Labor market equilibrium and conditional cash transfers Experimental evidence from Latin America

Transcription

1 Labor market equilibrium and conditional cash transfers Experimental evidence from Latin America Work in progress Comments welcome * This version: July María Laura Alzúa CEDLAS-Universidad Nacional de La Plata ** Guillermo Cruces CEDLAS-Universidad Nacional de La Plata and CONICET Laura Ripani Inter-American Development Bank Abstract: The rapid expansion of conditional cash transfers (CCT) programs throughout Latin America has spurred a wealth of empirical studies, but their impact on labor markets has only been sketchily covered in this literature. This is an important issue since potential general equilibrium effects could have consequences for the interpretation of their impact on the programs intended outcomes. This document analyzes the effect of CCTs on the labor supply of adults and remuneration levels in a series of countries in the region, using comparable results derived from homogeneous datasets and estimation methodologies. The document builds on the experimental evaluations of three programs implemented in rural areas: Mexico s PROGRESA, Nicaragua s Red de Protección Social (RPS) and Honduras Programa de Asignación Familiar (PRAF). Despite noteworthy results for some subsamples, none of the programs seem to have induced a major shift in the labor force participation or the hours worked among adults. However, there are strong positive effects of PROGRESA on wages and labor income in the short and medium term, for both eligible and ineligible households. This evidence suggests that while CCT programs in poor rural areas do not imply a significant disincentive to work, they might affect labor market equilibria. These indirect effects on labor income might in turn have an impact on the programs intended outcomes, which might be confounded with the direct impact of transfers and conditionalities on which most of the previous literature concentrates. The evidence presented in this paper might have relevant implications for future program and evaluation designs, and for the interpretation of previous findings. * Corresponding authors: malzua@cedlas.org, gcruces@cedlas.org and laurari@iadb.org. This paper originates in a section of the report Labor supply responses to conditional cash transfer programs. Experimental and non-experimental evidence from Latin America. The full report was prepared for the IDB by a team directed by María Laura Alzúa, Guillermo Cruces, Leonardo Gasparini and Laura Ripani. The team members, Javier Alejo, Andrés Ham, Sergio Olivieri and Mariana Viollaz, from CEDLAS-UNLP, provided outstanding research assistance in different stages of the report. The authors wish to thank Santiago Levy for encouraging this study, and Emanuel Skoufias for providing an early draft of his ongoing work. Comments by Laura Guardia, Julia Johannsen, Claudia Piras, Patrick Puhani, Graciana Rucci and Yuri Soares are much appreciated, as well as discussions with Sebastián Galiani, Fabio Veras Soares and Martín Tetaz. We also appreciate the comments received from Felipe Barrera, Sami Berlinski, César Bouillon, Santiago Levy, Florencia López Boo and other attendants to the 2008 LACEA Meetings in Rio de Janeiro, and those from Pablo Ibarrarán, Miguel Jaramillo and other participants in the AfrEA-NONIE-3ie Conference in Cairo, March The usual disclaimer applies. The opinions in this report correspond to the authors, and do not necessarily represent those of the institutions to which they belong. ** CEDLAS-UNLP. Centro de Estudios Distributivos, Laborales y Sociales, Facultad de Ciencias Económicas, Universidad Nacional de La Plata. Calle 6 entre 47 y 48, 5to. piso, oficina 516, (1900) La Plata, Argentina. Phone: +54-(221) cedlas@depeco.econo.unlp.edu.ar Website:

2 Index 1 INTRODUCTION LABOR SUPPLY AND CASH TRANSFERS MOTIVATION AND THEORETICAL CONSIDERATIONS POTENTIAL IMPACT OF CCTS ON LABOR MARKETS THE IMPACT OF CCTS ON LABOR MARKETS: PREVIOUS FINDINGS FOR LATIN AMERICA CONDITIONAL CASH TRANSFERS AND LABOR SUPPLY OF ADULTS: EVIDENCE FROM EXPERIMENTAL EVALUATIONS THE PROGRAMS AND THEIR EVALUATIONS ESTIMATING PROGRAM EFFECTS WITH EXPERIMENTAL DATA LABOR SUPPLY OF ADULTS IN CCT PROGRAMS WITH EXPERIMENTAL EVALUATION Descriptive statistics and the random assignment processes The effect of CCTs on labor supply of adults The effect of PROGRESA on wages and labor income CONCLUSIONS REFERENCES APPENDIX Mexico: PROGRESA Program Honduras: PRAF Program Nicaragua: RPS Program APPENDIX 2: WORKING DEFINITION OF ELIGIBLE HOUSEHOLDS IN PRAF, PROGRESA AND RPS

