Bayesian Analysis of Multi-Factorial Experimental Designs Using SEM.

Area of Science:

• Statistics
• Psychometrics
• Quantitative Psychology

Background:

• Latent repeated measures ANOVA (L-RM-ANOVA) offers advanced capabilities over traditional methods, including handling missing data and examining interindividual differences.
• However, L-RM-ANOVA's reliance on maximum likelihood limits its ability to incorporate prior information and can result in poor performance with small sample sizes.

Purpose of the Study:

• To demonstrate the integration of Bayesian estimation with L-RM-ANOVA to overcome limitations of maximum likelihood estimation.
• To illustrate the application of informative and weakly informative priors in Bayesian L-RM-ANOVA for improved statistical properties.

Main Methods:

• The study adapts L-RM-ANOVA for Bayesian estimation, enabling the incorporation of prior knowledge into the model parameters.
• Methods include placing informative priors on main and interaction effects and weakly informative priors on standardized parameters.
• A real empirical example is used to demonstrate the practical implementation and model specification.

Main Results:

• Bayesian estimation in L-RM-ANOVA can reduce Type 1 error rates and bias compared to traditional methods.
• The Bayesian approach has the potential to increase statistical power and efficiency, particularly when priors are appropriately chosen.
• The study provides guidance on necessary parameter estimates for specifying informative priors, moving beyond traditional ANOVA tables.

Conclusions:

• Bayesian estimation provides a robust alternative for L-RM-ANOVA, addressing limitations associated with maximum likelihood methods.
• The use of informative and weakly informative priors enhances the reliability and precision of L-RM-ANOVA results.
• This work advocates for more comprehensive reporting of parameter estimates to facilitate cumulative research in quantitative fields.