Report No. 31857-CA Shocks and Social Protection: Lessons from the Central American Coffee Crisis (In Two Volumes) Volume II: Detailed Country Cases December 28, 2005 Poverty Reduction and Economic Management and Human Development Sector Management Units Latin America and the Caribbean Region Document of the World Bank Shocks and Social Protection: Lessons from the Central American Coffee Crisis TABLE OF CONTENTS Volume I: Synthesis ofFindings andImplicationsfor Policy PREFACE....................................................................................................................................................... iv EXECUTIVE SUMMARY ............................................................................................................................ v CHAPTER 1 INTRODUCTION: Shocks and Social ProtectioninCentralAmerica................................ 1 CHAPTER2 THE WELFARE IMPACTS OF THE COFFEE CRISIS: A TALE OF FOUR COUNTRIES.................................................................................................................................................. 9 CHAPTER3 ELEMENTSOF A STRATEGY TO DEAL WITH SHOCKS ............................................. 34 CHAPTER4 IMPLICATIONS FORPOLICY............................................................................................ 53 Volume 11: DetailedCountryAnalyses PREFACE..................................................................................................................................................... ii 1. SHOCKS AND COFEE: Lessonsfrom Nicaragua(Renos Vakis, Diana Kruger andAndrew D.Mason).................................................................................................................... 1 2. COPINGWITH THE COFFEE CRISIS INCENTRAL AMERICA: The Roleofthe NicaraguanSocial Safety Net Program(John A. Maluccio) ......................................................... 58 3. ANALYSIS OF THE POVERTYAND SOCIAL IMPACTS OF THE COFEE CRISIS (Price Shock) INEL SALVADOR (Alvaro Trigueros andCarolinaAvalos)............................... 97 4. COPINGWITH THE COFFEE CRISIS INCENTRAL AMERICA: The Role of Social Safety Nets inHonduras (DavidCoady, Pedro Olinto, andNatalia Caldes)......................................................................................................................................... 149 5. THE COFFEE CRISIS: A Short Noteon GuatemalanFarmers(Renos Vakis)......................... 168 1 PREFACE Shocks are a familiar feature o f the Central American landscape. Since the mid-l990s, the countries o f Central America have experienced a number o f natural shocks, including Hurricane Mitch (1998), earthquakes (El Salvador, 2001), and a series o f seasonal droughts and floods (often associated El Nifio and L a Nifia); as small open economies, the Central America countries are also open to a variety o f economic shocks, whether in the form o f external terms-of-trade shocks, like that associated with the recent "coffee crisis", policy-induced terms o f trade changes, like those that will accompany the Central America Free Trade Agreement, or more generalized slowdowns inthe U.S.and global economies. This study is about the impact of shocks on people's welfare in Central America and about crafting effective public responses to shocks. It focuses on the on the lessons from the coffee crisis - an unprecedented decline in world coffee prices that occurred between 1997/98 and 2001/02. The report i s part o f an ongoing engagement between the World Bank and its counterparts in Central America on social protection, comprising both policy dialogue and operational support for strengthening governments' abilities to provide basic services to their poorest citizens and to protect the most vulnerable from the impacts o f shocks. This report was undertaken inresponse to requests from several Central American governments to better understand the impacts o f the coffee crisis on households' wellbeing. A key objective o f this study has thus beento bringtogether a rich set ofnew empirical evidence to enable a deeper understanding o f the crisis' impacts on household income, consumption, poverty and basic human development outcomes, such as education and nutrition. To accomplish this, the study has generated a body o f evidence based on an unusually rich collection o f household survey data from El Salvador, Guatemala, Honduras, and Nicaragua, including - including panels o f data from Nicaragua, El Salvador, and Honduras - to provide a more detailed, clearer understanding o f the crisis than has been available to date. Given the importance o f shocks in the Central American context, a second key objective o f the study is to draw out the broader lessons o f the coffee crisis to help strengthen the abilities of governments in the region to respond to the shocks that will inevitably hit their countries in the future. To do this, the report draws both on the new findings on the coffee crisis, as well as other recent evidence from the region - on shocks and on the role and efficacy o f various safety net programs - to identify a broader set o f lessons regarding shocks and social protection. By learning the lessons o f recent experience, the region's governments, as well as their development partners, can be better prepared to deal with a range o f economic and natural shocks in the future. To achieve its objectives, this report has been organized into two volumes. Volume Ipresents a synthesis o f the key findings and policy implications, focusing both on the specific impacts o f the coffee crisis and the more general lessons for government responses to shocks. Volume I1 focuses specifically on the impacts o f the coffee crisis, presenting the collection o f background studies commissioned specifically for this report. These background papers provide rich analytical detail on each o f the four study country for readers who are interested in a more in- depth assessment o f country-level impacts, inthe unique data sets underlying the analyses, and/or the methodologies underpinningthe analytical work summarized inVolume I. .. 11 ~~ SHOCKSAND COFFEE:LESSONSFROMNICARAGUA Renos Vakis, Diana Kruger andAndrew D.Mason* The World Bank Washington D.C. July 2004 ABSTRACT Using household level panel data from Nicaragua, this paper explores the impact o f the recent coffee crisis on rural households engaged in coffee production and coffee labor work. Taking advantage o f the panel structure o f the data, a number o f findings emerge: (i) overall growth between 1998 and 2001 was widespread in rural while Nicaragua, coffee households saw large declines invarious socioeconomic outcomes; (ii) coffeehouseholds,itissmallfarmhouseholdsthatwereaffectedthemost among and not poor labor households as previously expected; (iii) even though coffee households used various risk management strategies to address the shock, it was pre shock, ex-ante strategies (like income diversification) that were the most effective in allowing coffee households insulate against the shock. By contrast, the coffee households that used ex-post coping instruments did not manage to mitigate the adverse impact as well, with additional potential long run implications via extensive uses o f harmful coping strategies (like increases in child labor); and (iv) the coffee shock affected upward mobility and downward poverty vulnerability o f coffee households. Such findings seem to confirm the widespread impact o f shocks on overall household behavior and indicate the importance o f incorporating risk management inthe policy agenda o f poverty reduction. *Renos Vakis and Andrew D. Mason are economists at The World Bank. Diana Kruger i s on the faculty at Pontificia Universidad Cat6lica de Valparaiso, Chile. For comments contact rvakis@worldbank.org The authors are grateful to Caridad Araujo, Natalia Caldes, David Coady, Carlos Felipe Jaramillo, Bryan Lewin, Alessandra Marini, Hans Hoogeveen, John Maluccio, Pedro Olinto, Carlos Sobrado and Panos Varangis for helpful comments and insights. Excellent research assistant work was done by Kalpana Mehra. 1 CONTENTS I. INTRODUCTION .............................................................................................................. 3 2. DATA. COFFEE TYPOLOGYANDA BASELINE PROFILE OF COFFEE HOUSEHOLDS ................................................................................................................. 4 3. ASSESSING THEIMPACT OF THE COFFEE SHOCK .............................................. 7 4. RLSKMANAGEMENTSTRATEGIESANDRESPONDINGTOSHOCKS ...............19 5. SHOCKS. WLNERABILITYAND MOBILITY ........................................................... 44 6. PUBLICRESPONSE TO THE COFFEE CRISIS ....................................................... 50 7. MOVING FORWARD:LESSONSFOR CONSTRUCTINGA POLICY AGENDA 52 ... 8. REFERENCES ................................................................................................................ 55 9. Appendix 2: Attrition andpanel construction................................................................ 57 2 1. INTRODUCTION Coffee is by far the most important crop for the Nicaraguan economy. It i s the highest source o f agricultural export revenues in Nicaragua. Specifically, during the last 5 years, coffee exports have averaged $140 million (24 percent o f total export earnings).' It i s estimated that total employment in coffee production accounts between 20 and 40 percent o f the rural labor force; and that more than 65% o f those employed inthe sector are seasonal worker^.^ Nonetheless, for the last few years the coffee industry has been undergoing a worldwide structural change. The entry o f a number o fnew producers inthe late nineties (such as Vietnam), as well as technological improvements leading to increases in production in Latin American countries (e.g. Brazil) have dramatically increased production and as such, international coffee prices have been severely depressed. The collapse in prices has resulted in significantly lower revenues for coffee producers in Nicaragua. Between 1998 and 2001, average price receivedby coffee exporters decreased from $151 to $59 per hundredweight - a decrease o f 61%.4 By 2001, the price received by coffee producers (between $45 and $50 per hundredweight) was barely sufficient to cover production costs, which are estimated to be $35, $45, and $55 (per hundredweight) for low, medium and high-technology farms.5 This has seriously affected the Nicaraguan coffee economy. Many farmers have been forced to reduce and even abandon coffee production altogether. In addition, there i s concern about the social impact o f the crisis on the coffee laborers. Initial estimates suggested that 35,000 permanent and more than 100,000 seasonal coffee plantation workers may have lost their coffeejobs6 Still, the lack of in depth empirical evidence to understand the magnitude of the crisis impedes informed policy formation. Not only there i s a need to better measure the impact o f the shock but also identify the households that were affected the most and explore the various strategies utilized by these households to prevent, cope and mitigate the adverse effect o f the crisis. A better understanding o f these issues will be crucial indesigning appropriate instruments for policy response. This paper addresses these gaps in knowledge.Usinga householdpanel data that was collected in two periods (1998 where prices were relatively highand 2001when they were at their lowest) and by specifically exploring the sample heterogeneity to distinguish between coffee and non-coffee households, the paper describes the evolution o f household-level socio-economic welfare measures between the two periods and explores the various mechanisms and strategies employed to deal with the crisis. The paper is divided as follows: the next section describes the data and the various typologies and classifications used to define the coffee sector. An evaluation o f the impact o f the coffee crisis on a number o f socio-economic outcomes i s examined in section 111, while section IV explores risk * Source: I Banco Central de Nicaragua. Indicadores Economicos Mensuales. www.bcn.gob.ni FromLSMS data on employment and agricultural production, about 20 percent o f the rural labor force i s estimated to be directly employed inthe coffee sector while MAGFOR(2002) estimates this to be 40 percent. InterAmerican Development Bank (2001). The remaining 35% are permanent farm workers or farm owners. Government o fNicaragua, Ministry o f Industryand Commerce (MIFIC) and Center o f Export Transactions. These refer to international prices. Cf. 3. Ibid. 3 management strategies available to affected households. Section V addresses how the coffee shock may have influenced poverty mobility and vulnerability while a discussion o f public policy interventionsto address the crisis is presentedinsection VI. Section VI1concludes. 2. DATA, COFFEE TYPOLOGY AND A BASELINEPROFILE OF COFFEE HOUSEHOLDS 2.1 Data sources and coffee typology The maindata source is from theLiving StandardsMeasurementSurveys collectedinNicaragua in 1998 and 2001. The first survey was implemented in the summer o f 1998, while the second during the summer o f 2001. By then coffee prices had reached more than 60 percent o f their 1998 level (Figure 1).More than 4,000 households were surveyed each year, and approximately 3,000 of those surveyed in 1998 were also interviewed in 2001. Taking advantage o f the panel nature o f the data, 2,993 panel households are identified for which data on aggregate consumption and income exists in both years. Since the main focus i s to understand the impact of the coffee crisis (a mainly rural phenomenon), the analysis i s limitedlargely to rural households only and focuses on a final rural panel data o f 1,355 households.' - Figure1:Paneltiming and coffee prices(compositeindex) 200 LSMS2001 150 P k8 n 3 100 al 0 v) 3 50 0 II 1995 1996 1997 1998 1999 2000 2001 2002 Year Source: International Coffee Organization 'Preliminary analysis also includedurbanhouseholds to assess whether or not to incorporate them inthe analysis, While it i s likely that seasonalmigration from urban to ruralregions occurs during coffee harvests, the household survey reveals that most o f this migration occurs within rural areas. Inaddition, since isolating the impact of the coffee crisis per se is a challenging issue, focusing on rural areas alone facilitates this by eliminating any systematic biases inwelfare and other socioeconomic changes that could be due to urban- specific shocks. 4 In order to understand the impact of the coffee shock on households, a number of definitions are used to define how a household relates with coffee. The first definition focuses on household employment activities and classifies a household between "coffee" and "non coffee" based on whether any member o f a household worked in the coffee sector, either as a wage earner or as a producer. Specifically, a household i s defined as: (i) non-coffee ifit was not involved inany coffee activities ineither year; (ii) exiting coffee if it was only involved incoffee activities in 1998; (iii) entering coffee if it was only involved incoffee activities in2001; and (iv) coffee ifit was involved in coffee activities bothyears. The rural panel classifies 293 households involved in coffee activities in at least one of the years of the survey (Tables 1and 2). This represents 24 percent o f the rural panel households out o f which one third (8 percent o f the rural panel) remained inthe coffee sector over the period.' Table 1: Rural households coffee typology (sample sizes) N o n coffee - n o household involvement in coffee activities ineither year 1022 Exit coffee -involved incoffee activities in 1998not in2001 104 Enter coffee -not involved incoffee activities in 1998, yes in2001 117 Incoffee-both 1998and2001 112 Total 1355 Sources: Nicaragua LSMS 1998 and 2001; andNationalAgricultural Census 2001. Table 2: Rural sample structure, extended coffee categories (sample sizes) 2001 Non-coffee Coffee-labor Coffee farmer Total Non-coffee 1022 62 55 1139 2 Coffee-labor 66 31 11 108 Coffee farmer 38 11 59 108 Total 1126 104 125 1355 Sources: Nicaragua LSMS 1998 and 2001 The first definition further distinguishes coffee households between "labor" and "farm". This additional division i s crucial as one o f the key questions that this study tries to address i s how the impact o f the crisis compares among different types o f coffee households. Usingthis distinction, there are 3 1 coffee-labor households and 59 coffee-farm households that remained in coffee both periods While observinghouseholds enter the coffee sector during this periodis counterintuitive, there are two possible explanations: (i)households were already incoffee before the first survey but didnot have coffee income reported in 1998 due its perennial nature; (ii)households entered immediately after the 1998 survey, when coffee prices were still high. O f the 117 households that entered the coffee sector between 1998 and 2001, '62While are labor households and 55 are small farmers. these are weighted estimates usingthe ruralpanel, none o f the two surveys was designed to represent coffee households at the national or any sub-national level, and as such these estimates should only be treated as indicative. 5 (Table 2). It i s important to note that this latter category corresponds mainly to small-scale family farms with an average farm-size of 13 hectares and median o f 5.6 hectares." A third typology defines coffee households based on their activity during the baseline year. Since households may have entered or exited the coffee sector as a response to the shock, attributing changes in various outcomes such as poverty and consumption to the coffee shock cannot be separated from the strategy to "exit" or "stay" in coffee. Inthis sense, the two definitions above are "endogenous" to the outcome, which poses a challenge in measuring the coffee shock's impact. While this i s not always the case, classifying households based on the first year's (1998) affiliation to coffee i s used inthe empirical analysis as an instrument for the two previous definitions: (i) non-coffee ifit was not involved inany coffee activities in 1998; (ii) coffee labor if it was involved incoffee labor activities in 1998; and (iii) coffeefarm ifit was involved incoffee farming activities in 1998 Based on this definition, in 1998 there were 108 coffee-labor households, 108 coffee-farm households and 1139non-coffee households (Table 2). A final broader coffee classificationthat also serves for robustness checks is established using a geographicalbased index of coffee intensity. The small sample size o f coffee households usingthe previous definitions raises a concern about empirical inferences that could be made. Inaddition, given that there are possibly spillover effects betweenthe coffee and non-coffee sectors, it i s important to be able to assess the impact o f the coffee crisis on a more heterogeneous group o f households irrespective o f their direct involvement in coffee." As such, using the 2001 Censo Nacional Agropecuario (Agricultural Census), a municipality-level intensity o f coffee production i s defined as the share o f land dedicated to coffee cultivation. The benefit o f such geographical definition i s that it addresses the concerns above and serves as robustness check for the results obtained from the household definitions but can also look at the geographical aspects o f the impact (if any). Using the distribution o f coffee intensity three coffee regions are defined (low, medium, high).I2Based on the regional coffee definition, 288 households (21 percent o f the rural panel) reside in the high coffee region (Table 3). Box 1summarizes the four definitions above. ~~ loAs neither o fthe two household surveys was designed to represent coffee households at the nationalor any sub-national level, any conclusions should not be interpreted strictly as representing all coffee households in Nicaragua. For example, while the coffee crisis may directly affect the incomes o f agricultural workers, producers and anyone else involved in the production and marketing chain o f coffee, it may also affect the local non-coffee economy via lower demand for other goods or increases inthe labor supply for non-coffeejobs. A municipality is defined as Low coffee intensity ifless than 1.3 percent o fthe farmland is dedicated to coffee (corresponding to the first 3 quintiles o f the coffee intensity variable); medium coffee intensity i s a municipality where 1.4-10.7 percent o f farmland is used for coffee production (corresponding to the fourth quintile o f the coffee intensity variable); and high coflee intensity i s a municipality where 10.8 percent or more o f the total farmland is dedicated to coffee production. 6 Table 3: Regionalcoffee definitionusingcoffee intensity (sample sizes) L o w coffee intensity (< 1.3 % o f total cultivated land) 765 Mediumcoffee intensity (between 1.4 and 10.7 % o f total cultivated land) 302 Highcoffee intensity (> 10.8% oftotal cultivatedland) 288 Total 1355 Sources: Nicaragua LSMS 1998 and 2001; andNationalAgricultural Census 2001. The cultivated landpercentages correspond to the quintiles ofmunicipalities' share of cultivated landincoffee. Inparticular, the first 3 quintiles define the low intensity region, the fourththe mediumandthe fifth (highest) the highintensity region. 1 2 3 4 Any householdmember affiliatedincoffee sector: Coffee productionintensity Usingbothyears Usingbothyears Usinginitial year 1998 inmunicipality Non-coffee bothyears Bothyears: Non-coffee L o w intensity region Coffee-exit Coffee-labor Coffee-labor Mediumintensity region Coffee-enter Coffee-farmer Coffee-farmer Highintensityregion Coffee bothyears 3. ASSESSING THE IMPACT OF THE COFFEE SHOCK 3.1 Baseline Profile: 1998 The rural panel suggests that coffee labor households were among the poorest rural groups during 1998, while coffee farmers were the wealthiest. Inparticular, coffee labor households were the poorest group based on consumption and income levels as well as land assets (Table 4).13In fact, practically all coffee labor households were poor (Table 6). By sharp contrast, coffee farmers were by far the better-off group before the crisis in terms o f welfare and wealth, even compared to non-coffee households. Still, coffee farmers were the least diversified interms o f income sources (with almost 80 percent o f their income derived from farming), suggesting that they would be potentially less able to protect themselves from a coffee shock. I3All group comparisons presented inthis paper are statistically significant at the 90 percent level or more unless otherwise noted. 7 Table 4: Selected household characteristics, 1998 Non-Coffee Exit Coffee Enter Coffee Coffee bothyears Labor Farmer Consumptionper capita (cordobas) 4180 3309 3074 2259 5099 Income per capita (cordobas) 3697 3695 2820 3073 6031 Main income sources (%) Wage agriculture 12 37 21 65 3 Self-employment agriculture 20 29 29 11 78 Wage non-agriculture 31 17 18 7 1 Self-employment non-agriculture 14 6 8 4 2 N o n labor 22 11 25 14 15 Total 100 100 100 100 100 Mean farm size (hectares) 6.5 10.0 6.4 0.7 12.8 Median farm size (hectares) 4.2 4.0 4.2 2.1 5.6 Sources: Nicaragua LSMS 1998 and 2001 Table 6: Poverty Evolution by Coffee Definitions General Poverty Headcountrate Level Change % Change 1998 2001 Coffee labor, then exit 80.5 63.1 -17.4 -21.6 Coffee labor both years 95.5 91.9 -3.6 -3.8 Coffee farmer, then exit 69.3 61.7 -7.6 -10.9 Coffee farmer bothyears 60.9 67.2 6.3 10.3 Sources: Nicaragua LSMS 1998 and 2001. 3.2 Impact onpoverty Overall, the years between 1995 and 2001 are characterized by high economic growth in Nicaragua. Real GDP averaged annual growth rates o f about 5 percent between 1995 and 2001, while GDP per capita grew at a rate o f 2.1 percent per year.14 Partially in response to economic growth, overall poverty declined over this period. In particular, between 1998 and 2001, overall poverty inNicaragua declined by 4 percent to a headcount rate o f 46 percent (Table 5). Even though poverty i s still an overwhelmingly rural phenomenon (as more than two-thirds o f the Nicaragua's poor live inrural areas), poverty rates declined faster inrural areas than in urban areas. In 2001, 64 percent o f the rural were poor (a decline o f six percent from 1998), compared with only 29 percent among the urban population (a decline o f less than 2 percent). Similarly, almost 25 percent of the rural population was classified as extreme poor in 2001 (a decline o f 15 percent from 1998), while only six percent were extreme poor in urban areas (a decline o f less than 2 percent). l4Cf. footnote 1. 8 Table 5: Poverty evolution, by coffee definitions Extreme Poverty General Poverty Headcount Level % Headcount Level Change % rate Change Change rate Change 1998 2001 1998 2001 All Households (fullLSMS comparisons) All 17.3 15.1 -2.2 -12.7 47.9 45.8 -2.1 -4.4 Urban 7.6 6.1 -1.5 -19.7 30.5 28.7 -1.8 -5.9 Rural 28.9 24.7 -4.2 -14.5 68.5 64.3 -4.2 -6.1 Panel households All 21.4 12.7 -8.7 -40.7 46.8 40.1 -6.7 -14.3 Urban 10.1 5.6 -4.5 -44.6 30.2 26.3 -3.9 -12.9 Rural 35.1 21.4 -13.7 -39.0 67.2 58.5 -8.7 -12.9 Household Coffee Definition (rural panel) Non-Coffee (both years) 31.3 16.5 -14.8 -47.3 64.7 54.6 -10.1 -15.6 Coffee - Enter 56.7 43.8 -12.9 -22.8 77.8 76.4 -1.4 -1.8 Coffee - Exit 41.8 32.8 -9.0 -21.5 76.1 62.5 -13.6 -17.9 Coffee (both years) 35.3 37 1.7 4.8 73.6 75.4 1.8 2.4 Regional Coffee Definition (rural panel) L o w Coffee Intensity 31 13.8 -17.2 -55.5 66.1 53.5 -12.6 -19.1 MediumCoffee Intensity 35.3 22 -13.3 -37.7 60.5 54.6 -5.9 -9.8 HighCoffee Intensity 46.3 41.6 -4.7 -10.2 76.9 76 -0.9 -1.2 Sources: NicaraguaLSMS 1998 and 2001; and National Agricultural Census 2001. Nonetheless, the rural panel reveals that coffee-sector households did not benefit from these advances." Inparticular, the poverty rate among households involved inthe coffee sector inboth years increased by 1.8 percentage points to more than 75 percent (Table 5 and Figure 2). Similarly, households that entered the coffee sector before 2001 observed a moderate decline in poverty o f almost two percent. By contrast, poverty rates among households not involved in coffee inboth years and among households that exited coffee after 1998 decreased by more than ten percentage points to 55 and 63 percent, respectively. In fact, attributing (naively) the poverty rates differences between coffee and non-coffee households on the coffee shock alone would suggest that the crisis resultedin a poverty increase o f 11.9percentage points. l5Notethat fromthispoint forward, allcomparisons refer to the panelestimates. 9 Figure 2: Poverty ratechangesby coffee householddefinition 100 76.1 \ \ a 77.8 80 76.4 mrz I _I." 75.4 60 40 20 0 NonCoffee 1998 NonCoffee 2001 Exit 1998 Exit 2001 Enter 1998 Enter2001 Coffee 1998 Coffee 2001 Sources: Nicaragua LSMS 1998 and 2001. Similarly, reduction in extreme poverty was not shared among households involved in coffee activities. While extreme poverty decreased by 47 percent among non-coffee households, and by about 22 percent in households that entered and exited coffee, it increased by 5 percent among households involved in coffee in 1998 and 2001. A similar trend was observed with the regional coffee definition. Still, differentiatingbetween farm and labor households within the coffee sector reveals that while bothwere affected negativelyfarm householdswere hit the most. Infact, only coffee farm households experienced increases in poverty rates (seven percent). By contrast, poverty among labor households decreased by four percent even though it did at a lower rate compared to non-coffee households (Tables 5 and 6). This implies that while coffee labor households were poorer as noted earlier, the coffee crisis shock affected them less compared to coffee farm households. Understanding and comparing the various coping strategies betweenthe two groups i s therefore crucial. The regionalcoffee definition confirmsthe abovepatterns.Duringboth 1998 and2001, poverty in the high coffee intensity region was high compared to l o w and medium coffee intensity regions (Table 5 and Figure 3). Poverty rates among households in high coffee intensity regions remained above 75 percent while among households in low and medium intensity regions decreased by 13 and 6 percentage points, respectively. These trends and the corresponding impact o f the coffee shock on poverty rates using this definition (a suggested impact o f 11.7 percentage points) are both consistent with the household definitions discussed above. l6Extreme poverty declined inall regions, but the increase was more than 5 times greater among low-intensity coffee regions (56 percent) vis-a-vis high-intensity coffee regions, where extreme poverty fell by 10 percent. 10 Figure 3: Poverty rate changes by regional coffee 100 -b 80 76.9 7/; 60 40 20 0 Low Coffee 1998 Low Coffee 2001 MediumCoffee 1998 MediumCoffee2001 HighCoffee 1998 HighCoffee2001 ! urces: Nicaragua LSMS 1998and2001andNationalAgricultural Census2001. 3.3 Consumption Between 1998 and 2001, real consumption per capita inrural areas increased an average of 11.7 percent, or 470 Cordobas (Table 7). This increase was driven mainly by an increase inconsumption o f non-food items (e.g., non-durable household goods, clothing, transportation, etc.) o f 28.1 percent (or 9.4 percent per year). By contrast, average food consumption practically remained the same, increasing by less than 1percent over the three-year period. 11 Table 7: Nicaragua: changesinper capita consumption,by coffee definitions Type of Household 1998 2001 % Change All Rural Total Consumption 4,010 4,480 11.7 Food Consumption 2,440 2,457 0.7 Non-Food Consumption 1,570 2,012 28.1 Household Coffee Definition a Non-Coffee (both years) Total Consumption 4,180 4,806 15.0 Food Consumption 2,515 2,609 3.7 Non-Food Consumption 1,664 2,185 31.3 Coffee - Exit Total Consumption 3,309 3,812 15.2 Food Consumption 2,242 2,334 4.1 Non-Food Consumption 1,066 1,478 38.6 Coffee - Enter Total Consumption 3,074 3,113 1.3 Food Consumption 2,019 1,763 -12.7 Non-Food Consumption 1,055 1,336 26.6 Coffee (both years) Total Consumption 3,881 3,248 -16.3 Food Consumption 2,285 1,771 -22.5 Non-Food Consumption 1,596 1,477 -7.5 Regional Coffee Definition * Low Coffee Intensity Total Consumption 4,074 4,723 15.9 Food Consumption 2,485 2,596 4.4 Non-Food Consumption 1,589 2,109 32.7 Medium Coffee Intensity Total Consumption 4,363 4,911 12.5 Food Consumption 2,576 2,605 1.1 Non-Food Consumption 1,787 2,304 28.9 High Coffee Intensity Total Consumption 3,491 3,395 -2.1 Food Consumption 2,183 1,933 -1 1.5 Non-Food Consumption 1,308 1,463 11.8 Sources: Nicaragua LSMS 1998 and 2001; and National Agricultural Census 2001. All values are in 1998 cordobas (C$) per capita. Average exchange rate 1998: C$10.58 /US$ 1.00. a Household coffee definitions are based on the household's involvement in the coffee sector in either years. Specifically, a household i s defined as: (i)coffee household if it was involved in the coffee sector in both years (1 12 observations); (ii) non- coffee household if i t was not involved in any coffee activities in both years (1,022 observations); (iii)exiting coffee if the household was involved in coffee activities in 1998 but not in 2001 (104 observations); and (iv) entering coffee if a household was not involved in the coffee sector in 1998 but was in 2001 (1 17 observations). Regional coffee definitions are based on the municipal-level average o f proportion o f farm size dedicated to coffee production. Low = 0-1.3% (765 observations), medium = 1.4-10.7% (302 observations) and high = 10.8% or more o f average farm size is dedicated to coffee (288 observations) 12 In contrast, households that were involved in the coffee sector in both years experienced significant declinesinper capita consumption. While consumptionper capita increased 15 percent among non-coffee households, it decreased more than 16 percent among coffee households (Table 7 and Figure 4). Households that exited coffee production between 1998 and 2001 experienced an increase o f consumption o f 15 percent, whereas consumption remained unchanged among households that entered the coffee sector after 1998. Figure4: Ruralconsumptionper capita by householdcoffee definition 5.000 I i NonCoffee NonCoffee Exit 1998 Exit2001 Enter 1998 Enter2001 Coffee 1998 Coffee2001 1998 2001 ElFoodexuenditures Non-FoodExuenditures Sources: Nicaragua LSMS 1998 and 2001. Consistent with the poverty trends above, the consumption decline was more severe among farm as opposed to labor coffee households. Consumption per capita decreased more than 25 percent among farm households while consumption among coffee labor households remained the same (Table 9). Table 9: Consumptionandincomeamong coffee households Consumption 1998 2001 Level Change % Change Coffee labor, then exit 3,071 3,620 549 27.6 Coffee labor bothyears 2,259 2,219 -40 -1.8 Coffee farmer, then exit 3,679 4,113 434 11.8 Coffee farmer ,bothyears 5,099 3,790 -1,309 -25.7 Income Coffee labor, then exit 4,019 3,990 -29 -0.7 Coffee labor bothyears 3,074 2,976 -98 -3.2 Coffee farmer, thenexit 3,190 4,381 1,191 37.3 Coffee farmer, both years 6,03 1 3,696 2,335 -38.7 Sources: Nicaragua LSMS 1998 and 2001. 13 Similar patterns are observed using the regional coffee definition. Inparticular, total consumption per capita in low-intensity coffee areas increased by almost 16 percent between 1998 and 2001, in contrast with a 3 percent decrease in high-intensity regions (Figure 5)." This finding is consistent with the evolution o fpovertywithinthese regions. Figure 5: Rural consumption per capita by regional coffee definition 5,000 4,000 .-3e 3 k 33000 .r 0 c gP 2,000 s m E 1,000 0 Low 1998 Low 2001 Medium 1998Medium2001 High1998 High2001 I E3Food expenditures WNon-FoodExpenditures 1 Sources: Sicaragua LSMS 1998 and 2001 and National Agricultural Census 2001. The drop in overall consumption of coffee households was driven by a decline in food consumption. Decomposition o f consumption per-capita into its food and non-food components allows the identification o f the source in consumption changes. For non-coffee households, while food consumption was similar between 1998 and 2001, the non-food component increased by more than 30 percent (Figure 4 and Table 5). Conversely, while coffee households experienced drops in both consumption components, the largest drop was in food consumption (23 percent). Similar patterns hold usingthe regional coffee definition. 3.4 Income Mirroring the previous patterns, coffee households experienced large declines in incomes. Overall, between 1998 and 2001 real rural incomes per capita increased by 30 percent. Still, comparisons using the coffee definitions reveal distinct differences for each subgroup. For example, income per capita increased by 40 percent for non-coffee households (Table 8 and Figure 6). Similar increases are found in the low intensity coffee region. By sharp contrast, households involved in coffee inbothperiods suffered a decrease inper capita income o fmore than 25 percent. l7This decrease was not statistically significant. 14 IA 3 Figure6: Changesinper capitaincome 6,000 5,000 4,000 3,000 2,000 1,000 0 Coffee labor Exit Non-Coffee (both Coffee producer Coffee labor (both Coffee producer years) Exit years) (both years) Sources: Nicaragua LSMS 1998 and 2001. Nonetheless,coffee farmhouseholdswere hitthe worst. Infact, while they had the highest average incomes per capita in 1998, by 2001 it was among the lowest. Using the household coffee definition, income per capita for coffee farm households was 6,03 1 Cordobas, compared to 3,697 for non-coffee households in 1998 (Tables 8 and 9). This pattern completely reversed in 2001 with coffee farm households experiencing a 40 percent decrease in incomes while non-coffee households saw a 40 percent increase in incomes. On the other hand, incomes for coffee labor households changed little betweenthe 2 periods (Table 9), to a large part reflecting the price effect on agricultural income. 3.5 Health and Education Childmalnutritionremained unchangedwithin coffee regions between 1998 and 2001. Despite the fact that overall, incidences o f various malnutrition measures such as stunting, wasting and underweight showed improvement during the period (national declines o f 35, 11, and 73 percent, respectively), these gains were not enjoyed equally by children o f all regions." As figures 7 and 8 reveal, the Central Rural region -where more than 80 percent o f Nicaragua's coffee production i s concentrated - the incidence o f underweight children changed very little while for chronic malnutrition (stunting) actually appears to have slightly increased. Both malnutrition incidences for the Central Rural region were the highest inthe country duringboth periods and these trends suggests '*Stunting (height-for-age) reflects chronic malnutrition, which results from years o f retarded skeletal growth and is associated with poor economic conditions; wasting (weight-for-height) captures deficiencies infat tissue and indicates food loss from a short-term, emergency situation; and underweight (weight-for-age) combines the previous two measures and reflects total malnutrition. A child (of usually 5 years or less) i s considered "stunted", "wasted" or "underweight" if hidher corresponding anthropometric measure i s two or more standard deviations below the median o f the internationally recognized reference population. Also see Marini and Gragnolati 2002, and Chawla 2001. 16 that the coffee crisis had a negative effect on the nutritional status o f children younger than 5 years in the region (inthe sense o f at not enjoying the gains experienced elsewhere). Figure 7: Incidence of Underweight Children, 1998 - 2001 20 1s c e" 10 5 0 Managua Pacific Pacific Rural Central Central Rural Atlantic Atlantic Urban Urban Urban Rural 1 01998 w2001 1 Sources: Nicaragua LSMS 1998 and 2001. Figure 8: Nicaragua incidence of Stunting, 1998- 2001 - 30 * 20 a E6 10 0 Managua Pacific Urban Pacific Rural Central Central Rural Atlantic Atlantic Urban Urban Rural 1998 w2001 Sources: Nicaragua LSMS 1998 and 2001. In educational outcomes, despite large increases in enrollment rates at both the primary and secondary levels, overall, primary enrollment rates among coffee households fell and secondary enrolment rates hardly changed between 1998 and 2001. Among non-coffee households, primary net enrollment rates increased from 78 to 86 percent (Figure 9). By contrast, enrollment rates among 17 households involved in the coffee sector in both periods decreased from 77 to 72 percent. At the same time, secondary net enrollment rates almost doubled among non-coffee (to 40 percent), while remaining essentially unchanged among coffee-sector households over the period (at around ten percent; Figure 10). While not attributing these differences solely on the coffee crisis, it i s possible that these patterns reflect harmful coping strategies among coffee households. The next session addresses this issue inmore detail. Figure 9: Rural net primary enrollment rates (7-12 year olds) loo ~ 80 60 40 20 0 Non-Coffee (both Coffee Exit - Coffee Enter - Coffee (both years) years) 101998 H2001 1 Sources: Nicaragua LSMS 1998 and 2001. Figure 10: Rural net secondary enrollment rates,(13-17 year olds) 20 0 Non-Coffee (both Coffee-Exit Coffee Enter - Coffee (both years) years) 101998 W2001 I Sources: Nicaragua LSMS 1998 and 2001. 18 In summary, descriptive statistics suggest that households related to coffee activities did not benefit from an otherwise period of growth in Nicaragua. In fact, most socio-economic indicators for these households have worsened between 1998 and 2001, a period that saw coffee prices declinedby more than half. While accurately quantifying the impact that the coffee shock may have had i s challenging, the big magnitude cast little doubt that the coffee shock had a strong impact on coffee farm households and to a smaller effect coffee labor households. The next section explores the various strategies that these households used to mitigate, cope or prevent the shock and the extent by which informal insurance mechanisms to smoothconsumptionwere available. 4. RISK MANAGEMENT STRATEGIESAND RESPONDINGTO SHOCKS 4.1 Do households self-insure? The role of risk and insurance on householdbehavior is well documentedin the literature." As poor households make consumption decisions in uncertain environments, they face many risks: idiosyncratic risks that affect a specific household (illness, death, unemployment); or covariate risks that affect everyone within a particular region or group (droughts, hurricanes, terms o ftrade shocks or macroeconomic volatility). The question as to whether some households are better able to use formal or informal mechanisms to minimize the impact o f such risks on their consumption i s therefore key in designing policies that provide insurance or safety nets mechanisms. The previoussectionrevealedthat coffeehouseholdswere adversely affected by the coffee shock in terms a number of differentwelfare dimensions. Inthe context o f the coffee shock a number o f questions arise: were affected households able to protect against the negative income decline? H o w does their ability to insure (or not) compares with non-coffee households? Are there differences among coffee households? A number of empirical approaches have been used that address these questions of self- insurance and consumption smoothing. The most common is to fit an equation that looks at how changes in consumption correlate with income changes.20The typical specification i s derived from a consumption equation o f the initial form: where InC, i s the log o f consumption per capita o f household i inperiod t, lnYit i s the log o f income at time t, X, i s a vector o f socio-economic characteristics, a,/?and y are parameters to be estimated, vi i s a household fixed effect and wit i s an i.i.d. error term. By differencingequation 1(between the two years), the specificationbecomes: where A denotes changes over the two periods o f the respective variables. Estimating equation 2 will give unbiased estimates o f the coefficients. ''For example, Alderman and Paxson(1992), Townsend (1994), Jalan and Ravallion (1999). 