, for the system functions the same way regardless of how much the Center decides to commit in total. 4.56 To make the working of the proposed system quite clear, let us suppose that, based on the Household Survey daya, and using a government-determined national poverty line, 25 we calculate that the region X's needs for social assistance are 100. The central government would then disburse 80 as an earmarked block grant to be spent by X on social assistance only. In oirder to ensure that spending at the local level follows certain nation-wide rules, the central government must define precise rules regarding eligibility for social assistance (these rules, by the way, already exist in the Social assistance law), and should then proceed to periodic checks. Now, for all spending between 80 and 120, the Center uses the matching funds scheme. It means that each lat spent on social assistance and financed out of X's local sources is matched by a lat of transfers from the Center. For example, if total spending is 120, 100 would have been financed centrally, and 20 locally. All spending over 120, is either entirely financed out of X's funds, or is financed in part by the Center using a lower matching scale than 1-to-I (say, one lat from the Center is paid only against 4 lats of X's funding). The proposed system would therefore: *ensure national standard because it is based on a single nation-wide poverty line; eavoid open-ended commitments by the Center because the center limits its exposure; *thus ensuring that there is no increased expenditures from the Center; and, *ensure national monitoring of the rules because the Center can check whether the rules are observed. 4.57 The proposals outlined above have the following implications for the current system of transfers. The financing of social assistance would be entirely taken out of the 24 This can be made even more specific by limiting matching grants to specific spending, e.g. to single mothers with several children, the handicapped poor., or the poor elderly. 25 The selected poverty line would, of course, depend on a number of factors: likely number of takers, government fiscal position, political preferences, etc. 88 current system of state-local transfers and included in the central budget. A part of social assistance spending would be entirely guaranteed by the state; a part would be "controlled" by the state through the system of matching grants, and a part would be entirely left to local governments to finance and use as they see fit. 4.58 One can, of course, conceive of alternative proposals that similarly introduce national standards, and improve targeting of social assistance. One such proposal was presented in the Government's Concept for a Guaranteed Minimum Income System. It is similar to the above proposal in that a national standard for a poverty benefit is to be imposed on the municipalities. Municipalities must pay this benefits before they offer any other cash benefit to their citizens. The benefit is a means-tested cash allowance, which raises household income per capita up to a minimum level set by the Government. 4.59 This benefit would be financed from municipal resources and state budget resources. State budget resources would be transferred to the municipalities through the equalization fund, which should ensure that poor municipalities have adequate resources to pay the benefit. The transfer mechanism would be simple to administer and would use existing financing mechanisms. This benefit is estimated cost about 1% of GDP, or about 10% of local government revenues. The Government concept envisages an extensive program of training for local social assistance offices to ensure equal treatment, as well as monitoring and an adjudication mechanism. An evaluation would take place after the first year to verify targeting, as well as affordability at the municipal level. 89 ANNEX 1. THE DESCRIPTION OF THE LbTAVIA HOUSEHOLD BUDGET SURVEY (a) Frequency of the survey The new household budget survey (HBS) was initiated in September 1995. This survey has been prepared and implemented with technical assistance of The World Bank and the United Nations Development Program (UNDP).The survey is continuous; information from households is obtained every month. In a year it is envisaged to survey 7992 households. Each household included in the sampling is surveyed one month a year. Consequently, 666 households are surveyed in a month. During the next survey month these households are substituted by other. Half of the households surveyed in 1996 form the panel part of the survey. That means that households included in the panel shall be survey during the same rnonth as they were in 1996 in consecutive 3 years onwards. This type of panel survey shall allow to follow the budgets of the most typical groups of the households in development. (b) Coverage of the sample The target population of the HBS consists of all private households in Latvia. Persons living in the institutional households (elderly people boarding house, disabled children boarding house, student hostels, hotels, barracks, hospitals, sanatoriums, penal institutions, etc.) are excluded from current survey. (c) Coverage of the questions Three types of documents (application forms) have been prepared for the preparation of the survey: 1. HOUSEHOLD QUESTIONNAIRE 2. HOUSEHOLD DIARY 3. REPORT ON REASONS OF NON-RESPONSE Reference period of the HOUSEHOLD QUESTIONNAIRE is one month. Application form of Household Questionnaire consists of two interviews. The first is Introductory Questionnaire, which is filled in by the interviewer during the first visit to the household after it has agreed to participate in the survey. It contains following sets of questions: * description of the members of the household; * housing conditions; * arable land plot of the household; * employment status of 15 years old and elder members of the household. After carrying out preliminary interview, the interviewer shall give HOUSEHOLD