MASTER THESIS
Born and raised on PROGRESA: does receiving a conditional cash transfer have an impact on education and labor outcomes of young adults? Evidence from Mexico.
Author:
Patricia Pelayo Romero
Supervisor: Prof. Dr. E.J.S. Plug
Master in Economics
Abstract
I study the impact of the PROGRESA scholarship on young adults’ secondary school attendance, high school attendance, entering the labor market, and reporting positive income using data from the PROGRESA 2013 survey ESJóvenes. PROGRESA carries out periodic evaluations of the program in education, labor, and socioeconomic characteristics of the beneficiaries and other household members. By reduced form estimates, I find a positive and significant impact of the scholarship on education outcomes and little to no significance on labor outcomes across sub-samples. However, this estimation might be biased due to attrition issues. By implementing sensitivity checks to unintended outcomes – such as family size - and heterogeneity checks with different sub-samples, I am able to find similar results across sub-samples and find that the impact is consistent throughout them.
Table of Contents
Introduction……….4
Theoretical Framework and Literature Review………..8
Human Capital Investments………8
Cash Transfer Programs...10
Description of the PROGRESA Program……….15
Randomization………...18
Conditionality………19
Literature Review...………22
Data and Research Design………25
Data………...25
Attrition……….33
Estimates...………...35
Sensitivity and Heterogeneity Checks, and Discussion………41
Sensitivity Regressions………...41 Heterogeneity Regressions………45 Discussion……….52 Conclusion………55 Appendix……...………57 References……….58
Chapter 1
1. INTRODUCTION
Solid, coherent policies and plans are the bedrock on which to build sustainable education systems to achieve development goals. Education is often described as a crucial outcome for public policy and for public interest. At a private level, your education affects your earnings, your employment opportunities, your networks, and your chance of succeeding in life if your background is, from the start, especially in disadvantageous circumstances. It may also have an impact on your health, your family structure, and other numerous aspects that are the building blocks of well-being. At a public level, a country’s productivity is not only determined by capital and technology, but also by the capabilities of its labor force and their stock of skills. Uneducated citizens deter prosperity and output growth. Equality in opportunities is a major determinant to the degree of social and intergenerational mobility. (OECD, 2001)
Economic analysis can offer important insights for policy-makers. It is comprised of a strong quantitative approach, and a comprehensive theoretical framework for decision-making and policy creation. Moreover, it is focused on establishing links and impacts between policy and outcomes. These features make policy evaluation crucial for the creation of social welfare programs, particularly in developing nations. (Hanushek & Wößmann, 2010)
The accumulation of human capital is seen as an investment decision and a consumption trade off. Individuals forgo a proportion of income during the period of education in return for an increased future income. As any investment decision, it is only profitable if the benefits outweigh the costs. In some contexts, the costs are not only tuition fees and schooling supplies, but also the opportunity cost of not generating income for basic needs. (Hanushek & Wößmann, 2010)
I was interested in the PROGRESA program, a conditional cash transfer program that started in Mexico in the late nineties as a consequence of the economic crisis in 1994 for poverty alleviation. A relevant feature of the program was the fact that it tried to increase women’s bargaining power in the household and try to foster an increased interest in girls’ education in rural areas in Mexico. (Gobierno Ejecutivo Federal, 1997)
Regardless of the interest they had on gender equality, the main objective was to improve the well-being of the individuals in the most disadvantaged situation in Mexico and create opportunities that would break the intergenerational cycle of poverty. (Gutiérrez, Norman, & Alcalá, 2013) This policy was supposed to translate into an economic revolution that would break the ties of poverty in rural areas and increase human capital investment of the poorest households.
Reaching this objective required a high level of transparency, an increased monitoring procedure, education for the heads of the household, and an integral policy that would further improve the education effects. This implied important costs and resource allocation trade-offs for the Mexican government: health clinics had to be established, links between schools
and PROGRESA officials had to be made, etc. It was, and still is, the most important social program in Mexico. It was ambitious and expensive. (Gutiérrez et al., 2013)
Hence, the issue of efficiency must be raised: does the scholarship have a positive and significant impact on schooling and labor outcomes? Are benefits in terms of educational outcomes large enough to legitimize those investments? Did the program increase schooling attendance of girls? The aim of this analysis is to these questions by measuring the impact of being part of the PROGRESA treatment group on schooling and labor market outcomes.
To answer this question, I use the PROGRESA database. It is comprised of periodical surveys that are used to evaluate the progress of the policy. This database provides a rich and high-quality data for both of the surveys that I consulted: the pre-program ENCASEH 1997 survey as a baseline, and the ESJóvenes 2013 was used to measure the educational and labor market outcome of subsamples in the survey.
The data has a randomized controlled treatment design, which facilitates outcome evaluation, assuming that any differences between control and treatment groups are produced merely by chance. The effects of the treatment vary amongst subsamples, but remain positive and significant for educational outcomes, and generally insignificant for labor market outcomes. A group that is interesting to observe is the female group, which is expected to experience a higher impact of the treatment overall, given that there is a greater emphasis on the female group. (Gutiérrez et al., 2013)
This analysis is constructed as follows. The first part explores results in the literature on the topics of human capital, conditional cash transfers, the importance of conditionality, the impact of attrition and a description of the program. In the second section, I present the data used and its specificities. The third part displays the results and limitations of the reduced-form estimates of the impact of the treatment on education and labor outcomes of the sub-samples that were analyzed. In the fourth section, I perform sensitivity checks and I run heterogeneity regression to address the issues raised in the previous section and give validity to the model. Finally, I suggest further research that could be carried out with the database.
Chapter 2
2. THEORETICAL FRAMEWORK AND LITERATURE REVIEW
This analysis focuses on the influence of government sponsored scholarships on the schooling enrollment and potential labor market outcomes of young adults after fifteen years of being part of the PROGRESA conditional cash transfer program. The program started in 1997 with a baseline survey that would become the basis for randomization into treatment and control households. I use two surveys: the baseline survey of 1997 and the follow-up survey that was conducted in 2013, which focused on control and treatment individuals aged between 14 and 27 years old. This chapter explores the most important concepts used for the analysis: human capital theory, conditional cash transfer models, randomized controlled trials and the importance of conditionality for political economy. Additionally, I also describe the PROGRESA cash transfer program.