3 1 Introduction Conditional cash transfer programs (CCTs) have been hailed as a new paradigm in social policy in Latin America, and their influence has spread beyond the region. The innovation stems from an original combination of different aspects of social programs that were already present in previous experiences. As their name indicates, CCTs transfer resources to targeted households subject to a series of conditions. The conditionality of benefits existed in previous initiatives, for instance, in workfare programs. However, CCTs manage to combine short term objectives of poverty alleviation with long term goals, related to the accumulation of human capital, by means of subjecting benefit receipt to a series of verifiable conditions, such as school attendance, vaccination, and regular medical checkups, among others. Fiszbein et al. (2009, Chapter 2) review the economic rationale for this type of programs. The popularity of this type of initiatives can be partially traced to the implementation of an original and well-executed experimental impact evaluation methodology for the PROGRESA program (Programa de Educación, Salud y Alimentación renamed Oportunidades) in rural Mexico in the second half of the 1990s (see Behrman and Todd, 1999, Skoufias and Parker, 2001, and Gertler, 2004, among many others). The positive impacts of this program on the targeted outcomes, obtained through a reliable methodology, helped the dissemination of CCTs and of experimental evaluation methods. CCTs and similar programs were deployed in several other countries, both in Latin America (for instance, in Nicaragua, Honduras, Brazil, Argentina, Paraguay, Uruguay, Ecuador) and in the rest of the world (Mozambique, Turkey, the city of New York, and Yemen, among others). The rapid expansion of CCTs throughout developing countries has spurred a wealth of empirical studies. There is a vast literature on the impact of CCTs on the programs main target outcomes: educational and health results, nutrition, household income, and child labor (see the reviews of Rawlings and Rubio, 2003 and 2005, and Bouillon and Tejerina, 2006, for Latin America, and Fiszbein et al., 2009, for a global perspective). However, the impact of CCTs on labor markets has received less attention in this literature, with notable exceptions such as Parker and Skoufias (2000), Maluccio (2007), Skoufias and Di Maro (2008), and Angelucci and Di Giorgio (2009). The experimental setting in which these programs have been evaluated provides an ideal 3

4 source of identification of labor supply elasticity with respect to income, one of the salient issues in the empirical labor economics literature, for which there is limited evidence from developing countries. Moreover, the impact of the cash transfer component on other unintended outcomes, such as work incentives and wage levels, constitute an important issue. Potential general equilibrium effects of the programs could have consequences for the interpretation of their effects on the intended outcomes. This document analyzes the effect of CCTs on the labor supply of adults and remuneration levels in a series of countries in the region, using comparable results derived from homogeneous datasets and estimation methodologies. The document builds on the experimental evaluations of three programs implemented in rural areas: Mexico s PROGRESA, Nicaragua s Red de Protección Social (RPS) and Honduras Programa de Asignación Familiar (PRAF). Despite noteworthy results for some subsamples, none of the programs seem to have induced a major shift in the labor force participation or the hours worked among adults. However, there are strong positive effects of PROGRESA on wages and labor income in the short and medium term, for both eligible and ineligible households. This evidence suggests that while CCT programs in poor rural areas do not imply a significant disincentive to work, they might affect labor market equilibria. These indirect effects on labor income might in turn have an impact on the programs intended outcomes, which might be confounded with the direct impact of transfers and conditionalities on which most of the previous literature concentrates. This evidence is related to Angelucci and Di Giorgio s (2009) findings on indirect effects of PROGRESA on the consumption of ineligible households. The evidence presented in this paper might have relevant implications for future program and evaluation designs, and for the interpretation of previous findings. The document is organized as follows. Section 2 discusses the theoretical underpinnings of the impact of cash transfers on labor supply, and presents a review of the empirical literature for countries in Latin America. Section 3 discusses the evidence from three programs: PROGRESA from Mexico, Red de Protección Social (RPS) from Nicaragua and Programa de Asignación Familiar (PRAF) from Honduras. Conclusions follow. 4

5 2 Labor supply and cash transfers 2.1 Motivation and theoretical considerations The spread of conditional cash transfer programs (CCTs) throughout Latin America have made them a permanent aspect of social policy in the developing world for the foreseeable future. Moreover, the constantly growing literature on their evaluation has created a wealth of knowledge of their impact on the main target outcomes: educational and health results, nutrition, poverty reduction, and child labor. According to the results from a number of evaluations, cash transfers, especially when combined with conditionalities, have proved successful in increasing welfare and human capital accumulation in recipient households, and lowering child labor (see the review by Rawlings and Rubio, 2003 and 2005). The policy question that motivates this document is only indirectly linked to the intended outcomes of these programs. Besides the explicit objectives of the interventions, it is also important to establish whether the programs have had other relevant but unintended consequences. This study attempts to establish whether there are any incentive effects of the cash transfers (and their associated conditionalities) on the labor supply of adults in recipient households, the effect on market wages and/or on ineligible individuals. These are relevant questions that have dominated the discussion of welfare, workfare and related programs in developed countries (Meghir and Phillips, 2008). Moreover, welfare policy changes have provided plausible sources of identification for the impact of different variables on labor supply decisions (see the discussion in Blundell and MaCurdy, 1999). The evaluation strategy of some CCTs presents some advantages over the case studies of welfare policies in developed countries. The empirical work on the latter are usually based on policy experiments which provide plausible sources of identification, but concerns always remain about the true exogeneity of the observed changes, and thus on the validity of the constructed control groups. The experimental evaluation of some CCTs, where policy treatments were randomly assigned, provides a unique source of identification of labor supply elasticities. The results obtained could thus be appealing not only for the development field but for the labor economics literature at large. 5

