University of Southern California Dissertations and Theses

Essays on child labor and poverty in the context of a conditional cash transfer program in Nicaragua

(USC Thesis Other)

Essays on child labor and poverty in the context of a conditional cash transfer program in Nicaragua

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON CHILD LABOR AND POVERTY IN THE CONTEXT OF A CONDITIONAL CASH TRANSFER PROGRAM IN NICARAGUA by Ximena V. Del Carpio ___________________________________________________________ A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (POLITICAL ECONOMY AND PUBLIC POLICY) May 2010 Copyright 2010 Ximena V. Del Carpio ii Acknowledgments I want to thank my advisor Jeff Nugent for his guidance and patience, and my committee members Carol Wise, Erik Heikkila and Peter Rosendorff for their advice and support. I am also indebted to my colleagues and friends Renos Vakis and Karen Macours for their support and instrumental role in the design and implementation of the project from which this dissertation draws. Special thanks also go to the program team at the Ministerio de la Familia, the Center of studies in Nicaragua (CIERUNIC) and colleagues at the World Bank in Washington and Managua. I owe special thanks to Norman Loayza for his endless guidance and tireless encouragement and support throughout this process; lastly, I want to thank my family, especially my grandfather John Vion who inspired me to keep going with my education and my son Joshua Prits for allowing me to spend countless hours away from home to collect the data and write this dissertation. Versions of these chapters have been presented at several conferences around the World. Among them, the Impact Evaluation Workshop in Managua, the Latin American Economic Association (LACEA) conference in Rio, Understanding Children’s Work (UCW) conference in Madrid, The World Bank labor seminar series, The World Bank Independent Evaluation methods seminar, both in Washington DC, the Economic Research Services seminar (USDA-ERS) in Washington DC, and RAND in Santa Monica. A different version of the third chapter (co- authored with Karen Macours) has been submitted and accepted by a peer reviewed journal. iii Table of Contents Acknowledgments ii List of Tables iv List of Figures vi Abstract vii Chapter 1: Political Economy Concerns of Conditional Cash Transfer Programs 1 I. Arguments for and against CCTs 3 II. Other Dimensions Influencing the Viability of CCTs 7 III. A Successful Program in an Unfriendly Political Environment 8 III.1. Program design and targeting 10 III.3. Overview of some of the results from the program 15 IV. Conclusion 15 Chapter 2: Does Child Labor Always Decrease with Income? An Evaluation in the Context of a Development Program in Nicaragua 17 I. Related Literature 19 II. Theory 22 III. Data and Stylized Facts 30 IV. Empirical Strategy 31 V. Results 34 V.1. Income and all child labor 35 V.1.i. Analysis of the effect of income on child labor 35 V.1.ii. Income and child labor by types 37 V.2. The program and child labor 40 V.2.i. Analysis of the program-child labor relationship 41 VI.2.ii. The program and child labor by types of children 45 VI. Conclusion 49 Chapter 3: Leveling the Intra-household Playing Field: Compensation and Specialization in Child Labor Allocation 51 I. Theoretical Framework 58 II. Background on the Program 64 II.1. Program randomization 66 III. Data, Descriptive Patterns, and Hypotheses 73 III.1. Child labor allocation patterns 77 III.2. Hypotheses 78 IV. Impact of Atencion a Crisis on Intra-household Child Labor Allocation 80 IV.1. Gender and age: Compensation and specialization 81 IV.2. Past academic achievement 91 V. Conclusions 99 References 101 Appendix A: Details of the Nicaraguan CCT Pilot Program 107 Appendix B: Intra-household Child Labor Analysis 119 iv List of Tables Table 1-1. Program Benefits by Interventions 12 Table 2-1. Analysis of income and total hours worked of all children 38 Table 2-2. Analysis of income and total hours worked, by age and gender 40 Table 2-3. Effect of the program on various types of child labor 43 Table 2-4. Effect of each program intervention on various types of child labor 44 Table 2-5. Analysis of the program on various labor activities, by gender 46 Table 2-6. Analysis of the program on non-physical labor, by child type 47 Table 2-7. Analysis of the program on physical labor, by child type 48 Table 3-1: Randomization results 69 Table 3-2: Child labor by gender: Intrahousehold allocation 73 Table 3-3: Child labor by age: Intrahousehold allocation 74 Table 3-4: Difference in child labor between children attending school and those not attending: Intrahousehold allocation 75 Table 3-5: Difference in child labor of children below their optimal grade level versus others: Intrahousehold allocation 76 Table 3-6: Intrahousehold heterogeneity of impacts by gender and age: All eligible households 82 Table 3-7: Alternative specifications for intrahousehold heterogeneity of impacts by gender: All eligible households 87 Table 3-8: Intrahousehold heterogeneity of impacts by gender and age: For basic package and productive investment package 93 Table 3-9: Intrahousehold heterogeneity of impacts by past academic achievement: All eligible households 97 Table A1: Means for variables in the analysis by gender 108 Table A2: Descriptive child labor data by age and gender 109 v Table A3: Analysis of Community Variables with a Significant P-Value 112 Table A4: Labor participation and hours by Income quintiles 117 Table A5: Randomization results for sub-sample (with children 6-15 years) 118 Table B1: Number of hours in each activity, conditional on participation 120 Table B2: Child labor by gender: Intra-household allocation 120 Table B3: Child labor by age: Intra-household allocation 121 Table B4: Difference in child labor between children attending school and those not attending: Intra-household allocation 121 Table B5: Difference in child labor of children below their optimal grade level versus others: Intra-household allocation 121 vi List of Figures Figure 2-1: Optimal child labor as a function of income 26 Figure 2-2: Income and predicted child labor (both genders) 36 Figure 2-2a: Income and predicted child labor, by gender 37 Figure A1: Mean hours worked in non-agricultural labor (including children with 0 hours) 113 Figure A2: Mean hours worked in non-agricultural labor (excluding children with 0 hours) 114 Figure A3: Percent of children who had diarrhea last month and children who were sick with something other than diarrhea last month 115 Figure A4. Food consumption per capita distributions 115 vii Abstract This dissertation presents the political economy dimension of conditional cash transfer programs. It also delves into the evaluation of one such program implemented in rural Nicaragua and measures the impact of the program on poor households, specifically on how the program affects children’s work. The analysis throughout this dissertation focuses on the impacts of transfer programs on beneficiary wellbeing as well as beneficiary behavior toward their children. In particular, the focus is on the decision of adults to engage children in various labor activities while receiving monetary benefits from the program. In the face of chronic poverty and negative economic shocks, people resort to different strategies to cope and survive. Those fortunate enough to have some physical assets use them in times of need and, if necessary, deplete them. Most people attempt to expand their household income by increasing adult participation in the labor force and by having children provide work inside or outside the house. While these strategies may be indispensable in the short run, they may have negative consequences in the long run. Asset depletion puts the family in a vulnerable situation in the future. Similarly, some forms of child labor may jeopardize the children’s ability to acquire both the educational skills and physical strength to become productive and well- functioning adults. Governments can help families caught in dire situations with programs that lift them above a minimum living standard. Among the battery of social protection programs used to target the poor are conditional cash transfers (CCT). Although CCTs represent a small portion of the safety net toolkit they become one of the preferred poverty alleviation mechanisms. Preference for CCTs stem from their potential to increase household asset endowments directly by viii supplementing incomes while fostering human capital investments through various conditionalities. Many countries around the world, particularly in Latin America, have adopted traditional CCT programs. Few countries have experimented with alternative CCT models that add components beyond health, nutrition, and education. In general CCTs seek to improve risk- coping and poverty management mechanisms so that households do not have to make disinvestments that can harm their future potential. Traditional CCT programs have been shown to limit the impact of shocks by decreasing the need of selling assets, pulling children out of school, increasing children’s labor participation, and reducing nutritional intake of household members. The evidence of the impact of traditional CCT programs is overwhelmingly positive, especially on human development outcomes. However, programs range in size, modalities, and designs and there are clear limitations that must be noted. Questions remain whether CCTs are feasible and effective in difficult contexts; more specifically, in places with low implementation capacities and low political support. Additionally, it is also unclear whether operational capacities of poorer countries are too limited so that they cannot appropriately target the needy and/or implement a program in a manner that will lead to measurable improvements. For example, it is unclear how CCTs can be implemented in places where social services, such as education and health, are weak or not available. Similarly, the exportability of these projects also remains unclear. One of the largest CCT in the world is Brazil’s Bolsa Familia; its design and success was unique in Brazil and copied in nearby countries in Latin America (such as Mexico, Nicaragua, Jamaica, and Honduras). However, the context in other regions, and poorer countries in Latin America, differs drastically, making it difficult to ascertain the replicability of the program and its success. ix Another area of research still unclear in the literature has to do with the sustainability of impacts. Specifically, the effects of CCT programs on the next generation of children living in benefited households. Unfortunately, not enough time has passed from the first generation of CCTs to assess the sustainability of program impacts and their effects. For example, Mexico’s Progresa (or Oportunidades) is still relatively young (less than 10 years) to show long-term impacts and sustainability of change (Parker et. al. 2007). The first chapter in this dissertation presents an overview of the strengths and weaknesses of CCTs, explores the political economy behind their feasibility and their adoption, and studies their implementation experiences in various contexts. The second part of the first chapter introduces a CCT program, designed for a very poor and vulnerable context, using an alternative approach suitable to the development needs and reality of the region. The program presented in this chapter was implemented in the north-central region of Nicaragua. Its design includes traditional CCT components—nutrition, health, and education—as well as components aimed to improve the employability of young adults and diversify the household’s income source. The objective of introducing this program in this chapter is twofold, first, to qualitatively explore the political economy implications of designing, implementing, and evaluating a customized CCT program; and second, to illustrate certain features of the program—mainly its design and evaluability—that can help improve the design of future projects elsewhere. The second chapter investigates the relationship of household income with child labor. The chapter presents a simple model that relates child labor to household income, beneficiary preferences, and production technology. The chapter also presents an empirical application to test this relationship, expanding the analysis by stratifying the sample by age and gender. Lastly, the chapter investigates the effect of the program on child labor, dividing labor into two types – physically demanding labor and non-physical labor– to reveal the impact of the program. x The third chapter analyzes changes in the allocation of child labor within households in reaction to the CCT program. The chapter starts by formulating a simple model that investigates how children and household characteristics interact to render a specific distribution of tasks among children. Then, the chapter analyzes the data with the purpose of identifying the characteristics that determine the allocation of child labor in the sample of households prior to the implementation of the program. Finally, the chapter evaluates whether the exogenous shocks created by the CCT program resulted in compensation or reinforcement of pre-program differences in child labor allocation and human capital accumulation. All chapters are motivated through the relevant literature in order to place the chapter into the wider literature and highlight the contribution of this work. The second and third chapters also formulate theoretical models to clarify and motivate basic relationships and offer policy implications and ideas for future research. An essential component of this dissertation is its original data set. Data did not come from secondary sources; it was obtained directly from my work and that of a team of researchers. The data collection exercise was conducted through several field visits that resulted in a two-year panel (2005 ‐ 2006) of randomly targeted households, providing detailed information on household characteristics--such as assets, income, consumption patterns-- and individual education and labor outcomes, including child labor. 1 Chapter 1: Political Economy Concerns of Conditional Cash Transfer Programs The socio-economic consequences of shocks evidenced in the last few years highlight the fragility of the achievements to help the poor and the danger of overturning gains made since the last decade. The effects of past economic crises continue to linger among the poor. Additionally, the recent food and economic crisis pushed over 130 million people into poverty, and over 50 million more are expected to fall into poverty as a result of the global economic crisis (World Bank, 2009). Moreover, negative impacts of various micro and macro level shocks not only reduce incomes but also threaten gains in various human welfare dimensions. Given the recurrence of shocks, and the fact that the poor are often hit the hardest, governments are eager to re-engage by enacting social assistance programs that serve as safety nets to protect people from falling further into poverty. Almost all safety net programs—defined as noncontributory programs that are targeted to the poor and vulnerable (Grosh et. al. 2008)— aim to shield the poor from destitution in the short term and expand their opportunities in the long term. More specifically, the main goal of safety nets is to protect vulnerable people from social risks while fostering the redistribution of resources and enhancing economic opportunities by encouraging better investments in human and physical capital. Conditional cash transfers (CCT) are one of many types of anti-poverty programs in the safety net toolkit. They are specifically designed to improve human development outcomes— especially in education, nutrition, and health—conditional on a set of rules and based on the objectives sought by the program. Though CCTs are widely applicable, they are not designed to target all poverty dimensions; instead, they are designed to deal with populations whose binding constraint for getting out of poverty is low human capital. CCTs have been adopted worldwide in the last decade (Fizbein and Schady, 2009) despite clear political economy implications resulting from their resource intensiveness 2 requirements. In many countries, CCT programs absorb a large percentage of the total social assistance budget, which has direct implications on other social expenditures. In Brazil and Ecuador for example, expenditures are estimated to be around half of one percent of gross domestic product. Thus, given the significance of these programs it is crucial to understand the political economy arguments before implementing a new program or enlarging an existing one. Some of the emerging research in the literature finds that a key factor in the adoption of effective social policies to help the poor is the degree of their empowerment and their ability to influence policy makers (Ravallion, 2009). Moreover, recognizing the arguments for and against these programs requires understanding basic elements in program design and their ability to target well enough to affect poverty at an aggregate level. The onset of cash transfer programs over the last decade has enabled researchers to accumulate rich evidence of what types of programs (and modalities) reduce poverty in various settings. Much of the evidentiary base is derived from results measured through rigorous impact evaluations. For the most part, results show that CCTs consistently deliver positive results on the following indicators: household consumption, access to health and education services, and children’s anthropometric indicators (Rawlings and Rubio 2003). Given the evolving nature of these programs, a new wave of modified (or alternative) CCTs has developed in the last five years, setting forth an opportunity to enrich the evidentiary base with evaluations of alternative CCT designs. Though it is clear that traditional CCTs positively affect basic indicators, it remains unclear whether non-standard indicators can also be positively affected by alternative programs. Thus, development researchers are slowly broadening their evaluation focus and beginning to shed light on the effectiveness of alternative CCTs on intended and unintended outcomes. Given that the size of CCT transfers typically exceeds 10 percent of household income, many critics argue whether these programs can have perverse effects in the long-term. Moreover, 3 CCT critics question whether these programs are capable of relaxing the household’s liquidity constraint and result in sustained improvements over the longer-term. The primary objective of this chapter is to highlight that despite the general success of CCTs, critics are still not convinced about their application to reduce general poverty. A second objective of this chapter is to raise some political-economy concerns raised by the critics in the literature; mainly, that voters are still skeptical of spending large amounts of resources on a small segment of the population. The latter part of this chapter introduces an example that illustrates how despite a program’s overwhelming success in reaching the poor and reducing poverty, a CCT program in Nicaragua did not survive political changes (and gain the voters’ support) and was not sustained after the pilot phase. I. Arguments for and against CCTs Transfer programs such as CCTs became very popular in Latin America because their design could be customized to address immediate needs of the poor through income transfers, while also protecting longer term human capital investments (Parker et. al. 2007). A recent compilation of impact evaluations of CCT programs concludes that the success of CCTs in reducing poverty and encouraging parents to invest in health and education of their children is due to their short and long-term approach (Fiszbein and Schady, 2009). People who support these programs argue that poor parents discount the returns to investing in their children more heavily than the optimal amount and need incentives (or conditions) to invest more in them. CCT proponents rely heavily on the overwhelming amount of positive evidence that has been documented in the last decade to justify their support; however, most of the evidence focuses on the gains, especially short-term, without contrasting them to the costs. The evidence contains few examples that calculate the benefits, be it in the short or long-term, against the costs; thereby, limiting policy makers from being able to assess the real gains in order to convince their constituents of the net gains to society. For example, Shultz (2004) shows that a component 4 (focusing on secondary school) of the national CCT program in Mexico, known as Progresa/Oportunidades, increased secondary enrollment rates of older students in benefited households. Meanwhile, de Janvry and Sadoulet (2006) show that over sixty percent of benefited students in the same country would have enrolled in school anyway, meaning that only ten percent of new students were incentivized to enroll as a result of the program. In other words, the costs of that component far exceeded the gains from it; thus, potentially risking the public perception and support for the program as a whole. The arguments against CCTs can be grouped along two related strands; these are: First, the political economy implications of spending scarce resources directly linked to the cost-benefit ratio of these programs to society; and second, the potential negative incentives that these programs can have among people and the lack of viable program exit strategies. On the first strand, the literature shows that public (or voter) support of a program can be heavily influenced by the perception of the cost-benefit ratio to society. Therefore, politicians must understand well the political environment they operate in. A recent paper by Zucco (2009) shows that the large Brazilian CCT, known as Bolsa Familia, shifted voter support for the incumbent government from more developed regional areas to poorer ones. The paper argues that the change in the composition of the electorate is partly attributed to the effect of the program; though there are other confounding factors, such as the general pro-poor economic growth (Carraro et.al., 2007), that seemed to contribute to the large shift. Critics argue that redistribution to a subset of people, particularly when the amount is so significant, reduces general welfare. This is especially true when governments have scarce resources, as in most developing contexts, and there are no certainties that the return on investment will be positive. Most voters understand that there are positive social externalities from having more educated and healthier children. Thus, they want to be assured that program beneficiaries will be required to invest in their children at or above the socially optimal level. To 5 deal with this concern, program designers spend a lot of resources on targeting the poorest, conditioning transfers to induce socially optimal behaviors, and monitoring beneficiary outcomes regularly. With respect to targeting, some authors argue for a combination of targeting methods that include self-targeting into a program and a ―proxy means‖ approach based on household poverty scores, obtained using administrative data, in order to reduce errors of benefiting the non-poor. Coady and Parker (2009) find that there is a gap of knowledge in the literature on the relative contributions of different targeting methods and their effect on program results. Their work also shows that self-selection, based on knowledge of the program alone, can reduce leakage to wealthier households by making the eligibility criteria clear. Unfortunately, the self-selection method does not help increase participation of the poorest groups, or destitute, because they are usually living in unreachable areas or have no access to the materials, thus remaining unaware of the program. On the other hand, administrative targeting through a poverty score helps increase coverage of poorer households; however, it carries the risk of having non-poor households misreport their welfare to join the program, emphasizing the importance of using complex algorithms to reduce cheating. This is why Coady and Parker highlight in their work that a combination of both methods can improve program targeting performance. On whether to condition benefits to behavior or not, many critics or reformers question whether conditionalities are really necessary to ensure fulfillment of socially optimal activities (De Brauw and Hoddinnot, 2007). Lindert and Vincensini (2008) find that the conditional nature of Bolsa Familia makes the program more widely accepted by all sides of the political spectrum. Unfortunately, conditioning comes at a heavy cost and reduces the amount of resources devoted to benefits by increasing administrative costs. For example, the Mexican program spent about sixty-five percent of the total budget of the program on targeting and eight percent on implementation the first year (de Janvry and Sadoulet, 2006). Fortunately, follow-up evaluations 6 and expenditure reviews show that administrative costs were substantially reduced over time, as the program became institutionalized. There are clear trades-offs that policy-makers face; trade-offs are encountered at two levels: when assessing what to invest in or ―between‖ programs, and when assessing how to design the program. The first is a standard problem whereas the second is not. Focusing on the cost-benefit ratio within the program is newer and potentially more sensitive; devising the most suitable design that maximizes benefits and reduces unnecessary costs can gain general support. For example, program designers must decide whether to impose a strict targeting mechanism that reduces leakage and impose conditionalities to maximize impact and effectiveness or provide an unconditional model that reduces monitoring costs and can be more efficient. The second strand of criticism argues that the provision of free cash can create distortions of incentives, thereby fomenting more bad than good in the long-term (Levy, 2009). This has also been addressed through conditionalities and structuring them as co-responsibilities between the state and beneficiaries where they are partners in ending poverty. The core of this partnership is based on a reward system for good behavior and the promise that society at-large benefits from the programs in the long-term due to increases of productive citizenry. This strand of critics also focus their attention to the potential traps that CCTs can create; they argue that the program invites people to get hand-outs without having a clear exit strategy, thus fomenting an inter- generational transmission of dependency on social hand-outs rather than diminishing poverty. Levy (2009) argues that the incentive structure implicit in some social programs may lead to behaviors that reduce future productivity. Similarly, other authors find that certain features of these programs can offset benefits; some offsetting effects are increases in child labor and fertility/migration, decreases in adult labor and other transfers (Fiszbein and Schady, 2009). In the case of child labor it can be increased by embedding features and/or conditionalities that affect adult labor participation and children have 7 to substitute them (Macours and Del Carpio, 2009, chapter 3). Fertility and in-migration can be triggered by setting the provision of benefits to vary with the number of dependents in the household. Decreases in adult labor due to perverse incentives are a significant concern; Maluccio and Flores (2005) found that adult men in benefited households reduced their labor hours by almost the equivalent of one day of the week. Fortunately, the vast majority of evaluations of other CCT programs find no evidence of this effect, likely because people perceive these programs to be temporary in nature. Many evaluations of various CCTs show little or no evidence of crowding-out of other transfers (mainly remittances), indicating that this is not a concern. II. Other Dimensions Influencing the Viability of CCTs Another interesting aspect worth exploring in this chapter is whether this sort of program is a good vehicle for delivering other benefits (to other target populations) and whether the results will be equally successful. A CCT program described later in this chapter begins exploring this by measuring the effect of a modified CCT design where adults are also benefited. In other words, does it make sense from an investment stand-point to add components to the traditional CCT platform in order to offer a more integral program? Previous evaluations show positive results on the overall effectiveness of CCTs on increasing children’s outcomes. Some of the evidence looks at the differential effects on men and women, and in some cases authors investigate the effect of some of the potentially offsetting characteristics of CCTs such as impact on adult labor and child labor. With respect to the relationship between child labor and CCTs, the majority of authors (Skoufias and Parker, 2001; Maluccio and Flores, 2004; Glewwe and Olinto, 2004; and Attanasio and others, 2006) show that traditional CCTs do not exacerbate its incidence among beneficiaries. However, most of their work measures the relationship between traditional CCTs, and none test the impacts on CCTs with alternative designs, especially those targeting adults as well. 8 III. A Successful Program in an Unfriendly Political Environment The political backdrop of the Componente de Atencion a Crisis (CAC) pilot program was contentious. The political party in power was concluding its term in office and all successes attributed to the program were minimized by the opposition party, making the program a potential victim of political change. The project was put in place building on an existing conditional cash transfer program Red de Proteccion Social (RPS) implemented in 2000 by the Ministry of Social Protection of Nicaragua and the World Bank (Maluccio and Flores 2005). In that program, like it was envisioned for the pilot program evaluated in this dissertation, the government agreed to outsource health services to private providers in order to reach the most remote areas. Similarly, teachers received a regular bonus for helping monitor student attendance and to encourage their active participation (and reduce truancy among them). The new CAC pilot was designed using a similar approach but was modified to test three new components that target adult outcomes directly. The program was planned for one year (from November 2005 to the last transfer in December 2006) and it targeted approximately 3,000 households for private transfers in a region that recently experienced high levels of crop loss due to drought. Program recipients were women, who were also the main participants in various components of the program, though transfers were fungible and could be used by all household members. Politically, the region contained a wide representation of both relevant political parties in Nicaragua; the Liberales, representing the right-wing in power during the pilot phase of the project, and Sandinistas, representing the left-wing power with strong historical significance in the region and a re-surge of support. Most of the six pilot municipalities were led by the Liberales, making opponents perceive the program as a clientelistic reward, locking-in party backers; this appears more evident as Sandinista supporters were gaining power in local elections, eroding the ruling party’s core base. 9 The plurality of parties became more evident during electoral cycles and when the presidential political campaign commenced, with some people blaming the local leaders for excluding them from the program or favoring some communities over others. The region played a significant role during the Sandinista revolution a few decades before due to its geographic location, with many people uprooted from their villages among other unpleasant memories. Overall, the political environment in the pilot region was characterized by tension between core backers of one or the other party, coupled with very low levels of economic investment and development. A paper by Diaz-Cayeros, Estevez and Magaloni (2009) finds evidence of clientelistic behavior in Mexico, with an inverted J-shaped relationship with development. The incumbent party transferred large amounts of social resources in the form of private transfers to municipalities with large but eroding numbers of core supporters; thus, using overall targeting as a means to reward loyalty, punish defection and motivate political support. Apart from the six municipalities suggested by the government for targeting in the Nicaragua pilot program, politics had limited influence on the communities and actual beneficiaries selected because the approach was based on a random selection of both, communities and households. Thus, the primary reason to conduct a very rigorous evaluation of the program was to minimize political capture and provide credible evidence of the impact of the program to inform the existing government and the future one, and hopefully gain support for a full scale-up effort in subsequent years. Unfortunately, the program did not gain much support from the incoming government and was never institutionalized by them as part of its social policy agenda; on the contrary, CAC and its predecessor RPS were perceived as an ―assistentialist‖ approach of the previous government, used for clientelistic purposes. Previous evidence, from countries such as Brazil and Mexico, highlights the important role of transparent impact evaluations in times of transition. The evaluations from Brazil and Mexico show that despite political change, programs were kept as one of several social safety net 10 efforts; however, they had to undergo cosmetic changes (such as the change in name of the program) in order to gain ownership and political brand association. Despite the closure of CAC and RPS there are many important lessons that policy makers can take away from these programs. Among these lessons are their innovative designs, transparent targeting through open lottery systems, usefulness of a well-developed monitoring-information-system (not covered in this chapter), and the role of a rigorous evaluation. III.1. Program design and targeting The program consists of three components or packages of private transfers. The purposeful selection of the treatment region (six drought affected municipalities) seems to indicate that the government of Nicaragua engaged in vote-buying behavior by implementing the program to reward its core voters and decrease erosion of support. On the other hand, the random selection of the communities and beneficiaries into one of three of the components ensured that political capture did not affect the design and evaluation of the program; otherwise, the distribution of benefits may have been affected as previous theoretical work by Diaz-Cayeros, Estevez and Magaloni (2007) shows, with higher private benefits to places with higher electoral volatility and lower private transfers to communities with a stable core electorate. On the design, the first component resembles any traditional CCT. All beneficiaries receive transfers conditional on children’s school and health service attendance; this component resembles other traditional CCT programs. The second component is the first modification to the traditional CCT design; it includes a feature that targets young adults through a labor training scholarship. About one third of households received a scholarship to participate in a technical training course aimed at providing them with new labor skills. This package entails coverage for all costs (i.e., course, transport) as well as compensation for lost wages while training (up to 6 months). 