DIARY No.1, which in the form of self-registering is filled in by the household during 15 days. Afterwards the interviewer visits the household, examines the accuracy of notes given, asks few detailed questions if necessary and gives Household Diary No.2. This diary shall be filled in by the household until the end of the month. Household Diary No.2 differs from the first not demanding to report on state of cash at the beginning of reference period and Part 1 has stated the number of days from the date 16 to 31. As to the rest they are identical. Each household fills the household diary with the following information: 90 * all household money expenditure on food and non-food commodities, services etc. payments; * goods and services received and consumed in the household free of charge; * revenues and outgoings of agricultural production in the household (in money and kind). * number of meals (breakfast, dinner, supper) and number of persons present; Interview is concluded by the Final Questionnaire. It is the second part of household questionnaire application form and comprises rather intricate questions, which might be asked only when mutual confidence has been established between interviewer and respondent. This part of the questionnaire includes questions where the data acquisition by self-registration could be ineffective. Final questionnaire includes the following questions: * social characteristic of the members of household (sources of livelihood, description of the place of work, vacation etc.); * personal revenue of each of the members of the household. This information is obtained on each of the member of the household; - revenue of the household as a whole; - financial transfers - equipment in durable goods; * self-evaluation of the living conditions of the household; * visits to educational, cultural, health care etc. institutions (frequency, distance, time consumption). The task of the REPORT ON REASONS OF NON-RESPONSE is to find out the reasons for replacement (refusal) of the households included in the survey sample, as well as to discover the consequences that arise from deficiencies of the sampling frame (in urban areas - the Population register, in rural areas - lists of households). (d) The survey approach see answers to point (c) too. Preliminary interview is performed during the first contact with the household after it has given its consent to participate in the survey. After the preliminary interview, the interviewer shall give Household Diary No.1, which in the form of self-registering is filled in by the household during 15 days. Afterwards the interviewer visits the household, examines the accuracy of notes given, asks few detailed questions if necessary and gives Household Diary No.2. This diary shall be filled in by the household until the end of the month. Household Diary No.2 differs from the first not demanding to report on the financial balance at the beginning of the survey and Part 1 has stated the number of days from the date 16 to 31. As to the rest they are identical. Actually the interviewer must visit the household in between the above mentioned visits, in order to ascertain on the manner the household fills in the diary, whether it needs any assistance and additional explanations. It is rather often that reports are not made on a sufficient regular basis, and additional inquiry on purchases made in between the visits, 91 should be made by inte:rviewers in the diary. Such assistance fairly often is necessary to the old people, as well as to those with poor eye-sight and other restrictions as to their activities. The inquiry is completed by the Final interview. It is the second part of the Household Questionnaire, which includes fairly complex questions that may be put up only when mutual trust is achieved between the interviewer and the household. This part of the questionnaire includes a wide range of questions, including those on income where self- registering may turn out not to be as effective. The final interview consists of the following questions: (e) the size of the selected sample, the total number of households in the country, the percentage response, and whether response is uniform' or different' The total annual sample size of the new HBS is equal to approximately 7992 households (666 per month with a. complete monthly renewal till December 1996). Half of every month sample of the year 1996 will be used as a panel in the sample of the corresponding month for the next 3 yeatrs' period. The monthly sample size is distributed across the major strata in the following way: 222 HHs in Riga (capital of Latvia); 116 in 6 other largest cities; 90 in middle-sized towns (over 7,000 inhabitants);, 33 in small towns; 205 HHs in the rural areas. A sample for HBS has been selected separately for the following domains: * Riga and 6 large cities; * middle and small towns, and * rural areas. The distribution of the samples (distribution of the sample community) in different territories of the survey between towns and rural areas as well as among 5 historically - ethnographic regions of Latvia (Kurzeme, Zemgale, Vidzeme, Latgale and Riga region) is described by the table below: REGIONAL (AREA) DISTRIBUTION OF THE MONTHLY SAMPLES OF A SURVEY Number of primary sampling units, Number of households in the REGIONS (AREAS *) where the survey is made sample _urban Frural **) I total urban | rural total TOTAL 23 29 52 461 205 666 Kurzeme 4 5 9 56 35 91 Zemgale 4 6 10 37 41 78 Latgale 5 6 11 62 42 104 Vidzeme 4 7 11 37 49 86 Riga region L 6 5 11 269 38 307 *) relative (unofficial) distribution **) In some initial sampling units some of the parishes have been merged because of having small number of inhabitants The Population Register was used to obtain the sample of households in urban area, that is, in two first domains. For rural areas lists of parishes (as PSUs) and lists of households were used as sampling frames. 