Essays in Optimizing Social Policy for Different Populations: Education, Targeting, and Impact Evaluation
MetadataShow full item record
CitationNadel, Sara B. 2016. Essays in Optimizing Social Policy for Different Populations: Education, Targeting, and Impact Evaluation. Doctoral dissertation, Harvard University, Graduate School of Arts & Sciences.
AbstractIn the first chapter of this dissertation, I look at the relationship between preference sets among students in similar majors, compared with different majors, in Peru. I find that students within majors share preference sets that differ from students in other majors. I further find that students from households without a formal labor market participant have made decisions that are more consistent with predicted professional opportunities compared with students with a formal labor market participant. These differences are systematic and not related to the general industrialization level of the city where the student lives. This research suggests that the difference between students and workers from households with formal labor-market familiarity and those from households without formal labor-market familiarity are not accidental or due to lack of familiarity.
In the second chapter, I evaluate whether proxy-means testing as a method of targeting for Mexico's Conditional Cash Transfer program caused spending distortions among (potential) recipients. The income and wealth effect of participating in Progresa complicate a simple comparison of members of the control and treatment group in the acquisition of assets. To resolve this, I look at reduced asset acquisition just above the cutoff point. Because an imperfect implementation of the eligibility evaluation may have reduced treatment villagers’ perceived benefit of distorting, I also look for evidence of increased spending in non-assets and of increasing the number of eligible-aged children in the home to increase the size of the transfer. I do not find evidence of lack of investment in assets along the eligibility cutoff, but I do find evidence of increased spending as a percentage of income on items not included in the PMT, as well as evidence of increases in eligible-aged children among the poorest families in treatment villages.
In the final chapter, which is joint with Lant Pritchett, we propose that many development programs, projects and policies are characterized by a high dimensional design space with a rugged fitness function over that space. In nearly any project/program/policy there are many design elements, and each design element has a number of possible choices, and the combination produces a high dimensionality design space. If different program designs produce large changes to outcomes/impact, this implies that the "fitness function" or "response surface," the mapping from program design to outcomes/impact, is rugged. We motivate this investigation using as an example a skill-set signaling program for new entrants to the labor market in Peru. We present a simulation model which compares two alternative learning strategies: "crawling the design space" (CDS) and a standard randomized control trial (RCT) approach. In this artificial world, we demonstrate that with even modest dimensionality of the design space and even modest degrees of ruggedness, the CDS learning strategy substantially outperforms the RCT learning strategy. Moreover, we show that the greater the ruggedness of the fitness function, the higher the variance of the RCT results relative to CDS and hence the lower the reliability of RCT results even with "external validity" across contexts. We suggest that RCT results to date are consistent with a world in which social programs exist in a high dimensional design space with rugged fitness functions and hence in which the standard RCT approach has limited direct practical application.
Citable link to this pagehttp://nrs.harvard.edu/urn-3:HUL.InstRepos:33493361
- FAS Theses and Dissertations