I have a two class GMM and Iím examining auxiliary variables using du3step for continuous variables and de3step for binary variables. The de3step output shows 999.00 for the mean, S.E., chi square and p-value for binary variables where the proportion of 1ís is less than 10%. What does this mean for my analyses?
I also used du3step to analyse differences in proportions for binary variables between classes. All the chi square tests that were significant in de3step became non-significant. Does this suggest a problem with my de3step analyses or model?
Thank you for your help with this. I tried to run the DCAT/DCON analyses using 7.11, however it did not work because I am using data with sampling weights (which I did not mention in my initial post). Given this issue with my data, do you have any further suggestions for how I could look at differences between my two trajectories?
The WEIGHT option cannot be used with these options. If your entropy is .8 or higher, you can use most likely class membership.
Mike Todd posted on Wednesday, September 11, 2013 - 11:11 am
I'm trying to implement the Lanza approach in an LPA model with a continuous distal outcome ("TotPA" in this case), but I'm not sure I'm setting up the syntax properly.
I tried the following specifications:
(1) AUXILIARY = TotPA(DCON); statement under the VARIABLE command, along with MODEL command statements setting associations between latent means and TotPA at 0 (e.g., C1#1 on TotPA@0;) and labeling TotPA's mean and variance (e.g., [TotPA] (1)). This failed to run, returning an error message indicating that TotPA was an unrecognized variable.
(2) Same as (1) above, but without the fixed and labeled parameters. This ran and gave me class-specific means and the omnibus and pairwise between-class tests of differences among/between the means. In this analysis, can I assume that the auxiliary variable didn't affect estimation of the latent class part of the model?
I am examining an auxiliary continuous distal outcome variable in an LPA model. The automatic 3-step process (DE3Step and DU3Step) returned the previously mentioned error with 999's for the means, S.E.
The Lanza (DCON) method was successful. However, I am interested in controlling for covariates such as gender and age. Is there any way to include control variables?