Message/Author 


Dear, In growht mixture models: can you use the t statistic to evaluate the significance of the variance of the latent slope ? I have encountered this in a couple of articles. Should one not use a mixture of chisquare distributions to test for this ? Thanks for your advice, 


A mixture of chisquares is the proper thing to use in principle. Many in the field seem to use the z scores as an approximation (which may not be good enough). 


The test statistic is calculated as a difference in twice the log likelihood not ? Or can one use the square of the zstatistic ? Greetings, Liesbeth 


Z squared is equal to two times the loglikelihood difference. Neither of them is a mixture of chisquares. Both of them are used in practice. 


Dear Linda, I was wondering. Suppose I have a GMM, with two classes. For both classes I specify a random intercept and slope. To test the significance of the variance of the slope growth factor I fit a model including a random slope and a model with a fixed slope. I compare the log likelihood values of these two models. Is this allowed ? Are this still nested models; what if the partitioning of the patients in two classes is different in these two models ? Thanks for your reaction, Liesbeth 


Neither the ztest nor the LL difference test would be appropriate here although they are probably used in practice as an approximation. 

Back to top 