Message/Author 

Anonymous posted on Tuesday, July 24, 2012  9:01 am



I am looking for a guide as to how Mplus calculates 'results in probability scale' having estimated the latent class model. Can you advise? 


The formula is p = 1 / (1 + exp(threshold)) 

Anonymous posted on Tuesday, July 24, 2012  11:41 am



Thank you. And how are the class membership probabilities calculated? 


Mplus prints estimates of [c#1 c#2...]; that is the logits corresponding to the intercepts/means of the multinomial regression on covariates (no covs implies that the intercepts are the means). Then use UG Chapter 14 formulas on pages 441445 to translate from those logits to probabilities. 


To follow up on this question, how does Mplus obtain the SEs for the results in probability scale, following an LCA? As short explanation or reference would be great, thank you. 


Sorry I should have looked into this more closely before posting. It seems that the SEs for the results in probability scale can be obtained using the delta method and the output in TECH3. Thanks, Peter 


Right. 


I am running a threestep LPA using syntax that I have successfully used before and the output is not giving me the results in probability scale. I have checked that I have not made any mistakes in the syntax, and it's the same just with different variable names. I've tried running the LPA using BCH, R3STEP, and DE3STEP, and nothing is giving me the full output with the results in the probability scale. I can send my syntax but unfortunately cannot send my data as they're federally restricted data and my license prohibits that. 


BCH and DE3STEP are for continuous distal outcomes. See Tables 6 and 7 of Mplus Web Note 21. Use DCAT. 


Thanks for the quick response. My outcomes are continuous (high school GPA and number of courses), so BCH should be OK. I reran and still am not getting the results in probability scale. I'm sort of at a loss for what I'm doing differently and wrong. A colleague recommended (and I've been reading about) the manual 3step approach. Might that be an appropriate solution here? 


You can send your output to Support along with your license number. Also, try the manual approach described in web note 21. 


Note that probability output would be relevant only for a categorical outcome not a continuous one. 


Thanks, I will look at the note and keep working. Where do I find comparable output to help me interpret the nature of the classes? (Sorry, I'm used to working with LCA, this is my first time with LPA...) 

Back to top 