2.1 Human capital
Worldwide there are more than 200 million children under the age of five years who live in poverty. Their potential for growth, cognition, and socioemotional development is restricted by economic shortcomings. Infants and toddlers growing up in poverty are exposed to ailments that aggravate their condition further. They are subject to poor sanitation, large family size, and fewer household resources. As they grow up, children in developing nations, become comparatively underdeveloped to healthier adults, and are likely to have lower
wages, poorer health and fewer opportunities. Thus, perpetuating the effect of poverty and creating a vicious cycle that is quasi-impossible to break later in life. (Grantham-McGregor, S., Cheung, Y. B., Cueto, S., Glewwe, P., Richter, L., Strupp, B., & the International Child Development Steering Group., 2007)
In Mexico, specifically, in the period from 2008 to 2012, the poverty levels observed in the underage population was significantly higher (53.8%) than in the adult population (40.7%). Which seems to indicate that this segment is more sensitive to poverty effects than any other age group in the country. The economic shortcomings are not limited to credit constraints, but also to opportunity inequalities. Extreme poverty observed in 2012 in the population from 0 to 17 years old was 16.6%. As for the population aged 18 to 64, the extreme poverty ratio was 13.1. (Gutiérrez et al., 2013)
Early childhood is a period of rapid socio-physiological development and is paramount to address individuals in this period to optimize intervention. Early experiences can modify biochemistry and the architecture of neural circuits. Such experiences happen within a limited time frame. This period is termed as sensitive. Intuitively, if late investment in human capital is a good substitute for early investment, the early years are not critical. If it is not a good substitute, then the early period is critical. (Cunha, Heckman, Lochner, & Masterov, 2006)
The primary method to best intervene on child future schooling advancement in many developing countries – specially in Latin America – is through conditional cash transfers. Thereby, addressing the larger issue of poverty alleviation. In traditional cash
transfers or welfare programs, families receive cash benefits because the household falls below a certain income cutoff or lives within a geographically targeted region. These were, previously, the only criteria determining eligibility for participation in welfare programs. In the case of the PROGRESA program, the aid reception relies on meeting human capital investment conditions. (Grantham-McGregor, S. et al., 2007)
Based on human capital theories and given the national economic crisis in Mexico in 1994-1995, the Mexican government explored the possibility of creating a poverty alleviation program that covered the three most important dimensions of human capital investment: education, nutrition, and health. And thus, they launched the PROGRESA conditional cash transfer program. (Poder Ejecutivo Federal, 1997)
2.2 Conditional Cash Transfer Model
Conditional cash transfer programs aim to reduce poverty and break the intergenerational cycle of poverty by making welfare transfers conditional upon the beneficiaries’ actions. The institution transfers the money on the condition that those households make pre-specified investments in the human capital of their children. The established criteria which may include: school enrollment, health check-ups, vaccinations, etc. The core theoretical case in support of cash transfers revolves around a sequence of intended positive impacts and human capital investment. Additionally, these human capital investments can be translated into long-term effects such as capital accumulation, modification of livelihood strategies, thus reducing household vulnerability and increasing economic resilience. (Bastagli et al., 2016) Most CCT programs transfer the money to the mother of the household or to the student in some circumstances. (Fiszbein & Schady, 2009)
The impact of cash transfer programs worldwide has explored the intended and unintended effects, such as changes in bargaining power and gender relations, social interactions, psychosocial well-being, and fertility rates. There is also a potential spillover effect on the potential role of cash transfers in affecting community-level dynamics, productivity, local economic growth, changes in local labor markets and existing social networks – as well as macroeconomic outcomes. (Bastagli et al., 2016)
Literature is mostly focused on household level outcomes which can be classified in three categories: first-level outcomes, second-level outcomes, and third-level outcomes. The first level outcomes refer to the effects that are triggered as a direct consequence of receiving the transfer: changes in consumption, savings, and investments. The second-level outcomes concern the behavioral changes that derive of the first-level outcomes: education and health effects, primarily. Lastly, the third-order outcomes, which refer to the long-term or medium-term impacts, one can find discussions about schooling performance, livelihood strategy diversification, and psychosocial wellbeing. (UNICEF, 2007)
According to the large body of research that has evaluated the different CCT programs that have spread around developing countries since the 1990’s, by and large, CCT programs have increased consumption levels among the poor. As a consequence, there have been substantial reductions in poverty among beneficiaries – especially when the transfer is generous, well-targeted, and structured in a way that encourages recipients to modify their behavior, as is the case of PROGRESA. Virtually all CCTs have seen increases in female bargaining power, school enrollment, and health provider visits. (Fiszbein & Schady, 2009)
While these are important accomplishments, literature also shows that the evidence of CCT impacts on third-level outcomes in health and education – achievement and cognitive development rather than enrollment, child height-for-age rather than growth monitoring – is mixed. An important challenge for future research and policy design is to better understand how and which complementary actions are necessarily to ensure greater impact on final outcomes. Nevertheless, there is a consensus in that these complementary actions fall into two categories: improving quality of the supply of health and education services and ameliorating the environments in which children develop, be that their communities or the households themselves. (Fiszbein & Schady, 2009)
The purpose of this paper is to assess second-level outcomes of the beneficiaries of the CCT program PROGRESA in Mexico. I was primarily concerned with the schooling and labor outcomes of the early beneficiaries of the program who, at the time of the second survey used, were of an age when school attendance and early labor outcomes could potentially be studied and assessed. Additionally, the children in the subsample used, were of an infant age or unborn at the beginning of the program. Hence, they would have benefited the longest from the program.