20See Townsend (1994), Ravallion and Chaudhuri (1997) and Grimard (1997) for some examples. 19 The basic test of consumption insurance i s the extent to which householdincome co-varies with consumption. Ifhouseholds are fully insured against income shocks, then changes inincome do not affect consumption and Z=O. The extent to which 3 differs from zero indicates how insulated (or exposed) a household's consumptioni s to income shocks.21 In the case of the coffee shock, an additional empirical challenge is to correctly model the coffee crisis since it i s covariate shock that only affects a subgroup of the population. Specifically, it i s important to be able to test for differentiated impacts on consumption among different types o f households, based on whether they participated in coffee activities or resided in a coffee region (as defined earlier). Nonetheless, two o f the coffee definitions are endogenous in the sense that the decision to enter, exit or stay in coffee i s endogenous to consumption changes. As such, the final empirical strategy implemented here i s to estimate coffee-group specific models using equation 2. That is, for each coffee classification, a consumption changes i s regressed on income changes (AY ) and household size changes (L W ) This~ . ~ avoids the endogeneity issue since the only interest is to test the specific group's ability to se1f-insu1-e.'~ The overall results reject the full insurance hypothesis. Estimating Equation (1) suggests that more than fourteen percent o f an income shock i s passed onto current consumption (Table These effects are similar by estimating this on food andnon-food consumption. 21The intercept a captures aggregate income risk. 22This is a similar estimation strategy adopted by Jalan and Ravallion (1999). 23An alternative approach would be to estimate an augmented equation 2 using coffee dummies interacted by income changes to test the full insurance model and exploring differentiated insurance ability among various coffee categories. This approach has the advantage o f using the entire sample, w b c h is attractive due to the small sample sizes o f coffee categories using the specification o f equation 2. While estimating this specification resulted in similar results, they are not reported due to concerns o n the endogeneity of some o f the coffee classifications. 24These magnitudes are consistent with the ones typically found inthe literature. See also Skoufias and Quisumbirng(2002). 20 Table 10: Consumption smoothing: incomechanges coefficients Total Food Non-Food Allrural 0.14*** 0.14*** 0.13*** Coffee definitions Non-Coffee 0.14*** 0.14*** 0.13*** Exited Coffee 0.07 0.12 -0.01 Enteredcoffee 0.07 0.01 0.13 Coffee labor both years 0.22* 0.43* 0.08 Coffee farmer both years 0.20** 0.12 0.34** Initial coffee classifications Non-Coffee in 1998 0.14*** 0.13 0.13*** Coffee labor in 1998 0.12* 0.18** 0.14 Coffee farmer in 1998 0.19*** 0.16** 0.24* Regional coffee definitions L o w coffee intensity 0.14*** 0.14*** 0.12*** Mediumcoffee intensity 0.14*** 0.12** 0.13*** Highcoffee intensity 0.13*** 0.11* 0.16*** Dependent Variable: L o g o f change inconsumption per capita Each coefficient comes from estimating a fixed effects model o f consumptionper capita changes regressed on income per capita changes and household size changes for the corresponding coffee classification. Bothregressors are treated as exogenous. The municipal level fixed effects are jointly significant for all the specifications. * significant at 10%; ** significant at 5%; *** significant at 1% Estimation of equation 2 using coffee-specific models suggests that income shocks have a heterogeneousimpact among different rural subgroups. For example, using the first two coffee definitions, given an overall impact o f income shocks on consumption that i s similar for coffee and non-coffee households, the former are significantly less able to self-insure (Table 10). Specifically, for every dollar o f income decrease, coffee-labor households decrease consumption by 22 cents while coffee labor households by 20 cents. Comparing self-insurance abilities for food consumption, the results indicate that coffee-labor households are vulnerable to insuring food consumptionwhile coffee-farm households are not. Specifically, more than 43 percent o f an income shock among coffee-labor households i s passed through food consumption decreases. By contrast, among coffee-farm households, the effect i s not significant suggesting that income shocks do not translate into food consumption decreases. To the extent that coffee-labor households were the poorest in both periods, these findings imply that they were also the most vulnerable to income risks. As such in improving insurance mechanisms and risk reduction inrural Nicaragua, special attention on the poorer and more vulnerable populations (such as coffee labor households) may be a priority. This finding i s consistent with literature from other countries that suggest that the poorest households are also those least able to smooth c o n ~ u m p t i o n . ~ ~ The abilityto insurenon-food consumptionagainst incomeshocksis smaller among coffee-farm households.For example, among households that remained in coffee farming in both periods, non- food consumption changes decreased by 34 cents for every dollar decrease in income. A similar pattern i s observed usingthe other coffee household definition (even though the overall magnitude i s smaller). 25Jalan and Ravallion (1999). 21 Interestingly, households that exited and entered the coffee sector seem to be able to "insure" against income fluctuations. The non-significance o f the income coefficient for both groups suggests that these households were better able to insulate their consumption from income shocks (Table While for households that exited the coffee sector, this could be suggesting that mobility and adaptability to changing economic conditions may be important in determining how households insure against shocks, it i s unclear as to why that may be the case for household that entered coffee (but the small sample sizes for both groups may explain these results). Nonetheless, as discussed below, income diversification innon-agricultural activities seems to have allowed some households to stabilize consumption patterns. Understanding the process o f coffee entering or exiting may therefore be important. 4.2 Risk management strategies Exposure to risk in general does not necessarily translate in adverse outcomes. In fact, if households have access to a sufficient portfolio o f options that can allow them to manage the realization o f risk (the shock), then exposure to risk i s not an issue. This i s not the case in most cases and the results above do suggest that rural households in Nicaragua are not able to fully protect themselves against risk exposure. As such, a better understanding of the various risk management strategies employed by rural households to cope with risksis important. Typically it i s useful to separate such strategies into ex- ante and ex-p~st.~'Ex-ante mechanisms address what households (and to that extent, public and private instruments) can do to reduce or prevent the occurrence o f risks and mitigate the impact o f risk if an adverse event occurs. Some examples o f ex-ante mechanisms are crop insurance, exiting a risky occupation, income diversification. On the other hand, ex-post mechanisms address the ability o f households to respond after a risk has been realized (for example taking children out o f school or selling assets). Exploring whether these risk management strategies and mechanisms exist or vary across different households is also instrumental for policy design. This section exploreswhat strategies,if any, have allowed rural households to address exposure to various risks, with emphasis on the coffee shock. To facilitate the analysis, inaddition to ex-ante and ex-post strategies, risk management strategies are hrther grouped in: (i)labor market adjustments; (ii) precautionary savings; and (iii) informal insurance. Inprinciple, all three strategies can be both ex-ante and ex-post. Finally exiting the coffee sector as a response to the shock i s also considered as a coping strategy. Empirically, there are a number of approaches to explore the role of various risk management mechanisms on householdwelfare. Typically, data on a household's response as a result o f realized risks can be used to assess the existence and use o f the various mechanisms mentioned above. Since the Nicaragua survey did not collect such information a few alternative methodological strategies are implemented. Denoting 2 to be a vector o f potential risk management instruments available to the household the initial period (for example assets, labor supply), the first approach entails estimating a consumptiongrowth model o f the form: ''Holzmannand 26Similar results were obtained with changes infood and non-food consumption. Jorgensen (2000). 22 where xiand Zi are as previously defined above, 6,,6, and 8, are parameters to be estimated and vi is an iid. error term. Estimating equation 3 can allow indirect inferences on the existence of a particular risk management instrument vis-ti-vis consumption growth. Specifically, testing whether a specific instrument2is correlated withconsumption growthover the periodis interpretedas weak evidence of a positive role for that instrument in addressing risk. For example, finding a positive relationship between the initial level of remittances and consumption growth i s interpreted as evidence that migration was a potentially important strategy for households (and possibly against exposure to risk). As with the insurance models above and due to the similar endogeneity concerns, equation 3 is estimated for each of the coffee definitions separately so as to assess the existence of risk management instruments among each specific subgroup. The results are presented in Tables 11 through 19, the dependent variable being the change in total, food, and non-food consumption, respectively. Table 11:Consumptiongrowthand coping,by coffee householddefinition Non-Coffee Exited Entered Labor-bothyear Farm-both years Baseline period household characteristics (1998) Family size 0.04*** 0.07 0.04 0.04 0.03 Maximumyears o f education inhousehold 0.02** 0.02 -0.01 0.01 -0.01 Number ofkidsworkers -0.01 0.00 0.03 -0.03 -0.12 Number of adult workers -0.02 -0.13 0.07 -0.23 0.03 Number ofincome sources 0.04* 0.13 -0.21** 0.04 -0.17 Land owned (hectares) 0.00 -0.00 0.00 0.04 0.01 Receivedremittances (yes=l) 0.06 0.08 0.14 -0.15 -0.30 Distance to Managua (10 minute intervals) -0.00 0.02 0.02 0.08 0.03 Elevation (100 meters) 0.00 0.01 0.05 -0.08 0.09 Affectedby Mitch(yes=l) 0.00 0.00 0.00 0.00 0.00 Constant -0.27** -1.20 -0.59 -0.97 -1.35 Observations 1022 104 117 31 59 R-squared 0.23 0.55 0.74 0.86 0.55 DependentVariable: Change in(log) per capita consumption. Additional controls: municipality fixed effects. *significant at 10%;** significant at 5%;***significant at 1%. 23 Table 12: Foodconsumptiongrowth and coping, by coffee household definition Non-Coffee Exited Entered Labor-both Farm-both year years Baseline period household characteristics (1998) Family size 0.04*** 0.05 0.06** 0.02 0.01 Maximumyears of education in 0.02** 0.04 0.03 0.02 0.00 household Number of kidsworkers -0.00 -0.02 -0.06 0.11 0.05 Number of adult workers -0.01 -0.07 -0.13 -0.18 0.03 Number of income sources 0.04 0.10 -0.20 0.07 -0.12 Land owned (hectares) 0.00 -0.00 0.01 -0.01 -0.01 Receivedremittances (yes=l) -0.02 -0.08 0.19 -1.05 -0.36 Distance to Managua (10 minuteintervals) -0.00 0.04 0.01 0.10 0.03 Elevation (100 meters) 0.00 0.05 0.10 -0.14 0.09 Affected by Mitch(yes=l) 0.00 0.00 0.00 0.00 0.00 Constant -0.41"" -1.79 -0.93 -1.27 -1.66 Observations 1022 104 117 31 59 R-squared 0.22 0.50 0.77 0.79 0.59 DependentVariable: Change in(log) per capita food consumption. Additional controls: municipality fixed effects. * significant at 10%;** significant at 5%;***significant at 1%. Table 13: Non-foodconsumptiongrowth and coping, by coffee householddefinition Non-Coffee Exited Entered Labor-both Farm-both year years Baselineperiod household characteristics (1998) Family size 0.04*** 0.11** 0.00 0.10 0.07* Maximumyears of education inhousehold 0.00 -0.01 -0.05 -0.05 -0.02 Numberofkids workers -0.01 -0.01 0.17 -0.29 -0.38*** Numberof adult workers -0.04 -0.23* 0.31** -0.49 0.06 Numberof income sources 0.04* 0.17 -0.24 0.15 -0.29 Land owned (hectares) 0.00 -0.00 -0.00 0.15 0.02** Receivedremittances (yes=l) 0.15"" 0.39 0.20 0.99 -0.23 Distanceto Managua (10 minuteintervals) -0.01 0.00 0.02 -0.00 0.04 Elevation (100 meters) 0.01 -0.06 -0.04 -0.04 0.10 Affected by Mitch(yes=l) 0.00 0.00 0.00 0.00 0.00 Constant -0.02 -0.23 0.35 0.55 -1.25 Observations 1022 104 117 31 59 R-squared 0.23 0.65 0.61 0.80 0.56 DependentVariable: Change in(log) per capita non-food consumption. Additional controls: municipality fixed effects. *significant at 10%; ** significant at 5%;***significant at 1%. 24 Table 14: Consumptiongrowthand coping, by 1998 coffeehouseholddefinition Activity in 1998 Non-Coffee Coffee labor Coffee farmer Baseline periodhousehold characteristics (1998) Family size 0.02 0.02 0.04* ** Maximumyears of education inhousehold -0.01 0.03 0.02** Number o fkidsworkers -0.00 -0.06 -0.01 Number of adult workers -0.14 0.05 -0.01 Number of income sources 0.03* -0.11 0.03 Land owned (hectares) 0.04 -o.oo* * 0.00 Received remittances (yes=l) -0.03 0.34 0.07 Distance to Managua (10 minute intervals) 0.02 0.01 -0.00 Elevation (100 meters) -0.01 0.09* -0.01 Affected by Mitch(yes=l) 0.00 0.00 0.00 Constant -0.10 -0.97 -0.20 Observations 108 108 1139 R-squared 0.61 0.44 0.22 DependentVariable: Change in(log) per capita consumption. Additional controls: municipality fixed effects. * significant at 10%;** significantat 5%;***significant at 1%. Table 15: Foodconsumptiongrowthandcoping, by 1998 coffee householddefinition Activity in 1998 Non-Coffee Coffee labor Coffee farmer Baseline period household characteristics (1998) Family size 0.00 0.02 0.04*** Maximumyears of education inhousehold -0.00 0.01 0.02** * Number ofkidsworkers -0.01 -0.01 -0.00 Number ofadult workers -0.11 0.04 -0.01 Number ofincome sources -0.03 -0.01 0.02 Land owned (hectares) 0.05 -o.oo* 0.00 Received remittances (yes=l) -0.19 0.11 -0.01 Distance to Managua (10 minuteintervals) 0.01 0.03 -0.00 Elevation (100 meters) 0.02 0.15** -0.02 Affected by Mitch(yes=l) 0.00 0.00 0.00 Constant -0.17 -2.24* ** -0.29* Observations 108 108 1139 R-squared 0.59 0.49 0.20 DependentVariable: Change in(log) per capita food consumption. Additional controls: municipality fixed effects. * significant at 10%;** significantat 5%;***significant at 1%. 25 Table 16: Non-foodconsumptiongrowth and coping, by 1998 coffeehouseholddefinition Activity in 1998 Non-Coffee Coffee labor Coffee farmer Baseline period household characteristics (1998) Family size 0.07* 0.05 0.04*** Maximumyears o f education inhousehold -0.03 0.05 0.00 Number of luds workers 0.01 -0.17* -0.00 Number of adult workers -0.25** 0.06 -0.03 Number of income sources 0.15 -0.27* 0.03 Land owned (hectares) 0.03 -o.oo* 0.00 Receivedremittances (yes=l) 0.15 0.76* 0.15** Distance to Managua (10 minuteintervals) 0.01 -0.02 -0.01 Elevation (100 meters) -0.10 0.00 0.01 Affectedby Mitch(yes=l) 0.00 0.00 0.00 Constant 0.38 0.77 0.00 Observations 108 108 1139 R-squared 0.56 0.40 0.22 Dependent Variable: Change in(log) per capita non-food consumption. Additional controls: municipality fixed effects. * significant at 10%; ** significant at 5%; *** significant at 1%. Table 17: Consumptiongrowth and coping,by regionalcoffee definition Coffee intensity inmunicipality Low Medium High Baseline period household characteristics (1998) Family size 0.04*** 0.05*** 0.02* Maximum years of education inhousehold 0.01 0.01 0.03** Number ofludsworkers -0.01 -0.03 -0.00 Number of adult workers -0.01 -0.01 -0.04 Number of income sources 0.03 0.01 0.03 Land owned (hectares) 0.00 -0.00 -0.01*** Receivedremittances (yes=l) 0.06 0.10 -0.02 Distance to Managua (10 minuteintervals) -0.00 -0.03** 0.00 Elevation (100 meters) 0.02 -0.04 0.01 Affected by Mitch(yes=l) 0.00 0.00 0.00 Constant -0.26" 0.20 -0.47 Observations 765 302 288 R-squared 0.21 0.21 0.25 DependentVariable: Change in(log) per capita consumption. Additional controls: municipality fixed effects. * significant at 10%;** significant at 5%;*** significant at 1%. 26 Table 18: Foodconsumptiongrowth and coping, by regionalcoffee definition Coffee intensity inmunicipality L o w Medium High Baselineperiod household characteristics (1998) Family size 0.03** 0.05*** 0.02 Maximumyears o f education inhousehold 0*02* 0.00 0.02* Number o f kids workers 0.00 -0.03 0.00 Number o f adult workers 0.02 -0.04 -0.03 Number o f income sources 0.01 0.04 0.03 Land owned (hectares) 0.00 -0.00 -0.01 *** Received remittances (yes=l) -0.04 0.06 -0.15 Distance to Managua (10 minute intervals) -0.00 -0.02 0.02 Elevation (100 meters) 0.03 -0.04 0.02 Affected by Mitch (yes=l) 0.00 0.00 0.00 Constant -0.40** 0.12 -0.86** Observations 765 302 288 R-squared 0.19 0.19 0.26 Dependent Variable: Change in(log) per capita food consumption. Additional controls: municipality fixed effects. * significant at 10%; ** significant at 5%;*** significant at 1%. Table 19: Non-food consumptiongrowth and coping, by regionalcoffee definition Coffee intensity inmunicipality L o w Medium High Baselineperiod household characteristics (1998) Family size 0.04*** 0.06** 0.04** Maximum years o f education inhousehold -0.00 -0.00 0.01 Number o f kids workers -0.00 -0.03 -0.03 Number o f adult workers -0.04 -0.01 -0.03 Number o f income sources 0.04 -0.02 -0.01 Land owned (hectares) 0.00 0.00 -0.01 *** Receivedremittances (yes=l) 0.20** 0.18 0.12 Distance to Managua (10 minute intervals) -0.00 -0.03** -0.01 Elevation (100 meters) 0.01 -0.02 0.00 Affected by Mitch (yes=l) 0.00 0.00 0.00 Constant -0.00 0.49 0.20 Observations 765 302 288 R-squared 0.22 0.19 0.19 DependentVariable: Change in(log) per capita non- food consumption. Additional controls: municipality fixed effects. *significant at 10%; ** significant at 5%;*** significantat 1%. 27 A second approach is to directly test whether a household used a specific coping instrument. Empirically this can be implementedby estimating a probability model o f the form: Prob(AZi =1)=f (Eek coffeeik ti> Y-1 + (4) k=l where Mi denotes a positive use o f that risk management instrument. For example, Mi could be the change in a household's child labor allocation over the period. In this case, by differentiating among households based on their affiliation with coffee activities, a positive B for say, coffee laborers, would suggest that these households were more likely to engage in harmful coping mechanisms such as child labor due to the coffee shock. To further explore coping abilities among coffee households, equation 4 i s also estimated controlling for whether a household was poor in 1998, capturing heterogeneous coping ability between poor and less poor coffee households. The results for these estimations are presented inTables 20 through 23. The results from both approaches described above, complimented by descriptive statistics are summarized below.'' 4.3 Labor market adjustments Household diversification in non-agricultural activities plays an important role for rural welfare and coping with shocks. Non-coffee households that were more income diversified in 1998 (measure by the number o f different agricultural and non-agricultural income sources in the household) were more likely to experience consumption growth (for example Tables 11, 13 and 14).29 By contrast, diversification among coffee labor and farm households did not affect consumption growth. One important distinction that may explain these patterns i s the observation that while non- coffee households were diversified in both agricultural and non-agricultural activities, coffee households were mainly "diversified" only within the agricultural sector (Tables 4 and 8, Figures 11- 13). As such, these patterns suggest that access to non-agricultural activities may be a key instrument for bothriskmitigation and consumption growth in general. 28All models discussed inthis section also control for municipality level fixed effects, andwhether the household resides ina hurricane Mitch affected municipality, the other covariate shock during this period. 29This is consistent with Beneke and Gonzalez-Vega (2000) who find positive effects o f income diversification on income growth inEl Salvador. 28 Figure 11: Sources of rural income per capita by householdcoffee definition 6000 , 5000 'E 4000 d 3 kE 3000 - 2000 1000 0 Non- Non- Exit 1998 Exit 2001 Entry Entry Coffee Coffee Coffee Coffee 1998 2001 1998 2001 1998 2001 0Wage agriculture HSelf-employmentagriculture 0Wage non-agriculture 0Self-employmentnon-agriculture W Non labor Sources: Nicaragua LSMS 1998 and 2001. Figure 12: Sources of ruralincome per capita by regional coffee definition 6000 5000 'E 4000 d 3 kE 3000 -8c 2000 1000 0 Low 1998 Low 2001 Medium 1998 Medium 2001 High 1998 High 2001 OWage agriculture W Self-employment agriculture 0Wage non-agriculture 0Self-employment non-agriculture W Non labor Sources: Nicaragua LSMS 1998 and 2001 and National Agricultural Census 2001. 29 Figure13: Adjustmentsto incomeby incomesource 1 2000 I 1000 0 -1000 Exit coffee farming Coffee-laborboth years years -2000 i -3000 ' E lWage agriculture W Self-employmentagriculture 0Wage non-agriculture 0Self-employmentnon-agriculture Non labor Sources: Nicaragua LSMS 1998 and 2001 andNational Agricultural Census 2001. Consistent with the above, examination of income portfolio adjustments indicates that households that increased non-agricultural incomes faired better. For example, among households that exited coffee over the period, the main income increases were due to increases in non-agricultural income (Table 8 and Figure 13). In addition, while coffee labor households who exited coffee mainly diverted their efforts to non-agricultural labor (wage) activities, coffee farm households that exited coffee shifted labor to non-agricultural enterprises (self-employment). This i s indicative o f the constraints for poorer households (coffee labor) to take advantage o f higher return occupations in the non-agricultural sector. Nonetheless, the fact that these households did exit coffee highlightsthe importance o f understanding the determinants o f both upward income mobility and the ability to diversify into non-agricultural activities. The empiricalresultsalso implythat coffee households engagedin harmfulcoping activitiesvia increases in child labor, directly affecting school enrollment.Over the period o f the study, child labor incidence increased in rural Nicaragua by 24 percent (Figure 14). While this incidence has decreased among coffee households (Figure 14), the average total weekly hours worked by children among coffee households significantly increased compared with a decrease for child workers in non- coffee households (Figure 16 and 17).30Inaddition, households residing in the high coffee intensity region were significantly more likely to increase child labor (Table 21). Consistent with these trends, school attendance decreased among children in coffee households while it increased for non-coffee households (Figures 9, 10 and 18). 30The labor force participation among coffee households may be due to a shrinking demand for labor, corroborated by the higher unemployment rate among coffee households (Figure 15). 30 Figure 14: The coffee crisis and childlabor: labor force participation (ages 6-14) 30 c 0 .- 20 .-0 6 (I P 10 i? 0 Non coffee both Coffee Exit Coffee Entry Coffee both years years ID1998 W20011 I ;ources: Nicaragua LSMS 1998 and 2001. Figure 15: The coffee crisis and childlabor: unemployment (ages 6-14) Non coffee both Coffee Exit Coffee Entry Coffee both years years 101998 W2001 I Sources: Nicaragua LSMS 1998 and 2001. 31 Figure 16: The coffee crisis and childlabor: hoursworked (ages 6-14) 40 30 7% 2 a 20 L: B x P 0 10 0 Noncoffee both Coffee Exit Coffee Entry Coffee both years years (01998 M2001I Sources: Nicaragua LSMS 1998 and 2001. Figure 17: The coffee crisis and child labor: hoursworked (ages 6-14) Coffee both years labor - Coffee both years- selfemployed 101998 02001 1 I Sources: Nicaragua LSMS 1998 and 2001. 32 M M Figure 18: The coffee crisis and childlabor: primary school enrolment (ages 6-14) Coffee both years labor Coffee both years self employed /Ill1998 m2001 1 Sources: Nicaragua LSMS 1998 and 2001. The use of child labor as a coping strategy was more prevalent among coffee farm households. Inparticular, even though childrenworking inlabor and farmhouseholdsbothworked more andwent to school less, the impact in terms o f increases in hours worked was stronger among coffee farm households (Figure 16 and 17). This i s also confirmed by looking at the results in equation 4 that imply that coffee fann-households were up to 21 percent more likely than non-coffee households experience child-labor increases (Tables 20 and 21). These patterns raise serious issues about the need o f policy interventions that can protect children's human capital against adverse shocks. While partial evidence seems to suggest that remittances are important for consumption smoothing, migration per-se does not seem to be a widespread strategy adopted among coffee households. While the empirical results o f equations 3 suggest that both coffee and non-coffee households receiving remittances in 1998 were more likely experience non-food consumption growth (Tables 13, 16 and 19), the results from the coping equation 4 imply that migration was not a coping strategy implementedby coffee households (Tables 20-23).31 31 Nonetheless, migration as a coping strategy was suggested during various informal interviews in rural Nicaragua and it consistent with similar studies such as Beneke and Gonzalez-Vega (2000) who find that the existence of international migrants within a household was correlated with higher income growth during a downturn inagricultural production inEl Salvador. 34 * ? 0 4.4 Precautionary savings In addition to adjustments to income portfolios, precautionary savings can help households cope with shocks allowing them to liquidate available assets. Still, coffee labor households were the most asset-poor among all households inrural Nicaragua. As such, their ability to use such assets to cope with shocks was limited. By contrast, coffee farmers during 1998 were among the wealthier households in terms o f asset holdings. Exploring the changes o f various assets like land or livestock indicates that some o f these assets were used as coping mechanisms, still in a limited way (Figures 19-22). Figure 19: The coffee crisis and assets: land - fP 15 VI .c vc U g 10 u -m .- -- P3 5 J 0 Non coffee Coffee Coffee Coffee both Coffee both both years Exit Entry years years Itdl998 H2001 I Sources: Nicaragua LSMS 1998 and 2001. 37 Figure 20: The coffee crisis and assets: cattle 12 10 a 6 4 2 0 Non coffee Coffee Exit Coffee Coffee both Coffee both both years Entry years labor years self employed (!fl1998 B2001 1 Sources: Nicaragua LSMS 1998 and 2001. Figure 21: The coffee crisis and assets: pigs 2 , I Non coffee Coffee Exit Coffee Entry Coffee both Coffee both bothyears years labor years self employed ID1998 E2001 1 Sources: Nicaragua LSMS 1998 and 2001. 38 Figure22: The coffee crisis andassets:value of equipment 1 20000, I1 15000 v1 n m eO 10000 8 5000 Non coffee Coffee Exit Coffee Entry Coffee labor Self-employed both years both years both years Sources: Nicaragua LSMS 1998 and 2001. Furthermore,equation4 suggests that poor farmers were less likelyto use assetsin responseto the coffee shock. By differentiating between coffee poor and non-poor households based on their 1998 classifications, the results suggest that poor coffee farmers were 13 percent less likely to sell land and 9 percent less likely to sell (or consume) cattle compared to non-poor coffee farm households (Table 21). Interestingly, poor coffee households were more likely to experience decreases in the number o f poultry owned, suggesting partial coping via own animal consumption (Tables 20,21 and Figure 21). These trends overall indicate the importance o f assets and highlight the limited capacity amongpoorer households to use physical assets as a major coping strategy.32 4.5 Informal insurance The use of informalinsurancemechanisms canbe another instrument bywhich householdmay use to address shocks. For example, informal social networks established by households through memberships in civic, religious, or neighborhood organizations can provide them an alternative source o f resources in the event o f an adverse shock. Inaddition, strong ties with migrant household members o frelatives may result help in the form o fremittances or informal gifts duringcrises. The empirical analysis shows that at least partially,the role of family networks is important.As discussed earlier, remittances (used as a proxy for the existence o f a family network) were positively correlated with non-food consumption growth for both coffee labor and labor households (Table 16). The impact seems to be stronger for coffee labor households implying that informal coping mechanisms may be more important for the poorer coffee households. 32This finding is similar to results inConning, Olinto andTrigueros (2000) who findthat households owning land or other productive assets were better able to protect their income during economic downturns. 39 4.6 Exiting coffee As indicated earlier, a significant number of households in the survey exited the coffee sector during this period. This "exit option" was higher among coffee laborers partially explained by the short run inability o f coffee farmers to exit the coffee sector due to their land commitment to the coffee production (Table 24). The observation that households that exited coffee did overall better in terms o f (socio)-economic outcomes suggests that that it would be useful to explore the attributes o f those exiting in order to understand the characteristics associated with higher mobility to get out o f coffee. While the data does not permit the distinction between those households that exited coffee due to lack o fjobs or farm business failure with those that have used exit as a risk management strategy, a model exploring a number o f initial (1998) characteristics and how they correlate with the exit decision o f the following i s estimated as follows: Prob(ExitC~ffee,,~~~~ I Coffee,,,,,, =1)= 4 4+4 Coffee,P +A2 (T/t:'Coffee,) +xi (5) where W i s a vector of initial (1998) household and regional attributes and i s a dummy identifying coffee farmers capturing a differentiated impact o f an attributing between coffee labor and farm households. As earlier, ` O ,4 and4 are parameters to be estimated while * 7T. . i s an i.i.d. error term. The estimation also uses municipality level fixed effects. Table 25 presents the results. Table 24: Transition matrix between coffee and non-coffee work (in %) 2001 COFFEE- COFFEE-FARMER NON- TOTAL LABOR COFFEE COFFEE- 35 9 56 100 LABOR 1998 COFFEE- 10 54 37 100 FARMER NON-COFFEE 5 4 91 100 Assets, wealth status and income diversification in non-agricultural jobs are important correlates with a household's ability to exit coffee. Less poor households were more likely to exit coffee suggesting that poorer households are less mobile. In addition, conditional on whether a household i s a coffee laborer or farmer, higher consumption increases the probability for coffee laborers to exit coffee comparedto farm coffee households (see also Figure 25). Similarly, while farm households were less likely to exit (since by definition their land investment inthe production process i s fixed), after controlling for land size, larger farmers were more likely to exit the coffee sector, indicating that if land can be interpreted as wealth, assets are important in allowing households engage in new activities. Finally, coffee households that were more income diversified in non- agricultural activities were more likely to exit coffee. This i s consistent with the earlier findings that show that the ability to enter the non-agricultural sector has been key in mitigating the negative shocks o f the shock. 40 Table 25: Mobility out of coffee: who can exit? Model 1 Model 2 Model 3 Model 4 interacted with coffee farmer dummy N o No Yes Yes With fixed effects No Yes No Yes Baseline period household characteristics (1998) Coffee farmer -0.22* ** -0.20* -0.71* -0.68 Number of adults aged 19-64 0.04 0.03 0.09* 0.03 Interaction -0.05 0.06 Age o f head o fhousehold -0.003 -0.004 -0.01 -0.003 Interaction 0.003 -0.0002 Average years o f education inhouseholds -0.02 -0.02 -0.03 -0.06 Interaction 0.01 0.02 Cultivated land owned (inhectares) 0.0001 -0.002 -0.04** -0.08" Interaction 0.04** 0.09* Received credit (yes=l) 0.10 0.12 0.43** 0.54** Interaction -0.39* -0.49** Income diversification index (O=not diversified) 0.27* 0.40" -0.10 0.11 Interaction 0.78** 0.50 Annual per capita consumption (in cordobas x1000) -0.01 0.01 0.01** 0.07* Interaction -0.12*** -0.1* Affected by Hurricane Mitch (yes=l) -0.14* 0.14 -0.33"" 0.36 Interaction 0.35* 0.40 Coffee farm intensity ("A o f total cultivable land) -1.43*** -2.62*** Interaction 1.76** 1.33 Distance to Managua (in 10 minute intervals) -0.01** -0.01 -0.01** -0.04" Interaction 0.01 0.01 Loe likelihood: " -122 -85 -107 -75 Adjusted percentage o fcorrect prediction: 0.39 0.33 0.50 0.44 Observations: 216 151 216 151 Dependent variable: Coffee activity status in2001 conditional on being incoffee in 1998. Marginal effects reported * significant at 10%; **significant at 5%; ***significant at 1% 41 Figure 25: Mobility out of coffee -AverageCoffee-labor __ -----Coffee-self employed -AverageCoffee-labor __ __-__ Coffee-self employed 0 1 4I I I 40 I 0 I 0I 20 $0 80 I Regional coffee intensity Distance to Managua 1bo __ -AverageCoffee-labor ----_Coffee-self employed __ -AverageCoffee-labor ____- Coffee-self employed 1 0 1 .i 4 .6 .8 1 0 10000 20600 3000C Household income diversification in 1998 Consumption per capita in 1998 42 Access to credit i s associated with a higher probability to exit coffee. The role o f credit can be crucial in mitigating the impact for shocks by both helping to cope and diversify in other activities. Credit has a stronger impact on the probability to exit coffee among labor coffee households as opposed to farmers, perhaps highlightingthe lack o f assets among coffee labor households. Finally, a number of attributes describing the local economic context are correlatedwith exiting coffee. For example, distance to Managua or residing within the coffee region are both negatively correlated with the probability to exit coffee. Both o f these attributes capture the existence o f non- coffee activities and opportunities (in addition to controlling for the shock for the latter). Interestingly, residingina region affectedby hurricane Mitch also decreases the probability o f exiting the coffee sector, presenting an example o fthe adverse effect o f multiple shocks on household^.^^ While separating the decision to exit from a forced exit is challenging, these findings seem to indicate the critical importance of assets and opportunities on upward mobility and coping capacity. They reinforce the fact that in the presence o f shocks, those households that can protect themselves using instruments that either detach them from exposure to risk or minimize its impact if the risk i s realized, are better able to cope. 4.1 The role of ex-ante risk management To summarize the results in this section, coffee households have used a mixture of coping mechanisms in response to the coffee crisis. While harmful coping mechanisms such as increases in child labor and - to a lesser extent - selling or consuming physical and animal assets were utilized among coffee households, a number o f ex-ante management instruments such as exiting coffee, receivingremittances or income diversificationwere also used (Table 26). Table 26: Use of riskmanagement mechanisms and ruralheterogeneity, by coffee definitions Type o f strategy N o n Coffee Coffee coffee labor farmer Income diversification ex-ante Yes Labor market Child labor ex-post Yes Yes adjustments Ex-post migration ex-post Exit coffee ex-ante/ ex-post Yes Yes Precautionary Sale o f physical assets ex-post Yes savings Consumption o f owned animals ex-post Yes Yes Informal Remittances ex-ante Yes Yes Yes insurance While a formal test cannot explicitly compare the two, the findings suggest that households that used ex-ante as opposed to ex-post mechanisms were better insulated from the coffee shock. For example, since much o f the explanatory variables in the consumption growth models are all based on the initial pre-crisis household income strategies, their positive role on consumption growth can be interpreted as the realization o f ex-ante risk management actions taken by these households. For example, by diversifying the income sources or having migrant members before the coffee shock, coffee households were better able to mitigate the adverse impact o f the crisis. Similarly, higher education (using the maximum level o f education in the household in 1998) was associated with a 33Hurricane Mitch hit the regioninOctober 1998, right after collection of the first survey. 43 four percent increase in consumption growth, which -while not testable - i s consistent with the hypothesis that human capital may have allowed households to mitigate the negative impact from the crisis by either finding higher return occupations or increasing farm efficiency. Comparing the effectiveness o f ex-ante and ex-post strategies i s beyond the scope o f this study. Still, the dominant role of ex-ante strategies among coffee households for consumption smoothing and the observation that households that predominantly usedex-post coping mechanisms did worse suggests that, at least qualitatively, ex-ante strategies have been more effective. 5. SHOCKS,VULNERABILITY AND MOBILITY The previous sections outlined the extent by which the coffee crisis has affected rural households and explored the various mechanisms affected households utilizedto cope with the shock, While households do not seem to be able to fully insure against unanticipated income fluctuations, a number o f coping strategies were used among rural coffee households that mitigated the impact o f the coffee shock. For households affected by the coffee crisis, a heterogeneous set o f mechanisms such as ex-ante income diversification or ex-post increases in child labor have allowed households to deal partially with the shock. Nonetheless, prioritizing among the identified strategies and mechanisms explored above is a complextask. For example, the results suggestthat the coffee shock had a bigger impact on farmers rather than labor households. Still, coffee farmers had the lowest poverty rates, highest level o f assets while labor households are chronically poor. As such, further exploring the linkages between shocks and poverty dynamics may allow buildinga more comprehensive policy agenda. 5.1 Poverty dynamics To this end, this section provides an analysis on the impact of shocks on poverty dynamics. Specifically, two questions are addressed: (i) the coffee shock increased household vulnerability has to decreases in welfare; and (ii)did the ability o f households to escape poverty (mobility) changed due to the shock? Inthe case of ruralNicaragua, poverty is dynamic.For example, between 1998and2001, almost a third o f non-coffee households moved in and out o f poverty (Table 27, Figures 23 and 24). In addition, non-coffee households were less likely to exit poverty (upward mobility) than falling into poverty, consistent with the overall poverty rate decreases observed duringthis period. 44 Table 27: Ruralpovertydynamics,by coffee definitions(YOof households) Povertv in2001 Poor Non Poor Total Non-Coffee Poor 46 19 65 NonPoor 9 26 35 Total 55 45 100 Exit Poor 67 11 78 NonPoor 9 13 22 03 m Total 76 24 100 2 Enter .sx Poor 52 24 76 F NonPoor 11 14 24 0 Total 63 37 100 a Bothyears-Labor Poor 90 5 95 NonPoor 2 3 5 Total 92 8 100 Bothyears- farmer Poor 51 10 61 NonPoor 17 23 39 Total 67 33 100 Sources: Nicaragua LSMS 1998 and 2001. Figure23: Poverty mobility by householdcoffee definition Non-Coffeeboth Coffee Exit - Coffee Enter - Coffee-labor both Coffee self- years years employed both years I ~ ~~ E4 Poor in both years EEntry(non-poorin 1998,yes in OExit (poor in 1998, not in 2001) 0Non-poorin bothyears Sources: Nicaragua LSMS 1998 and 2001. 