2021/10/26 by Alexander Kreiß, Kreiß, Alexander, Rothe, Christoph · 1 citation
Mathematics · #Advanced Causal Inference Techniques #Applications (stat.AP) #Econometrics (econ.EM) #FOS: Computer and information sciences #FOS: Economics and business #Statistical Methods and Inference #Statistical Methods in Clinical Trials
paper · pdf · doi:10.48550/arxiv.2110.13725
openalex publication_date 2021/10/26 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
We study regression discontinuity designs in which many predetermined\ncovariates, possibly much more than the number of observations, can be used to\nincrease the precision of treatment effect estimates. We consider a two-step\nestimator which first selects a small number of "important" covariates through\na localized Lasso-type procedure, and then, in a second step, estimates the\ntreatment effect by including the selected covariates linearly into the usual\nlocal linear estimator. We provide an in-depth analysis of the algorithm's\ntheoretical properties, showing that, under an approximate sparsity condition,\nthe resulting estimator is asymptotically normal, with asymptotic bias and\nvariance that are conceptually similar to those obtained in low-dimensional\nsettings. Bandwidth selection and inference can be carried out using standard\nmethods. We also provide simulations and an empirical application.\n