The potential of growth mixture modelling

NOT GMM VERSUS LCGA, BUT BOTH IN A GENERAL LATENT VARIABLE FRAMEWORK The authors bring up the important issue of choice of model within the general framework of mixture modelling, especially the choice between latent class growth analysis (LCGA) techniques developed by Nagin and colleagues versus GMM developed by Muthen and colleagues. LCGA specifies that all individuals in a trajectory class behave the same, whereas GMM allows for within-class variation. Nagin’s writings show reservations about the virtues of GMM (see Nagin, 2005; Nagin & Tremblay, 2005) and some of this is reflected in the current paper. Unfortunately, in my view, Nagin’s writings on GMM contain many misconceptions. One is that the inclusion of within-class variation clouds the meaning of the resulting classes. Another is that LCGA is superior to GMM by avoiding normality assumptions on the growth factors, instead using an unrestricted, non-parametric representation with latent classes capturing the latent variable distribution. What is lost in Nagin’s writing is that GMM and LCGA are closer in spirit than what first impressions might suggest. Furthermore, statistics can help choose the model that fits the data best. If the GMM model gives a considerably better log likelihood value for fewer (or at least not many more) parameters than the LCGA, GMM should clearly be chosen over LCGA. Having access to the general latent variable modelling framework of the 3B2 ICD : 482 PROD.TYPE: COM ED:DURGAS PAGN: KN.JAGADISH SCAN: pp.1^3 (col.¢g.: NIL)