A Primer on Inverse Probability of Treatment Weighting and Marginal Structural Models

Abstract

Emerging adulthood researchers are often interested in the effects of developmental tasks. The majority of transitions that occur during the period of early/emerging adulthood are not randomized; therefore, their effects on developmental trajectories are subject to potential bias due to confounding. Traditionally, confounding has been addressed using regression adjustment; however, there are viable alternatives, such as propensity score matching and inverse probability of treatment weighting. Propensity scores are probabilities of selecting treatment given values on observed covariates. Inverse probability of treatment weights are also based on estimated probabilities of treatment selection and can be used to create so-called pseudo-populations in which confounders and treatment are unrelated to each other. In longitudinal models, such weighting can occur at multiple time points. This article provides a primer on these weighting methods and illustrates their application to studies of emerging adulthood. We provide annotated computer code for both SPSS and R, for both binary and continuous treatments.