Analysis of Longitudinal Data with Missing Values in the Response and Covariates Using the Stochastic EM Algorithm

TL;DR

This paper introduces a stochastic EM algorithm combined with multiple imputations to handle non-random missing data in longitudinal studies, providing a way to obtain unbiased parameter estimates and their standard errors.

Contribution

It develops a novel SEM-based method with a Monte Carlo approach for standard error estimation in longitudinal data with non-random missingness.

Findings

Method performs well in simulation studies.

Application to real data demonstrates practical utility.

Provides unbiased estimates under non-random dropout.

Abstract

In longitudinal data a response variable is measured over time, or under different conditions, for a cohort of individuals. In many situations all intended measurements are not available which results in missing values. If the missing value is never followed by an observed measurement, this leads to dropout pattern. The missing values could be in the response variable, the covariates or in both. The missingness mechanism is termed non-random when the probability of missingness depends on the missing value and may be on the observed values. In this case the missing values should be considered in the analysis to avoid any potential bias. The aim of this article is to employ multiple imputations (MI) to handle missing values in covariates using. The selection model is used to model longitudinal data in the presence of non-random dropout. The stochastic EM algorithm (SEM) is developed to obtain the model parameter estimates in addition to the estimates of the dropout model. The SEM algorithm does not provide standard errors of the estimates. We developed a Monte Carlo method to obtain the standard errors. The proposed approach performance is evaluated through a simulation study. Also, the proposed approach is applied to a real data set.