Modeling actor and partner effects in dyadic data when outcomes are categorical.

Area of Science:

• Social Psychology
• Statistical Modeling
• Relationship Science

Background:

• Relationship outcomes are influenced by individual and partner inputs.
• Linear mixed models (LMMs) are standard for Gaussian dyadic data analysis.
• Actor-partner interdependence model (APIM) is commonly used.

Purpose of the Study:

• To evaluate generalized linear mixed models (GLMMs) for non-Gaussian dyadic outcomes.
• To assess performance of GLMM approximation techniques for actor and partner effects.
• To identify superior methods for analyzing dyadic data with non-normal distributions.

Main Methods:

• Investigation of approximation techniques within standard software for GLMMs.
• Comparison of GLMM performance against generalized estimating equations (GEE).
• Focus on estimating actor effects, partner effects, and within-dyad correlation.

Main Results:

• GLMM approximation techniques showed unsatisfactory performance for actor-partner effects and within-dyad correlation.
• Performance issues were pronounced with negative within-dyad correlations and small sample sizes.
• Generalized estimating equations (GEE) emerged as a viable alternative for non-Gaussian dyadic data.

Conclusions:

• Standard GLMMs and their approximations are insufficient for many non-Gaussian dyadic data analyses.
• GEE provides a more reliable approach for estimating effects in non-Gaussian dyadic relationships.
• Further research into robust statistical methods for dyadic data is warranted.