Latent growth curve models (LGCM) became in recent years a very popular technique for longitudinal data analysis: they allow individuals to have distinct growth trajectories over time. These patterns of change are summarized in relatively few parameters: the means and variances of the random effects (random intercept and random slope), as well as the covariance between intercept and slope (Bollen & Curran, 2006). Although the specified model structure imposes normality assumptions, the data analyst often faces data deviations from normality, implying mild, moderate or even severe values for skewness and or kurtosis. A traditional approach for generating data that deviates from the normal distribution was proposed by Vale-Maurelli (1983). Recently, Foldnes and Olsson (2016) proposed the independent generator transform approach to generate multivariate non-normal distributed data. Following this new approach, in the current paper a Monte Carlo simulation study was conducted in R, using lavaan, in order to investigate the effect of observed data deviations from normality on the standard errors and goodness of fit indices. LGCMs with unconditional linear growth are considered. Three and four time points, and sample sizes ranging from 50 to 500 observations are used. The impacts of such deviations on parameter estimates, standard error and fit measures are discussed.

--