6 Welfare policy, work incentives, and their combination, salient issues in labor economics, have received less attention in the development literature. This line of research has potentially relevant payoffs for our understanding of labor markets in general, and especially in how they interact with the development process. Labor supply decisions are crucial to comprehend the income generation process of the poor and to assess potential poverty traps. Moreover, these studies can shed light on labor markets in developing countries: households make decisions amidst a complex setting characterized by unemployment, labor informality, migration, cultural norms and high transaction costs in general. As in developed economies, social protection programs and income transfers may affect the labor supply decisions of households. Finally, research on these issues is fundamental for a better understanding of the current programs, and for those that will be implemented, replicated or scaled-up. It can also inform the policy formulation process to anticipate and deal with unintended consequences of CCTs. The relevance of this research and its potential to unveil complex interactions is apparent in the existing literature. Ardington, Case and Hosegood (2009), for instance, find that changes in old-age pensions in South Africa increased the labor supply of other adults in the household, by loosening the constraints faced by potential migrant workers. These findings constitute a good example of what can be learned in terms of welfare policies and labor supply decisions in the multifaceted context of labor markets in developing countries. Regarding CCTs in Latin America, the results presented by Parker and Skoufias (2000) and Skoufias and Di Maro (2008) on the effects of Mexico s PROGRESA on adult labor supply are relevant for the design of future initiatives in the region. The potential impact of transfers on work decisions is especially relevant in countries like Brazil, Colombia and Mexico, where programs which originally had limited coverage have been scaled up (for instance, from rural to urban areas) and have gradually increased the amounts of the cash transfers. The estimates presented in this document can provide some guidance on the potential scope of unintended labor supply effects among adults, which is especially relevant given the dissemination of programs in the region and their increasing coverage within countries. Moreover, besides the existing evidence on the impact of CCTs on child labor, the empirical results presented here will also shed some light on which members (if any at all) change their labor supply decisions in response to the programs. 6

7 The following pages discuss the rationale and the plausibility of labor supply impacts of CCT programs, and review the available evidence for the region. 2.2 Potential impact of CCTs on labor markets As briefly described in the introduction, CCT programs combine short term poverty alleviation (through cash transfers) with long term objectives through incentives to human capital accumulation (school attendance, health check-ups, improved nutrition, and reduction of child labor). Within this common framework, each program differs in the subsidy level per child or household, in the duration of the benefits, in the age profile of the beneficiaries, and on the specific conditionalities implemented, among other aspects. With the exception of minor training components in some programs, the overall design for CCTs in Latin America and the Caribbean is not directly related to adult employment in beneficiary households there are no work requirements or restrictions, and households can earn labor income without losing the benefit. 1 New interventions are incorporating aspects related to the job market for instance, one of the components of Opportunity New York City, a CCT recently implemented by the City Government of New York, is designed to improve workforce participation, requiring adults to work at least 35 hours a week to receive the subsidy. The lack of work requirements does not mean, however, that the programs are neutral in terms of adult labor supply and participation incentives. Since they modify the income of the household and, potentially, the allocation of time through their impact on children s work and school enrollment, there is a potential for effects on the labor supply of adults in beneficiary households. Economic theory suggests several ways in which CCTs can affect work decisions within recipient households. The simplest way to think about these issues is through a standard static model of labor supply, representing the choice between consumption and leisure. In such a model, the components of CCTs might play a role through at least four channels. 1 Some large emergency cash transfer programs, notably Argentina s Heads of Household Program (Programa Jefes de Hogar), are sometimes classified as CCTs with work requirements. However, they constitute in fact poverty alleviation programs targeted through unemployment, given the ambiguity of the work requirement, and the difficulty to provide tasks and to monitor compliance in the context of massive emergency interventions. See Gasparini et al. (2007) for more details on work requirements and their effects in the context of Argentina. 7

