Education Choices in Mexico: Notes on PROGRESA Evaluation

Education Choices in Mexico: PROGRESA Evaluation

Introduction

• This paper analyzes data from the PROGRESA randomized social experiment in Mexico to evaluate its impact on school participation using an economic model.
• It emphasizes the usefulness of experimental data in estimating structural economic models and the importance of such models in interpreting experimental results.
• The study also incorporates general equilibrium effects into simulations, enabled by the experiment's availability.

Main Findings

1. Grant Impact: The PROGRESA grant has a stronger impact on school enrollment than an equivalent reduction in child wages.
2. Enrollment Effect: The program positively affects children's enrollment, especially after primary school, which is well-replicated by the structural model.
3. Child Wages: The program has noticeable effects on child wages, but this only marginally reduces the program's effectiveness.
4. Revenue-Neutral Change: A revenue-neutral change that increases grants for secondary school children while eliminating them for primary school children would significantly increase enrollment of the former, with minor effects on the latter.

Program Overview

• PROGRESA (now Oportunidades): Aims to improve human capital accumulation in poor communities through conditional cash transfers in nutrition, health, and education.
• The education component provides grants to mothers for keeping their children in school, starting in third grade and increasing until ninth grade, conditional on enrollment and attendance.

Evaluation Design

• 506 communities were identified, and a longitudinal dataset was collected starting in 1997-1998.
• 186 communities were randomized out of the program for about 2 years to serve as a control group.
• The survey covered all households, identifying potential beneficiaries in control villages.

Importance of School Attendance

• Deficits in human capital accumulation are a major reason for the modest growth performance of Latin American economies.
• The evaluation of PROGRESA, based on a randomized experiment, was highly successful.

Limitations of Experiment-Based Estimates

• Estimates from experiments can only answer limited questions about the specific program implemented.
• Policy decisions require answers to more refined questions, such as extrapolating to different groups or altering program parameters.

Goals of the Paper

• Analyze the impact of monetary incentives on education choices in rural Mexico.
• Discuss effective design of interventions aimed at increasing school enrollment of poor children.
• Illustrate the benefits of combining randomized experiments with structural models.
• Estimate a structural model of education choices using data from the PROGRESA randomized experiment to simulate changes to program parameters.

Comparison with Previous Studies

• Todd and Wolpin (2006) identify the impact of the program from variation in child wages across villages where the program is not implemented.
• This study uses both treatment and control villages, exploiting the grant-induced variation from the randomized experiment.
• It recognizes that the marginal utility of the grant differs from other income sources due to preferences, within-household decision making, and perception of the grant.

Income Pooling

• Ex-ante evaluation using non-experimental data requires assuming that conditional on the child's activity (education or work), child income and other household income have the same effect on utility.
• The PROGRESA grant is given to the mother, while child labor income may be controlled by the father or the child.
• Results indicate that differences between income sources are crucial.

Feasibility of Ex-Ante Evaluation

• Ex-ante evaluation may be feasible with suitable observational data, such as data from a population where some children receive scholarships of varying amounts.
• Experimental information is important because the variation induced is guaranteed to be exogenous.

General Equilibrium (GE) Effects

• The paper also explicitly considers and estimates the GE effects of the program on the wages of young children.
• PROGRESA was randomized across localities, allowing for the estimation of GE effects.
• The program's effectiveness in increasing school participation could reduce child labor supply, increasing child wages and attenuating the program's impact.
• The study establishes that the program led to an increase in child wages in treatment municipalities by decreasing the labor supply of children.

Estimation Approach

• The model includes the presence of habits in the utility from schooling, creating an initial condition problem.
• The increasing availability of schooling is used as an instrument to control for the initial condition problem and disentangle state dependence from unobserved heterogeneity.
• Unlike TW's model, this paper does not solve a family decision problem including trade-offs between children and fertility, as the program did not affect fertility.

Program Specifics

• PROGRESA provides incentives and conditions for participant households to comply with to keep receiving benefits.
• It targets alleviating poverty in the short run while fostering human capital accumulation to reduce it in the long run.
• The nutritional subsidy is paid to mothers who register children for growth and development check-ups and vaccinate them, as well as attend courses on hygiene, nutrition, and contraception.
• Education grants are paid to mothers if their school-age children attend school regularly.
• Programs similar to PROGRESA have been implemented in other countries like Colombia, Honduras, Nicaragua, Argentina, Brazil, and Turkey.

Program Targeting

• PROGRESA is first targeted at the locality level and then within each community by proxy means testing.
• Eligibility depends on a single indicator: the first principal component of variables like income, house type, presence of running water, etc.

Education Grants

• The size of the grant increases with the grade and is slightly higher for girls than for boys in secondary education.
• Beneficiaries also receive a small annual grant for school supplies.
• Children must attend at least 85% of classes to keep the grant.
• A child is still entitled to the grant for the same grade upon not passing a grade once, but loses eligibility if they fail again.

Program Expansion

• The program was phased in slowly but became very large, expanding to more than 50,000 localities by the end of 1999.
• It included about 2.6 million households, or 40% of all rural families and one-ninth of all households in Mexico.

