Variance constraints strongly influenced model performance in growth mixture modeling: a simulation and empirical...

Area of Science:

• * Quantitative Psychology
• * Developmental Psychology
• * Statistical Modeling

Background:

• * Growth Mixture Modeling (GMM) is widely used for analyzing developmental trajectories.
• * Common issues include convergence problems and unreliable estimates due to variance parameter constraints.
• * Variance constraints can mitigate issues but may introduce bias if variances differ.

Purpose of the Study:

• * To identify optimal variance parameter constraints for Growth Mixture Modeling.
• * To evaluate the performance of different constraint strategies across various sample sizes.
• * To validate findings using a real-world longitudinal dataset.

Main Methods:

• * A simulation study was conducted to assess GMM performance under different variance constraint conditions.
• * The simulation varied sample sizes to determine constraint effectiveness.
• * Results were validated using the TRacking Adolescent Individuals' Lives Survey (TRAILS) cohort.

Main Results:

• * Unconstrained GMMs performed best when variance parameters differed across classes and time.
• * Constrained models, particularly those limiting random effect and residual variances across classes, showed poor performance.
• * Small sample sizes (N=100) presented challenges for all tested GMMs.

Conclusions:

• * Fit Growth Mixture Models without variance constraints whenever feasible.
• * If constraints are necessary, explore various specifications and avoid relying solely on default settings.
• * Thorough reporting of variance structures and adherence to guidelines like the GRoLTS-Checklist are crucial.