Across Latin America, school dropout is a growing concern, because of its negative social and economic consequences. Although a wide range of interventions hold potential to reduce dropout rates, policy makers in many countries must first address the basic question of how to target limited resources effectively for such interventions. Identifying who is most likely to drop out and, therefore, who should be prioritized for targeting, is a prediction problem that has been addressed in a rich set of research in countries with strong education system data. This paper makes use of newly established administrative data systems in Guatemala and Honduras, to estimate some of the first dropout prediction models for lower-middle-income countries. These models can correctly identify 80 percent of sixth grade students who will drop out in the transition to lower secondary school, performing as well as models used in the United States and providing more accurate results than other commonly used targeting approaches.

*Featured in the World Bank’s Research Digest (Fall 2017).

This paper studies whether providing information on funding opportunities and average earnings by degree-college pairs affects higher education decisions. We conducted a randomized controlled trial in Bogotá, Colombia, on a representative sample of 120 public high schools, half of which received an informational talk delivered by recent college graduates. Using survey data linked to administrative records, we analyze student beliefs and evaluate the intervention. Findings show that most students overestimate college premiums, but the talk does not affect these beliefs. It does improve knowledge of financing programs, especially among the poor. There is no evidence that information disclosure affects post-secondary enrollment. However, students in treated schools who do enroll choose more selective colleges. These positive effects are mostly driven by students from better socioeconomic backgrounds. We conclude that information policies are ineffective to raise college enrollment in contexts with significant barriers to entry, but may influence intensive margin choices.

*Featured in the World Bank’s Development Impact Blog (Jul 2016).

We study the effect of a large increase in access to gambling on crime by exploiting the expansion of video gambling terminals in Illinois since 2012. Even though video gambling was legalized by the State of Illinois, local municipalities were left with the decision whether to allow it within their jurisdiction. The City of Chicago does not allow video gambling, while many adjacent jurisdictions do. We take advantage of this setting along with detailed incident level data on crime for Chicago to examine the effect of access to gambling on crime. We use a difference-in-differences strategy that compares crime in areas that are closer to video gambling establishments with those that are further away along with the timing of video gambling adoption. We find that (i) access to gambling increases violent and property crimes; (ii) these are new crimes rather than displaced incidents; and (iii) the effects seem to be persistent in time.

We study whether Honduran children exposed to a conditional cash transfer report lasting effects on human capital and labor market outcomes in early adulthood. While the government randomized transfers by municipalities, the control group was ultimately contaminated due to political reasons. Despite contamination, this program provides an opportunity to test an unanswered question: how does differential exposure to demand, supply, or both incentives for education impact long-run outcomes? Using municipal-level panel data, effects are estimated by difference-in-differences. We find that the form of delivering cash transfers affects the degree to which these programs achieve their long-term goal of reducing poverty. Relative to comparison municipalities who received three years of demand-side support, separate exposure to demand or supply-side incentives has no lasting impact. Joint exposure to both incentives leads to higher schooling and greater labor market attachment, especially for women. These differences are significant and robust to multiple hypothesis adjustments.

  • “How important is spatial correlation in randomized controlled trials?” with Kathy Baylis.

Randomized controlled trials (RCTs) have become the gold standard for impact evaluation since they provide unbiased estimates of causal effects. This paper focuses on RCTs that allocate treatment status over clusters in geographical proximity. We study how omitting spatial correlation in outcomes or unobservables at the cluster-level affects difference-in-difference estimates at the individual-level. Using Monte Carlo experiments, we identify bias and efficiency problems and propose solutions to overcome them. Our framework is then tested on data from Mexico’s Progresa program. Results show that spatial correlation may affect both the precision of the estimate and the estimate itself, especially when geographic dependence is high.