45 Figure 24: Poverty mobility by regional coffee definition 100% 80% 60% 40% 20% 0% Low Medium High 0Poorinbothyears HEntry(non-poorin 1998, yes in 2001) OExit (poor in 1998,not in 2001) Non-poor in both years I Sources: Nicaragua LSMS 1998 and 2001 andNational Agricultural Census 2001. In addition, a number of interesting patterns related to the coffee shock emerge with respect to poverty changes. First, while almost a third o f coffee farm households experienced similar movements in and out of poverty compared with the overall trends above, they were more likely to enter poverty (Table 27). In addition, coffee labor households were virtually trapped in chronic poverty. Almost 90 percent o f coffee labor households remained in poverty and experience little upward mobility. Coffee households were also more likely to experience a consumption decrease. Only ten percent o f non-coffee experienced a fall in their "ranlung" interms o f consumption quintiles (Table 28). This compares with a quarter o f coffee labor households and half o f the coffee farmers. In addition, comparing households based on whether consumption in general decreased over the period, while almost 40 percent o f non-coffee household experienced consumption decreases, more than two thirds o f coffee farm households and 56 percent among coffee labor households suffered a drop in consumption. 46 Table 28: Consumptiondecreases,by coffeedefinitions(YOof households) % o fhouseholds experiencing a consumption decrease: Level Quintile Coffee typology N o n coffee both years 38 11 Exit coffee 46 27 Enter coffee 48 30 Coffee both years 61 39 Coffee farmers both years 56 47 Coffee labor both years 65 23 Regional coffee definition L o w 36 15 Medium 43 8 High 52 25 Sources: Nicaragua LSMS 1998 and 2001. These resultsindicatethat the coffee shock may have affected coffee households' ability to enter or exit poverty. Further exploring how the coffee shock may have affected these dynamics i s addressed below. 5.2 Vulnerability topoverty Vulnerability is a dynamic concept capturingthe probabilitythat a householdwill experiencea negativeloss initswelfare.34The main idea of vulnerability is that it measures a household's ability to insure or protect against exposure to risk. In fact, while exposure per se i s not sufficient to infer vulnerability, observing a differential behavior among exposed households or between exposed and non-exposed households i s indicative o f the degree that a household will suffer welfare losses in the event o f the risk being realized, therefore measuring its vulnerability to risk exposure. For the purposesofthe study, three definitionsfor vulnerabilityare used:(i) likelihood that a the household's consumption fell below the poverty line during the two periods covered by the data; (ii) the probability that a household's experienced a decrease in its consumption level; and (iii) the probability that a household's initial ranking based on consumption quintiles decreased. To address the first definition, the following model o f the probability that a household - which was not poor in 1998 - entered poverty in2001i s estimated: where Coffeek, Xi and Zi are as defined earlier, zi i s an i.i.d. error term. In addition, while 5k tests whether a household's exposure to the coffee crisis increase the probability (and therefore vulnerability) to fall into poverty, p and ty reveal the extent where a number o f householdattributes are correlated with vulnerability to poverty.35 ~ 34Holzmann2001. 35To control for municipal-level characteristics related to the coffee crisis, the regression also includes the municipality-level intensity incoffee production. 47 Similarly, using the second definition, the probability that household i experienced a fall in consumption level is given by: while for the last definition, the probability that ahousehold's consumption ranking fell canbe estimated using: N-1 Prob(Quintilei,,,ol < Quintilei,1998) = z5k Coffee, +p.Xi +ty +zi Zi (8) k=l The results from these models arepresentedinTable 29. Table 29: Poverty dynamics: examining vulnerability and mobility Probability to: Experienced a fall in consumrAion Fall into poverty Level Quintile Exitpoverty Baseline period household characteristics (1998) Coffee labor 0.07 0.07 0.05 0.09 Coffee farm -0.09 0.01 -0.00 0.14 Family size 0.01 0.01 0.01 -0.04* ** Maximumyears of education inhousehold -0.03*** -0.03*** -0.03 *** 0.03*** Number of kids workers 0.03 0.00 -0.01 -0.00 Number of adult workers 0.01 -0.02 0.00 0.03* Number of income sources -0.02 -0.01 -0.02 0.05** Land owned (hectares) -0.00 -0.00 -0.00 o.oo*** Receivedremittances (yes=l) -0.03 -0.03 -0.01 0.06" Distance to Managua (10 minuteintervals) o.oo** 0.00 0.00 -o.oo** Coffee farm intensity inmunicipality 0.33 0.67** * 0.43*** -0.78 *** Affected by Mitch (yes=l) 0.03 0.07** 0.06** -0.06** Sample Non-poor in 1998 All rural All rural Poorin 1998 Observations 505 1355 1355 850 Log likelihood: -306 -936 -880 -481 Adjustedpercentage of correct prediction: 0.72 0.59 0.65 0.77 Dependent Variable for model 1: Poverty status in2001 conditional on being nonpoor in 1998. Dependent Variable for model 2: Dummy on whether a household experienced a decrease inconsumption level between 1998 and 2001. Dependent Variable for model 3: Dummyon whether a household experienced a decrease inconsumption quintile ranking between 1998 and 2001. DependentVariable for model4: Poverty status in2001 conditional onbeingpoor in 1998. Additional controls: initial period consumption quintile ranking for 2nd and 3rd models. *significant at 10%; ** significant at 5%;*** significant at 1%. 48 Households residing in the coffee region were more vulnerable to welfare losses, suggesting that the coffee shock increased vulnerability. While participation in the coffee sector (using the initial coffee classification) did not have statistically significant effect in household's vulnerability to welfare loss, the regional coffee definition suggest that households in the coffee region were more likely to experience a fall in consumption (Table 29). This finding implies that exposure to the coffee shock risk has increased vulnerability to welfare losses among exposed households. Exploring further the concept of vulnerability to poverty and consumption loss, a number of interesting points arise. For example, higher levels o f education significantly reduce vulnerability to poverty. This reinforces the importance o f human capital accumulation as an ex-ante instrument to minimizing vulnerability. Inaddition, residing ina municipality affectedby hurricane Mitch increases the probability that a household will experience reductions in welfare. Again, this confirms the hypothesis that shocks negatively influence poverty dynamics, inthis case vulnerability. 5.3 Upwardmobility An alternative exercise in understanding poverty dynamics is to explore the factors that are correlated with households' mobility to exit poverty. To address this, a model o f the probability that a household exited poverty in2001 conditional on beingpoor in 1998 i s estimated: P ~ o ~ (>C~ov~ine,,,,1~POO~,,,, =1)= x 5 k Coffeek+p xisi w z +zi `I I ~ , ~ ~ ~ * (8) k=l where the regressors are the same as inequation 6 and 7. The results are discussed below. Households residing within coffee regions were less likely to exit poverty. Mirroring the results on vulnerability, while the household-level classifications o f affiliation in coffee activities were not significant, this finding illustrates the aggregate impact o f the exposure to the coffee crisis inupward mobility (Table 29). A number of other factors are correlated with the ability to exit poverty. First, income diversification increases the probability to exit poverty (Table 29). This provides empirical support to the current policy efforts to promote diversification in rural areas, as it indicates that it i s not only a successful coping strategy (among coffee farmers) but also important in enhancing upward income mobility.36It i s also important to point out, however, that the diversification measure used here refers to income from different sources (agriculture, non-agriculture, wage and self-employment), and not to diversification in agricultural production. Indeed, an alternative specification using crop diversification found no significant correlation with poverty dynamics. In addition, households receiving remittances were more likely to exit poverty. This result indicates that migration as a strategy to access higher-return opportunities, i s important for economic mobility and reinforces the role o f social capital and informal networks in poverty alleviation. Furthermore, both human capital (education) and physical (land) assets were also positively correlated with exiting poverty. Finally, distance from Managua i s inversely related to the ability to exit poverty. To the extent that this captures the local economic environment, it shows that more isolated areas offer fewer income options for households. 36Ministerio Agropecuario y Forestal (2003), Varangis et. a1(2003). 49 To summarize the poverty dynamics analysis vis-his the coffee crisis, predicted probabilities to fall or escape poverty are calculated. First,households affiliated with the coffee sector were the most vulnerable to decreases in welfare and least mobile to exit poverty compared to non-coffee households, suggesting that the coffee crisis has indeed affected their mobility and vulnerability (Table 30). These results are robust as they hold independent o f the coffee definition or typology used.37 Table 30: Poverty dynamics: predictedprobabilities,by coffee Household(YOof households) Predicted probability to: Experienced a fall inconsumption: Fall into poverty Level Quintile Exitpoverty Household definition N o n coffee both years 27 45 33 24 Exit coffee 32 52 39 27 Enter coffee 36 47 34 16 Coffee labor both years 44 55 40 17 Coffee farmer bothyears 30 61 47 17 Initialyear classification Non coffee in 1998 27 45 34 23 Coffee-labor in 1998 39 52 38 19 Coffee-farm in 1998 29 59 45 23 Regional definition L o w coffee intensity 27 43 32 27 Medium coffee intensity 26 46 35 23 Highcoffee intensity 33 56 41 14 Overall 28 47 35 22 Finally, while coffee laborers-the poorest rural group in the survey were the most adversely - affectedwith respect to vulnerability and mobility with respect to poverty, coffee farmers were mostly affected in terms of the probability to experience consumption declines. These results, suggest that while for coffee farmers the shock may have been more transitory innature, it may have accentuated poverty traps among the chronically poor coffee laborers. This raises the need for distinct policy interventions for each o f the two groups. 6. PUBLIC RESPONSE TO THE COFFEE CRISIS While Government and private support for the coffee sector was significantly delayed inNicaragua, a number o f programs addressing the coffee crisis have since been established. A short summary i s presentedbelow. 37 The probability to exit poverty among non-coffee households is not statistically significant with that o f coffee-farm households usingthe initial coffee classification. 50 6.1 Debt restructuring By 2002, coffee-farm debts totaled approximatelyUS$105 millionin Nicaragua.38As the ability o f coffee farmers to repay these loans diminished, it presented a potential crisis in the country's already stressed financial system. As such, the Government intervened by promoting, coordinating and providing funds for different debt-restructuring programs. These programs varied according to the type of debt held by a coffee producer, with the following main restructuring categories being created: (i)debts to solvent commercial banks (US$55 million - 684 cases); (ii) debts to bankrupt commercial banks (US$32 million - 665 cases); (iii) tomicro-financeorganizations (US$6million-7,520 cases); and(iv)debtstoexporting debts firms (US$12 million - 2,300 cases). The first two categories targeted mainly medium and large coffee farmers (with farms sizes o f at least 20 manzanas), the third focused on small farmers (5 manzanas or less) while the final category did not distinguishbased on farm size. It i s important to note that the majority o f the government restructuring schemes (more than 80 percent) has focused on large coffee farmers. As of May 2003, 100% of the debts in categories (i)and (iii)had been resolved, where the Government played an active role. While the Government did not get involved in re-structuring producers' debts to exporting firms (category iv), these appear to be getting resolved in an efficient manner by the stakeholders (usually an exporting firm and a producer). 6.2 Socialprotection interventions The Government of Nicaragua implemented a "Food-for-work on Coffee Farms" program through the Ministry of Agriculture (MAGFOR). The program took place in 2002 in 21 coffee municipalities, costing US$574,336 and providing family food rations to 8,212 households: 6,3 17 o f them were small coffee farm owners (6 manzanas or less), and 1,895 were coffee farm workers. Participating households receivedthe food complement inexchange for workmg on various activities on coffee farms.39 6.3 Indirect benefitsfrom existing (non-coffee specific) programs A number of existing public programs may have indirectly mitigated the impact of the coffee crisis. First, the Government's "Libra por Libra" program which started in 2002 has led to higher productivity o f small farmers' production o f basic grains for own-consumption via the disbursement o f genetically improved and certified seeds for basic grain production, and technical assistance. An estimated 72,000 small farmers, many o f which reside in coffee regions have participated in the program. During 2003, and in part due to the coffee crisis, MAGFOR doubled the amount o f seed distributed insome coffee regions.40 In addition, the "Red de ProteccionSocial", a conditionalcash program in Central Nicaragua that supplements poor rural households' incomes seems to havemitigated the adverse impact of 38NicaraguanCoordination and Strategy Secretariat o f the Presidency (SECEP). 39Prior to this program, the Government financed a small scale workfare program benefiting 300 coffee workers (representing about 1,000 family members) in2001. 40MAGFOR. 51 the coffee shock. In particular, a recent impact evaluation o f the program finds that program beneficiary households involved inthe coffee sector have faired better in a number o f socio-economic outcomes compared to non-participatingcoffee ho~seholds.~~ 6.4 Supportfrom other agencies USAID financed a US$2.5 million coffee relief, food-for-workinitiative in 2002. The program's objectives was to provide reliefto unemployed coffee laborers, provide incentive to coffee farmers to continue employing their full-time labor force on a full-time basis, ensure that essential crop maintenance i s performed and provide limited support to rehabilitate public infrastructure. An estimated 13,394 coffee laborers inten coffee municipalities benefited from the USAIDprogram. Finally, the German government's assistance agency (KDR) financed a large infrastructure project to increase the supply o fpotable drinkingwater inthe departments o f Jinotega and Matagalpa. This project was initiated in 2001, and it generated approximately 10,000 to 15,000 temporary jobs, potentially coffee laborers. While the programs described above may have temporarily alleviated some of the adverse impacts o f the coffee crisis, it i s unclear as to whether they have fully addressed its structural nature. In fact, none o f the coffee-specific programs discussed above seem to have had a long-term objective but instead aimed at addressing the short runcoping capacity o f affected households. Inaddition, the majority o f the public resources were targeted in a regressive way, mainly directed to medium and large coffee farmers. 7. MOVINGFORWARD: LESSONSFORCONSTRUCTINGA POLICY AGENDA Using household level panel data from Nicaragua, this paper explores the impact o f the recent coffee crisis on rural households engaged in coffee production and coffee labor work. Taking advantage o f the panel structure o f the data, a number o f findings emerge: (i) while overall growth between 1998 and 2001 was widespread in rural Nicaragua, coffee households saw large declines in various socioeconomic outcomes; (ii) small coffee-farm households were affected the most, and not poor labor households as previously expected; (iii) among the various risk management strategies coffee households usedto address the shock, pre-shock, ex-ante strategies (like income diversification) were more effective in allowing coffee households insulate against the shock. By contrast, the coffee households that used ex-post coping instruments did not manage to mitigate the adverse impact as well, with additional potential long run implications via extensive uses o f harmful coping strategies (like increases in child labor); and (iv) the coffee shock affected upward mobility and downward poverty vulnerability, Based on the finding above, a number o f lessons emerge in terms o f pushing forward the policy agenda related to the coffee crisis and shocks in general. They are discussed below. 7.1 "Understand the shock and those affected" Initial attention onthe coffee crisis focused onthe impactof the shock on labor employment. The analysis shows that it was small coffee farmers, rather than poor coffee laborers, that appear to have experienced the most serious effects from the crisis. This was partly due to the fact that while labor workers 4' Maluccio (2003). 52 were mobile in moving from coffee employment to other low paying labor jobs, coffee farm households were stuck inlong-term perennialinvestments withlittle flexibility to complement their incomes. These insights have important implications about the choice of a short-run safety net one could potentially consider. While shocks that result in open unemployment are typically addressed through workfare programs by providing support to unemployedworkers untilrenewed labor demand draws them back into the labor market, the fact the laborers were able to substitute for potential labor losses via alternative low payingjob opportunities seems to implythat such interventions were notnecessarily critical. By contrast, whle the immediate debt relief efforts discussed above may have allowed large farmers to cope with falling coffee prices and cost increases, the low participation in such programs by small scale farmers and the lack o f alternative coping mechanisms for them seems to explain to a large extent the large welfare impacts o f the crisis on these small, immobile farm households. As such, understandingwhich populations shocks affect and how is key for designing appropriate interventions. 7.2 "While householdsuse a diverseset of informal risk management instruments, they are onlypartially effective" Coffee householdsuseda multitude of riskmanagementmechanismsto address the crisis. Some examples include informal support systems such as receiving remittances from family, income diversification to sales o f assets (land or animals) or sending children to work. Nonetheless, the absence o f formal insurance instruments available to these households implies that such self- insurance and risk management instruments are unlikely to be fully effective inprotecting them from risk exposure. Indeed, the results indicate that coffee households, especially the poorer coffee-labor ones, were extremely vulnerable to insuring food consumption, with more than 43 percent o f the income shock among coffee-labor households being passed through food consumption decreases (and 13 percent among coffee-farm households). Such findings reinforce the need for improving formal insurance mechanisms and enhancing informal risk management instruments. They also suggest that interventions should pay special attention on the poorer and more vulnerable populations. 7.3 "Enhancing households' ex-ante set of risk management instrument base is crucial" The findings suggest that households that used ex-ante as opposedto ex-post mechanismswere better insulated from the coffee shock. For example, coffee household that diversified their incomes, invested inhuman capital or exited the coffee sector altogether before the crisis hit (and thus fully dissociated themselves from the coffee risk exposure) were better positioned to deal with the coffee crisis. By contrast, coffee households that did not have the ability or did not use such risk management instruments were not only affected worse, but they also used some coping mechanisms with potential long-term adverse implications (such as taking children out o f school). Policies that enhance the ability and adoption o f ex-ante risk management strategies should therefore be at the center o f the policy agenda. 7.4 "Shocks influence long run welfare dynamics" Coffee households were the most vulnerable to fall into poverty and the least mobile to exit poverty by taking advantage of the overall growth in rural Nicaragua over the period of the study. Still, while coffee farmers were affectedthe most interms o f levels, even after the crisis hit they were still among the wealthiest rural groups inNicaragua. By sharp contrast, coffee laborers -by far the poorest rural group in the survey - were the most adversely affected with respect to their increased probability to fall and lower probability to exit poverty. These insights seem to indicate the 53 distinction between the impact o f shocks with respect to chronic and transient poverty. To some extent, while for coffee farmers the shock may have been more transitory in nature, it may have accentuated poverty traps among the chronically poor coffee laborers. This raises the need for distinct policy interventions for each o f the two groups better addressing structural versus transient poverty. Some potential areas for further exploration on this comes out o f the analysis by observing the various factors that are correlated with the ability to fall or exit poverty. Such factors include the role o f human capital and its importance as an ex-ante instrumentto minimizingvulnerability and enhance upward mobility, the ability to have a diverse income portfolio by including non-agriculture income sources or the role o f the local context and infrastructure in providing alternative income opportunities to risk exposed households. 7.5 "Long-run investmentsfor short-run protection?" While not a direct outcome from the study, some of the insights seem to suggest that longer-term interventions such as cash transfers conditional on householdinvestments inhouseholdmembers' (such as children) health and education can partially allow households affectedby shocks to better cope with shocks by insulating them from their adverse impacts. Indeed, "Red de Protection Social" beneficiary households involved in the coffee sector seem to have faired better in a number o f socio-economic outcomes compared to non-participating coffee households (such as the significant higher children's education attainment outcomes among beneficiary ho~seholds).~~ Suchprogramsarenotdesignedto dealwith shocks and are not "insurance" schemes per se. Still, the observed positive impact inthe coffee crisis example suggeststhat by incorporating risk exposure inthe design o f such programs' eligbility rules, or by allowing additional flexibility in terms o f scaling up or down such interventions to address large shocks on-demand i s worth fkther examination to understand whether these programs can serve as alternative riskmanagement instruments. 7.6 "Agricultural interventions: structural shocks require structural changes" While this is beyond the scope of the paper, a number of insightswith respect to the potential role of agricultural or coffee-industry specific interventions can be outlined. First, improving crop insurance schemes seems to be an important direction for further analysis. Introduction o f such a market based ex-ante instrument can greatly improve households' ability to make decisions under uncertainty. This issue still remains highly understudied. Second, promoting product differentiation in coffee i s another area for policy discussion. Infact, the fact that only ten percent o f the current coffee production in Nicaragua i s specialized (e.g. organic, fair trade) suggests that at least exploring its feasibility and pre-requisites o f scaling up such practices i s crucial.43 In addition, enhancing marketing practices and channels by promoting local and external demand also seem important areas for policy design and intervention. Finally, as the analysis shows, facilitating coffee households to exit the coffee sector altogether may be a desired policy. To the extent that such as policy can be targeted at small farmers that engage in lower quality coffees or farm in marginal lands, complemented by promoting alternative livelihoods for such households seems to be a direction by which policy can strengthen householdadaptability and mobility. Such structural changes can only be part o f large comprehensive vision for rural development, poverty reduction and risk management schemes and as such, adapting these to the specifics parameters o f regional and household realities will be essential. 42Maluccio (2003). 43Varangis (2003). 54 REFERENCES Beneke de Sanfeliu, Margarita and Claudio Gonzalez-Vega. 2000. "Dynamics of Rural Household Incomes inEl Salvador: 1995- 1997 PanelResults," Mimeo. Chawla, Mukesh.2001. "Malnutrition Among Pre-School Children." Backgroundpaper for Nicaragua Poverty Assessment, Vol.II,ReportNo. 20488-NI, The World Bank, Washington, D.C. Conning, Jonathan, Pedro Olinto, and Alvaro Trigueros. 2001. "Managing Economic Insecurity inRural ElSalvador: Therole of asset ownership andlabor marketadjustments," Williams University-EconomicsDepartment Working Paper Series, Williamstown, Ma. Davis, Benjamin and Marco Stampini. 2002. "Pathways towards prosperity inruralNicaragua; or why households drop inand out ofpoverty, and some policy suggestions onhow to keep them out," FA0 and Scuola Sant'Anna, Pisa. Glewwe, Paul and GiletteHall. 1998. "Are some groups more vulnerable to macroeconomic shocks than others?Hypothesistests based on panel data from Peru." Journal of Development Economics. Vol. 56, pp.181-206. Grimard, F. 1997. "Household Consumption SmoothingThrough Ethnic Ties: Evidence from Cote d'Ivoire," Journal of Development Economics, 53, August 1997: pp.391-422. Holzmann, R. and SteenJorgensen. 2000. "Social RiskManagement: aNew Conceptual Framework for Social Protection and Beyond." Social ProtectionDiscussionSeries No. 0006. The World Bank, Washington, D.C. Inter American DevelopmentBank. 2001. Transicidn Competitivapara el Cafk Centroamericano: Crisis International del Cafk y su Impact0 en Nicaragua, Mimeo, Washington, D.C. Jalan, Jyotsna and Martin Ravallion. 1999. "Are the poor less well insured? Evidence on vulnerability to income riskinrural China," Journal ofDevelopment Economics,Vo1.58, pp. 61- 81. Klugman, Jeni, Diana Kruger and Kate Withers. 2002. "Consumption Riskand Smoothing DuringDisasters: The Case ofHurricaneMitchinNicaragua." The World Bank, Mimeo. Lard6 de Palomo, Anabella, Claudio Gonzalez-Vega, and Aida Arguello de Morera. 2000. "Household Integration to the Market as a Determinant of Rural Incomes inEl Salvador," Working Paper No.OOP15, Ohio StateUniversity, Department of Agricultural, Environmental, and Development Economics, BASIS Research. Maluccio, J. 2003. "Coping with the Coffee Crisis: inCentral America: the Role o f the Nicaraguan Social Safety Net Program. IFPRImimeo. Marini, Alessandra and Michele Gragnolati. 2002. "Malnutrition and Poverty inGuatemala." Policy ResearchWorking Paper No.2967, The World Bank. Ministerio Agropecuario y Forestal. 2003. "Estrategia para reconversih y la Diversificacih Competitiva de la Caficultura enNicaragua." Mimeo. Narayan, Deepa. 1999. "Bonds and Bridges: Social Capital and Poverty." The World Bank, 55 Poverty Group PREM. Ravallion, M.and S. Chaudhuri. 1997."Risk and Insurance inVillage India: Comment," Econornetrica, 65 (l), 1997: pp. 171-184. January Skoufias, Emmanuel and Agnes R. Quisumbing.2002. "Consumption Insurance and Vulnerability to Poverty: A Synthesis o f the Evidence from Bangladesh, Ethiopia, Mali, Mexico, and Russia." Mimeo. The World Bank. 2003. Nicaragua Poverty Assessment: Raising Welfare and Reducing Vulnerability.Report No. 26128-NI, The World Bank, Washington D.C. Townsend, Robert M. 1994. "Risk and Insurance inVillage India," Econornetrica, 62, May 1994, pp. 539-91. Varangis, Panos, Paul Siegel, Daniele Giovannucci, and Bryan Lewin. 2003. "Dealing with the Coffee Crisis inCentral America: Impacts and Strategies." The World Bank, Policy Research Working Paper No. 2993. 56 Appendix2: Attrition and panelconstruction An extensive analysis o f the attrition inthe Nicaraguapanel used inthis paper can be found inDavis and Stampini (2002). They conclude that while almost a third o f the original sample was not interviewed in 2001, attrition i s not a major problem inthe sample. Infact, the only exception intheir analysis i s among urban non-poor households, where they find some weak evidence o f non-random attrition. In addition, there does not seem to be a systematic difference between coffee households (both labor and farm) with non-coffee households (Table 31). As such, and since this paper focuses exclusively on rural households, attrition i s not consideredto be a problem. Table 31: Panelattrition N o n coffee Coffee Households All Households Labor Farmer All coffee Number % Number % Number % Number % Number % Dropped in2001 1109 28.8 61 30.5 46 28.1 107 29.4 1216 28.9 InPanel 2736 71.2 139 69.5 118 71.9 257 70.6 2993 71.1 Total 3845 100 200 100 164 100 364 100 4209 100 57 COPING WITH THE COFFEE CRISISINCENTRAL AMERICA: THE ROLEOF THE NICARAGUAN SOCIAL SAFETY NETPROGRAM John A. Maluccio InternationalFoodPolicy Research Institute Washington D.C.20006-1002 202-862-5693 j.maluccio@cgiar.org FirstDraft: October 2003 This Draft: February 2004 ABSTRACT The international and local Nicaraguan media have widely reported on the "coffee crisis" in Latin America and there i s substantial evidence that there has been a downturn and that this has been more severe in the coffee growing regions. Using household panel data from a randomized community- based intervention carried out in both coffee and non-coffee growing areas, Iexamine the role o f a conditional cash transfer program, the Red de Proteccidn Social (RPS) during this downturn. While not designed as a traditional safety net program in the sense o f reacting or adjusting to crises or shocks, RPS has performed like one, with larger estimated program effects for those who were more affected by the downturn. For example, it protected households against declines in per capita expenditures and, while not significantly depressing labor supply relative to before the program, muted additional labor effort for beneficiaries in coffee growing areas, relative to their counterparts without the program. The evidence i s more mixed, however, as to whether RPS enabled households to reallocate their resources in a fashion consistent with the historical downward trends in coffee prices. Beneficiaries who participated in the coffee industry as laborers were more likely to have exited the coffee industry, whereas those who participated as producers were less likely to have exited. The findings are consistent with the existence o f credit constraints inhibiting such transitions inthe absenceofthe program. Overall, then, RPS appears to beplaying animportant part inthe "risk" coping strategies o f households, a conclusion also supported by a separate analysis o f individual household-level idiosyncratic shocks. This researchbuilds on the evaluation o fthe Red de Proteccidn Social by the International FoodPolicy Research Institute. Ithank the Red de Proteccidn Social team, for their continued support. Nathlia Caldts, Dave Coady, Andrew Mason, Nancy McCarthy, Pedro Olinto, Laura Rawlings, Ferdinand0 Regalia, and Renos Vakis providedmany helpfulcomments, as have seminar participants at NEUDC and the World Bank. Igratefully acknowledge funding for this research from the Inter-American Development Bank, through the Norwegian Fundfor SocialInnovation, and from the World Bank. 58 1. INTRODUCTION Inspite of some recovery in 1994 and 1998, world coffee prices have been declining since the mid 1980s-in 2001, real prices were at their lowest levels in more than 50 years. The continued downward trend and the recent substantial decrease in prices have had adverse implications for incomes within many o f the coffee producing countries in Central America. These have been widely reported on in the international and local media as the "coffee crisis." In some cases, prices have reached levels below typical production costs. Though only limited micro level empirical evidence exists regarding the magnitude and nature of the effects o f the price trend, there i s a perception that one consequence i s that poverty i s rising, at least among certain groups and incertainrural areas. In this paper, Iexplore the effects of the price decline in some of the poorest rural regions of Nicaragua, using household-level panel data collected as part o f a randomized evaluation o f a conditional cash transfer program. Ialso examine the role played by the program, the Red de Proteccibn Social (RPS), in protecting well-being, as well as its effects on labor market supply and activities. To do this, Icontrast the behavior and outcomes o f households who were benefiting from the program to those who were not. About half o f these households live in coffee growing areas and many are involved to some extent in the coffee industry. These data are brought to bear on the following questions: 0 H o w have households in coffee growing areas without the program fared over the period 2000-2002? 0 Were households in coffee growing areas with the program better able to protect household expenditures (particularly on food) and educational and nutritional outcomes than their counterparts in coffee growing areas without the program? That is, how effective was RPS as a social safety net duringthe downturn? 0 Were labor supply and the agricultural versus non-agricultural mix o f activities within the household different among households in coffee growing areas with and without the program? That is, did RPS enable different labor responses to the downturn? The overarching research question i s whether, and how, the program enabled alternative responses to the downturn. While much o f the emphasis in the paper i s on the so-called coffee crisis, the results have broader implications to the extent they demonstrate how safety net programs like RPS condition behavior duringan economic downturn. The findings show that, while not designed as a traditional safety net program inthe sense o f reacting or adjusting to crises or shocks, RPS has performed like one, with larger estimated program effects for those who were more affected by the downturn. For example, it protected households against declines in per capita expenditures and, while not significantly depressing labor supply relative to before the program, muted additional labor effort for beneficiaries incoffee growing areas, relative to their counterparts without the program. The evidence i s more mixed, however, as to whether RPS enabled households to reallocate their resources in a fashion consistent with the historical downward trends incoffee prices. The evidence i s more mixed, however, as to whether RPS enabled households to reallocate their resources in a fashion consistent with the historical downward trends in coffee prices. Beneficiaries, who participated in the coffee industry, as laborers, were more likely to have exited the coffee industry, whereas those who participated as producers were less likely to have exited. The findings are consistent with the existence o f credit constraints inhibiting such transitions 59 in the absence of the program. Overall, RPS appears to be playing an important part in the "risk" coping strategies o f households, a conclusion also supported by a separate analysis o f individual household-level idiosyncratic shocks. 2. DESIGNAND IMPLEMENTATIONOF THERED DEPROTECCIbNSOCIAL44 In recent years, increasing emphasis has been placed on the importance of human capital in stimulating economic growth and social development. Consequently, investing in the human capital o f the poor i s widely seen as crucial to alleviating poverty, particularly inthe long term. At the same time, there i s growing recognition o f the need for social safety nets to protect poorer households from poverty and its consequences during the push for economic growth (World Bank 1997). At first glance apparently conflicting strategies for economic development, both are important, and, potentially complementary (Morley and Coady 2003). Effective social safety nets may directly contribute to economic growth via improved human capital, particularly inthe long term. Consistent with this view, several Latin American countries have introduced programs that integrate investing inhuman capital with access to a social safety net. One o f the first, and largest, programs o f this type was the Programa Nacional de Educacidn, Saludy Alimentacidn (PROGRESA) inMexico, begun in 1997. Another large program i s the Programa de Asignacidn Familiar ( P W ) inHonduras. In this paper, Iexamine a third such program, the Nicaraguan Red de Proteccidn Social (RF'S) or "Social Safety Net." Modeled after PROGRESA, RPS i s designed to address both current and future poverty via cash transfers targeted to households living in extreme poverty in rural Nicaragua. The transfers are conditional, and households are monitored to ensure that they are undertaking prescribed actions intended to improve their children's human capital development; when they fail to fulfill those obligations, they lose their eligibility for the program. By targeting the transfers to poor households, the program alleviates short-term poverty. By linking the transfers to investments in human capital, the program addresses long-run poverty. RPS's specific objectives include: 0 Supplementing household income for up to three years to increase expenditures on food, 0 Reducing school desertion during the first four years o fprimary school, and 0 Increasing the healthcare and nutritional status o f children under age five. Designed in two phases over a period o f five years starting in 2000, the pilot phase (also known as Phase I) was for three years with a budget o f US $11 million funded by a loan from the Inter- American Development Bank (IDB), and represented approximately 0.2 percent o f GDP or 2 percent o f annual recurring government spending on health and education (World Bank 2001, annex 21). To permit an assessment o f how a complex program like RPS has altered behavior o f households during an economic downturn, it i s first necessary to describe how the program operates and how it has evolved over time. 44Section 2 draws from Maluccio and Flores (2004) who provide a more complete description of the program. 60 2.1 Program Targeting In the design phase of RPS, rural areas in all 17 departments of Nicaragua were eligible for the program. The focus on rural areas reflects the distribution o f poverty inNicaragua-of the 48 percent o f Nicaraguans designated as poor in 1998, 75 percent resided in rural areas. For the pilot, the Government o f Nicaragua (GON) selected the departments o f Madriz and Matagalpa from the northern part o f the Central Region o f Nicaragua on the basis o f poverty as well as on their capacity to implement the program. This region was the only one that showed worsening poverty between 1998 and 2001, a period duringwhich both urban and rural poverty rates were declining nationally, and this downturn has been attributed in part to the decline in coffee prices (World Bank 2003). Approximately 80 percent o f the rural population o f Madriz and Matagalpa was poor, and half o f those, extremely poor in 1998 (IFPRI2002). Coffee i s grown in many parts o f Matagalpa, which lies at altitudes appropriate for its cultivation (above 800 meters). Both departments had easy physical access and communication (including being less than a one-day drive from the capital, Managua, where RPS i s headquartered), relatively strong institutional capacity and local coordination, and reasonably good health post and school coverage. By purposively targeting, RPS could avoid devoting a disproportionate share o f its resources during the pilot to increasing the supply o f educational and health services. In the next stage o f geographic targeting, six (out o f 20) municipalities were chosen based on criteria similar to those used at the department The six were well targeted in terms o f poverty. Between 36 and 61 percent o f the ruralpopulation ineach o f the chosen municipalities was extremely poor and between 78 and 90 percent was extremely poor or poor (IFPRI 2002), compared with national averages o f 21 and 45 percent, respectively (World Bank 2003). While not the poorest municipalities in the country, or in the chosen departments for that matter, the proportion o f impoverishedpeople living inthese areas was still well above the national average. Inthe last stage of geographic targeting, a marginality index based on information from the 1995 National Population and Housing Census was constructed and an index score was calculated for all 59 rural census "comarcas" (hereafter coma~cas~~) in the selected municipalities. The index was a weighted average o f a set o f poverty indicators (including family size, access to potable water, access to latrines, and illiteracy rates) inwhich higher index scores were associated with more impoverished areas (Arcia 1999).47The 42 comarcas with the highest scores were eligible for the pilot phase's first stage and were included inthe evaluation. 