8 Firstly, the cash transfer component of the program constitutes an increase in unearned non labor income. As such, it induces a pure income effect, which loosens the budget constraint of the recipient households. 2 In this sense, CCTs constitute a much simpler policy instrument than welfare programs in developed countries: as a pure subsidy, CCTs do not include the taper of the negative income tax credit, nor induce steep replacement rates like in traditional welfare programs. Moreover, the lack of work requirements and employment premiums imply that CCTs lack the complexity of welfare-to-work initiatives such as the US Earned Income Tax Credit (Eissa and Liebman, 1996) or the UK s Working Family s Tax Credit (Meghir and Phillips, 2008). Conditionalities constitute the second channel through which CCTs might induce behavioral responses in the labor supply of adults. The requirements related to children s human capital accumulation might have an impact on the household s allocation of time. While there is evidence that CCTs place the burden of the benefit administration on women in the household (Parker and Skoufias, 2000), the main potential impact on adult s time arises from freeing up time dedicated to childcare, at least in the cases were programs successfully induce an increase in school enrollment, given the negative impact of fertility on mother s labor supply in Latin America (Piras and Ripani, 2005, and Cruces and Galiani, 2007). The third channel originates in the potential fall in household income due to the reduction in the labor supply of children, which reduces the net impact of the cash transfers in households were working children are induced to reduce their participation in work activities. Finally, the fourth channel operates through the effect at the community level beyond eligible households, for example on wages, or Angelucci and Di Giorgio s (2009) results on consumption of ineligible individuals. The combination of these four channels implies that the overall effect of CCTs on adult labor markets is ambiguous from a theoretical point of view. The rise in non labor income might increase or decrease hours of work, depending on whether leisure is a normal or an inferior good. If leisure is assumed to be a normal good (as pointed out by the available evidence see for instance Imbens et al., 2001), we could expect 2 Given the small value of the cash transfers, we ignore the impact of the potential change in labor income labor on the program s eligibility means test. 8

9 a negative impact on labor supply. However, the other two channels introduce further ambiguity on the sign of the overall effect. On the one hand, even if leisure is considered to be a normal good, the presence of fixed hour or money costs (such as commuting or childcare see Cogan, 1981, and Bhattarai and Whalley, 2003) implies that a lump-sum transfer might increase labor supply. Moreover, the positive impact of CCTs on children s school attendance might free time previously spent in childcare, further reducing the cost of work (Baker et al., 2005). On the other hand, the increase in income through benefit receipt might be reduced in net terms by the foregone income from the reduction of child labor, and might even result in a net reduction in total household income. This effect might mitigate any potential disincentive in the labor supply of adults, and might even induce a positive response to compensate for the foregone child labor income. Moreover, as noted above, Ardington et al. s (2009) results from South Africa indicate that transfers might affect even more complex within-household interactions, inducing unexpected labor supply responses. 3 There might also be potential heterogeneity in the effects in the gender and age dimensions, among others. The predictions from standard labor economics models are thus ambiguous: the existence of any impact, and its direction, is ultimately an empirical question. This conclusion is further reinforced by the existing evidence on differential labor supply elasticities by population subgroups (Blundell and MaCurdy, 1999). After a brief review of the impact of the CCTs covered in this study, the empirical analysis in the following sections seeks to establish the presence of labor supply effects for the whole adult population and for the relevant subgroups identified in the literature. 2.3 The impact of CCTs on labor markets: previous findings for Latin America Most of the literature evaluating the impact of CCTs focuses on the program s effects on years of schooling, child labor and health status (mainly morbidity, anemia and z scores). While results vary from country to country, program evaluations show to some degree a positive effect on years of schooling and improvement in some key 3 While for simplicity this discussion of the potential impacts of CCTs on adult labor supply only deals with an individual s decision to work, the results presented below cover joint labor supply decisions at the household level. See the references in Blundell and Macurdy (1999) for more details. 9

10 health indicators. 4 While the available evidence is broadly positive, there are still several unanswered questions, related mainly to the external validity of the evaluation results (in terms of their scalability, their general equilibrium impacts, and the potential for replication under different country conditions), and on the sustainability of the short term welfare impacts. In the following paragraphs we review only the main results of CCTs for the countries considered in this study, highlighting (when available) previous findings on the labor supply of adults. Most of the studies, however, cover school attendance, health and child labor, the main target outcomes of most initiatives. 5 Skoufias and Parker (2001) study the effects of PROGRESA on child labor and schooling, exploiting the initial experimental evaluation setting in which communities were randomly assigned to treatment and evaluation groups. 6 They find a significant reduction in the participation of boys and girls in work activities, with a stronger reduction for boys. These significant reductions in child labor contrast with the lack of impact of PROGRESA on the labor supply of adults in beneficiary households. Parker and Skoufias (2000) and Skoufias and Di Maro (2008) exploit the experimental design of PROGRESA s evaluation to study this issue. Both studies find no significant effects of the program on adult labor force participation, for both eligible and non eligible households in program localities, for men and women, and for most of the age cohorts considered. While hours of work are not studied as an independent variable, the two papers also fail to find any significant impact on leisure time. 7 The empirical results section below compares the estimation strategies in this paper with that of Parker and Skoufias (2000) and Skoufias and Di Maro (2008). Angelucci and Di Giorgi (2009) present some evidence that adult equivalent labor earnings are not affected by the program. In a related study, Skoufias et al. (2008) assess the impact of a food program (Mexico s Programa Apoyo Alimentario, PAAL) on consumption and adult labor 4 For a thorough review of CCTs impact on education and health outcomes see Rawlings and Rubio (2003). 5 Health impacts are beyond the scope of this report and thus not covered in this section. See Gertler (2004) for a review of PROGRESA and further references. 6 The experimental evaluation design of PROGRESA, RPS and PRAF are described in detail in the empirical results section below. 7 These results are in line with that of the food stamp program for the US. See Keane and Moffit (1998) and Fraker and Moffit (1988), and the references therein. 10

