Work

A Monte Carlo Evaluation of Growth Mixture Models: Effects of Varying Distributional Parameters on Grouping Outcomes

Year: 2019
Type: article
Abstract: Statistical methods designed to group individuals into homogenous subsets have existed for decades.Historically, when research identifies weakness in one method, new methods are developed in attempts to provide more valid subgroups.At present, growth mixture modeling (GMM) and its variants, which group individuals based on similar longitudinal growth trajectories, is popular.However, research addressing the validity of latent subgroups is limited.This Monte Carlo simulation study tests the efficiency of GMM in correctly identifying known numbers of subgroups in data sets across various conditions.GMM was applied to randomly simulated data, and to data containing 1, 2, 3, and 4 latent trajectory classes.Effects of varying skew and kurtosis were also examined, as were effects of sample size, intercept effect size, and proportions within latent subgroups.Growth patterns were set at zero (no growth), linear growth, and quadratic growth.A total of 2,090 combinations of parameters were examined, with 1,000 replications each.Results were mixed.When data were completely random, GMM often failed to converge, or identified one group with errors in model fit.When one group was simulated with varying skew and kurtosis, GMM often identified multiple groups.When two groups were simulated, GMM performed well with steep linear growth or quadratic growth.When 3 to 4 groups were simulated, GMM was most effective when intercept effect sizes and sample sizes were large.In other cases, GMM often underestimated the correct number of groups when the true number was between 2 and 4.These results indicate significant limitations of GMM that Beauchaine, for his support and encouragement in my research pursuits and his guidance and patience as he helped foster my writing skills.Further, my committee, Dr. Mike Vasey and Dr. Paul DeBoeck for their feedback and support in the dissertation process.I would also like to acknowledge Dr. Jennifer Cheavens, who often provided me with kind words and support.From my training before OSU
Language: English
Open Access status: green
OpenAlex ID: W7002197552
FWCI:
Cites: 125
Cited by:
Related to: 10