92 It is necessary to remind that in each domain different sample design has been used. For this reason estimation procedure will be considered for each domain separately. (f) The stratification (sample design) of the survey In urban areas the Population Register formed during 1991 -- 1992, which comprised more than 99% of all population of Latvia, was used as sampling frame. It should be noted also that the use of population register to acquire household sampling complicates procedures of data processing considerably. Namely, sampling unit is a person if population register is used as a frame. At the same time survey unit is a household. It means that the population register for the sample selection of households is a multiple frame -- the household will be included in the sample independently of the member selected. Hence, the selection probability for the given household is larger for larger households. To determine the inclusion probability for the given household it is necessary to know the number of registered members of the household. In spite of imperfections of the population register mentioned above it is the only acceptable existing register at the moment and it is used as a frame for sampling in urban areas. Alternatives of sampling frame were sought for in rural areas. As such alternative turned out to be complete lists of households of the lower level self-governments -- parish. These lists were compiled and up-dated back in the times of the former USSR for the purposes of agricultural and demographic statistics. Complete lists of parish's households were successfully applied when sampling households for Agricultural survey. Therefore it was necessary to prepare identical lists for only 21 parish because for 11 parishes of 32 sampled during the first stage of sampling, the complete household lists were already made for the needs of the Agricultural survey. In a multi-functional survey as Household Budget Survey, it is exceptionally significant to receive from the households included in the sampling lists as exhaustive responses to the questions included in the survey programme as possible. In the case if households refuse to participate in the survey seriously or respond to the questions of the survey, or they are not to be found at the address given, it may materially affect precision of acquired results. So that due to the factors of refusal or other non-response, the amount of acquired information would not change, sequential sampling approach is used. Non-respondent household is being replaced by another from the reserve list. The last subsequently is surveyed in place of the non-respondent. According to the survey procedure households for re-placement are taken in strict order. Considering that the reserve list has been made on a random selective basis, the household replaced by the household from the reserve list has the same probability of being selected. Since the beginning of the survey the number of households replaced from the reserve list reached 30% from the total number of households included in the sampling list. The total response level is characterised by the following table. 93 NONRESPONSE LEVEL BY REASONS OF NONRESPONSE. IV Q I Q 1996 II Q | III Q IVQ 1995 1996 1996 1996 Number of households included in the 1998 1998 1998 1998 1999 basic list Number of households taken from the 916 818 862 888 833 reserve list Total number of households laken 2914 2816 2860 2886 2832 from both lists Percentage of households excluded 12.4 10.2 6.2 5.7 7.7 from the lists due to different sampling frame imperfections Percentage of correctly identified 7.5 7.5 8.0 10.3 8.3 households that were not at home Percentage of correctly identified 12.4 10.2 14.1 11.2 12.5 households that refused to participate Illness, not able to participate, 3.5 4.2 4.2 6.9 5.0 percentage Other reasons of non-response, 3.2 3.7 3.2 2.7 2.4 percentage Total response level, percentage 73.4 74.4 70.5 68.9 71.8 As it is evident from the table above, there are mainly "objective" non-response factors. It is connected with problerrms of sampling frame and these were greater in cities, where data of Population Register serve as the sampling frame. FRAME IMPERFECTION CASES BY QUARTERS AND RESIDENTIAL AREA ._________ __________ Of which Quarter Urban-Rural Total Register not Wrong Institutional Duplicates Other _ updated address households IV Q Urban 327 252 23 18 4 30 1995 Rural 35 19 4 2 3 7 l Q Urban 232 154 31 19 5 23 1996 Rural 55 23 6 1 6 19 II Q Urban 131 63 19 24 5 20 1996 Rural 46 20 3 3 13 7 III Q Urban 107 54 11 13 16 13 1996 Rural 57 27 7 0 12 11 IVQ Urban 139 62 19 24 21 13 1996 Rural 80 32 2 3 31 12 Since March 1996 interviewers were allowed to include those households in the survey found at the address given by the sampling list, in case the person included in the sampling 94 list did not live at the given address and his/her actual address could not be identified or it was located outside the borders of the territory to be surveyed. This fact has diminished the level of non-response as of II quarter 1996. As evidences by the data of the survey mainly hired labour and pensioner- households prevail in the households surveyed. At the same time results of the survey prove that there is a low number of entrepreneurs, self employed people and their households. Non- response plays a significant part in the fact that households contain a low number of entrepreneurs and other persons connected with private business. That is not a secret that entrepreneurs and other self-employed persons would not willingly participate in the survey. (g) How the survey data are grossed up to provide national estimates All parameters will be estimated at domain level first, i.e. large cities (Riga and 6 large cities), other towns, and rural areas. National estimates will be obtained by summing up through domain estimates or weighting them respectively. The estimation is confined here to the following parameters: averages, totals, ratios and percentages. Let us use the following notations: U - population (the set of all households), R - the set of households responding during the survey, D - domain of study (the set of all households under interest), (D c U), R D=DrR. The mean value of variable X in the domain of study D, i.e., the value XDI _ 1 XD~ND isl XiD is estimated by XD, ie RD X~D= NE I L Wi iE RD where ND is the number of elements of the set D, xi. is the value of variable X of the i-th household within the corresponding set (D or R D). D is the set of indices corresponding to the set of households D, RD is the set of indices corresponding to the set of households R D. Pi is the inclusion probability of i-th household of the set R D, 1 wi -is the corresponding weighting coefficient. Pi 95 The total XD, XD XiL, ieD is estimated by XD, XD w E ixi ic-RD The ratio of two totals RD, XD XD ieD YD YD = Yi is estimated by RD, wixi i(E-RD ie RD Remarks. 