Research

“Information, Role Models and Perceived Returns to Education: Experimental Evidence from Madagascar” (Job Market Paper) [PDF]
Abstract: This paper shows that increasing perceived returns to education strengthens incentives for schooling when agents underestimate the actual returns. I conducted a field experiment in Madagascar to study alternative ways to provide additional information about the returns to education: simply providing statistics versus using a role model—an actual person sharing his/her success story. Some argue that role models may be more effective than providing statistics to a largely illiterate population. However, this proposition depends on how households update their beliefs based on the information the role model brings. Motivated by a model of belief formation, I randomly assigned schools to the role model intervention, the statistics intervention, or a combination of both. I find that providing statistics reduced the large gap between perceived returns and the statistics provided. As a result, it improved average test scores by 0.2 standard deviations. For those whose initial perceived returns were below the statistics, test scores improved by 0.37 standard deviations. Student attendance in statistics schools is also 3.5 percentage points higher than attendance in schools without statistics. Consistent with the theory, seeing a role model of poor background has a larger impact on poor children's test scores than seeing someone of rich background. Combining a role model with statistics leads to smaller treatment effects than statistics alone, also consistent with the theory. The key implication of my results is that households lack information, but are able to process new information and change their decisions in a sophisticated manner.

Research in Progress

“Improving Management in the Educational System in Madagascar: An Experimental Approach” (with G. Lassibille)
Abstract: This paper analyzes data from a field experiment in Madagascar that provided operational inputs to school actors and other administrators, and promoted local monitoring via parent-teacher meetings and school report cards. We ask whether this program changed the administrators’ behaviors and students’ learning outcomes. We find an improvement in various observable performance measures when the intervention was implemented at the school level. Teachers tracked student attendance more; they wrote lesson plans and gave exams more frequently. Meanwhile, teacher attendance and communication with parents did not improve, consistent with predictions of the multi-tasking principal-agent problem. The net effect on student attendance was positive: 4.3 percentage point increase from 85% mean attendance. Test scores improved by 0.1 standard deviations. We are also able to test for effort spillovers at the district level, and find little program impact on other untreated schools in the same districts.