2.2 Program Design RPS has two core components: Food security, health. and nutrition. Each eligible household receives a bimonthly cash transfer known as the bono alimentario or "food security transfer," contingent on attendance at bimonthly educational workshops and on bringing any children under age five for scheduled healthcare appointments. The workshops, held within the communities, educate mothers inhousehold sanitation 4sThe six were Totogalpa and Yalagliina municipalities in the department of Madriz, and Terrabona, Esquipulas, El Tuma-La Dalia, and Ciudad Dario municipalitiesin the department ofMatagalpa. 46Census comarcas are administrative areas within municipalities that includebetween one and five small communities averaging 100 households each. 47IFPRI (2002) describes the WS targeting in more detail. 61 and hygiene, nutrition, reproductive health, breastfeeding, and related topics. To ensure adequate supply in these poor, rural communities, RPS trained (and paid) private providers to deliver the specific healthcare services required by the program-a child growth and monitoring program known as VPCD. These services, provided free o f charge to beneficiary households, are directed toward children and also include vaccination and provision o f anti-parasites, vitamins, and iron supplements. Education: Each eligible household receives a bimonthly cash transfer known as the bono escolar or "school attendance transfer," contingent on enrollment and regular school attendance o f children ages 7-13 who have not completed 4* grade o f primary school. Additionally, for each eligible child, the household receives an annual cash transfer intended for school supplies (including uniforms and shoes) known as the mochila escolar or "school supplies transfer," and contingent on enrollment. Unlike the school attendance transfer, which i s a fixed amount per household regardless o f the number o f children in school, the school supplies transfer i s a per-child transfer. To provide incentives to the teachers, who have some additional reporting duties and were likely to have larger classes after the introduction o f RPS, and to increase resources available to the schools, there i s also a small cash transfer, known as the bono a la oferta or "teacher transfer." Table 1 summarizes the eligibility requirements and demand and supply side benefits o f RPS. Nearly all (about 95 percent) o f the households were eligible for the food security transfer, and this cash transfer was a fixed amount per household, regardless o f household size. Households with children ages 7-13 who had not yet completed the fourth grade o f primary school were also eligible for the education component o f the program. 62 Table 1: NicaraguanRPS eligibilityandbenefitsinthe PilotPhase PROGRAM COMPONENTS ELIGIBILITY Geographic All householdsawith children ages 7-1 3 targeting All householdsa who have not yet completed fourth grade o f primary school DEMANDSIDE BENEFITS Bono alimentario Bono escolar Monetary flood security transfer) (school attendance transfer) transfers C$2,880 per householdper year C$1,440 per householdper year (US$224) (US$112) Mochila escolar (school supplies transfer) C$275 per child beginning o f school year SUPPLY SIDE BENEFITS Bimonthly health education workshops Services Child growth and monitoring Bono a la oferta provided and -Monthly (0-2 year olds) (teacher transfer) monetary -Bimonthly (2-5 year olds) C$60 per child per year given to transfers Provision o f anti-parasites, teacher/school (US$5) vitamins, and iron supplements Vaccinations (0-5 year olds) a. As described inthe text, a small percentage of households were excluded. The amounts for each transfer were initially determinedinUS.dollars (US$) and then converted into Nicaraguan C6rdobas (C$) in September 2000, just before RPS began distribution. Table 1 shows the original US$ annual amounts and their C$ equivalents (using an exchange rate o f 12.85 C6rdobas per dollar): the food security transfer was US$ 224 a year and the school attendance transfer, US$ 112. On its own, the potential food security transfer represents about 13 percent o f total annual household expenditures in beneficiary households before the program. A household with one child benefiting 63 from the education component would receive additional transfers o f about eight percent, yielding a total potential transfer o f approximately 21 percent o f total annual household expenditures. Over the two years, the actual average monetary transfer (excluding the teacher transfer) was nearly SC 3,800 (or 18 percent o f total annual household expenditures). This is approximately the same percentage o f total annual household expenditures as the average transfer in PROGRESA, but more than five times as large as the transfers given in PRAF. The nominal value o f the transfers remained constant, with the consequence that due to inflation the real value o f the transfers declined by about eight percent duringtwo years o f transfers inthe pilot phase. The value o f the supply side services, as measuredby how much RPS paid to the health care providers and the value o f vaccines, anti-parasites, etc., was also substantial, on average over US$ 100per beneficiaryhousehold. To enforce compliance with program requirements, beneficiaries did not receive all components o f the transfer when they failed to carry out any o f the conditions shown in Table 2. The table demonstrates that there are four different "types" o f beneficiary households in the program, who receive different transfers and have to fulfill different requirements. Households with no children in the targeted age ranges are only eligible for the food security transfer, but at the same time need only attend the health education workshops in order to qualify for continued receipt o f the transfers. Households with children under age 5 (but without children ages 7-13 who have not completed the fourth grade) are also eligible for the food security transfer only, but have more requirements to fulfill, related to their young children. Households with children ages 7-13 who have not completed the fourth grade are eligible for both the food security and education transfers and requiredto comply with the schoolingrelated conditions. If,in addition, there are children under age 5 inthe household, it is eligible for the same transfers but has more requirements to fulfill; inparticular, those related to the health controls for young children. RPS allows households to receive a partial transfer if they comply with the health requirement and not the education requirement or vice versa. During the first two years o f delivering transfers, approximately ten percent o f beneficiaries were penalized at least once and therefore did not receive, or received only part of, their transfer. It was also possible for households to be expelled from the program, e.g., for repeated non-compliance. At the start o f the program, about 90 percent o f the households in the intervention areas were participating in the program. Less than one percent of households were expelled from the program duringthe first two years o f delivering transfers, though five percent voluntarily left the program, e.g., by electing to drop out and/or by migrating out o f the program area. 64 Table 2: NicaraguanRPS beneficiaryco-responsibilitiesmonitoredinthe PilotPhase HOUSEHOLDTYPE Households Households with Households with with no children ages children ages 7- targeted 0-5 13 who have not children completed 4" grade (A) (B) (C) (B) (C) + PROGRAMREQUIREMENT Attend bimonthly health education workshops J J J J Bringchildren to prescheduled healthcare appointments Monthly (0-2 years) J J Bimonthly (2-5 years) Adequate weight gain for children under 5' J J Enrollment ingrades 1to 4 o f all targeted children inthe household J J Regular attendance (85 percent, i.e., no more than 5 absences every two months without valid excuse) o f all targeted children inthe J J household Promotion at end o f school year J J Deliver teacher transfer to teacher J J Up-to-date vaccination for all children under 5 years J J a. The adequate weight gain requirement was discontinued inPhase I1starting in2003. b.Condition was not enforced. Only the designated household representative could collect the cash transfers, and where possible, RPS appointed the mother to this role. As a result, more than 95 percent o f the household representatives were women. These representatives attended the health education workshops and were responsible for ensuring that the requirements for their households were fulfilled. 65 2.3 Program Impact Before examining the role o f RPS during an economic downturn, Isummarize the findings fi-om the evaluation, based on a randomized, community-based intervention with measurements before and after the interventioninboth treatment and control comarcas. Overall, RPS had positive (Le., favorable) and significant double difference estimated average effects on a broad range o f indicators and outcomes. Where it did not, it was often due to similar, though smaller, improvements inthe control group. Nearly all estimated effects were larger for the extremely poor, often reflecting their lower starting points (e.g., percentage o f children matriculating before the program)-among poorer beneficiaries there was simply more potential for improvement on many o f the indicators. As a result, the program has reduced inequality o f these outcomes across expenditure classes. RPS in its pilot phase supplemented per capita annual total household expenditures by 18 percent, on average. For beneficiary households, this increase compensated for the large income loss experienced by non-beneficiaries during this period, while producing a small overall increase in expenditures. Most o f the increase in expenditures were spent on food; the program resulted in an average increase o f $C 566 in per capita annual food expenditures and an improvement in the diet o f beneficiary households. Expenditures on education also increased significantly though there was no discernable effect on other types o f investment expenditures. Labor market participation was apparently little changed with the program, though there was an indication o f slightly fewer hours worked on average in the last week. While not designed as a traditional safety net program in the sense o f reacting or adjusting to crises or shocks, the economic difficulties experienced by these communities enabled RPS performed like one, aiding households duringa downturn. For schooling, RPS produced a massive average net increase in enrollment o f 18 percentage points and an even larger increase (23 percentage points) in current attendance for the target population. Examining the number o f children in grades 1-4 who advanced two grades between 2000 and 2002, RPS led to an average increase o f 7 percentage points, despite the fact that advancement past 4thgrade was not a formal requirement o f the program. Intandem with the increased schooling, the percentage o f children ages 7-13 that were working declined by 5 points. Regarding child healthcare, RPS inducedan average net increaseof 11percentagepointsinthe participation of children under three years of age in the VPCD program. At the same time, the services provided by the program, as measured by process indicators including whether the child was weighed and whether their health card was updated, improved even more. Participation by children ages 3-5 also increased substantially. While not possible to statistically demonstrate that RPS increased vaccination coverage for children ages 12-23 months in the intervention group (relative to the control group), it was demonstrated that vaccination rates climbed over 30 percentage points in the intervention and control areas at a time when they were on average decreasing in the remaining comareas in the very same municipalities.One would be hard pressed not to attribute at least some part of this substantial improvementto RPS. Finally, the more varied household diet and increased use o f preventive health care services for children have been accompanied by an improvement in the nutritional status o f beneficiary children age five. The net effect was a five-percentage point decline in the percentage o f children who were stunted. This decline i s more than 1%times faster than the rate o f annual improvement seen at the national level between 1998 and 2001-very few programs inthe world have shown such a decrease 66 instuntinginsuch a short time. Despiteimprovements inthe distribution ofiron supplementsto these same children, however, RPS was unable to improve hemoglobin levels or to lower rates o f anemia. 3. DATA SOURCES, THE SETTING,AND METHODOLOGY 3.1. Data Sources The evaluation design was based on a randomized, community-based intervention with measurements before and after the intervention in both treatment and control communities. One-half o f the 42 cornarcas were randomly selected into the program; thus, there are 21 cornarcas in the intervention group and 21 distinct cornarcas in the control group (IFPRI 2001). Given the geography o f the program area, control and intervention cornarcas are in some cases adjacent to one another, a theme I return to below. The selection was done at a public event with representatives from the cornarcas, GON, IDB, IFPRI, and the media present. The 42 cornarcas were ordered by their marginality index scores and stratified into seven groups o f six each. Within each stratum o f six cornarcas, randomization was achieved by blindly drawing a colored ball without replacement (starting with three blue for intervention and three white for control) from a box after the name of each cornarca was called out. Thus, three cornarcas from each stratum were randomly selected for inclusion in the program, leavingthe other three as controls for the evaluation. The original evaluation was designed to last for one year; that is, the control group was meant to be a control for only one year since there was not sufficient capacity to implement the intervention everywhere at the same time. Due to delays in funding for RPS as a result o f a governmental audit unrelated to the program, incorporation o f beneficiaries in the control cornarcas was postponed until 2003, extending the possible length o f the treatment-control evaluation by more than a year. The control cornarcas had to wait a little over two years before being fully incorporated into the program. The data used here are from an annual household panel data survey implemented inboth intervention and control areas o f RPS before the start o f the program in 2000 and after the program began operations, in 2001 and 2002.48The questionnaire was a comprehensive household questionnaire based on the 1998 Nicaraguan Living Standards Measurement Survey (LSMS) instrument, expanded in some areas (e.g., child health and education) to ensure that all the program indicators were captured, but cut in other areas (e.g., income from labor and other sources) to minimize respondent burden and ensure collection o f high quality data from a single visit interview. As a result, one area where it i s weaker than the typical LSMS comprehensive household survey i s the employment module; the RPS survey only covers activities carried out in the last week and does not ask about earnings from those or any other activities. An anthropometry module for children under age five was also implemented in 2000 and 2002, but not in 2001 and a cornarca-level community survey was implemented in2001. Table 3 outlines the primary data sources used inthe analysis. 48Results reported on here are based on the September 2003 release of the RPS evaluation data. 67 Table 3: NicaraguanRPS data sources I Survey RPS 2000 census RPS 2000 Baseline Follow-up RPS 2001 RPS 2002 Follow-up Month o f survey M a y August/September October October Household Includesweeklast LSMS with shortened LSMS with LSMS with questionnaire labor activities and labor module shortened labor shortened labor land holdings module module Anthropometry Children < 5 questionnaire Children < 5 years years Comarca-level Coffee questionnaire cultivation and comarca shocks The survey sample i s a stratifiedrandom-sample at the comarca level from all 42 cornarcas described above, half o f which were randomly selected for the program. As such, the areas represented comprise a relatively poor part o f the Central Region inNicaragua but the sample i s not statistically representative o f the six municipalities or any other areas o f Nicaragua, for that matter. At the same time, there i s no reason to believe that responses to the program by these households are systematically different from other rural Nicaraguan households. Forty-two households were randomly selected from each comarca using a census carried out by RPS three months prior to the survey as the sample frame and yielding an initial target sample o f 1,764 household^.^^ The first wave o f fieldwork was carried out in late August and early September 2000, without replacement; that is, when it was not possible to interview a selected household, another household was not substituted in its place. While there was a great deal o f progress in getting RPS started throughout 2001, it was not possible to design and implement all the components according to timelines established at the outset. In particular, the healthcare component o f the intervention was not initiated until June 2001. There were also some delays in the payment o f transfers to households duringthe year, due to the governmental audit (mentioned earlier) that effectively froze RPS funds. For these reasons, the RPS 2001 follow-up survey was postponed to the beginning o f October, to allow additional time for the interventions to take effect and for five o f the scheduled six payments to be effected. Of course, the advantage o f the original plan, with the scheduled RPS 2001 follow-up at exactly the same time o f year as in the 2000 baseline, was that it obviated the need to control for seasonal variation, for example in expenditures or labor force participation. With a randomized control group, however, the possible biases introduced by seasonality can be controlled for using the double-difference techniques described below. This difference in the timing o f the survey, then, does not present a serious problem for the 49IFPRI (2001)describes the sample size calculations and Maluccio and Flores (2004) describe the baseline and follow-up samples inmore detail. 68 estimates o f program effects presented, though it i s a potential problem for making definitive statements about changes over time within the control group, a concern addressed in Section 4.1. First round non-response and attrition in the survey are also potential concerns for the analysis. Overall, 90 percent o f the random sample was interviewed in the first round yielding 1,581 successfully completed household interviews, a little more than ten percent o f the area's population (see Table 4). In a handful o f cornarcas the coverage was 100 percent, but in six it was under 80 percent. For the follow-up surveys in 2001 and 2002, the target sample was limited to these 1,581 first round interviews. In 2002, just over 90 percent o f these were interviewed, on a par with similar surveys in other developingcountries (Thomas, Frankenberg, and Smith 2001; Alderman et al. 2001). Again, however, coverage in six o f the cornarcas was substantially worse, where less than 80 percent were successhlly re-interviewed. This attrition i s unlikely to have been random, a theme taken up in section 4.5. Because the same target sample was used in 2002 as in 2001, regardless o f whether the household was interviewed in 2001, some households that were not interviewed in 2001 were interviewed in 2002, and vice versa. The sample o f households for which there i s a complete set o f observations (one in each o f the three survey rounds) i s 1,396, smaller than the 1,434 shown in the first row of the third column o f Table 4. The households are about evenly divided between intervention and control groups, indicating that the level o f attrition, at least, was not significantly different betweenthem. Table 4: Nicaraguan RPS evaluation survey non-response and subsequent attrition 2000 2001 2002 CompletedInterview 1581 1490 1434 (89.6) (94.2) (90.7) Completedinterviewinall 3 rounds 1396 1396 1396 (79.1) (88.3) (88.3) ...ofwhich Intervention 706 706 706 (percent of targeted intervention sample) (80.0) (87.2) (87.2) Control 690 690 690 (percent of targeted control sample) (78.2) (89.5) (89.5) Not Interviewed Uninhabiteddwelling 60 51 83 Temporaryabsence 100 28 46 Refusal 16 6 12 Urban(misclassifiedas rural) 6 0 0 Lost questionnaire 0 6 6 Target Sample 1764 1581 1581 Notes: Percent of target sample inparentheses. 69 3.2. The importance of coffee in Nicaragua and the %offee crisis" Coffee production in Nicaragua more than doubled from 932 thousand quintals (or hundred-weight) in 1990 to 2,083 in 2000, but declined to 1,800 in 2001.50Over this ten-year period, productivity increased dramatically, with on farm average yields more than doubling. The vast majority o f coffee produced in Nicaragua i s exported, and most o f it i s strictly high grown (SHG) arabica (grown at altitudes greater than 800 meters) and therefore commands a high price; indeed Nicaraguan coffee often sells at a premium (e.g., in July 2002 it sold for a $3 premium over the New York coffee C contract price for September delivery at the exchange for strictly high grown per quintal). Over the last 5 years, coffee exports averaged US$140 million, or about one-quarter o f total export earnings, and it was the single most important agricultural export (Kruger, Mason, and Vakis 2003). Inaddition, the coffee sector is a major employer inthe rural economy. Estimates of the importance o f coffee inrural labor markets vary substantially, from 20 percent o f the rural labor force employed at some point in the year in the coffee sector (Kruger, Mason, and V a h s 2003) to 40 percent (Varangis et al. 2003). Even if the more conservative estimates are taken, it i s clearly an important source o f rural employment. Approximately two-thirds o f this employment i s seasonal, while the remainder i s self-employed or permanent farm workers (Varangis et al. 2003). Hence, despite the fact that Nicaragua i s only a minor producer on the world stage-and therefore a price-taker in world marketsnoffee i s a major export crop and employer for the Nicaraguan economy, and declining prices have had important effects on the economy. International nominal year-end prices inUS.cents per pound reported for Arabica coffee were nearly 160 in 1997 but had dropped over 50 percent to 71 cents in 2000, and again by a third to 46 and 40 cents in 2001 and 2002, re~pectively.~~ Unit export values declined in recent years in tandem with the price declines, from $121 per quintal in 1997 to $81 in2000, and $54 in2001. The latter prices are unlikely even to cover production costs for some producers (Lewin and Giovannucci 2003). It i s this fact that, while perhaps not coming as a surprise (to coffee analysts, anyway), leads many to refer to the current situation as a crisis. Many farmers have been forced to reduce or abandon coffee production, and it has been estimated that 35,000 permanent and 100,000 seasonal jobs have been lost (IDB 2001). 3.3. Coffee cultivation in the RPS sample Via a cornarca-level survey that accompanied the household-level instrument in 2000 and was administered to key informants, 21 o f the 42 cornarcas in the sample were identified as being areas where coffee i s cultivated, 10 in the intervention group and 11 in the control group (see Table 5). Because the cornarcas are spread across six municipalities in two departments, however, this apparently even allocation masks the fact that all the coffee producing areas are located in the department o f Matagalpa, which i s about 100 kilometers closer to Managua than Madriz. As such, in addition to analyzing the complete sample, Iwill also assess whether, and how, limitingthe sample to cornarcas inMatagalpa changes any o f the results presentedbelow. Except where otherwise cited, statistics cited in this paragraphare drawn from Varangis et al. (2003). ''The world coffee prices are December values from the InternationalCoffee Organization. 70 Table 5: Coffee cultivationinRPS sample at comarca level Type o f comarca 1 Coffee cultivating Non-coffee cultivating Total Intervention 10 11 21 Control I 21 I I 11 10 II Total 21 21 21 I I From the labor force participation questions asked about the previous week (in each survey), we can glean partial information on the extent to which individuals and households are participating in the coffee industry. For agricultural labor activities, the type o f crop was not numerically coded. Interviewers, however, were trained to write a brief description o f the activity and when coffee was involved, the descriptiontypically included the word "caf6." All jobs inwhich "caf6" was noted down are treated as coffee sector jobs-therefore figures presented below most likely represent a lower bound for individual and (to a lesser extent) household-level participation in the last week since it seems likely that interviewers at times neglected to specify coffee when the work was in coffee. In addition, given the seasonal and sporadic nature o f coffee production, the one-week reference period i s almost certainly inadequate to capture all those who ever work in coffee, and i s also very likely to miss many o f those who occasionally work in coffee (for example, only during the harvest season), further understating involvement in the industry, though it i s difficult to say by how much using only the RPS evaluation data.s2In describing patterns and descriptive regressions using this information, then, Iemphasize that in comparison to the other analyses, results that pertain to household-level participation inthe coffee sector are less definitive. While inAugust and September 2000 nearly eight percent o f those reporting that they had worked in the previous week indicated that they worked in coffee, this percentage had dipped to under five percent in 2001 and 2002. These workers were spread across 14 percent o f the households in 2000 and 10 percent in 2001 and 2002. These percentages appear to be low in comparison with estimated levels o f about 20 percent from the rural subsample o f the 2001 LSMS (Kruger, Mason, and Vakis 2003). Hardly any o f the coffee workers resided outside an identified coffee growing area, however, so the percentage participating in those areas alone i s twice as large. This pattern i s consistent with the demarcation o f coffee and non-coffee regions and suggests that the comarca-level information i s broadly accurate. As with most crops, the demand for casual labor in coffee rises during the harvest season, which begins in October but peaks inDecember and January. The decline between 2000 and 2001/2, then, i s somewhat surprising since during a typical year the seasonal difference in the timing Another source of information on participation inthe coffee sector is the food expenditure and consumptionmodule inthe questionnaire, where those food items consumedfrom own-production,or received as in-kindpayment, are indicated.Ineach year, asmallpercentageof householdsreport consumingcoffee from these sources in the previoustwo weeks. When this informationis contrastedwith those who worked incoffee inthat last week, however, only about half of those indicating consumption of coffeefrom these sourcesreportedworking inthe coffee sector, suggestingthat the shortreference periodand use of written job descriptionsmaybe understatingparticipationinthe sector by as muchas one-half. 71 o f the survey would lead to more reported coffee work inthe October period, not less. This i s the first piece o f evidence suggesting that participationinthe coffee sector i s declining. Between 10 and 15 percent o f those reporting workmg in the coffee sector indicated that they were self-employed farmers (from less than two percent o f all households), and this percentage changed little over the three surveys. Over two-thirds o f those working in coffee are men and only ten percent are children. In2000, seven percent o f those working in coffee indicated that they were employed as permanent workers on a coffee farm, but virtually none did in 2001 or 2002, consistent with local media reports that larger coffee farms (which are the ones that employ permanent laborers) had to release labor in recent years. As a result o f these small sample sizes, it i s not feasible to distinguish between coffee farmers (the sample has on average 30 o f them in each year) and laborers inmost o f the analyses presented below. A simple comparison o f per capita expenditures across these two groups in the coffee sector, however, does show that coffee farmers were substantially better o f f in 2000, with 30 percent higher average expenditures than households with coffee laborers who were not self-employed. The average percentages across the years belie the fact that many individuals and households report moving in and out o f coffee-only one-third o f the households reporting participation in coffee in 2000 also report participation in 2002, for example. This movement i s shown in household-level transition matrices between 2000 and 2001 (Table 6a) and between 2001 and 2002 (Table 6b). A household i s defined to be in the coffee sector if any adult (aged 15 or older) in the household reported any sort o f participation in coffee in the last week. Between 2000 and 2001 there seems to have been significant exit from the coffee sector (and this despite the timing o f the survey which favors greater participation in October than in August and September)53 whereas on net only one percent exited between 2001 and 2002. Much o f the variability or churning (e.g., in2001-2002 where nearly as many households entered as exited) i s almost certainly due to the short reference period considered and likely does not reflect longer run changes. Thus the patterns seen here are consistent with the media representation o f a crisis in the 2000-2001 season, and consequent reduced labor demand on coffee farms; households appear to have been adjusting and "exiting" coffee over time in these areas. Of course, this description o f the data considers the entire sample and therefore conflates effects o f the crisis with those o f RPS. Similar, though slightly weaker patterns emerge when the analysis i s restrictedto households in the control group only. Ianalyze these transitions into and out o f coffee more formally insection 4.3 below. ''Supportingthe hypothesisthat there is seasonal variation inparticipation in coffee is evidence taken from the May 2000 RPS population censuswhich shows only four percent of those reporting working incoffee in the last week since labor demand is even lower duringthat part of the coffee season. 72 Table 6a: Coffee cultivation in RPS sample at householdlevel (2000-2001) I I Coffee cultivating in Non-coffee cultivating in 2001 2001 Total Coffee cultivating in 74 128 202 2000 (5.3) ( 9 4 (14.5) Non-coffee cultivating in 65 1129 1194 2000 (4.7) (80.8) (85.5) I I Total 139 1257 1396 (10.0) (90.0) (100.O) I I Table 6b: Coffee cultivationin RPS sample at household level (2001-2002) Coffee cultivating in Non-coffee cultivating in 2002 2002 Total Coffee cultivating in 59 80 139 2001 ( 4 4 (5.7) (10.0) Non-coffee cultivating in 72 1185 1257 2001 (5.2) (84.9) (90.0) I I I Total 131 1265 1396 (9.4) (90.6) (100.0) I I 3.4. Econometric methodology The empirical approach exploits two key features o f the data allowing one to overcome the vast majority o f typical concerns regarding econometric estimation and causal inference: 1) the randomized design o f the evaluation and 2) the panel structure, i.e., the fact that the same households were interviewed over time, before and after RPS was implemented and in both intervention and control comarcas. Iestimate a series o f reduced form specifications that essentially estimate program effects, differentiating them for households residing incoffee or non-coffee growing areas.54 ''Thisapproach differs from that of Coady, Olinto, and Caldes (2003) inthat they estimate structural equations. As in the work presented here, identificationrestsprimarily on the randomization of the intervention. An advantage to their approach i s that imposing more structure (and its attendant assumptions) facilitates exploration of the pathways of effects and all coefficients on variables that are interacted with 73 The methodology used i s based on difference-in-difference techniques and yield what i s commonly referred to as the "average program impact."" The resulting measures can be interpreted as the expected effect o f implementing the program ina similar population elsewhere. The method i s shown inTable 7. The columns distinguishbetween groups with and without the program (denoted by Ifor intervention and C for control) and the rows distinguish between before and after the program (denoted by subscripts 0 and 1). Anticipating one o f the analyses presented below, consider the measurement o f school enrollment rates for children. Before the program we would expect the average percentage enrolled to be similar for the two groups, so that the quantity (Io -Co)wouldbe close to zero. After the program has been implemented, however, we would expect differences between the groups as a result o f the program. Furthermore, because o f the random assignment, we expect the difference (I1 -C,) tomeasuretheeffectdirectly attributabletotheprogram.Indeed,(Il - C,) i s a valid measure o f the average program impact under this experimental design. A more robust measure o f the effect, however, would account for any preexisting observable or unobservable differences between the two randomly assigned groups: this i s the double difference obtained by subtracting the preexisting differences between the groups, (b - Co),from the difference after the program has been implemented, (Il -C,). Table 7: Calculation of the double-difference estimate of average program effect Measurement Interventiongroup with Control group Difference across RPS program without RPS program groups Follow-up C1 11 - CI Baseline I O CO I o -co Double-difference Difference across time 11 - I o CI -co (11 - - (Io - C,) CO) An alternative interpretation o f the double-difference estimator emerges if one first considers the differences within the (intervention or control) groups. This approach begins with a nake estimator o f the program effect, the difference over time for the intervention group, (I1 This i -Io).snakebecause it would include all changes over time in enrollment rates, regardless o f what is causing them. For example, if increases in public investment nationally were improving school access and leading to changes in enrollment, these effects would show up in the difference over time in the intervention group, in addition to the effects attributable to the program. The obvious measure for the non-program related change over time inthe intervention group i s the change over time in the control group, (C, - Co). Thus we estimate the average program impact by first considering the total change over time in the intervention group, and then subtracting from this the change over time in the control group. As above, this yields the double-difference estimator. program availability, such as ownership ofcoffee land, are consistentlyestimated. Ido not follow their approach mainly because of the relativelypoor quality o fhousehold-levelinformation on participationin the coffee sector in the RPS evaluation data. 55Ravallion (2001) provides a useful, and enjoyable, discussion on this and related evaluation tools. 74 The alternative interpretation i s probably best illustrated graphically as shown in Figure 1. For an arbitrary indicator that we measure over time, we assume (for the graph) that as a result o f the randomization both the intervention and control groups start at the same level. No change in the indicator over time would lead to the outcome depicted by point A in2002; ifwe were only following the intervention group we would then naively calculate the effect o f the program as the quantity (C - A). However, as the control group makes clear, there was a trend over time that led to an improvement, even in the absence o f the program, o f (B - A). Were we to ignore this we would overstate the effect o f the program. Instead, our estimate o f the program effect i s (C - B); this i s the double difference estimate. In the case where rather than increasing, the trend line for the control group were declining, ignoring that effect would tend to understate the program effect. Figure 1: Illustration of the double difference estimate o f average program effect I b 2000 2002 For this work, the technique just described i s extended to a triple difference, where the third difference i s whether or not the household resides in an area where coffee i s cultivated. A simple triple difference estimating equation for two periods i s shown inequation (1). Where Eict=outcome variable o finterest for householdi incomarca c at time t Y, = (1) ifsecondperiod (or year) K, =(1) ifcoffee cultivatingcomarca c P, = (1) ifprogram comarca c I&= all (observed andunobserved) comarca-level time invariant factors =all (observed andunobserved) household-level time invariant factors vict= unobserved idiosyncratic household and time varying error 75 and all the aj and 6, are unknown parameters. The parameters o f interest are ti2,the "double difference" estimator o f the average program effect in non-coffee growing areas and 63, the "triple difference" estimator o f the average program effect in coffee growing areas relative to non-coffee growing areas. The total estimated program effect in coffee growing areas, then, i sa2+ 63. In section 4.4, Islightly modify this framework in order to introduce household-level shock information for the 2-year period (collected in the follow-up surveys). Rather than estimate the household fixed-effects regression in levels form as presented above, Itake first differences (year 2001 - year 2000 and year 2002 - year 2001) and then introduce on the right hand side indicators for the shocks (described inmore detail insection 4.4). One potential concern about the classification into coffee growing areas i s that because coffee cultivation requires specific agro climatic conditions, the opportunity set for production technologies may differ across areas that do and do not cultivate coffee. Put another way, the fact that coffee i s grown in an area i s related to other production and labor market decisions in the area. A second concern i s that households choosing to live incoffee or non-coffee cultivating regions are different in other ways that may be directly associated with the outcome variables under consideration. These suggest possible correlation betweenthe coffee region indicator and b or b,that, ifnot controlled for would contaminate estimates o f all coefficients on any variable including the coffee indicator, with the exception o f those in which it i s interacted with the random program dummy variable, P,. To avoid this possibility, household fixed effects are explicitly included in all the analyses, thereby controlling for any time-invariant unobserved heterogeneity that may be associated with the location o f the household. Another implication o f including household fixed effects i s that all estimated results implicitly include comarca-level fixed effects as well, so that Icontrol for the potential problem that the coffee indicator i s correlated with any omitted fixed factors at the comarca level. Of course, all other time-invariant factors, such as K,, now drop out o f the relation and (extending to all three survey years) we are left with (2) as the main estimating equation with Y1(Y2)a dummy for 2001 (2002). 621is the "double difference" estimator for 2001 (relative to 2000) and 822for 2002 (relative to 2000). 831and832are the respective "triple difference" estimators. The coffee versus non-coffee classification at the comarca level i s necessarily crude-neither the coffee labor nor the coffee production markets are completely segregated across cornarcas in the sample. The fact that coffee i s cultivated in an area does not imply that all households in the area participate in the coffee industry (via labor, production, marketing, etc.) and those not directly participating could be affected less by coffee price declines though they could still be affected, for example by changes inlabor supply for non-coffee sector jobs. Conversely, households living inareas where coffee i s not planted may still participate in the industry as (temporary) migrant laborers, a common practice during harvest periods, so that they could be directly affected by price declines. As a result, the estimated coefficient on an indicator o f whether an area has coffee cultivation (relative to one that does not) interacted with the year 2001, for example, will tend to understate the size o f the effect o f the price decline for those households actually participating in the industry, and therefore directly affected, a result similar to the difference between the intent to treat effect and the treatment effect. The majority o f analyses presented below estimate this "intent to treat" effect, in large part 76 because o f the data limitations on coffee participation at the household level. The approach has the advantage o f being conservative, thereby increasing confidence in the results when significant differences are found. Finally, Iemphasize that program effects are identifiedby the randomized design o f the intervention. They do not, for example, condition on household choices or rely on treating the rapid coffee price decline as a shock-an assumption that i s hard to maintain in the face o f historical trends and given that the recent downturn inprices began inthe late 1990s whereas the data Iexamine start in 2000.56 As such, however, the results presented below are likely to understate the effects o f RPS in coffee growing regions to the extent that households have already undertaken various strategies in reaction to the continued price declines. This i s in the same direction as the possible biases described above and reinforce the claim that the methodology used i s a conservative one. With the exception o f section 4.4 where Iexamine household-level idiosyncratic shocks, Iemphasize throughout that the estimates presented refer to the effects o f the program during an economic downturn, and not in response to an economic shock. Inthe analysis that follows, Iwork with all (relevant) individuals or households from the balanced panel sample (of 1,396 households interviewed three times each) to keep from changing sample composition in estimating the differences between 2000 and 2001 and between 2000 and 2002. Inall but the few instances that are indicated, the estimates control for household-level fixed effects as described above. 