11 The third component is a business grant aimed at productive investments in livestock or non-agricultural activities for household adults. It was given to one-third of the beneficiaries, and it is in addition to the traditional CCT package. Beneficiaries were conditioned on writing a business plan, with technical assistance, and outlining how it would diversify the household’s income generation options. The mean household yearly expenditure is $1577 US dollars and per capita is approximately $340. Program benefits range, depending on the package, from approximately 15 to 25 percent of total annual income. Details of transfers are listed in Table 1-1; total per household varies on the number of children and type of package. 12 Table 1-1. Program Benefits by Interventions Transfer Amount Comments # beneficiary households Traditional CCT All 3000 Food transfer $145/household per year Partial transfer every 2 months over 1 year Education transfer $90/household per year Partial transfer every 2 months over 1 year School ―backpack‖ (supplies) $25/child per year 1 time transfer at the beginning of the school year School ―supply-side‖ transfer $1.3/child Every 2 months over 1 year Health transfer $90/household per year Was to be paid to health provider (but was never implemented) Occupational training (Traditional CCT above plus) +/-1000 Opportunity cost transfer Up to $90/household per year $15 per month for the duration of the course, up to 6 months. Paid every 2 months Course costs Up to $140/household per year Paid directly to course provider upon selection of course Matching Grant transfer (Traditional CCT above plus) +/-1000 Matching Grant transfer $200/household 1 time transfer upon successful completion of a business development plan (in 2 payments: 175 $ + 25$) III.2. Transparency embedded in the design and the evaluation The strengths of the program, apart from the private social benefit to the poor, are the methods by which it selected its beneficiaries—and averted political capture--and the rigour of its evaluations (experimental) aimed to reduce (or eliminate) biases in the selection of beneficiaries that often inhibit researchers from obtaining a true estimate of the impact of the intervention 13 (Duflo et. al., 2007); thus making the evaluation results more credible and less subject to manipulation. The random selection of beneficiaries led to increasing the perception of transparency among most beneficiaries and non-beneficiaries; the selection was done as a multi-stage process. First, six poor municipalities in the north of the country were targeted due to their low level of economic development and high susceptibility to droughts; though, it is not clear if the selection of these municipalities was based on political motivations because the region was not randomly chosen. National poverty maps were used to justify the selection and initial targeting of the general region, highlighting two important criteria for social programs: poverty and vulnerability. Once a comprehensive list of all communities within the six municipalities was assembled, 56 intervention and 50 control communities were randomly selected through a lottery process. The mayors of each municipality participated to observe the process. Due to budget constraints, only 3,000 households could be selected for treatment and 1,000 for control (no-treatment). To ensure comparability between the control and treatment communities, these were paired up based on proximity and similar characteristics (e.g., access to services, climate, and crop mix). There were some cases where two small communities would serve as a control or treatment of a large one. Geographic maps were used to ensure these were close to each other but sufficiently separate to ensure minimal spillover effects. Once the treatment and control communities were selected, the second step (identification of treatment households) of the selection process took place. The criteria for eligibility, within treatment communities, were set using the proxy means methodology developed for the previous CCT ―Red de Proteccion Social‖ and using the national household survey collected in 2001. The data collection process was completed in 2005; this included a census of households in treatment communities and a random sample of households from the controls. All households were placed in ranking order with respect to their economic status using their own household data and proxy 14 estimates. Ranking was used to identify the poorest households and determine the appropriate sample. Community leaders helped the ministry and implementing team identify possible exclusion and/or inclusion errors; however, to avoid potential political capture from community leaders, changes were limited to the most obvious cases (physical visits took place before inclusions or exclusions were made). These small corrections led to 3.72 percent of households being re-assigned from eligible to non-eligible and 3.65 percent the same in reverse. Given a potential for selection bias resulting from the re-assignments, all estimates in the following essays reflect intent-to-treat estimates as defined by the proxy means estimation and methodology. Communal registration assemblies were organized; one or more depending on the community size, and female household members were invited. The assemblies had multiple objectives. The first objective was to introduce members to the program and its components. The second objective was to recruit women volunteers for several implementation and monitoring tasks; and third, to conduct a lottery (random allocation) of the packages among all households in assistance. All households invited attended and only a handful in the entire process chose not to participate. An evaluation of this program by Macours and Vakis (2009) finds that beneficiary women were significantly empowered by the program. More specifically, the authors’ find that the program improves the attitudes of women toward their future through improved exposure to leaders, the process of decision making, and allows them to gain experience in leadership roles. The third stage of the selection process is the lottery of households to receive one of three packages. Given heterogeneity within communities, and to increase the power of the statistical tests, all three packages were offered in each community. During the assembly households drew one of three colored balls from a bag and the color was recorded. At the end of the assembly each color was assigned to a package, based on the same lottery mechanism. Families were then 15 recorded as being part of the traditional package, the vocational training package or the small grant package. Appendix A describes the data collection and evaluation approach of the program. III.3. Overview of some of the results from the program The first phase of the evaluation, nine months after the program has started and two months before the last transfer, shows overwhelmingly positive results. Among them are nutritional and health care improvements; as a result, cognitive development outcomes were also positively affected. As expected from the evidence of other CCTs, education enrollment improved for the target group (children ages 6 through 15), household income and consumption increased. In terms of asset creation, the program led to increases in both productive and non- productive assets, which are expected to help with the sustainability of impacts. Income from economic activities, mostly in small commerce and non-agriculture increased significantly. There are some interesting positive spillover effects resulting from the design of the program. Mainly, the program strengthened local social dynamics by making people interact to fulfill the productive aspects of the program as well as regularly participate in community meetings. The program had an effect on leadership among women in beneficiary communities; likely due to the fact that they received the benefits and had to ensure their family and other community members were compliant. IV. Conclusion Much of the evidence obtained from impact evaluations of CCT programs confirms that these programs are effective in increasing human capital outcomes such as education participation, health prevention and nutritional gains for children. Similarly, these programs can also serve as political instruments; the private transfer characteristic of CCTs makes it an optimal instrument of tactical redistribution, where governments can use to reward loyalty among poorer supporters and avert future voter erosion by indirectly threatening to remove the program. In the case of the pilot program under review in this dissertation, there is no formal investigation of this 16 point; however, anecdotally there is some indication that the initiative for targeting the north central region supports the clientelistic view, though full capture was averted by the random selection of communities and beneficiaries. This chapter provides evidence that alternative CCT designs, such as CCTs that include benefits targeted to non-children or aimed at improving outcomes beyond human development (education, health and nutrition) also result in positive welfare gains. It remains unclear whether conditionalities are necessary to ensure positive outcomes or are simply a tool likely to ensure impacts and/or gain societal (and political) support for concentrating scarce resources on a subset of the population. The answer to this point will not be possible until CCT programs purposively omit conditionalities. From the broader perspective, CCT programs face a number of challenges. These challenges range, from reaching the neediest groups to fostering transparency and accountability; this is especially true among similarly poor groups that do not receive the program. Overcoming political economy barriers may be one of the most difficult and important problems program designers have to account for; previous evidence has shown that this can improve by using a transparent system of beneficiary selection and measuring/reporting results. However, the case in Nicaragua showed to be an exception, demonstrating that political support is unpredictable. 17 Chapter 2: Does Child Labor Always Decrease with Income? An Evaluation in the Context of a Development Program in Nicaragua Prior theoretical work on child labor has attempted to establish a link between poverty and child labor by modeling the relationship and obtaining interesting, but with disparate results. Parallel to this, a whole body of empirical work designed to study this seemingly important relationship emerged; a few notable papers mix both, theory and empirics, in order to improve the knowledge of the causes and consequences of child labor and provide us with the groundwork necessary to evaluate this complex relationship. Unfortunately, good panel data on child labor (e.g., children’s time use, activity types) and income data in the same survey are rare; as a result, the literature has been limited from obtaining conclusive evidence on why child work is present in households at various levels of the income distribution. The research question in this chapter is evaluated in the context of a CCT program targeted at improving human capital outcomes of school aged children and the productive possibilities of a sub-set of households in a poor rural region in Nicaragua (see chapter 1 for detailed program information). In this chapter I ask if child labor always increases with income. Moreover, I expand the research question to incorporate the CCT program into the analysis and pose a secondary question, how does a cash transfer program, with three distinct interventions and meant to improve the economic welfare of the families affect child labor? The empirical section makes use of a rich panel data-set collected for a randomized experiment. The chapter contributes to the child labor literature through a simple yet comprehensive model and an empirical application by revisiting the relationship between income and child labor; more specifically the chapter evaluates what happens to child labor in households that are distinctly placed at various levels of the income distribution. The model illustrates the relationship of different forms of child labor with each other and with household production and 18 assets; it allows for the inclusion of a transfer program through a tax on child labor or a subsidy to alternative activities (such as education or leisure). Given that the conditional cash transfer (CCT) program offers standard benefits plus income to promote productive activities in non-agricultural work, the expected effect of the program on child labor ex-ante is ambiguous. For the purposes of this chapter I use several definitions that influence the theoretical and empirical approach. I follow the definition of human capital used by Bardhan and Udry (1999) where they use the term human capital to cluster factors such as nutrition, health, formal education, and on the-job-training. I follow the international labor office (ILO) definition for child labor, that is any activity other than study or play that is remunerated or not for children fifteen years or younger. I build on this definition in part of the analysis by disentangling the word labor into two types: non-physical labor—this is a combination of commercial savvy, calculation ability, intellectually oriented tasks—and, physical labor—based almost solely on physical strength. The definition for non-physical labor is consistent with the objective of the program, to improve the economic opportunities of households by diversifying their labor activities from agricultural-dependency to non-agricultural activities. More specifically, the sample households primarily dependent on non-agricultural work, to include tradesmen, professionals, merchants, service providers etc. earned 21 percent more income in 2005 than households mostly dependent on agricultural work. Many people lack the necessary skills to work in these activities and the costs to obtain the necessary skills are high enough to make them inaccessible for many. The program seeks to increase the earning potential of beneficiaries by promoting non-agricultural activities; it does this through vocational training, business creation, and business training in order to enhance the skills of beneficiaries and reduce their exposure to risks stemming from agricultural dependency. 19 Given that the program was designed to reduce income vulnerability due to risk from weather shocks, I consider that a skill enhancing activity is one that changes occupational practices of people and increases their productive capacity in the future. These skills can be used through non-household employment as well as self-employment but not be directly dependent on land or animal breeding purely for self-consumption. Any type of activity that has value added such as services, commerce or trade (to include animal trading and commerce of animal products), formal or informal employment in a non-farm setting and/or profession qualifies in this category. A physically oriented activity on the other hand is one that depends on the land and/or animals directly, entails farming and harvesting farm products primarily for household consumption, raises animals primarily for household consumption not commerce, gathers goods (i.e., wood, water) for household use and takes care of household chores. The remainder of the chapter is organized as follows. Section I presents the background literature on child labor that helped motivate this research. Part II develops the theoretical model of income and child labor. Section III presents some stylized facts and descriptive statistics from the dataset, focused on child labor and other factors that influence the behavior of households toward child work. Section IV discusses the empirical implementation of the model. Section V presents the results and a brief discussion, and part VI contains concluding remarks. I. Related Literature One side of the child labor and wealth (income) literature agrees with Basu and Van (1998) where they propose that having the child not work is a luxury that poor families can rarely afford, and as income increases the poor family can afford more leisure: ―Luxury axiom‖. Other studies find that child labor is a consequence of inequality in the distribution of non-labor income (Swinnerton and Rogers, 1999) or economy wide poverty reflected at the household level (Grootaert and Kanbur, 1995). In the case of poverty, studies find that schooling is often traded 20 for work during difficult economic times (Edmonds, 2003; Edmonds and Turk, 2004; Beegle, Dehejia, and Gatti, 2005). This finding is extremely important because it shows that parents recognize the importance of human capital accumulation and want to do what it takes to increase the earning potential of the kids in the future, such as enrolling them into school or engaging them in skill- forming activities, when resources permit it. This is consistent with altruistic arguments presented in the literature and stemming back to Becker and others (Becker and Lewis, 1973) but is not inconsistent with the possibility of parent’s wanting to improve the human capital of their kids for reasons other than altruism. Previous work also finds that when parents are faced with liquidity constraints, particularly in the absence of functioning capital markets (Baland and Robinson, 2000; Ranjan 2001; Dehejia and Gatti 2002) they are more likely to engage their children in work, despite their preference for having children not work at all or only in certain types of activities. These studies directly or indirectly argue that parents would be willing to borrow against the children’s future earnings to potentially fulfill their preference of increasing the human capital acquisition of their children today, but in the absence of credit markets, they are forced to remove their kids from school (or reduce their study-leisure time) and in most cases have children work. The opposite strand of the literature presents a wealth paradox argument that challenges the luxury axiom. This literature finds that child labor increases in periods of economic growth and children in wealthier families—in terms of assets and land which are found to be correlated to the household’s income—work more than children in asset-poor households (Parsons and Goldin, 1989; Bhalotra and Heady, 2003; Rogers and Swinnerton, 2004). One of the arguments for this challenge is the presence of imperfect labor and land markets in most developing countries. Another explanation is that children have to take up more domestic chores as adults are engaged in household enterprises (Hazarika and Sarangi 2005). The unequal distribution of land in poor 21 countries, with heavy reliance on land production in agrarian societies, is biased toward wealthier households. Wealthier households who are unable to hire laborers, due to labor shortages, have a larger incentive to employ their own family; particularly when the marginal product of labor is increasing in land size. Other reasons given for preference for family over a hired hand are: moral hazard, easiness of shirking in volatile weather areas, reliance and overall trust (Deolalikar and Vijverberg, 1987; and Foster and Rosenzweig, 1994). In this chapter I consider an alternative reason, that not all child labor is deemed negative or harmful (Edmonds, 2007) and parents, who seek to enhance the human capital of their children and/or recognize their capacity to learn, can potentially distinguish between the prospective contributions of some work activities. Moreover, not all child labor occurs at the expense of schooling or interferes with human capital accumulation, particularly for younger kids who tend to combine both activities (Cartwright and Patrinos, 1999); in some cases child work enables school attendance by increasing the household income as well as improves the productive capacity of the child through the attainment of labor skills. Other potential advantages of child labor explored in this chapter are consistent with findings in previous studies. They show that child labor seems to serve as training experiences for the children who obtain and enhance new skills well before adulthood (Rogers and Swinnerton, 2002; Raju, 2005). Beegle, Dehejia and Gatti (2005) find that young adults who attained work experience as children had higher earnings in wages and farm work than others with less or no experience. More interestingly, the author’s find that the loss in wages due to early abandonment of formal schooling was fully offset by the work experience obtained as a child. These results however, are applicable for a few years after the work experience is attained as the returns to work experience decrease over time. 22 A literature review by Dar et al. (2002) presents thirteen empirical studies where the effects of household welfare on child labor are evaluated and find that there is strong support for the luxury axiom presented by Basu and Van (1998). The literature review also concludes that although welfare has a significant impact on child labor and the relationship appears to be inversely related there are some studies that cast doubt in the strength of the relationship pointing to other factors as being equally or more important. In an effort to reconcile the diversity of findings in the poverty and child labor relationship Basu, Das, and Dutta (2007) examine the possibility of an inverted U-shape relationship. This relationship provides the theoretical basis to this chapter and the model developed in the chapter. This research is done in the context of a social program that transfers income and skills (through training) to families conditional on various types of human capital investments (i.e., school attendance, nutrition, vocational training) and physical capital investments (e.g., goods, equipment). More specifically, for one third of the treated families there is an added conditionality of starting a non-agricultural business while another third of families are conditioned on enrolling a family member in a vocational training course. II. Theory In this section I present a unique simple model, with some resemblance to a previous model by Basu, Das and Dutta (2007) in terms of how benefits and costs of child labor are specified but distinct in every other dimension. In this model, different types of child labor are combined with adult labor and capital to obtain household production. Apart from exploring income effects on child labor, the model allows us to study the effect of a CCT on households’ choices with respect to two types of child labor. The model endogenizes the household decision regarding child labor. Its purpose is to illustrate the relationship of different forms of child labor with each other and with household production and assets. I focus on poor economies with simple production technologies and 23 rudimentary markets. For simplicity I assume that all households produce a single good that cannot be saved or stored. Therefore, there is no trade across households, and in each of them consumption equals production. The analysis of this economy can then be addressed from the perspective of any household at a given period of time. There are four factors of production: physical capital (including land), K; adult labor, Q; non-physical child labor, H; and physical child labor, L. They are combined according to a constant-returns-to-scale Cobb-Douglas production function. There are appealing features to the use of Cobb-Douglas; one is that it allows for partial substitutability between production factors (as well as partial complementarity). The second is that I can identify the optimal use of a factor of production as a function of overall production, the marginal product is a constant multiple of the factor intensity times the average product. The third feature is the ability to characterize the factor intensity as a simple parameter, thereby allowing us to do comparative statics. The production function looks as follows: 1 Y AL H Q K           (1) where the factor intensities, , ,, and (1- - - ), are inside the interval (0,1). Moreover, I assume that the production function is so rudimentary that it uses considerably more intensively physical- labor than non-physical child labor. Therefore,   (2) Capital and adult labor are household endowments and, thus, are provided at fixed supplies that are determined exogenously. The two types of child labor are flexible and derived 24 endogenously as a family decision based on utility optimization. The utility function considers two main components that establish a trade-off for child labor: if children worked more intensively, more could be produced and consumed by the family but the children would be less happy and less developed. Specifically, then, the first component of the utility function is the consumption of the single good, Y. For simplicity and in order to consider decreasing marginal utility of consumption, I assume that the utility function is quadratic in consumption, with a positive coefficient on the linear term and negative one on the squared term. The second component is the disutility of child labor, which I model as a linear function of each type of child labor, both with negative coefficients. The utility function is then given by, 2 2 P u MY Y lL hH     (3) where, the parameters M and P and the possibility of production Y are assumed to be such that the household will never reach negative utility of consumption. Families dislike child labor because it takes time and energy away from formal education and leisure, and because it may even hurt the health and normal development of the child. It then stands to reason that non-physical child labor imposes lower utility costs than the physically demanding type. This implies that hl  (4) Optimal child labor, L* and H*, is obtained by maximizing the utility function in equation (4) with respect to the control variables, L and H. Substituting the production function into the utility function and taking the corresponding partial derivatives, 2 2 u Y P Y M Y l L L L           25 YY M PY l LL     And 2 2 u Y P Y M Y h H H H           YY M PY h HH     Then, applying the first-order conditions for utility maximization, 0 u L    and, I obtain the expressions for optimal child labor, 2 * M Y P Y L l    (5) 2 * M Y P Y H h    (6) Figure 2-1 presents these functions, along with some critical points for L*, H*, and Y. It illustrates the inverted-U-shape relationship between child labor and income/production. It also shows the differences and similarities between the two types of child labor: if the production function is sufficiently rudimentary, physically demanding child labor tends to be more prevalent than the non-physical type. Both, however, move in the same direction with respect to changes in income. 