| Message/Author |
|
|
|
|
| I have isolated a 4-class Growth Mixture Model with gender as a knownclass grouping variable. Because we are estimating group structure using the knownclass option, we cannot use the newer auxiliary options to estimate mean differences between class on continuous distal outcomes (e.g., dcontinuous). I have been trying to use the manual 3-step option, however I did not get the table with logits for the classification probabilities for the most likely latent class by latent class in my output when I ran it in V7.2. I am aware that there were issues in V7.11 with switched rows and columns for the logit output. Do you recommend calculating the logits by hand using average latent class probabilities from the output or is there a way to get them in the output in V7.2? |
|
|
|
|
| If I understand correctly you have 4 latent classes + gender as a knownclass so total 8 classes. First note that the mean of the distal outcome will also be different across gender. One possible solution is to do the following. Use the parameter estimates from the 8 class model and run the 4 class male and female separately with all parameters fixed at the 8 class parameter values. You have to be quite careful here especially with the [C#i] parameters since for one of the genders they won't be explicit. Once you get the nominal indicator and the tables run the males and females third step separately. |
|
|
|
|
| Thank you for your prompt reply. I ran the models separately for each gender with all parameters fixed at the 8 class parameter values as you suggested and was able to get logit tables. However, could you please clarify what you mean that I need to be careful with the [C#i] parameters since for one of the genders they won't be explicit? Would this still be the case if I am to run the third step separately for each gender? |
|
|
|
|
| My comment about [C#i] is for the case when you are running the 8-class solution since you need C|gender in logit scale. If you are not running the 8 class model then you don't have the problem. |
|
|
|
|
Thank you for clarifying, this was very helpful. I have two follow up questions. First, does this mean that I cannot compare classes between genders on distal outcomes (e.g. girl class 1 to boy class 3)? Second, when I compare classes using the following model test command, I get a warning that the wald test cannot be computed because of a singular covariance matrix.
MODEL TEST:
DM1=DM2;
DM1=DM3;
DM1=DM4;
DM2=DM3;
DM2=DM4;
DM3=DM4;
Based on suggestions from previous postings, I shortened the model test command to include only two contrasts and changed dm1=dm2 to 0=dm1-dm2 and did not get any warnings (command reproduced below). Does this mean I can only conduct two contrasts at a time?
MODEL TEST:
0 = DM1-DM2;
0 = DM1-DM3; |
|
|
|
|
The correct specification to do a joint comparison is:
MODEL TEST:
0 = D2 - D1;
0 = D3 - D1;
0 = D4 - D1; |
|
|
|
|
| Thank you for your reply. Does this mean it is not possible to compare across all four classes within one model? For example, with the above specification, comparisons between class 2 and class 3 would not be included in the overall test. When I include these comparisons, I get warnings that the wald test cannot be computed because of a singular covariance matrix. |
|
|
|
|
| You can do a second run comparing just class 2 and class 3. |
|
|
| Back to top |
