서강대학교

초록

In this study, different types of growth mixture model (GMM) regard to distributional assumptions are compared. Normal GMM and skew-t GMM are the focus of comparison. Normal distribution being specific type of skew-t distribution with skew parameters of 0 and degree of freedom of infinite. A Monte Carlo simulation was performed for examining the performance of fit indices for model selection and the accuracy of parameter estimation under incorrect and correct model specification. Two different types of data sets were generated: one from a normal GMM and the other from a skew-t GMM. Then each type of data was analyzed using normal GMM as well as skew-t GMM with differing numbers of classes. The results showed that BIC performed the best in terms of choosing the correct model. When skew-t GMM was fitted to normal data, all the fit indices preferred the correct number of classes. However, the parameter estimation was not accurate. When normal GMM was fitted to skew-t data, all the fit indices led to over-extraction of classes. Under correct model specification, skew-t GMM performed better than normal GMM. These results showed the importance of choosing a correct model, and how the result could be misleading if an incorrect model were fitted to data. The implications, limitations, and possible extensions of this study are discussed.