26 Figure 2-1: Optimal child labor as a function of income Beyond this graphical presentation, the equations for optimal child labor allow us to formally derive the following conclusions: —First, L* and H* are concave functions of Y, which implies that optimal child labor first increases and then decreases with household production (or income). In fact, the marginal effect of production on child labor is given by, * ( 2 ) L M PY Yl     (7) * ( 2 ) H M PY Yh     (8) The turning point of this marginal is ' 2 M Y P  . Below this threshold, both types of child labor increase as production expands, although at gradually lower rates. Above the threshold, child labor declines as production increases. The intuition for this result is as follows: for very poor families, consumption is so low that they apply child labor to increase production when the opportunity arises (in the form of larger endowments of the other production factors). When families achieve a certain level of income, the cost and grief of child labor start to weigh more 27 than the corresponding foregone consumption, and, therefore, child labor decreases as production opportunities arise. —Second, the impact of the endowment of physical capital, K, and adult labor, Q on optimal child labor, L* and H*, depends on both the level of income, Y, and the relative scarcity of the corresponding factor endowment (i.e., Y/K and Y/Q). Specifically, the marginal effects are given by the following equations. For physically intensive child labor, * (1 ) ( 2 ) LY M PY K K l                        (9) * ( 2 ) LY M PY Q Q l            (10) For non-physical child labor, * (1 ) ( 2 ) HY M PY K K h                        (11) * ( 2 ) HY M PY Q Q h            (12) The mechanism by which changes in K and Q affect L* and H* goes through income: the endowed production factors affect income and consumption, and this in turn determines optimal child labor. Thus, in all cases, the second term in brackets is the marginal effect of income on the respective type of child labor ( * L Y   , and * H Y   ), and the first term is the marginal product of the corresponding endowed factor of production ( Y K   , and Y Q   ). Then, an increase in capital or adult labor would produce a rise in child labor if the household is relatively poor and would lead to a reduction only if it is sufficiently rich. 28 Moreover, the effect on child labor would be larger (in absolute value) if the endowment of the changing production factor (K or Q) is relatively scarce (so that it’s marginal product is large). For illustration, consider the following example. Suppose a household is sufficiently rich (so that Y>Y*), and this wealth is based on a large supply of adult labor despite a low endowment of land capital. The result just presented implies that, for this household, an increase in land capital would have a larger reducing impact on child labor than a rise in adult labor would. — Third, physical child labor, L*, will be larger than non-physical child labor, H*, if the household production function is relatively backward (in the sense of being more intensive in the use of unskilled labor) and the difference in the utility loss from the two types of child labor is sufficiently small (which happens, for instance, when the learning opportunities of non-physical child labor are not substantial). Specifically, taking the ratio of equations (4) and (5), * * Lh Hl    (13) Then, since  is much larger than  , and only a little smaller than l , then * 1 * L H  . Note that although L* and H* are functions of Y, the ratio of L* to H* is only a function of technological and preference parameters. It is important to realize that the utility loss of child labor can be different depending on the child’s gender and age; moreover, this utility loss can also be affected by policy, as in the case of enforced prohibitions of specific types of child labor and pecuniary or in-kind transfers to promote certain others. In order to analyze these alternatives, it is interesting to assess how the size of one type of child labor relative to the other varies with the parameters h and l. Taking partial derivatives of equation (13), * () * 0 L H hl      (14) and 29 2 * () * 0 L h H ll         (15) As expected, the impact of an increase in the cost of non-physical labor is a rise in the ratio L*/H*. Conversely, this ratio will decline if the cost of physical demanding child labor increases. For example, if the cost of L is lower for girls than for boys, then the proportion of girls doing agricultural labor will be larger than that of boys. Also, if the cost of H declines with age (because the opportunity cost of formal education is lower as the child gets older), then the proportion of children doing work using calculation skills at work will be smaller in older cohorts. Finally, if a conditional transfer program provides at the same time an implicit subsidy for children’s time away from work (e.g., by rewarding school attendance) and creates more opportunities for non-physical labor (e.g., by promoting non-traditional business opportunities for the household), then both L* and H* will decline but the ratio L*/H* will increase. — Fourth, the impact of Y on L* is greater in absolute value than the impact of Y on H*, except at the turning point, Y=Y*, at which they are the same. In fact, except at the turning point, the ratio of the partial derivatives of L* and H* with respect to Y is exactly the same as the ratio of L*/H*: * ( 2 ) * ( 2 ) L M PY Yl H M PY Y h          when ' 2 M YY P  , then ** 0 LH YY    . Otherwise, * 1 * L h Y l H l Y h             30 Therefore, for relatively poor families (Y<Y*), physical intensive child labor rises faster than the non-physical type as income increases. By the same token, for relatively rich families (Y>Y*), physically demanding child labor declines faster when income rises. III. Data and Stylized Facts I used a two round panel (2005 and 2006) data from the CCT program introduced in the previous chapter. All details on the program and the 4,200 household data are presented in chapter 1; here I present basic descriptive related to the analysis. The data permits me to observe the same household over a years time which can help separate changes of child labor over time that are attributable to the conditional cash transfer program under evaluation. For this analysis I disaggregate work hours from a dichotomous outcome to actual time (hours) use. The data has detailed information on activity types worked, to include domestic chores. Methodologically, the use of hours versus binary outcomes allows for a more robust estimation approach and more accurate investigation of the research question. The data are structured so that the two work types presented in the model and defined earlier in the chapter can be differentiated: physical labor work in the definition includes farm work, livestock raising for consumption, day laborer as a peon in a farm, water gathering, wood cutting and gathering and household chores such as manning the house, cleaning and caring for siblings. The second type is non-physical work which in the definition is a combination of commercial activity, retail, service, non-agricultural employment and professional activities. From the previous chapter (descriptive tables in appendix A) I can recall that more than 50 percent of children between 8 and 15 years of age work in non-domestic work; the majority in agriculture activities. Domestic work increases labor hours to over 10 hours per week; females working more than boys in these activities. The average household income in dollar amount is $1,400 and the household average size is 7. Communities are on average almost more than 1.5 hours away from the city center which indicates the remoteness of these villages. 31 IV. Empirical Strategy I depart from the basic empirical specifications and present various possibilities that enable me to investigate the research question posed for this chapter. I first establish the relationship of income and all child labor without the program. I then allow for children type to vary in order to spot potential differences in the effect of income on child labor that may arise in the sub-groupings. I analyze how the program affects labor and then analyze how each of the program interventions affects total child labor and child labor by the two types of activities presented in the model. I conclude this part of the analysis by sub-grouping children into types (by age and gender). To ensure that no outliers bias the results I cut 1 percent in the upper tail of the distribution for all (2005 and 2006) continuous variables in the analysis. Some data are missing or made missing in the survey due to sample attrition (2 percent) or digitizing errors; this is a small percent of the total sample and does not affect my findings. All estimations are clustered at the community level, following the data collection design. Lastly I conduct several robustness exercises (bootstrapping and widening of the sample to include younger children) and find the results are unchanged. I begin my exposition of the basic specification by investigating the relationship between income and child labor. I then proceed to investigate how a program that increases household income as well as contribute to human capital opportunities through various investments affects child labor. In the previous chapter I established the success of the randomization which allows me to use the cross section of households in the second period to test the hypothesis that the relationship between income and child labor is linear up to a point and non-linear thereafter. I test the relationship using all three treatment groups and control group in the sample 0 1 2 3 ( , , , ) n t t t t t  . 32 I investigate the effect of income on child labor. The intensity of child labor is measured by the total average hours worked (where LH represents total child labor, L for physical child labor and H for non-physical child labor) in all activities, including domestic work, as a dependent variable. Domestic work accounts for a non-trivial number of children who reported zero hours worked in other non-domestic activities but had more than zero hours in domestic duties. The child labor data has a left hand side censoring problem, particularly for non-physical labor, resulting from our inability to observe negative work hours for children who would normally have less than zero if that were possible. I use a special case of censored regressions (tobit) to accurately calculate the effect of the variables of interest on child labor. I use a tobit for all estimations to keep all results consistent and easily comparable with each other. I first estimate: , 1 , 0 , 0 ihc p h p h p LH Y S         (16) where ,1 ihc p LH represent the dependent variables related to the intensity of child work estimated for child i in household h and community c (cluster) and in the latter period, ,0 hp Y  is household income lagged, ,0 hp S  is a control for household size (later included in the full vector of controls), is a constant and is a normally distributed error term.. In order to test the inverted-U relationship of poverty and child labor I included a squared term for 2 h Y  in all specifications. The sample of children included in the analysis is restricted to kids who are currently 8 through 15 years of age in 2006 (7-14 in 2005) because a great part of kids do not enter primary education until the age of seven and this is the target group for the program. I include a vector of individual ,( 0) ip X characteristics and household ,( 0) hp X attributes in all estimations as controls; all right hand side variables for household and community are lagged for 2005. Community 33 characteristics that affect the supply and demand of child labor are also included in the estimations ,( 0) cp X . Revised equation is: 2 , 1 , 0 , 0 , 0 , 0 , 0 ihc p h p h p i i p h h p c c p LH Y Y X X X                 (17) I explore the gender and age dimensions in the study by looking at sub-samples of young girls and boys (8-12) and older girls and boys (12.1-15). From the qualitative work I know that gender differences exist in terms of typical work activities performed by children in this region of Nicaragua; boys are typically engaged in agricultural work while girls care for the home and help their parents. Young children are typically assigned domestic work and basic agriculture while older children assist adults in more difficult tasks in both non-physical and physical work. I explore this estimation for L and H separately. For the individual characteristics included in the i X vector, I include: age of child, labor is expected to increase with age (Patrinos and Psacharopoulos, 1997; Cartwright and Patrinos, 1999 etc.) and gender of child when appropriate. In h X I include: household size at baseline, number of children in various age cohorts, all affecting the incidence of labor in the child labor literature (Kruger and Berthelon, 2003; Edmonds, 2006; Ponczek and Portela Souza, 2007). For community characteristics c X I include two variables: total population in the community divided by total manzanas of land owned in the community; this variable controls for propensity of agricultural work in the community; and, total number of kids in the 7-14 age range in 2005 to total population in the community; this indicates availability of child laborers in the community. Other controls are age, gender and education of the household head (Dar et. al., 2002; Emerson and Portela, 2007). I also include territorial variables that proxy for remoteness by measuring the proximity to major services and markets. Previous work on poverty in Nicaragua finds that distance deters children, particularly girls and young children, from attending school due to the danger of access particularly during the winter months (Del Carpio, 2007). I add 34 distance, measured in terms of time, to the nearest elementary school, health center and the nearest large market (municipal headquarters). Three quarters of the sample received the program; it is of interest to apply the empirical estimation to investigate the effects of the program, and each of its interventions, on several child labor outcomes. I first include a dummy hh T  (1=treated, 0=control) and then include all three intervention dummies where I change hh T  to 0 1 2 3 ( , , , ) t t t t in the specification, using the non- intervention group (control) as a benchmark. The parameter h  is the coefficient for impact of the program and then each intervention by type. The complete specification with controls is: 2 , 1 , 0 , 0 , 0 , 0 , 0 , 0 ihc p h p h p h p i i p h h p c c p LH T Y Y X X X                    (18) I repeat the empirical exercises in the program section by separating L and H from each other and evaluating the effect of the program on each separately. I expect that the implicit subsidy of the program on children’s alternative activities away from work will reduce child labor; by the same token, I expect that as the new productive opportunities in the household arise it may be viewed as an opportunity for some children, particularly those unbound by the conditionality, to contribute to the households’ production as well as gain some new skills. Therefore I expect hh T  to decrease L* and increase H* for the older cohort. V. Results I present the results of the empirical work in two main parts: (1) the income-child labor relationship, including the analysis for various types of children as presented in the first result in the theoretical section, (2) the effect of the CCT program on child labor, including the analysis of the effect on two types of work activities and a differentiation by various types of children. 35 V.1. Income and all child labor In this section I first present the results from the empirical application of ' 2 M Y P  in the theoretical part. In the model I also find that the utility loss of child labor can be different depending on the child’s gender and age; these are represented in the model by the multiplicative terms that give the intensity of child labor. I develop this part of the model empirically by stratifying the sample into gender and age groups to account for various types of children. V.1.i. Analysis of the effect of income on child labor In the model I assume there is a disutility of child labor, modeled as a linear function of each type of child labor, both with expected negative coefficients. The quadratic term indicates that the relationship between income and child labor has a concave shape (inverted-U) and parents have distaste for child labor, conversely since not all child labor is the same I illustrate the differences in the model and present them as they relate to each other and with household production. As the model shows, when the family is on the left hand side of the inverted-U and to the left of the maximum turning point (poor family), I see an increase in total child labor. A basic empirical exercise (Table A1 in Appendix A) tests the U-relationship using total number of work hours (physical and non-physical work combined) and shows that indeed income is increasing (positive) up to a point and decreasing (negative) thereafter; this relationship is statistically significant when I group all kids together, regardless of type. These results confirm the first finding of the model, that there is a concave function in the relationship between income and child labor that indicates that child labor first increases and then decreases as household production passes the maximum point. This can be interpreted from an economic viewpoint and a statistical standpoint. The economic interpretation is that both coefficients are going in the direction expected for an inverted-U shape, however given the low 36 levels of income of the population sampled in this study, it is only possible to see an inverted-J relationship, where the majority of the households fall to the left side of the maximum point estimate or the point where the curvature takes place. In other words, the sample mean and the majority of the households fall well below the point where an extra dollar of income leads to a decrease in child labor. From a statistical standpoint, the inverted-U relationship is observed in the appropriate signs for all children. In the income figure below (Figure 2-2), the shape of the curve appears to be more or less uniform, as if household incomes for the sample were evenly distributed; however, note that 95 percent of the households fall to the left of the thick solid line ($3,247) and the peak of the curve is at $3,531, beyond the 95 percentile of the distribution. Figure 2-2a shows the shape by each gender. Figure 2-2: Income and predicted child labor (both genders) Note: calculations done on all children 8-15 years of age, Confidence Interval 95 percent. 37 Figure 2-2a: Income and predicted child labor, by gender V.1.ii. Income and child labor by types I am also interested in the effect of the intensity factors  and  which are part of the production function, and l and h which represent the cost of labor in the utility function. This multiplicative term in equations (7) and (8), allow me to distinguish between various types of children (young and older boys and girls) and evaluate how high families maximize labor given their income level. I stratify the sample first by gender, where I observe that the income-labor relationship for girls and boys have an inverted-U shape but statistical significance only holds for girls (Table 2-1); I obtain the same result when I add a full set of control variables. 38 Table 2-1. Analysis of income and total hours worked of all children tobit (1) tobit (2) tobit (1) tobit (2) tobit (1) tobit (2) Income for household (in 1000) 2005 1.258* 1.748** 1.583* 1.947** 0.907 1.446 (1.8) (2.42) (1.88) (2.31) (1.01) (1.58) Income for household squared -0.187* -0.252* -0.270* -0.322* -0.102 -0.158 (-1.86) (-1.74) (-1.66) (-1.91) (-0.59) (-0.9) age of child in 2006 1.372*** 1.248*** 1.483*** (18.74) (12.4) (11.87) gender of child (boy=1, girl=0) 2006 0.323 (0.83) household size 2005 -0.193*** -0.836*** -0.120 -0.337 -0.263*** -1.315*** (-2.69) (-3.21) (-1.39) (-0.97) (-2.59) (-4.02) education level of head 2005 -0.350 0.017 -0.696** (-1.54) (0.07) (-2.46) age of head in 2005 0.052** 0.033 0.074** (2.5) (1.3) (2.54) gender of household head 2005 -0.010 -0.801 0.789 (-0.02) (-1.41) (0.95) # of children under 5 years 2005 1.424*** 0.971* 1.819*** (3.5) (1.82) (3.32) # of children 5-14 years 2005 0.582* 0.076 1.050** (1.74) (0.17) (2.65) # of children 15-24 years 2005 0.113 -0.202 0.417 (0.39) (-0.52) (1.05) dist. in time to municipal hq 2005 -0.052 -0.173 0.048 (-0.24) (-0.76) (0.15) dist. in time to primary school 2005 0.724 1.373* 0.112 (1.05) (1.67) (0.12) dist. in time to health center 2005 0.297 0.393 0.235 (0.9) (1.18) (0.53) tot community owned land/tot population -0.042** -0.058*** -0.024 in community 2005 (-2.26) (-2.65) (-0.93) tot # of kids age group in comm /tot comm -17.689*** -14.30975** -20.058** population 2005 (-2.75) (-2.39) (-2.28) Observations 4256 4200 2083 2053 2173 2147 Pseudo R-squared 0.04% 1.38% 0.05% 1.37% 0.05% 1.5% Note: Absolute value of t statistics in parentheses. Note2:* significant at 10%; ** significant at 5%; *** significant at 1% All kids age 8-15 Girls age 8-15 Boys age 8-15 Total # of hours worked in all activities When I look at the coefficients for boys and girls I observe striking differences between the two genders; boys showing a higher amount than girls. The interpretation is that boys require substantially more income (approximately $1,500) than girls for total labor hours to begin decreasing their labor time. Taking it back to the theory, this can be explained through the differences in types of children and the idiosyncratic costs that each type has and intensity that each type of kid is assigned. The utility loss for child labor will vary by child type; for example is 39 the context in which the child lives biases favoring one gender over the other the utility loss of having one child work versus the other will influence the work allocation strategy of the family. I stratify the sample further, to incorporate age into the mix. In Table 2-2 I observe that the relationship for older girls and boys (age 12.1-15) is concave and significant; the coefficient for the linear term is substantially larger than the quadratic coefficient. An increase in income for older boys appears to have a stronger increasing effect on total labor than that of older girls; however the decreasing effect observed through the negative coefficient on the quadratic term is slightly larger for older girls than for older boys. I can’t say anything conclusively for young kids (ages 8 through 12) because the statistical significance is lacking in the specifications. One can assume that given that most of the households’ income fall well before the maximum point, and only less than 10 percent of households pass the maximum point estimate, the relationship is more like an inverted-J. Particularly for girls and older boys, whose sign on the income coefficients indicate that they follow the inverted-U pattern. 40 Table 2-2. Analysis of income and total hours worked, by age and gender tobit (1) tobit (2) tobit (1) tobit (2) tobit (1) tobit (2) tobit (1) tobit (2) Income for household (in 1000) in 2005 1.138 1.598* -0.148 0.105 1.7186 2.405* 2.461** 3.012** (1.2) (1.68) (-0.13) (0.09) (1.42) (1.9) (2.01) (2.43) Income for household squared -0.184 -0.238 0.049 0.037 -0.324 -0.419* -0.332 -0.383* (-0.91) (-1.14) (0.23) (0.16) (-1.39) (-1.63) (-1.46) (-1.68) age of child in 2006 1.737*** 1.931*** 0.788** 0.828* (9.95) (8.64) (2.05) (1.8) gender of child (boy=1, girl=0) 2006 household size 2005 -0.223** -0.646* -0.225** -0.636 0.028164 0.008 -0.392*** -2.192*** (-2.16) (-1.94) (-1.7) (-1.63) (0.19) (0.01) (-2.8) (-4.72) education level of head 2005 0.029 -0.642 -0.016 -0.766* (0.11) (-1.65) (-0.04) (-1.69) age of head in 2005 0.051* 0.058 0.026 0.097*** (1.66) (1.56) (0.81) (2.33) gender of household head 2005 -0.686 0.503 -0.929 1.188 (-0.98) (0.52) (-1.01) (1.07) # of children under 5 years 2005 1.333** 0.809 0.538 3.029*** (2.52) (1.39) (0.56) (4.03) # of children 5-14 years 2005 0.077 0.217 0.142 2.109*** (0.2) (0.45) (0.17) (3.4) # of children 15-24 years 2005 -0.055 -0.069 -0.393 1.113** (-0.15) (-0.15) (-0.53) (2.03) dist. in time to municipal hq 2005 0.088 -0.281 -0.538 0.431 (0.33) (-0.78) (-1.52) (0.98) dist. in time to primary school 2005 1.546* 0.395 1.136 -0.319 (1.91) (0.36) (0.99) (-0.28) dist. in time to health center 2005 0.175 0.253 0.610 0.257 (0.45) (0.59) (1.38) (0.39) tot community owned land/tot population -0.057** -0.027 -0.065** -0.015 in community 2005 (-2.16) (-1.01) (-2.24) (-0.44) tot # of kids age group in comm /tot comm -17.794*** -13.764 -9.555 -28.898*** population 2005 (-2.63) (-1.38) (-1.06) (-2.59) Observations 1153 1135 1119 1184 930 918 974 963 Pseudo R-squared 0.08% 1.51% 0.05% 1.22% 0.05% 0.37% 0.13% 0.66% Note: Absolute value of t statistics in parentheses. Note2:* significant at 10%; ** significant at 5%; *** significant at 1% Total # of hours worked in all activities Young girls age 8- 12 Young boys age 8- 12 Older girls age 12.1-15 Older boys age 12.1- 15 V.2. The program and child labor Here I evaluate the effect of the program on all child labor by identifying the relationship between the program and each type of intervention on total labor hours. I present the effect of the program on two types of labor activities: physical and non-physical work. These two types of labor activities are derived endogenously as the family decides based on its utility optimization 41 strategy how much to allocate children to each one. In one part of the analysis I hold children type constant (all children 8-15) to isolate the effect of the program on each type of labor activity; I introduce types of children to see how the effect of the program varies. I begin by comparing the performance of each intervention and the control groups on various types of child labor outcome variables. V.2.i. Analysis of the program-child labor relationship I introduce the conditional cash transfer program that benefited three-quarters of the sample. The income-child labor relationship was established in the previous section without controlling for the program, here I seek to evaluate the performance of the program and see if the inverted-U shape persists despite the program. I disaggregate child labor (LH as represented in the empirical part) into two types of child labor, L and H. I know from the model that the impact of an increase in the cost of non-physical work is a rise in the ratio L*/H* and the opposite is also true, if the cost of physical child labor increases. Based on the model presented earlier, I expect the program to reduce child labor because it serves as a tax on child labor and reduces the utility cost of the household. Any transfer that subsidizes education for example, makes education cheaper and the extra money received by the household for abiding to the conditionality increase the opportunity cost by increasing the labor costs, irrespective of labor type (l or h). From the utility function I know that as the cost for l and h rise, all child labor is expected to decrease. When I evaluate each intervention and the corresponding conditionality alone, the amount of child labor by type changes depending on the intervention received by the household and the specifics of the intervention. For example, the basic intervention introduces a subsidy to the household making the cost for labor higher when the child does not attend school; if the household is very poor it will increase its production when the opportunity arises by seeking the optimal level of H* and L* 42 In the case of the business grant intervention, the effect on child labor is different as the intervention introduces a new work opportunity in the household that directly or indirectly motivates child labor in one type of activity H, by lowering the cost of h compared to l. The household will seek to have an optimal level of H* and L*. In Table 2-3 I include the program as a dummy variable (1 for program households and 0 for control households), control for other determinants, and find that the program has a decreasing effect on physical labor and total work hours. The effect is reversed for non-physical work with a positive and significant coefficient, meaning that the program promotes non-physical labor while decreasing physically demanding work. The labor-income relationship exhibits the expected signs but remains significant for total labor hour only. In Table 2-4, I disaggregate the program into its three interventions and find that the training intervention is the only one that significantly decreases physical labor and total labor hours for all kids. The basic intervention has a negative sign for all types of work but none have statistical significance. The third intervention (productive investment/business grant), has a positive effect on non-physical work and no effect on physical labor and total labor hours. 43 Table 2-3. Effect of the program on various types of child labor physical labor non-physical labor total labor hours program household (yes=1, no=0) in 2005 -1.178* 3.504*** -1.102* (-1.93) (4.92) (-1.64) income for household (in 1000) in 2005 1.064 1.1719 1.723** (1.56) (1.18) (2.39) income for household squared -0.153 -0.0513 -0.024* (-1.12) (-0.26) (-1.66) age of child in 2006 1.234*** 0.832*** 1.374*** (19.23) (7.64) (18.76) gender of child (boy=1, girl=0) 2006 0.235 -1.603*** 0.356 (0.66) (-2.98) (0.92) household size 2005 -0.786*** -0.350 -0.843*** (-3.2) (-0.75) (-3.24) education level of head 2005 -0.296 0.450 -0.343 (-1.46) (1.3) (-1.52) age of head in 2005 0.054*** -0.004 0.052*** (2.81) (-0.14) (2.55) gender of household head 2005 0.044 0.000 0.031 (0.08) (0) (0.06) # of children under 5 years 2005 1.448*** 0.482 1.426*** (3.81) (0.7) (3.55) # of children 5-14 years 2005 0.548* -0.389 0.601* (1.76) (-0.65) (1.79) # of children 15-24 years 2005 0.234 -0.360 0.115 (0.84) (-0.66) (0.4) dist. in time to municipal hq 2005 -0.098 0.428 -0.084 (-0.49) (1.18) (-0.38) dist. in time to primary school 2005 0.892 0.628 0.612 (1.38) (0.65) (0.89) dist. in time to health center 2005 0.298 -0.465 0.322 (1.03) (-1.11) (0.99) tot community owned land/tot population -0.039*** -0.043 -0.048*** in community 2005 (-2.32) (-1.2) (-2.59) tot # of kids age group in comm /tot comm -11.340* -8.021 -16.189** population 2005 (-1.86) (-1.11) (-2.44) Observations 4198 4130 4200 Pseudo R-squared 1.34% 3.10% 1.4% Note: Absolute value of t statistics in parentheses. Note2:* significant at 10%; ** significant at 5%; *** significant at 1% All kids age 8-15 44 Table 2-4. Effect of each program intervention on various types of child labor physical labor non-physical total labor hours basic intervention -1.052 1.414 -1.143 (-1.56) (1.47) (-1.52) training intervention -1.407** 1.155 -1.620** (-2.09) (1.41) (-2.25) business grant intervention -1.013 6.288*** -0.533 (-1.58) (7.77) (-0.75) income for household (in 1000) in 2005 1.150* 1.440 1.7902*** (1.75) (1.49) (2.61) income for household squared -0.180 -0.087 -0.261** (-1.41) (-0.47) (-1.92) age of child in 2006 1.241*** 0.834*** 1.381*** (19.25) (7.38) (18.98) gender of child (boy=1, girl=0) 2006 0.268 -1.736*** 0.422 (0.73) (-3.33) (1.08) household size 2005 -0.739*** -0.280 -0.793*** (-2.98) (-0.64) (-3.06) education level of head 2005 -0.255 0.411 -0.311 (-1.31) (1.19) (-1.45) age of head in 2005 0.053*** -0.012 0.051** (2.75) (-0.42) (2.48) gender of household head 2005 -0.149 -0.190 -0.200 (-0.28) (-0.21) (-0.37) # of children under 5 years 2005 1.366*** 0.283 1.311*** (3.58) (0.44) (3.31) # of children 5-14 years 2005 0.511 -0.456 0.558* (1.62) (-0.78) (1.66) # of children 15-24 years 2005 0.194 -0.549 0.057 (0.7) (-1.05) (0.2) dist. in time to municipal hq 2005 -0.103 0.248 -0.095 (-0.51) (0.67) (-0.43) dist. in time to primary school 2005 0.865 0.571 0.626 (1.34) (0.65) (0.92) dist. in time to health center 2005 0.317 -0.227 0.359 (1.08) (-0.54) (1.1) tot community owned land/tot population -0.037** -0.029 -0.046** in community 2005 (-2.13) (-0.85) (-2.34) tot # of kids age group in comm /tot comm -10.905* -6.805 -15.456** population 2005 (-1.78) (-1) (-2.35) Observations 4101 4040 4103 Pseudo R-squared 1.33% 5.11% 1.4% Absolute value of t statistics in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% Child labor by activity types All kids age 8-15 45 VI.2.ii. The program and child labor by types of children I now allow for type of children to vary by age and gender (bringing in the intensity factor in the production function over the cost factor in the utility function) as presented in equations (7) and (8). I expect to see that as the cost of L is lower for one type of child versus another, the proportion of that type of child in participating in physical labor will be greater than the other type of labor. Moreover if the opportunity cost of a non-working activity, such as education, is lower for older children not targeted by the program and are not subject to the school attendance conditionality I expect to see a larger proportion of children in this age cohort participating in work activities. I stratify the sample by gender and then by age and gender. Table 2-5 shows that the basic intervention reduces total child labor hours for boys and increases non-physical labor for girls; all other relationships are not statistically significant; tables 2-6 and 2-7 show results by type of child labor activity. The vocational training intervention reduces physical labor for boys and total labor hours for boys. As expected, I find that the business intervention positively affects non-physical labor for boys and girls; this intervention has no effect on physically demanding labor and total labor hours. 