Estimation of a Semiparametric Natural Direct Effect Model Incorporating Baseline Covariates

Area of Science:

• Causal inference
• Statistical methodology
• Observational studies

Background:

• Establishing cause-effect relationships is fundamental in empirical science.
• Understanding mediation, the extent to which an effect is direct or indirect through a third variable, is crucial.
• Recent semiparametric theory enables multiply robust estimation of causal effects in observational studies.

Purpose of the Study:

• To extend existing semiparametric theory for causal mediation analysis.
• To handle parametric models of natural direct and indirect effects, considering pre-exposure variables.
• To address challenges in high-dimensional covariate settings where direct estimation is infeasible.

Main Methods:

• Extension of semiparametric theory to parametric models with identity or log link functions.
• Identification of the curse of dimensionality as a barrier to estimation in unrestricted models.
• Development of multiply robust estimation strategies under a more general model assumption.

Main Results:

• Demonstration that direct estimation is often infeasible due to high-dimensional covariates.
• Proposal of a generalized modeling approach assuming partial validity of working models.
• Application of multiply robust estimation to overcome estimation challenges.

Conclusions:

• The proposed methods extend causal mediation analysis to parametric models.
• Multiply robust estimation offers a viable approach for estimating direct and indirect effects in complex observational settings.
• The findings address limitations of previous theories in high-dimensional scenarios.