A1125
Title: Model selection for structural nested models using double/debiased machine learning in causal inference
Authors: Tatsuki Nishihata - Keio University (Japan) [presenting]
Yoshiyuki Ninomiya - The Institute of Statistical Mathematics (Japan)
Abstract: Double/debiased machine learning (DML) provides a framework for achieving doubly robust estimation in high-dimensional settings. By combining machine learning methods for estimating nuisance parameters with orthogonalization and sample splitting techniques, DML improves the convergence rate of estimators for target parameters. Recent advances in causal inference have seen widespread application of doubly robust methods in marginal structural models (MSMs) and structural nested models (SNMs). In this context, the development of valid model selection criteria is essential. For MSMs, an asymptotic Cp criterion has been proposed as a mathematically sound approach to selecting the mean structure. Building on this foundation, the purpose is to develop an asymptotic Cp-type model selection criterion for SNMs, incorporating doubly robust estimation via DML for the mean structure.