4. RESULTS 4.1. The changing environment: Control group Ifirstconsiderpatternsofexpenditures,laborforceparticipation,schoolenrollmentrates,andchild nutritional status inthe control areas during the period 2000-2002, contrasting coffee and non-coffee growing areas. T o summarize what follows, the evidence shows that expenditures have declined over the period while work effort has intensified, and these changes are on average substantially larger in coffee-growing areas. Primary enrollment rates improved modestly over the period, somewhat more so withincoffee growing areas, and the incidence o f child labor for young boys declined inall areas. Inthe top panel ofTable 8, Ipresent descriptive results for expenditure and labor force participation measures over time for the control group. Real per capita annual household total expenditures measured inbase year 2000 Cordobas dropped by nearly 20 percent between 2000 and 2001, but held 56While coffee prices have declined substantially inrecent years, the current crisis was not unexpected since the industry has been anticipating large increases inBrazilian output for some time. "Apart from over-supply, there are two principal factors underling the current crisis: a structural change inthe nature o f supply, particularly increases inboth the quantity and quality o f Brazil and Vietnamese coffees, and 0 Structural changes indemand, comprising increasing demand for high-end specialty products, new technology allowing greater flexibility inblending, and generational shifts inthe appeal of different types o f coffee products" (Lewin and Giovannuci 2003, page 5). 77 steady between 2001 and 2002.57358similar pattern, but with a larger percentage decline, is observed A for per capita household food expenditures. At the same time that expenditures were declining, labor supply was increasing, as measuredby the total number o fhours worked by householdmembers and the average number o f hours worked per worker (shown in the third and fourth columns, respectively). In2001 and 2002, workers reported working on average more than one additional hour a week, compared to in 2000. Households in the control group faced declining expenditures despite increasedlabor hours. An important consideration is whether these patterns hold for bothpoor and non-poor households. To examine this, Iconsider two separate categorizations o f households, one usingtheir predicted poverty status based on a proxy means model that predicts per capita expenditures for each household based on a set o f indicators measured at the household level that are highly correlated with logarithmic per capita expenditures (R2> 0.50) and a second based on the size o f their landholdings being less or greater than one he~tare.~'Both o f these show that while expenditures decreased for both poor and non-poor groups over the period, the decline was concentrated in non-food expenditures and sharper for the less poor, the group that may have been better able to withstand reductions; the labor force participationtrends were similar across groups (results not shown). There were two possible factors leading to a downturn in the area, a drought in 2001 and the continued decline in coffee prices.60 In the 2001 comarca-level survey, 38 o f 42 comarcas reported the drought as a significant event during the past year, indicating that it was pervasive inthe program areas. T o explore whether the decline was specifically related to coffee, Icompare coffee and non- coffee growing areas (the bottom panel o f Table 8). Households in coffee growing areas started out with somewhat higher (by about 10 percent) expenditures but this advantage was reversed over the period as they experienced significantly larger declines in expenditures. Average per capita expenditures in the non-coffee growing areas were about ten percent lower in 2001 and 2002 than in 2000, but a further 20 percent lower in coffee growing regions. Total hours worked in the last week by household members increased modestly in 2001, by about 5 hours per week, and substantially in 2002, about 18 hours per week. Average hours per worker in the last week increased in coffee growing areas inboth 2001 and 2002. Changes inhouseholds in coffee growing areas are driving the overall average trends depicted inthe first panel o f Table 8. j7 All reported expenditures havebeendeflated to 2000base Cordobas usingthe Nicaraguanconsumer price indexreported by the Central Bank of Nicaragua for which there was approximately four percentinflationper year in2001and 2002. 58 The drop inexpenditures in the controlgroup was not due to changesin householdsize or family composition, which did not significantly change.Another possibility is that there are biases inthe reporting of expenditures.For example, incontrol areas it is possible that non- beneficiarieswho had learnedabout the programunderstatedtheir expendituresin an effort to appear more inneedof the program. However, at this stage, the programwas beingimplemented usingonly geographicaltargeting, and beingmore or less poor would not have affected their eligibility Additional evidencethat the decline in expenditures is real comes from the decline innutritionalstatus of children, which i s not subject to the same sort of possiblereportingbias. j 9Iusepredictedpovertystatusratherthanactualmeasuredexpenditurepovertysincethelatterislikelytoleadtoregressiontothemean given measurementerrors in expenditures. IFPRI(2002) contains the details. 6o HurricaneMitch (October 1998)did not severely affect the RPS programareas so it is not a concernfor interpretingthe results. 78 Table 8: Expendituresand labor force participationinthe controlgroup,2000-2002 Lnper Lnper Avg hours capita real capita real Total hours Per annual annual worked worker expend. food worked expend. last week last week Year 2001 -0,1895 *** -0.2473 *** 2.0391 1.4082 *** (7.50) (8.55) 5.9565 *** (1.03) (2.92) Year 2002 -0.1767 *** -0.2331 ** 1.1056 ** (6.99) (8.06) (3.01) (2.29) Constant 8.0166 *** 7.6370 *** 84.1884 *** 25.5777 *** (448.9) (373.6) (60.25) (74.97) Year 2001 -0.0928 *** -0.1324 *** -0.1701 -0,0088 (2.77) (3.46) (0.07) (0.01) Year 2002 -0.1053 *** -0.1661 *** -3.6031 -1.7101 *** (3.15) (4.34) (1.38) (2.70) Year 2001 x coffee -0.2208 *** -0.2626 *** 5.0476 3.2377 *** (4.37) (4.54) (1.28) (3.38) Year 2002 x coffee -0.1629 *** -0.1531 *** 21.8415 *** 6.4331 *** (3.22) (2.64) (5.55) (6.72) Constant 8.0166 *** 7.6370 *** 84.1884 *** 25.5777 *** (451.9) (376.1) (60.95) (76.14) F-test Year 2001 + 68.41 *** 82.80 *** 2.73 * 20.22 *** Year 2001 x coffee [<0.01] [<0.01] [O. lo] [0.07] Joint test year 2002 50.05 *** 54.06 *** 38.14 *** 43.25 *** +Year 2002 xcoffee [<0.011 [<0.01] [<0.01] [<0.01] F-test overall 22.94 *** 28.57 *** 10.83 *** 13.72 *** regression [<0.01] [<0.01] [<0.01] [<0.01] Number o f observations 2070 2070 2070 2070 Notes: Household-level fixed-effects estimation incontrol group only. T-statistics reported inparentheses and p-values inbrackets. * indicates significance at lo%, ** indicates significance at 5%, and *** indicates significance at 1%. 79 Inextconsiderhowschoolenrollmentratesandchildlaborhavechangedovertimeinthecontrol group. Since schooling and child labor decisions depend on the opportunity cost o f children's time as well as costs o f schooling and the resources the household commands, it i s not possible a priori to predict the direction o f the effects o f an economic downturn since opportunity costs and resources may both be changing, with opposing influences for household decisions. In 2000, though less than 20 percent reportedworlung, children ages 7-12 were more likely to report having worked inthe last week in coffee growing areas versus non-coffee growing areas (19 versus 12 percent). Possibly reflecting these differing work patterns, net primary enrollment rates for the same children were substantially lower incoffee growing areas (66 percent versus 87 percent). Inthe first two columns ofTable 9, Ipresent household-level fixed-effects estimates ofthe changes in enrollment rates for girls and boys over time in the control group, conditional on age in years. Enrollment rates were substantially higher in 2000 for girls (83 percent) than for boys (74 percent). For both girls and boys there was hardly any change from 2000 to 2001, but enrollment rates were up significantly for both groups in 2002 (relative to 2000), and more so for boys, who made relative gains over the period. Turning to the bottom panel o f the table in which Iagain distinguishbetween coffee and non-coffee growing areas, we see that most o f the gains over the period were concentrated in coffee growing areas; by 2002, about one-third of the gap between net primary enrollment rates that existed in2000 between coffee and non-coffee growing areas hadbeen overcome. While girls inthis age group rarely reported working (on average less than 10percent do), about one- quarter o f the boys ages 7-12 reportedworking in2000. By 2002, however, this had declined to about 15 percent in both coffee and non-coffee growing areas (see third and fourth columns o f Table 9). The same pattern holds for their older siblings between ages 13-17 (not shown). It would seem that the downturn did not adversely affect enrollment and, if anything, had negative effects on the incidence o f child labor for young children, possibly because o freduced labor demand. 80 Table 9: Primaryenrollment, child labor, and child nutritionalstatus inthe control group, 2000-2002 (1) if (1) if (1) if (1) if HAZ 7-1 2 year 7- 12 year 7-12 year 7-12 year children old old old old 6-48 enrolled: enrolled: working: working: months o f GIRLS BOYS GIRLS BOYS age Year 2001 -0.0107 0.0004 -0.0094 -0.0953 *** (0.51) (0.02) (0.65) (4.11) Year 2002 0.0468 ** 0.0701 *** -0.0178 -0.0933 *** -0.1480 * (2.09) (2.87) (1.15) (3.74) (1.77) Age inyears 0.0133 ** 0.0202 *** 0.0191 *** 0.0741 *** -0.0045 (months infinal (1.98) (2.67) (4.12) (9.63) (1.32) column) (1) ifmale d a d a d a d a 0.0613 (0.73) Constant 0.6997 *** 0.5581 *** -12.8903 *** -0.4558 *** -1.6337 *** (11.40) (8.06) (3.03) (6.46) (13.93) Year 2001 -0.0139 -0.0288 -0.0057 -0.0772 *** (0.5 1) (0.97) (0.31) (2.54) Year 2002 0.0407 0.0067 -0.0401 ** -0.0870 *** -0.0639 (1.42) (0.21) (2.03) (2.68) (0.63) Year 2001 x coffee 0.0079 0.0712 -0.0077 -0.0428 (0.19) (1.57) (0.27) (0.92) Year 2002 x coffee 0.0149 0.1457 *** 0.0551 * -0.0163 -0.1616 (0.34) (3.12) (1.83) (0.34) (1.37) Age inyears 0.0132 ** *** *** *** (months inlast (1.97) 0.0205 0.0186 0.0743 -0.0046 column) (2.72) (4.02) (9.64) (1.35) (1) ifmale d a d a d a d a 0.0684 (0.82) Constant 0.7002 0.5538 *** -0.1247*** -0.4563 *** -1.6347 *** (11.39) (8.03) (2.94) (6.46) (13.95) F-test Year 2001 + 0.04 1.49 0.37 11.41 *** Year 2001 x coffee [ O M ] [0.22] [OS41 [<0.01] Joint test year 2002 2.66 * 18.01 *** 0.41 7.88 *** 2.38 * +Year 2002 xcoffee [0.10] [<0.01] [OS21 [<0.01] [0.09] F-test overall 3.21 *** 7.19 *** 4.53*** 20.28 *** 1.84 regression [<0.01] [<0.01] [<0.01] [<0.01] [0.12] Number o f observations 1196 1190 1196 1190 774 Notes: Household-level fixed-effects estimation in control group only for fust four columns; ordinary least squares estimation with robust standard errors allowing for heteroskedasticity in the final column (Stata Corporation2001). T-statistics reported inparentheses and p-values inbrackets. * ** indicates indicates significance at 109'0, significance at 5%, and *** indicates significance at 1%. 81 One concern withthe above analysis relates to the timingo f the surveys since the baseline was carried out in August/September and the follow-up surveys in October. It i s possible that seasonal variation inconsumption or work could leadto part or all ofthe observed changes. Indeed, when broken down by recall period, the higher frequency periods show declines in expenditures across the surveys but the longer recall periods (that include non-food items) o f 1month, 6 months, and 12 months do not. If all periodicities, including the longer recall periods, showed a decline, we could more confidently say that the observed declines are not due to seasonal variation inexpenditures.61 The first piece o f evidence Ibringto bear on whether the results presented inTable 8 are due solely to seasonality comes from an independent source o f information, a quality control survey carried out on a five percent sample o f the households interviewed in the 2000 baseline. This survey was implemented in October 2000-so that the timing exactly matches the follow-up surveys. The estimate o f the mean and median change in the logarithm o f per capita expenditures and per capita food expenditures shows that they increased slightly over the period-in both coffee and non-coffee producing areas, though these increases are not statistically significant. A comparison o f number o f workers, total hours worked, and average hours worked per worker also show slight (but insignificant) increases. Thus, at least in2000, the baseline survey year, there was no dramatic decline in expenditures between August and October, supporting the interpretationthat the changes we see between2000 and 2001/2002 are real changes resulting fi-om the economic downturn. The second piece o f evidence supporting the hypothesis that the downward trend in expenditures reflects a real economic downturn and not merely seasonality i s shown inthe final column o f Table 9 where Ipresent the findings for height-for-age z-scores o f children ages 6 4 8 months o f age.62Due to planning difficulties, the anthropometry survey in 2000 was carried out separately from the main household survey work-in September and early October 2000. Thus, for anthropometry, the 2000 and 2002 surveys were implemented closer together, so that seasonal variation i s not a concern when comparing them. There was a significant decline in the nutritional status o f children in the control areas over the period, and this decline appears to have been more severe for households in coffee growing regions (see joint F-test in third to bottom row). When broken down by sex, the height-for- age z-scores for boys, which on average was slightly higher than girls at the outset, deteriorated more severely with the result that in2002 the two were nearly identical (controlling for age)-none o f these differences by sex are statistically significant, however. 4.2. Effect of theRPS on households in coffee cornarcas Governmental responses in Central America to the decline in coffee prices, including those o f the Nicaraguan government, were slow to materialize and initially have focused attention on producers, traders, and exporters, rather than laborers, even though it i s the latter who appear to be more vulnerable. Further, because the initial responses tended to be directed via the financial sector, they favored medium and large enterprises to the detriment o f the small-scale producers prevalent in Nicaragua.63 Since many o f these households also cultivate other crops, and the downturn in prices was accompanied by a drought (at least in 2001), their livelihoods were doubly threatened (Varangis et al. 2003). 6'O f course, for seasonalvariation to be driving the difference between coffee and non-coffee areas, there would also need to be different patterns of seasonal variation between the groups. 62Unlikethe other regressionsreported inTable 9,the height-for-agez-score specification is not estimated usinghousehold-levelfixed effects because the sample for which there is a child between the age of 6 and 48 months from the same household measuredinboth 2000 and 2002 i s too small for precise estimation. Varangis et al. (2003) estimate that 90 percent of producers in Nicaraguaproduce less than 100quintals. 82 Varangis et al. (2003) outline a variety o f possible responses to the decline in prices, ranging from price risk management instruments (see McCarthy and Sun 2003 for a discussion o f these in Honduras) to food-for-work programs. They call for improving social safety net programs, o f which RPS i s one example, making this analysis complementary to theirs. One recently begun Nicaraguan program they describe i s Plan Cufk, which aims to help both large producers and laborers alike. Participants are employed on private coffee farms and are paid inpart by the farm owners and inpart by the government, inthe form of food. Unfortunately, the RPS evaluation data analyzed inthis paper were collected before this program was widely implemented so an assessment o f its effects i s not possible here. The results presented below should be interpreted as what happens in the absence o f a governmental response. In Section 2, Idescribed the average effects of RPS on a variety of outcomes. In this section, I demonstrate that RPS has had greater average impacts in coffee growing versus non-coffee producing regions, for many indicators. Of course, this i s not surprising since there was more "potential" for the program to have impact where the situation was worse or deteriorating more rapidly. This i s similar in spirit to the general finding in the overall RPS evaluation that double difference estimated average impacts tend to be larger among the poorer groups in the sample, where there was often more potential, for example due to lower enrollment rates among the extreme poor (Maluccio and Flores 2004). Inboth years, the program positively and significantly improved per capita total annual household expenditures and per capita household food expenditures. Across all program areas, RPS increased these expenditure measures by nearly 20 percent on average (see Table 10). In 2001, the program effect in coffee producing areas was substantially larger than in non-coffee areas. This differential, however, was substantially smaller in 2002 where, again, the program had a significant impact on expenditures though it was not significantly larger in coffee versus non-coffee growing areas. As I argue below, this may reflect increased labor effort between 2001 and 2002 by non-program recipients incoffee growing areas. 83 Table 10: The effect of NicaraguanRPS on expenditures and labor force participation, 2000-2002 Lnper capita Lnper capita Total hours Avg hours real annual real annual worked last per worker expenditures food worked last expenditures week week Year 2001 -0.0928 *** -0.1324 *** -0.1701 -0,0088 (2.92) (3.57) (0.07) (0.01) Year 2002 -0.1053 *** -0.1661 *** -3.6031 -1.7101 *** (3.31) (4.48) (1.45) (2.85) Year 2001 x coffee -0.2208 *** -0.2626 *** 5.0476 3.2377 *** (4.59) (4.69) (1.35) (3.57) Year 2002 x coffee -0.1629 *** -0.1531 *** 21.8415 *** 6.4331 *** (3.39) (2.73) (5.82) (7.09) Year 2001 x RPS area 0.1816 *** 0.2781 *** -3.9191 -0.4825 0.1749 *** (4.02) (5.28) (1.11) (0.57) Year 2002 x RPS area 0.2618 *** 0.3406 0.7732 (3.97) (4.97) (0.10) (0.91) Year 2001 x coffee x 0.2789 *** 0.2553 *** -13.0845 ** -4.2388 *** RPS area (4.14) (3.25) (2.49) (3.33) Year 2002 x coffee x 0.0657 0.0561 -23.4683 *** -5.2571 *** RPS area (0.97) (0.71) (4.46) (4.13) Constant 8.0599 *** 7.6714 *** 81.9047 *** 25.2518 *** (679.85) (555.47) (88.56) (112.87) F-test Year 2001 + Year 84.58 *** 83.61 *** 18.95 *** 24.97 *** 2001 x coffee [<0.01] [<0.01] [<0.01] [<0.01] Joint test year 2002 + 23.99 *** 29.71 *** 35.06 *** 22.52 *** Year 2002 x coffee [<0.01] [<0.01] [<0.01] [<0.01] F-test overall regression - 17.31 *** 19.19 *** 8.88 *** 8.97*** [<0.01] [<0.01] [<0.01] [<0.01] Number o f observations 4188 4188 4188 4188 See notes to Table 8. Overall, the program had little significant effect on either total number o f hours worked last week or hours worked per worker, but within coffee growing areas it had a negative effect on both.64These effects were larger in 2002 than in 2001, possibly explaining the weaker program effect on expenditures in that year as households without the program worked harder to make up for lost consumption. The estimated effects are driven largely by male labor which comprises about 90 percent o f the total reported labor; excluding women does not change the findings and estimating the relationship for women alone leads to similar conclusions. The negative estimated impact on labor 64The program also did not affect the number o f adult householdmembers. Notice that unlike the discussion earlier regardinga concern that the timing o f the survey may affect changes across rounds, the double (and in this case triple) difference estimator controls for this possibility so it is not a concern here. 84 supply does not, however, reflect a large decline in labor supply for program beneficiaries, which dropped about 8 hours per week in 2001 but only 2 hours a week in 2002, but rather reflects the substantial increase inhours worked by their coffee region counterparts who are not beneficiaries. In the absence o f the program, then, beneficiary households in coffee producing regions would have had to devote substantially more time to work (and at the same time would have suffered declines inper capita expenditures) . Ina separate section of the questionnaire, households report on remittances received over the past year. Inregressions similar to those in Table 10 (but not shown), where the dependent variable i s an indicator o f whether a household received remittances in the past year (or the amount o f remittances received), Ifind that RPS had a negative effect on the probability a household receivedremittances in 2001-but only in non-coffee growing areas. In coffee-growing areas, which underwent a more severe decline over the period, the program had no significant effect on receipt o fremittances. Despite coming with conditionality that may substantially increase private costs to household^,^^ RPS transfers likely relax beneficiaries' budget constraints allowing them to re-optimize and thereby improve both their current and future situations. Given the long-term downward trend in coffee prices, one would be hard pressed to argue for entry in the industry and we might expect to see exit (if,indeedit has not already begun) over the mediumto long term, fixed costs for coffee production notwithstanding. If, for example, households are credit constrained, they may not be able to reallocate their activities immediately, and so remain in coffee. It is possible that the access to additional resources provided by RPS allow this credit constraint to be broken and, in addition to changes in hours worked we would see changes inthe type o f work being carried out. A separate credit constraint pathway via which RPS may work i s posited in Coady, Olinto, and Chldes (2003). Presenting a simple two period model for small coffee farmers, they demonstrate how unconditional transfers can have a direct income effect on labor supply (for all households) but also an indirect effect for credit constrained coffee farmers who, instead o f having to seek off-farm labor are able to devote more time to maintaining their coffee trees, thereby raising the marginal productivity o f their coffee land. That the transfers are conditioned on child attendance at school introduces a third effect, substitution between child and adult labor. If these are the underlying mechanisms, incontrast to the argument inthe previous paragraph, one would see more labor devoted to coffee, rather than less. Inthe baseline 2000, fully three-quarters o f the households (inboth coffee and non-coffee growing areas alike) indicated that they were credit constrained in the sense that either they had requested a loan (from either formal or informal sources) but not received it or that they had not requested a loan but didnot do so because they felt they would not receive it. Because o f the predominance o f credit- constrained households, Ido not report results distinguishing program effects between whether a household was credit constrained or not before the program, noting that any time invariant components o f differences between these types o f households i s already controlled for inthe analysis. When Ido consider to what extent results differ for credit-constrained versus credit-unconstrained households, Ifind that most effects tend to be slightly larger for credit-constrained households, but not significantly so. Inowexaminethetypesofworkhouseholdscarryoutwithandwithouttheprogramtoseeifin addition to changes in total hours, the program induces adjustments along other dimensions of labor 6sCaldes and Maluccio (2004) provide some estimates ofprivate costs for beneficiary women of around 40 hours per year and C$40 in additional transportation costs. 85 supply. The results are presented in Table 11. In the first column, Iassess the impact on total hours dedicated to agriculture in the last week-RPS reduced the total number of hours dedicated to agriculture on average for coffee producing areas, by around 10 hours a week. Nonetheless, despite these large declines, when Iconsider the fraction of labor hours in the household dedicated to agriculture, the RPS effect was negative for households in non-coffee growing areas in 2001, but positive for households in coffee growing areas in 2002. (Clearly, the effect o f RPS on total hours was also negative and larger than that on agricultural hours alone, in coffee growing areas.) The evidence for small business participation i s consistent with these patterns-program beneficiaries in coffee growing cornarcas are less likely to be undertaking small business activities than their counterparts innon-beneficiary cornarcas. Table 11: The effect of Nicaraguan RPS on occupational choice, 2000-2002 Total hours Fraction o f (1) if small (1) if dedicated to labor regular allocated to business small last week agriculture agriculture week activity last business last week activity Year 2001 1.3119 0.04389 *** -0.2113 *** 0.0515 *** (0.58) (2.60) (7.62) (2.60) Year 2002 4.3686 ** 0.1026 *** -0.2010 *** -0.0309 (1.95) (6.07) (7.25) (1.56) Year 2001 x coffee 2.1848 -0.0245 0.1352 *** -0.0085 (0.64) (0.96) (3.22) (0.28) Year 2002 x coffee 7.8036 ** -0,1194 *** 0.1348 *** 0.0674 ** (2.30) (4.69) (3.21) (2.25) Year 2001 x RPS area -2.1229 -0.0053 0.1011 *** -0.0568 ** (0.67) -0.0633 *** (0.22) (2.57) (2.02) Year 2002 x RPS area -4.0562 0.0619 0.0021 (1.27) (2.63) (1.57) (0.07) Year 2001 x coffee x .11.5277 ** -0.0297 -0.1665 *** -0.0663 RPS area (2.42) (0.82) (2.83) (1.58) Year 2002 x coffee x -9.6113 ** 0.1208 *** -0.0972 * -0.1339 *** RPS area (2.02) (3.36) (1.65) (3.18) Constant 65.1655 *** 0.8094 *** 0.1841 *** 0.1218 *** (77.90) (128.28) (17.80) (16.47) F-test Year 2001 + Year 14.93 *** 1.70 2.24 15.53 *** 2001 x coffee [<0.01] [0.19] [0.13] [<0.01] Joint test year 2002 + 14.97 *** 4.66 ** 0.65 17.82*** Year 2002 x coffee [<0.01] [0.03] [0.42] [<0.01] F-test overall regression 6.07 *** 7.16 *** 16.42 5.83 *** [<0.01] [ chi2 =O.OOOO Log likelihood= -284.80475 PseudoR2 = 0.1892 COFFEE97 Coef. Std. Err. z P>z [95%Conf. Interval1 intensity 3.552645 ,3299456 10.77 0.000 2.905964 4.199327 -cons -1.107939 .0774568 -14.30 0.000 -1.259751 -.9561261 The variable "intensity" (=area in square kilometers of coffee plantation over area in square kilometers of the municipality) i s a good predictor of the probability of a household being involved in coffee activities. We graph the predicted probability o f working in the coffee sector against intensity inFigure3. Both, the significant coefficient o f the coffee-area-intensity variable and the graph show that this variable will be a good instrument inregression analysis. Also, taking the subcategory o f households that are coffee growers the coffee area intensityindex i s a good predictor of the probability of a household participating as coffee grower in the economy. Giventhat this subcategory i s available independentlyfor each of the four years of the panel, predicted probabilities could be estimated for each year. Figure 4 shows that the probability that a household participating as coffee grower increases with the coffee area intensity index, but it also shows that the probabilities are lower for 1997 and 2001. Value Added of the coffee sector in El Salvador was smaller in 1997 and 2001, compared to 1995 and 1999 (see Table No. 1 in the statistical annex), suggesting that the survey i s able to capture these 72The sample size in this regression includes only those households for which we were able to identify the municipality. 105 movements in coffee production in El Salvador, and that there i s a tendency to move away from this sector. Figure3: PredictedProbabilityof a HouseholdInvolvedinCoffeeActivities as aFunction I I I I I I 0 .2 .4 .6 .8 1 intensity 106 Figure 4: PredictedProbability of a Household Involvedas Coffee Grower as a Functionof CoffeeArea Intensity, 1995,1997,1999 and 2001. To see the movements in and out o f coffee production for the category o f coffee growers it is convenient to construct a transition probability matrix. The transitionprobability matrix describes the change o f a categorical variable over time. A symmetric diagonal matrix with values equal to 100 inthe diagonal indicates that households do not change categories over time; on the contrary the smaller values o f the diagonal indicate more changes o f categories over time. Table 7 shows that from households in the panel data set, 5.76% participated as coffee growers on average over the four years covered by the survey (1995, 1997, 1999, and 2001). For coffee growers the probability o f moving out o f this activity i s about one third. On the other hand, from those households that did not participate on coffee activities, there i s only a two-percent probability o f moving into the sector. This supports the idea that categorical mobility i s very high for coffee growers. HouseholdType Others Coffee Growers Total Others 97.96 2.04 100.00 Coffee Growers 33.33 66.67 100.00 Total 94.24 5.76 100.00 107 Coffee intensity and alternative measures of welfare Before we continue with the analysis o f the BASIS/FUSADES data set it i s worth mentioning two sources o f information o f social indicators that are available and that might be helpful to assess the effects o f the coffee crisis in El Salvador. These indicators are malnutrition and the percentage o f stunted children. These measures o f health and nutrition are widely use inempirical research as outcomes that result from decision o f households facing prices and by resources that they hold(see Behrman and Deolalikar 1988). PERCENTAGESTUNTEDVERSUS COFFEEINTENSITYINMUNICIPALITIES One source o f information about nutritional status i s stunting (height to age) o f children at the municipal level. The source o f this information i s the school census o f 2000 fielded by the Ministry of Education (MINED). The next graph presents plots of the height-for-age data and coffee area intensity at the municipal level for all the country and for selected departamentos. At glance, with the full sample it i s impossible to claimthat there i s any correlation between stunting and coffee areas, suggesting that the spread o f nutritional deficiencies inthe school population i s explained by other factors. The simple correlation between stunting and coffee intensity i s equal to 0.1340. One problem with the data, i s that it does not differentiate between urban and rural areas. To dig a little bit more into the data, similar plots are presented for those departments where coffee area intensity i s higher, Ahuchapin, Santa Ana, Sonsonate, and Usulutan. In this cases there i s a positive correlation between coffee are intensity and the percentage o f children suffering from malnutrition, meaning that high intensity coffee areas suffer from high poverty rates. However, this outcome does not implythat the coffee crisis i s having an impact on the nutritional status o f children, because it i s looking only at a cross section o f municipalities. W e are interested in the changes in nutritional status from one year to another, with the hypothesis that those households living in areas where coffee activities are predominant might suffer more the effects or face more difficulties to satisfy their basic needs. Also, these thoughts send a signal o f warning to those that think that aid should go directly to areas where coffee i s grown more intensively. W e need to look for other factors that are also important in determining which kinds o f households would benefit the most with aid programs. 108 Figure 5: Height-for-Age Ratios and Coffee Area Intensity by Municipalities. Retardo en Talla e lrstensidad porArea de Cafe Todos 10s departamntos Ahuachapan Santa Ana 0lntensidad4por areade.8caf61 .2 .6 Source: MINED andPROCAFE. CHANGEINMALNUTRITIONBETWEEN1998AND 2001 Usingfigures publishedby FESAL (Encuesta Nacional de Nutrici6n Familiar) for the country as a whole the percentage o f underweight children fell from 11.8% in 1998 to 10.3% in 2002/2003. However, looking at the departmental level and crossing the information with the intensity o f coffee areas by departments, those departments with higher coffee intensity have a trend to show a smaller reduction, or even an increase in the percentage o f underweight children. InFigure 6, the horizontal axis is the coffee area intensity index, and the vertical axis represents the change in the percentage o f underweight children under 5 years old using data from FESAL 1998 and 2002/2003. Clearly the department with the highest coffee area intensity index, Ahuachapan, has increased the percentage o f underweight children. And the other three departments with high coffee area intensity index, Santa Ana, L a Libertad and Sonsonate, have decreased the percentage o f underweight children, but by a smaller amounts that departments with low coffee area intensity index, such as Usulutan, Morazan, L a Paz and L a Union. 109 Figure 6: Change inthe percentageof underweightchildrenbetween1998 and 2002 by departmentsand coffee area intensity. a - 0AHUACHAPAN w z Q, C a, OSAN SALVADOR 0 SAN VlCf&!!? 0 SANTA ANA 0 CABAWS OCUSCATLAN '' -0 C 0CHALATENANGO *LA PA2 0 - 0 LA UNION s .- C Q) w 0 MORAZAN m C c 0 s- 0 USULUTAN I I I I I Although the graph suggests that there i s a positive correlation between malnutrition and coffee area intensity, it i s necessary to control for different factors to be able to establish if the coffee crisis i s related to this underperformance o f coffee intensive department^.'^ Given the small sample size--14 departamentos- the degrees o f freedom for regression analysis i s too small and in several specifications none o f the coefficients are significant.74 Table 8 presents simple correlation coefficients for nutritional change, coffee intensity, rural female literacy rates and the change inreal per capita household income between 1998 and 2002. The strongest correlation i s between nutritional change and coffee intensity (0.42). The correlation between female literacy rates and the change in nutritional status i s positive but smaller (0.17), with an unexpected sign, since higher literacy rates are positively correlated with an increase in the percentage o f underweight children. The change in real per capita household income and the change in percentage o f underweight children are negatively correlated (-0.13); thus, increases in real per capita income are associated with improvements in nutrition o f children. Overall the strongest association i s between nutritional change and coffee intensity. 73At the end o f the statistical annex we present a graph associating the change inmalnutrition with the percentage o f coffee-worker households in the BASIS 1997 data set, and the conclusion are similar. 74Simple and multiple regressions were performed for the change in the percentage o f underweight children against coffee intensity, rural female literacy rates in 1998, and change in real per capita household income at the departamental level. 110 ' Table 8: Simple correlationmatrix of changein percentage of underweight children and potencialcorrelates. Percentage Change in % of change in real underweight Female literacy percapita children 1998- Coffee intensity rates in rural household 2002 areas in 1998 income, 1998- I 2002 Change in % of underweight children 1998- 1 2002 Coffee intensity 0.4212 Female literacy rates in rural 0.1725 areas in 1998 Percentage change in real percapita household -0.1286 -0.1883 -0.0052 1 income, 1998- 2002 ;htchildren 1998-2002), PROCAFE :offee intensity), and EHPM 1998 and 2002 @emale literacy rates inrural areas in 1998 and percentage change inrealper capita householdincome, 1998-2002). Comparative analysis of socio-economic characteristics between coffee and non-coffee households. The classification o f households as non-coffee, coffee workers and coffee growers i s helpful to identify some socio-economic differences, and this distinction i s critical to the story that unfolds below. As many o f the statistics show, forming a group o f coffee-related households may be misleading, as growers and workers are very different group. 111 DEMOGRAPHICDIFFERENCES Indemographics, non-coffee households have approximately 6 members per family, while coffee workers have more than that and coffee growers have fewer." The difference inhouseholdsize i s statistically significant between coffee workers and coffee growers. The average age o f the head o f the household i s similar between non-coffee households and coffee workers (the difference i s not statistically different from zero). But it i s clear that for coffee growers the head o f the household i s more than 10 years older than the other type of households which might explain why they have fewer children, and hence, fewer members (the difference i s statistically significant at 1% significance level). As expected, the average age o f the head o f the household also increases over time for all type o f households, showing that the panel data captures the life cycle o f these households as they get older. The percentage o f households where the head i s female has increased for non-coffee households from 7 percent in 1997 to 18 percent in 2001, it remained around 10 percent for coffee workers, and it increased from 8 to 19 percent for coffee growers. Therefore among coffee workers there i s a predominance o f males as head o f households, while in the other two groups, women's participation as head of the family has increased over time. This difference is statistically significantbetweencoffee workers andnon-coffee households only in2001, due the increase inwomen's participation as headofhousehold over the periodfor non- coffee workers. DIFFERENCESINEDUCATION The average o f years o f schooling o f the head o f household i s smaller for coffee workers; around 2.5 years o f studies for coffee workers and 3 for non-coffee households (the difference i s statistically significant only in 2001). Between coffee growers and non-coffee households the difference inyears o f schooling o f the head o f households i s not statistically different from zero; but between coffee growers and coffee workers it is statistically different from zero (in 1995 and 2001) in favor o f coffee growers. In terms of stocks and flows of human capital accumulation in education coffee growers outperform non-coffee households, and non-coffee households outperform coffee growers. The average years o f schooling for adults are higher for coffee growers than for non-coffee households and coffee workers (the differences are statistically different from zero in 2001). In terms o f investment in education, the differences between non-coffee and coffee workers are not statistically different from zero (the percentage o f children ages 6-12 enrolled in school); but coffee growers have significantly a higher percentage o f enrolled children between ages 6 and 12, than the other two types o f households. However, for the percentage o f children between 13 and 18 years o f age enrolled in schools, there are no differences (statistically) among all types o f households, suggesting that opportunities or access to education i s more difficult at the secondary level for all types o f household. Also, investment in education improves over time, with increasing attendance ratios for all types o f households. ''Thissection is based on Tables 8.1-8.2 and Tables 9.1-9.3 in the statistical annex. 