1. To estimate the total or mean of some variable X or ratio of two variables X and Y in the whole population in the formulas given above it is necessary to replace D by U. 2. D can be an arbitrary subset of U, for example, all households of some socio- economic group living in urban areas, or all households living in Riga. Nevertheless it is necessary to take into account that estimates will be inefficient in the case if there are small number of representatives in the sample from the domain D. (h) Whether, and if so how, adjustments are made to the raw survey data Data entry and primary data control is performed at the supervisor's office which is located in administrative region of the regional state statistical office. For this purpose each supervisor has a PC with a printer. This equipment provides all data entry procedures and print-out of all errors and logical misappropriations discovered during the entry. About two hundred checks were implemented into the entry program. All messages about negative check's results are printed. Data entry operators print these messages for each household separately and discuss them with supervisors. After reception of data files from data entry, data 'cleaning' is done at the central sector of data processing. After this procedure objection is given to the law-governed use of pressed data entry procedure up to 20%. Adjustment for alcohol and tobacco has not been taken in account due to lack of reliable source of information on purchase or consumption of them. (i) To allow for non-response or differential response The analysis of non-response of the Household Budget survey in rural areas in the first quarter of 1996 shows that the percentage of single person households among non- respondents is 48%. At the same time it is only 18% among respondents. It means that for 96 rural areas the small size households are seriously underrepresented in our survey. Therefore we are making corrections of the individual weighing coefficients of households in rural areas in two steps. At the first step we are replacing the planned sample size of household (the same correction coefficient for all households of the same size in rural areas) to obtain the estimated distribution of households in the rural areas by size equal to the distribution of households by sizes in the frame (in the lists of households of 29 PSU). At the same time the correction coefficients are chosen in such a way that the estimated population size in rural areas is equal to the one given. 97 ANNEX 2. "MAPPING" OF LATVIA'S LOCAL GOVERNMENTS INTO THE CLASSIFCATION PROVIDED BY THE HOUSEHOLD BUDGET SURVEY Region Riga Large Cities Small Cities Rural Kurzeme Liepaja Liepajosrajons Kuldlsraons Ventpils rajons Saidus rajons Talsu rajons Zemgale Jelgava Bauskas rajons Aizkraukles raions: Jau:njelgava Dobeles rajons Jaunjelgavas rural territory fllkste Jelgavas rajons Daudzeses pagasts Subate Aizkraukles rajons Kurmenes pagasts Jakabpils (partially) Mazzalves pagasts Akniste Daugavpils rajons Neeretas pagasts Viesite (partially) Pilskalnes pagasts Jakabpils rajons Seces pagasts Serenes pagasts Staburaga pagasts Sunakstes pagasts Valles pagasts Zalves pagasts Daugavpils raions: Subates rural territory Bebrenes pagasts Demenes pagasts Dvietes pagasts Eglaines pagasts Kalkuines pagasts Laucesas pagasts Medumu pagasts Pilskalnes pagasts Salienas pagasts Skrudalienas pagasts Sventes pagasts S6deres pagasts Tabores pagasts Vecsalienas pagasts Jekabpils rajons Aknistesrural territory Viesites rural territory Asares pagasts Abelu pagasts Dignajas pagasts Dunavas pagasts Elksnu pagasts Garsenes pagasts Kalna pagasts Leimanu pagasts Rites pagasts Rubenes pagasts Salas pagasts Saukas pagasts Selpils pagasts Zasas pagasts 98 Region Riga Large Cities [ Smali Cities Rural Latgale: Daugavpils Balvu rajons Daugavpils ra.jows Razekne Kraslavas rajons Ambelu pagasts Varakjani Ludzas rajens Bikermieku pagasts Preilu rajons Dubnas pagasts Rezeknes rajons Kalupes pagasts Daugavpils rajons Uksnas pagasts (partially) Malinovas pagasts Jekabpils rajons Naujenes pagasts (partally) Nkcgales pagasts Madonas rajons Vaboles pagasts (panially) Vijku pagasts Jekabpils rajons At$ienes pagasts Krustpils pagasts Kllku pagasts Meares pagasts Vanesu pagasts VTpes pagasts Madonas rajons Barkavas pagasts Munmastienas pagasts _a_a__a_u gasts Vidzeme: Aizkraukle Aluksnes rajons Aizkraukles rajons Plavinas Casu rajons Aivieckstes pagasts Madona Gulbenes rajons Aizkraukles pagasts Cesvaine Limbazu rajons Bebru pagasts Lubana Valkas rajons lrKu pagasts Valmiieras rajons Klintaines pagasts Aizkraukles rajons Kokneses pagasts (partially) Skrtveru pagasts Madonas rajons Vietalvas pagasts (partially) Madonas rajons Cesvaines rural territory Aronas pagasts Barzaunes pagasts Dzelzavas pagasts iglu pagasts Indrgnu pagasts Jumurdas pagasts Kalsnavas pagasts Lazdonas pagasts Liez&res pagasts Laudonas pagasts Marcionas pagasts Metrienas pagasts Osupes pagasts Praulienas pagasts Sarkanu pagasts Sausnrjas pagasts Vestienas pagasts Riga Region: Riga Jurmala Rlgas rajons Ogres rajons Tukuma rajons Note: shaded areas show the areas that are