Governments of developing countries and NGO’s are aware of the vicious circle of poverty and allocate resources to early childhood interventions with the aim to provide disadvantaged children in developing countries a better start. (Rosero & Oosterbeek, 2011) Additionally, literature also suggests that family size plays an important part in the success and outcomes of the cash transfer programs. There is consistency in the literature findings of
a negative correlation between desired outcomes and family size. (Haan, Monique De; Plug, Erik; Rosero, 2013)
Taking into account these shortcomings in previous policy design, the Mexican government created a policy that had a greater chance of impacting young children human capital investment and household welfare, thus becoming a role model for the first generation of conditional cash transfer programs. Since its establishment in 1997, this model has been replicated in 52 countries around the world in very different contexts: in Latin America, Asia and Africa. (Jere R. Behrman, 2010)
Several factors allowed this model to be replicated in different countries. However, it should be stressed that conditional cash transfer programs have emphasized evaluation, which has enabled their impact to be demonstrated. Additionally, assessment allows for the program design to be modified and adapted according to the circumstances of the beneficiaries and context of each country. (Barrientos, & Sabate-Wheeler, 2017)
Even the best-designed CCT program cannot meet all the needs of a social protection system. It is only one part of a system which includes: workfare, employment and social pension programs. The poorest households are more prone to be affected by economic shocks, as developing countries navigate a period of crisis, it is vital to design and implement social protection systems that help those vulnerable household weather shocks, while maximizing the efforts of developing countries to invest in their children’s human capital. CCTs are not the only programs fitting for this purpose, but they surely can be a compelling part of the solution. (Fiszbein & Schady, 2009)
Additionally, there are various reasons why CCTs may have had only a modest effect on third-level outcomes in education and health. One possibility is that some important constraints at the household level are not addressed by CCTs as currently designed. These constraints could include poor parenting practices, inadequate information, or other inputs into the production of education and health. Low quality of education and health services in developing countries make that mitigate the effect that increased use of these services could have in the poorest sectors. (Barrientos et al., 2017)
An important concern of researchers and policy-makers when CCTs were first launched was that they would modify labor market participation of adults – either because they would choose to forfeit working hours for leisure at higher income levels, or because they would supply less work in order to appear to be “poor enough” to be eligible for the cash transfers. In some contexts, at most, CCTs appear to have prompted a modest disincentive effect on adult work. Nevertheless, literature focused on Cambodia, Ecuador, and Mexico shows that adults in household that received the transfers did not reduce their work effort. (Fiszbein & Schady, 2009)
The role and design of CCT programs is evolving. Early successes of the basic model are prompting countries to address second and third level outcomes. Greater emphasis is being put on the supply side of education and health services to complement the cash transfer efforts. Conditionality and the range of conditions is also something that is continuously revised by evaluators: should performance be rewarded instead of service utilization? Ensuring long-term effects is a challenge for policy-makers and the literature is now
exploring the adjustments that the basic design of conditional cash transfers need to implement to actually break the intergenerational cycle of poverty.(Barrientos et al., 2017)
2.3 Description of the program
In 1997, the government of Mexico implemented PROGRESA (Programa de Educaciόn, Salud, y Alimentaciόn), an integrated approach to poverty alleviation through the development of human capital. PROGRESA was one part of a larger poverty alleviation strategy, and its role was to lay the groundwork for a healthy, well-educated population who could successfully contribute to Mexico’s economic development and break the inter-generational cycle of poverty. (Gantner L., 2007) Each beneficiary family would receive a scholarship for basic education1 for their children, health services for all household members,
and nutritional supplements for children up to two years old and their mothers. In addition, all components were to be received and managed by the mothers. (Poder Ejecutivo Federal, 1997).
The program aimed to increase families’ investment in human capital through generous cash transfers. To achieve this goal, the cash transfer was conditioned on children’s enrollment and regular school attendance as well as health clinic consultations. The transfers correspond, on average to a 22 percent increase in the income levels of the beneficiary families. In Mexico, as well as in other contexts, it is observed that children who suffer from malnutrition are likely to drop out of school or repeat years of school, implying that attempts
1 The International Standard Classification of Education (ISCED) defines basic education as primary and
to ensure that children go to school will be more effective if combined with nutrition and health programs, thereby helping to break the vicious cycle of intergenerational poverty transmission. (Levy & Rodríguez, 2005)
According to Parker and Teurel (2005), perhaps the first innovation of PROGRESA is the fact that it combines three different components, that is, education, health, and nutrition. The reason for linking these three components in one program, as stated before, and in the document proposing PROGRESA (Poder Ejecutivo Federal, 1997), was based on the philosophy that the interactions between the components would increase the effectiveness of an integrated program over and above the separate benefits of each component.
The program is subject to a rigorous evaluation effort in rural and urban areas that included an experimental design. On its inception, a subset of communities and individuals eligible to receive Progresa were interviewed to be later designated to control and treatment groups. For the approximately twenty-four thousand households living in these communities, the household interviews were carried out both prior and after program implementation. The International Food Policy Research Institute was commissioned to provide an external evaluation on the impacts of PROGRESA. (Jere R. Behrman, 2010) These impacts included direct effects (e.g., education, health, and nutrition) as well as other potential impact indicators, including child and adult work, consumption patterns, women’s status and transfers. (Parker & Teruel, 2005)
The intervention began in 1997 a few months after the initial survey. Low-income communities were randomly assigned to be enrolled in the conditional cash transfer program.
(Fernald, Gertler, & Neufeld, 2008) The program eligibility was established in two stages in rural areas. First undeserved or marginalized communities were identified based on the proportion of households living in very poor conditions from data from the 1997 census and survey. Households within those communities were identified on the basis of a socioeconomic characteristics and were later randomized into treatment and control groups. Once enrolled, households received benefits for a minimum of 3 years, conditional on meeting the program requirements, after which, they were to be reassessed for eligibility. (de Janvry & Sadoulet, 2006)
As for urban areas, the program was advertised and households would self-identify as potential beneficiaries. They would be surveyed and, according to socioeconomic characteristics, they would be classified as eligible or not. Subsequently they would be randomly assigned to either the control or treatment groups. Randomization was a priority of the program given that it was designed to facilitate policy evaluation. (Fernald, Gertler, & Neufeld, 2008)
Conditional cash transfer programs, according to de Janvry and Sadoulet (2006), have a dual objective: immediate and long term poverty reduction. Conditional cash transfers have become widely used, in particular, to induce beneficiary households to invest in their children’s human capital, thus creating a long-term effect. The approach presumes that the supply side of social services for education and health is in place, and that stimulating demand through income effects is insufficient to induce major changes in human capital investment. Therefore, a conditionality needs to be attached to the cash transfer. (de Janvry & Sadoulet, 2006)
2.4 Randomized Controlled Trials
The policy evaluation of PROGRESA is carried out as a randomized controlled trial (RCT), which is a way of doing impact evaluation in which the population receiving the program or intervention is chosen at random from an eligible population, and a control group is also chosen at random from the same eligible population. The strength of an RCT is that it provides a very powerful response to questions of causality, helping evaluators and program implementers to know that, what is being achieved, is as a result of the intervention and not anything else. (UNICEF, 2014)
In the case of PROGRESA, it was one the policy makers’ priority to create a setting that would allow for a proper RCT since the beginning of the program implementation. Participation in the program was carefully controlled with the experiment in mind, given that RCT’s cannot be undertaken retrospectively. (UNICEF, 2014) The evaluation design was heavily influenced by the works of economists in the field of evaluation in conjunction with the program design. It is of outmost importance to assess the program impact periodically for policy makers in Mexico. (Parker & Teruel, 2005)
Correspondingly, at the inception of the program, the government randomly chose 320 communities for intervention in seven states. Random assignment was generated without weighting by use of randomization commands in Stata version 2.0. Thus, every participant was given equal chance of being included. None of the sites or individuals were told they would be participating in the study, and information regarding timing of rollout was not made public. (Fernald et al., 2008)
Within each community, all households, regardless of poverty status, were interviewed. This was done partly for evaluation purposes and partly due to the fact that most of the cost of interviewing in communities lies in actually getting to the community. Once there, the marginal cost of carrying out extra interviews is relatively low. This is another important feature of the evaluation design, and it implies that within each community, data exists for both control and treatment groups. (Parker & Teruel, 2005)
PROGRESA officials tried to conceal the purpose of the pre-program surveys to not create pressure or expectations in all surveyed households. It was imperative to obtain accurate information to ensure the reliability of the evaluation and identify pre-program differences. (Parker & Teruel, 2005)
2.5 Conditionality
Conditional cash transfer programs were first introduced in Latin America more than two decades ago. Linking cash benefits to families’ investments in human capital, particularly school attendance, has become wildly popular in poverty alleviation programs. Well over 30 countries now have, as part of their social policy, CCT programs. Most of which include substantial schooling conditionalities. (Behrman, Parker, & Todd, 2010)
Some of the most relevant impacts of conditional cash transfers, including their long-term effects on schooling and labor, can only be measured directly after significant amount of time has passed since the program started operating. Notably, there is no significant
information regarding the long-term effect of the most important schooling programs, although efforts at systematically evaluating programs using controlled experiments and nonexperimental statistical methods have expanded. Evaluations are, for the most part, dedicated to assessing the impact of a fairly short exposure to a new program (that is, for a year or two) on outcomes over short periods. (Behrman et al., 2010) The contribution of this analysis is the evaluation of long-term effects of the early beneficiaries of PROGRESA.
