2021/06/11 by Claudia Noack, Noack, Claudia
Engineering · Mathematics · #Advanced Causal Inference Techniques #Econometrics (econ.EM) #FOS: Economics and business #Nuclear reactor physics and engineering #Statistical Methods and Inference #Statistical Methods in Clinical Trials
paper · pdf · doi:10.48550/arxiv.2106.06421
openalex publication_date 2021/06/11 · openalex created_date 2021/07/05 · openalex updated_date 2026/07/28
In this paper, we develop a method to assess the sensitivity of local average\ntreatment effect estimates to potential violations of the monotonicity\nassumption of Imbens and Angrist (1994). We parameterize the degree to which\nmonotonicity is violated using two sensitivity parameters: the first one\ndetermines the share of defiers in the population, and the second one measures\ndifferences in the distributions of outcomes between compliers and defiers. For\neach pair of values of these sensitivity parameters, we derive sharp bounds on\nthe outcome distributions of compliers in the first-order stochastic dominance\nsense. We identify the robust region that is the set of all values of\nsensitivity parameters for which a given empirical conclusion, e.g. that the\nlocal average treatment effect is positive, is valid. Researchers can assess\nthe credibility of their conclusion by evaluating whether all the plausible\nsensitivity parameters lie in the robust region. We obtain confidence sets for\nthe robust region through a bootstrap procedure and illustrate the sensitivity\nanalysis in an empirical application. We also extend this framework to analyze\ntreatment effects of the entire population.\n