receiving less than half of social assistance which they "should" receive according to the poverty rates calculated from HBS. 99 Appendix A: Composition of poverty Table Al: Composition of poverty Non-Door Poor All Household Size 1 16.2 5.3 14.1 2 25.7 14.5 23.5 3 24.4 20.4 23.6 4 22.0 25.6 22.7 5 7.7 20.8 10.2 6 2.5 7.7 3.5 7 0.8 3.1 1.3 8+ 0.6 2.7 1.0 100.0 100.0 100.0 Numbier of children 0 56.5 32.7 51.9 1 24.1 27.4 24.7 2 15.3 24.4 17.1 3+ 4.1 15.5 6.3 100.0 100.0 100.0 Number of elderly 0 71.9 77.1 72.9 1 20.2 18.4 19.9 2+ 7.9 4.6 7.3 100.0 100.0 100.0 Household composition 0 children 14years 0 elderly persons 33.8 20.0 31.1 1 elderly person 15.9 9.6 14.7 2+ eldlerly persons 6.8 3.1 6.1 1 child 14 years 0 elderly persons 20.8 22.8 21.2 1 elderly person 2.6 4.1 2.9 2+ elderly persons 0.7 0.5 0.6 2 children 14 years 0 elderly persons 13.8 20.4 15.1 1 elderly person 1.2 3.2 1.6 2+ elderly persons 0.3 0.8 0.4 3+ chilcren 14 years 0 elderly persons 3.6 13.8 5.5 1 elderly person 0.4 1.4 0.6 2+ elderly persons 0.1 0.2 0.1 100.0 100.0 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determ-ined using per capita consumption. 100 Table A2: Composition of poverty (cont.) Non-Poor Poor All Age and gender Male <5 yrs 1.9 3.4 2.2 5-9 3.2 5.1 3.6 10-14 3.8 5.6 4.1 15-19 2.7 4.2 3.0 20-29 5.3 6.2 5.4 30-39 5.6 7.3 6.0 40-49 5.7 5.2 5.6 50-59 5.4 4.3 5.2 60-69 5.2 2.9 4.7 270 3.4 1.7 3.0 Female <5 yrs 1.8 3.3 2.1 5-9 3.0 5.2 3.4 10-14 3.6 5.3 3.9 15-19 2.9 3.5 3.0 20-29 6.2 7.3 6.4 30-39 7.0 7.6 7.1 40-49 7.6 7.3 7.6 50-59 8.4 4.8 7.7 60-69 9.1 4.5 8.2 270 8.3 5.5 7.7 100.0 100.0 100.0 Age and gender of household head Male <30 yrs 4.3 5.1 4.5 30-39 9.9 13.4 10.6 40-49 10.0 10.2 10.1 50-59 7.9 7.5 7.8 60-69 7.3 5.2 6.9 270 4.3 2.3 3.9 Female <30 yrs 5.8 6.5 5.9 30-39 10.8 13.7 11.3 40-49 11.7 14.3 12.2 50-59 10.6 8.6 10.2 60-69 9.8 6.2 9.1 >70 7.7 6.8 7.5 100.0 100.0 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. 101 Table A3: Composition of poverty (cont.) Non-Door Poor All Gender aind education, persons l15 years Male Primary or less 11.2 18.6 12.5 Vocational 2.3 3.5 2.5 Secondary 21.1 20.5 21.0 Higher 5.7 1.5 5.0 Female Primary or less 17.1 22.4 18.0 Vocational 1.1 2.2 1.3 Secondary 31.7 29.2 31.3 Higher 9.9 2.1 8.5 100.0 100.0 100.0 Gender and education of household head Male Primary or less 8.9 12.9 9.6 Vocational 2.2 3.4 2.4 Secondary 25.0 24.9 25.0 Higher 7.7 2.6 6.7 Female Primary or less 12.0 19.6 13.5 Vocational 1.0 2.1 1.2 Secondary 32.9 32.3 32.8 Higher 10.2 2.1 8.7 100.0 100.0 100.0 Socio-economic group Male Wage earner 25.9 23.3 25.4 Self-employed 2.9 2.2 2.8 Pension 9.3 11.2 9.7 Other social benefit 0.5 1.8 0.7 Other incorne 3.7 7.3 4.4 Female Wage earner 31.5 25.7 30.3 Self-employed 3.1 2.7 3.0 Pension 17.5 15.7 17.2 Other social benefit 0.7 2.3 1.0 Other income 5.1 7.7 5.6 100.0 100.0 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. 102 Table A4: Composition of poverty (cont.) Non-Poor Poor All Gender and labor force status, persons Ž15 years Male Employed 22.8 21.0 22.5 Unemployed 2.6 8.9 3.7 Not in labor force 14.8 14.2 14.7 Female Employed 25.1 20.1 24.2 Unemployed 2.9 7.3 3.6 Not in labor force 31.8 28.5 31.2 100.0 100.0 100.0 Gender and labor -force status of household head Male Employed 29.1 27.2 28.8 Unemployed 1.8 6.1 2.6 Not in labor force 12.9 10.5 12.4 Female Employed 28.6 24.4 27.8 Unemployed 2.7 6.7 3.5 Not in labor force 25.0 25.1 25.0 100.0 100.0 100.0 Presence of food plot Riga N6 plot 28.4 15.0 25.8 Has plot 7.7 3.1 6.8 Other urban No plot 25.1 27.6 25.6 Has plot 11.5 9.2 11.0 Rural No plot 20.0 36.3 23.2 Has plot 7.3 8.8 7.6 100.0 100.0 100.0 Locality Riga 36.1 18.1 32.6 Other urban 36.6 36.8 36.6 Rural 27.3 45.1 30.8 100.0 100.0 100.0 Region Rigas region 49.3 29.5 45.5 Kurzene 12.9 17.3 13.8 Vidzeme 11.1 14.7 11.8 Zemgale 12.8 13.7 13.0 Latgale 13.9 24.7 16.0 100.0 100.0 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. 103 Appendix B: Additional tables Table BI: Household consumption budget shares (percent) Riea Other Urban Rural Latvia Poor All Poor All Poor All Poor All Food .59.4 40.4 57.4 42.7 45.3 38.8 52.3 40.8 Rent 3.3 4.0 3.6 3.2 0.7 0.9 2.3 2.8 Other housing 11.9 20.1 13.1 16.4 7.0 9.6 10.1 15.7 Education and culture 3.6 4.9 3.2 4.4 2.1 3.4 2.8 4.3 Health 2.8 3.9 2.4 4.1 2.4 4.1 2.5 4.0 Transport 5.2 7.7 2.1 5.1 2.5 5.6 2.8 6.1 Clothing 3.0 5.2 3.4 5.5 3.4 4.8 3.3 5.2 Private transfers given 0.8 3.0 1.0 3.2 0.9 2.5 (.9 2.9 Home consumption expenditure 2.0 1.7 6.5 5.6 29.2 21.6 16.0 8.9 Other 7.8 9.2 7.2 9.7 6.5 8.8 7.0 9.3 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. Table B2: Poverty rates by gender and age (percent) 0=1 660.8 660.6 Male <5 yrs 29.7 24.4 21.7 5-9 27.7 22.9 18.9 10-14 26.2 21.2 18.8 15-19 26.8 22.9 20.9 20-29 21.9 18.8 18.0 30-39 23.9 20.9 19.0 40-49 18.1 17.6 16.8 50-59 16.1 15.5 16.8 60-69 11.9 12.9 15.2 270 10.7 12.3 16.2 Female <5 yrs 31.5 24.9 20.9 5-9 29.4 24.8 20.5 10-14 26.0 21.3 17.3 15-19 22.1 19.3 18.3 20-29 22.1 19.2 18.2 30-39 20.7 18.4 16.7 40-49 18.7 17.6 17.9 50-59 12.2 12.3 15.0 60-69 10.7 11.6 17.1 >70 13.7 16.6 23.9 Source: Author estimates based on HBS, 1997, 1998. 104 Table B3: Poverty rates by gender and age of household head (percent) 0=1 0.8 6=0.6 Male <30 years 22.3 19.6 18.4 30-39 24.6 19.6 16.5 40-49 19.8 17.8 14.2 50-59 18.6 15.8 16.6 60-69 14.7 14.1 15.6 Ž70 11.3 12.9 16.0 Female <30 years 21.2 18.8 18.8 30-39 23.5 21.4 19.6 40-49 22.9 20.5 19.9 50-59 16.4 15.3 16.5 60-69 13.3 13.5 18.5 270 17.6 20.0 27.4 Source: Author estimates based on HBS, 1997, 1998. Table B4: Poverty rates by gender and education level (percent) Population Riga Other Urban Rural Latvia share (Latvia) Gender and education level, persons Ž15 years Male Primary or less 14.8 23.1 32.4 25.9 12.5 Vocational 17.4 21.7 31.0 24.2 2.5 Secondary 10.0 18.9 23.8 17.0 21.0 Higher 3.7 7.5 4.9 5.1 5.0 All 10.0 19.0 27.2 18.7 Female <=primary 14.0 19.9 27.0 21.6 18.0 vocational 22.7 31.3 38.3 29.3 1.3 secondary 9.3 17.6 24.3 16.2 31.3 university 3.0 5.4 5.9 4.3 8.5 All 9.3 16.9 24.5 16.4 100.0 Gender and education level of household head Male Primary or less 16.3 18.1 31.9 25.9 9.6 Vocational 7.9 25.0 34.1 27.5 2.4 Secondary 10.8 20.1 26.3 19.3 25.0 Higher 4.8 12.1 7.5 7.6 6.7 All 9.7 18.8 27.6 19.4 Female Primary or less 16.6 25.5 36.6 28.2 13.5 Vocational 32.8 35.0 32.8 33.6 1.2 Secondary 11.9 21.1 27.1 19.1 32.8 Higher 4.2 5.1 6.3 4.8 8.7 All 11.4 20.0 29.4 19.4 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. 