Impact on Beneficiaries

• The nutritional component alone corresponded to 8% of the beneficiaries' income.
• Education grants can reach up to 52% of the average beneficiary's income.

Evaluation Sample

• 506 localities were chosen randomly and included in the evaluation sample.
• Surveys were conducted, and households were classified as poor or non-poor based on their entitlement to the program.
• In 320 localities, the program started immediately, while in the remaining 186, it started almost 2 years later.
• Extensive data was collected on consumption, income, transfers, and a variety of other variables. Information was also collected on age, gender, education, labor supply, earnings, school enrollment, and health status for each household member, including children.

Locality Questionnaire

• Provided prices, wages, and information on institutions in the village, along with distances to the closest primary and secondary schools.

Data Classification

• Households were defined as either eligible or non-eligible, with some non-eligible households reclassified as eligible before the program started, though some experienced delays in receiving the program.
• The structural model considers as beneficiary a household that actually receives the program.

Impact Measurement

• PROGRESA was assigned randomly between treatment and control villages, making it straightforward to estimate the impact on school enrollment.
• Randomization implies statistical identity between treatment and control samples, allowing estimates of program impacts to be obtained by comparing means.

Balance Checks

• The availability of pre-program data allows for checking whether the evaluation sample is balanced between treatment and control groups.
• Differences in school enrollment among non-eligible households were controlled for when estimating impacts.

Focus of the Study

• The focus is on aspects of the data that are pertinent to the model and sample used to estimate it.
• By describing the impacts of the program in the sample used to estimate the structural model, a benchmark is set against which it will be fitted.
• Only results for boys are reported, as the structural model will be estimated on boys.

Estimated Impact

• The impact of the PROGRESA program on enrollment has a marked inverted U-shape.
• The program impact is small at age 10, increases considerably past age 10 to peak at age 14, and then declines for higher ages.
• The average impact for boys aged 10-16 years is about 5 percentage points, while for boys aged 12-15 years it is as high as 6.6%.

Impact on Non-Eligible Children

• Estimates indicate a large impact on non-eligible boys, possibly due to pre-existing differences between treatment and control towns.

Difference in Difference Estimates

• Using 1997 data to obtain difference in difference estimates, the pattern of impacts among eligible children remains largely unaffected, while impacts on non-eligible children become insignificant.
• This justifies the interpretation of the evidence as being caused by pre-existing un-observable differences for non-eligible children and justifies the use of a "non-eligible control" dummy in the empirical specification.

Model Framework

• A simple dynamic school participation model is used, where each child (or his/her parents) decides whether to attend school or to work based on economic incentives.
• Parents are assumed to act in the best interest of the child; interactions between children are not considered.
• Children have the possibility of going to school up to age 17; all formal schooling ends by that time.
• If children decide to work, they receive a village/education/age specific wage; if they go to school, they incur a (utility) cost and, with a certain probability, progress a grade.

Dynamic Considerations

• The model is dynamic because one cannot attend regular school past age 17, and going to school now provides the option of completing some grades in the future.
• State dependence is allowed for; the number of years of schooling affects the utility of attending in this period.
• The initial condition problem is addressed, and related identification issues are discussed at length below.

Anticipation Effects

• The model addresses anticipation effects and the assumptions required for their identification.
• There is no evidence of anticipation effects in the data.

Instantaneous Utilities

• The model assumes linear utility. In each period, going to school involves instantaneous pecuniary and non-pecuniary costs, as well as losing the opportunity of working for a wage.

Uncertainty in the Model

• There are i.i.d. shocks to schooling costs, modeled by (logistic) random term.
• Pupils may not be successful in completing the grade.

Return to Education

• After age 17, individuals work and earn wages depending on their level of education. A terminal value function depends on the highest grade passed.

Value Functions

• Schooling choices involve comparing the costs of schooling now to its future and current benefits.

Wages and GE responses

• An increase in wages will reduce school participation. These wages may be affected by the program because the latter reduces the labor supply of children.

Section 5. Estimation Identifying the Effect of the Grant

• Ideal experiment randomized potential amounts offered across villages or within villages.
• The amount of the grant varies by the grade of the child. Given the demographic variables included in the model and given treatment for initial conditions, this variation can be taken as exogenous. The way that the grant amount changes with grade varies in a non-linear way, which also helps identify the effect.

Key Identifying Factors

• Comparing across treatment and control villages
• Comparing across eligible and ineligible households
• Comparing across different ages within and between grades.

Section 6. Summary of Results of Model Estimation and Alternative Models for Control Samples

• Different Model Specifications

  • The first constitutes the basic model.
  • The second, controlling for the pre-program difference in enrolment rates among non-eligible individuals.
  • Estimates from the control sample.

• The discount factor, estimated consistently to be between 0.95 and 0.98.

Section 7. Simulations Conducted Using the Various Model Specifications

Validating the Models

• The impact of the grant and the relationship that key variables have on policy.

• The main focus for policy makers, the wage and the grant coefficients.

• An interesting conclusion of the study. That experimental data information may not be available when performing standard observational data.