ABSTRACT
Confirmatory factor analysis of multitrait-multimethod data (CFA-MTMM) can be complex and frequentist estimation techniques become unstable or non-trustworthy in small samples. Bayesian estimation allows researchers to include prior information for the parameters in a CFA-MTMM model. However, constructing informative priors can be challenging, especially if previous studies have used different CFA-MTMM models. We propose a model-free approach to derive informative priors for any CFA-MTMM model that is consistent with the MTMM approach by Campbell & Fiske (1959). In this chapter, we illustrate how the model-free approach can be used to derive informative priors for the parameters in the single and multiple indicator CTC(M-1) model based on commonly reported descriptive statistics. The model-free approach can be combined with meta-analytical tools to obtain a range of plausible values of the key model parameters (e.g., variance of the prior distribution). Results of a small simulation study show that Bayesian estimation with moderate or strong informative priors outperforms maximum likelihood estimation in small samples (N ≤50). Finally, we provide practical guidelines and cautionary notes for the application of Bayesian estimation with informative priors in small samples.