105 Table B5: Poverty rates by gender and socio-economic group (percent) Ponulation Riga Other Urban Rural Latvia share (Latvia) Male Wage earner 9.9 18.3 29.9 17.8 25.4 Self-employed 5.7 13.3 17.9 15.7 2.8 Pension 15.6 22.9 28.5 22.5 9.7 Other social benefit 12.9 44.9 74.6 48.4 0.7 Other income 16.1 35.3 35.4 32.3 4.4 All 11.5 20.9 29.6 20.7 Female Wage earner 9.2 16.4 29.5 16.5 30.3 Self-employed 10.2 9.4 19.8 17.3 3.0 Pension 11.8 19.1 22.8 17.8 17.2 Other social benefit 24.0 43.3 64.0 46.2 1.0 Other income 9.9 30.2 33.2 26.8 5.6 All 10.3 18.5 27.5 18.4 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. Table B6: Poverty rates by gender and labor force status (percent) Population Riga Other Urban Rural Latvia share (Latvia) Gender and labor force status, persons >15 years Male Employed 7.2 14.1 27.1 16.2 22.5 Unemployed .31.4 44.5 51.0 41.8 3.7 Not in labor force 9.2 17.3 23.6 16.8 14.7 All 10.0 19.0 27.2 18.7 Female Employed 6.4 13.4 26.5 14.4 24.2 Unemployed 21.2 40.8 48.8 35.1 3.6 Notinlaborforce [10.1 16.1 21.7 15.8 31.2 All 9.3 16.9 24.5 16.4 100.0 Gender and labor force status of household head Male Employed 7.9 16.8 28.0 18.3 28.8 Unemployed 28.6 45.3 58.6 45.4 2.6 Not in labor force 10.8 15.1 21.9 16.4 12.4 All 9.7 18.8 27.6 19.4 Female Employed 9.5 15.1 31.5 17.1 27.8 Unemployed 24.1 46.0 52.0 37.5 3.5 Not in labor force 11.4 21.9 26.5 19.5 25.0 All 11.4 20.0 29.4 19.4 100.0 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. 106 Table B7: Selected housing characteristics by household poverty status Riga Other Urban Rural Latvia Poor Non-poor Poor Non-poor Poor Non-poor Poor Non-poor % of households with amenity Running water 98.7 99.1 81.9 88.9 45.3 57.3 69.4 84.3 Hot water 70.9 79.5 45.5 62.8 12.4 23.8 36.1 58.7 WC or bathroom 82.9 91.3 73.5 84.1 39.5 47.6 60.7 77.1 Central gas 85.0 86.5 45.5 49.9 10.0 12.5 37.6 53.6 Central heating 77.7 84.5 68.7 78.3 22.7 33.9 50.8 68.9 Electricity 97.8 99.7 98.6 99.9 97.1 99.0 97.8 99.6 Telephone 70.6 85.9 51.3 73.7 33.5 51.7 47.2 72.4 % of households owning durable Car 6.3 21.7 12.1 27.5 19.2 31.1 14.0 26.3 Television (B/W) 29.1 20.6 35.3 25.1 42.3 35.8 37.1 26.3 Television (color) 68.7 80.3 55.5 75.2 43.7 59.7 52.9 73.0 Refrigerator/freezer 92.7 94.7 80.9 92.6 70.3 85.2 78.5 91.4 Auto washing machine 65.7 73.4 66.6 74.9 59.4 70.1 63.4 73.1 Sewing machine 42.3 58.0 43.7 54.8 31.9 50.6 38.4 54.9 Personal computer 0.7 2.9 0.5 1.2 0.2 0.8 0.4 1.7 VCR 13.0 25.9 10.8 18.9 5.9 10.5 9.1 19.3 Stereo 24.7 31.1 24.6 32.6 31.6 31.8 27.6 31.8 Microwave 1.8 6.9 2.0 3.4 0.5 1.5 1.4 4.2 Motorcycle 0.0 0.6 1.4 2.2 2.7 4.4 1.7 2.2 Source: Author estimates based on HBS, 1997, 1998. Note: Poverty status is determined using per capita consumption. 107 Table B8: Per capita consumption regressions, by locality Rica Other urban Rural Dependent variable: locfconsumption/household size' coefficient p-value coefficient p-value coefficient p-value Household composition number of children <14 years -0.265 0.000 -0.214 0.000 -0.164 0.000 number of male adults -0.128 0.000 -0.118 0.000 -0.152 0.000 number of female adults -0.135 0.000 -0.158 0.000 -0.092 0.000 number of elderly persons -0.159 0.000 -0.135 0.000 -0.093 0.000 Education of household head primary or less -0.177 0.000 -0.126 0.000 -0.213 0.000 vocational -0.102 0.115 -0.093 0.096 -0.158 0.004 higher 0.189 0.000 0.247 0.000 0.335 0.000 Age of household head -0.019 0.000 0.004 0.402 0.014 0.002 (age of household head)x102 0.000 0.001 0.000 0.323 0.000 0.009 Head is female -0.094 0.000 -0.060 0.009 -0.076 0.001 Head is inactive -0.262 0.000 -0.156 0.000 -0.066 0.038 Head is unemployed -0.268 0.000 -0.519 0.000 -0.329 0.000 Number of unemployed in household -0.282 0.000 -0.242 0.000 -0.209 0.000 Household situated in Riga region 0.075 0.013 0.234 0.000 Household situated in Vidzeme -0.007 0.831 0.078 0.021 Household situated in Zemgale 0.030 0.310 0.181 0.000 Household situated in Latgale -0.102 0.000 0.040 0.262 Household has access to food plot 0.070 0:013 0.117 0.000 0.098 0.000 Constant 5.190 0.000 4.358 0.000 3.733 0.000 N 2385 2861 2444 R2 7.266 .256 .242 Source: Author estimates based on HBS, 1997, 1998. Notes: See notes to Table 2.12. 108 Table B9: Consumption regressions, main breadwinner is head 0=1 6=0.8 0=0.6 Dependent variable: o[elconsumntion/(household size)91 coefficient p-value coefficient p-value coefficient p-value Household composition number of children S14 years -0.219 0.000 -0.150 0.000 -0.082 0.000 number of male adults -0.158 0.000 -0.075 0.000 0.008 0.474 number of female adults -0.150 0.000 -0.062 0.000 0.025 0.028 number of elderly persons -0.132 0.000 -0.041 0.004 0.051 0.000 Education of household head primary or less -0.164 0.000 -0.165 0.000 -0.166 0.000 vocational -0.131 0.000 -0.132 0.000 -0.134 0.000 higher 0.253 0.000 0.253 0.000 0.253 0.000 Age of household head 0.003 0.119 0.003 0.114 0.003 0.109 (age of household head)x102 0.000 0.092 0.000 0.058 0.000 0.036 Head is female -0.076 0.000 -0.084 0.000 -0.092 0.000 Head is inactive -0.215 0.000 -0.220 0.000 -0.225 0.000 Head is unemployed -0.417 0.000 -0.423 0.000 -0.428 0.000 Number of unemployed in household -0.229 0.000 -0.224 0.000 -0.220 0.000 Household situated in Riga region 0.127 0.000 0.127 0.000 0.127 0.000 Household situated in Vidzeme 0.025 0.299 0.023 0.335 0.021 0.375 Household situated in Zemgale 0.088 0.000 0.091 0.000 0.093 0.000 Household situated in Latgale -0.047 0.033 -0.047 0.032 -0.047 0.031 Household situated in capital (Riga) 0.040 0.055 0.037 0.068 0.035 0.085 Household situated in rural area -0.055 0.000 -0.060 0.000 -0.066 0.000 Household has access to food plot 0.086 0.000 0.088 0.000 0.090 0.000 Constant 4.359 0.000 4.323 0.000 4.286 0.000 N 7690 7690 7690 R 2 .287 .233 .226 Source: Author estimates based on HBS, 1997,1998. Notes: See notes to Table 2.12. 