112 DIFFERENCESINLABORPORTFOLIOS As a first approximation, labor time has been divided into two categories, wage labor, and self- employed labor, referring to income generating activities. Households with coffee workers participate more as wage laborers than non-coffee households and coffee growers, both in absolute and in relative terms, and their portfolio o f time allocation i s less diversified. Table 9 shows that households with coffee workers have a larger share o f their time devoted to wage labor, around seventy percent, and that the average o f this share i s statistically different between coffee workers and the other two types o f households. These suggests that both, coffee growers and non-coffee growers may find other income generating activities as self employed, while for coffee growers it i s more difficult. In addition, the share o f time devoted to wage labor has increased for coffee growers, but has decreased for non-coffee households between 1995 and 1997, and then remained at around forty percent. For households with coffee workers the share of time devoted to wage labor has also decreased over time, but still remaing higher that the other to types o fhouseholds. Table 9: Average of share of working time devoted to wage labor at household level. * Significant at 1% level. A second approximation is to divide productive time between agricultural and non agircultural activities. Table 10 shows that in 1995 and 1997 households with coffee workers spent less time innon-agricultural activities than non-coffee households, butbeginning in 1999 these differences are no longer statistically different. Also, for all types o f households there i s a tendency to increase the share o f time devoted to non-agricultural activities, and coffee households have greater changes. This coincides with aggregate data that shows the declining importance o f agriculture in GDP. 113 Table 10: Average of share of working time devotedto no agricultural activities at householdlevel. * Significant at 1% level. Wage labor between 1995 and 2001. The evolution o f wage labor over time shows that an important structural change i s undergoingin rural El Salvador. The number o f households with at least one member engaged in wage labor outside the house decreased from 375 to 319 in 1997, and afterwards it increased on every year in the sample, but it i s still below the 1995 level. However, this evolution differs by type o f household. For instance, for coffee growers the number o f households egaged in wage labor increased between 1995 and 2001, for coffe growers it did not changed, while for non coffee households there was a sharp decline in 1997, and an increase in 1999 and 2001, but staying well below the 1995 level (See Table 10 instatistical annex). By looking at sector-specific changes it is evident that a structural change is under way in the country, the number o f households involved with wage labor in the agricultural sector have declined every year; however for coffee households as workers and growers wage labor participation in agriculture increased between 1995 and 1997, but after that year it decreased considerably, while for non-coffee households occurred the opposite, their wage labor participation inagriculture declined in 1997, increased in 1999, and stay at the same 1999-level in 2001. On the other hand, as opposed to wage labor participation in agricultural activities, participation in non agricultural sectors increased for both coffee and non-coffee households, suggesting that households are moving away from the agricultural sector and increasing their wage labor participationin other sectors. Coffee households have increased considerable their participation in non-agricultural sector, especially in 1999 and 2001, suggesting that reallocation o f time could be an important coping mechanism for these households (see Table 10 instatistical annex). The average number of hours devoted to wage labor i s considerably higher for households with coffee workers only, compared to other types o f households (t-test for the equality o f average are statistically significant at 1% level; see Table 11inthe statistical annex), suggesting that they are more dependent o f wage labor. For households with coffee workers, the average number o f hours devoted to wage labor declined in 1997 and gradually recovered to the 1995 level by 2001; mainly as a result o f an increase inthe average number o f hours inthe non agricultural sector that offset the decline of the number o f hours inagriculture. 114 Furthermore, in 1995 the average number wage-labor hours spent in agriculture by households with coffee workers was greater than the average number o f hours spent in non-agricultural sector, and for coffee growers it was about the same; but by the end o f the period for both types o f coffee households the average number o f hours spent in non-agricultural sector i s greater than inthe agricultural sector. This shows that households are allocatingmore time to non-agricultural wage labor and that there are more labor opportunities in this other sectors, as a result o f the structural change inthe economy. The total number o f hours devoted to wage labor supports a similar idea. Ingeneral, total time devoted to wage labor decline in 1997, starts recovering in 1999, and reaches in 2001 a similar level to that o f 1995. However, they are decreasing wage-labor time spent inagriculture along the whole period, while for non agricultural sector, there was a decline in 1997, and a rapid recovery in 1999 and 2001, reaching muchhigher levels. Coffee households devoted in2001 63.9% more hours to wage labor in the non-agricultural sector than in 1995, while for non-coffee households there was a 14.75% increase between these two years, showing that the change has been bigger for coffee households. Self-employed labor between 1995 and 2001. The number o f households participating in income-generating activities as self employed, such as microenterprises increased steadily over time; both in agriculture and non-agricultural activities, suggesting that rural households are diversifying their time portfolios. This increase i s sharper for households with coffee workers and non-coffee households. The increase in the number o f households as self employed between 1995 and 2001 i s about 100 additional households; both in agricultural and non agricultural activities (see Table 10 inthe statistical annex). It i s important to notice that households as coffee growers and non-coffee households participate more than households with coffee workers as selfemployed in non-agricultural activities, meaning that the former are more diversified intheir activities. From the point o f view o f total number o f hours as self employed, overall there i s a steady increase from 1995 to 1999, and then it remained stable in 2001. But this trend differs by type o f household and activity. For coffee growers there i s a decline inthe total number o f hours devoted to self employment, with a sharp decline in 2001; this decline occurs mainly in the agricultural sector, as expected from the decline in the sector, in part due to the drop in coffee prices. For coffee workers the total number o f hours as self employed increased between 1995 and 1999, but it declined in 2001; these movements come from both, agricultural and non-agricultural sectors. For non-coffee households, it increased considerably between 1995 and 1997, and then at slower rates in 1999 and 2001. Between 1995 and 1997 there was an increase in self employment in agriculture for non-coffee households, it remain steady in 1999, but it decrease in2001; while for non-agricultural activities there was a sharp increase between 1995 and 1997, further increase in 1997 and 1999, and an important increase in2001(see Table 10 inthe statistical annex). Ina similar trend, the average number of hours devoted to selfemployment increased steadily from 1995 to 1999, and it remain stable in 2001. It i s evident that there i s a very sharp and steady increase in the average number o f hours as selfemployed in non-agricultural activities. On the 115 other hand, alfthough the average number o f hours as self-employed in agricultural activities i s higher, there was an increase between 1995 and 1997, remaining at the 1997 level in 1999, and then in decreased in 2001, but it was still higher than in 1995. By type o f households, there i s a considerable decline in the average number o f hours as self-employed in agriculture for coffee growers, while the average hours as self-employed in non-agricultural activities increased in every year o f the survey. It i s important to notice that the average number o f hours as self employed in agriculture was higher and statistically significant for coffee growers, compared to households with coffee workers in 1995 and 1997andto non-coffee households in 1995, but the differences were not statistically signficant afterwards (see Table 12 inthe statistical annex). The other important statistically significant difference i s that the average number o f hours as self- employed i s higher for non-coffee households as compared to households with coffee workers in all years. INCOME DIFFERENCESBETWEEN1995 AND 2001. Total per capita income differs among different types o f households and in different years. Coffee workers have smaller per capita income (and per capita income net o f remittances) than non- coffee households (the difference i s significantly different from zero at 1% level o f significance) in 1997 and 2001, but not in 1995 and 1999. Coffee growers had significantly higher per capita household income than non-coffee households in 1995 and 1999 only, suggesting that in bad agricultural years their incomes are similar. In all years total per capita income i s higher for coffee growers than for coffee workers (the differences are significantly different from zero) suggesting that there i s a clear difference i s socioeconomic status between these two groups. i s much higher for coffee growers than for any other type o fhousehold. The sources o f income differ among types o f households. First, remittances have a positive trend for all types o f ho0useholds. From 1997 onwards non-coffee households receive more remittances than coffee workers (the difference i s statistically significant), suggesting that the former group i s more likely to be more diversified in their sources o f income. Between non- coffee households and coffee growers remittances receipts are not statistically different, nor betweencoffee workers and coffee growers. The mean o f agricultural income has different trends among the different types o f households. For non-coffee households it increases between each year, for coffee workers it increases from 1995 to 1999, but it falls dramatically in 2001, when coffee prices reached the lowest level, and for coffee growers it decreased between each year, with the sharpest decline inpercentage terms in 2001. It is remarkable that in 1995 and 1997, both, coffee workers and coffee growers had statistically different (and higher) agricultural income that non-coffee households, butnot in 1999 and 2001, reflecting the decline in value added in the coffee sector. Furthermore, for coffee growers, agricultural income i s even smaller than for non-coffee households and coffee workers in2001, reflecting the fact the coffee growers depend more on this crop, and that their decline in agricultural income i s therefore higher, due to the drop incoffee prices. Third, non-agricultural income shows a positive trend in all types o f households. While in 1995 and 1997 non-coffee households received higher non-agricultural income than coffee workers (the differences are statistically different from zero), in 1999 and 2001 the differences in non- 116 agricultural income are not statistically different from zero. This suggests that there i s a strong shift to non-agricultural income among coffee workers. Between non-coffee households and coffee growers, and between coffee workers and coffee growers, the differences are not statistically different from zero inany year. Fourth, there i s a clear patternthe source o f income generated inside and outside the house differs between non-coffee households and coffee workers in 1997, 1999, and 2001. Clearly, coffee workers depend more on labor income outside the house-wage income-than non-coffee households; income outside the house i s higher (differences are statistically different from zero) for the former. On the other hand, income inside the house i s higher (differences are statistically different from zero) for non-coffee households than for coffee workers. However, the differences in income inside and outside the house between coffee growers and non-coffee households are not statistically different from zero. Income outside the house i s higher for coffee workers than for coffee growers (the difference i s statistically different from zero in 1997 and 2001). Finally, income inside the house i s greater for coffee growers than for coffee workers (the differences are statistically different from zero in 1995, 1997, and 1999, but not in2002 suggesting that the drop incoffeepriceshada greater effect oncoffee growers). Also, non-coffee households receive about fifty percent o f their income from labor outside the house, and fifty percent from labor inside the house-or entrepreneurial activities- while coffee workers rely more on income outside the house-or wage income-on a 80:20 ratio with respect to labor inside the house. Coffee growers have higher income diversification from this point o f view, than coffee workers, since they obtained 54 percent o f their income outside the house and 46 percent inside the house by 2001. However, in 1995 coffee growers earned 65 percent o f their income inside the house and 35 percent outside. Evolution of income between 1995 and 2001 by type of household. This section describes the evolution o f per capita household income over time, compares it with coffee and non-coffee households, and brakes the latter into coffee growers and non growers, to see if those households involved in coffee sector as laborers only have a different income pattern over time and across households. Inthis section the classification between coffee and non-coffee households corresponds to 1997 only, that is, we observe the evolution o f income for those households that were involved incoffee activities in 1997. The first observation i s that overall the data show a positive trend o f average per capita household income, but there are differences by regions and type o f ho~seholds.'~The second observation i s that there are significant income differences among non-coffee households, households involved in coffee production as laborers only and coffee growers. Coffee growers have considerably higher incomes than other type o f households in any o f the four years, while coffee households as 76This trend corresponds to the sample in the panel, and therefore does not correspond to the evolution of averagehousehold income at national level. At national level, data from the Multipurpose Household Survey (EHPM) of DIGESTYC, and general macroeconomic indicators show that household income has increased very little over the period. Meanwhile income from this sample has increased.One plausible explanationi s that becausethe BASISRUSADESsurvey keeps track of the evolution of the same households it includes the life cycle movements of income of a typical household, which implies that at certain level of education and experience, income goes up for young households. See Table 8 in statistical annex and the section describing socioeconomic characteristics of the households. 117 coffee workers only, have considerably lower income than non-coffee households and even lower than coffee growers. Therefore our classification on households based on participation in coffee activities highlight important differences inhouseholdincomes (see Table 11). A third observation is that while non-coffee households present a positive trend in per capita income over the entire period, coffee households have more variation in their in per capita income. Coffee households as workers only show a decline in income between 1995 and 1997 o f 10.2 percent, and between 1999 and 2001 o f 5.3 percent. Similarly, those that were coffee growers in 1997 experience a decline in income in the same periods, but in the 2001 shock the decline i s greater suggesting that they have a different pattern in the evolution on income compared to other coffee households. It i s not a coincidence that aggregate production data inthe coffee sector show a similar trend; the growth rate o f GPD in the coffee sector was -5.1 percent bOetween 1995 and 1997, 5.7 percent between 1997 and 1999, and -26.4 percent between 1999 and 2001 (see Table 1inAnnex). Source: BASIS surveys, 1995, 1997, 1999,2001. The sample usedis for the 1995/1997/1999/2001panel. A fourth observation is that average household per capita income increases over time inall four regions considered in this study. Similarly, if the look at regional patterns by different type o f households it i s not possible to identify different patterns by regions that could be related to coffee shocks. For instance, for coffee growers income fell in all four regions between 1995 and 1997, and increased in all o f them between 1997 and 1999, except in the west region i s the most coffee intensive region (see Table 10 in Annex). For coffee households as workers only income fell between 1995 and 1997 in all regions, except the West region which i s the most coffee intensive region o f the country. Inthat sense, if we want to identify the effects of the coffee crisis inrural households it would be better to stick to the classificationof coffee households based in the 1997 survey. The data on per capita income provide a first approximation of household's welfare during the period, but it includes information on subsidies and family remittances from relatives leaving abroad. The variable therefore does not reflect household's ability to generate income from productive activities, and i s not a good measure to try to capture the effects o f the coffee crisis on households in the panel, although it's a better approximation to household's welfare. Alternatively, Table 12 presents information on per capita net income (subtracting remittances and other subsidies) classifying households by type. In this case, both coffee households and coffee growers present a decrease in income between 1995 and 1997, and between 1999 and 2001; reflecting that 1997 and 2001 were both bad years for coffee production inEl Salvador and that it affected negatively on average both types o f households. Also the difference between per capita income and net income i s particularly large in 2001, suggesting that remittances play an 118 important role in this year after the shock o f earthquakes and the coffee crisis. In fact, for coffee households as workers only the decline o f net percapita income was 11.2 percent, but after remittances and other subsidies it was 5.3 percent, meaning that remittances and other type o f subsidies ameliorated the negative effects that the coffee crisis and earthquakes might have produced in these households. Similarly, for coffee growers net per capita income declined by 38.2 percent in the same period, but after including remittances and subsidies per capita income decreased by 20.8 percent. Inabsolute terms the average decline inincome was o f 3,203 colones, while average remittances were 1894 colones for coffee growers. Source: BASIS surveys, 1995, 1997, 1999,2001. The sample usedis for the 1995/1997/1999/2001panel. Exploring in more detail the possibility that coffee growers may have different coping mechanisms we can look at differences in changes in per capita income before and after remittances in the two periods where there are negative income shocks. In the first period, between 1995 and 1997 the negative income shock must have come from a weather related shock since in 1997 the average price for coffee is the highest. In fact there i s evidence that in 1997 El Salvador was affected by El Niiio, and it has effects on temperatures and rainfall.77The change inper capita net income was negative for both coffee growers and coffee households as workers only, and it was even larger for coffee growers. After including remittances as part o f income, the change in income was also negative for both types o f households, but the negative change were even higher than net per capita income inboth types of households. The reason i s that for both types the average o f remittances received fell between 1995 and 1997. However, for coffee workers average remittances in 1997 were larger than the negative income shock, 126.7 colones versus 103.2. But, for coffee growers average remittances were smaller than the average net income shock, 206.6 colones versus 263.7, suggesting that remittances, even though increase household's income, were not big enough to offset the negative effect o f the weather shock. On the other hand, between 1999 and 2001 the negative income shock i s related to two different types o f shocks, the dramatic fall in coffee prices and the two huge earthquakes that hit the country in January and February 2001. In this case, net per capita income fell for both types o f coffee households, but it increased for non-coffee households. The difference with the previous shock i s that after including remittances the negative change in per capita income for coffee households as workers only decreased from 11.2 to 5.3 percent, implyingthat remittances were able to offset the negative effects o f both shocks inincome (remittances increased with respect to 1999). The change in average net income amounted to a decline o f 492 colones, while "Forinstance,Tables1and5inthestatisticalannexshowthatGDPandthevolumeofvolumeofproductioninthecoffeein1997 and 2001 were smaller than in 1995 and 1999. Also, Table 7 inthe statistical annex shows that yields per manzana and total production of coffee were smaller in the 97/98 and 01/02 agriculturalyears. See also Angel (1998). 119 remittances in 2001 amounted to 733 colones. As was mentioned before, for coffee growers remittances were not able to offset the negative effect on income; the change in average net income amounted to 3,203 colones, and average remittances were 1894.3 colones. Hence, the size o f remittances for coffee households i s muchlarger, maybe because coffee growers own land, and therefore, with secure property rights they had the right incentives for reconstruction o f their homes and therefore received and important influx o f remittances for this purpose. On the other hand, the other type o f households i s less likely to own land and without adequate property rights they do not receive the necessary aid for home reconstruction. Remittances for coffee growers are almost four times larger than o f coffee households as workers only. INCOME MOBILITY FOR COFFEEAND NON-COFFEEHOUSEHOLDS The economic position o f households may change for a variety o f reasons, such as GDP growth, sector specific shocks, weather shocks, or other events. With the panel data set it i s possible to study the movement o f households through the distribution o f per capita income over time, in order to establish how dependent their current position i s from a past position, and how this relates to different circumstances, such as their participation in the coffee sector. To achieve this we use a transition probability matrix o f income quintiles. Income quintiles give the position o f each household in the income ladder in any particular year, so that a specific householdmightbe in a different quintile in every period, or may stay in the same one. The entries in a transition probability matrix indicate what fraction o f individuals starting in a particular quintile ends up in another quintile in some other period, hence each row sums to one hundred percent. The percentage probability i s calculated based on the information o f the position o f the household in each o f the four years. The transition probability matrix for all households shows that households in the lower two quintiles are very likely to remain in low income quintiles, and only about 10 percent o f these households manage to move to the upper quintile. The same occur for households in the higher quintile, they tend to remain in the highest quintile-about 50 percent o f these households--, and less that 10 percent fall to the lowest quintile. This proves that households move up and down in the income distribution over time. Are different types o f households more or less mobile? Coffee growers in general appear to have more presence in the upper quintiles o f the distribution, which i s consistent with the fact that on average they enjoy a higher income. However, regarding mobility they show a pattern o f upward and downward movements in the income distribution, except for those at the top quintile which have a 60 percent chance to remain in that quintile. Coffee Households as workers only are concentrated towards the center o f the distribution since over time 26.1 percent are in the middle, and less than 20 percent are in each o f the lowest and highest quintiles. Income mobility for this type o f households i s also high.Non-coffee households on the other hand have less participation in the middle quintile and more participation in the lowest and highest quintile, and they also show some mobility, at it i s smaller in the two extremes. Therefore, non-coffee households show less mobility than coffee households and they tend stay more time in the richest or lowest quintiles. On the other hand, coffee households show more mobility over time whether they are coffee growers or coffee workers only; but coffee growers are more concentrated in the upper quintiles, while households as coffee workers are concentrated inthe middle quintiles. 120 Table 13: TransitionsProbability Matrix of Per Capita Income Quintiles for Different Type of Households. All Households 30.7 26.9 22.4 11.o 18.2 20.6 24.9 24.5 11.9 10.7 13.7 21.o 32.1 22.5 5 9.0 5.6 12.3 24.3 48.9 Total 21.5 18.9 19.0 20.3 20.4 Coffee Households Quintiles I 1 2 3 4 5 1 25.4 34.9 20.6 15.9 3.2 2 28.0 28.0 22.7 14.7 6.7 3 9.3 23.3 33.7 25.6 8.1 4 12.5 12.5 31.8 25.0 18.2 5 8.1 6.5 12.9 27.4 45.2 Total 16.3 20.9 25.4 21.9 15.5 Quintiles 1 2 3 4 5 1 28.3 37.7 20.8 11.3 1.9 2 24.6 29.0 24.6 15.9 5.8 3 8.2 24.7 35.6 23.3 8.2 4 15.7 14.3 27.1 25.7 17.1 5 9.7 12.9 16.1 29.0 32.3 Total 17.6 24.3 26.4 20.6 11.2 Quintiles 1 2 3 4 5 1 10.0 20.0 20.0 40.0 10.0 2 66.7 16.7 0.0 0.0 16.7 3 15.4 15.4 23.1 38.5 7.7 4 0.0 5.6 50.0 22.2 22.2 51 6.5 0.0 9.7 25.8 58.1 'Total I 11.5 7.7 21.8 26.9 32.1 31.8 26.5 22.2 9.5 10.1 22.8 19.2 20.4 24.0 13.8 14.2 15.9 35.5 24.6 9.2 5.3 12.1 23.3 50.0 Total I 23.5 18.1 16.5 19.7 22.2 121 Evolution of school attendance School attendance has increased for households in the sample, especially at preschool and primary education levels, a ten percent point increase at the secondary level, but with a small decline at higher educational levels. There are differences among types o f households. With the broader classification o f coffee and non-coffee households there are some differences. For instance coffee households have a lower attendance rate at the preschool and higher educational levels, while at primary and secondary education the differences between the two groups as smaller. At primary levels o f education, between 7 and 12 years o f age, school attendance i s very similar to coffee and non-coffee households, with a positive trend over the period. Although coffee households begin with lower attendance ratios that non-coffee households, it seems that the differences tend to narrow over time. A further breakdown o f coffee households between coffee grower and households as coffee workers only shows that the differences among groups are wider. For instance, the attendance ratio for households as coffee workers and preschool level was less than have o f that o f non-coffee households; however over time this gap disappeared. At the higher educational level the gap between households as coffee workers and non-coffee households remain over time. On the other hand, coffee growers have clearly higher attendance ratio than any other group suggesting that they enjoy a higher level o f social and economic welfare, which i s consistent with their higher incomes (see Table 14). The positive trend o f school attendance at the lower level o f the education system, and the fact that coffee households did not experience a deterioration of school attendance ratio during the coffee and earthquakes shocks could be a result o f public policy in El Salvador that has given priority to education expenditure inrural areas over the past decade. 122 100.0 13-15 16-18 32.9 33.3 30.0 38.8 19-25 15.7 11.9 21.7 14.5 Total 53.4 47.5 46.4 52.8 51.7 ITotal I 58.71 54.01 51-31 71.71 57.41 Source: BASIS surveys, 1995, 1997, 1999,2001. The sample usedi s for the 1995/1997/19991`2001panel. A breakdown by regions shows similar conclusions. The West region, the region where coffee plays a greater role compared to other regions, also has a lower proportion o f children going to school in 1995, but over time it caught up with other regions (see Table 11 in Annex). Based on this breakdown is not possible to claim that the coffee crisis had a negative impact on school attendance, on the contrary, school attendance has improved in those places where coffee activities predominate. Even though school attendance has improved over time, due to particular educational policies that favored rural areas over the past decade, with these observations alone we cannot claim that the coffee crisis did impair some o f the positive effects o f educational policy. What these results suggest i s that these types o f policies are in general positive and can ameliorate or more than offset the negative impact coming fkom economic shocks. 123 PROBABILITYOF SCHOOL ATTENDANCE AND COFFEE VARIABLES The decision to send a child to school depends on the incentives that households face to send their children to school and the opportunity sets that they have. We based this section on a model that states that the decision o f sending a child to school i s a function o f individual characteristics, such as sex and age; and demographic and socioeconomic characteristics o f the household, such as the number o f siblings o f different ages, the number o f adults and of elderly, education o f the father and the mother, whether the father or the mother are missing, per capita household income net o f remittances and o f child wages, per capita remittances, land ownership, and distance to school. After controlling for these factors it is possible to examine whether the links to a particular sector have a separate effect on the probability o f attending school. The advantage o f this approach i s that if there are some specific effects o f the coffee crisis on the probability o f sending children to school, the model will help to identify possible mechanisms through which these effects work. A random-effects probit model o f school attendance for childrenbetween 7 and 18years of age is estimated for all years in the complete panel data set usingthe framework presented inTrigueros (2002) and we added indicator variables o f either the type o fhousehold (coffee versus non-coffee households in 1997, the dummy variable COFFEE97) or the intensity o f coffee activities in the local community (coffee area intensity). Table 15 shows explanatory variables to be used in the estimation. The addition o f the coffee variables help to assess if controlling for all other factors the linkages to coffee activities have some additional effect on investment in education in a particular year. In order to control for other shocks, dummy variables are included in specific years where a shock could be identify in the neighborhood o f a household, such as hurricane Mitch in 1998 and the earthquakes o f 2001. 124 Y MAMAED Years of schooling of the mother. MISSING-FA (=1) if father i s missing. MISSING-MO 1(=1) ifmother i s missing. COFFEEHOUSEHOLDS (=1)control (=1)ifthe household grows coffee in 1997 (=1) ifhousehold's membersworked inthe coffee sector in 1997 but did not prow coffee. PCNINCOME Per capita income net of remittances and childwages in thousands o f colones (US$1=8.75 colones en todos 10s aiios). PCREMITT Per capita remittances inthousands of colones colones. OWNLAND (=1) ifthe household owns land.. DIST-PRIMARY Distance to primary school inkm.Takes the value of zero for childrenolder than 12years. DIST--SECONDARY Distance to primary school inkm.Takes the value of zero for Hurricane Mitch(1998) (=1) ifhousehold lives ina region affected by the hurricane. Earthquakes (2001) (=1) ifhouseholdlives inamunicipality seriously affectedby the earthquakes (more than 25% of homes have some damage) 125 Table 16: LandomEffects I Probit Models ir School Atten tnce. 1995 1997 1999 2001 (1) (2) (3) (4) (5) (6) (7) (8) FEMALE 0.219 0.223 0.000 0.021 0.067 0.088 0.251 0.259 (0.148) (0.148) (0.163) (0.162) (0.165) (0.165) (0.143) (0.142) AGE 1.013'* 1.016" 0.813" 0.828" 0.870'* 0.879*' 0.462** 0.470*' (0.203) (0.203) (0.204) (0.203) (0.212) (0.213) (0.173) (0.173) AGE2 -0.048** -0.048'* -0.043" -0.043" -0.043'* -0.043** -0.025** -0.026"' (0.008) (0.008) (0.008) (0.008) (0.009) (0.009) (0.007) (0.007) NSlBl2 -0.117 -0.119* -0.161" -0.165" -0.104 -0.107 0.059 0.060 (0.060) (0.060) (0.062) (0.062) (0.072) (0.073) (0.051) (0.051) NSIB13-18 -0.106 -0.102 -0.139 -0.140 -0.074 -0.057 -0.111 -0.106 (0.108) (0.107) (0.116) (0.111) (0.114) (0.115) (0.088) (0.087) NMEN19-59 -0.058 -0.037 -0.139 -0.137 -0.008 -0.061 0.164 0.163 (0.137) (0.139) (0.166) (0.166) (0.145) (0.143) (0.105) (0.105) NWOMEN19-59 0.387* 0.368* 0.309 0.328 0.313 0.334* 0.126 0.114 (0.171) (0.174) (0.172) (0.174) (0.169) (0.170) (0.108) (0.107) NMEN-6OP -0.106 -0.022 0.341 0.304 0.002 -0.050 0.183 0.200 (0.306) (0.309) (0.350) (0.350) (0.382) (0.391) (0.253) (0.251) NWOMEN-6OP -0.085 -0.047 0.807 0.880 0.400 0.514 -0.189 -0.187 (0.462) (0.467) (0.593) (0.605) (0.609) (0.614) (0.447) (0.439) PAPAED 0.126*' 0.130" 0.072 0.083 0.038 0.043 0.055 0.055 (0.045) (0.045) (0.053) (0.053) (0.050) (0.049) (0.036) (0.035) MAMAED 0.049 0.039 0.114 0.109 0.181** 0.175** 0.162** 0.164*' (0.051) (0.051) (0.059) (0.061) (0.057) (0.056) (0.040) (0.040) MISSING-FA 0.026 0.088 0.478 0.573 -0.047 -0.141 0.307 0.291 (0.463) (0.470) (0.499) (0.495) (0.415) (0.412) (0.292) (0.289) MISSING-MO 0.415 0.457 1.179 1.476 1.034 0.982 1.171* 1.141' (0.739) (0.745) (1.653) (1.848) (0.858) (0.860) (0.581) (0.573) PCNlNCOMEl 0.119" 0.132* 0.006 0.009 0.008 0.003 0.003 0.003 (0.056) (0.059) (0.041) (0.042) (0.030) (0.030) (0.016) (0.016) PCREMITT 0.559* 0.583' 0.048 0.038 0.080 0.072 0.099* 0.102' (0.230) (0.230) (0,099) (0.099) (0.071) (0.072) (0.046) (0.046) OWNLAND 0.445 0.508* 0.549 0.602 0.543* 0.451 0.277 0.281 (0.231) (0.237) (0.472) (0.467) (0.269) (0.268) (0.180) (0.185) DIST-PRIMARY -0.085 -0.072 -0.066 -0.066 0.094 0.090 0.167 0.161 (0.094) (0.094) (0.099) (0.099) (0.167) (0.167) (0.138) (0.137) DIST-SECONDARY -0.052' -0.053' -0.047 -0.048 -0.083' -0.089' -0.014 -0.016 (0.023) (0.023) (0.032) (0.032) (0.038) (0.038) (0.027) (0.027) intensity -0.811 0.709 1.716* 0.112 (0.604) (0.701) (0.796) (0.503) coffeeworkers -0.329 -0.021 -0.028 0.113 (0.245) (0.281) (0.281) (0.211) coffeegrower -1.094 -0.667 1.684 0.698 (0.761) (0.752) (1.059) (0.676) mitch 0.330 0.271 (0.250) (0.248) terremoto 0.150 0.152 (0.176) (0.174) Constant -4.480** -4.556** -2.518 -2.600* -3.800** -3.568** -1.975 -2.042 (1.210) (1.216) (1.307) (1.305) (1.359) (1.356) (1.131) (1.125) Observations 705 705 722 729 678 683 617 617 Number of hhid 275 275 265 266 255 256 236 236 Rho 0.58 0.58 0.65 0.65 0.61 0.62 0.24 0.23 LR test CHI2 50.87 51.68 74.40 75.86 49.32 50.40 7.03 6.49 Standard errors in parentheses significantat 5%; ** significantat 1% 126 The estimation o f the random effects model for the probability o f school attendance for rural households in the sample shows mixed results among variables and years. The most consistent result i s age, as children grow older the probability o f going to school increases, but around age 10 it starts to decrease, reflecting the attendance ratio at the different educational levels. The number o f siblings inthe household has a negative effect on the probability o f school attendance, but only in the earlier years, suggesting that there was a trade off between the quantity and the quality inthe number o f children, but in 1999 and 2001 the trade o f f has disappeared. The level o f education o f the father andor the mother increases the probability o f school attendance in all years except in 1997, suggesting that there mightbe some intergenerational drag on education for those households with lower education o f parents. Per capita net income had a positive effect on the probability o f school attendance only in 1997; afterwards the coefficients were not statistically different from zero. Per capita remittances also increases the probability o f school attendance but only in 1995 and 2001, and the coefficients differ from those o f per capita net income, suggesting that different sources o f income have a different effect on the probability o f income. Infact, some households receive remittances to support children that were left behind by their parents, and therefore this might be a good reason to expect a positive association between remittances en school attendance. Land ownership had a significant positive effect on school attendance, but only in 1995 and 1999 when agncultural GDP increased as opposed to 1997 and 2001 when it fell. Distance to primary school, reflecting part o f the cost to attend school, does not have an impact on school attendance, but distance to secondary school had a negative impact, meaning the living closer to a secondary school increases the probability o f school attendance. After controlling for all these factors, the model does not show a statistical difference between non-coffee households and coffee households inthe probability o f school attendance o f children. However, when looking at the intensity o f coffee activities at the local level, the variable for coffee area intensity was statistically different from zero in 1999, when coffee value added increased inthe country. This suggests that innormal or badyears those children living inregions with high coffee intensity have the same probability o f assisting to school than children inother areas, but in a good year, like in 1999, those children living in areas with high coffee intensity have a higher probability o f attending school, even after controlling for all other factors. One possible explanation i s that in a good year for coffee production households are better able to support their families without imposing extra work on children, increasingthe opportunities to go to school. Even though we cannot find evidence that the coffee crisis have had some effects on investment on education, if improving the education o f children i s a concern in rural areas special attention must be paid to landless households that are less likely to receive remittances, have parents with lower levels o f education and live farther from secondary schools. ENROLLMENTRATIOSREGRESSIONS To take advantage o f the panel information we pooled together all years and households following the approach in Conning, Olinto and Trigueros (2000), defining attendance ratio o f enrolled children to the total number o f children inthe respective school age category, Si,,where the age categories correspond to primary (ages 6-12) and secondary (13-18) education. W e assume that for each school category there i s a linear relationship between attendance ratio and households characteristics, Tit,shock variables and dummy variables for years, Zi, a household specific random error to capture unobserved characteristicas, vi,and a general random error, uit. 127 Household characteristics included inthe model are household size, whether head o f household is female, education o f the head o f household, per capita income net o f remittances, per capita remittances, size o f land owned. Shock variables includes dummy variables for each year, with 1995 as reference year, variables that link households to coffee activities at regional level (intensity o f coffee area, and an interaction o f coffee intensity with the 2001 dummy) and at household level (dummy variables for type o f household, with non-coffee households as reference type, an interaction term for households as coffee growers and coffee workers with the 2001 dummy to capture the coffee-price shock), and a dummy variable for areas affected by the 2001 earthquakes. The equation was estimated using a random-effects tobit model to take advantage o f the panel information o f the data set. A household with a male as head o f household, with larger size, lower years o f schooling for the head o f household, less per capita income net o f remittances, less remittances, smaller land plots, have a lower chance to send children 6 to 12 years old to school (Table 16). This provides a clear example about the difference between insecurity and vulnerability. Given that in an evironment with perfect markets education is a function o f expected future income, when markets are incomplete, investment in education will be a function o f present income or availability o f assets that may help to cope with an income shock (Jacoby and Skoufias, 1997). In these results, investment ineducation i s a function o f current asset position, and inthat sense, those households with worse asset position (no land or land o f small size) and low income are vulnerable, since they are more likely to keep their children out o f school. On the other hand, households ina better position may suffer insecurity, but are less vulnerable. The dummy variables for each year are also significant, showing that after controlling for these factors, there was an improvement in attendance ratios over time, in part explained by government efforts to improve education coverage in rural areas. On the other hand, the dummies that attempt to control for the coffee price shock and for the earthquakes are not significant, even if we do not control for household characteristics. Only for coffee growers the coefficient i s positive and significantly different from zero. What this means, i s that once we control for household's characteristics the specific indicators for coffee involvement are not useful. Since we h o w that households with coffee workers have a profile that resembles poor households, we can conjecture that many o f these households have been affected by the coffee crisis. But it i s more important to identify them according to their asset position and characteristics (education o f head o f household, family size, land ownership, remittances) rather than to derive conclusions based on their linkage to coffee activities; given the diversity o fhouseholds involved inone activity. For enrollment in secondary education at household level, the key variables that increase the chance o f sending children to school are years o f education o f the head o f the household, per capita remittances, size o f land owned, and distance to the closest secondary school. The conclusion i s similar to primary education enrollment ratios, except that the relevant variables are different. In this case, distance to secondary education seems to be a binding constraint to send children to school, which reflects that policy efforts had been oriented mainly to primary education where distance to school was not significant. In this case, households current asset position i s also important (years o f schooling o f head o f household and land size), meaning that those households with worse current asset position are more vulnerable and less able to protect their long runinvestment in education. Yearly dummies are not significant, expect for 2001, wich maybe reflects a new wave o f policy efforts to increase attendance at this level. After controling 128 for household charcteristics external shocks dummies are not significant. Ifwe do not control for household characteristics, the dummy for coffee workers has a significant negative coefficient, but not the interaction term with year 2001, suggesting that lower attendance ratio at secondary level for households with coffee workers respond more a structural problem represented by householdcharacteristics. Table 17: Random effect tobit model for enrollment ratios enrollment ratio for 6-12 years old enrollment ratio for 13-18years old Explanatoryvariables (11 (2) (3) (4) (5) (6) (7) (8) HHSiZE -0.061'" -0.064** -0.061'* -0.022 -0,019 -0.021 (0.023) (0.023) (0.023) (0.021) (0.021) (0.021) FEMALEHEAD 0.379* 0.380' 0.359 0.016 0.011 0.011 (0.192) (0.194) (0.192) (0.151) (0.153) (0.151) HEADSCHOOL 0.098*' 0.102** 0.096** 0.137** 0.137*' 0.137" (0.021) (0.022) (0.021) (0.019) (0.020) (0.019) NETPClNCOMEt 0.064*' 0.070** 0.063* 0.006 0.006 0.005 (0.025) (0.025) (0.025) (0.014) (0.014) (0.014) PCREMlTTt 0.231'* 0.227'* 0.231" 0.103** 0.098" 0.096" (0.074) (0.074) (0.075) (0.037) (0.037) (0.037) LANDSIZE 0.045* 0.042' 0.042* 0.033** 0.035" 0.032" (0,019) (0.018) (0,018) (0.010) (0.011) (0.010) DISTANCE1 -0.064 -0.043 -0.062 (0.044) (0.045) (0.044) DISTANCE2 -0.026* -0.025' -0.025' (0.012) (0.012) (0.012) intensity -0.497 -0.087 0.141 0.755 (0.355) (0.401) (0.337) (0.397) intensity-D2001 0.490 0.171 -0.060 -0.094 (0.577) (0.649) (0.562) (0.627) earthquake 0.079 0.079 -0.054 -0.266 -0.271 -0.337 (0.208) (0.209) (0.204) (0.192) (0,191) (0.189) D-coffee-growers 0.868 1.114* -0.019 0.002 (0.512) (0.546) (0.381) (0.427) D-coffee-workers -0.174 -0.303 -0.234 -0.474*' (0.158) (0.185) (0.149) (0.178) Dgrowers-D2001 0.150 0.085 -0.252 0.151 (0.942) (0.896) (0.697) (0.690) Dworkers-D2001 0.116 0.019 0.162 0.082 (0.259) (0.286) (0.236) (0.264) Constant 1.680*' 1.708'* 1.702** 1.760** 0.377* 0.334 0.410* 0.566'" (0.221) (0.224) (0.222) (0.122) (0,191) (0.195) (0.193) (0.100) D1997 0.354** 0.363*' 0.364** 0.367" 0.164 0.199 0.184 0.217 (0.128) (0.130) (0.129) (0.128) (0.123) (0.126) (0.123) (0.125) D1999 0.299' 0.311' 0.308' 0.442" 0.200 0.197 0.218 0.282' (0.128) (0.128) (0.128) (0.124) (0.124) (0.124) (0.124) (0.122) D2001 0.493** 0.414' 0.440** 0.669'" 0.276" 0.385' 0.362* 0.502" (0.136) (0.173) (0.168) (0.168) (0.128) (0.166) (0.159) (0.158) Observations 1638 1615 1638 1620 1427 1407 1427 1458 Uncensored obs. 243 240 243 242 334 328 334 336 Left-censored obs. 168 166 168 167 458 449 458 471 Right-censoredobs. 