QUANTITATIVE RESEARCH METHODS WORKSHOP
Abstract: Leave-k-out (LKO) robustness checks are widely used to assess whether excluding specific observations meaningfully changes the results of an empirical investigation. However, critics argue that this practice lacks a clear theoretical justification and arbitrarily alters the sample. This paper addresses this concern by demonstrating that the LKO set can function as a diagnostic for unobserved confounding through a novel conceptual tool we call the “corrosive covariate.” The key insight is that a worst-case omitted variable can selectively negate the influence of specific observations without actually removing them from the sample. This framework centers discussions of unobserved confounding around the plausibility of the corrosive covariate simultaneously providing a justification for LKO sets and a novel form of sensitivity analysis based on focused case studies. This approach is applied to a reanalysis of three studies on divided government.
Austin Jang is a Ph.D. candidate in the joint Statistics & Data Science and Political Science program at Yale University, specializing in quantitative methods. His research combines political methodology, statistics, and computational social science to bridge the gap between statistical output and scientific understanding, providing tools and frameworks that help empirical researchers interpret results, diagnose problems, and assess robustness. Jang’s dissertation, “Essays on Computational Methods for Causal Inference,” develops computational and conceptual tools for understanding research findings. His work includes designing adversarial simulation studies using machine learning, reinterpreting leave-k-out diagnostics through the lens of omitted variable bias, and creating practical applications of influence function-based diagnostics for modern statistical methods. His research also addresses fundamental questions about the scope and validity of empirical inference, including work on temporal validity, the foundations of design-based causal inference, and applications in democratic governance. Jang’s work has been published in Research & Politics, with additional work under review at Political Analysis.
The Quantitative Research Methods Workshop is open to the Yale community. To receive regular announcements and invitations to attend, please subscribe at this link.
The Quantitative Research Methods Workshop series is sponsored by the ISPS Center for the Study of American Politics and The Whitney and Betty MacMillan Center for International and Area Studies at Yale with support from the Edward J. and Dorothy Clarke Kempf Fund.