Short-term estimates based on exposure to a program of a year or two have been used to extrapolate to long-run program impacts (Behrman, Sengupta & Todd, 2005; Schultz, 2004). However, there are a number of reasons why short-term impacts might not be useful for this purpose. For instance, at least one recent controlled experiment (Banerjee et al., 2007) reports that fairly substantial initial program effects in the first two years largely faded after the program was terminated. Fortunately, the data for the PROGRESA program allows for both long-term and short-term assessments. (Behrman et al., 2010)
There are both public and private perspectives that provide good reasons for conditional cash transfers to be conditional. From the public perspective, governments may perceive that they know what actions or behaviors will benefit the poor better than the beneficiaries do themselves, and that conditioning transfers can modify behavior to better match those beliefs. For instance, governments may place greater weight on the intrinsic value of educating girls than do families, which was, in fact, the case for the Mexican officials.
Conditioning may help the government overcome information asymmetries and increase the accuracy of evaluation methods. Governments may be aware of the benefits associated with immunization or screening for chronic diseases but individuals may be unaware or unconvinced of these benefits. Which is the main reason why clinic consultation was part of the integral program. When other approaches to such informational problems, such as public health campaigns, have failed, conditioning transfers can be seen as a means of modifying behaviors. (De Brauw & Hoddinott, 2011)
Finally, conditionality might have an impact for political economy reasons. Politicians and policy makers are often assessed by performance and growth indicators such as changes in school enrollment or use of health clinics and general well-being. By conditioning transfers on behaviors that improve these indicators, politicians and policy makers can factually demonstrate accomplishments long before the more important evidence of poverty reduction, in the form of increased productivity or better adult health, occurs. Therefore, politicians can perceive that conditioning transfers is a useful tool to help them stay in office and pursue their political agenda in future terms. (de Janvry & Sadoulet, 2006)
From the private perspective, the conditional component of conditional cash transfers can also have potential benefits. Disagreements may exist within households regarding the allocation of resources, specially in communities that suffer from male dominance. Imposing conditionality on cash transfers can strengthen the bargaining position of individuals whose preferences are aligned with the government's preferences, who may otherwise lack bargaining power within the household. Conditioning may overcome stigma effects otherwise associated with welfare payments. The stigma attached to welfare payments may
discourage those with valid claims from taking them up. From the beneficiary's point of view, conditioning can be seen as part of a social contract between themselves and the state and may legitimize the transfer, overcoming the stigma. This feature is particularly important for indigenous groups. (De Brauw & Hoddinott, 2011)
Finally, work in behavioral economics emphasizes that when households have hyperbolic discount functions, they undertake actions that can reduce their own welfare (Laibson, 1997). In such circumstances, households are better off when constraints are imposed that reduce or limit their ability to trade-off future for present consumption. Conditionality can be seen as such a constraint. (De Brauw & Hoddinott, 2011)
2.6 Literature review
Previous literature has highlighted the importance of human capital investment at an early age for long-term outcomes. (Cunha et al., 2006) According to Heckman (2006), much of the effectiveness of early childhood intervention comes from boosting non-cognitive skills and fostering motivation. These skills are usually developed in the household. Thus, family time and early factors matter for adult performance. Despite the growing body of research in early childhood intervention, there is little evidence for long-term outcomes of low income individuals. My research focuses on outcomes of early childhood intervention in Mexico of the PROGRESA female recipients and the observed differences between genders.
According to Theodore Schultz (1961), investments in human capital can be described as direct expenditures on education, health and internal migration to take advantage
of better job opportunities and increase future consumption. The evidence that human capital investments are profitable in the long-run, under the conditions that PROGRESA beneficiaries experience, is thin. My research aims to fill this gap. Schultz (1961), in his paper, stated that there is a strong correlation between education and earnings in adult life. To assess the relationship between education and labor outcomes and early childhood investments – and the gender differences therein – I link the PROGRESA scholarship to the outcomes of young adults sixteen years after the start of the program.
According to Patrinos (2016), the Mincer equation—arguably the most widely used in empirical work—can be used to explain a host of economic, and even non-economic, phenomena. One application involves explaining – and estimating – employment earnings as a function of schooling and labor market experience. The Mincer equation provides estimates of the average monetary returns of one additional year of education. Based on this theory, this article estimates the relationship between the scholarship and schooling and labor outcomes of young adults, with particular interest in girls. The time frame, however, did not allow for a Mincerean model, given that the sample was too young for years of experience and earnings to be significant.
According to Fiszbein and Schady (2009), the contagion of CCT programs in developing countries, many containing gender equality goals, augments the significance of the impact of these programs in gender equality. Studying the program structure of PROGRESA will help illuminate some of the its impact for girls. This article’s focus on how the scholarship at an early childhood impacts schooling and labor outcomes for the beneficiaries. This relationship has important policy implications for the ability of CCTs to
promote gender equality, and long-lasting effects such as breaking the intergenerational inheritance of poverty.
Berhman (2010) does a cost and benefit analysis of being part of the program for 6 years and the changes in the labor market they can expect for young adults. However, he only makes projections of how the scholarship will impact schooling and labor outcomes. In this analysis, I use the survey from 2013 to assess the realized impact of early childhood intervention and scholarship on young adults.