1227 1209 1227 1211 635 630 635 651 Number of nu 691 682 691 682 647 640 647 650 Rho 0.40 0.42 0.39 0.50 0.31 0.30 0.31 0.40 LR test CHI2 71.98 72.69 75.08 34.92 83.58 82.88 86.84 21.96 Chi 2 for LR test for random effects 72.32 77.03 71.I2 118.59 51.48 50.45 51.91 91.28 Standard errors in parentheses * significant at 5%; **significant at 1% Conclusions 129 When studyingthe effects o f the coffee crisis inrural households inEl Salvador it i s necessary to recognize that over the past decade the economy has experienced important structural changes, among them, a reduction o f the share o f agriculture intotal GDP and the share o f rural population in the ~ountry.'~These effects have been captured by the BASIS/FUSADES panel data set in several ways. Households have increased the total and the average number o f hours in non- agricultural activities, while at the same time decreased them in agricultural activities. Given these trends, one has to be careful about attributing all the observed changes in coffee households' labor and income patterns to the most recent changes in international coffee prices, as some o f the shifts may be attributable to broader structural changes inthe economy. Nonetheless, the data set show that on average and over time fewer households are involved in the coffee sector. The probability o f being a coffee grower i s lower for 1997 and 2001 (especially 2001), precisely when the value added in the coffee sector was smaller compared to the earlier two years. This could imply that the impact o f the coffee crisis might not be very big in El Salvador, given the persistent trend o f movements away from the sector. Inother words, given the fact that over the past decade household have been moving away from agricultural activities, including coffee related activities, any negative impact that the most recent coffee crisis might have had i s smaller than if those previous changes had not occurred. This does not imply that for those households that have remained in the sector the effects o f the crisis may not have been devastating. Therefore, it i s important to identify who and where are those households. Household's socioeconomic characteristics show that those involved in the coffee sector with some members participating as wage laborers in the sector in 1997 have lower income and per income, less remittances, larger-than-average family size, and below average years o f education o f the head o f the household and education o f adults in general. In the other extreme, coffee growers have larger than average income and income percapita, larger remittances, higher levels o f education o f the head o f the households and adult members and smaller family size. Non- coffee households are in between. This points to the conclusion that the distinction between coffee-worker-only households and coffee growers i s important, with workers living in a more vulnerable situation (withrespect to poverty) than growers. Using alternative data sources it was possible to show a negative association between the intensity in coffee areas and declines in the percentage o f underweight children at the departmental level. In general, between 1998 and 2002, the percentage o f underweight children decreased more in those departments where the percentage o f land devoted to coffee production was smaller. An extreme case i s Ahuachapan, the department with the highest coffee intensity, where malnutrition actually increased between 1998 and 2002. Santa Ana i s similar in coffee intensity, but showed a small decline in child malnutrition. Putting together the fact that Ahuachapan and Santa Ana are two o f the departments with more coffee involvement, that the percentage o f coffee-workers households i s highest inthese two departments, and that these type o f households present socioeconomic characteristics that put them at the bottom o f the welfare scale in society, we see at least circumstantial evidence that the negative effects o f the coffee crisis were greater in these places. L a Libertad, Sonsonate, and San Salvador, with high coffee participation and intensity also show poor performance in reduction o f malnutrition, but not as 78The share o f agriculture as a percent of GDP, the share of the population that is rural, and the share o f the labor force that works in agriculture have all tended to decline across Central America since 1990. These trends reflect broader structural changes that were occurring in the region's economies. At the same time, cross-country data suggest that such structural changes, particularly in the labor force, may have been occurringrelatively rapidly in El Salvador (see World Development Indicators 2005). 130 bad as Ahuachapan and Santa Ana. As suplementary information, data at municipal level inthose departments with highcoffee intensity show that there i s positive correlationbetween coffee-area intensity and height-for-age ratios, supporting the idea tha coffee intensive areas present problems o f child nutrition, and therefore policy efforts should focused on programs to reinforce child nutrion on those areas, paying additional attention to efforts to focus the aid on those who need it most. From a dynamic perspective, income o f coffee-worker and coffee-grower households changes over time, following the same general pattern as that o f aggregate figures in national accounts, although their income i s more volatile than non-coffee households. Coffee household's income decreased in 1997 and 2001, and they also show or appear to have more mobility than non-coffee households. However, coffee growers are concentrated in the upper quintiles, while the distribution o f coffee workers households i s more concentrated inthe middle quintiles. Changes in per capita income and net income for coffee households show that the latter has more variation, and that i s explained in part by changes in family remittances, which turn out to be a source o f income smoothing related to migration and risk management. The 2001 negative income shock on coffee households was larger than that o f 1997, but it i s hard to isolate the changes between the shock incoffee prices and the earthquakes. In2001 remittances were higher for both types o f households, i.e., coffee-workers and coffee-growers. But for coffee growers, while in 1997 remittances were not able to offset the negative income shock, in 2001 they were. The difference mightbe that after the earthquakes, coffee growers, who are land owners, received considerable larger amounts o f remittances. One plausible explanation i s that by being land owners, they have more secure property rightsand, therefore, the incentives for reconstructionare higher. Regarding investments in education, we were not able to find a direct negative effect o f participating in the coffee sector on the probability o f sending a child to school. Econometric analysis shows that households without land, with l o w parents' education, with fewer remittances and that are farther away from school have more difficulties in sending their children to school. Coffee related activities do not seem to have a direct impact on schooling after controlling for other factors. However, given current asset position o f households with coffee workers, such as lower years o f schooling o f parents and smaller or no land, and that such households are poorer and receive fewer remittances, they might face difficulties in sending their children to school. But other types o f households also have similar characteristics. Even though the coffee crisis might have had a negative effect through the income effect, its impacts may have been offset by an aggressive public policy campaign to increase school attendance inrural areas through several programs o f school constructions, EDUCO, and Escuela Saludable (Healthy School) a school program that includes provision o f food supplements. 131 BIBLIOGRAPHY Angel, Amy. (1998): "El Fen6meno De "ElNifio" y El Sector Agropecuario y Pesquero Salvadoreiio, "Informe D e Coyunhra," San Salvador: Ministerio de Agricultura y Ganaderia. El Salvador. pp 124-137. Behnnan, Jere R. and Ani1B. Deolalikar. 1988. Health and Nutrition. Handbook of Development Economics. Volume 1.Hollis Chenery andT.N. Srinivasan editors. Elsevier. Birdsall, Nancy and Carol Graham, Editors. 2000. "New Markets, New Opportunities?Economic and SocialMobility ina ChangingWorld". Brookings InstitutionPress, Washington D.C. Conning, Jonathan, Pedro Olinto, and Alvaro Trigueros Argiiello, 2000. "Land and labor adjustment strategies duringan economic downturn inrural El Salvador". Technical Report, BASIS. Deaton, Angus. 1997. "The Analysis of Household Surveys: A Microeconometric Approach to Development Policy". Johns Hopkins UniversityPress. Ferranti, David, Guillermo E.Perry, Indermit Gill and Luis Servkn. 2000. "Asegurando el futuro en una economia globalizada".Banco Mundial. Washington, D.C. Jacoby, H.and E. Skoufias (1997): "Risk, FinancialMarkets, and Human Capital ina Developing Country," Review of Economic Studies, 64.Jacoby, Lewin, Bryanand Giovannucci, Daniele. 2002. Global Supply and Demand: New Paradigms in the Coffee Markets.WorldBank Fields, Gary S. 2001, "Distribution andDevelopment: A new look at the developing world". The MITPress.2001. Stiglitz, Joseph. 2000. "Reflections on Mobility and Social Justice, Economic Efficiency, and Individual Responsibility". Chapter inBirdsall and Graham (2000). Trigueros, Alvaro. 2002. "The Economics of Schooling and Child Labor for Boys and Girls in Rural Households inEl Salvador: 1995-1999". Ph.D. Dissertation. Vanderbilt University. Nashville, Tennessee. 132 m m ri .-P > > Y CI .-Y .-b CI 0 'E0 'E0 8 28 LLI w c 0 Lc 0 c c C 0 m E m E 4 - bm u 6 I I f : I 1 tI f I ! 1 ! I1 ! 1 f 1 1 I !I i cI .! I uI 1I . 1 II ! I 4 1 I CI I I I I C / 0 : 2 : a i zt c " i , Table 8.2: Mean and Standard Deviation of Variables Describing Household Characteristic 1995 1997 1999 2001 VARIABLE Mean S.D. Mean S.D. Mean S.D. Mean S.D. Demographics Household Size 5.38 2.38 5.54 2.87 5.00 2.64 5.19 2.84 # of Children 0-5 years old 0.73 0.83 0.73 1.25 0.46 0.86 0.42 0.81 #of Children 6-12 years old 0.54 0.76 0.81 0.90 0.73 0.87 0.88 1.07 #of Children 13-18years old of Adults 0.58 0.95 0.69 1.05 0.46 0.71 0.50 0.71 19-59years old 2.62 1.33 2.42 1.53 2.19 1.47 2.15 1.54 # of Adults 60 years old or more 0.92 0.84 1.04 0.82 1.15 0.88 1.23 0.86 Characteristicsof Household Head Age of Head of Household 57.85 16.02 61.65 11.33 64.35 11.09 66.15 10.96 Dummy variable for Female Head 0.08 0.27 0.15 0.37 0.15 0.37 0.19 0.40 Years of Schooling of Head of Household 4.42 3.43 3.31 3.03 3.62 3.35 3.23 3.18 Education Average Years of Schooling of Adults* 5.52 3.01 5.56 3.44 6.32 3.60 5.49 3.51 Number of Children 6-12 years old in school 0.50 0.71 0.81 0.90 0.65 0.85 0.85 1.01 Number of Children 13-18years old in school 0.42 0.81 0.35 0.63 0.38 0.70 0.38 0.70 Percentage of Children ages 6-12 enrolled in L school* 0.95 0.16 1.00 0.00 0.88 0.31 0.97 0.09 Percentage of Children ages 13-18 enrolled in school* 0.68 0.47 0.48 0.45 0.78 0.44 0.70 0.48 Distance to closest primary school* 1.36 1.39 1.27 1.29 0.62 0.55 0.85 0.84 Distance to closest secondary school* 4.72 3.57 3.07 2.91 3.08 3.00 2.33 1.99 Work At least one household member works in a farm 0.27 0.45 0.42 0.45 0.31 At least one household member works in agriculture 0.27 0.45 0.42 0.45 0.31 Income Remittances 2,395 6,825 5,853 12,216 Agricultural Income 18,217 15,389 3,572 4,566 Non Agricultural Income 7,380 24,702 20,502 18,755 Incomeoutside home 8,475 16,954 11,650 14,861 Incomeinside home 15,685 24,663 9,859 15,882 Total Income 28,269 27,936 31,098 26,128 Total Per Capita Income 5,921 8,393 7,082 5,776 Net Income 25,597 28,443 24,074 19,476 5,327 8,693 5,187 138 0 b, .E .E L 8 e0P 504 Table 10: Allocationof time amont differenttypes of activities and households. Source FUSADESBASIS suwep, 1995, 1997, 1999and 2001 The sample us&ISforlhe 1995119971199912001balanced panel(n=451) 142 1 I- r o o 3 o a ? 9 " 3 0 0 3 N a n o m ?3c m" q o 1995/1997/1999/2001panel. 145 199511997/1999/2001 panel. 146 13-15 76.1 77.4 69.1 77.5 75.6 16-18 48.1 47.5 53.1 53.0 50.2 19-25 11.2 11.8 12.7 13.6 12.4 Total 57.2 58.7 54.7 57.7 57.4 Source: BASIS surveys, 1995, 1997, 1999,2001. The sample used i s for the 1995/1997/1999/2001 panel. 147 Figure7: Changeinthe percentageof underweightchildrenbetween 1998 and 2002 by departmentsandpercentageof coffee-workershouseholdsinthe department. 0AHUACHAPAN OSAN SALVADOR DSANV I C E N ~ S ~ ~ 0SANTAANA 0CABARAS 0SONSOnTTBSC%!db!BERTAD OCHA!ATENANGOeLA PAZ OLA UNION OMORAWN OUSULLlTAN I I I I 0 10 30 40 148 Copingwith the CoffeeCrisisinCentralAmerica: The Role o f Social Safety Nets inHonduras David Coady, Pedro Olinto, and Natalia Caldes InternationalFood Policy Research Institute Washington D.C. 1. Introduction Inspite ofsomerecoveryin1994and 1998, worldcoffeepriceshavedeclineddramaticallysince the mid 1980s. Real coffee prices are now at their lowest levels in more than 50 years. Such a substantial decrease in prices has obviously had enormous adverse implications for the incomes o f many o f the coffee producing countries in Central America and for coffee producers in particular. However, in spite o f the widespread perception o f rising poverty, little rigorous empirical evidence exists regarding the magnitude and nature o f the poverty impact o f the crisis. Even less evidence exists with regard to the potential role for social safety net programs both in protecting poor households from such shocks as well as facilitating more efficient response^.'^ This paper contributes to filling some of these knowledge gaps. Honduras i s one o f the poorest countries in Latin America with around 70% o f its population classified as poor. Unlike other countries in the region, poverty i s predominantly a rural phenomenon, with 49% o f the total population in 1998 (around 6.5 million) living in rural areas where poverty and, in particular, extreme poverty, are substantially higher (World Bank, 2001; Morris et al, 2002). Also, the extent o f poverty in rural areas has increased in recent years and much o f this i s often attributed to the continued decline incoffee prices over this period. The analysis inthis paper uses household-level survey data collected as part o f an evaluation o f a recently introduced social protection program in Honduras (called Programa de Asignacion Familiar, PRAF). These data provide a very rich source o f information for the purpose o f addressing the above empirical issues. We use these data to evaluate the impact o f the ongoing crisis on some o f the poorest rural regions and households in Honduras. In particular, we evaluate the effectiveness o f PRAF in enabling poor households to protect their welfare against such shocks. The analysis may also help to improve our understanding o f the role o f social safety net programs more generally in protecting against economic shocks as well as o f the nature o f riskcoping strategies available to poor households. InSection2 we provide abriefoverview ofcoffee inthe contextofthe Honduran economy. In Section 3 we describe the program and survey design. Section 4 provides some descriptive analysis o f the data, with special emphasis on the extent o f involvement in coffee in the program area as well as differences insocio-economic characteristics across coffeehon-coffee households. 79See Morris et a1(2002) for a discussion of the impact o f Humcane Mitch in late October 1998 on some o f the poorest municipalities inHonduras, inparticular the discussion on the extent and distribution offoreign aid. The tropical stormMichelle also affected substantial areas in Honduras in the fall of 2001. 149 Section 5 evaluates the impact o f the program on household consumption and Section 6 examines impacts on labor supply on coffee farms. Section 7 provides some concluding remarks. 2. CoffeeinHonduras Coffee i s extremely important to Honduras in terms o f output, employment, and export earnings (McCarty and Sun, 2003; Partners, 2002; Varangis, 2003). Interms o f output, in 2002/2003 the area under coffee cultivation reached 233,750 hectares, accounting for 65% o f total permanent crop area.80 A large amount o f the coffee i s also grown on small farms in isolated, high altitude areas; for 92% o f growers annual coffee production i s less than 100 quintales per farm, most o f which i s sold to the market through intermediaries. Coffee i s also seen as a major source o f employment, employing up to one third o f the labor force in rural areas (Hearne et al, 2002). Coffee i s the second most important export crop in Honduras in value terms and accounts for almost 25% o f national GDP. In 2000 the value o f coffee exports was $345.2 million, which represents over 26% o f total export revenue for that year. During2000 and 2001, continued increases inworld supply led to some of the lowest real coffee prices over the last few decades. Since mid 2001, both the nominal and real prices o f Arabica coffee, both domestically and on internationalmarkets (Figure l), have continued to decline. Yet, in spite of this, coffee production in Honduras (as elsewhere) has been increasing over this period. Figure 1: World andHondurasDomesticCoffeePrices andProduction 250000 200000 t50000 .-u v1 0 e u 100000 V .50000 ia 3 - 0 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 -WorldPrice of Arabica. Nominal-- Domestic ProducerCoffee Prices. Honduras- CoffeeProduction. HondurasI Source: International Coflee Organization, FA0 (Reproducedfrom McCarty and Sun, 2002). The main coffee variety produced in Honduras is the Arabica coffee variety. Inrecent years there has been concerns about the quality o f the coffee being produced and its economic consequences, with Honduran coffee being subject to penalties o f between 5-20 cents per pound on world markets over the past few years (Heme et al, 2002). The European Union is the predominant destination for Honduran coffee exports. 150 3. The Program and SurveyDesign Inlate 1998, the Government ofHonduras initiatedthe PRAF/IDB Phase I1Project, which is one o f the largest social investments in the history o f Honduras. The program covers households in 70 (out of 297) o f the most disadvantaged municipalities in 7 departments (Copan, Intibuca, Ocotepeque, F. Morazan, L a Paz, Sta. Barbara and Lempira), these all being located in the western part o f Honduras and covering many o f the major coffee-growing departments. Program transfers are roughly 3% o f total household consumption on average, but are substantially higher for the poorest households. For the purpose of selecting participating households, the government first identified the poorest 70 municipalities based on the average height-for-age o f children in first grade.81 These municipalities were then categorized into five strata based on the same variable and, within each stratum; municipalities were randomly allocated to four program evaluation groups (Le. demand- and supply-side transfers - 20 municipalities, demand-side transfers only - 20 municipalities, supply-side transfers only - 10 municipalities, and 20 control municipalities). All households with children less than 13 years or with a pregnant woman were considered eligible for the program. As part o f the evaluation, a randomly selected sample o f households was surveyed before and after the program intervention. This survey contains information on 5,484 households from the 70 municipalities covered by the program evaluation (i.e. 80 households per municipality). Households from 20 o f these municipalities were eligible to cash transfers conditioned on attendance at school and health clinics, 10 were eligible for only increased investments inschools and health clinics, 20 were eligible to transfers and supply-side investments, and 20 were not eligible to any component o f the program and were used as a control for the purposes o f program evaluation. The survey questionnaires were implemented between August-December 2000 (baseline) and June-September 2002 (follow-up). The surveys contain sixteen modules (including information on household conditions, household composition, remittances, education, expenditures, matemakhild health, anthropometrics, womerdchild time use, quality o f schools/clinics, other programs, goods/animals, community survey) containing over 500 variables. In this paper we will focus on three outcome variables, namely total household consumption, food consumption and labor supply decisions. Two features o f the data collection process have important implications for the way in which we analyze the impact of the program on consumption and labor supply across coffee and non-coffee households. Firstly, the baseline and follow-up surveys were collected at different times o f the year, with the 2000 baseline being collected over the months August to December and the 2002 follow-up survey collected over the months June to September. Secondly, for operational reasons, the baseline survey was implemented first in the treatment municipalities (i.e. in those where households were later to receive cash transfers). Both these timing discrepancies mean that great care needs to be taken when comparing differences in consumption (and other) outcomes across program treatment/control groups as well as over time. W e will address such issues inmore detail below. See the Appendix A for a more detailed discussion of the programand survey design. 151 4. Data Descriptionand Coffee Involvement Inthe dataset, we have 5,484 households inthe baseline survey. The survey includes two sets of variables we could use to classify households into coffeehon-coffee households: (i) Aquestionthatasksifahouseholdhaslandand,ifso,howmuchofthisisdevotedto coffee. This question is asked only in the 2002 follow-up survey and is in a module relating to landholdings and their allocation in2001. (ii) A question in each survey asking how each working individual's time was allocated between various activities over the seven days prior to the survey, including coffee- related activities mainly referring to coffee-harvesting activities. With respect to the second question relating to involvement in coffee activities, since the coffee harvesting season covers the months September-November, in the follow-up survey over June- September households are much less likely to report involvement in coffee activities since it i s prior to the harvest. This i s not the case in the baseline survey over August-December, which covers the harvest season. Therefore, one cannot gain any insights, other than possibly about seasonal patterns, from comparing the extent o f households labor involvement in coffee-related activities across survey rounds. The different seasonal timing o f the surveys thus means that one expects labor activity to be greater inthe baseline survey. W e start by examining the pattern o f coffee involvement in the program region. For this purpose we categorize sample households into four categories according the amount o f land they own: landless, small holders with one manzana or less, medium holders with 5 manzanas or less, and large holders with over 5 manzanas. From Table 1 we can see that 30% o f households are landless, 23% are small holders, 29% are medium holders, and 18% are large holders. As indicated earlier, because o f the different timing o f the surveys we expect to observe less labor activity in coffee in the follow-up survey and this i s indeed borne out in Table 1 with 24% o f households reporting labor activity in coffee in the baseline survey and only 11% reporting such activity in the follow-up survey. Inboth cases, the incidence o f labor activity i s highest amongst households with land and also slightly higher for large holders. In Table 1 we can also see that the proportion of households with land reporting some land allocated to coffee i s quite high at 42%, ranging from 39%-44% across the three land size categories. Below we use two definitions o f a "coffee household". Inthe first, we simply use a binary variable that indicates whether or not a householdhas some land devoted to coffee. From Table 1 we see that 42% o f households with land report having some land devoted to coffee. In the second, we use a binary variable that indicates whether or not a household lives in a coffee cluster, i.e. in a cluster where at least one household reports having some land devoted to coffee production. This i s obviously a much broader definition o f a coffee household and 76% o f households are classified as coffee households under this definition (Table 1). To the extent that labor hiring for coffee activities i s local, one expects the additional households incorporated under the second definition to include coffee laborers. Note also that, under this definition o f a coffee household, there i s no substantial difference in the proportion thus involved in coffee across landless households or households with different land holdings. Below we will refer to these distinct groups "coffee land", "other coffee", and "non-coffee" households. 152 Labor Activity Landholding Share o f 2ooo Have Land Under InCoffee Category Households 2002 Coffee Cluster Landless 0.30 0.18 0.06 0.72 <=1 manzana 0.23 0.26 0.12 0.44 0.80 <=5 manzana 0.29 0.26 0.12 0.39 0.74 >5 manzanas 0.18 0.29 0.14 0.43 0.81 Total 1.oo 0.24 0.11 0.42 0.76 Table 2a presents the some descriptive statistics for those with land under coffee, those without coffee land but living in "coffee clusters", and the remaining "non-coffee" households - see below for more discussion on these definitions. Those households growing coffee have larger household size, more adult household members, are less likely to have a female head of household, and have higher total consumption and food consumption. Table 2b presents the results of a probit regression where the dependent variable indicating whether or not a household allocates some land to coffee, conditional on having land. The coefficients indicate that households with land on a hillside are more likely to allocate land to coffee as are households where the head has secondary education. Also, those with larger farm holdings are also more likely to allocate land to coffee. Characteristics such as household size, number of adult males, gender of head of household and age of head do not show up as being significantly associated with the decision to allocate landto coffee. Table 2a: DescriptiveStatisticsby Coffeemon-CoffeeClassification. Coffee Land Other Coffee Non Coffee Household size (persons) 5.84 5.49 5.79 Persons<=6years 1.20 1.23 1.27 7<=Persons<=lO 0.91 0.88 0.95 11<=Persons<= 19 1.21 1.04 1.15 20<=Persons<=49 1.87 1.76 1.73 50=Persons<=64 0.44 0.37 0.45 Persons>=65 0.2 1 0.21 0.24 Age of head 45.1 44.0 45.6 Female head 0.15 0.21 0.19 Headhas no education 0.26 0.26 0.26 Headhas primary education 0.65 0.67 0.69 Headhas secondary education 0.05 0.04 0.03 Total landholdings (ha) 22.6 25.5 18.5 Per capita landholdings (ha) 4.64 5.40 5.17 Per capita total daily consumption 22.2 19.5 18.0 (2000 lempiras) Per capita food daily consumption 13.0 12.2 12.2 (2000 lempiras) Numberofhouseholds 1597 2220 1213 Note: Coffee Landindicates household has some landunder coffee; Other Coffee indicates householddoes not have land under coffee but lives ina cluster where at least one household does have land under coffee; Non-Coffee indicates households not inf i s t two categories. 153 I I IIIII I I IIIIIII IIIIII I -I- L 5. Coffee and Consumption Table 3 presents data on the pattern o f the change in total household per capita consumption and household per capita food consumption across coffee and non-coffee households.'' Overall, we observe a 16% decline in total real per capita consumption levels between 2000 and 2002. It i s also noteworthy that households with coffee land experience a substantially larger 20% decline, compared to a 16% decline for other-coffee households and a 12% decline for non-coffee households. % Change inPer Capita Household Consumption Samde Size Samde Share Total Food Coffee Land 1597 0.32 -0.20 -0.16 Other Coffee 2220 0.44 -0.16 -0.13 Non Coffee 1213 0.24 -0.12 -0.09 Total 5030 1.oo -0.16 -0.13 The final column presents the change in (real) per capita food consumption across household groups. In aggregate, per capita food consumption decreases by 13%. A similar pattern o f changes i s observed across household categories, with coffee-land households experiencing a 16% decline, other-coffee households a 13% decline, and non-coffee households experiencing only a 9% decline. InTable 4 we presentthe percentage change inreal consumptionby economic quintile grouping. Because consumption i s measured with a substantial amount o f error (or, equivalently, has a large transitory component), we use an asset index as our proxy for "permanent" (or long-run) household income. The first column o f numbers shows the ratio o f mean per capita total consumption to that o f the lowest asset quintile. Although all in these communities may be perceived as being poor, they are obviously not equally poor. For example, the consumption o f the most asset-rich group i s over 2.5 times that o f the most asset-poor group. *'Total household per capita consumption is in real terms using the change in the exchange rate changes as our proxy for inflation. Note that although one cannot sensibly compare changes across program groups because o f the different survey timing, this is not the case for comparisons across coffeehon-coffee groups since these are distributed uniformly across survey periods. However, the timing differences between the survey rounds (as well as within the baseline) means that the time pattern contains an element o f seasonality. For this reason we focus more on relative changes across groups. 155 c e s and CoffeeIntensity By EconomicQuintiles. %Change Consumption 2000-2002 Consumption Per Capita Per Capita Coffee Coffee Ratio Total Food Land Cluster Asset Quintiles L o w 1.oo -0.11 -0.09 0.23 0.77 2 1.09 -0.17 -0.11 0.27 0.74 3 1.21 -0.15 -0.12 0.32 0.74 4 1.51 -0.19 -0.15 0.32 0.74 High 2.52 -0.18 -0.16 0.43 0.81 Note: Consumption i s inreal 2000 prices. The next two columns present the percentage change in per capita total and food consumption over the two-year period 2000-2002. In the case o f total per capita consumption, all groups exhibit large declines although the decline i s somewhat smaller for the poorest quintile. Inthe case o f food consumption, all households experience a decline with this decline being higher for more asset-rich households. The final two columns describe the coffee-intensity o f the different asset quintiles. The asset-rich households are clearly more coffee-intensive in terms o f having land devoted to the crop. However, with the broader definition o f coffee intensity, we observe less differences across asset quintiles suggesting that any adverse coffee shock may affect all households adversely. 6. Modelingthe programimpact on consumptionand crisis mitigation Inthissectionwe takeamorestructural approachandinvestigatethe impactofthe drop incoffee prices and the cash transfer program on household allocation decisions and expenditure outcomes. We do so by studying the channels via which the drop in coffee prices may have affected household expenditures and labor allocation, and identifying the impact o f transfers in mitigating the shock. InAppendix B,we develop atheoretical model do guide our empirical analysis. As describedin the appendix, the key result o f the model i s that household's response to an unexpected coffee price shock will depend on their access to credit, and the additional cash from the program will only impact their allocation decisions if they are credit constrained. Intuitively, we expect the impact o f a shock to depend on the ability o f households to smooth consumption over time. Since access to credit is a key factor in a household's ability to smooth consumption, we expect responses to shocks to depend on whether or not the household has access to credit. Having access to credit allows the householdto smooth consumption so that we would not expect a shock to impact much on either consumption or production decisions o f these households. Without access, households may have to decrease consumption and reallocate labor off-farm (i.e. away from on-farm coffee investment activities, e.g. maintaining coffee trees) in order to support current consumptionlevels. Therefore, cash transfers should not affect the labor supply o f credit-unconstrained households, but they are likely to induce credit-constrained households to supply more labor to maintaining coffee trees. So, for these credit-constrained households, the program's cash transfers should have both a substitution (Le. due to conditioning) effect and a liquidity effect on labor allocation decisions. Accordingly, credit-unconstrained households would not need to generate as much 156 additional current income to smooth consumption as credit constrained-households would because they can borrow freely. Therefore, they would sell less labor off-farm and would be able to better maintain their coffee trees than credit-constrained households. This would result in a higher shadow value o f coffee land for credit-unconstrained households. W e test these two implications o f the model below. 6.1 Transfers, credit constraintsand labor allocation on coffee farms W e first test the implications o f the theoretical model for the household's decision to supply labor to coffee related activities. Ignoring the conditionalities o f the program transfers for a moment, the model presented in the appendix suggests the following relationship between labor supply to coffee activities and access to cash transfers: -=0,ifthehouseholdisliquidityunconstrained; dl= dz ->0,ifthehouseholdisliquidityconstrained, dl az where 1' represents the number o f hours supplied by the household to coffee related activities in the last 7 days, and z i s the amount o f cash received by the household. Hence, without conditionalities, cash transfers would only impact labor allocation via a liquidity effect on credit- constrained households. Under the conditionality o f the program, however, children aged 6-13 o f beneficiary households need to be attending school to ensure continued eligibility o f the household. This may result in a drop on the supply o f labor from these children, and a simultaneous increase in the supply o f labor to coffee related activities from members older than 13. Thus, such conditional cash transfers (CCTs) may affect labor allocation o f adults via both a liquidity effect and a labor substitution effect. Inthe analysis that follows we econometrically estimate two labor supply equations to test the implications o f the model above: the first for hours worked in coffee related activities by household members 14 and older, and the second for children aged 6-13. For the dependent variable we use the information on the main activity performed by each member o f the household six years o f age and older. We only include activities related to maintaining coffee trees, weeding and fertilization o f coffee land, and exclude activities related to harvesting coffee. We do so because the former i s more likely to respond to price shocks and cash transfers that affect current consumption since they represent investment for future consumption. Since harvest related activities can be transformed into cash for current consumption more easily, we only estimate the labor supply equations for the data collected in 2002 (because o f the timing o f the surveys, more o f the labor activities reportedin2000 relate to harvesting activities). As for determining whether the household was credit unconstrained or constrained, we use the data for the application to formal and informal credit providers. Households that had all their credit applications denied were classified as credit constrained. So were households that did not apply to any credit provider, but reported that they would be denied credit from all providers if they had applied. 