Seventh Mplus Users' Meeting
Abstract – Combining approximate zero constraints for measurement invariance and cross-loadings: An application of dual process growth curve models with panel data
The paper investigates a common developmental pattern of hedonistic beliefs and associations with delinquent peer groups in adolescence. Based on theory and previous research it is assumed, that the two latent variables (‘hedonism’ and ‘association’) are partly intertwined as some indicators seem to be co-determined from both dimensions. Thus, a modeling strategy for the parallel processes has to take into account measurement invariance (MI) and cross-loadings (CLs) simultaneously. With Bayesian structural equation modeling (BSEM), exact zero parameter constraints for MI and CLs can be relaxed with zero mean, small variance priors to yield approximate zero constraints, and to assure a more reasonable representation of the assumptions. The simultaneous assessment of approximate MI and CLs in multivariate latent growth models is demonstrated using panel data for n =1,186 adolescents from the German criminological study “Crime in the modern city”.