46 Table 2-5. Analysis of the program on various labor activities, by gender basic intervention 2.462** -0.147 -0.992 -1.056 -0.551 -1.675* (2.17) (0.11) (1.27) (1.26) (-0.66) (-1.78) training intervention 1.240 0.974 -1.127 -1.598* -1.126 -2.073** (1.31) (0.75) (1.54) (1.76) (-1.48) (-2.06) business grant intervention 5.766*** 6.852*** -0.812 -1.185 -0.076 -0.989 (6.74) (5.76) (1.17) (1.36) (-0.1) (-1.01) income for household (in 1000) 2005 1.394 1.491 1.438* 0.789 1.899** 1.6252* (1.22) (1.03) (1.81) (0.91) (2.28) (1.81) income for household squared -0.127 -0.004 -0.251 -0.089 -0.311* -0.196 (0.60) (0.01) (1.54) (0.54) (-1.82) (-1.16) age of child in 2006 0.782*** 0.905*** 1.160*** 1.311*** 1.247*** 1.504*** (5.14) (4.88) (11.69) (11.66) (12.28) (11.77) household size 2005 0.133 -1.038 -0.298 -1.155*** -0.335 -1.222*** (0.28) (1.62) (0.90) (3.52) (-0.96) (-3.77) education level of head 2005 0.830** -0.084 -0.002 -0.496** 0.027 -0.622** (2.47) (0.15) (0.01) (2.07) (0.11) (-2.32) age of head in 2005 0.023 -0.062 0.031 0.078*** 0.031 0.073** (0.74) (1.20) (1.37) (2.85) (1.24) (2.5) gender of household head 2005 -0.331 -0.046 -0.960 0.686 -0.984 0.561 (0.29) (0.04) (1.62) (0.83) (-1.63) (0.66) # of children under 5 years 2005 -0.161 1.114 0.910* 1.767*** 0.886* 1.668*** (0.26) (1.19) (1.80) (3.26) (1.66) (3.08) # of children 5-14 years 2005 -0.720 0.139 0.067 0.904** 0.045 1.014*** (1.13) (0.17) (0.15) (2.43) -(0.1) (2.63) # of children 15-24 years 2005 -0.721 -0.160 -0.099 0.467 -0.206 0.300 (1.28) (0.23) (0.26) (1.19) (-0.52) (0.75) dist. in time to municipal hq 2005 0.274 0.234 -0.196 -0.023 -0.178 -0.026 (0.74) (0.44) (0.98) (0.07) (-0.76) (-0.08) dist. in time to primary school 2005 0.942 -0.061 1.472* 0.282 1.342* -0.038 (0.82) (0.04) (1.91) (0.35) (1.66) (-0.04) dist. in time to health center 2005 -0.431 0.037 0.383 0.277 0.452 0.307 (0.87) (0.07) (1.22) (0.68) (1.36) (0.69) tot community owned land/tot population -0.042 -0.016 -0.054** -0.021 -0.059*** -0.030 in community 2005 (1.23) (0.32) (2.44) (0.85) (-2.57) (-1.12) tot # of kids age group in comm /tot comm -8.402 -5.017 -7.596 -13.274 -12.288** -17.518* population 2005 (0.94) (0.45) (1.40) (1.57) (-2.1) (-1.88) Observations 1972 2068 2002 2099 2004 2099 Pseudo R-squared 4.50% 6.47% 1.38% 1.41% 0.0139 1.57% Absolute value of t statistics in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% Boys age 8-15 Total labor hours Girls age 8-15 Boys age 8-15 non-physical labor Girls age 8-15 Girls age 8-15 Boys age 8-15 physical labor 47 Table 2-6. Analysis of the program on non-physical labor, by child type basic intervention 3.526** -3.615* 1.858 2.651 (2.15) (1.77) (1.43) (1.55) training intervention 1.737 0.486 0.948 1.830 (1.17) (0.30) (0.79) (1.02) business grant intervention 6.311*** 5.776*** 5.393*** 8.484*** (5.17) (4.27) (4.31) (5.39) income for household (in 1000) 2005 -0.058 1.976 2.337 1.456 (0.04) (1.00) (1.58) (0.84) income for household squared 0.072 -0.054 -0.263 -0.031 (0.26) (0.16) (0.91) (0.09) age of child in 2006 1.056*** 2.286*** 0.276 1.204** (3.05) (4.17) (0.60) (2.58) household size 2005 0.622 -0.349 -0.276 -1.726* (1.07) (0.51) (0.42) (1.87) education level of head 2005 0.738 -0.495 1.003** 0.380 (1.59) (0.69) (2.17) (0.60) age of head in 2005 -0.027 -0.061 0.071 -0.052 (0.67) (1.13) (1.59) (0.77) gender of household head 2005 -1.071 1.091 0.049 -0.886 (0.73) (0.62) (0.04) (0.58) # of children under 5 years 2005 -0.521 0.409 0.116 1.632 (0.60) (0.37) (0.14) (1.52) # of children 5-14 years 2005 -1.250* -0.544 -0.142 0.724 (1.76) (0.58) (0.16) (0.65) # of children 15-24 years 2005 -1.292* -0.896 -0.228 0.496 (1.66) (1.03) (0.32) (0.51) dist. in time to municipal hq 2005 0.078 -0.045 0.409 0.494 (0.17) (0.08) (0.78) (0.70) dist. in time to primary school 2005 -0.828 0.864 1.897 -1.388 (0.53) (0.74) (1.37) (0.72) dist. in time to health center 2005 0.026 0.477 -0.756 -0.420 (0.04) (0.74) (1.16) (0.49) tot community owned land/tot population -0.013 -0.043 -0.074* 0.007 in community 2005 (0.33) (0.76) (1.71) (0.10) tot # of kids age group in comm /tot comm -17.387 -18.041 -0.635 7.366 population 2005 (1.51) (1.36) (0.05) (0.56) Observations 1101 1144 871 924 Pseudo R-squared 5.28% 8.64% 3.89% 7.00% Absolute value of t statistics in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% Older boys age 12.1-15 Total # of hours worked in non-physical labor Young girls age 8- 12 Young boys age 8- 12 Older girls age 12.1-15 48 Table 2-7. Analysis of the program on physical labor, by child type basic intervention -1.123 -1.160 -0.879 -0.691 (1.31) (1.30) (0.82) (0.55) training intervention -1.818** -1.682* -0.346 -1.360 (2.27) (1.80) (0.34) (1.04) business grant intervention -1.325* -0.298 -0.196 -1.964 (1.70) (0.32) (0.19) (1.52) income for household (in 1000) in 2005 1.372 -0.395 1.540 2.197* (1.46) (0.36) (1.40) (1.95) income for household squared -0.203 0.107 -0.302 -0.323 (0.94) (0.47) (1.36) (1.59) age of child in 2006 1.594*** 1.682*** 1.002*** 0.615 (9.92) (8.62) (2.63) (1.48) household size 2005 -0.586* -0.652* 0.050 -1.770*** (1.85) (1.78) (0.08) (3.83) education level of head 2005 -0.047 -0.454 0.040 -0.571 (0.20) (1.38) (0.11) (1.56) age of head in 2005 0.046 0.067* 0.029 0.089** (1.63) (1.95) (0.99) (2.21) gender of household head 2005 -0.514 0.680 -1.478 0.784 (0.79) (0.70) (1.57) (0.77) # of children under 5 years 2005 1.341*** 0.904 0.325 2.769*** (2.60) (1.64) (0.35) (3.55) # of children 5-14 years 2005 0.030 0.381 0.172 1.557*** (0.08) (0.87) (0.21) (2.65) # of children 15-24 years 2005 -0.006 0.028 -0.247 1.051* (0.02) (0.06) (0.34) (1.95) dist. in time to municipal hq 2005 0.053 -0.319 -0.545* 0.259 (0.22) (0.93) (1.71) (0.56) dist. in time to primary school 2005 1.783** 0.481 1.134 0.097 (2.11) (0.47) (1.13) (0.10) dist. in time to health center 2005 0.168 0.260 0.565 0.414 (0.44) (0.65) (1.34) (0.66) tot community owned land/tot population -0.058** -0.028 -0.052* -0.005 in community 2005 (2.28) (1.03) (1.80) (0.17) tot # of kids age group in comm /tot comm -10.173* -6.568 -4.321 -22.459** population 2005 (1.69) (0.70) (0.50) (2.14) Observations 1108 1158 894 941 Pseudo R-squared 1.53% 1.15% 0.41% 0.62% Absolute value of t statistics in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% Total # of hours worked in physical labor Young girls age 8-12 Older girls age 12.1-15 Older boys age 12.1-15 Young boys age 8- 12 49 I look at young and older children by gender and activity type and find that all interventions have positive coefficients for non-physical labor and negative coefficients for physical labor. The basic intervention has an increasing effect on young children of both genders in non-physical activities and the business intervention has an increasing effect on children of all ages and both genders on the same activity. With respect to physical labor, the vocational training intervention has a reducing effect for young children and the business grant intervention has a reducing effect for young girls. VI. Conclusion I began this chapter by noting that good theoretical foundation combined with empirical applications is not commonplace in the child labor literature due mostly to data restrictions. In this chapter I attempt to provide a model that is applied to a rich dataset collected for a randomized experiment to answer whether child labor always increases with income; and, how a productively focused CCT with three distinct interventions affects child labor? When I use total number of work hours the data shows that indeed income is increasing up to a point and decreasing thereafter; this relationship is statistically significant when I group all kids together, regardless of type. I conclude that the income and child labor relationship is concave but the heterogeneity in income of the sample indicates the actual shape; an inverted-J in samples with mostly poor households (left of the turning point) like the one analyzed here instead of an inverted-U. I observe that the income effect on labor for girls and boys exhibits an inverted-U shape; statistical significance however only holds for girls. The stratification by gender also allows us to observe a large difference in the income level necessary for labor to begin decreasing between the genders, girls requiring far less total household income than boys for income-labor relationship to be negative. Once I analyze kids by gender and age group combined I find that the concavity observed earlier remains strong for older kids but becomes ambiguous, in terms of statistical 50 significance, for younger children. The model provides a plausible explanation, families may look for the child with the highest potential to contribute to the households’ production and assign it a high labor intensity factor. In general, older children are comparatively more productive than younger children thereby making their intensity factor higher and the whole labor intensity ratio higher. When I include the CCT program into the analysis, I find that it serves as a tax on child labor making it less appealing for parents to send their children to work. When I disaggregate child labor into physical and non-physical work, I find that the program helps decrease physical work and increase non-physical labor among all children. In other words, the program makes the cost of physical work higher while unaffecting or decreasing the cost of non-physical labor. Specific components of the program have diverse effects. Children bound by the conditionality of the program and receiving either the vocational or business grant interventions experience a decrease in physical work; while children receiving the basic intervention do not appear to have such an effect. Conversely, the business grant intervention and the basic intervention increase non-physical labor for the younger cohort. Given the objectives and mechanisms of the program of providing a subsidy for school- age children’s time away from work while rewarding school attendance and creating opportunities for intellectually driven activities in the household (encompassed in non-physical work), I find that the intended effect of the program spills over children. In other words, the decline in physical labor and total labor for children and increase in non-physical labor indicates a change in the labor composition of the household, even for children. 51 Chapter 3: Leveling the Intra-household Playing Field: Compensation and Specialization in Child Labor Allocation Child labor in developing countries is a topic of debate and concern for many policy makers. The literature on child labor has discussed the complicated trade-offs that are often involved in parental decision-making on child labor, and has shed light on how various household characteristics and different contexts might affect such decisions (Basu, 1999; Edmonds, 2007). Less is known about child-specific characteristics parents take into account when assigning responsibilities for the various work tasks within the household. Parents’ decisions could either reinforce existing differences between children by investing more in those children that have accumulated higher human capital or more natural or social endowments, or they could compensate for deficiencies through targeted investments. A positive shock might lead households to compensate for or reinforce existing differences. To the extent that such investments compete with, or possibly complement, children’s time working, we would expect this to be reflected in the intra-household child labor allocations. This chapter therefore aims at analyzing how the allocation of tasks among children within a household changes in response to a social program. It does so by first presenting a simple theoretical model on the relationship between child characteristics and potential occupations in the presence of external interventions (such as a cash transfer program). This is followed by the main contribution of the chapter, which consists of an evaluation of Nicaragua’s Atención a Crisis program, a conditional cash transfer intervention. The particular focus of this evaluation relates to the impacts on intra-household child labor allocation in beneficiary families. Thus, the chapter shows that child-specific characteristics can help shed light on the allocation tasks among children in a household. It then analyzes whether the exogenous shocks created by a social program resulted in compensation or reinforcement of pre-program differences in child labor 52 allocation and human capital accumulation. As such, it emphasizes the heterogeneity of program impact within the household, including secondary program effects resulting from the reallocation of child labor between children of the same household. In particular, we analyze whether such reallocation helped to compensate, or rather exacerbate, disadvantages of certain types of children within the households, considering various categories based on gender, age and academic achievement. As mentioned above, the program we analyze is called Atencion a Crisis, a conditional cash transfer program (CCT) in Nicaragua. Women in randomly selected treatment communities received a bi-monthly sizable cash transfer conditional on school enrollment and attendance of primary school children. A random subset of the beneficiaries in addition received a grant aimed at increasing their productive capacity in a nonagricultural activity. The empirical identification strategy in this chapter relies on this two-staged randomized design to analyze the various factors that might affect child labor allocation. To understand the intra-household allocations specifically, we use a household fixed effects model. This allows controlling for the many observed and unobserved household characteristics that could affect child labor. More interestingly, the use of a household fixed effects approach allows looking within the household to investigate whether and how child labor gets reallocated between siblings, when a program relaxes budget constraints and imposes conditionalities on children’s schooling. The use of the fixed effects implies that the sample considered only includes households that have at least 2 children, but these are exactly the households for which reallocation between children is relevant. To my knowledge, there are only two papers that consider intra-household child labor reallocation in the context of cash transfer programs. Filmer and Schady (2008) show that a conditional fellowship program targeted at individual children in Cambodia reduced child labor for eligible children, and did not affect work of ineligible siblings. On the other hand, analyzing impacts of a conditional cash transfer program targeted at individual children in Colombia, 53 Barrera-Osorio et al. (2008) find negative spillover effects on other children in the household. This chapter differs as we do not address whether parents reallocate child labor to children not affected by the conditionality; instead, we consider whether the program leads to a reallocation of child labor that helps compensate, or instead exacerbates, past disadvantages by age, gender, and past academic achievement. Other studies that analyze intra-household differences in child labor mainly focus on heterogeneity by birth order (Behrman and Taubman, 1986; Edmonds, 2006; Ejrnaes and Portner, 2004; Emerson and Souza, 2008; Manacorda, 2006). Intra-household heterogeneity along other characteristics has received less attention. Yet, there is a large related literature on intra- household differences in investments in education, health and nutrition (Behrman, Pollack and Taubman, 1986; Das Gupta, 1987; Rosenzweig and Schultz, 1992; Foster, 1995). Differences in intra-household bargaining power between spouses often can lead to differences in resource spending between children of the same household (Thomas, 1997; Duflo, 2003). This can be reinforced by households’ coping mechanisms in the face of negative shocks (Behrman, 1988; Behrman and Deolalikar 1988). Vice versa, a positive income shock can help to compensate for existing differences (Rose, 1999; Mansuri, 2006). And expected return can affect parental decisions. Rangel (2008) provides striking evidence that skin color helps explain differences in human capital investments between siblings in Brazil. In this chapter, we first consider whether the program helped to compensate for existing imbalances in child labor along gender lines. We then analyze whether the program helped parents to compensate for lags in academic achievement by reducing child labor more for children with lower past academic achievement. Both gender and past academic achievement are factors that parents are likely to take into account when considering the returns to schooling and the assignment of different child labor tasks within the household. 54 If parents decide on child labor allocations according to the expected returns, existing specialization patterns in society that require boys and girls to be prepared for different types of tasks might matter for child labor allocation. However, in the presence of a social intervention, the expected returns can change, potentially favoring children who would not be favored in regular circumstances. This is particularly the case given that CCTs give families direct transfers conditional on all primary-school-age children attending school. In addition, in households that received the CCT and the productive investment grant, women are likely to dedicate themselves more to nonagricultural economic tasks. This might increase return to girls’ labor in these nonagricultural activities, but also in domestic tasks to the extent they need to substitute for mother’s work. Parents also may consider past academic achievement when considering returns to both specific child labor tasks and returns to further schooling, which can compete with child labor. Past academic achievement is likely to capture a combination of innate ability and accumulated skills, which might be the result of past disadvantages, negative shocks, and investment decisions that could have affected different children in the households differently. It is a priori unclear whether the positive program income shock linked with the conditionalities would lead to compensation or reinforcement of existing differences. Parents may reduce child labor more for children without lags and exacerbate intra-household differences, or they could decrease child labor more for children with existing lags, possibly helping to compensate for past delays. While academic achievement is a measure of both innate ability and accumulated skill, this question relates to the debate on parents human capital decisions as a response to innate abilities. Becker (1991) predicts that parents will invest more in the human capital of abler children, but, in the case of rich families, parents make compensatory transfers to less able children. Empirical results for the US are mixed, as some find that parents compensate for deficiencies in children’s endowments or prejudices from cultural biases (Becker and Tomes, 1976; Wilhelm, 1996; 55 Ashenfelter and Rouse, 1998; Ermish and Francesconi, 2000), while other results suggest that investments by parents reinforce genetic endowment (whether intelligence or gender) and/or labor-market biases (Kim, 2005). This chapter first shows that child labor is distributed unequally within the household, and that there appear to be clear patterns of specialization. In particular, while total amount of hours worked do not differ significantly by gender, boys work more hours than girls in economic activities, and this difference comes primarily from work in agriculture. Girls work more in domestic activities and livestock. The trade-off between schooling and work appears to be stronger for boys, possibly due to the specialization in agriculture that competes, in terms of when the activity takes place, more directly with schooling. Intriguingly however, children that are attending school are working more hours in nonagricultural activities than their siblings. The chapter then shows that the program helped compensate for some of these intra- household differences, but exacerbated others. In particular, it reduced total hours worked more for older boys, and for boys with low past academic achievements. Households that, in addition to the CCT, randomly received the productive investment grant show an increased specialization of older girls in nonagricultural and domestic work, but no overall increase in older girls’ child labor. By analyzing child labor patterns in response to Atencion a Crisis, this chapter contributes to the more general literature of the effect of CCTs on child labor. In general, the findings of this literature appear to be mixed. While Skoufias and Parker (2001), Yap, Sedlacek, and Orazem (2008), and Filmer and Schady (2009) find relatively large reducing impacts for programs in Mexico, Brazil and Cambodia, Attanasio et al. (2006) and Glewwe and Olinto (2004) find no significant impacts for Colombia and Honduras. Often there appears to be marked heterogeneity in program impacts, both by child characteristics, and by the type of work considered. In Mexico, the largest overall impacts were found for older boys, while there was a 56 reduction in domestic work for girls (Skoufias and Parker, 2001, see also Djebbari and Smith, 2008). In Colombia, where there is no overall impact on child work, the program did lead to a reduction of time allocated to domestic chores (Attanasio et al, 2006). In Ecuador, impacts on child labor are especially large for those children vulnerable to transitioning from school to work, with impacts concentrated in work-for-pay outside of the home (Edmonds and Schady, 2008). For the specific case of Nicaragua, the Red de Proteccion Social (RPS), has been shown to reduce child labor substantially (9%) and more so for older children. Dammert (2009) has analyzed heterogeneity of impacts of RPS along household and community welfare indicators. This chapter contributes to, and is distinct from, the above literature, by focusing on the intra-household reallocation. The chapter further differs from most previous work because the program has three different intervention packages, which allows shedding further light on specialization patterns within the household, when more economic activities become available together with higher income from the transfer program. Macours and Vakis (2009) show that beneficiary households who received the productive investment grant in addition to the CCT had indeed higher incomes from nonagricultural self-employment. Del Carpio (2008) shows that the program led to a shift of child labor to such nonagricultural activities, and an overall decline in total child labor hours. In related findings, Macours and Vakis (2005) show that the program increased school enrollment and attendance, while Macours, Schady, and Vakis (2008) show parents also increased investment in preschool children, leading to improvements in cognitive development. As has become common in the child labor literature, we consider not only the total amount of hours worked, but also the composition of labor by disaggregating work in hours by various non-domestic and domestic activities. Edmonds (2006) and Kruger and Berthelon (2007), among others, have demonstrated that the inclusion of domestic work can be key to shed light on gender differences, as girls might be disproportionately assigned to domestic tasks. The 57 differentiation between different tasks is also important as some tasks are more likely to compete with schooling in terms of timing, while parents could consider experience in other tasks as complementary to human capital investment through schooling. This could be the case because some child labor might result in learning new skills---such as counting and handling money in a small shop or engaging in commerce while selling goods in the community (Edmonds, 2007), or learning about agricultural practices which might increase future returns in agriculture (Beegle, Dehejia, and Gatti, 2006). Another contribution of this chapter is a simple theoretical model of family choice regarding the intra-household allocation of child labor. Keeping the empirical emphasis of the chapter, the theoretical model is introduced with the purpose of motivating the empirical exercise, formulating the main hypothesis of interest, and interpreting the results. Thus, the model is developed to examine the relationship between specific child characteristics (such as gender and ability) and the choice of potential child occupations (e.g., school, agricultural and non- agricultural labor, and household chores). Moreover, the model serves to clarify how various types of interventions (e.g., cash transfers and training) affect the family’s decision on children’s occupations. The interaction between child characteristics, returns to potential occupations, and external interventions renders a pattern of economically optimal family decisions. The model is built in steps of complexity to highlight the role of each element in the final labor allocation of the family. The rest of the chapter is organized as follows. Part I presents the theoretical model and derives its main implications. Part II provides the necessary background on the Atención a Crisis program and its randomization strategy. In part III we turn to the data, discuss the patterns in intra-household labor allocation, and present the main hypotheses related to the program impact. In part IV we present the main results of the chapter, and show how the program impacts differ 58 between siblings with different gender and with differences in past academic achievements. Part V concludes. I. Theoretical Framework We now present a simple theoretical model of the family decision regarding children’s occupation. We develop the model in increasing complexity. In the first case, we allow for only schooling and one type of child labor. In the second case, we consider two types of child labor, each suited to different child characteristics. And in the third case, we allow for one of the types of child labor to be skill-enhancing. For simplicity, each family is assumed to live for two periods. In the first, children work or study and acquire skills; and in the second period, children’s skills are used to produce income. In this simple setting, the objective of the family is to maximize the present value of income brought by each child. This income can be pooled together with other sources of income and then distributed among members of the family, but the model is silent regarding this stage of welfare maximization. Although some of the findings of the theoretical model have implications beyond those explored in the empirical section, the model helps motivate the empirical exercise by exploring the child-labor decision-making process, relating it to child characteristics, potential occupations, and external interventions (such as the CCT evaluated in this chapter). Case 1. Assume that in order to generate income from a given child, the family has one type of child labor activity and one comprehensive set of productive assets, which may include adult labor, land, and capital. In the first period, the production function is represented as follows, ps l Lf Y l f l   ln 1 (1) where, Y 1 is income in the first period. L represents family endowments such as adult labor and productive assets in the household (e.g., tools, land, and capital). Note that a transfer program focused on increasing productive assets in the household is likely to augment this factor. In this 59 first case, there is only one type of child labor, l and l f represent the characteristics of the child (e.g., gender, ability etc.) that make him or her productive. Also in this case, schooling, s, is the only skill-enhancing activity; and p is the cost associated to schooling. Note that in the context of a CCT, the cost of schooling can be purposely reduced. Finally, we assume that labor and schooling exhaust the total time available to the child, normalized for simplicity to 1. That is, labor and schooling compete for child’s available time, and 1  l s . In the second period, the production function can be represented as follows, 2 Y Rs   (2) where 2 Y is income in the second period, R is the rate of return to schooling, and  represents the percentage transfer of income from children to the family in the second period. In this first case, the only control variable is the child’s labor, l, in the first period. Then, the family’s optimization problem with respect to this child is 12 max l YY   , or ) 1 ( ) 1 ( ln max l R l p l Lf Y l l       (3) where,  represents the rate of time preference, indicating the degree of patience for consumption in the future. The first order condition of this maximization problem is, 0 * 1             R p l Lf l Y l  (4) which leads to the solution for optimal child labor, l*, p R Lf l l    * (5) 60 From this result, we can see that increases in productive assets, the productivity of children, and the cost of education lead to increases in child labor in the first period. The opposite is the case when the family has more patience, the expected return on schooling increases, and the expected portion of second-period income to be shared with the family rises. In preparation for the empirical exercise, let’s contextualize these results into a more realistic scenario. Suppose the family lives in an agricultural setting and owns some land (positive L). In this rural area, education is likely to be relatively expensive and of poor quality (high p and low R). If family income is low, they are bound to be impatient and prefer to increase their consumption in the present (low  ). In addition, suppose that the child is a boy, strong and qualified for tough agricultural work (high f l ). According to the model, all these characteristics point to a high allocation of labor to the boy in question. Case 2. Given that our empirical exercise aims to differentiate across types of child labor, we now augment the theoretical model to include two types of labor: the first is physically- demanding labor (l), and the second one is skilled-oriented labor (h). Each type of child labor requires a specific asset to generate production, respectively L and H. In the context of our empirical exercise, L would correspond to land and agricultural tools, while H would represent business or service-related assets. In this case, the production function is given by, ps h Hf l Lf Y h l    ln ln 1 (6) Rs Y   2 (7) and the time constraint per child is, 1    h l s (8) 61 The family’s optimization problem is to maximize lifetime income per child, with respect to each type of labor, ) 1 ( ) 1 ( ln ln max , h l R h l p h Hf l Lf Y h l h l          (9) The first-order conditions and corresponding optimal child labor are, therefore, 0 * 1             R p l Lf l Y l  p R Lf l l     * (10) 0 * 1             R p h h Hf h Y  p R Hf h h     * (11) From these expressions, we can derive the optimal relationship between physical and skilled labor: h l f H f L h l    * * (12) Assuming an interior solution, this formula indicates that the optimal ratio of physical to skilled child labor depends on both family endowments of complementary assets and child- specific abilities. For example, a smart girl (high f h , possibly low f l given her gender) who lives in a household that owns a town store whose operations require some math and communication skills (high H) would likely be allocated to more skilled than physical labor. Assuming an interior solution, this formula indicates that the optimal ratio of physical to skilled child labor depends on both family endowments of complementary assets and child- specific abilities. For example, a smart girl (high f h , possibly low f l given her gender) who lives in 62 a household that owns a town store whose operations require some math and communication skills (high H) would likely be allocated to more skilled than physical labor. Case 3. As reviewed in the introductory section, the literature on child labor finds that certain labor activities are skill enhancing and