Children who grow up in extreme poverty experience a hindered skill accumulation over the course of their child life, they will therefore increasingly fall short. According to human capital theory, family time and family income – or lack thereof – may lead to this restricted development. Poverty alleviation will lead to less financially restricted families, and thus to improved child outcomes. (Cunha et al., 2006)
Chapter 3
DATA AND RESEARCH DESIGN
My analysis uses household survey data generated from the implementation and evaluation of Mexico’s CCT program. PROGRESA has attributes that help exploit its randomization and is a trademark conditional cash transfer program in Latin America that has been replicated in other contexts. There are numerous issues in Mexico that make education and human capital investment a priority for social development programs. In the grander scheme of things, the Mexican government aimed to solve social problems through poverty alleviation and education. (Gobierno Ejecutivo Federal, 1997)
The program has evolved since 1997 and has been renamed from PROGRESA to OPORTUNIDADES and, since 2014, to PROSPERA. Regardless of the spillover effects that the Mexican government hoped for, the program has been an adequate measure of social development in a poverty context in Mexico. (Gutiérrez et al., 2013) In this section, I will describe the two surveys that I used for my analysis and will detail the reasons for creating a subsample and will thus describe the subsample.
3.1 Data
The ENCASEH (Household survey of socioeconomic characteristics) was carried out in 1997 in preparation for the PROGRESA launch in August 1997, which is when the school
year begins in Mexico. A total of 140,544 households in 3,269 locations were incorporated into the program, all with children eligible children for the program. All households were surveyed for parental and family background to be included in the database that would later be constructed. (Gobierno Federal Ejecutivo, 1997) The ENCASEH was said to be a representative sample of the 53 million individuals living in different degrees of poverty in Mexico. From this database, I selected a subsample of children from 0 to 5 years old that could potentially match the sample from the 2013 follow-up survey.
The 1997 survey had a total of 45,180 observations of underage household members. My subsample, for comparison purposes, is comprised of 15,877 observations.2 To impose stronger sample selection, after deleting the observations of individuals over 5 years old, I matched the 1997 survey to the observations of the 2013 survey. It was important to also delete the observations that had missing variables on the relevant characteristics such as indigenous background, gender, or parental education.
From the reduced sample, I find that the control and treatment groups are quite similar in age, the mean of both is 2.7 years old. In both groups genders are balanced, though in the treatment group 50.4% are male and in the control group 49.5% are. As for the parental characteristics, the indigenous background is quite similar in both groups. Parental education was determined by the highest level one of the parents had. In the control and treatment groups we find that the education levels are quite similar. From these, we can assume that the treatment was, in fact, randomized.
It was imperative for the surveys to be as comprehensive and as accurate as possible given that the program would be closely monitored. The program was partially sponsored by the Inter-American Development Bank and by the Mexican Government. The program represented roughly over 40% of the federal governments annual poverty budget in 1997 and would increase over the following years and governmental administrations. (Parker & Teruel, 2005)
The ENCASEH’s main objective was to answer the following question: “what are the ex-ante characteristics of the eligible population?” The aim was not only to measure poverty and education, but to fundamentally understand the shortcomings that aggravated their economic situation. Questions were based on real-life contexts and everyday life problems. They ranged from health and nutrition, to background and education. Additionally, they addressed gender differences. Gender differences were another focus of the program because the government had the theory that women’s bargaining power decreased with income, specially in poor households in rural Mexico. (Poder Ejecutivo Federal, 1997)
Despite the evolution of the program, the cash transfers associated with health services and education were focalized in households whose socioeconomic conditions and income deter the household members from fully developing their economic potential. Furthermore, in addition to the income-related criteria, the program integrated other conditions for eligibility for the program. Priority for eligibility was given to households whose monthly income per capita was lower than the minimum of the designated well-being, and to the presence in the household of individuals under the age of 22 years of age and
women of reproductive age. This allowed to focus on household with individuals in the phase of human capital accumulation. (Gutiérrez et al., 2013)
The follow-up survey of 2013 focused on individuals between the age of 14 and 27. It’s in this group where the Mexican government expected to identify the socioeconomic mobility that the program seeks to achieve. In 2013, the program reached every municipality in the country, covering 5.8 million households and 6.6 million young adults aged between 14 and 27. The survey, however, focused on a representative sample of young adults whose households were interviewed in the early stages of the program. (Gutiérrez et al., 2013)
The follow-up survey “Follow-up Survey of Youth in Oportunidades Households, 2013” (ESJóvenes 2013) was carried out in the second semester of 2013 in Mexico. The survey focused on a representative sample of respondents. The respondents were 5,203 young adults from 4,532 different households. The survey included information regarding health, nutrition, education, labor and general well-being. The survey had three different questionnaires: the household survey, the young adult survey, and the proxy for young adult survey – in case that either he or she was not present in the household, it was answered by either the mother or the father. (Gutiérrez et al., 2013)
The survey had specific questions on the following subjects: socioeconomic characteristics, schooling characteristics, labor information, labor training, entrepreneurship and small business ownership, health, skills or capabilities learned in the household, credit and savings, asset holdings, sexual and reproductive information, risk conducts, tendency to
crime, use of leisure time, migration, intergenerational mobility, contact information, and anthropometric measures. (Gutiérrez et al., 2013)
From that sample, to ensure validity of the model’s estimates and heterogeneity of the analysis, I drafted two subsamples. One with individuals aged 14 – 18 years old, which would coincide with the children that were either born or raised under the program and could potentially reflect the most accurate estimates for my variables of interest. Additionally, since a portion of that sub-sample was yet to be born in the 1997 survey, I created a subsample of young adults aged between 16 and 18. The first subsample had 1,860 observations, the latter was comprised of 1,296 observation. These observations were on the individual level, not household.
The random assignment of the treatment ensures the assumption that any differences between the treatment and control groups are merely a result of chance. There is however, an endogeneity issue with fertility, it is possible that the cash transfer stimulated fertility, this could imply that an outcome variable was selected when I analyze the families that had not yet had children in 1997, or the families who became larger because of the treatment. This is explored further in Chapter 5. Additionally, there’s an attrition problem with the data. It is unclear in the government report (Gutiérrez et al., 2013) why the sample lost so many observations in the 2013 survey. Almost 90% of the observations were lost in the 2013 sample. However, the balancing table shows that the treatment and control groups share similar background characteristics. See Table 1 below for descriptive statistics of both the control and treatment groups of the sub-sample in the follow-up survey (ESJóvenes 2013).
