Research

Below are current projects and selected papers.

• Birds of a Feather Collude Together: Subnational Alignment and Corruption
  Joint with Leopoldo Fergusson, Arturo Harker, and Carlos Molina
  [Paper] (Conditionally accepted at APSR)
  
  Abstract [+]

  We examine how subnational partisan alignment influences corruption in clientelistic environments, focusing on the fabrication of “ghost” students to inflate education transfers to local governments in Colombia. Using a Regression Discontinuity Design, we find that partisan alignment between municipal mayors and departmental governors increases ghost students by 0.3 standard deviations, without improving genuine enrollment or student performance. Alignment also leads to more discretionary hiring, patronage-based outsourcing, and increased electoral fraud risk. The effects are strongest in municipalities with weaker institutions and entrenched clientelism. Alignment also raises the likelihood that mayors’ relatives are appointed to departmental posts and governors’ relatives to municipal posts, consistent with reciprocal patronage. These findings support the view that resource diversion benefits politicians with few benefits for local constituencies. Aligned politicians also experience better future electoral prospects, suggesting a breakdown in accountability. Our results highlight how clientelistic networks distort public service delivery, reinforcing the persistence of political corruption.

• Poverty Targeting with Imperfect Information
  [Paper] [Slides – World Congress] (Submitted)
  
  Abstract [+]

  A key challenge for targeted antipoverty programs in developing countries is that policymakers must rely on estimated rather than observed income, which leads to substantial targeting errors. The policy problem is not only to predict income, but to decide how noisy income estimates should be translated into feasible transfers. I formulate this as a statistical decision problem in which a policymaker chooses transfers to minimize a poverty-targeting loss subject to a fixed budget and a no-taxation constraint. I show that the standard plug-in rule, which treats estimated incomes as true, is inadmissible. I develop a nonparametric empirical Bayes targeting rule that assigns transfers using posterior distributions of true poverty gaps. Although the budget and no-taxation constraints make the targeting rule nonsmooth, Bayes regret is governed by the accuracy of the posterior functionals that determine the oracle allocation. In simulations using household survey data from nine African countries, the empirical Bayes rule reaches substantially more poor households and systematically improves poverty reduction relative to plug-in OLS and machine-learning benchmarks.

• When and How to Pilot: Design Rules for Two-Wave Experiments
  [Draft available upon request] [Slides – AFE 2025]
  
  Abstract [+]

  Researchers often run pilots before the main wave of an experiment, but it is unclear how much a small pilot should change the eventual experimental design. This paper studies how pilot evidence should guide treatment assignment probabilities in two-wave experiments. Balanced assignment is safe but cannot adapt to differences in outcome variability across arms. By contrast, feasible Neyman allocation can improve precision but may overreact to noisy pilot estimates in finite samples. I propose a conditional minimax-regret rule that moves away from balance only when the pilot provides defensible evidence to do so. The rule uses a finite-sample confidence set for the variance pair and reports a bound on the remaining precision loss. It is simple to compute, recovers Neyman allocation at the root-pilot rate in regular cases, and matches the uniform minimax-regret rate. I also characterize when a pilot is worth running, and extend the framework to multi-arm, stratified, and cluster-randomized designs.

• Two-Way Effects Models: a Nonparametric Empirical Bayes Approach
  Joint with Cole Davis
  [Draft available upon request]
  
  Abstract [+]

  Researchers estimate two-way effects models to decompose outcomes into unit and cluster components, such as workers and firms, teachers and schools, or individuals and regions. Existing empirical Bayes approaches for this setting rely on parametric prior assumptions. We develop a nonparametric empirical Bayes framework that allows the distribution of unit effects to vary with latent cluster effects, so unit and cluster components need not be independent and units within a cluster can be correlated. We propose a feasible estimation procedure and characterize the resulting shrinkage rules. Simulations show mean squared error close to an oracle benchmark and improvements over i.i.d.-based methods.