11 supply. The study compares the impacts of in-kind and cash transfers on household welfare by means of a random assignment of beneficiaries to either type of benefit. The PAAL program imposes some conditionalities in some of the treated groups, mainly related to school attendance. The main results show positive impacts of the transfer (irrespective of type) on consumption, but no significant effect on the labor supply of adults in beneficiary households. However, Skoufias et al. (2008) also found that the program induces households to move away from agricultural activities. According to the authors, this finding supports the view that agricultural production is intended to guarantee food in the event of negative income shocks, and that the PAAL program works as an insurance mechanism, allowing individuals to allocate less time to agricultural activities. The impact of Nicaragua s Red de Protección Social (RPS) was analyzed in depth by Maluccio and Flores (2005). The program is a CCT modeled broadly in the line of PROGRESA, and the initial evaluation strategy also included the random assignment of communities into treatment and control groups. According to the results from this study, RPS had a negative effect on child work: for all the age groups considered, the percentage of children working was lower after participation, with reductions ranging from -4.6 to -5.6 percentage points. Moreover, this effect on child labor was found to be twice as large for boys than for girls. The potential impact of RPS on the labor supply of adults is analyzed by Maluccio (2007). Unlike the case of the two previous studies of PROGRESA, Maluccio (2007) concentrates on total hours of work at the household level rather than on individual labor force participation. The results indicate a small but significant effect of the program on the labor supply of adults in beneficiary households as measured by total hours of work per household, with most of the negative impact on time spent in agricultural activities. 8 The results in this document complement Maluccio s (2007) study by presenting further estimates of the impact of RPS on labor force participation and hours of work at the individual level. 8 Maluccio s (2007) empirical strategy is discussed in the empirical results section below. 11

12 3 Conditional cash transfers and labor supply of adults: Evidence from experimental evaluations This section studies the effect of CCT programs on labor outcomes of adults using experimental evaluation data from three interventions: PRAF (Honduras), PROGRESA (Mexico), and RPS (Nicaragua). While some previous work on this issue exists, at least for PROGRESA (Parker and Skoufias, 2000; Skoufias and Di Maro, 2008) and RPS (Maluccio, 2007), this document presents comparable results for the three programs based on a common estimation strategy and in common criteria for the definition of the relevant variables. We start by briefly outlining each program s experimental evaluation setting. We then discuss alternatives for the estimation of treatment effects in this context. Finally, we describe the treatment and control samples, and we present difference in difference results of the impact of the programs on a series of labor supply indicators. The results are presented by gender and by household type to capture possible heterogeneities in treatment effects. 3.1 The programs and their evaluations The data used in this section of the document corresponds to ad hoc surveys carried out to evaluate each specific program. The three data sources were harmonized following common criteria to achieve maximum comparability, using the methodology described in CEDLAS (2009). The structure of each program can be seen in a detailed appendix at the end of the paper. The three interventions covered in this section were targeted at rural areas in poor regions of the respective countries. The following paragraphs briefly describe the three programs evaluation strategies. 9 All the programs had an experimental design and they were randomized at the village level. In 1997, Mexico began implementing the first phase of PROGRESA. It was geographically targeted by locality, based on a poverty index. From an initial group of 506 localities selected for the first round, 320 were randomly selected to participate in 9 Besides the documents referenced in the previous section, detailed descriptions of the programs at the time of the evaluations may also be found in Todd (2004) for PROGRESA, Glewwe and Olinto (2004) for PRAF, and Adato and Roopnaraine (2004) for RPS. 12

13 the PROGRESA program (i.e., qualifying households in those localities would be eligible to participate), while the program was not deployed in the remaining 186 localities. Households in the latter localities were still subject to the data collection process, and thus constituted the control group for the program s evaluation. 10 The data employed in this study originates in the PROGRESA Evaluation Survey ENCEL. The estimates shown in the paper are based on the initial baseline survey, and three follow-up rounds 11 implemented in the subsequent semesters after the program implementation. The surveys collected socio-demographic and labor market information at the household and individual levels for the all the households in both treatment and control communities. Honduras PRAF was implemented in a set of 50 randomly selected municipalities of a total of 70, with the 20 additional municipalities forming the control group. The data in this document corresponds to a baseline survey carried out in the last quarter of 2000 and a follow-up survey in 2002, with a reasonably low sample attrition of approximately 8 percent. In contrast to the case of PROGRESA, where all households in treatment and control localities were interviewed, the PRAF surveys covered a sample of households. The corresponding sampling weights are used when possible in the empirical work below. For the case of Nicaragua s RPS, half of the poorest 42 localities were randomly assigned to the treatment group. The data used in this document corresponds to the initial baseline survey carried out in the third quarter of 2000, and the first follow-up survey conducted in The sample attrition rate is approximately 7 percent. As with the PRAF evaluation data, the survey consists of a sample of the targeted population, and sampling weights are used when possible. 3.2 Estimating Program with Experimental Data The random assignment of localities in the context of the three programs under study constitutes the ideal setting for plausible differences-in-differences (DD) estimation. A standard DD model with controls takes the form: 10 The evaluation followed a phase-in process: PROGRESA was deployed in the control localities when the program expanded its coverage in Baseline data was gathered between November 1997 and March The first, second and third follow ups correspond to November 1998, March 1999 and November 1999 respectively. 13