can serve in the adult life of the child. To take this into account, case 3 extends the basic model to allow for one type of child labor (h) to compete with schooling in the formation of skills. Apart from added realism, this extension is important because the CCT under evaluation in this chapter affects the optimal choice of child labor by reducing the cost of schooling (p) and, in some cases, increasing household non-agricultural assets (associated with H). That is, in order to understand the effects of the CCT, we need a model that considers not only two types of child labor but also the possibility that one of them be skill-enhancing. In terms of our mathematical formulation, this extension is captured in the production function of the second period. Specifically, ] [ 2 h R s R Y h s    (13) where R s and R h represent, respectively, the rate of return to schooling and skilled child labor. The family’s maximization problem per child becomes, h R h l R h l p h Hf l Lf Y h s h l h l            ) 1 ( ) 1 ( ln ln max , (14) The first-order conditions and corresponding optimal labor allocations are, 63 0 * 1             s l R p l Lf l Y  p R Lf l s l     * (15) 0 * 1              h s R R p h h Hf h Y   p R R Hf h h s h     ) ( *  (16) The optimal allocation of schooling now depends not simply on the returns to schooling but on the difference between the returns to schooling and the returns to skilled labor. Considering the time constraint and the first-order conditions, optimal schooling is given by, p R R Hf p R Lf s h s h s l       ) ( 1 *   (17) Clearly, given competing child occupations, the chosen level of schooling will only be significant if the returns to schooling are sufficiently high and its costs sufficiently low. Therefore, an intervention can be successful at achieving higher schooling attendance if it works through these two margins, education returns and costs. However, if it affects other aspects of the problem (by, for instance, providing assets that complement child labor), its impact on schooling may be ambiguous. Coming back to the allocation of child labor in different activities, the model allows us to identify the relationship between physical and skilled labor. This is given by the ratio,                          p R R f f H L h l s h h l   1 * * (18) This indicates that when child labor has the potential of enhancing the skills that can be used in adult life, then the relative allocation across types of child labor not only depends on 64 family endowments and child characteristics but also depends on factors affecting the skill- forming decision. In fact, whereas the first two terms of the right-hand side of the equation refer to present family and child characteristics, the latter term involves the calculation of the benefits and costs of skill formation for the future of the child. In summary, the schooling and child labor decisions depend on a combination of family characteristics (such as asset endowments and patience), social characteristics (such as cost of schooling and returns to skilled child labor), and child characteristics (such as gender and ability). An intervention can affect these decisions through several possible channels. Depending on what channels are affected, therefore, an intervention can have different results on child labor and schooling outcomes. This theoretical analysis highlights the complexity of the child labor decision and guides the specification of the empirical evaluation. II. Background on the Program The Atención a Crisis program was a one-year pilot program implemented between November 2005 and December 2006 by the Ministry of the Family in Nicaragua. The program was implemented in the aftermath of a severe drought and had two objectives. First, it aimed to serve as a short-run safety net by reducing the impact of the aggregate shock on human and physical capital investments. This was facilitated via cash transfers, which were envisioned to reduce the need for ex-post, adverse coping mechanisms, such as asset sales, taking children out of school or reductions in food consumption. Second, the program also intended to promote long run upward mobility and poverty reduction through asset creation by enhancing households’ asset base and income diversification capacity. In order to achieve these objectives, and building on the already existing and successful conditional cash transfer model in Nicaragua (Red de Protección Social - RPS), the program introduced 3 different packages in order to evaluate and compare the effectiveness of each to reach the objectives stated above. Specifically, a total of 3,000 households were selected to 65 participate in the program. These households were allocated one of three different packages through a participatory lottery, organized in each community: (i) the basic CCT; (ii) the basic CCT plus a scholarship for an occupational training; and (iii) the basic CCT plus a grant for productive investments. All selected beneficiary households received the basic CCT consisting of cash transfers conditional on children’s primary school and health service attendance. The school conditionality specifically implied that older children, who had not yet completed primary school, had to enroll and regularly attend school. Note that while children in principle can finish primary school by the age of 12, few children do, which is why the program included older children in the conditionality in the first place. In the data, the schooling conditionality is binding for 88 percent of children between 7 and 15. School enrollment and attendance were monitored by the ministry, through data received from the primary school teachers, and this monitoring was successfully implemented (Aguilera et al., 2006). In addition to the CCT, one third of the beneficiary households also received a scholarship that allowed one of the household members to choose among a number of vocational training courses offered in the municipal headquarters. However, due to implementation delays, the vocational training courses had not started yet at the moment of the follow-up survey. Finally, another third of the beneficiary households received, in addition to the basic CCT, a grant for productive investments aimed at encouraging recipients to start a small non-agricultural activity with the goal of asset creation and income diversification. This grant was conditional on the household developing a business development plan, outlining the investments outside of subsistence farming in new livestock or non-agricultural income generating activities. This package included technical assistance and training in basic commercial skills. Henceforth, the term ―productive investment package‖ refers to the entire package received by this group of households (the combination of the CCT, the productive investment grant, and the technical 66 assistance and basic commercial training). The beneficiaries of this productive investment package had received the largest amount of benefits at the moment of the follow-up survey: 2 to 3 months before being surveyed they had received $175 to invest. II.1. Program randomization The program was targeted to 6 municipalities of the drought region in the northwest of Nicaragua. These were municipalities that met both criteria of having been affected by a drought the previous year and by the high prevalence of extreme rural poverty based on the national poverty map. From the list of all communities in the 6 municipalities, 56 intervention and 50 control communities were randomly selected through a lottery to which the mayors of the 6 municipalities were invited to attend and participate. Baseline data were then used to define program eligibility based on poverty and vulnerability, resulting in the identification of 3,000 households to participate in the program. Finally, from each eligible household, the female household member that was reported as the children’s primary caregiver was invited to a registration assembly. At the end of each assembly, all the beneficiaries participated in a lottery process through which the three packages described above were randomly allocated among the eligible households. The random assignment was successfully implemented. Table 3-1 presents the randomization results for the sample of eligible households relevant for the analysis in this chapter, i.e., households with at least 2 children between 6 and 15 years old. The differences between households in treatment and control communities are small and not statistically significant. Similarly, the differences between households with the productive investment package and households in the control communities are generally small and not significant. Finally, take-up of the overall program among eligible households was 95 percent, with the main selection due to exclusion by leaders (see footnote 6) and some outmigration. Take-up of the productive investment grant among households in the program was near 100 percent. 67 III. Data, Descriptive Patterns, and Hypotheses The data come from a household panel in the control and treatment communities. In treatment communities, data were collected from all households. In control communities, a random sample of households was selected to obtain a control group of equal size as each of the three intervention groups. The follow-up data was collected 9 months after the start of the program. The attrition rate of the second round was 1.3 percent of the original households. Attrition is uncorrelated with treatment—in a regression of an indicator for attrited households on a dummy for treatment the coefficient is -.004, with a standard error of .005. The main household survey, collected in both rounds, contains household and individual level data on various socio-economic indicators on approximately 4400 households. In the follow-up survey, additional questions on child labor were added, to capture children’s work in chores (wood and water gathering) and domestic work. The quantitative data was complemented with qualitative work, based on focus groups and semi-structured interviews with a wide set of beneficiaries and other local actors in treatment and control communities (see Aguilera et al., 2006). The data in Table 3-1 characterizes the socio-economic context, which helps frame the results of this study. Households are very poor, with average expenditure per capita around 250 US$ per year, and household heads have on average less than 3 years of education. Almost all households dedicate themselves to semi-subsistence agriculture (about 90 percent of households is self-employed in agriculture), and about 50 percent have some livestock. Nonagricultural self- employment is relatively rare, while around 20 percent of households have income from nonagricultural wage work. More households complement their self-employment income with wage work in agriculture, which often occurs in other regions of the country and leads to seasonal migration of many adult men, as well as some adult women (Macours and Vakis, 2008b). The 68 dependence on agriculture and the temporary absence of men clearly might affect the demand for child labor. On average, households have almost 4 children below 15 years. It is because of these relatively large households that we have enough variation in the data to analyze intra-household child labor allocation along various dimensions. In particular, Table 3-1 shows that 60 percent of households have at least one boy and one girl between 6 and 15 years old, and almost 80 percent of households has at least one child between 6 and 10 years old, and one child between 10 and 15 years old. Table 3-1 further shows that when we divide children in finer categories (by age- gender groups, or academic achievement-gender groups) we still have a reasonable number of households with children in the different categories, suggesting the results in this chapter are unlikely to be driven by outliers. Table 3-1 further shows that these shares are similar in the different treatment and control groups. 69 Table 3-1: Randomization results Baseline Household Characteristics # Obs. Control (C) Treatment (T) Productive investment package (T3) P-value (T-C) P-value (T3-C) Age household head 1596 43.65 43.59 44.28 (0.94) (0.50) Male household head 1597 0.850 0.859 0.875 (0.71) (0.37) Literate household head 1597 0.618 0.652 0.654 (0.28) (0.36) # years education household head 1496 2.734 2.721 2.785 (0.94) (0.80) Household size 1597 6.882 6.918 7.041 (0.83) (0.41) Number of hh. members under 5 1597 0.804 0.766 0.778 (0.60) (0.74) Number of hh. members 5-14 1597 2.824 2.919 2.934 (0.16) (0.19) Number of hh. members 15-24 1597 1.060 1.050 1.081 (0.89) (0.82) Number of hh. members 25-64 1597 2.039 2.050 2.067 (0.81) (0.63) Number of hh. members 65 or older 1597 0.143 0.119 0.148 (0.37) (0.87) Number of rooms in the house 1597 1.652 1.601 1.626 (0.61) (0.81) Own land 1597 0.681 0.656 0.667 (0.56) (0.76) Distance to school (hours) 1597 0.327 0.261 0.262 (0.10) (0.11) Distance to health center (hours) 1597 1.216 1.156 1.146 (0.70) (0.66) Distance to municipal headquarters (hours) 1597 1.693 1.578 1.623 (0.53) (0.72) 70 Table 3-1: (continued) Baseline Household Characteristics # Obs. Control (C) Treatment (T) Productive investment package (T3) P-value (T-C) P-value (T3-C) At least one household member active in: Self-employment in agriculture 1597 0.911 0.880 0.901 (0.26) (0.73) Self-employment in livestock 1597 0.536 0.474 0.488 (0.36) (0.52) Wage labor in agriculture 1597 0.592 0.651 0.697 (0.19) (0.04) Elaboration food products 1597 0.060 0.075 0.079 (0.50) (0.45) Manufacturing (self-employment) 1597 0.036 0.029 0.036 (0.67) (0.97) Commercial activities 1597 0.082 0.080 0.066 (0.91) (0.38) Services (self-employment) 1597 0.053 0.049 0.056 (0.80) (0.89) Wage employment 1597 0.174 0.191 0.191 (0.58) (0.59) Skilled wage employment 1597 0.036 0.047 0.056 (0.43) (0.20) 71 Table 3-1: (continued) Baseline Household Characteristics # Obs. Control (C) Treatment (T) Productive investment package (T3) P-value (T-C) P-value (T3-C) Total consumption per capita (cordoba) 1584 4039 4248 4255 (0.42) (0.47) Total income per capita (cordoba) 1584 3278 3602 3611 (0.13) (0.23) Income from agricultural wage labor 1589 2150 1959 2024 (0.49) (0.67) Income for elaboration food products 1590 115.9 120.5 124.4 (0.92) (0.88) Income from commercial activities 1589 175.4 205.7 154.6 (0.62) (0.76) Income from manufacturing (self-employment) 1589 26.97 21.5 30.0 (0.73) (0.86) Income from services (self-employment) 1590 155.7 125.0 105.0 (0.63) (0.43) Income from nonagricultural wage labor 1578 681.3 975.6 904.2 (0.12) (0.34) Income from temporary migration 1592 2663 2838 2432 (0.69) (0.62) Monetary income from agricultural self-employment 1591 728.5 829.5 865.4 (0.50) (0.41) Monetary income from livestock 1591 737.1 651.2 558.1 (0.49) (0.20) 72 Table 3-1: (continued) Baseline Household Characteristics # Obs. Control (C) Treatment (T) Productive investment package (T3) P-value (T-C) P-value (T3-C) Composition children between age 6 and 15 % with at least 1 boy and 1 girl 1597 0.61 0.62 0.63 (0.86) (0.68) % with at least 1 child 6-10 years and 1 child 10-15 years 1597 0.79 0.78 0.76 (0.60) (0.35) % with at least 1 boy 6-10 years and 1 boy 10-15 years 1597 0.34 0.36 0.35 (0.32) (0.60) % with at least 1 girl 6-10 years and 1 girl 10-15 years 1597 0.35 0.32 0.34 (0.41) (0.80) % with at least 1 boy 6-10 years and 1 girl 6-10 years 1597 0.24 0.24 0.25 (0.82) (0.79) % with at least 1 boy 10-15 years and 1 girl 10-15 years 1597 0.26 0.29 0.31 (0.29) (0.19) % with at least 1 child at grade level and 1 child below grade level 1597 0.32 0.35 0.33 (0.25) (0.68) % with at least 1 boy at grade level and 1 boy below grade level 1597 0.09 0.12 0.09 (0.17) (0.81) % with at least 1 girl at grade level and 1 girl below grade level 1597 0.10 0.11 0.13 (0.70) (0.36) % with at least 1 boy at below grade level and 1 girl below grade level 1597 0.16 0.16 0.19 (0.68) (0.26) NOTE: Households with at least 2 children between 6 and 15 years of age. P-values account for clustering at community level. All values refer to intent-to-treat 73 Table 3-2: Child labor by gender: Intrahousehold allocation Number of hours per week child between 6 and 15 years worked in: Non-agricultural Agriculture and livestock Agriculture Livestock Chores Domestic work All econ activity All work Boy 0.1295 4.324 3.418 0.879 4.542 1.242 4.564 10.346 Girl 0.218 1.797 0.651 1.116 3.605 3.734 2.125 9.463 Difference -0.0885 2.527 2.767 -0.237 0.937 -2.492 2.439 0.883 P-value (0.277) (0.000) (0.000) (0.045) (0.000) (0.000) (0.000) (0.186) NOTE: Control communities only. Results from estimations with household fixed effects. Standard errors corrected for clustering at community level. Children 6-15 years old, in households with at least 2 children in this age range. Number of hours in each activity trimmed for 0.5 % highest values. N = 1088 (from 413 households) 74 Table 3-3: Child labor by age: Intrahousehold allocation Number of hours per week child between 6 and 15 years worked in: Non-agricultural Agriculture and livestock Agriculture Livestock Chores Domestic work All econ activity All work Age 6-10 0.048 1.223 0.533 0.668 3.167 1.766 1.327 6.259 Age 10-15 0.338 5.327 3.864 1.427 5.200 3.496 5.845 14.54 Difference -0.290 -4.104 -3.331 -0.759 -2.033 -1.730 -4.518 -8.281 P-value (0.001) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) NOTE: Control communities only. Results from estimations with household fixed effects. Standard errors corrected for clustering at community level. Children 6-15 years old, in households with at least 2 children in this age range. Number of hours in each activity trimmed for 0.5 % highest values. N = 1088 (from 413 households) 75 Table 3-4: Difference in child labor between children attending school and those not attending: Intrahousehold allocation Number of hours per week child between 6 and 15 years worked in: Non-agricultural Agriculture and livestock Agriculture Livestock Chores Domestic work All econ activity All work ALL Difference 0.195 -2.962 -2.892 -0.0524 -0.469 -0.502 -2.759 -3.730 P-value (0.083) (0.006) (0.008) (0.720) (0.170) (0.140) (0.013) (0.000) BOYS Difference 0.208 -5.971 -5.825 -0.110 -0.318 0.453 -5.744 -5.610 P-value (0.17) (0.002) (0.002) (0.55) (0.54) (0.14) (0.003) (0.002) GIRLS Difference 0.185 -0.566 -0.558 -0.00659 -0.588 -1.263 -0.383 -2.234 P-value (0.22) (0.35) (0.35) (0.98) (0.090) (0.009) (0.53) (0.000) NOTE: Controlling for gender and age through series of age dummies by gender. Control communities only. Results from estimations with household fixed effects. Standard errors corrected for clustering at community level. Children 6-15 years old, in households with at least 2 children in this age range. Number of hours in each activity trimmed for 0.5 % highest values. N = 1088 (from 413 households) 76 Table 3-5: Difference in child labor of children below their optimal grade level versus others: Intrahousehold allocation Number of hours per week child between 6 and 15 years worked in: Non-agricultural Agriculture & livestock Agriculture Livestock Chores Domestic work All econ activity All work ALL Difference 0.014 0.220 0.312 -0.101 0.003 -0.051 0.210 0.162 P-value (0.90) (0.71) (0.60) (0.39) (0.99) (0.80) (0.74) (0.83) BOYS Difference 0.108 0.272 0.309 -0.038 -0.245 -0.344 0.391 -0.199 P-value (0.51) (0.78) (0.76) (0.80) (0.48) (0.21) (0.70) (0.87) GIRLS Difference -0.082 0.143 0.295 -0.168 0.231 0.220 -0.002 0.449 P-value (0.47) (0.81) (0.59) (0.35) (0.56) (0.48) (1.00) (0.60) NOTE: Controlling for gender and age through series of age dummies by gender. Control communities only. Results from estimations with household fixed effects. Standard errors corrected for clustering at community level. Children 6-15 years old, in households with at least 2 children in this age range. Number of hours in each activity trimmed for 0.5 % highest values. N = 980 77 III.1. Child labor allocation patterns Before considering the program impact, we describe the child labor patterns among children from the control group. Child labor is measured in number of hours worked per week in economic activities, chores and domestic activities. Economic activities include labor in agricultural and livestock activities, as well as labor in nonagricultural activities. Child labor in agricultural and livestock mostly consists of help with the crops or livestock self-employment activities of the household, but also includes some wage labor in agriculture. Labor in nonagricultural activities consists of help by children in the commercial or manufacturing self- employment activities of the household, or possibly wage-employment outside of agriculture. Chores consist of wood or water gathering, and domestic activities include cooking, cleaning, washing and caring for younger siblings. Table 3-2 shows that including chores and domestic work is important as they constitute a large part of child labor in this setting. Table 3-3 shows very clear gender patterns in the allocation of child labor within the household. Boys work more hours in economic activities, in particular agriculture, and also spend more time carrying water and wood. On the other hand, girls work more hours in domestic activities. As a result, there are no significant differences between girls and boys in total number of hours worked. Table 3-3 shows that boys work on average almost 2.5 hours per week more than their sisters in economic activities. The difference between boys and girls falls to less than an hour, and is not significant, when chores and domestic activities are included. Overall, these patterns suggest within-household specialization along gender lines. Not surprisingly, the data also indicates that older children work more hours than younger children in all activities. Given the schooling conditionality, we analyze whether school attendance and child labor appear to compete for children’s time. Table 3-4 shows that, after accounting for differences in age and gender, children who attend school indeed work on average 4 hours less per week. Children work in particular almost 3 hours less in agriculture, with the remaining hour mainly 78 coming from chores and domestic work. Intriguingly however, school-going children work more in nonagricultural activities. None of these differences capture any household level variation, as the fixed effects control for household socio-economic status and other household unobservables. A possible explanation of the finding on nonagricultural activities is related to the low education levels of the parents whom might need the help of school-going children for basic math and accounting necessary in such activities. Table 3-4 also shows results for boys and girls separately, and indicates that the negative correlation between child labor in economic activities and school attendance is completely driven by the results for boys. Girls work equal amounts in economic activities, whether they are attending school or not, but boys work on average almost 6 hours less when they attend school. This hence suggests that for boys, school and work might be substitutes, while this is much less the case for girls. This is consistent with the timing of work in agriculture—which occurs mainly in the mornings at the same time of classes—and with the specialization of girls and boys in different tasks. Finally, we consider whether children are at their grade level or below. Children are classified as below their grade level if their accumulated years of education are less than the level they should have attained if they enrolled at age 7 and passed grades every year. Children are classified as at grade level if there is no age-grade distortion. Table 3-5 shows that there are no significant differences in child labor allocation between children who are below grade level, and their siblings who are at grade level. The relationship between past academic achievement and child labor hence does not appear to be straightforward. III.2. Hypotheses Does the Atencion a Crisis program strengthen or offset existing child labor allocation patterns within the households? The answer likely depends on a number of program design features, as highlighted by the theoretical analysis. First, the cash component of the program 79 might reduce the need for child labor. It is possible for instance, that the cash would be used to hire day laborers in agriculture that could substitute for the boys’ work. Second, the school conditionality related to the program, together with the cash that allows buying school materials, might decrease the cost of schooling thus increasing enrollment and attendance to school and reducing the number of hours the children work. This would have a larger impact on those children that would have been working more in the absence of the program. The descriptive results hence would suggest that both the cash and the conditionality might affect work hours of older boys in particular, and as such possibly reduce the gender discrepancies. On the other hand, work done in other countries around the world (Duflo, 2003 for South Africa; Thomas for Brazil, Ghana and the US) has shown that resources in the hands of women might favor investment in girls. While it is unclear whether this pattern holds in Nicaragua, it could possibly lead to a higher reduction of child labor for girls (see also Emerson and Portela Souza, 2007). Yet the program – and in particular the productive investment grant – might also lead to an increased need for help in nonagricultural activities. These are activities in which older girls tend to specialize, and it is unclear whether the increased demand for labor would reinforce or weaken this specialization pattern. Households with the productive investment package possibly also have an increased need for help with domestic work. As the female beneficiary takes up her new activity, this could affect older girls in the household, and lead to a reinforcement of the age and gender patterns. Also, given low levels of literacy in the region, adult program participants might need to rely on children with more advanced math skills (higher schooling levels) for help with the accounting part of the new activity. Finally, the finding that there are no significant differences in child labor between children who are at grade level versus those that are behind, might indicate that on average parents do not put extra labor burden on children that have fallen behind in school. This could be because higher ability or accumulated skill can increase both the return to child labor and the 80 return to schooling, or because higher ability or accumulated skill can make it easier for children to combine both. If this is the case, one might expect that the additional cash combined with the conditionalities might help parents to compensate for past lags in academic achievement. IV. Impact of Atencion a Crisis on Intra-household Child Labor Allocation Given that the decision-making in households with the productive investment package needs to account for a number of additional factors, we first analyze the impact on the intra- household allocation of all households from the treatment communities, and then compare the impacts of households with the third package with those with the basic package. We rely on the randomized design, and estimate the impacts using simple differences between the treatment and control households. Hence let  Y ij   0   1 T i   2 X ij   3 T i X ij   i   ij (19) where Y ij is the child labor hours (in a specific type of activity) for child j in household i corrected for any gender-specific age-trends based on the estimated trends for the control group (see below). T is a dummy variable indicating the intent-to-treat for household i, X ij a key characteristic of child j in household i that could affect child labor allocation (gender or past academic achievement),   i captures all unobservable characteristics of household i, while   ij captures the unobservable characteristics of child j in household i. We estimate the model using household fixed effects, implying that both the term   1 T i and   i cancel out. The estimate of   3 , the coefficient of interaction of the intent-to-treat dummy with the various variables of interest then sheds light on the intra-household reallocation in child labor. The model hence allows isolating the heterogeneity in impacts within households along a number of dimensions, while controlling for all unobservable household factors. We first consider 81 differences between children of different gender, and then focus on whether past academic achievement of the child is related to differences in program impact within the household. All standard errors are corrected for clustering at the community level. As the   1 T i term cancels out when we include household fixed effects, we also estimate a random effects model in order to facilitate the interpretation of the estimate for   3 . As correlation between explanatory variables and household unobservables might bias the random effects model, the fixed effects model is our preferred specification. Yet the results of the random effects model provide us with a base of comparison that helps to interpret the meaning of the coefficients. Further, given that the age of the child and possibly gender might be correlated with other variables of interest (such as grade achievement), and in order to increase the precision of the estimates, we first normalize the outcome variables by regressing each outcome on a series of dummies for age (in months) by gender for the control. For outcomes used in the fixed effect estimates, we also include household fixed effects in these estimations. We then obtain the residual by subtracting the estimated outcome for each category from the observed measure. In this chapter, we hence measure how child labor hours differ from the average number of hours of a child of the same gender and age. IV.1. Gender and age: Compensation and specialization Table 3-6 sheds light on the reallocation of child labor within the household as a result of the intervention. The top panel shows that child labor decreases more for boys than for girls, and this is primarily the result of larger decreases for boys in agriculture and livestock activities. Both child labor in all economic activities and total child labor reduce more for boys than for girls, leading to a reduction of the gaps in total numbers worked with 1.5 hours. 82 Table 3-6: Intrahousehold heterogeneity of impacts by gender and age: All eligible households Fixed Effects Model Number of hours per week child between 6 and 15 years worked in: COEFFICIENT nr obs/hh Non- agricultural Agricultur e & livestock Agricultur e Livestoc k Chores Domesti c work All econ activity All work male*itt 4253/1594 -0.00231 -1.179** -0.824* -0.374** - 0.0395 -0.223 -1.256** -1.518** (0.11) (0.47) (0.45) (0.17) (0.25) (0.27) (0.49) (0.63) female*age1015*itt 4253/1594 0.209 0.575 0.264 0.311 -0.180 0.663* 0.707 1.191 (0.182) (0.488) (0.447) (0.197) (0.302) (0.350) (0.511) (0.723) male*age1015*itt -0.0281 -2.786*** -2.014*** - 0.805*** -0.154 -0.0823 -2.840*** - 3.076*** (0.133) (0.784) (0.725) (0.215) (0.330) (0.234) (0.794) (0.906) male*itt 0.137 0.743 0.476 0.267 - 0.0584 0.217 0.775 0.933 (0.132) (0.542) (0.498) (0.226) (0.380) (0.364) (0.591) (0.883) P-value female*age1015*itt = male*age1015*itt 0.586 0.001 0.015 0.000 0.909 0.088 0.001 0.000 83 Table 3-6 (continued) Random Effects Model Number of hours per week child between 6 and 15 years worked in: COEFFICIENT nr obs/hh Non- agricultural Agricultur e & livestock Agricultur e Livestoc k Chores Domesti c work All econ activity All work male*itt 4253/1594 -0.0580 -1.741*** -1.296*** - 0.413*** 0.229 -0.142 -1.755*** -1.478** (0.11) (0.48) (0.46) (0.15) (0.22) (0.24) (0.52) (0.61) itt 0.192** -0.104 -0.288** 0.177 - 0.707** 0.0735 0.00238 -0.747 (0.090) (0.23) (0.12) (0.19) (0.35) (0.25) (0.26) (0.63) female*age1015*itt 4253/1594 0.181 -0.0594 -0.0295 0.0467 -0.255 0.810** 0.0364 0.805 (0.151) (0.291) (0.221) (0.201) (0.285) (0.329) (0.344) (0.614) male*age1015*itt 0.0502 -3.405*** -2.851*** - 0.672*** -0.0134 0.163 -3.199*** - 3.011*** (0.114) (0.817) (0.782) (0.184) (0.336) (0.194) (0.824) (0.914) female*itt 0.0864** -0.0664 -0.269*** 0.151 -0.558* -0.399* -0.0165 -1.208** (0.043) (0.201) (0.103) (0.166) (0.321) (0.213) (0.219) (0.543) male*itt 0.106*** 0.117 0.0591 0.153 -0.471 -0.161 0.0917 -0.489 (0.038) (0.255) (0.183) (0.144) (0.305) (0.176) (0.314) (0.494) P-value female*age1015*itt = male*age1015*itt 0.529 0.000 0.002 0.004 0.253 0.204 0.000 0.001 84 Note on Table 3-6: Standard errors corrected for clustering at community level in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Dependent variables corrected for gender-specific age trends by regression of outcomes in control on series of age dummies for each gender. Number of hours in each activity trimmed for 0.5 % highest values. All estimates include a gender dummy, and interactions for gender and age. All households with at least 2 children between 6 and 15. age1015 is a dummy variable = 1 for children between 10 and 15. Fixed effects model controls for household fixed effects, random effects model controls for household random effects. 