Research

Working Papers

Multi-Margin Selection in Leniency Designs - Job Market Paper

Author: Lonjezo Sithole
PDF

Abstract. Judge leniency designs are widely used in empirical economics, but their conventional interpretation relies on a single-index assumption: that judges differ only in overall strictness. I develop a latent-regime framework in which judge assignment can shift treatment along multiple latent margins, so decision-makers may differ in which case features they weight, not just in how strict they are overall. Under weak assumptions, the model identifies how many latent regimes are active; under stronger shape or parametric restrictions, it can point-identify the regime-specific outcome laws and mixture weights. The conventional Wald ratio then decomposes into regime-specific treatment effects with weights that depend on which judges are compared, yielding a direct test of whether a scalar leniency interpretation is adequate. In an application to felony bail decisions, the standard scalar leniency score fails the rank-1 benchmark for Black defendants but not for Non-Black defendants, implying that the same leniency variation has different economic content across groups. For the sentencing outcome, the estimated cross-race gap in the Wald ratio is driven primarily by which decision environments the instrument activates rather than by differences in what incarceration does within a given regime.

Semisupervised Causal Inference with Unstructured Data - Draft coming soon

Author: Lonjezo Sithole

A Locally Robust Semiparametric Approach to Examiner IV Designs - Reject and Resubmit, Journal of Econometrics

Author: Lonjezo Sithole
PDF (replaces an older ArXiv version)

Awarded the 2024 Outstanding Third Year Paper Prize.

Abstract. Examiner IV designs often require rich covariate adjustment. In such settings, the leniency instrument must be estimated, and current practice typically relies on unbiased jackknife IV (UJIVE), which is well suited to saturated linear first stages but restrictive when there are many examiners and many covariates. This paper develops a semiparametric estimator and asymptotic inference procedure for the standard covariate-adjusted examiner-IV estimand when the treatment propensity is estimated flexibly. I derive an examiner-specific orthogonal score, show that the associated Riesz representers admit explicit conditional-expectation formulas, and establish a multiple-robustness property of the resulting moment condition. Under mean-square consistency and product-rate conditions, the estimator is root-n consistent and asymptotically normal, and cross-fitting removes own-observation bias while reducing the effect of first-step estimation error on second-step inference. Monte Carlo evidence shows coverage close to nominal in a correctly specified benchmark design and sizable bias and RMSE improvements over naive plug-in estimation and linear UJIVE in a harder nonlinear design of the same sample size.

Nonparametric Testability of Slutsky Symmetry - Submitted

Authors: Florian Gunsilius, Lonjezo Sithole
PDF

Abstract. Economic theory implies strong limitations on what types of consumption behavior are considered rational. Rationality implies that the Slutsky matrix, which captures the substitution effects of compensated price changes on demand for different goods, is symmetric and negative semi-definite. While empirically informed versions of negative semi-definiteness have been shown to be nonparametrically testable, the analogous question for Slutsky symmetry has remained open. Recently, it has even been shown that the symmetry condition is not testable via the average Slutsky matrix, prompting conjectures about its non-testability. We settle this question by deriving nonparametric conditional quantile restrictions on observable data that permit construction of a fully nonparametric test for Slutsky symmetry in an empirical setting with individual heterogeneity and endogeneity. The theoretical contribution is a multivariate generalization of identification results for partial effects in nonseparable models without monotonicity, which is of independent interest. This result has implications for different areas in econometric theory, including nonparametric welfare analysis with individual heterogeneity for which, in the case of more than two goods, the symmetry condition introduces a nonlinear correction factor.

Maximum Overlap Component Analysis - Draft coming soon

Authors: Florian Gunsilius, Lonjezo Sithole

Work in Progress

Nonparametric Tests of Weak and Strong Separability

Authors: Lonjezo Sithole, Florian Gunsilius, Per Hjertstrand