rdlasso: A Stata command for high-dimensional regression discontinuity designs

The rdlasso command implements regression discontinuity designs (RDD) with high-dimensional covariates in Stata. The procedure is based on the methodology developed by Kreiss and Rothe (2023), and extends it to both sharp and fuzzy designs. Covariate selection is performed through a lasso-based local estimation, ensuring valid inference under approximate sparsity. The command is built using Stata’s Python integration via the SFI module and automates all steps of the estimation process—from covariate selection to bandwidth choice and bias-corrected treatment-effect estimation. The syntax allows for flexible user control while remaining fully embedded in the Stata environment. rdlasso enables Stata users to apply machine learning techniques for causal inference without requiring programming in external platforms such as R or Python. The command generates output variables that can be used for further postestimation analysis within the same session. An option automatically distinguishes between sharp and fuzzy designs, making the tool both user-friendly and methodologically complete. The implementation is illustrated through a step-by-step example and an empirical application. The command contributes to the growing set of tools for modern causal analysis in Stata, particularly in high-dimensional settings.