Table 1 — Descriptive Statistics of sample of young adults aged 14 – 18 (ESJóvenes 2013) Control Treatment Meana SDb Meana SDb CHILDREN CHARACTERISTICS Age 16.53 1.085 16.31 1.198 Female 0.462 0.499 0.495 0.5 PARENTAL CHARACTERISTICS
Highly educated parents 0.209 0.408 0.234 0.424
Indigenous Background 0.080 0.280 0.150 0.360
Rural 0.509 0.501 0.735 0.441
OUTCOMES
Attended secondary school 0.616 0.487 0.93 0.254
Attended high school 0.167 0.374 0.347 0.476
Entered the labor market 0.532 0.5 0.449 0.497
Reported positive earnings 0.192 0.394 0.108 0.311
Monthly earnings in $MXN 3,558.12 2,910.44 3,060.38 2,363.93
NUMBER OF OBSERVATIONS 167 1,693
Source: Author calculations. PROGRESA data from the Encuesta de Seguimiento de Jóvenes 2013.
aMean: For descriptive binary variables, the mean is the ratio of the sample with those characteristics. b SD: standard deviation.
For my analysis, 1,860 individuals from seven different states in Mexico were selected based of the following criteria: age range from 14 to 18, and no missing data in whether or not they were part of the treatment or control groups. Hence, 3,343 observations were drawn from the sample. Nonetheless, some of these observations had missing values for labor outcomes, hence, I deleted them. The final sample is comprised of 1,861 observations. Afterwards, I constructed a binary variable based on whether each individual
was a beneficiary or not. The treatment variable was equal to 1 if they had been PROGRESA recipients, and 0 if they were not beneficiaries, but had been surveyed as part of the control.
The ratio of treated and control units varies because, despite being important, over time the main goal was not to create an empirical setting that was suitable for evaluation, but to reach all sectors of the population and alleviate poverty (Gutiérrez et al., 2013). Moreover, there are more male individuals than female in both the treatment and control groups. In this case, it could be indicative that females left the household either because of marriage or migration. Finally, the mean of the qualitative variables corresponds to the ratio of the sample that has the variable’s characteristics, not to an average.
Many variables included in the dataset and used in this analysis were not directly drawn from the survey responses to the questionnaire but were constructed from questionnaire items, as they cannot be observed directly (e.g. highly-educated parents, reported positive earnings, etc.) Those derived variables were constructed in the following manner: Simple indices were constructed from arithmetical transformation, or recoding of questionnaire items.
For the sensitivity analysis, I had to create an indicator for family size. The data did not allow for number of siblings in the household, which would’ve helped to assess more clearly the impact of the treatment on an additional child. Small families are those that have three children or less, and large families are those that have more than three children and thus, from the fourth child on, would be left out of the program as direct beneficiaries.
Data on the indigenous background (ind_bck) was also selected. This index was constructed from whether or not they could speak an indigenous language, whether or not they identified as indigenous, or whether or not one of the parents was indigenous. I made no distinction between first-generation indigenous and the rest. Individuals with missing responses for all of those questions were assigned missing values for this variable.
For parental education, I use the variable on the highest education level of parents (HISCED) that is constructed from the items on maternal education and paternal education levels. The classification is based on ISCED (International Standard Classification of Education) 1997, with 7 levels: (0) None, (1) ISCED 1 (primary education), (2) ISCED 2 (lower secondary), (3) ISCED Level 3B or 3C (vocational/prevocational upper secondary), (4) ISCED 3A (general upper secondary) and/or ISCED 4 (non-tertiary post-secondary), (5) ISCED 5B (vocational tertiary) and (6) ISCED 5A and/or ISCED 6 (theoretically oriented tertiary and post-graduate). Those that responded Level 3B or 3C and up were classified as “highly educated”. For this analysis, upper secondary education was considered “highly educated”, I took the parent with the highest education level, which in most cases is the father, and created the variable.
The outcomes of interest in my analysis are all binary variables and can be described as follows:
• Attendance to secondary school: whether the individual had enrolled and attended secondary school. In Mexico, this corresponds to grades: 7th, 8th, and 9th. Typically, children enter secondary school at age 12 or 13.
• Attendance to upper secondary school: whether the individual had enrolled and attended high school. In this context, it corresponds to grades: 10th, 11th, and 12th. Generally, children attend upper secondary school at age 15 or 16.
• Entered the labor market: whether or not the individual had previously held a job outside of the household.
• Reported positive earnings: if the individual had, in fact, entered the labor market, I created an index on whether or not he or she had reported positive earnings for any of his or her previous jobs.
Finally, data regarding earnings was constructed on the previous earnings that they had reported items, similarly to parental education level, the value that was used for the analysis was the highest level of earnings that they reported. There were some flaws in this particular question in the survey, some of the reported earnings were very low (around 10 cents of a dollar), which could indicate a mistake in data input. Subsequently, I constructed another binary variable on whether or not they had reported positive earnings. The non-respondents were also assigned missing values for the variable.
3.2 Attrition
The main evaluative strength of randomized controlled trials is that each group is generally balanced in all characteristics, with any imbalance occurring merely by chance. However, during many trials participants are lost to follow-up. Such attrition is problematic if it correlates with the treatment status. Attrition can also happen when participants have
missing data at one or more points. Researchers need to be explicit about losses due to follow-up, especially if rates are high. (Dumville, 2006)
Fieldwork protocols in Progresa dictated revisiting only the original households, proceeding to interview whomever was present, if anyone. As a consequence, attrition has turned out to be a problem in the ENCEL surveys (Encuesta de Evaluación de los Hogares). For instance, by the end of the November 2000 ENCEL survey, approximately 16.01 percent of households and 21.89 percent of individuals originally interviewed in the fall of 1997 were no longer in the database. Most of this attrition is caused by apparent changes of residence or migration (close to 80 percent), and the rest are related to nonresponse and deaths. (Teruel & Rubalcava, 2003).
Literature has highlighted the point that movers may be very different than stayers in the context of longitudinal surveys and stressed the importance of following movers. For the current evaluation, there are two related issues. First, if attrition in the surveys differs between treatment and control groups, as Teruel and Rubalcava (2003) suggested, this attrition may bias the estimated impacts of the program, and thus it would be useful to have information on the attritors to correct any bias. Second, with high percentages of migrant individuals, it is obviously of interest to know the impacts of the program not only on the population remaining in their home communities but also for the population that moves. This is likely to be particularly true for youth, who are the most likely to leave their household and community of origin, and also presumably the group most likely to be impacted by the program benefits. (Parker & Teruel, 2005) In this analysis, we find that the background characteristics are balanced, despite the high attrition rate.
