The Impact of Measurement Noninvariance on Latent Change Score Modeling: A Monte Carlo Simulation Study

Measurement invariance (MI) overtime is required for meaningful interpretation of changes in the latent change score model (LCSM). In this simulation study, we investigate the impact of measurement noninvariance on the estimation of LCSM when proportional change model, dual change model, and their bivariate versions are used with mean composites or MI-assumed measurement models. The results show that even noninvariance in one item could result in severe bias in the LCSM parameter estimates and false statistical inferences (e.g., Type I error). However, the impact depends on the simulation factors, fitted models, and estimated parameters. For example, the intercept-factor mean is biased in proportional change model but not in dual change model. In bivariate LCSM the parameter estimates of only the noninvariant variable are biased except coupling effects. Model fit indices such as RMSEA and SRMR are insensitive to the ignored noninvariance. Discussions and implications of findings are presented