drlate: Doubly Robust and Kappa-Weighting Estimation of the Local Average Treatment Effect in R

Estimating the local average treatment effect (LATE) with covariates is a routine task in applied instrumental-variables analysis. The estimators with the strongest theoretical foundations- the doubly robust and Abadie-kappa weighting estimators of Słoczyński, Uysal, and Wooldridge- have until now been available only through Stata commands that report robust or cluster-robust standard errors and leave the weak-instrument problem of a ratio estimand unaddressed. The drlate package brings the complete program to R behind a single three-formula interface and, more consequentially, equips every estimator with inference that to our knowledge no other software provides in any language: weak-instrument-robust Fieller confidence sets for the covariate-adjusted LATE, cluster-robust and bootstrap standard errors, and a doubly robust Hausman test of unconfoundedness. All of it follows from one jointly stacked M-estimation system whose sandwich variance carries first-stage propensity-score uncertainty into the variance of the LATE. Around the estimators the package adds the diagnostic workflow applied work needs: overlap and covariate-balance checks, a formal balance test, complier profiling, sampling weights, and a one-call estimator comparison. We describe the methodology, the moment-block software architecture that makes the catalog extensible, and the full workflow on simulated and survey data. The core estimators reproduce the authors' Stata commands to within 10−6 (point estimates) and 10−4 (standard errors) across 38 scenarios; the probit and fractional families and the postestimation diagnostics are verified against their standard references; and a Monte Carlo calibration confirms the analytic, cluster-robust, and Fieller inference.