Lonjezo Sithole

University of Michigan · Economics · 2026–27 job market

Fields

Econometrics, Applied Micro

Job market paper

Multimargin Selection in Leniency Designs

Leniency designs exploit quasi-random assignment of cases to decision makers with different propensities to assign treatment. When decision makers weight unobserved case characteristics differently, a higher overall treatment rate can mask a lower probability of treating certain case types, with consequences for the interpretation of the IV estimand. I develop a model with finitely many latent regimes, each representing a case type with its own potential outcome distributions and decision-maker-specific treatment propensities. When each decision maker's treatment assignment within each regime and covariate cell preserves the marginal distribution of each potential outcome, observed outcome distributions admit a finite mixture representation. I show that pairwise Wald estimands weight regime-specific average treatment effects by each regime's relative contribution to the first stage. Because these weights can be negative, the IV estimand can be negative even when every regime's average treatment effect is positive. I provide conditions for identifying the number of regimes, regime-specific average treatment effects, and IV weights using either parametric restrictions on outcome distributions or auxiliary pretreatment measurements. I also characterize when the weights are nonnegative and when they are constant across decision-maker pairs, and develop pointwise asymptotically valid bootstrap tests of both restrictions for a fixed number of regimes. In an application to Miami-Dade County bail decisions, I find substantial variation across latent case types in the effects of pretrial release on sentence duration.