Inverse Probability Weighted Estimation for Monotone and Nonmonotone Missing Data

Abstract

Missing data is of common occurrence in epidemiologic research. In this paper, three data sets with induced missing values from the Collaborative Perinatal Project, a multisite United States study conducted from 1959 to 1974, are provided as examples of prototypical epidemiologic studies with missing data. Our goal is to estimate the effect of maternal smoking behavior on spontaneous abortion while adjusting for numerous confounders. At the same time, we do not necessarily wish to evaluate the joint distribution among potentially unobserved covariates, which is seldom the subject of substantive scientific interest. The inverse probability weighting approach preserves the semiparametric structure of the underlying model of substantive interest, and clearly separates the model of substantive interest from the model used to account for the missing data. However, inverse probability weighting often will not result in valid inference if the missing data pattern is nonmonotone, even if the data are missing at random. We describe a recently proposed approach to model nonmonotone missing data mechanisms under missing at random for use to construct the weights in inverse probability weighted complete-case estimation, and we illustrate the approach in the three data sets described in the companion manuscript (Am J Epidemiol. 2017;000(0):000-000) of this issue of the journal.