Finite Sample Effects in Group-Based Trajectory Models

Abstract

Two desirable properties of maximum likelihood-based parameter estimates are that the estimates are asymptotically unbiased and asymptotically normally distributed. In this article, the authors test whether the asymptotic properties of maximum likelihood estimation are achieved in sample sizes typically used in applications of group-based trajectory modeling. Through empirical results generated by resampling of population data, they find that the maximum likelihood estimates obtained in group-based trajectory models still provide reasonably close estimates of their true population values and have approximately normal distributions, even when estimated with a sample size as small as n = 500. Furthermore, and more important for the users of these types of models, the authors find similarly good performance in the model’s ability to estimate the transformed quantities of main interest: the group trajectories and mixing probabilities.