14 Y A B cx I (3.1) ist s t ist st ist where Y ist denotes the outcome variable of interest for individual i in group s for time t, I st is an indicator variable representing treatment status for group s in time t (or alternatively, an interaction between a treatment group indicator and time effects), A s and B t are group and time effects, respectively, X ist is a matrix of individual characteristics and ist is an error term. The estimate of the program impact is the coefficient. Without the X ist controls and with two time periods, the estimate of by ordinary least squares (OLS) is simply the difference in mean outcomes between the treatment and control groups in the two time periods. The more general case, with more than two time periods, adds a full set of time controls and interactions to account for differential evolutions over time. Despite the relatively simple interpretation of the standard DD model, the specific evaluation setting of PROGRESA, PRAF and RPS and the nature of the outcomes under analysis still require some discussions. It is still necessary to establish whether the panel nature of the database is accounted for in the estimation, whether linear or non linear estimators must be used for binary outcomes, and what type of adjustment should be made to the standard errors to reflect the structure of the program implementation process. In the empirical results presented below, we have chosen to report only results from linear models either OLS or fixed effects estimations of equation. The standard errors in the estimations need to account for the structure of program deployment. In the context of the three CCTs under study, the random assignment process did not apply directly over beneficiary households or individuals. The allocation was instead done at the geographical level. In terms of the equation above, randomization occurs at group level (s) instead of individual or household level (i). Since the eligibility for the program is defined at the group level, the standard errors of the DD estimates should account for the likely intra-cluster correlation to avoid a potential bias. Donald and Lang (2007) attribute this bias to the fact that many of the outcomes analyzed in the literature are serially correlated, which is not usually controlled for in DD estimation see also the discussion in Bertrand, Duflo and Mullainathan (2004 BDM henceforth). This issue might be particularly predominant in the case of the labor market outcomes covered in this study. Not accounting for this correlation across the randomization groups makes the usual OLS 14

15 standard errors inconsistent, and leads to erroneous inferences of the program s causal effects. BDM propose two methods to correct the standard errors of estimates in equation (3.1): (i) taking into account serial correlation of the outcome variable in each group s; this is known as cluster-robust variance estimation (CRVE) and is implemented by clustering observations by the assignment groups (e.g., municipalities or localities); and (ii) estimating standard errors using block bootstrap with replacement. 12 The two methods are implemented in the empirical results presented below. However, BDM report uncorrected OLS standard errors for their block bootstraps. We instead adopt the method proposed by Cameron, Gelbach and Miller (2008), and present results from bootstrapped CRVE-corrected standard errors. 13 These corrections to the covariance matrix yield unbiased estimates of household or individual level outcomes in geographic targeting settings, accounting for potential serial correlation across groups. As a robustness check, all the estimates presented below apply both corrections. 3.3 Labor supply of adults in CCT programs with experimental evaluation Descriptive statistics and the random assignment processes The tables in this section present a series of descriptive statistics at the time of the baseline survey for both treatment and control groups for the three programs discussed above. This allows us to verify the comparability of the treatment and control groups in terms of observables, and to establish whether it is necessary to control for pre-program differences when estimating the program s effects. Tables 3.1 present demographic statistics by treatment and control groups for the three programs. Household sizes in all three programs are high, greater than six (as expected in programs targeted at poor rural areas in developing countries), with approximately 70 to 80 percent including both parents. The calculations in the tables 12 While in standard bootstrap methods individual observations are randomly selected with replacement from the original sample, in block bootstrap a series of consecutive observations are selected instead. In the context of program evaluation with random assignment at the geographical level, the blocks correspond to the geographical units over which the treatment was allocated. 13 BDM also propose a third correction, by aggregating the data into group-year cells and estimating this model. However, we opt for using individual-level data in our estimates. 15

