Seventh Mplus Users' Meeting
Abstract – Different approaches to Bayesian Measurement Invariance modeling and testing
Jean-Paul Fox (This is joint work with Research master student Vera Broks).
Construct measurements from persons clustered in groups can only be correctly interpreted when measurement invariance assumptions hold. Different Bayesian models have been proposed to model violations of measurement invariance to correctly compare scores of persons from different groups. Results of a conditional (IRT) model, where the violation(s) of measurement invariance is explicitly modeled, can be obtained using Mplus. Inferences can be made about scale differences across groups, and measurement invariance assumptions can be tested. This approach is compared to a new Bayesian marginalized IRT model, where possible violations of measurement invariance are represented in correlated response observations given the construct variable. In a simulation study both Bayesian procedures will be compared.