Latent growth curve (LGC) modeling is emerging as a preferred method of longitudinal analysis, which uses the structural equation modeling (SEM) framework to demonstrate growth or change (Meredith & Tisak, 1990). The purpose of this dissertation was to examine the performance of commonly utilized measures of model fit in LGC modeling data environments. A Monte Carlo simulation was conducted to examine the influence of LGC modeling design characteristics (i.e., sample size, waves of data, and model complexity) on selected fit indexes (i.e., x², NNFI, CFI, and RMSEA) estimated in correct LGC models. The CFI performed the best, followed by the NNFI, x², and finally, the RMSEA showed the least desirable characteristics. The RMSEA was found to over-reject correct models (i.e., suggest poor model fit) in conditions of small to moderate sample size (N = 1,000) and few waves of data. The x² over-rejected correct multivariate models with more waves of data and small sample sizes (N = 100). The NNFI over-rejected unvariate and multivariate models with small sample size (N = 100) and three waves of data. Six guidelines were proposed for LGC modeling researchers, including: maximizing the chance of obtaining a plausible solutions, cautioning the use of the x², adopting the novel LGC modeling cutoff values, using multiple fit indexes, and assessing the within-person fit. As LGC modeling applications escalate in the social and behavioral sciences, there is a critical need for additional research regarding LGC model fit, specifically, the sensitivity of fit indexes to relevant types of LGC model misspecification.

Download Full Article