Residual Balancing: A Method of Constructing Weights for Marginal Structural Models

• DOI: 10.1017/pan.2020.2
• Publication Date: 2020-03-04T07:16:30Z
• Authors: Xiang Zhou; Geoffrey T. Wodtke
• Keywords: Marginal structural model; Inverse probability weighting; Spurious relationship; Conditional probability distribution

When making causal inferences, post-treatment confounders complicate analyses of time-varying treatment effects. Conditioning on these variables naively to estimate marginal effects may inappropriately block pathways and induce spurious associations between the outcome, leading bias. To avoid such bias, researchers often use structural models (MSMs) with inverse probability weighting (IPW). However, IPW requires for conditional distributions is highly sensitive their misspecification. Moreover, relatively inefficient, susceptible finite-sample difficult continuous treatments. We introduce an alternative method constructing weights MSMs, which we call “residual balancing”. In contrast IPW, it modeling means rather than treatment, therefore easier Numeric simulations suggest that residual balancing both more efficient robust model misspecification its variants in a variety scenarios. illustrate by estimating (a) cumulative effect negative advertising election outcomes (b) controlled direct shared democracy public support war. Open-source software available implementing proposed method.