16 show that treatment and control households are not significantly different in their demographic composition, with a few exceptions for some subgroups. Table 3.2 shows educational statistics for each program. Enrolment rates are above 60 percent for children aged 6-11 in the three programs, with slightly higher rates when both parents are present. Since the programs are targeted at poor areas in each country, the overall educational distribution is concentrated on lower levels of attainment. The differences in educational achievement and enrollment rates between the treatment and control groups are relatively minor. Finally, Table 3.3 presents a set labor market indicators for control and treatment groups in the three programs. The datasets only allow for a simple definition of participation individuals report if they work or if they do not (with the exception of RPS). Employment varies from 52.2 (PROGRESA) to 67.9 (PRAF), and it is much higher for men than for women (about 35 percent in PRAF, 60 percent in PROGRESA and 22 percent in RPS). Employment is higher in households with children and in single-parent households. These individuals are mostly salaried and self-employed individuals, and we observe in RPS a high predominance of workers with no pay, who work full-time (more than 40 hours per week). The unconditional means of socio-economic and demographic statistics indicate some pre-program differences between treatment and control groups at the individual and household level. The same is apparent in a conditional framework. Tables present the results of a probit regression of the probability of being selected into the treatment sample for each program as a function of observable household and individual characteristics. Since the focus of this document is to estimate differentiated treatment effects across population subgroups, estimates are presented for the entire adult population aged 15-65, and separately by gender, with standard errors clustered at the locality level. 14 The results reveal some significant differences between treatment and control groups in all three programs. The results from PRAF indicate that treatment and control localities varied in two important dimensions. Households in the treatment sample had a significantly higher proportion of children attending school, and a significantly lower proportion of children employed. There do not seem to be any significant differences by gender. 14 A number of other models were estimated using further disaggregated characteristics. However, the results (available from the authors upon request) were qualitatively similar. Some of the differences disappear once the probit regression accounts for clustering at the locality level. 16

17 In localities selected for PROGRESA deployment, there was a significantly higher proportion of individuals employed, and this effect was accounted for mainly by men. The treatment and control samples from RPS, according to the probit regression, seem to be more balanced. None of the included variables is significantly associated with the probability of participation in the program. These results are in general agreement with preexisting reports on these programs, which also find differences at the household and individual level, but little significant differences at the locality level (see Behrman and Todd, 1999, for PROGRESA, Glewwe and Olinto, 2004 for PRAF, and Maluccio and Flores, 2005, for RPS). Given the random assignment process in the three programs, it is likely that these differences arise because of the small number of effectively randomized units (localities). They reflect the differences between the resulting samples rather than selection into treatment the randomization processes rule out the latter possibility. The estimations below control for some of these individual characteristics The effect of CCTs on labor supply of adults This section presents an analysis of labor market outcomes for adults in PRAF, PROGRESA and RPS. All estimates for individuals refer to those between 15 and 65 years old, while for household outcomes we restrict the sample to household heads in the same age range. 15 Since the evaluation surveys focused primarily on the programs outcomes, such as children s health and education, and household well-being, we only have a reduced set of labor market indicators when compared to periodic household surveys with large coverage. In the three data sources employed in this study, the adult population could only be partitioned into two alternate states, working and not working it is thus not possible to distinguish between inactivity and unemployment. 16 In what follows, we refer to working, employment rates and labor supply interchangeably. 15 The results do not differ substantially if we also include individuals aged 66 and more. 16 It is possible to make this distinction for RPS, but to maintain comparable results we chose to work with the same variable for the three programs and we settled for the lowest common denominator. 17

18 The outcomes of interest are an indicator of whether the individual is employed, the number of hours worked in all occupations in a week (for those with positive hours), an indicator for employment in agricultural activities (for those employed), and the total hours worked in the household by members aged 15 to 65 (this variable is computed and estimated at the household level, for households with positive hours). As stated above, we present two specifications for each outcome of interest. On the one hand, we estimate OLS models with a series of controls (the X variables in equation 3.1): Controls for individual outcomes: gender (if applicable), household size, an indicator for bi-parental household, number of children of the head of household, age of the individual, age squared, educational indicators (complete primary through complete college) and locality effects. Controls for household outcomes: the gender of the household head (if applicable), household size, an indicator for bi-parental household, number of children of the head of household, a dummy variable indicating if at least one child in the household attends school, indicators for the household head s educational level, and locality effects. The fixed effect estimations, on the other hand, do not include any controls besides time and treatment effects. As described in the previous section, we present two sets of standard errors for all estimations: cluster-robust standard errors and cluster-robust standard errors by block-bootstrap at the locality level with 100 replications. Finally, we present estimates for the whole samples and for a series of subgroups. For individual outcomes, we define the following subsamples: All individuals aged 15-65, males and females in the same age range. Females aged in single-parent households. Eligible individuals aged Eligible females aged The definition of eligible households is derived from the administrative rules for each program. A household is deemed eligible irrespective of its residence in a control or treatment locality according to the age of the children, and in the case of PROGRESA also according to the poverty level of the household compared to a given threshold, which were included in the program database. See the appendix for the age ranges of children that determine our simplified definition of eligibility for each program. 18

