Multiple robust estimation of marginal structural mean models for unconstrained outcomes.

Area of Science:

• Causal inference
• Longitudinal data analysis
• Statistical modeling

Background:

• Marginal structural mean models are used for causal effect estimation from longitudinal observational data.
• Existing inverse probability of treatment weighted and double robust (DR) estimators have limitations.
• DR estimation requires specifying potentially incompatible models for counterfactual outcomes.

Purpose of the Study:

• To develop improved estimators for marginal structural mean models.
• To address the model compatibility issues in double robust (DR) estimation.
• To propose a novel multiple robust (MR) estimator offering greater protection against model misspecification.

Main Methods:

• Utilizing a likelihood parameterization proposed by Robins et al. (2000b) for compatible parametric models.
• Extending this parameterization to construct a novel double robust (DR) estimator.
• Developing a multiple robust (MR) estimator based on the compatible parametric models.
• Implementing methods through iterative fitting of weighted regressions.

Main Results:

• The proposed methods are shown to be easy to implement.
• The novel multiple robust (MR) estimator provides enhanced robustness against model misspecification compared to DR estimators.
• The study fills a gap in exploiting the Robins et al. (2000b) parameterization for DR estimation.

Conclusions:

• The developed multiple robust (MR) estimator offers a more reliable approach for causal inference from longitudinal data.
• The iterative weighted regression approach facilitates practical implementation.
• This work advances the methodology for handling model misspecification in causal effect estimation.