157 As for cash transfers, we use an indicator variable (dummy variable) that indicates whether the family was eligible or not for the program. We use eligibility instead o f actual receipt because current labor supply to coffee activities i s likely to be determined by whether the household expects to receive cash soon rather than whether they actually received cash in the past. This i s especially important for the data collected in 2002 because the majority o f the interviews were carried out concurrent with the distribution o f cash inthe region. Table 5 below presents the results for the Tobit estimation o f the two labor supply equations just described (see Appendix Table 1 for more detailed results). The estimates are for the sub-sample of coffee growers only (1756 households).This makes it more likely that the labor supply reported refers to work on their own land and not work provided to other farmers. Note first that the coefficients on the coffee land variables indicate that both adult and child labor allocated on-farm i s significantly higher on larger coffee farms, the latter highlighting a potential conflict between investments in coffee land and investments inchildren's human capital. The results for the labor supply of adult members (14 and older) indicatethat there is no impact of the cash transfer on labor supply for credit-unconstrained households, as predicted by the theoretical model. This indicates that the conditionality of the program does not induce an increase in labor supplied to coffee activities by older members, that is, there is no substitution effect. However there does appear to be some liquidity effect of transfers on adult labor supply in credit- constrained households - although this effect is significant only at the 10% level. Since the dependent variable is the log of hours, we need to manipulate the coefficients to derive the marginal impact. Let T be the dummy for treatment, Ci be the dummy for credit constrained households, and Xi other explanatory variables. Then our specification is: ln(y 1) = + mi+yT,.C,+6Xi + E so that, = +UT 4 +&+E) -1 and, Then the marginal impact i s calculated as: From the regressions we have the mean value o f (y+l) i s 10.06, p=0.955 and y=1.394. Therefore, for credit constrained households: E id r - i 2 (p+fli)E[(y+l)] +1.394)x = = (0.955 1.5 = 3.27 That is, the transfer (or eligibility for it) seems to increase the supply o f adult labor in credit constrained households to coffee activities by 3.27 hours per week. So even the small cash transfers given by PRAF do appear to slightly alleviate the liquidity constraints faced by some coffee growers and enable them to redirect their labor to short-term investments focused on maintainingthe productivityof their coffee crops. 158 Table 5: Tobit estimatesof coffee-relatedlabor supply by coffee growers. Hours Hours supplied by supplied by persons I 4 children aged Explanatory Variables: and older 6 to under 14 Area under coffee (manzanas) 0.215** 0.926** (0.048) (0.303) Area under coffee squared -0.004** -0.025* (0,001) (0*012) Dummy1ifcredit constrained -0.455 0.243 (1.OSl) (2.682) Eligiblefor transfer 0.955 -2.274 (0.567) (2.4 14) Eligible for transfer X credit constrained dummy 1.394 -21.153** (1.511) (2.693) Observations 1756 1756 Note: Dependent variable is log of (labor hours plus one). Regressions also include education o f household head, household composition, land elevation, location; these results are presented in Appendix table 1. Standard errors inparentheses:+ significant at 10%; * significant at 5%; ** significant at 1% The results in the second column also indicate that the liquidity effect is more important than the substitution effect induced by the conditionality of transfers. As the results indicate, it seems that the conditionality of the transfer is not affecting the labor supply to coffee activities of the young in credit-unconstrained households. This is consistent with, for example, children from unconstrained households only working on such activities outside of school hours. It is also consistent with the results of the estimated impact of PRAF on enrollment, which suggest that enrollment rates for children aged 6-13 are quite high, especially for higher income households (Glewwe and Olinto, 2004). However, additional liquidity to credit-constrained households does induce a lower supply of labor by children to under 14 to coffee related activities. Fromthe regressions we have the mean value o f (yt-1) i s 1.499, p=-2.274 and y=-21.153. Therefore, for credit constrained households: E - = @+;tC,)E[(y+l)]=(-2.274-21.153)~1.5 = -35.14 [:I That is, the transfer (or eligibility for it) seems to decrease the supply o f child labor to coffee activities by about 35 hours per week. Since on average households have 1.41 children aged 6 to 13 this implies an average decrease o f nearly 25 hours o f work per childper week. This decrease in labor supply by children is quite substantial and one would therefore expect a corresponding increase in time allocation elsewhere, e.g. more time in school. However, the evaluation of the program impact on education outcomes indicates that enrollment rates increased by only 2.6 percentage points as a result of the program, absenteeism decreased by just under 2 days per month, the probabilityof a child dropping out of school decreased by around 6 percentage points, and there was no evidence of any statistically significant program impact on child labor force participation (Glewwe and Olinto, 2004). In total, the program is expected to increase the schooling of 14 year olds in poor rural Honduras by 0.7 years. In nearly all cases, the beneficial educational impacts were substantially higher for the poorest households. One interpretation of the above results is that the decrease in coffee labor time by children in households that are 159 credit constrained results both in an increase in time at school but, more importantly, in an increase in the time availableto children either for educationactivities in the home or leisure. Thus, together the results from both equations suggest that the extra liquidity provided by transfers is helping cash constrained coffee growing households cope with the coffee crisis maintain their coffee trees productivewithout compromisingthe schooling of their young. That is, transfers seem to be helping coffee households maintain their investment in human capital despite the likely effect of sharp drop in prices on current consumption. 6.2 CCTs,credit constraints and household expenditures W e now look at the channels through which conditional transfers may have impacted per-capita expenditures (PCE) o f coffee and non-coffee households in western Honduras. As indicated by the model presented in Appendix B, there are three channels via which transfers may have impacted PCE. First, as indicated by the analysis o f the impact o f transfers on labor supply presented above, PRAF beneficiaries may have been better able to maintain the productivity o f their coffee land by selling less labor off-farm and investing more on coffee tree maintenance, soil fertilization and weeding. This investment effect would translate into a higher shadow value o f coffee land for transfer beneficiaries. Secondly, transfers may have provided sufficient additional income to compensate for the drop in coffee prices. This would be a direct income effect o f transfers, which could be stronger for coffee farmers since they have suffered a coffee price reduction. Finally, transfers may have provided sufficient liquidity to credit-constrained coffee farmers thus allowing them to smooth consumption. Table 6 presents the results of an instrumental variable estimation of a PCE function on several household and municipality characteristics, including whether the household is credit constrained, whether it received transfers from PRAF, and whether it owns coffee land. For transfers we use a dummy indicating take up of the program and not just eligibility. Given that take up is likely to be endogenous, we instrument it with the exogenous eligibility variable. The regressions also include municipal dummies to control for municipal leveleffects. 160 Table6: IV estimationof the per-capitaexpenditure (PCE) functions for 2002 and 2000. PCEin PCEin 2002 - 2002 2000 2000 Received CCT 0.038 -0.031 0.068 (0.05 7) (0.048) (0.075) Received CCT X area under coffee (manzanas) 0.008 0.010 -0.003 (0.010) (0.008) (0.013) Received CCT X credit constrained dummy 0.061 -0,010 0.071 (0.069) (0.057) (0.089) D u m m y 1 ifhouseholdheadhas incomplete primary 0.109** 0.150** -0.041 (0.019) (0.021) (0.028) Dummy1ifhouseholdheadcompleted primary 0.414** 0.409** 0.005 (0.032) (0.035) (0.048) Dummy=l ifhouseholdheadhas incomplete secondary 0.815** 0.770** 0.045 (0.053) (0.069) (0.087) D u m m y 1 ifhouseholdhead completed secondary 1.155** 1.206** -0.051 (0.066) (0.069) (0.095) Dummy=l ifhhheadhas somepost secondary schooling 1.355** 1.146** 0.209 (0.107) (0.110) (0.153) Log o fhouseholdpopulation -0.425** -0.336** -0.088* (0.028) (0.026) (0.038) Area under coffee (manzanas) 0.034** 0.030** 0.004 (0.00 7) (0.005) (0.008) Non coffee land area (manzanas) 0.002+ 0.001 0.001 (0.001) (0.001) (0.001) Dummy=l if householdis credit constrained -0.095+ -0.056 -0.038 (0.048) (0.035) (0.060) Share o f householdpopulation under 6 -0.932** -0.876** -0.056 (0.073) (0.078) (0.107) Share o f householdpopulation 6 to under 10 -0.639** -0.695** 0.056 (0.079) (0.101) (0.128) Share o f householdpopulation 10 to under 14 -0.471** -0.578 ** 0.106 (0.087) (0.093) (0.127) Share o f householdpopulation 14 to under 18 -0.262** -0.290** 0.028 (0.077) (0.077) (0.109) Share o fhouseholdpopulation 40 to under 60 -0.056 -0.003 -0.052 (0.048) (0.064) (0.080) Share o fhouseholdpopulation 60 and above -0.371** -0.310** -0.061 (0.060) (0.066) (0,089) Minutes walking to closest public transportation -0.002** -0.001 ** -0.000 (0.000) (0.000) (0.000) Days from May 1'' -0.002 -0.015 0.013 (0.003) (0.012) (0.012) Days from May lst squared 0.000 0.000 -0.000 (0.000) (0.000) (0.000) Elevation inmeters above sea level -0.000 -0.000 -0.000 (0.000) (0.OOO) (0.000) Elevation inmeters above sea level squared 0.000 0.000 0.000 (0.000) (0.000) (0.000) Constant 3.943** 4.77 1 ** -0.952 (0.197) (0.934) (0.948) Observations 5663 10744 R-squared 0.42 0.41 +Standarderrors 10%; * significantat 5%; ** inparentheses significantat significant at 1% 161 The first column of the table presents results of the estimation using only the 2002 survey data. The second and third columns present the results from a regression that pools the 2000 and 2002 data, but interacts all variables with a dummy variable for 2002. This allows all parameters to change from 2000 to 2002, and allows us to test whether the shadow value of each variable has changedover the period. The third column presents the coefficients o f all interacted variables, i.e. the difference between the 2002 coefficients and the 2000 coefficients presented in the second column. As can be seen, the only coefficient that seems to significantly change from 2000 to 2002 i s the coefficient on "Log of household population". However, ajoint Wald test cannot reject the nullhypothesis that all interaction coefficients are zero (P-value = 0.76). Therefore, we focus the discussion on the results for 2002 presented inthe first column. As can be seen in the first column, PRAF transfers seem to have no direct income effect on current PCE. Also, while credit constrained households do appear to spend less per-capita (at the 10% significance level) than unconstrained households, the transfers given by PRAF do not seem sufficient to induce higher levels o f consumption, that is, there seems to exist no liquidity effect o f transfers on PCE.This indicates that the liquidity effects found in the labor supply analysis are not enough to affect current consumption and that the adult and child labor effects are likely to cancel out. Finally, although the returns to coffee land appear to be higher relative to non-coffee land, the results also indicate that there i s no effect o f transfers on the shadow value o f coffee land. Thus, it seems that the investment effect described inthe previous section i s not sufficiently large to affect the productivity o f coffee land in2002. For such effects to emerge it i s likely that the size o f the cash transfer needs to be considerably larger. 7. Concluding Remarks In this paper we have described the nature of coffee involvement in some o f the poorest rural municipalities in western Honduras. Our analysis indicates that the region i s very coffee intensive with a large proportion o f land devoted to coffee. W e found that households with larger landholdings are more likely to be involved in coffee as are households whose heads have greater than primary education. Also, hillside farms are more likely to grow coffee. Our analysis o f the consumption data provides a picture o f the change in total household consumption over the period 2000 to 2002, a period when households were hit with two economic shocks, namely, a drought and a continuing decline in international coffee prices. We find that the percentage decreases in total household per capita consumption and in per capita food consumption are highest for households with coffee land, followed by households living in coffee clusters who are likely to rely indirectly on the coffee economy. Inthe second halfofthe paper we evaluate the impact of a transfer program implemented inthe area over the period on labor supplied to coffee related activities and on household per-capita expenditures. It i s noticeable that household labor time o f both adults and children appear to be substantially higher on larger coffee farms, the latter highlighting a potential conflict between investments in coffee land and investments inthe human capital o f children. Our results indicate that the cash transfers given out by PRAF, which are also conditioned on keeping kids in school, have significantly affected the labor allocation decision o f credit-constrained coffee farmers. The additional liquidity provided by the transfers seems to have allowed families to maintain their children inschool and increased the time dedicated by adults to maintaining coffee trees. Inother 162 words, the fact that the transfers have been conditioned on investments in child education seems to have ensured that higher on-farm investment labor activities have not come at the expense o f investments inchildren's human capital. The results from the per-capita expenditures regressions also indicate that, although credit- constrained households have lower consumption levels, the transfers may not have been large enough to increase current consumption o f either credit-constrained or unconstrained households. This is not surprising since the program was designed to induce human capital accumulation by simply covering the opportunity cost o f children's time. It was not intended to increase current household expenditure and reduce current poverty. It seems however that an increase in the amounts transferred would provide much needed current poverty alleviation and could possibly contribute further to alleviating liquidity constraints o fpoor households. 163 References FA0 (Food andAgriculture Organizationof the UnitedNations). At Glewwe, P., and P. Olinto (2004): "Evaluating the impact of conditional cashtransfers on education: An experimental analysis ofHonduras' PRAFprogram", Final Report for USAID, January, unpublished. Hearne, R., B.Barbier and J.M. Gonzalez (2002): "Development of a minimumcost, incentive- basedplan for the implementationof atechnology standard for coffee processing inHonduras". Paper presented at the 2002 AAEA Annual Meetings, LongBeach, CA. IC0 (International Coffee Organization). At Morris, S., 0.Neidecker-Gonzales, C. Carletto, M.Munguia, J.M. Medina and Q.Wodon (2002): "Hurricane Mitchand the livelihoods of the ruralpoor inHonduras", World Development, 30(1), ~ ~ 4 9 - 6 0 . Partners (Partners o fthe Americas, Washington, DC) (2002): "Small scale coffee production in Honduras". At Varangis, P., P. Siegel, D.Giovannucci and B.Lewin (2003): "Dealing with the coffee crisis in Central America: Impacts and strategies", Policy ResearchWorking Paper, No. 2993, World Bank. World Bank (2001): Honduras: Poverty Diagnostic 2000. ReportNo. 20531-HO, World Bank, Washington, DC. 164 Appendix A: Program and Survey Design The evaluation experiment was conducted in 70 municipalities in the west o f Honduras with a total population o f 660,000 in 2001 [Instituto Nacional de Estadistica, 20021. The municipalities were selected because they had the highest prevalence o f malnutrition in the country according to a height census o f first-grade primary school children conducted in 1997 [Government o f Honduras, 19971. For some benefits (see below), eligibility was restricted to households whose residence in a particular municipality had been recorded in a special census of the area conducted inmid-2000. Two packages o f conditional cash transfer interventions were planned and implemented inthis areabythe program. The first, whichwe termthe health and nutritionpackage,was a cash transfer paid to pregnant women and women looking after children less than three years o f age in households o f established residence. Each eligible household received up to a maximum o f two freely exchangeable vouchers worth 55 Lempiras per beneficiary per month. The second package, the educationalpackage,was targeted to households with children 6 to 12years o f age (inclusive) enrolled inprimary school in grades one through four. These receivedup to three vouchers worth 80 Lempirasper beneficiary per month for ten months o f the year. The vouchers were distributed on three occasions between the baseline and post- intervention surveys reported here: in November o f 2000, and in May/June and October/November o f 2001. A fourth round o f voucher distribution coincided with the post- intervention survey in 2002. The payments were conditional in that all beneficiaries were informed that their payments would be suspended if they did not keep up to date with routine ante-natal care and preventive health care for children under three, and the school enrolment and minimumattendance requirement (85% o fclasses). The total number o f trial municipalities was determined by the budget available to the program. Because the entire population o f each municipality could be receiving program benefits, it was necessary to restrict data collection activities to a representative sample o f households in each municipality. Sample size calculations took into account the cluster-randomization, and were based on an ex-post comparison o f 20 intervention and 20 control municipalities, with 80% power to detect a significant difference (P=0.05, two-sided). We used the formula presented by Murray [1998; 368-91 for group-randomized trials with repeat observations of groups. The final size o f the evaluation cohort was eighty households per municipality. The representativeness o f the evaluation cohort at the municipality level at baseline was ensured by: (i) randomly sampling eight census enumeration areas in each municipality with probability proportional to size; (ii) mapping all the dwellings in the enumeration area and numbering them consecutively; (ii) choosing a random start-point, and conducting interviews in ten consecutive inhabited dwellings following the direction o f the numbering on the map. The same households were interviewed inthe post-intervention survey. Women and young children in these households who had moved inthe intervening period were followed up intheir new homes (referred to as `derived' households), provided these were located in one o f the seventy trial municipalities or an adjacent one. Each o f the seventy municipalities was randomly assigned to one o f four groups: (a) the household-level demand-side package alone, (b) the supply-side package, (c) both packages, and (d) control group. Before randomization, the municipalities were stratified into five groups of fourteen on the basis of the prevalence o f stunting reported in the 1997 school height census 165 [Government o f Honduras, 19971. Within each stratum, municipalities were randomly allocated to the various evaluation groups. The randomization was carried out by children in the presence o f legal authorities and representatives o f Honduras' agency for administrative probity. The aperture o f the box was sufficiently small that once the child had insertedhisher arm, it was impossible for himher to see the colored balls. Fromthe day o f the randomizationonwards, there was no attempt to conceal the allocation. Because o f the 2000 residence requirement, no household could become eligible for the cashtransfers by movinghouse after randomization. 166 Appendix Table 1: Tobit estimatesof coffee related labor supplyby coffee growers. Hours Hours supplied by supplied by persons I 4 children aged Explanatory Variables: and older 6 to under I 4 Area under coffee (manzanas) 0.215** 0.926** (0.048) (0.304) Area under coffee squared -0.003** -0.025+ (0.001) (0.0 13) Dummy=l ifcredit constrained -0.456 0.243 (1-08) (2.682) Eligible for CCT 0.955+ -2.274 (0.567) -21.153** (2.414) Eligiblefor CCTX credit constrained dummy 1.394 (1.51) (2.693) Household population under 6 -0.158 0.131 (0.186) (0.584) Householdpopulation6 to under 10 0.191 1.679* (0.250) (0.724) Household population 10to under 14 -0.190 1.934** (0.220) (0.466) Household population 14to under 18 0.389+ 1.088+ (0.230) (0.523) Householdpopulation 18 to under 40 0.989** -0.390 (0.166) (0.544) Householdpopulation40 to under 60 0.646* * -0.079 (0.274) (0.666) Householdpopulation60 and above -0.069 -2.188* (0.307) (0.841) Dummy=l ifhouseholdheadhas incompleteprimary 0.145 -0.025 (0.418) (1.022) Dummy=lifhouseholdheadcompletedprimary -0.292 -3.785* (0.497) (1.778) Dummy=l ifhouseholdheadhasincomplete secondary -2.146* -24.306** (1.088) (2.695) Dummy=l ifhousehold head completedsecondary -2.058 -24.471** -21.273** (1.470) -22.603** (1.948) Dummy1ifhhheadhas some post secondary schooling (0.978) (1.709) Elevation inmeters above sea level 0.015* -0.004 (0.006) (0.131) Elevation inmeters above sea level squared -o.ooo* 0.000 (0.000) (0.000) Constant -15.086** -34.993** (4.18) (8.996) Observations 1756 1756 Note: Dependent variable i s log o f (labor hours plus one). Standard errors inparentheses: + significant at 10%;*significant at 5%; ** significant at 1%. Regressions also include municipality dummies as explanatory variables. 167 T h e coffee crisis: a short note on Guatemalan farmers RenosVakis The World Bank Washington D.C. Introduction The worldwide structural change o f the coffee industry i s seriously affecting Guatemala. Coffee has always played an important role for the Guatemalan economy. It i s the most important export o f the country with receipts o f more than $570 million in 2000 (20 percent o f total export earnings).83In fact, Guatemala i s the fifth largest coffee exporter in the world. The coffee sector provides bothpermanent and temporary employment to thousands o f people, many o f them poor. Nonetheless, the recent entry in coffee production from a number o f countries (particularly Vietnam), as well as above average yields in some Latin American countries (such as in Brazil) have severely depressed coffee prices, resulting in significantly lower revenues for coffee producers in Guatemala. Thisnote explores cross sectional data to evaluate how coffee farmhouseholds inGuatemala may have been affected by the crisis and what mechanisms they have employed to mitigate its impact. Despite the fact the data are static innature, information on shocks experienced by the households as well as coping strategies employed to mitigate their effect i s used to partially address the question o f the impact o f the crisis on coffee farm households. T o summarize the findings, the data reveal that coffee farmers inthe ENCOVI2000: 0 comprise about 10 percent o f the rural population; 0 are mainly small-scale coffee producers; 0 are significantly poorer compared to their regional non-coffee counterparts; 0 during2000, they were more likely to experience adverse shocks, resulting inincome and assets loses; 0 used a number o f coping strategies to mitigate the impact o f these shocks including increases inlabor supply, depleting savings and decreases inconsumption patterns. The next session briefly describes the data and coffee definition used in the note, followed by a profile o f coffee-farmers, and an overview o f welfare trends and income portfolios. An examination o f coping strategies employed by coffee farmers i s presented in the next section, while the last session concludes. 83World Development Indicators(2001). 168 The ENCOVI 2000, sample structure and coffeefarmers The main source of quantitative information used in this work is the Living Standards Measurement Survey (ENCOVI-2000). The ENCOVI, executed by the Guatemalan National Institute of Statistics (INE), covers a sample of about 7,300 households and is statistically representative at the national level and for a number of strata including: (a) urbanand rural areas; (b) eight regions(andurbanand rural areas inthese regions). While the survey allows identification o f coffee farm households, it does not permit identification o f coffee laborers. Unfortunately, there was no question that distinguishes wage earners in the coffee sector. Nonetheless, the self-employment section includes information on crop production and as such the identification o f farm households that produce coffee i s possible. It i s important to note that because the ENCOVI i s a household survey, plantations owned by entities other than households (such as corporations or banks) are not captured in the data collected in the survey. As such, the estimatespresentedhere are not representative for all coffeeproducers. The ENCOVI 2000 data reveal that seven percent o f all households produce coffee (Table 1). This corresponds to about 160,000 farm households that receive income from coffee production. Distinguishingbetweenurban and rural areas, two percent o furbanhouseholds are coffee farmers while 11 percent o f rural households produce coffee (Figure l).84 Since it i s very likely that welfare differences exist between urban and rural households, we report the findings separately for the two groups. Table 1: Samples structure:householdcoffee farmers. Urban Rural All Number o fhouseholds producing coffee 21,171 138,424 159,595 % o fhouseholds producingcoffee - 2 11 7 Sample size 123 467 590 Source: World Bank calculations using ENCOVI2000, Instituto Nacional de Estadistica - Guatemala. 84It is important to note that, becausethe ENCOVIis a household survey, plantations owned by entities other than households (such as corporations or banks) are not captured inthe data collected inthe survey. As such, the estimates presentedhere are not representative for all coffee producers. 169 Figure1:Samplestructure: coffeefarmers and non-coffeehouseholds. Urban Rural 2% 11% mCoffee Farmers Non-coffee coffee Farmers ENon-coffee In addition, the regional distribution of coffee farmers suggests that there is a strong concentration interms o f coffee production. Specifically, almost 80 percent o f the coffee farmers in the survey reside in 3 regions: Norte, Suroriente and Noroccidente (Table 2). Infact, coffee farmers in each o f these regions comprise more than 20 and in the case o f Norte more than 30 percent o f the household populationwithin each o f these regions, indicating the strong geographic concentration o f coffee production. Table 2: Samples structure: coffee-farmhouseholds(YO),by region. Across Within Poverty rate (%) Metropolitana 3 0.1 18.0 Norte 30 30 84.0 Nororiente 2 2 51.8 Suroriente 23 19 68.6 Central 7 5 51.7 Suroccidente 9 3 64.0 Noroccidente 26 17 82.1 Peten 0.1 0.4 68.0 Total 100 56.2 Source: World Bank calculations using ENCOVI 2000, Instituto Nacional de Estadistica- Guatemala. 170 Who are the coffee farmers There are a number o f differences between urban and rural coffee farmers. For example, rural coffee farmers have larger families, are more likely to be indigenous and have significantly lower levels o f education compared to their urban counterparts (Table 3). Inaddition, rural farmers own less land than urban coffee farmers, and are less likely to rent additional land (Table 4). Still, most coffee farmers inthe survey are small-scale landowning households irrespective of the area o f residence. More than 88 percent o f rural and 96 percent o f urban coffee farmers own land (Table 4). For rural coffee farm households, the average land holdings are 2.7 hectares (3.1 among landowners). This compares with 3.6 hectares (3.8 among landowners) for urban coffee farmers. Table 3: Household characteristics. Urban Rural Non- Coffee Non- Coffee coffee coffee Household composition # o f children (<14 years) 1.8 1.9 2.7 2.9 # o f adults (15-60 years) 2.5 2.9 2.6 2.8 # o f adults (>60 years) 0.3 0.5 0.4 0.4 Household size (number) 4.6 5.3 5.7 6.1 Indigenous("3) 25 41 48 59 Household head is male (%) 77 87 84 92 Household head education (years) 6.5 5.7 2.3 1.6 Migrant inthe household(yes=l) 9 13 12 12 Household members would like to work more 68 69 76 82 (%) Source: World Bank calculations using ENCOVI2000, Instituto Nacional de Estadistica- Guatemala. Table 4: Land use. Urban Rural Non- Coffee Non- Coffee coffee coffee Land ownership (%) 10 96 46 88 Tenants 4 4 19 12 Average land owned (hectares) 0.2 3.6 1.3 2.7 Average land owned among landowners 1.6 3.8 2.8 3.1 (hectares) Average land rented in (hectares) 0.1 1.1 0.4 0.4 Average land rented in among renters 0.7 3.3 1.o 0.8 (hectares) Source: World Bank calculations using ENCOVI 2000, Instituto Nacional de Estadistica- Guatemala. Perhaps the most striking difference between urban and rural households is the fact that while urban coffee farm households are relatively not poor, rural coffee farmers have high poverty incidences. Specifically, only one third o f coffee farmers in urban areas are classified as poor (Table 5). This compares with more than 77 percent among rural coffee farmers. Interestingly, both poverty rates are significantly higher compared to the average poverty rates for rural or urban areas respectively. This i s also true for regional poverty rates comparisons, where the 3 171 regions where 80 percent of the coffee farmers in the survey reside have higher poverty incidences than any other region (Table 2). Welfare comparisons: coffee farmers versus non-coffee households Poverty and consumption As discussed above, poverty rates among rural coffee farmers in 2000 were higher than urban coffee farm households. Still, compared with their regional counterparts, coffee farm households had higher poverty rates than non-coffee households. For example, the headcount ratio in urban areas for coffee farm households i s 35 percent compared to only 20 for non-coffee households (Table 5 and Figure 2). In addition, comparing households in rural areas (where poverty i s extensive), coffee farm households have a poverty rate of 77 percent while non-coffee of 64 percent. Similar patterns emerge usingextreme poverty indicators. Table 5: Householdwelfare indicators, by coffee. Urban Rural Non- Coffee Non- Coffee coffee coffee Poverty rate (%) 19.6 35.3 64.1 77.3 Extreme poverty rate (%) 1.6 2.2 17.2 22.4 Annual consumptionper capita (in 12,194 9,467 4,459 3,584 auetzales) Overall ' 27.1 74.5 Source: World Bank calculations using ENCOVI 2000, Instituto Nacionalde Estadistica- Guatemala. Non-coffee 1 Coffee Non-coffee I Coffee Urban Rural 172 Reflecting the poverty measures above, consumption per capita patterns also show similar differences. For example, consumption per capita inrural areas i s more than 30 percent higher for non-coffee households (Table 5). Inurban households consumption per capita among coffee farm households was 49,500 compared to more than 412,100 among non-coffee households. Incomes and income portfolios As the previous patterns suggest, coffee farmers inthe ENCOVI 2000 were poorer compared to their regional-specific non-coffee counterparts. As expected, income trends suggest similar patterns. Specifically, incomes per capita among coffee farmers in both urban and rural areas were about 10 percent lower compared to incomes o f non-coffee households in their respective area (Table 6). Table 6: Income portfolios. Urban Rural Non- Coffee Non- Coffee coffee coffee Income sources (%): Agricultural wages 2 3 15 10 Agriculturalfarming 1 23 13 43 Non-farm salaries 50 24 29 14 Non-farm self-employment 19 16 17 11 Non-labor a 27 34 26 22 Total 100 100 100 100 Total annual income per capita (in 11,685 10,964 3,733 3,318 quetzales) aThis includes returns to capital, private andpublic transfers as well as pensions. Source: World Bank calculations using ENCOVI2000, Instituto Nacional de Estadistica -Guatemala. Still, examination o f the income portfolios for the various types o f households reveals a number o f interesting insights. First, rural coffee farm households heavily depend on agricultural incomes. In particular, more than half o f the income i s derived from agricultural wages or farm activities (Table 6 and Figure 3). In fact, coffee sales among rural coffee farmers comprise more than 23 percent o f total income (Table 7). Table 7: Coffee sales. Urban Rural Average coffee sales (Quetzales) a 15,430 4,525 Share o f coffee sales to total household 23.4 23.3 incomea a For coffee producers only. Coffee sales does not refers to net revenues but only coffee receipts (coffee price times quantity sold). Source: World Bank calculations using ENCOVI2000, InstitutoNacional de Estadistica- Guatemala. 173 Figure3: Incomeportfolios. 100% 80% 60% 40% 20% 0% Non-coffee 1 Coffee Non-coffee 1 Coffee Urban Rural 1 0 Agricultural wages a Agricultural farming 0 Non-farm salaries 0 Non-farm self-employment a Non-laborI Second, urban coffee farmers are less dependent on agricultural income. Only about a quarter of their income i s derived from agricultural activities (Table 6). Instead, more than 40 percent o f their income comes from non-agricultural salaried employment or self-employment. Still, coffee sales among urban farmers also represent about 23 percent total household income (Table 7). Finally, non-coffee households in both areas depend significantly more on non-agricultural incomes. As expected, while rural households are diversified in both agricultural and non- agricultural activities, urban non-coffee households derive practically no income form agricultural employment (Table 6). Insightsfromthe shocks module As the previous sections suggest, coffee farmers are poorer compared to regional non-coffee counterparts, and are more likely to have significantly lower socio-economic indicators. Understanding the extent by which the observed patterns are due to the coffee crisis shock or other differences i s important. While the data i s static innature, the survey did collect information related to various shocks affecting households and available coping mechanisms employed. This information i s used here to assess at least partially how the coffee crisis may have affected coffee farmers. Three shocks that can be related to "coffee" can be distinguished: (i)whether a household member has lost a job; (ii) whether the household has experienced income losses; (iii) whether the price o f a product producedby the householdhas decreased. While none o f these are specific to coffee farmers, it i s expected that these are the types o f shocks that would be associated with exposure to the coffee crisis. As such, comparing the incidence o f these types o f shocks among coffee farmers as well as non-coffee households i s useful. 174 Overall, coffee farmers are significantly more likely to experience any o f the three shocks discussed above. In particular, 33 and 16 percent o f urban and rural coffee farmers respectively have been affected by one of the shocks (Table 8 and Figure 3). This compares with 26 and 12 percent for the non-coffee counterparts. Still, experiencing product price decreases i s by far more widespread among coffee farmers compared to non-coffee households. By contrast, the incidences of loss o f income or employment are relatively similar between coffee farmers and non-coffee households. As such, these trends suggest that while shocks are not uncommon among all households in both rural and urban areas, coffee farmers have been affected more from product price decreases compared to non-coffee households, which i s expected. Table 8: Shocksincidences. Urban Rural Non- Coffee Non- Coffee coffee coffee Household experiencing (%) Loss of employment 14 15 4 3 Loss of income 17 15 5 8 Productprice decrease 4 14 4 7 Any of the shocks above 26 33 12 16 Among those who suffered any of the shocks above: Shock resulted in (%): Income loss 93 100 84 91 Asset loss 2 0 1 2 Both income and assets losses 2 0 12 6 No loses 3 0 3 1 Total 100 100 100 100 Main coping mechanisms to address shocks Did nothing 24 20 27 29 Increased hours worked 24 19 21 15 Decrease in consumption 22 19 26 33 Usedsavings 8 17 8 10 Borrowed 5 4 3 3 Sold assets (animals, land,property) 1 0 4 3 Other 15 21 11 7 Total 100 100 100 100 Have the losses being completely recovered 23 9 22 11 (yes=l) When doyou anticipate tofully recover from shock: Less than 6 months 9 0 9 2 Between 6-12 months 8 2 6 7 More than a year 8 48 8 10 Not sure 75 50 77 81 Total 100 100 100 100 Source: World Bank calculations using ENCOVI 2000, Instituto Nacional de Estadistica- Guatemala. 175 Figure 4: Shocks incidences. Coffee -m Non-coffee I I I I 0 10 20 30 40 Loss of employment Loss of income Product price decrease Any of these shocks The najority of the households affected by these shocks also experienced income and asset declines. For example, among rural coffee farmers affected by any o f these shocks, only one percent reported not been affected in terms o f income or asset declines (Table 8). Still, experiencingincome losses was the most frequent impact on householdwelfare. From a policy perspective, exploring what coping mechanisms are employed by affected households i s important to understand gaps and potential interventions that can be implemented. Interms of coffee farm households affectedby a shock, the main coping mechanisms to mitigate the adverse impact were increased labor supply (15 percent for rural coffee farmers and 19 percent for urban), decreasing household consumption (more than 33 percent for rural coffee farmers) and used own savings (ten percent for rural and 17 percent for urban coffee farmers). Still, 29 percent o f rural coffee farmers did nothing to mitigate the shocks, implying that there may be lack o f coping instruments (Table 8). Nonetheless, similar patterns emerge for non-coffee households in terms o f available coping mechanisms. For example, about a fifth o f non-coffee households affected by a shock worked more and about ten percent used savings. Still, coffee farmers are more likely to decrease consumption compared to non-coffee households. These patterns are not conclusive, but the seem to suggest that while shocks do affect a number o f heterogeneous households in both rural and urban areas, coffee farmers may have had fewer coping mechanisms compared to non-coffee households, which could partially explain the lower welfare indicators discussed earlier. Finally, the shock module also included questions as to whether households had recovered from the adverse effect of the shocks and ifnot how long would it take to do so. The majority o f coffee farmers (90 percent for both urban andrural) hadnot recovered from shock (Table 8). Inaddition, while rural coffee farmers are unsure about the time it will take to recover, urban coffee farmers expect it will take more than a year. These patterns suggest a highlevel o f uncertainty about the ability o f these households to regaintheir original welfare levels. 176 Discussion This note highlighted the situation for small scale coffee farmers inGuatemala. To the extent that the static nature o f the data limits inferences about the potential impact o f the coffee crisis on affected households, the analysis from the ENCOVI 2000 indicates that the coffee farmers may have been affected by the coffee crisis households in several dimensions. Inparticular, coffee farmers seem to have experienced declines in incomes and consumption as well as increases in the incidence o f poverty. In addition, coffee farm households were more likely to experience shocks that adversely affectedincomes and assets compared to non-coffee households. Inorder to cope with the shocks, coffee farmers increasedlabor supply, decreased consumption and to a lesser extent have used up savings. Such coping strategies may be harmful to the extent that they may imply taking children out o f school or deterioration o f the nutritional intake o f householdmembers, especially for children. The fact that at least 10 percent o f all rural households are involved in coffee production, and since more than 80 percent o f the coffee production i s concentrated in 3 regions suggests that the overall impacts o f the coffee crisis may be even greater, especially if one takes into account the effect of the crisis on households whose income depends on wage employment in the coffee industry. According to ANACAFE, an estimated 200,000 people are permanently employed in the coffee industry. This figure increases to more than 500,000 duringthe coffee harvest. Most laborers Cjornaleros) in the coffee sector are seasonal migrants from poor households that depend on the coffee sector to augment their incomes. While exploring coffee laborers using this data i s not possible, an estimated 40,000 jobs related to coffee production were expected to be lost in 2002 (ANACAFE puts this figure at 60,000).85 As most o f these jobs are expected to be low-end jobs, the effect on the poor i s likely to be greater. As such, to further understand the interactionand potential impact o f the crisis, panel data will be required in order to adequately assess changes in welfare indicators, examine household responses and strategies to cope with the crisis, explore the impacts on human capital outcomes as well as assess the government's response to the crisis. 85MinistryofAgriculture. 177