19 For household weekly hours of work: Households with head aged Mono-parental female headed households with heads of household aged Eligible households with heads of household aged We proceed to estimate equation (3.1) by OLS (with controls) and by fixed effects (FE) for all three programs. The estimates for PRAF and RPS correspond to the simple two-period case, while estimates for PROGRESA include the baseline and the three consecutive yearly follow up surveys. 18 The tables only report the relevant coefficient for the treatment effects (the coefficient ). 19 Tables present the main results for the employment status of adults for the three programs. The columns show the estimated coefficient of the time-treatment indicator in the regression, with the two alternative estimates of the standard errors (cluster-adjusted and block-bootstrapped). The results in Honduras s PRAF indicate a small and negative effect of the program on employment for all adults (0.7 percentage points for OLS and 1.7 for FE), and much higher negative impacts for females in single parent households (of 6.4 percentage points). While the estimated coefficients are negative, none is significantly different from zero at the standard levels. The results from Mexico s PROGRESA are presented in Table 3.8. The OLS estimates show a significant reduction of about percentage points in the labor supply of all adults for the second and third follow up surveys, but the FE estimates are smaller (about half) and not significantly different from zero. The only two cases with significant impacts for both the OLS and FE estimates are the coefficients for males in the third follow up survey, with falls in the percent range, and for females in single parent households, with reductions of percentage points for the second follow up survey. As in the case of PRAF, most coefficients are negative, but very few are significant at the standard levels. This pattern is similar to that found by Skoufias and Di Maro (2008), although these authors estimate probit models for eligible households only, and look in more detail at the impacts by age. 18 The tables report the effect by year, and correspond to the percentage difference between that year and the baseline (pre-program) levels. These effects are estimated jointly by multiple time and treatment interactions. 19 Full regression outputs for all estimations are available from the authors upon request. 19

20 Table 3.9 in turn presents the results for employment for Nicaragua s RPS. All coefficients are negative and similar in range to the other programs: for all adults, labor supply falls by 1.3/0.5 percentage points (OLS/FE), while for females in single parent households the estimate is -7.3/-6.5 percentage points (OLS/FE). However, as in PRAF, none of the results are statistically significant at the standard levels. The general pattern of non-significant effects of the programs in labor market outcomes is also evident in tables , which present the impact on hours worked for individuals which report positive hours. For PRAF, there is an increase in weekly hours for all adults of 0.73/0.922 hours (OLS/FE), significant at the 10 percent level. This impact seems mostly driven by individuals in eligible households. In PROGRESA, there is no overall effect, but for females in single headed households the program seems to increase hours of work by 3.5/2.8 per week (OLS/FE) in the second follow up survey, and these coefficients are significantly different from zero. While not significant, the overall impact for PRAF and PROGRESA is positive, but negative and relatively large (2.4/2.1 hours) for RPS. 20 Beyond employment and hours, a third dimension of labor market outcomes that the programs might affect is occupational choice. This is especially relevant in the case of rural areas in Latin America given the results of Skoufias et al. (2008), who find that the PAAL program in Mexico (see the review in section above) induced workers to move away from agricultural work. For this purpose, we construct an indicator for employed individuals equal to one if she works in agricultural occupations, and zero otherwise. According to tables , the impact of the program on occupational choice (agricultural versus others, for those employed) is positive for individuals in PRAF and negative for those in RPS, but none of the estimates are significantly different from zero for these two programs. For PROGRESA there are no significant aggregate effects either, but for the second follow up survey there are significant reductions in agricultural activities for men and for females in single parent and in eligible households, with impacts in the range of percentage points (the larger corresponding to women). 20 The OLS coefficient for females in single parent households is unusually large and significant at the 5 percent level, but the fixed effect regression estimate is not significantly different from zero. 20

21 Finally, tables present the estimates for total hours of work at the household level. There does not seem to be a clear pattern for PROGRESA, but there are some small positive impacts for all households and for eligible households (of 3.3 and 2.7 hours per week respectively) in PRAF. Most surprisingly however (given the lack of individual effects), there are large and significant reductions in total hours of work (of about 7 hours) for all households and for eligible households in RPS. These results are compatible with Maluccio s (2007) findings based on random effects estimation with follow up rounds of the RPS evaluation. Summing up, none of the three CCT programs analyzed seem to have large effects on employment for the whole population, although we did find negative impacts on some population subgroups at some points in time. For employed individuals we found a positive impact in hours worked for some subgroups in PRAF and PROGRESA, but a negative impact in RPS, especially for women. The sectoral allocation of work was not fundamentally transformed by either of the programs, although there is some evidence of a shift away from agricultural activities in PROGRESA. Total hours of work at the household level seem to have increased in Honduras with PRAF, but fell significantly for RPS. It should be noted that the results for hours at the individual level for RPS were mostly not significant, but we did nonetheless find a significant impact for total household hours. The lack of significant results might reflect the lack of statistical power, although this argument seems better fitted for the relatively smaller evaluation samples of PRAF and RPS than for the large, multi-period evaluation of PROGRESA. Despite a few noteworthy results, none of the three CCT programs analyzed in this section seem to have induced major shifts in the labor supply of individuals in eligible or beneficiary households. Moreover, the few significant impacts imply, if anything, adjustments in the intensive (hours) rather than the extensive margin (participation). The evidence from experimental evaluations presented in this document suggests that CCT programs do not modify substantially the labor supply of eligible individuals, and do not necessarily imply a disincentive to work. However, the external validity of these results must still be established. The three programs analyzed were deployed in poor rural areas, where employment opportunities (and opportunities in general) are likely to be limited. Moreover, there are important differences in the size of the average cash transfers, from about 4 21