A Monte Carlo evaluation of growth mixture modeling

Area of Science:

• Developmental Psychology
• Clinical Science
• Quantitative Psychology

Background:

• Growth mixture modeling (GMM) is widely used to identify distinct developmental trajectories.
• However, the validity of latent subgroups identified by GMM requires further investigation.

Purpose of the Study:

• To evaluate the accuracy of GMM in identifying known subgroups under diverse conditions.
• To assess the influence of distributional characteristics, sample size, and growth patterns on GMM performance.

Main Methods:

• A Monte Carlo simulation was employed, examining 1,955 parameter combinations across 1,000 replications.
• Simulations varied skew, kurtosis, sample size, intercept effect size, growth patterns (none, linear, quadratic, exponential), and group proportions.
• Standard fit indices were used to assess GMM's ability to recover known group structures (k=1-4).

Main Results:

• GMM frequently misidentified the number of true groups, particularly when data deviated from normality.
• Accuracy decreased with fewer steep growth trajectories or smaller sample sizes/effect sizes.
• GMM often underestimated the true number of groups (2-4) under less ideal conditions.

Conclusions:

• Caution is advised when interpreting GMM-derived subgroupings, especially in real-world research.
• The study highlights the sensitivity of GMM to data characteristics and potential for inaccurate subgroup identification.
• Researchers should critically evaluate GMM assumptions and results in light of these findings.