Chapter 4
ESTIMATES
In the following section I will describe the methodology of the analysis. I will also address the issues that arose because of attrition rate and mention heterogeneity and sensitivity tests that evaluated the RF estimates and assessed the validity of the model.
Given the nature of the data and the attributes of experimental design, I used reduced-form estimates to explore the relationship between receiving the PROGRESA scholarship and schooling outcomes, as well as for labor market outcomes of young adults in Mexico. The adoption of an experimental design in the early stage of the implementation of the program, allows the measurement of program impact by comparing outcomes between beneficiary households – the treatment group – with similar households that were not covered by the program – the control group. The randomization allows to interpret the differences amongst groups in a causal fashion and attribute those differences to the cash transfer program.
First, I estimate the effect of the treatment on secondary school attendance, high school attendance, entered the labor market, and reported positive earnings with reduced-form estimates. I estimate the effect with no controls, and controlling for a set of covariates. This set includes age of the child, and dummies for gender, urban/rural indicator, and highest level of education of the parents. The regression I ran is the following:
Subscript i indexes the child for whom the mother receives the transfer. 𝑇" is a binary variable indicating whether or not the child is part of a control or treatment household. 𝑋" is the set control variables included in the regressions. 𝑌" is the educational or labor outcome. The coefficient I am interested in is 𝛼(, since it reflects the relationship between the treatment and the schooling or labor outcome.
See below the reduced form estimates for subsamples taken from the survey (ESjóvenes, 2013) both the subsample of young adults aged 14 to 18, and the subsample aged 16 to 18, which participated in both surveys.
Table 2 — Reduced form regression estimates Full sub-sample
Estimates of CCT
treatment Secondary school Upper secondary school Labor market Positive earnings Age 14-18 0.314*** 0.225*** 0.180*** 0.225*** -0.083*** -0.040 -0.083** -0.048
(0.038) (0.031) (0.031) (0.031) (0.040) (0.038) (0.031) (0.0301)
R-squared 0.283 0.283 0.468 0.424 0.498 0.130 0.062 0.062
Observations 1,860
16-18 year olds sub-sample
Age 16-18 0.294*** 0.298*** 0.281*** 0.278*** -0.073 -0.053 -0.083** -0.059 (0.042) (0.042) (0.038) (0.038) (0.045) (0.042) (0.037) (0.036)
R-squared 0.093 0.107 0.029 0.066 0.002 0.107 0.005 0.056
Observations 1,296
Controls X X X X
Source: Author calculations. Progresa Data from the Encuesta de Seguimiento de Jóvenes 2013
a Age, gender, paternal education, rural and urban indicators as controls
Standard errors in parentheses below the estimated coefficients *** Significant at the 1 percent level.
** Significant at the 5 percent level. * Significant at the 10 percent level.
Without controls, the reduced form coefficient of the treatment, for the full sub-sample, displays a positive, large and significant impact of being a beneficiary of PROGRESA on secondary and upper secondary school attendance. All estimates show an increase of around 20% of attendance on both levels. When adding controls, the value of the coefficient on treatment decreases for secondary school attendance, and it increases for high
school attendance. However, adding the controls does not change much the estimates much. In a true RCT fashion, the estimates remain similar.
As for the labor outcome outcomes, for the sub-sample 14-18, the reduced form coefficients without controls for entering the labor market displays a negative and significant impact of approximately 8% of the treatment on entering the labor market. Which is in fact, what one would expect of the treatment, given that the goal was to keep children in school and out of the labor market. It is quite high given that most jobs in low income communities do not require an upper secondary school education. When adding controls, the absolute value of the coefficient on treatment decreases, so the negative effect of treatment on the child working estimates is slightly lower.
When estimating the impact of the treatment on reporting positive earnings, without controls for the full sub-sample, the reduced form coefficients display a decrease of around 4% of reporting income and the estimates are statistically significant. Subsequently, when adding controls, the absolute value of the coefficient on treatment decreases, so the negative effect of treatment on the child reporting positive earnings is lower. The difference between the estimates with and without controls is small.
On the other hand, when looking at the sub-sample of individuals aged 16 – 18 years old, without controls, estimates of the treatment display a large, positive and significant increase in secondary and upper secondary school attendance, in comparison to the full sub-sample, the coefficient is lower without controls. The estimates are almost as large as for the full sample around 29%. This is an impressive increase given that this is group that was born
in the baseline survey. When comparing to the previous sub-sample, both the estimates with and without controls are larger for 16 – 18 sub-sample, than for the 14 – 18 sub-sample. This could also be due to the fact that the 14 year olds are not old enough to attend upper secondary school.
As for the labor outcome coefficients, for the 16 – 18 sub-sample, estimates without controls for entering the labor market display a small (approximately 4%), negative and statistically insignificant impact of the treatment on entering the labor market. This is a nice feature, because this sub-sample is the most heterogeneous of the analysis and one would expect the treatment to not have a direct effect of entering the labor market for this age cohort. When adding controls, the absolute value of the coefficient on treatment decreases, so the negative effect of treatment on the child working estimates is lower, but it again insignificant. When compared to the previous sub-sample one should underline the fact that for the previous one, the impact was significant without controls, and with this sub-sample it is insignificant, but equally small.
The impact of the treatment on reporting positive earnings, for the sub-sample of young adults aged 16 – 18, without controls, the reduced form coefficients are negative and significant, approximately around 5%. Subsequently, when adding controls, the absolute value of the coefficient on treatment increases, so the negative effect of treatment on the child reporting positive is higerr, but, again, it loses its significance and is quite similar (also around 5%). Compared to the previous sub-sample, the estimates are quite similar.
