Research

Publications

Advisor Value-Added and Student Outcomes: Evidence from Randomly Assigned College Advisors, with Serena Canaan and Pierre Mouganie. American Economic Journal: Economic Policy, 2022, 14(4):151-91. 

Abstract: This paper provides the first causal evidence on the impact of college advisor quality on student outcomes. To do so, we exploit a unique setting where students are randomly assigned to faculty advisors during their first year of college. We find that higher advisor value-added (VA) substantially improves freshman year GPA, time to complete freshman year and four-year graduation rates. Additionally, higher advisor VA increases high-ability students’ likelihood of enrolling and graduating with a STEM degree. Our results indicate that allocating resources towards improving the quality of academic advising may play a key role in promoting college success. 

Working Papers

The Impact of Religious Diversity on Students’ Academic and Behavioral Outcomes,with Serena Canaan and Pierre Mouganie.  (Revise and Resubmit at Journal of Labor Economics)  

Abstract: This paper explores how religious diversity affects college students’ academic performance and behavior towards members of other religions. Our setting is a secular four-year university located in Lebanon, a country that is deeply divided along religious lines. To identify causal effects, we exploit the university’s random assignment of first-year students to peer groups. We proxy students’ religious backgrounds by whether they attended secular, Christian or Islamic high schools. We find that exposure to peers from different religious backgrounds increases Muslim students’ enrollment in classes with non-Muslim instructors, suggesting that contact improves openness towards members of other religions. Inter-religious contact also impacts students’ academic performance. We show that while exposure to peers from non-Islamic high schools increases Muslim student GPA, exposure to peers from Islamic backgrounds reduces the GPA among students from secular high schools. These asymmetric effects highlight the heterogeneous academic returns to inter-religious mixing in a divided society.

Estimating treatment-effect heterogeneity across sites in multi-site randomized experiments with imperfect compliance, with Clément de Chaisemartin.

Abstract: We consider multi-site randomized controlled trials with a large number of small sites and imperfect compliance, conducted in non-random convenience samples in each site. We show that an Empirical-Bayes (EB) estimator can be used to estimate a lower bound of the variance of intention-to-treat (ITT) effects across sites. We also propose bounds for the coefficient from a regression of site-level ITTs on sites' control-group outcome. Turning to local average treatment effects (LATEs), the EB estimator cannot be used to estimate their variance, because site-level LATE estimators are biased. Instead, we propose two testable assumptions under which the LATEs' variance can be written as a function of sites' ITT and first-stage (FS) effects, thus allowing us to use an EB estimator leveraging only unbiased ITT and FS estimators. We revisit  Behaghel et al. (2014), who study the effect of counselling programs on job seekers job-finding rate, in more than 200 job placement agencies in France. We find considerable ITT heterogeneity, and even more LATE heterogeneity: our lower bounds on ITTs' (resp. LATEs') standard deviation are more than three (resp. four) times larger than the average ITT (resp. LATE) across sites. Sites with a lower job-finding rate in the control group have larger ITT effects

A Framework for Using Value-Added in Regressions

Abstract: As increasingly popular metrics of worker and institutional quality, estimated value-added (VA) measures are now widely used as dependent or explanatory variables in regressions. For example, VA is used as an explanatory variable when examining the relationship between teacher VA and students' long-run outcomes. Due to the multi-step nature of VA estimation, the standard errors (SEs) researchers routinely use when including VA measures in OLS regressions are incorrect. In this paper, I show that the assumptions underpinning VA models naturally lead to a generalized method of moments (GMM) framework.  Using this insight,  I construct correct SEs' for regressions that use VA as an explanatory variable and for regressions where VA is the outcome. In addition, I identify the causes of incorrect SEs when using OLS, discuss the need to adjust SEs under different sets of assumptions, and propose a more efficient estimator for using VA as an explanatory variable. Finally, I illustrate my results using data from North Carolina, and show that correcting SEs results in an increase that is larger than the impact of clustering SEs.

Clustering and External Validity in Randomized Controlled Trials , with Clément de Chaisemartin.

Abstract: In the literature studying randomized controlled trials (RCTs), it is often assumed that the potential outcomes of units participating in the experiment are deterministic. This assumption is unlikely to hold, as stochastic shocks may take place during the experiment. In this paper, we consider the case of an RCT with individual-level treatment assignment, and we allow for individual-level and cluster-level (e.g. village-level) shocks to affect the potential outcomes. We show that one can draw inference on two estimands: the ATE conditional on the realizations of the cluster-level shocks, using heteroskedasticity-robust standard errors; the ATE netted out of those shocks, using cluster-robust standard errors. By clustering, researchers can test if the treatment would still have had an effect, had the stochastic shocks that occurred during the experiment been different. Then, the decision to cluster or not depends on the level of external validity one would like to achieve.