85 When accounting for heterogeneity by age when considering gender differences in impact, it becomes clear that the reductions in agriculture, livestock, domestic work and total work are particularly large for older boys, when compared to their siblings. In contrast, impact on child labor allocation for older girls does not seem to be larger than for their younger sisters, and there is some indication of an increase in domestic work, relative to their younger sisters. Yet, in terms of total hours worked the impact for girls does not decrease significantly as age increases, which contrasts the results found for boys. In fact, the point estimate for the comparison of impacts on older girls versus younger girls is positive, though not significant. Overall, the P- values in Table 3-6 show that the differences in differential impacts by age for boys compared to girls are very significant for all activities except nonagricultural work and chores. Table 3-7 shows that the finding that child labor decreases more for older boys compared to both younger boys and girls is robust to different alternative specifications. First we show that results are similar when controlling for age and gender trends in the regression instead of measuring the dependent variable as a deviation from the age-gender specific mean. Second, we add a number of additional child-specific control variables. While the randomization eliminates the need for controls, they could possibly add some precision to the estimates. The child specific controls that are added are number of years of education, a binary variable whether the father of the child lives in the household, the child’s rank among all children below 15 in the household, and the child’s rank among all children of the same gender below 15 in the household. Each of these control variables is also interacted with the binary variable for gender. The results show that the findings are similar, but that the contrast between older boys and older girls, as well as the contrast between older girls and younger girls become stronger. In a third specification, only households that have some child labor in the specific activity considered are included to addresses a potential censoring concern (appendix B elaborates on this point). Note first that because the dependent variable is measured as deviations from the mean for 86 children of the same age and gender in the control, there is no clear censoring in this variable. Nevertheless, there might be a concern related to censoring at 0 of the original child labor variables. Specification 3 excludes all the households that had no children working for each activity, to shed some light on this issue. As can be seen this results in relatively few households for some of the activities (for example nonagricultural work), but overall there are few households (about 68 out of 1594) where none of the children work. More importantly the results are quite similar to the estimates on the full sample. Indeed overall, all three alternative fixed effects specification show that child labor decreased substantially more for older boys when compared both to younger boys, and to their female siblings 87 Table 3-7: Alternative specifications for intrahousehold heterogeneity of impacts by gender: All eligible households Fixed Effects Models COEFFICIENT Non- agricultural Agriculture & livestock Agriculture Livestock Chores Domestic work All econ activity All work Number of hours per week child between 6 and 15 years worked in each of the activities 1. Without removing age trends and with age*gender dummies as control female*age1015*itt 0.426** 0.198 -0.0289 0.238 -0.0213 0.470 0.661 1.109 (0.190) (0.512) (0.442) (0.229) (0.360) (0.434) (0.541) (0.853) male*age1015*itt 0.110 -2.342** -1.968** -0.357 -0.211 0.0800 -2.128** -2.259** (0.0942) (0.944) (0.876) (0.276) (0.372) (0.256) (0.949) (1.116) male*itt 0.138 0.341 0.202 0.117 0.00149 -0.137 0.484 0.348 (0.153) (0.590) (0.551) (0.212) (0.374) (0.389) (0.620) (0.911) P-value female*age1015*itt = male*age1015*itt 0.148 0.038 0.078 0.135 0.696 0.438 0.024 0.020 2. Including additional control variables female*age1015*itt 0.285 0.543 0.241 0.309 -0.0167 0.792** 0.776 1.552** (0.201) (0.497) (0.456) (0.198) (0.309) (0.360) (0.518) (0.724) male*age1015*itt -0.0139 -2.563*** -1.779** - 0.808*** -0.258 -0.131 -2.584*** - 2.972*** (0.146) (0.813) (0.749) (0.220) (0.347) (0.241) (0.829) (0.980) male*itt 0.166 1.095* 0.787 0.304 0.209 0.216 1.155* 1.579* (0.145) (0.557) (0.503) (0.240) (0.379) (0.366) (0.601) (0.886) P-value female*age1015*itt = male*age1015*itt 0.244 0.004 0.037 0.001 0.611 0.027 0.002 0.001 88 Table 3-7 (continued) COEFFICIENT Non- agricultural Agriculture & livestock Agriculture Livestock Chores Domestic work All econ activity All work 3. Only households with some child labor in the type of activity considered female*age1015*itt 1.827 0.939 0.831 0.712* -0.0528 0.821** 1.319* 1.401* (1.142) (0.665) (1.109) (0.385) (0.317) (0.363) (0.670) (0.731) male*age1015*itt 0.495 -2.770** -1.987 -0.857* 0.164 0.0460 - 2.773*** -2.811*** (1.108) (1.080) (1.366) (0.464) (0.360) (0.296) (1.019) (0.907) male*itt 0.667 1.052 1.726 0.427 -0.144 0.0882 0.866 0.954 (1.058) (0.827) (1.182) (0.452) (0.400) (0.410) (0.858) (0.905) P-value female*age1015*itt = male*age1015*itt 0.468 0.012 0.148 0.023 0.657 0.083 0.004 0.001 Nr. obs/hh 534/200 2497/914 1418/504 1777/663 3710/1369 3430/1259 2687/990 4098/1528 89 Table 3-7 (continued) COEFFICIENT Non- agricultural Agriculture & livestock Agriculture Livestock Chores Domestic work All econ activity All work Whether child between 6 and 15 years worked in: female*age1015*itt 0.0360 -0.0168 -0.0589* 0.0410 -0.0355 0.0892* -0.0180 0.00297 (0.0224) (0.0431) (0.0341) (0.0379) (0.0437) (0.0451) (0.0407) (0.0404) male*age1015*itt 0.00357 -0.0501 -0.0418 -0.0686* -0.0490 0.00581 -0.0453 -0.0133 (0.0178) (0.0408) (0.0395) (0.0361) (0.0395) (0.0412) (0.0407) (0.0381) male*itt 0.0213 -0.0184 -0.0111 0.0119 0.0379 0.0274 0.00531 0.0160 (0.0181) (0.0376) (0.0327) (0.0335) (0.0531) (0.0542) (0.0383) (0.0392) P-value female*age1015*itt = male*age1015*itt 0.257 0.592 0.768 0.041 0.821 0.194 0.640 0.745 NOTE: Standard errors corrected for clustering at community level in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Dependent variables in 2nd, 3rd and 4th specification corrected for gender-specific age trends by regression of outcomes in control and on series of age dummies for each gender. Number of hours in each activity trimmed for 0.5 % highest values. All estimates include a gender dummy, and interactions for gender and age, and a household fixed effect. In addition, specification 2 also includes controls for number of years of education, a binary variable indicating whether the father of the child lives in the household, the child's rank among all children below 15 in the household, and the child's rank among all children below 15 from the same gender in the household. Each of these control variables is also interacted with a binary variable of gender. All households with at least 2 children between 6 and 15. age1015 is a dummy variable = 1 for children between 10 and 15. Number of observations in specification 1, 2, and 4 is 5213 (1594 households). 90 In a final specification, we consider a binary variable for child labor in each activity, as opposed to the continuous variable. This sheds light on whether parents adjust by reducing the number of hours worked of some siblings compared to others, or rather withdraw some children entirely from working. While the point estimates of these estimations mostly point in the same direction, they are generally not significant, suggesting parents adjust the intensity of work, rather than relieving some children completely from their work duties. Table 3-8 re-estimates the model only including households with the productive investment package and control households, while also providing, for comparison, the model estimates including only households with the basic package and control households. Given that the productive investment grant might have increased the return to activities in which women specialize, one can wonder whether this package led to a shift of girls’ child labor to those activities. While the standard errors are higher because of smaller sample sizes, the results first show that for this group, child labor for boys decreases significantly more than for girls, which appears to be driven by labor in agriculture and livestock. Table 3-8 also shows that the productive investment package result in a shift of older girls to nonagricultural activities and domestic activities, when compared with their siblings. As a result of both these mechanisms, the gender differences are larger than for the basic package. The productive investment package hence appears to reinforce intra-household specialization of older girls in nonagricultural activities and domestic work. Note that the random effects estimation suggests that the overall impact on child labor of older girls is not significant - in fact the sum of the coefficients is close to 0 (-1.6 + 1.7). Hence while the productive package led to less child labor for younger girls and older boys, it did not significantly affect overall child labor of younger boys and older girls. Overall then, these results suggest that the conditional cash transfer helped to narrow intra-household gender and age differences in child labor, and older boys in particular appear to have benefited most. This is consistent with the descriptive statistics in Table 3-4 that suggested 91 that school attendance and boys’ work in agriculture are negatively correlated with each other. This could indicate that the school conditionality, by guaranteeing that children are in school at the moment they otherwise would be working in the field is helping compensate for the higher number of hours that older boys were working, when compared to their sisters, and their younger siblings. At the same time, there is evidence of increased intra-household specialization of older girls in nonagricultural activities and domestic work for households who received the productive investment package. IV.2. Past academic achievement We now consider whether the program helped compensate for lags in past academic achievement, and in particular whether reductions in child labor are larger for children that were lagging behind. The variables we use to measure past academic achievement are a dummy variable indicating whether the child was below the grade level that corresponds to its age at baseline, and a variable indicating the number of years the child was below grade level at baseline. These variables are likely to capture innate capabilities, academic skills (or lack thereof), and past disadvantages that might have led the children to have lower academic performance than their siblings. For example, a child might have been disadvantaged because of a drought shock in early childhood that affected his or her cognitive development during a critical stage (see e.g., Alderman, Hoddinott and Kinsey, 2006). Such disadvantage might afterwards have been aggravated (or not) by resource allocation decisions of the parents. Or children with lower cognitive abilities might themselves have been more likely to drop out of school, and therefore might have higher child labor participation. On the other hand, parents of children with very low abilities might not make them work, as the returns to their work might be very low. Independently of the reasons why certain children in a household are below grade level, it is interesting to see whether parents shifted child labor away from those children and as such might help compensate for lags. Table 3-9 shows the fixed effect estimation that accounts for 92 heterogeneity of impacts by past academic achievement and gender. The results show that boys that were below grade level had a much larger reduction in child labor in economic activities, in particular in agriculture when compared to their brothers; labor for boys that are at least 1 grade behind reduced with 3 hours per week more than labor for other boys. Interestingly however, these boys seem to be shifted into domestic work, as compared to their brothers, and as a result the effect is much smaller, when considering all hours worked. The relative reduction of child labor in economic activity among boys below grade level is even larger when considering only households with the productive investment package. Labor in agriculture, and as a result in total economic activity reduces by almost 4 hours, as compared with their brothers at grade level. As for the full sample, this effect is partially offset by an increase in hours spent in domestic work (results available from the authors). When considering the nonagricultural work, there is some indication that the increase in child labor in this activity is muted for children of both genders who are behind grade level. This is consistent with patterns of specialization in nonagricultural activities by children with higher schooling levels, as shown earlier in the chapter. Other than that, we do not find any significant differences in child labor impacts between girls with low versus high past academic achievement. Overall, the results in past academic achievement hence indicate that the program did help compensate for past lags, but only for boys. Parents responded to the program by reallocating boys with lower skill or ability away from agriculture, but other boys had larger reductions in domestic work. 93 Table 3-8: Intrahousehold heterogeneity of impacts by gender and age: For basic package and productive investment package Fixed Effects Model: Beneficiaries from basic package versus control Number of hours per week child between 6 and 15 years worked in: COEFFICIENT nr obs/hh Non- agricultural Agriculture & livestock Agriculture Livestock Chores Domestic work All econ activity All work male*itt -0.0528 -1.013* -0.760 -0.303 0.0748 -0.384 -1.133** -1.442* (0.141) (0.598) (0.579) (0.254) (0.323) (0.351) (0.570) (0.740) female*age1015*itt 0.0553 0.862 0.869 0.0177 -0.311 0.492 0.810 0.991 (0.237) (0.668) (0.590) (0.293) (0.395) (0.524) (0.707) (1.133) male*age1015*itt -0.00484 -2.962*** -1.978** - 0.987*** 0.00413 -0.353 - 3.019*** - 3.368*** (0.179) (1.075) (0.949) (0.332) (0.466) (0.328) (1.111) (1.231) male*itt -0.0157 1.190 0.893 0.264 -0.121 0.120 1.069 1.069 (0.214) (0.775) (0.692) (0.335) (0.507) (0.482) (0.774) (1.201) P-value female*age1015*itt = male*age1015*itt 0.851 0.006 0.018 0.041 0.606 0.157 0.007 0.012 94 Table 3-8 (continued) Fixed Effects Model: Beneficiaries from productive investment package versus control Number of hours per week child between 6 and 15 years worked in: COEFFICIENT nr obs/hh Non- agricultural Agriculture & livestock Agricult ure Livesto ck Chor es Domes tic work All econ activity All work male*itt 2143/806 -0.0680 -1.480** -0.914 - 0.558** 0.065 4 - 0.0643 -1.760** -1.759* (0.23) (0.62) (0.58) (0.23) (0.35 ) (0.33) (0.67) (0.93) female*age1015*itt 2143/806 0.915*** 0.665 0.0801 0.561* - 0.245 0.946* * 1.527** 2.228** (0.307) (0.640) (0.585) (0.299) (0.39 7) (0.431) (0.725) (1.047) male*age1015*itt 0.0499 -3.319*** - 2.404** * - 0.910** * - 0.390 - 0.0754 -3.254*** - 3.720** * (0.197) (0.875) (0.807) (0.306) (0.46 1) (0.324) (0.897) (1.096) male*itt 0.430* 0.853 0.545 0.300 0.154 0.525 1.031 1.710 (0.247) (0.682) (0.647) (0.286) (0.48 5) (0.450) (0.782) (1.126) P-value female*age1015*itt = male*age1015*itt 0.184 0.002 0.031 0.001 0.984 0.177 0.000 0.001 95 Table 3-8 (continued) Random Effects Model: Beneficiaries from productive investment package versus control Number of hours per week child between 6 and 15 years worked in: COEFFICIENT nr obs/hh Non- agricultural Agriculture & livestock Agricult ure Livesto ck Chores Domest ic work All econ activity All work male*itt 2143/806 -0.205 -1.586*** -1.169** - 0.413* * 0.482 0.0997 -1.806*** -1.186 (0.19) (0.54) (0.52) (0.19) (0.30) (0.29) (0.59) (0.77) itt 0.621*** -0.0413 -0.292** 0.265 - 0.963* ** -0.143 0.539 -0.595 (0.15) (0.26) (0.15) (0.23) (0.37) (0.27) (0.34) (0.67) female*age1015*itt 2143/806 0.895*** 0.0848 -0.124 0.271 -0.402 1.026* ** 0.982** 1.748* * (0.236) (0.362) (0.284) (0.276) (0.355) (0.396) (0.445) (0.794) male*age1015*itt 0.314** -3.699*** - 2.779** * - 0.928* ** -0.335 0.163 -3.266*** - 3.503* ** (0.153) (0.867) (0.847) (0.269) (0.449) (0.254) (0.886) (1.068) female*itt 0.0920 -0.0959 -0.222 0.103 - 0.728* * - 0.746* ** -0.0460 - 1.631* * (0.0860) (0.273) (0.148) (0.238) (0.350) (0.235) (0.330) (0.669) male*itt 0.233*** 0.515 0.147 0.389 -0.283 -0.144 0.622 0.241 (0.0805) (0.363) (0.272) (0.243) (0.366) (0.203) (0.435) (0.664) 96 Table 3-8 (continued) P-value female*age1015*itt = male*age1015*itt 0.049 0.001 0.007 0.017 0.884 0.015 0.000 0.000 NOTE: Standard errors corrected for clustering at community level in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Dependent variables corrected for gender- specific age trends by regression of outcomes in control and on series of age dummies for each gender. Number of hours in each activity trimmed for 0.5 % highest values. All estimates include a gender dummy, and interactions for gender and age. All households with at least 2 children between 6 and 15. age1015 is a dummy variable = 1 for children between 10 and 15. Fixed effects model controls for household fixed effects, random effects model controls for household random effects. 97 Table 3-9: Intrahousehold heterogeneity of impacts by past academic achievement: All eligible households Fixed Effects Model Number of hours per week child between 6 and 15 years worked in: COEFFICIENT nr obs/hh Non- agricultural Agriculture & livestock Agriculture Livestock Chores Domestic work All econ activity All work male*at least 1 grade behind 3228/1551 0.108 0.272 0.309 -0.0382 -0.245 -0.344 0.391 -0.199 (0.16) (0.96) (0.99) (0.15) (0.34) (0.27) (1.01) (1.21) male*at least 1 grade behind*itt -0.332 -2.734** -2.560** -0.132 0.683 0.966*** -3.087** -1.438 (0.21) (1.21) (1.22) (0.29) (0.44) (0.34) (1.27) (1.43) female*at least 1 grade behind -0.0820 0.143 0.295 -0.168 0.231 0.220 -0.00201 0.449 (0.11) (0.58) (0.54) (0.18) (0.39) (0.31) (0.61) (0.85) female*at least 1 grade behind*itt -0.232 0.682 0.374 0.328 -0.194 -0.269 0.310 -0.153 (0.22) (0.73) (0.64) (0.27) (0.45) (0.43) (0.74) (1.04) male*itt -0.158 0.0639 0.401 -0.354 -0.707 -1.452*** -0.266 -2.425** (0.23) (0.67) (0.71) (0.26) (0.49) (0.50) (0.73) (1.11) P-value female*grbehind*itt = male*grbehind*itt 0.745 0.012 0.032 0.232 0.181 0.033 0.016 0.465 98 Table 3-9 (continued) Number of hours per week child between 6 and 15 years worked in: COEFFICIENT nr obs/hh Non- agricultural Agriculture & livestock Agriculture Livestock Chores Domestic work All econ activity All work male*number grades behind 3192/1543 0.0151 0.218 0.184 0.0174 - 0.0145 -0.181** 0.251 0.0555 (0.043) (0.45) (0.47) (0.055) (0.13) (0.076) (0.45) (0.49) male*number grades behind*itt -0.116* -1.242** -1.081** -0.127 0.120 0.344*** -1.429*** -0.966* (0.058) (0.50) (0.51) (0.095) (0.15) (0.10) (0.50) (0.54) female*number grades behind -0.00916 0.105 0.0908 0.0211 0.0909 0.0208 0.110 0.222 (0.045) (0.17) (0.14) (0.084) (0.10) (0.14) (0.18) (0.28) female*number grades behind*itt -0.123* -0.0632 -0.0531 -0.0234 - 0.0392 -0.0648 -0.287 -0.391 (0.068) (0.24) (0.19) (0.12) (0.13) (0.18) (0.26) (0.36) male*itt -0.200 0.514 0.778 -0.355 -0.464 - 1.536*** 0.190 -1.810 (0.23) (0.84) (0.84) (0.25) (0.43) (0.53) (0.92) (1.24) P-value female*grbehind*itt = male*grbehind*itt 0.941 0.022 0.042 0.441 0.404 0.043 0.033 0.363 NOTE: Standard errors corrected for clustering at community level in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Dependent variables corrected for gender-specific age trends by regression of outcomes in control on series of age dummies for each gender using a household fixed effects model. Number of hours in each activity trimmed for 0.5 % highest values. All estimates include gender dummy and household level fixed effects. All households with at least 2 children between 6 and 15. Academic achievement measured at baseline. age1015 is a dummy variable = 1 for children between 10 and 15. 99 V. Conclusions This chapter analyzed whether parental decisions in response to a social program in rural Nicaragua appear to compensate or reinforce pre-program differences in child labor allocation. The chapter starts by providing a theoretical framework that relates various types of child labor to family characteristics (such as asset endowments and parental preferences), child characteristics (such as gender and ability), and social characteristics (such as schooling costs and rates of return). This serves to guide the empirical specification and the interpretation of results. The chapter then evaluates the Atención a Crisis social program regarding the allocation of child labor, showing that the program helped compensate for some of the pre-program intra-household differences, but exacerbated others. In particular, it reduced total hours worked for older boys, and for boys with low past academic achievements, and these results are driven by reductions in agriculture and livestock. On the other hand, the productive investment package reinforced existing specialization in specific tasks within the household for older girls in particular. Girls in households that received the productive investment package are more likely to increase work in nonagricultural activities and domestic work, when compared to their siblings. This suggests that increased potential for nonagricultural activities in the household reinforced specialization by girls in these tasks. At the same, overall child labor did not increase for these older girls, suggesting that combining the productive investment package with the conditional cash transfer might have been important to avoid an overall increase of child labor of girls. A possible explanation of these differences in impacts by gender relates to the timing of the different activities. Agricultural work tends to be done in the mornings, at the same time of classes, while nonagricultural work, domestic work, and chores can be done at a time that does not directly compete with class. Moreover, boys’ work in agriculture can be substituted for with hired labor, while this is more difficult for the tasks in which the girls specialize. The program increased the likelihood of using wage labor in agriculture with about 20 percent (P-value = 100 0.024), indicating that such substitution indeed might have taken place. On the other hand, the new nonagricultural activities were typically small scale and did not involve any hired labor. Given that boys--older boys and boys that had fallen behind in school--before the program worked more hours in economic activities compared to their siblings these findings suggest that the program helped level the playing field to a certain extent. While the chapter cannot identify whether it is the cash or the conditionality feature of the program design that helped trigger this response, it is consistent with substitution between child work in agriculture and schooling, and with the program impacts on school enrollment and attendance. On the other hand, for both genders, child labor in nonagricultural economic activities and schooling appear to be complements, even within the household, indicating that the return to children’s schooling may be higher for such activities, possibly because of low education levels of the adults. Overall, the findings in this chapter suggest that child characteristics, family endowments, and the specific characteristics of the intervention are essential factors to understand the impact of a program, such as Atencion a Crisis, on child labor. 101 References Aguilera, V., X. Del Carpio, C. Herrera, K. Macours, M. E. Moncada, C. Obregón, and R. Vakis, 2006. ―Estudio Cualitativo del Componente de Atención a Crisis (CAC) del Ministerio de la Familia, Nicaragua‖, mimeo, Johns Hopkins University. Alderman, H., J. Hoddinott, B. Kinsey, 2006. ―Long Term Consequences of Early Childhood Malnutrition‖, Oxford Economic Papers, 58(3):450-474. Ashenfelter, O., and C. Rouse, 1998. ―Income, Schooling and Ability: Evidence from a New Sample of Identical Twins‖, Quarterly Journal of Economics, 113: 253-284. Attanasio, O., E. Fitzsimons, A. Gomez, D. Lopez, C. Meghir and A. Mesnard, 2006. ―Child Education and Work Choices in the Presence of a Conditional Cash Transfer Program in Rural Colombia‖, The Institute for Fiscal Studies Working Paper No. 06/10. Baland, J.M., and J. A. Robinson, 2000. ―Is Child Labor Inefficient?‖ Journal of Political Economy, 108(4), August, pp. 663-679. Bardhan, P., 1984. Land Labor and Rural Poverty. Oxford, Oxford University Press. Bardhan, P., and C. Udry, 1999. ―Human Capital and Income Distribution‖, chapter 10 in Development Microeconomics, Oxford Press. Barrera-Osorio, F., M. Bertrand, L. Linden, and F. Perez-Calle, 2008. ―Conditional Cash Transfers in Education: Design Features, Peer and Sibling Effects. Evidence from a Randomized Experiment in Colombia. NBER working paper 13890. Basu, K., 1999. ―Child labor: Cause, consequence, and cure, with remarks on international labor standards‖, Journal of Economic Literature 37: 1083-1119. Basu, K., S. Das and B. Dutta, 2007. ―Child Labor and Household Wealth: Theory and Empirical Evidence of an Inverted-U‖, IZA, Discussion Paper No. 2736. Basu, K., and P. H. Van, 1998. ―The Economics of Child Labor,‖ American Economic Review, 88(3), June, pp. 412-427. Becker, G., 1991. A Treatise on the Family, University of Chicago Press. Becker, G.S., and H.G. Lewis, 1973. ‖On the interaction between the quantity and quality of children‖, Journal of Political Economy 81, 279-288. Becker, G., and N. Tomes, 1976. ―Child Endowments and the Quantity and Quality of Children‖, Journal of Political Economy 84(4), 143-162. Beegle, K., R. Dehejia, and R. Gatti, 2005. ―Why Should We Care About Child Labor? The Education, Labor Market and Health Consequences of Child Labor‖, the World Bank Policy Research Working Paper 3479. 102 Beegle, K., R. H. Dehejia, and R. Gatti, 2006. "Child Labor and Agricultural Shocks‖, Journal of Development Economics, 81:80-96. Behrman, J.R., 1988. "Intra-household Allocation of Nutrients in Rural India: Are Boys Favored? Do Parents Exhibit Inequality Aversion?" Oxford Economic Papers, 40(1): 32-54. Behrman, J.R., and A. Deolalikar, 1988. "Health and Nutrition" Chapter 14 in Handbook of Development Economics, 631-711. Behrman, J.R., R. Pollack, and P. Taubman, 1982. ―Parental Preference and Provision for Progeny,‖ Journal of Political Economy, 90(1):52-73. Behrman, J.R., R. Pollack, and P. Taubman, 1986. ―Do Parents Favor Boys?‖ International Economic Review, 27(1):33-54. Behrman, J.R., and P. Taubman, 1986. ―Birth Order, Schooling, and Earnings,‖ Journal of Labor Economics, 4(3): 121-145. Behrman, T., and P. Todd, 1999. ―Randomness in the Experimental Samples of PROGRESA (Education, Health and Nutrition programs).‖ International Food Policy Research Institute, Washington DC. Bhalotra, S., and C. Heady. 2003. "Child Farm Labor: The Wealth Paradox." World Bank Economic Review 17(2), 197-227. Bourguignon, F., and P.A. Chiappori, 1994. ―The Collective Approach to Household Behavior‖ Chapter 3 in The Measurement of Household Welfare, 70-85. Bourguignon, F., F. Ferreira and P. Leite, 2003. ―Conditional Cash Transfers, Schooling and Child Labor: Micro-Simulating Bolsa Escola‖, DELTA Working Paper. Cartwright, K., and A. Patrinos, 1999. ―Child Labor in Urban Bolivia.‖ In Grootaert, C., and H. Patrinos, eds. The Policy Analysis of Child Labor: A Comparative Study. New York: St. Martin’s Press. Coady, D., and S. Parker, 2009. ―Targeting Performance under Self-Selection and Administrative Targeting Methods.‖ Economic Development and Cultural Change, Issue 57, pp 559-587. Dammert, A. C., 2009. ―Heterogeneous Impacts of Conditional Cash Transfer Programs: Evidence from Nicaragua.‖ Economic Development and Cultural Change, forthcoming. Dar, A., N. Blunch, B. Kim and M. Sasaki, 2002. ―Participation of Children in Schooling and Labor Activities: A Review of Empirical Studies‖. Social Protection Discussion Paper No. 0221. Das Gupta, M., 1987. "Selective Discrimination Against Female Children in India." Population and Development Review, 13:77-101. 103 De Brauw, A., and J. Hoddinott, 2007. ―Must Conditional Cash Transfer Programs be conditioned to be effective? The impact of conditioning transfers on school enrollment in Mexico.‖ Washington DC: World Bank. Dehejia, R., and R. Gatti, 2002. ―Child Labor: The Role of Income Variability and Access to Credit Across Countries‖, NBER Working Papers, W9018. Del Carpio, X., 2007. ―Voices of Nicaragua, a Qualitative and Quantitative Approach to Viewing Poverty in Nicaragua‖ chapter 2, Nicaragua Poverty Assessment, Report No. 39736, The World Bank. Del Carpio, X., 2008. ―Does Child Labor Always Decrease with Income? An evaluation in the context of a development program in Nicaragua‖, The World Bank, Policy Research Working Paper Series 4694. Deolalikar, A., and W. Vijverberg, 1987. ―A Test of Heterogeneity of Family and Hired Labour in Asian Agriculture‖, Oxford Bulletin of Economics and Statistics, 49:3, 291-303. Diaz-Cayeros, A., F. Estevez, and B. Magaloni, 2007. Strategies of Vote Buying: Social Transfers, Democracy, and Welfare in Mexico. Chapters 3 of 4, available: http://www.stanford.edu/~albertod/buyingvotes/buyingvotes.html Djebbari, H., and J. Smith, 2008. ―Heterogeneous Impacts in Progresa‖, Journal of Econometrics, 134(1-2): 64-80. Duflo, E., 2003. ―Grandmothers and Granddaughters: Old-age Pension and Intra-household Allocation in South Africa‖, World Bank Economic Review, 17(10): 1-17. Duflo, E., R. Glennerster, and M. Kremer, 2007. ―Using Randomization in Developing Economics‖ Chapter 61, Handbook of Development Economics, Volume 4. pp 3895- 3962 Edmonds, E., 2003. "Does Child Labor Decline with Improving Economic Status?" NBER Working Paper 10134. Cambridge, Mass.: National Bureau of Economic Research. Edmonds, E., 2006. ―Understanding Sibling Differences in Child Labor‖, Journal of Population Economics, 19(4), 795-821. Edmonds, E. 2006. ―Child Labor and Schooling Responses to Anticipated Income in South Africa‖, Journal of Development Economics, 81(2), 386-414. Edmonds, E., 2007. ―Child Labor‖ forthcoming in The Handbook of Development Economics, vol. 4. Edmonds, E., and N. Schady, 2008. ―Poverty Alleviation and Child Labor‖, World Bank Policy Research Working Paper No 4702. The World Bank, Washington DC. Edmonds, E., and C. Turk., 2004. "Child Labor in Transition" In P. Glewwe, N. Agrawal and D. Dollar, eds., Economic Growth, Poverty and Household Welfare: Policy Lessons from Viet Nam. Washington D.C.: World Bank. 104 Ejrnaes, M., and C. Portner, 2004, "Birth order and the intra-household allocation of time and education," Review of Economics and Statistics 86: 1008-1019. Emerson, P., and A. Portela Souza, 2007. ―Child Labor, School Attendance and Intra-Household Gender Bias in Brazil‖. World Bank Economic Review. Emerson, P., and A. Portela Souza, 2008. "Birth order, child labor, and school attendance in Brazil," World Development 36(9): 1647-1664. Ermisch, J., and M. Francesconi, 2000. ―Educational Choice, Families and Young People’s Earnings‖ Journal of Human Resources, 35(1):146-176. Filmer, D., and N. Schady, 2008. ―Who Benefits? Scholarships, School Enrollment and Work of Recipients and their Siblings‖, mimeo, World Bank. Fizsbein, A and N. Schady, 2009. ―Conditional Cash Transfers: Reducing Present and Future Poverty.