Covariate Balancing Inverse Probability Weights for Time-Varying Continuous Interventions

Abstract

In this paper we present a continuous extension for longitudinal analysis settings of the recently proposed Covariate Balancing Propensity Score (CBPS) methodology. While extensions of the CBPS methodology to both marginal structural models and general treatment regimes have been proposed, these extensions have been kept separately. We propose to bring them together using the generalized method of moments to estimate inverse probability weights such that after weighting the association between time-varying covariates and the treatment is minimized. A simulation analysis confirms the correlation-breaking performance of the proposed technique. As an empirical application we look at the impact the gradual roll-out of Seguro Popular, a universal health insurance program, has had on the resources available for the provision of healthcare services in Mexico.

Keywords

• causal inference
• covariate balancing propensity score
• inverse probability weights
• sequential ignorability
• time-varying continuous interventions