109 Table B1O: Poverty regressions (probit) 0=1 0=0.8 0.6 probability probability probability Dependent variable: household is poor derivatives p-value derivatives p-value derivatives p-value Household composition number of children <14 years 0.060 0.000 0.040 0.000 0.018 0.004 number of male adults 0.047 0.000 0.025 0.000 -0.014 0.078 number of female adults 0.032 0.000 0.006 0.387 -0.032 0.000 number of elderly persons 0.019 0.032 -0.006 0.507 -0.063 0.000 Education of household head primary or less 0.086 0.000 0.076 0.000 0.097 0.000 vocational 0.111 0.000 0.106 0.000 0.102 0.000 higher -0.076 0.000 -0.088 0.000 -0.080 0.000 Age of household head -0.003 0.027 -0.003 0.031 -0.005 0.002 (age of household head)x102 0.000 0.087 0.000 0.083 0.000 0.001 Head is female 0.022 0.007 0.018 0.035 0.028 0.005 Head is inactive 0.033 0.002 0.049 0.000 0.071 0.000 Head is unemployed 0.218 0.000 0.236 0.000 0.268 0.000 Number of unemployed in household 0.106 0.000 0.117 0.000 0.135 0.000 Household situated in Riga region -0.032 0.018 -0.045 0.002 -0.061 0.000 Household situated in Vidzeme 0.017 0.217 0.005 0.748 -0.007 0.668 Household situated in Zemgale -0.031 0.012 -0.047 0.000 -0.051 0.001 Household situated in Latgale 0.021 0.100 0.011 0.406 0.005 0.719 Household situated in capital (Riga) -0.035 0.006 -0.026 0.058 -0.021 0.181 Household situated in rural area 0.030 0.001 0.043 0.000 0.058 0.000 Household has access to food plot -0.035 0.000 -0.044 0.000 -0.050 0.000 N 7690 7690 7690 Pseudo R2 .172 .126 .108 Source: Author estimates based on HBS, 1997, 1998. Notes: See notes to Table 2.12. The derivatives are the predicted change in the probability of a household being poor, calculated at the mean of continuous variables and for a change from zero to one in the case of the dummny variables. The p-values indicate the significance of the underlying coefficient in the probit model. 110 Appendix C: Sensitivity Analysis Table Cl shows the poverty rate calculated using different poverty lines and welfare measures. Relatively small shifts in the poverty line results in quite marked changes in the headcount rate. For example, if the poverty line were to be increased by 10 percent when per capita consumption is the welfare measure, then the poverty rate would increase to 24.1 percent, an increase of 24.2 percent over the poverty rate calculated for the base poverty line. The table also shows the impact of using a welfare measure which takes account of the demographic composition of the household; when the OECD equivalence scale is used, the poverty rate under the base poverty line falls to 9.6 percent. Table Cl: Sensitivity analysis Alternative Poverty lines Base poverty 80% of 90% of 110% of 120% of line (LVL) base line base line base line base line 0=1 Poverty line 28.0 22.4 25.2 30.8 33.6 PO 19.4 11.4 15.3 24.1 28.5 %pt. change in P0 -41.2 -21.1 24.2 46.9 0=0.8 Poverty line 34.0 27.2 30.6 37.4 40.8 Po 17.8 10.2 14.0 22.3 27.2 % pt. change in P0 -42.7 -21.3 25.2 52.8 0=0.6 Poverty line 42.0 33.6 37.8 46.2 50.4 Po 18.2 10.2 14.0 22.6 27.3 % pt. change in P0 -44.0 -23.1 24.2 50.0 OECD per equivalent consumption- Poverty line 28.0 22.4 25.2 30.8 33.6 PO 9.6 5.1 7.1 12.6 15.6 % pt. change in P0 -46.9 -26.0 31.3 62.5 Source: Author estimates based on HBS, 1997, 1998. 1 The OECD equivalence scale is as follows: first adult in household = 1.0, additional adults 0.7, children less than 14 years = 0.5. 111 Appendix D: Functional form for regressions: linear versus semi-log In this paper, regression techniques are used to predict household welfare and to identify the variables that are significantly correlated with welfare. In the regressions, the dependent variable is log per capita consumption, while the independent variables are either continuous variables or dummy variables reflecting household characteristics. From a theoretical perspective, it is attractive to use the semi-log functional form: Inc = oc + o3x, where x is a continuous variable. This specification implies that the effects of household characteristics on welfare are proportional rather than linear. Thus education, for example, will increase household welfare in a fixed proportion rather than by a fixed amount and hence the absolute returns to education are lower for the poor. While there are theoretical reasons for using the semi-log functional form, it is useful to also econometrically test whether the data fit this functional form. This was done using the STATA ado program boxcox, which finds the maximum-likelihood Box-Cox transform. The Box-Cox transform Y(X) - y =X represents a family of data transformations. For instance: y(X)=y-l ifx=1 yA =ln(y) if X = ° I() - 1/y if X = -1 The STATA ado program boxcox finds the maximum-likelihood value of X for the model: Yi (;) = CC + oxi + ei where ei is assumed to be normally distributed and homoscedastic. The X obtained from boxcox is therefore the value that transforms y to being approximately normally distributed. An estimated X of one implies that the linear specification of the consumption function is appropriate, while an estimated A of zero supports the use of the semi-log functional form. The boxcox procedure was run for Latvia using per capita consumption and the independent variables from Table 5.1, and the estimated X was -0.078. While the test that X=O was rejected, the fact that the estimated X was much closer to zero than one led to the conclusion that the semi-log functional form is appropriate for use in the welfare regression. 