‖ Washington DC: World Bank. Foster, A., 1995. "Prices, Credit Constraints and Child Growth in Rural Bangladesh." Economic Journal, 105(43): 551-570. Foster, A., and M. Rosenzweig, 1994. ―A Test for Moral Hazard in the Labor market: Contractual Arrangements, Efforts and Health‖, Review of Economics and Statistics 76(2): 213-227. Glewwe, P., and P. Olinto, 2004. ―Evaluating the Impact Conditional Cash Transfers on Schooling: An Experimental Analysis of Hondura’s PRAF Program.‖ Final report for USAID. Washington DC: International Food Policy Institute. Grootaert, C., and R. Kanbur, 1995. ―Child Labor: An Economic Perspective‖, International Labor Review, 134(2), 187-203. Grosh, M., C. del Ninno, E. Tesliuc, and A. Oueghi, 2008. ―For Protection and Promotion. The Design and Implementation of Effective Safety Nets‖, World Bank, Washington DC Hazarika, G., and S. Sarangi, 2005. ―Household Access to Microcredit and Child Works in Rural Malawi‖, IZA Discussion Paper No. 1567, Kruger, D., and M. Berthelon (2003). ―How Households Economic Opportunity Affect Child Labor and Schooling in Nicaragua: Differential Effects by Gender‖, The Georgetown Public Policy Review 9, 1-16. Hoddinott, J., H. Alderman, and L. Haddad, 1997. ―Testing Competing Models of Intra- household Allocation‖, Chapter 8 in Intra-household Resource Allocation in Developing Countries, 129-141. Kim, H., 2005. ―Parental Investment Between Children With Different Abilities‖, mimeo University of Wisconsin at Madison. Kruger, D., and M. Berthelon, 2007. ―Work and Schooling: The Role of Household Activities Among Girls in Brazil‖, mimeo, Pontificia Universidad Católica de Valparaíso. 105 Levy, S., 2008. Good Intentions, Bad Outcomes: Social Policy, Informality, and Economic Growth in Mexico. Brookings Institution Press, Washington DC. Lindert, K., and V. Vincensini, 2008. ―Bolsa Familia nas Mancheres‖ Presentation at the World Bank, Washington DC. Macours, K., N. Schady, and R. Vakis, 2008. ―Cash Transfers, Behavioral Changes, and Cognitive Development in Early Childhood: Evidence from a Randomized Experiment.‖ World Bank Policy Research Paper, 4759. Macours, K., and R. Vakis, 2005. ―Weather risk and poverty in Nicaragua: Expanding Risk Management Options for the Poor: Pilot Program Objectives and Impact Evaluation Design‖, mimeo, Johns Hopkins University and The World Bank‖. Macours, K., and R. Vakis, 2008. ―Seasonal Migration and Early Childhood Development‖, World Development, forthcoming. Macours, K., and R. Vakis, 2009. ―Changing households’ investments and aspirations through social interactions: Evidence from a randomized transfer program in a low-income country‖, Policy Research Working Paper, TheWorld Bank Research Group. Maluccio, J., 2003. ―Education and Child Labor: Experimental Evidence from a Nicaraguan Conditional Cash Transfer Program‖, in Orazem, Sedlacek, and Tzannatos (eds.), Child Labor in Latin America. Maluccio, J., and F. Flores, 2004. ―Impact Evaluation of a Conditional Cash Transfer Program: The Nicaraguan Red de Proteccion Social‖, FCND Discussion Paper, No. 184, IFPRI, Washington DC Manacorda , M., 2006. ―Child labor and the labor supply of other household members: Evidence from 1920 America,‖ American Economic Review 96: 1788-1800. Mansuri, 2006. ―Migration, Sex Bias, and Child Growth in Rural Pakistan‖, World Bank Policy Research Paper 3946. Parker, S., L. Rubalcava and G. Teruel, 2007. ―Evaluating Conditional Schooling and Health Programs‖ Chapter 62, Handbook of Development Economics, Volume 4.pp-3963-4035 Parsons, D., and C. Goldin, 1989. ―Parental Altruism and Self-Interest: Child Labor among Late Nineteenth-Century American Families‖ Economic Inquiry 27, no. 4 (1989): 637-59. Patrinos, H., and G. Psacharopoulos, 1997. ―Family size, schooling and child labor in Peru – An empirical analysis.‖ Journal of Population Economics 10(4), 387-406. Ponczek, V., and A. Portela Souza, 2007. ―The Causal Effect of Family Size in Child Labor and Education‖, mimeo. Raju, D., 2005. ―Banning Harmful Child Labor: A Labor Market and Welfare Analysis‖, mimeo 106 Raju, D., 2006. ―The Effects of Conditional Cash Transfer Programs on Child Work: A Critical Review and Analysis of the Evidence‖, The World Bank. Rangel, M., 2008. ―Is Parental Love Colorblind? Allocation of Resources within Mixed Families.‖ BREAD Working paper No. 167. Ranjan, P., 2001. ―Credit Constraints and the Phenomenon of Child Labor.‖ Journal of Development Economics. 64(1): 81-102. Ravallion, M., 2009. Shocks, Crises, and Safety Nets. A presentation given in Cairo, Egypt at the Bailing Out the World’s Poorest, ERF conference. Rogers, C.A., and K. Swinnerton, 2002. ―The Theory of Exploitative Child Labor‖, mimeo. Rogers, C. A., and K. A. Swinnerton, 2004. "Does Child Labor Decrease when Parental Incomes Rise," Journal of Political Economy, Volume 112, (August) pp. 939-946. Rose, E., 1999. "Consumption Smoothing and Excess Female Mortality in Rural India," The Review of Economics and Statistics, 81(1):41-49. Rosenzweig, M., and T. P. Schultz, 1982. "Market Opportunities, Genetic Endowments and Intra- family Resource Distribution: Child Survival in Rural India," The American Economic Review, 72(4): 803-815. Shultz, T., 2004. ―School Subsidies for the Poor: Evaluating the Mexican PROGRESA Poverty Program‖ Journal of Development Economics, vol. 74, no. 1, Special Issue June 2004, pp. 199-250. Skouffias, E., and S. Parker, 2001. ―Conditional Cash Transfers and their Impact on Child Work and Schooling: Evidence from the Progresa Program in Mexico‖, IFPRI FCND Discussion Paper No. 123. Swinnerton, K. A., and C. A. Rogers, 1999. ―The Economics of Child Labor: Comment,‖ American Economic Review 89(5), December, pp. 1382-85. Thomas, D. 1994. ―Like Father, Like Son, Like Mother, Like Daughter: Parental Resources and Child Height.‖ Journal of Human Resources 29(4): 950-88. Thomas, D, 1997. ―Income, Expenditures, and Health Outcomes: Evidence on Intra-household Resource Allocation‖ in Chapter 9 in Intra-household Resource Allocation in Developing Countries, 142-164. Wilhelm, M., 1996. ―Bequest Behavior and the Effect of Heirs’ Earnings: Testing the Altruistic Model of Bequests‖, American Economic Review, 86(4):874-892. Zucco, C., 2009. ―Cash-transfers and voting behavior: An empirical assessment of the political impacts of the Bolsa Familia Program‖. Mimeo, Princeton University. 107 Appendix A: Details of the Nicaraguan CCT Pilot Program The main household survey, collected in both rounds, contains eleven sections of household and individual level data on various socio-economic indicators on approximately 4400 households in the North-region of Nicaragua. In the follow-up survey, additional questions on child labor were added, to capture particular children’s work in chores (wood and water gathering) and domestic work. Given the focus on child labor and child outcomes, Table A1 shows descriptive statistics for all households with children, first for all children together and then separated by gender of the child. The quantitative data was complemented with qualitative work, based on focus groups and semi-structured interviews with a wide set of beneficiaries and other local actors in treatment and control communities (see Aguilera et al., 2006). The households in this region are mostly subsistence farmers who rely on basic grain agriculture and some animal farming activities; agricultural participation of children is not uncommon and overall child labor, including domestic activity, is commonplace. Table A2 illustrates some descriptive statistics on child work in 2006 by gender and age groups; more than 50 percent of children between 8 and 15 years old work in non-domestic work, with the largest concentration of work hours taking place in agricultural activities, 4 hours for boys and 2 hours for girls. When I include domestic work into the total work hours calculation I get 10 hours on average for both boys and girls. Females spend on average 9.7 hours a week in non-skill enhancing activities while males spend little over 10 hours in this activity. The difference for control and treatment groups are slightly wider, 1 hour more for girls in untreated households and 1.2 hours more for boys in untreated households. 108 Table A1: Means for variables in the analysis by gender Income for household (in 1000) 2005 4256 1397.86 1389.68 1405.70 basic intervention 4181 24% 23% 25% training intervention 4181 25% 24% 26% business grant intervention 4181 25% 25% 25% age of child in 2006 4289 11.58 11.58 11.58 household size 2005 4289 6.91 6.95 6.87 education level of head 2005 4289 1.32 1.32 1.32 age of head 2005 4289 44.78 44.79 44.76 gender of the household head 4289 0.84 0.83 0.85 (male=1, female=0) in 2005 gender of child (boys=1, girls=0) 2006 4289 51% -- -- #of children 5 yr & under 2005 4289 0.72 0.74 0.71 #of children 6-15 years 2005 4289 2.85 2.87 2.83 #of children 15-24 years 2005 4289 1.13 1.15 1.10 dist. in time to municipal hq 2005 4289 1.59 1.60 1.57 dist. in time to prim. School 2005 4289 0.27 0.27 0.27 dist. in time to health center 2005 4289 1.17 1.18 1.15 tot community owned land/tot population 4235 7.88 8.01 7.75 in community 2005 tot # of kids in age group in the community /tot comm 4289 0.32 0.32 0.31 population 2005 Girls Boys Kids age 8-15 (7-14 in 2005) All kids Total Observations 109 Table A2: Descriptive child labor data by age and gender ALL Obs. All Obs. Con trol Obs. Treat Obs. Basic B. Obs. Train in g B. Obs. Gran t B. Child works in non-dome stic a ctivity 4287 54.4% 1096 55.3% 3191 54.1% 1043 52.7% 1079 53.1% 1069 56.3% Child works in non-a gro a ctivity 4287 8.7% 1096 4.7% 3191 10.1% 1043 7.0% 1079 6.4% 1069 16.8% Child works in a gro or ca ttle a ctivity, inc. a s a a g-da y la bor 4287 44.8% 1096 46.2% 3191 44.3% 1043 42.9% 1079 46.0% 1069 44.0% Child works in house hold chore s (dome stic a ctivity) 4287 91.6% 1096 91.7% 3191 91.6% 1043 90.5% 1079 92.2% 1069 92.0% Child works in dome stic a nd non-dome stic a ctivitie s 4287 91.6% 1096 91.7% 3191 91.6% 1043 90.5% 1079 92.2% 1069 92.0% Tota l hrs p/wk worke d in non-a gro a ctivity 4214 0.399 1078 0.195 3136 0.469 1031 0.34 1067 0.291 1038 0.781 Tota l hrs p/wk worke d by child in a gro (inc pe on) a nd ca ttle a ct. 4220 2.956 1071 3.218 3149 2.867 1031 2.84 1064 2.850 1054 2.915 Tota l hrs p/wk worke d in chore s 4235 7.050 1072 7.724 3163 6.821 1034 7.00 1069 6.698 1060 6.767 Tota l hrs p/wk in non-skill (physica l) work 4287 9.870 1096 10.699 3191 9.591 1043 9.75 1079 9.446 1069 9.583 Tota l hrs p/wk worke d in chore s a nd work 4289 10.482 1097 11.201 3192 10.234 1043 10.26 1079 9.881 1070 10.568 FEMALE Obs. All Obs. Control Obs. Treat Obs. Basic B. Obs. Training B. Obs. Grant B. Child works in non-dome stic a ctivity 2102 51.2% 564 53.9% 1538 50.3% 494 49.4% 516 50.0% 528 51.3% Child works in non-a gro a ctivity 2102 10.1% 564 5.5% 1538 11.8% 494 9.7% 516 7.6% 528 18.0% Child works in a gro or ca ttle a ctivity, inc. a s a a g-da y la bor 2102 41.4% 564 45.6% 1538 39.9% 494 38.3% 516 42.6% 528 38.6% Child works in house hold chore s (dome stic a ctivity) 2102 92.1% 564 92.7% 1538 91.8% 494 92.1% 516 92.1% 528 91.3% Child works in dome stic a nd non-dome stic a ctivitie s 2102 92.1% 564 92.7% 1538 91.8% 494 92.1% 516 92.1% 528 91.3% Tota l hrs p/wk worke d in non-a gro a ctivity 2063 0.4556 558 0.222 1505 0.542 485 0.495 510 0.335 510 0.794 Tota l hrs p/wk worke d by child in a gro (inc pe on) a nd ca ttle a ct. 2097 1.9512 563 2.287 1534 1.828 493 1.706 514 1.839 527 1.932 Tota l hrs p/wk worke d in chore s 2061 7.9014 544 8.480 1517 7.694 487 7.930 510 7.673 520 7.494 Tota l hrs p/wk in non-skill (physica l) work 2102 9.6938 564 10.462 1538 9.412 494 9.520 516 9.415 528 9.309 Tota l hrs p/wk worke d in chore s a nd work 2104 10.267 565 10.675 1539 10.117 494 10.253 516 9.857 529 10.244 MALE Obs. All Obs. Control Obs. Treat Obs. Basic B. Obs. Training B. Obs. Grant B. Child works in non-dome stic a ctivity 2185 57.4% 532 56.8% 1653 57.6% 549 55.7% 563 56.0% 541 61.2% Child works in non-a gro a ctivity 2185 7.3% 532 3.8% 1653 8.5% 549 4.6% 563 5.3% 541 15.7% Child works in a gro or ca ttle a ctivity, inc. a s a a g-da y la bor 2185 48.0% 532 46.8% 1653 48.4% 549 47.0% 563 49.0% 541 49.2% Child works in house hold chore s (dome stic a ctivity) 2185 91.2% 532 90.6% 1653 91.4% 549 89.1% 563 92.4% 541 92.8% Child works in dome stic a nd non-dome stic a ctivitie s 2185 91.2% 532 90.6% 1653 91.4% 549 89.1% 563 92.4% 541 92.8% Tota l hrs p/wk worke d in non-a gro a ctivity 2151 0.344 520 0.165 1631 0.401 546 0.198 557 0.251 528 0.768 Tota l hrs p/wk worke d by child in a gro (inc pe on) a nd ca ttle a ct. 2123 3.949 508 4.249 1615 3.855 538 3.875 550 3.795 527 3.898 Tota l hrs p/wk worke d in chore s 2174 6.242 528 6.946 1646 6.016 547 6.180 559 5.808 540 6.066 Tota l hrs p/wk in non-skill (physica l) work 2185 10.048 532 10.951 1653 9.757 549 9.954 563 9.474 541 9.851 Tota l hrs p/wk worke d in chore s a nd work 2185 10.688 532 11.759 1653 10.344 549 10.262 563 9.904 541 10.885 ages 8-15 ages 8-15 ages 8-15 110 There are no drastic differences between the three interventions types in this activity. The differences in skill-enhancing activities are striking; kids in treated households work on average .5 hours per week while kids in control households work .2 hours; females in treated and untreated households work more in this activity than males and kids in households receiving the business grant intervention work .8 hours on average. One of the main advantages of this data-set is that there are two rounds of data collected (2005 and 2006). The second round of the data however, was collected 9 months after the program began, 3 months before it concluded which may influence why some of the benefits had not been completely delivered. I can observe the same household over a years’ time which can help separate changes of child labor over time that are attributable to exogenous changes, such as the conditional cash transfer program under evaluation, as well as other economic environment or labor market changes. The multi-layer random process can potentially pose problems when estimations are at the household level (as in the subsequent essays) and the randomization is at the community level. Similar to other studies (Gitter, 2005), I test for biases caused by potential heterogeneity presented by the randomization design, between treatment and control groups, and find that on average, clustered at the community level, the sub-samples under evaluation were comparable to each other and the randomization did not suffer from selectivity problems. More specifically on the randomization tests, I followed Behrman and Todd (1999) by selecting sets of outcome variables and applying one of three types of tests. The type of estimation to test the equality of their means and distributions depends on whether the variable of interest is binary, continuous or categorical. I first test at the community level (stage I) and then the household level and within treatment sub-groups (stage II). In this part of the analysis I also conduct the same set of tests on control households as compared to treatment households and sub- 111 treatment groupings. Lastly I present the results obtained for the sample derived with the input of village leaders on beneficiary selection. The first test used is the Kolmogorov-Smirnov test (KS) to analyze the statistical equality of two populations when using continuous variables. This test makes no assumption about the distribution and it works by identifying the difference between two population distribution functions based on their two independent random samples. I hypothesize that in both samples are drawn from identical populations and both samples are drawn from significantly different populations. I use the p-value obtained by using the ksmirnov command in stata 9.1 to determine whether the samples compared are significantly different from each other. Pearson chi-squared (chi) test of equality of the cell proportions is used for analyzing all the categorical variables in the selected sample. This test uses frequencies rather than means and computes it in a two-way table. The chi-square test becomes increasingly significant as the numbers deviate from the sample patterns. The corresponding p-value are calculated and used to determine whether to reject or not. The third type of test used is the t-test. This test is specifically used for binary discrete variables taking on values of 0 and 1. This test evaluates the difference in means between two groups (for example: treatment and control), follows a standard normal distribution and tests the equivalence between the probability that each of the variables takes on the value 1. Further descriptions of each of these tests can be found in the appendix of Behrman and Todd (1999). I obtained community means by using the collapse command in Stata 9.1 and ended up with only proportions at the community level. Since percentages are continuous variables I only used the Kolmogorov-Smirnov test to analyze all community indicators in stage I. I used all tests for the other parts of the analysis, always clustered at the community level to allow for the variance of the residual to be community specific. 112 A quick glance of the summary table (Table A1) demonstrates how effective the randomization process was; but, I will, for the purposes of clarity, select a set of variables with significant p-values for further analysis. At the community level, 11.63 percent of variables had significant p-values; and only 3.1 percent have p-values less than .01. The analysis of community variables table (Table A3) lists all variables with a p-value smaller than .10, the mean for control and treatment groups, the standard deviation and the p-values from the means test. According to the means test I find that only four variables are significant at less than 90 percent, a low number that is worth noting. Table A3: Analysis of Community Variables with a Significant P-Value note: * .90%,**-95%, and ***-99percent significant. All calculations are clustered at the community level. It is also important to note that there appears to be no systematic biases toward the control or the treatment communities. For example from the means reported I observe that kids in the control community have a slightly lower early cognitive score in the TVIP test than those in treatment communities; however, the difference is only 1.6%. On the other hand, children in the control group are 7 percent more likely to have received Vitamin A in the last six months than KS p-val sd C mean T mean ttest p-val health consumption pc 0.044** 337.819 306.541 342.676 0.585 standardize TVIP score 0.081** 5.020 65.706 67.388 0.086* hours per week worked by child in nonag 0.004*** 0.817 0.264 0.593 0.037** servhh 0.065* 0.063 0.046 0.048 0.842 received vitamin A last 6 mo (<5 yrs) 0.085* 0.194 0.754 0.684 0.064* paid for treatment last mo (<5 yrs) 0.004*** 0.093 0.080 0.086 0.723 z-scores for underweight (<5 yrs) 0.081* 0.672 -0.943 -1.210 0.041** hh plans to insure against shocks to cope 0.047** 0.029 0.012 0.015 0.643 age of hh head 0.014** 5.160 47.311 45.749 0.120 basic services program 0.024** 0.139 0.121 0.136 0.568 hh suffered income shock 0.030** 0.062 0.052 0.066 0.241 currently pregnant 0.000*** 0.033 0.026 0.041 0.024** no cultivation 0.082* 0.306 0.687 0.682 0.927 C_T all Variables w/ significant p-values using the KS test STAGE I randomization at the community level 113 children in the program. Another example of how random the samples are is illustrated by the variable measuring how many kids have had a health check-up; 95 percent for both categories. The direction of these and other variables pertinent to the same age group indicate there is no systematic bias in either direction. In other words, kids in the control group do not appear to be better or worse off in any consistent manner. The distribution of the variable measuring mean hours per week of kids working in non- agriculture activities to analyze why the p-value in the KS test is significant. Figure A1 shows the distribution of the variable for both, control and treatment, including all kids in this age regardless of whether they report working or not in this activity. The mean hours are .26 and .59 for kids in the control and treatment groups respectively. Figure A1: Mean hours worked in non-agricultural labor (including children with 0 hours) I follow with Figure A2 where I show the same distribution comparison except in this figure I exclude children with 0 hours in this activity; the resulting p-value for the KS test is not significant (0.147) and the gap in the means is reduced substantially (.55 for the control group and .73 for the treatment group). 114 Figure A2: Mean hours worked in non-agricultural labor (excluding children with 0 hours) I turn from the community analysis to testing the experimentality at the household level. The results from analyzing the household level allocation into treatment types indicate that the randomization process was effective. Only 10 percent of all the variables tested turn out a significant p-value. As described earlier, this step of the randomization was done at the household level and test type differs based on the type of variable being analyzed. Small sample sizes may be biasing the results. An investigation of the source of non-randomness shows that the stunting indicator, for children aged 3 through 4, only includes 765 children in the sample; 376 in the traditional package (T1) and 389 in the vocational training package (T2). The immunization indicator is another example; this variable only encompasses 236 total children. Having the adequate sample size is important, as small sample numbers can yield to misleading results. Much like the community sample I find that variables are not systematically biased in one direction or another. Figure A3 show the share of children in the same age cohort who had diarrhea last month and were sick with something other than diarrhea. In panel A, Business grant package (T3) has the highest percentage of children with diarrhea in the prior month and the 115 vocational training package (T2) has the lowest percentage (24%). In panel B, 43 percent of kids in T2 reported being sick; the highest of all the sub-groups. The Kolmogorov-Smirnov test of equality between the empirical cumulative distribution functions of food consumption per capita distribution (Figure A4) appears similar in the figure below, and the t-test of the same variable reveals that the difference in means between T1 (3,615) and T2 (3,482) is not significantly large (p-value of .415). Figure A3: Percent of children who had diarrhea last month and children who were sick with something other than diarrhea last month Figure A4. Food consumption per capita distributions 116 It’s worth noting that while the randomization process was performed at the community level (in stage 1) and the household level (within treatment types in stage 2), some of the estimators were calculated at the individual level; child health outcomes being the most commonly estimated at the individual level. This may lead to some heterogeneity within treatment and control samples and some over or under estimation of program effects in the evaluation process. This analysis confirms the effectiveness of the randomization process and concludes that whether at the community level or the household level, the process successfully led to a randomly obtained sample. In other words, the treatment and control groups and treatment sub-groups are statistically equivalent to one another; therefore enabling their use as counterfactuals. In other words, the I conclude that ex-ante, all people in the targeted region had an equal probability of being part of the intervention. Given the type of interventions implemented and the focus of the following chapters on child labor, I present a set of outcome indicators using the follow-up survey to motivate the work. In Table A4 I use descriptive data to illustrate the intensity of work for all children (8-15 years of age) by income quintiles; this illustration is broken down into young kids (8-12) and older kids (12.1-15) for further evaluation. There is an increase in work hours for all activities in the higher income quintiles; domestic work is the only exception, exhibiting the exact opposite relationship. The data also show that older kids work approximately 4 hours more per week on average in total work hours, with equally large differences in non-physical work and physical labor activities. 117 Table A4: Labor participation and hours by Income quintiles In relation to the intra-household child labor allocation chapter, Table A5 presents the randomization results for the sample of eligible households with children between 6 and 15 years old. The differences between households in treatment and control communities are small and not statistically significant. Similarly, the differences between households with the business grant package and households in the control communities are generally small and not significant. All 8-15 both genders Obs. All Obs. All Obs. All Obs. All Obs. All Total hrs p/wk worked in non-agro activity 904 0.2782 927 0.4229 928 0.3125 816 0.428 639 0.6213 Total hrs p/wk worked by child in agro (inc peon) and cattle act. 895 2.4688 931 3.0773 924 2.704 815 3.21166 655 3.4886 Total hrs p/wk worked in chores 900 7.2928 929 6.9698 928 7.1331 823 6.94848 655 6.8374 Total hrs p/wk in non-skill (physical) work 909 9.6514 942 9.915 939 9.7103 834 9.99532 663 10.201 Total hrs p/wk worked in chores and work 911 10.02 942 10.478 939 10.006 834 10.9558 663 11.198 Age group 8-12 both genders Obs. All Obs. All Obs. All Obs. All Obs. All Total hrs p/wk worked in non-agro activity 514 0.2393 504 0.3433 523 0.2524 447 0.2774 355 0.4873 Total hrs p/wk worked by child in agro (inc peon) and cattle act. 515 1.7462 507 2.1302 526 1.6492 448 2.35156 363 2.6088 Total hrs p/wk worked in chores 513 6.5058 505 6.082 523 6.3002 450 5.99444 361 6.0166 Total hrs p/wk in non-skill (physical) work 516 8.2109 509 8.156 529 7.8686 453 8.28035 364 8.5687 Total hrs p/wk worked in chores and work 516 8.393 509 8.5921 529 8.069 453 8.92274 364 9.2912 Age group 12.1-15 both genders Obs. All Obs. All Obs. All Obs. All Obs. All Total hrs p/wk worked in non-agro activity 390 0.3295 423 0.5177 405 0.3901 369 0.60976 284 0.7887 Total hrs p/wk worked by child in agro (inc peon) and cattle act. 380 3.4482 424 4.2099 398 4.098 367 4.26158 292 4.5822 Total hrs p/wk worked in chores 387 8.3359 424 8.0271 405 8.2086 373 8.09946 294 7.8452 Total hrs p/wk in non-skill (physical) work 393 11.543 433 11.983 410 12.087 381 12.0344 299 12.189 Total hrs p/wk worked in chores and work 395 12.145 433 12.694 410 12.506 381 13.373 299 13.52 4th Quintile 1st Quintile 2nd Quintile 3rd Quintile 4th Quintile 5th Quintile 5th Quintile 1st Quintile 2nd Quintile 3rd Quintile 4th Quintile 5th Quintile 1st Quintile 2nd Quintile 3rd Quintile 118 Table 5: Randomization results Household Characteristics # Obs. Age household head 1596 43.65 43.59 44.28 (0.94) (0.50) Male household head 1597 0.850 0.859 0.875 (0.71) (0.37) Literate household head 1597 0.618 0.652 0.654 (0.28) (0.36) # years education household head 1496 2.734 2.721 2.785 (0.94) (0.80) Household size 1597 6.882 6.918 7.041 (0.83) (0.41) Number of hh. members under 5 1597 0.804 0.766 0.778 (0.60) (0.74) Number of hh. members 5-14 1597 2.824 2.919 2.934 (0.16) (0.19) Number of hh. members 15-24 1597 1.060 1.050 1.081 (0.89) (0.82) Number of hh. members 25-64 1597 2.039 2.050 2.067 (0.81) (0.63) Number of hh. members 65 or older 1597 0.143 0.119 0.148 (0.37) (0.87) Number of rooms in the house 1597 1.652 1.601 1.626 (0.61) (0.81) Own land 1597 0.681 0.656 0.667 (0.56) (0.76) Distance to school (min) 1597 0.327 0.261 0.262 (0.10) (0.11) Distance to health center(min) 1597 1.216 1.156 1.146 (0.70) (0.66) Distance to municipal headquarters 1597 1.693 1.578 1.623 (0.53) (0.72) At least one household member active in: Self-employment in agriculture 1597 0.911 0.880 0.901 (0.26) (0.73) Self-employment in livestock 1597 0.536 0.474 0.488 (0.36) (0.52) Wage labor in agriculture 1597 0.592 0.651 0.697 (0.19) (0.04) Elaboration food products 1597 0.060 0.075 0.079 (0.50) (0.45) Manufacturing (self-employment) 1597 0.036 0.029 0.036 (0.67) (0.97) Commercial activities 1597 0.082 0.080 0.066 (0.91) (0.38) Services (self-employment) 1597 0.053 0.049 0.056 (0.80) (0.89) Wage employment 1597 0.174 0.191 0.191 (0.58) (0.59) Skilled wage employment 1597 0.036 0.047 0.056 (0.43) (0.20) Total consumption per capita (cordoba) 1584 4039 4248 4255 (0.42) (0.47) Total income per capita (cordoba) 1584 3278 3602 3611 (0.13) (0.23) Income from agricultural wage labor 1589 2150 1959 2024 (0.49) (0.67) income for elaboration food products 1590 115.9 120.5 124.4 (0.92) (0.88) income from commercial activities 1589 175.4 205.7 154.6 (0.62) (0.76) income from manufacturing (self-employment) 1589 26.97 21.5 30.0 (0.73) (0.86) Income from services (self-employment) 1590 155.7 125.0 105.0 (0.63) (0.43) Income from nonagricultural wage labor 1578 681.3 975.6 904.2 (0.12) (0.34) Income from temporary migration 1592 2663 2838 2432 (0.69) (0.62) Monetary income from agricultural self-employment 1591 728.5 829.5 865.4 (0.50) (0.41) Monetary income from livestock 1591 737.1 651.2 558.1 (0.49) (0.20) P-value account for clustering at community level. All values refer to intent-to-treat Control (C) productive investment package (T3) Treatment (T) P-value (T-C) P-value (T3-C) Table A5: Randomization results for sub-sample (with children 6-15 years) 119 Appendix B: Intra-household Child Labor Analysis A possible concern with an analysis that investigates intra-household allocation of child labor in different categories is that the dependent variable is censored at 0. In case of low participation rates the results might then be driven by a few outliers with a large number of hours. As discussed, in this paper the data on the number of hours was trimmed for outliers, and we only consider allocation along dimensions for which there were sufficient households with children in different categories (see the discussion in section 3). To further reduce any concerns about outliers, this appendix presents descriptive statistics of children’s participation in each of the categories (a binary variable) and sheds light on the distribution of the conditional number of hours of work in each activity. Table B1 shows the conditional hours of work in each of the categories, separately for treatment and control at follow-up. Not surprisingly, the means differ, but the minimum and maximum values are in the same range for both groups and there are no clear outliers in either group. Tables B2 through B5 then show participation in child labor by age, gender, age-gender, and school attendance category for the control group (analog to the data on the number of hours). These data generally show the same patterns as we discussed before: more specialization of boys in agriculture and livestock, while girls specialize in domestic chores, livestock and non- agricultural work activities. They also show that participation is common for most of the activities considered. 120 Table B1: Number of hours in each activity, conditional on participation Table B2: Child labor by gender: Intra-household allocation 121 Table B3: Child labor by age: Intra-household allocation Table B4: Difference in child labor between children attending school and those not attending: Intra-household allocation Table B5: Difference in child labor of children below their optimal grade level versus others: Intra-household allocation

