Distal outcomes in mixture modeling: A guide for pairwise comparisons, multiplicity control, and effect size...

Area of Science:

• Behavioral Sciences
• Psychology
• Education

Background:

• Mixture models are widely used in behavioral sciences, but lack clear guidelines for testing distal outcome differences.
• Current practices often involve uncorrected tests and omit effect sizes, hindering interpretability and reproducibility.
• An audit revealed infrequent reporting of multiplicity adjustments and effect sizes in applied mixture modeling studies.

Purpose of the Study:

• To provide a framework for testing, correcting for multiplicity, and reporting distal outcome differences in mixture modeling.
• To adapt general statistical recommendations for the specific context of latent class analysis and distal outcomes.
• To enhance the transparency and consistency of reporting practices in applied research.

Main Methods:

• Synthesizing recommendations from statistical literature for mixture modeling contexts.
• Focusing on continuous distal outcomes and comparisons of class-specific means.
• Proposing a framework for defining pairwise comparisons, implementing multiplicity corrections (recommending Benjamini-Hochberg), and reporting effect sizes and confidence intervals.

Main Results:

• A principled framework is proposed for defining, testing, and reporting distal outcome comparisons.
• The Benjamini-Hochberg procedure is recommended as a default for multiplicity correction.
• Methods for computing and reporting global and pairwise effect sizes and confidence intervals are demonstrated using readily available software outputs.

Conclusions:

• Implementing the proposed framework enhances the interpretability and reproducibility of mixture modeling findings.
• Standardized reporting of multiplicity-corrected pairwise comparisons and effect sizes is crucial.
• The framework is applicable across common auxiliary-variable approaches like ML three-step and BCH methods.