CMStatistics 2021: Start Registration
View Submission - CMStatistics
B1396
Title: Post-machine learning causal inference: Uniformly valid inference and sensitivity analysis Authors:  Niloofar Moosavi - Umeå university (Sweden)
Tetiana Gorbach - Umea University (Sweden)
Jenny Haggstrom - Umea University (Sweden)
Xavier de Luna - Umea University (Sweden) [presenting]
Abstract: The recent literature on uniform valid causal inference is reviewed, and sensitivity analysis to unobserved confounders is discussed in this context. When inference is aimed at a low dimensional causal parameter, double robust estimation strategies combined with machine learning algorithms to fit high-dimensional nuisance models are becoming increasingly popular. This popularity is due to key theoretical results obtained in recent years that guarantee the uniform validity of the inference post-model selection (or post-machine learning). We discuss the costs and benefits of using uniformly valid causal inference. Another major threat to the validity of causal inference is the potential existence of unobserved confounders. We, therefore, present new results for dealing with uncertainty due to unobserved confounding in the context of uniformly valid causal inference.