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Political determinants and economic effects of HIV/AIDS: a push for the multisectoral approach

PDF

Against the wind: labor force participation of women in Iran

PDF

Social policy and presidential ideology in Latin America: the political economy of social spending and anti-poverty programs

PDF

From structure to agency: Essays on the spatial analysis of residential segregation

PDF

Essays on the econometrics of program evaluation

PDF

Essays on housing market behavior analysis within the international context

PDF

Building and sustaining effective school leadership through principal mentoring in an urban school context

PDF

Essays on the econometrics of program evaluation

PDF

Essays on political economy of privatization

PDF

Adaptation, assets, and aspiration. Three essays on the economics of subjective well-being

PDF

Mechanisms of motor learning in immersive virtual reality and their influences on retention and context transfer

PDF

The effect of active and passive listening condition on young children's musical understanding of familiar and unfamiliar melody through researcher-invented notation

PDF

Intergenerational transfers & human capital investments in children in the era of aging

PDF

The effect of active and passive listening condition on young children's musical understanding of familiar and unfamiliar melody through researcher-invented notation

PDF

Effects of a formulary expansion on the use of atypical antipsychotics and health care services by patients with schizophrenia in the California Medicaid Program

PDF

The impact of programs, practices, and strategies on student academic performance: a case study

PDF

Essays on health and aging with focus on the spillover of human capital

PDF

A study of one hundred unmarried mothers in Los Angeles

PDF

The determinants and measurement of human capital

PDF

The development implications of China’s Belt and Road Initiative for Russia, Kazakhstan and Belarus

Asset Metadata

Creator Carpio, Ximena V. Del (author)

Core Title Essays on child labor and poverty in the context of a conditional cash transfer program in Nicaragua

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 04/12/2010

Defense Date 05/16/2009

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Child labor,Economic development,OAI-PMH Harvest,Poverty,social program

Place Name Nicaragua (countries)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Nugent, Jeffrey B. (committee chair), Heikkila, Eric J. (committee member), Rosendorff, Peter (committee member), Wise, Carol (committee member)

Creator Email xdelcarpio@worldbank.org

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m2906

Unique identifier UC1451247

Identifier etd-carpio-3467 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-319179 (legacy record id),usctheses-m2906 (legacy record id)

Legacy Identifier etd-carpio-3467.pdf

Dmrecord 319179

Document Type Dissertation

Rights Carpio, Ximena V. Del

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu