2004年-世界发展银行全球_An_Evaluation_of_the_Performance____________of_Regression_Discontinuity_Design_on_PROGRESA_50页_367kb

Summary of "An Evaluation of the Performance of Regression Discontinuity Design on PROGRESA"

Core Content

This paper evaluates the performance of the Regression Discontinuity Design (RDD) as a quasi-experimental estimator for assessing the impact of the PROGRESA poverty alleviation program in rural Mexico. The study compares RDD estimates with experimental impact estimates to determine whether RDD can serve as a reliable alternative to randomized experiments, which are often politically and ethically challenging to implement in developing countries.

Main Viewpoints

• Reliability of RDD: The paper finds that the RDD estimator performs remarkably well, with estimates agreeing with experimental results in 10 out of 12 possible cases.
• Program Overview: PROGRESA, now known as Oportunidades, is a large-scale poverty alleviation program in Mexico that provides cash transfers to poor households conditional on children's school attendance.
• Eligibility Determination: Eligibility is determined using a discriminant score, which is based on household income and other characteristics. The selection process involved a "densification" phase to include more households, especially the elderly poor.
• RDD Application: The RDD method is applied using the discontinuity in eligibility based on discriminant scores. The analysis uses panel data from 24,000 households across 506 rural localities in seven Mexican states.
• Estimation Method: The RDD estimates are derived using one-sided kernel regressions and local linear regressions. The paper also discusses the theoretical underpinnings of RDD, including the assumption that individuals just below and above the threshold are comparable in their characteristics.
• Limitations of RDD: The RDD method fails to detect significant program impact in the first year of the program (round 3) for both boys and girls, which may be due to the program not being fully implemented or other administrative issues.

Key Information

• Data Source: The study uses a unique dataset collected for evaluating PROGRESA in rural Mexico.
• Sample Size: The sample includes 24,000 households across 506 localities, with 320 treatment and 186 control localities.
• Outcome Indicators: The paper focuses on two key indicators: child school attendance and child work.
• RDD vs. Experimental Estimates: The RDD estimates are compared with experimental estimates (cross-sectional difference in means) to assess their validity.
• Eligibility Thresholds: Each region has its own threshold score for eligibility, which is determined through discriminant analysis.
• Exclusions: Households that were later reclassified as eligible and received benefits are excluded to avoid bias in the comparison group.
• Time Period: The analysis is limited to survey rounds 1, 3, and 5 (October 1997, October 1998, and November 1999), as these are the most relevant for assessing program impact.

Structure of the Paper

1. Introduction: Discusses the importance of program evaluation in developing countries and the challenges of using randomized experiments.
2. Background on PROGRESA: Describes the program's objectives, eligibility criteria, and implementation.
3. Application of RDD to PROGRESA: Explains how RDD is applied in the context of PROGRESA, including the use of discriminant scores and the distinction between sharp and fuzzy designs.
4. Empirical Strategy and Findings: Outlines the methodological approach used to evaluate RDD performance and presents the results for school attendance and child work.

Conclusion

The paper concludes that the RDD method is a viable and reliable alternative to randomized experiments for evaluating social programs, especially in contexts where experimental methods are not feasible. The findings are particularly relevant for countries implementing similar poverty alleviation programs.