112 Appendix E: Labor Force Surveys LATVIA LABOR FORCE SURVEY (MAY 1998) ILO STANDARD UNEMPLOYMENT DEFINITION (UE 2) Path in the Questionnaire # of respondents Description Employed C01==1 7078 Worked last week "Working2" C05-=. & C05-=10 & C05-=4 170 Not work for 'excused" reason C03==1 64 Worked on family farm C04==1 or C04==2 285 Farmwork for sale or consumption Subtotal Employed: 7597 Unemployed F35==1 1125 Looked& could start in 2 weeks "UE2" C05==4 15 Involuntary Unpaid Leave Subtotal Unemployed: 1140 Subtotal in the Labor Force: 8737[ Unemployment Rate 13.0% Out of the Labor Force F37==2 17 ?sent to Training from SEB? "OOLF2" F37==3 orF37=4 53 No job in hand, but seems likely F37==9 & F40==7 138 Didn't look, "Out of work" F35==2 28 Looked & couldn't start in 2 weeks F37==5 or F37 ==6 502 Discouraged Workers F37==1 24 Entering Labor Force now F37==9 & (F40>=1 & F40 <=6) 5648 Didn't look, All other reasons Subtotal Out of the Labor Force: 64101 Total Respondents: 113 LATVIA LABOR FORCE SURVEY (MAY 1998) REGISTERED UNEMPLOYMENT DEFINITION (UE5) Path in the Questionnaire #of respondents Description Employed D06-=. ("working3"+C05==4) 7612 Answered Section D "Main Job" "Working5" Subtotal Employed: 7612 Note: D06-=.&G49==1 221 "Working", registered unempioyed Unemployed D06==. & G49==1 519 No "main job", register unemployed "UE5" F33==1 & G49-=1 42 Subtotal Unemployed: 561 Subtotal in the Labor Force: l 8173 Unemployment Rate 6.9% Out of the Labor Force Working5==O & UE5==O 6975 "OOLF5" Subtotal Out of the Labor Force: 69751 Total Respondents: 151481 114 LATVIA LABOR FORCE SURVEY (MAY 1996) ILO STANDARD UNEMPLOYMENT DEFINITION (UE2) Path in the Questionnaire # of respondents Description Employed COI =1 5354 Worked last week "Working2" C03~=.&C03-=10&C03-=4 237 Not work for "excused" reason Subtotal Employed: 5591 Unemployed F33 -=8 1487 Looked& could start in 2 weeks "UE2" C03==4 29 Involuntary Unpaid Leave Subtotal Unemployed: 1516 Subtotal in the Labor Force: I7107| Unemployment Rate 21.3% Out of the Labor Force F3 1==2 102 Looked & couldn't start in 2 weeks "OOLF2" F34 ==1 3 Entering Labor Force now F34==2 0 ?sent to Training from SEB? F34==3 or F34==4 6 No job in hand, but seems likely F34=5 or F34 ==6 151 Discouraged Workers F34==9 71 Other reasons for not looking F30 ==2 4046 Don't want to work Subtotal Out of the Labor Force: 4379 Total Respondents: 11486 115 LATVIA LABOR FORCE SURVEY (MAY 1996) REGISTERED UNEMPLOYMENT DEFINITION (UE 4) Path in the Questionnaire # of respondents Description Employed D04-=. ("workingl "+C03==4) 5620 Answered Section D "Main Job" "Working4" Subtotal Employed: 5620 Note: D04-=.&CT48==1 94 "WodKiing", registered UE Unemployed D04==.&G48==1 331 No "main job", register UE "UE4" F33==1 & G48,=1 11 Subtotal Unemployed: 442 Subtotal in the Labor Force: [6062 Unemployment Rate 7.3% Out of the Labor D06==. & G49,=1 5535 No Main Job, Didn't registerUE Force "OOLF4" Subtotal Out of the Labor Force: 55351 Total Respondents: 115971 116 REFERENCES Ackland, Robert (1999), "Poverty in the Republic of Latvia in 1997/1998", Mimeo, World Bank, May. Braithwaite, Jeanine, Christiaan Grootaert, and Branko Milanovic (1998), "Determinants of Poverty and Targeting of Social Assistance in Eastern Europe and the Former Soviet Union," mimeograph, World Bank: Washington, D.C. Braithwaite, Jeanine, Christiaan Grootaert and Branko Milanovic (2000), Poverty and Social Assistance in Transition Countries, London, New York: St. Martin's Press. Central Statistical Bureau of Latvia (1998), "Women and the Transition in Latvia", mimeo, Riga. Fischer, Stanley, Ratha Sahay, and Carlos Vegh (1998), "From transition to market - Evidence and growth prospects" IMF working paper WP/98152. Foster, J., Greer, J. and Thorbecke, E. (1984), "A Class of Decomposable Poverty Measures," Econometrica, 52, pp. 761-765. Gassman, Franziska (1998), "Who and Where are the Poor in Latvia", UNDP, June 1998 Government of the Republic of Latvia, Social Report 1996-1997, 1997. Gassman, Franziska and Chris de Neubourg (1999), "Coping with Little Means in Latvia: Quantitative Analysis of Qualitative Statements", UNDP Riga, mimeo, June. Goldman, Peter (1998), "Reaching the Poor During Transition: Administering the Latvian Social Assistance Reform," mimeograph, World Bank, June 11. Havrylyshyn, Oleh, Ivailo Izvorski, and Ron van Rooden (1998), "Recovery and growth in transition economics 1990-97: A stylized regression analysis" IMF woreking paper WP/98/141. Institute of Philosophy and Sociology (1998), "Listening to the Poor: Social Assessment of Poverty in Latvia: Report on Research Findings (March-June 1998)", Riga. Institute of Economics, Latvian Academy of Sciences, "Activities in 1997/1998" (1998), mimeo, Riga. IMF Background Papers for Staff Report (1998), "Labor Market Issues in Latvia" (March 13, 1998) Keune, Maarten (1998), "Poverty and The Labour Market in Latvia: Evidence from the Household Budget Survey and Labour Force Survey", UNDP, mimeo, October. 117 Lanjouw, Peter and Jeny Lanjouw (1997), "Poverty Comparisons with Non-Comparable Data: Theory and Illustrations," Policy Research Working Paper 1709, World Bank. Lanjouw, Peter, Branko Milanovic, and Stefano Patemostro (1998), "Poverty and the Economic Transition: How Do Changes in Economies of Scale Affect Poverty Rates for Different Households?" Poli.cy Research Working Paper 2009, World Bank: Washington, D.C. Levitan, Sar A. (1990), Programs in Aid of the Poor, Baltimore: Johns Hopkins University Press. Loza, Zana, Karlis Caunitis, and George-Stephan Barfub (1997), "Labor Market in Latvia 1996: Deployment and Analysis," Mimeo, Stockholm School of Economics. Ministry of Economy, Republic of Latvia (1998), Economic Development of Latvia, Riga, December. Milanovic, Branko (1998), Inicome, Inequality and Poverty During the Transition, World Bank: Washington, D.C. Oaxaca, Ronald L. and M'ichael R. Ransom (1994), "On Discrimination and the Decomposition of Wage Differentials," Journal of Econometrics; 61(1), March, pp. 5-2 1. World Bank (1995), "Latvia - Local government expenditures and resource transfers", Report No. 14470. July 20. 118