

One LC variable moderating effect of ... 

Message/Author 


I have a latent class variable CN (measured by u1u10) predicting distal outcome Y. I want to examine if another latent class variable CP (measured by u11u15) moderates this relationship. Would it be correct to do: Step 1a: LCA for CN on u1u10, save class membership N1 and get classification error logits Step 1b: LCA for CP on u11u15; save class membership N2 and get logits Step 3: Use syntax: VARIABLE: NOMINAL = N1 N2; CLASSES = CN(4) CP(3); ANALYSIS: TYPE = MIXTURE; MODEL: %OVERALL% Y on CN; %CN#1% (also for %CN#2%, %CN#3%, %CN#4%) [N1#1@logit] [N1#2@logit] [N1#3@logit] [N1#4@logit] %CP#1% (also for %CP#2%, %CP#3%) [N2#1@logit] [N2#2@logit] [N2#3@logit] Y ON CN; Thanks much for helping me figure this out! 


You need to add Model cn:... and Model cp: ... See UG examples with more than one latent class variable. Also, you can't have Y ON N because Mplus doesn't regress on nominal variables. Instead, simply say [Y]; to let the mean of Y vary across the classes. Web Note 15 points out, however, that the class percentages may change when you include Y as a DV in the model. You should check for that. This change is what the new DCON option tries to avoid. But DCON can be used only with one latent class variable. 


Thank you, Dr. Bengt Muthen. I'll definitely check class switching. As for the syntax, in addition to [Y] under %OVERALL%, I am not sure whether to put [Y] under MODEL CN or MODEL CP, or both. I am interested in whether CP moderates the association between CN and Y. Please let me know which I should do. Thank you! MODEL: %OVERALL% [Y]; MODEL CN: %CN#1% (also for %CN#2%, %CN#3%, %CN#4%) [N1#1@logit] [N1#2@logit] [N1#3@logit] [N1#4@logit] MODEL CP: %CP#1% (also for %CP#2%, %CP#3%) [N2#1@logit] [N2#2@logit] [N2#3@logit] [Y] 


I would actually delete [y] from Model cp, in which case I think the y mean will vary across all the combinations of classes (check this). This is the interaction you are looking for. 


Thank you. That makes sense! I'll try it. 

Back to top 

