Latent growth mixture models as latent variable multigroup factor models: Comment on McNeish et al. (2023).

Area of Science:

• Psychometrics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Growth mixture models (GMMs) are used to identify latent subgroups with distinct developmental trajectories.
• Covariance pattern growth mixture models (CPGMMs) have been proposed for their convergence and subgroup identification capabilities.
• The current study critically evaluates the assumptions and applicability of CPGMMs.

Purpose of the Study:

• To challenge the universal applicability of CPGMMs.
• To propose a more general framework for latent growth modeling.
• To demonstrate the importance of data-driven model selection in GMMs.

Main Methods:

• Conceptual critique of CPGMMs, framing them as special cases of a broader model.
• Discussion of psychometric models with varying slope factor loadings across latent subgroups.
• Methodological suggestions for improving GMM convergence rates.
• Reanalysis of a longitudinal dataset on posttraumatic stress disorder (PTSD) symptomatology.

Main Results:

• CPGMMs are argued to be a specific type of random intercept model, not a general solution.
• Alternative psychometric models with subgroup-varying slope factor loadings are presented as conceptually superior.
• Improved convergence rates for GMMs can be achieved through specific initialization and model specification techniques.
• A reanalysis of PTSD data identified a three-group exponential decline model, questioning previously reported four-group patterns.

Conclusions:

• No single latent growth model is suitable for all research contexts; models must be 'right-sized' to the data.
• The findings suggest that previously identified complex patterns in PTSD symptomatology may be artefactual.
• This work advocates for a nuanced approach to GMM selection and interpretation.