Based on these findings, the results of implementing the cash transfer program seem promising for the rest of their adult life. The main motivation to carry out the ESJóvenes 2013 survey was to assess the effects of the treatment and the human capital investment that was presumably carried out by the households, to break the poverty cycle and see whether or not this age cohort of young adults had achieved better labor outcomes. This outcome might still be quite premature given that the children that were born into the program have yet to show their labor potential and it is expected of them, at this stage, to still be students. Labor outcomes are not only contingent to schooling, but also to years of experience, amongst many other factors that these individuals have yet to develop. (Gutiérrez et al., 2013)
In the evaluation document, commissioned and published by the Mexican government of the 2013 survey, they expressly declare: “The objective of the program is to break the intergenerational cycle of poverty and it is precisely in the group of young people between 14 and 24 years of age that social mobility would be expected to be achieved. This is the group that the PROGRESA program wishes to monitor in order to periodically assess the effects that the improvement of their nutrition conditions, health consultations and human capital investments in their childhood will have in their adult life.”(Gutiérrez et al., 2013)
Chapter 5
5. SENSITIVITY AND HETEROGENEITY CHECKS, AND DISCUSSION
5.1 Sensitivity regressions
One of the main concerns of policy makers in Mexico, when the PROGRESA program was introduced was the effect that this would have on fertility. By giving a cash transfer per child in school in the household, it could be perceived by the heads of the household as if having an additional child could generate income. The government and policy makers were uneasy and presumed that this income effect could cause a distortion in household decisions and generate an increase in population suffering from poverty, which would have an adverse effect on breaking the intergenerational poverty cycle. (Gutiérrez et al., 2013)
Much of Latin America has experienced a steep decline in fertility over the last three decades. In Mexico, specifically, fertility rate went from 5.1 children per women in 1970, to 2.5 children per women in 2005. Despite the progress at the national level, disparities persist in Mexico, with poor, rural, and indigenous women in Mexico having lower contraceptive use rates and higher fertility rates than their urban, more educated peers. Additionally, early pregnancy is associated with adverse health outcomes, increased total fertility, and increased poverty. Reducing unintended pregnancies was a key strategy to decrease poverty and improve women’s health. (Darney et al., 2013)
For policy makers, it was important to examine whether the program had any direct effect on pregnancy experience and contraceptive use among young rural women, which was the segment of the population must vulnerable to suffer from this distortion. (Darney et al., 2013). In order to offset this potential effect, part of the program’s design was to give contraceptives in the health consultations and to offer permanent sterilization for women after the third child was born. One of the program’s specification was that households could enroll up to the third child to the program, consequently, child four and five and so forth would not be subject to being direct beneficiaries of the program. (Gobierno Ejecutivo Federal, 1997)
In order to address these concerns in my own analysis, I did sensitivity regressions of the impact of the treatment on family size. The data, unfortunately, did not allow me to see the age of the siblings, which would’ve helped me assess the situation more accurately with a cutoff of age and birthdate to see which were born before and after the inclusion into the program. It is likely that the increased fertility leads to underestimation of the treatment given that an extra child would lead the income of the scholarship to be allocated differently between household members. Less money would go to the child in school, because after the third child, the other siblings are not eligible for the cash transfer.
Without controls, the coefficient of the treatment displays a positive and statistically significant impact of being a beneficiary of PROGRESA on family size, around 9% increase without background variable controls and approximately 4% when adding the controls. When beneficiaries of PROGRESA. In other words, families are more likely to be larger than when being part of the program than the control group. The estimates being lower when adding
controls is to be expected given that observational data displays that fertility rates are higher amongst lower educated individuals in rural areas. (Darney et al., 2013)
Attrition was another issue that I faced when doing the analysis. The internal validity of a randomized trial’s results partly depends on the between group balance in characteristics of those who remain in the trial. In addition, important imbalances that are not readily apparent in the analyzed groups may become apparent when we examine the between group characteristics of those lost to follow-up. (Dumville, 2006) In this case, from Table 1, we can observe that the baseline characteristics of both groups seem balanced, and the only observation that can be made as to imbalances is the fact that there is a lower female rate in both. This was already addressed in the data section of the analysis.
As previously mentioned in this study, protocols dictated to only revisit households that pertained to the program, both for control and treatment groups. (Gutiérrez et al., 2013) Without controls, the reduced form coefficient of the treatment displays a negative and statistically significant impact of being a beneficiary of PROGRESA on attrition. Being part of the program decreases the attrition rate by approximately 13% with and without background characteristic controls.
These results could have several interpretations, being a beneficiary has a negative impact on attrition, which could mean that by being a beneficiary the individual is more likely to be part of the follow up survey and not have left the household. It could potentially mean that the individual has a higher sense of commitment to the program and will thus answer the
survey, it could mean that it decreases migration, or that the individual is still studying and has not left the household. Which in itself, is a positive outcome.
Attrition bias can lead to overestimation or underestimation of the impact of the treatment. In this case, the high attrition rate is likely to underestimate the impact of the program on labor outcomes and overestimate the education outcomes. It is likely that young adults that did not stay in the household migrated and have better economic possibilities or less education years than those that stayed in the household. An easy way to determine whether there is attrition bias is to draft a balancing table for the individuals lost to the follow-up. However, since information on the baseline characteristics of those lost to follow-up and those for whom data are analyzed is rarely reported, it is almost impossible to identify the effect of attrition on the study sample as a whole and therefore the result of the randomized controlled trial needs further validity checks. (Dumville, 2006) In this case, we can observe that both the 1997 survey and the 2013 survey are balanced and can proceed with the analysis.
Table 3 — Sensitivity Regressions Estimates with Reception of Progresa
Cash Transfer Family size Attrition
Treatment 0.092** 0.043** -0.135*** -0.126**
(0.032) (0.025) (0.034) (0.038)
R-squared 0.004 0.078 0.007 0.066
Observations 1,860
Controls X X
Source: Author calculations. Progresa Data from the Encuesta de Seguimiento de Jóvenes 2013
a Age, gender, paternal education, rural and urban indicators as controls
Standard errors in parentheses below the estimated coefficients *** Significant at the 1 percent level.
** Significant at the 5 percent level. * Significant at the 10 percent level. 5.2 Heterogeneity regressions
To test the validity of the model, the analysis required heterogeneity tests to be carried out. In order to observe the treatment heterogeneity on subpopulations and the policy impact among different types of individuals to see whether or not the estimates of the regressions in the entire population remain the same, or similar for those subgroups. It is important to extract the information from the controls and to consider the conditional average policy effects for individuals with different observed characteristics.
The objective of the heterogeneity test that was carried out was to study whether the policy treatment was beneficial for subpopulations of the samples that are defined by the covariate values of family size, gender, urban and rural indicators, and parental education.
The aim was to assess the impact on those subpopulations and to test the heterogeneous impact across said subpopulations. Those margins were chosen because I wanted to see if the estimates changed with family size, per gender, if it was in an urban or rural setting or with the parental education. I wanted to test whether there was inconsistency of the effect of the treatment for these different subpopulations. I wanted to see whether the effect was similar enough under these margins to have confidence in the fact that the variation of the effect was a product of chance and not of selection or attrition bias.
Heterogeneity regressions were conducted and the impact of the treatment can be observed in Table 4. The regressions that were carried out were for the four outcomes of interest: secondary school attendance for small and large families (more than three children in the household); high school attendance for small and large families; entered labor market for small and large families; and reported positive income for small and large families. This was carried out for all groups mentioned above: female and male, urban and rural, and high and low parental education.
