

Fmm for continuous variables 

Message/Author 

Tom Cheong posted on Wednesday, November 04, 2009  7:18 pm



1. For a 2 class model, is it correct that one of the alphas has to be zero and the other one will denote the difference in the factor means between two classes? Could they be constrained to be the same? 2. I included algorithm = integration in the analysis. I read from the other post that says that it should be removed for continuous variables. Is saving computing time the reason or that is not allowed? I seem to get the similar answers. 3. What does the parameter gamma(c) as reported in tech1 (with algorithm = intergration) denote? There isn't such a parameter when I commented out the algorithm = integration command. I could not find info. about that parameter in the user's guide. 4. I plotted a series plot for the results using the estimated means. Could one get s.e. for those means? Are those means computed using the formula nu + lambda*alpha? Thank you. 


1. Yes, the mean of the factor in one class must be zero. To constrain them to be the same, factor means must be fixed to zero in all classes. 3. You go down a different path in the program if you leave ALGORITHM =INTEGRATION in the input. The convergence criteria are different which could account for small differences in results. 3. Gamma (c) contains the multinomial regression coefficients for the categorical latent variable regressed on a set of observed covariates. See page 596 of the user's guide. 4. MODEL CONSTRAINT can be used to obtain standard errors of the predicted values using the formula you give. You can use the standard errors to create confidence intervals. 

Tom Cheong posted on Tuesday, November 10, 2009  9:20 pm



Thank you, I looked up the user's guide on p. 555558 on Model Constraint and have not been able to find out how to figure out the standard error of the estimated means. My understanding is for the formula for the estimated mean, I'll label those parameters of interest, then use the NEW keyword to create the new variable, but I'm not sure how to get the se. Here is my understanding: 1.lambda: f by x1 (p1) x2 (p2) x3 (p3); 2. nu [x1] p4; 3. alpha [f] p5; Then MODEL CONSTRAINT; NEW(estex1); estx1 = p4 + (p1 * x5); Am I on the right track? How would I be able to get the se. of estx1? Thanks. 


When you define a new parameter in MODEL CONSTRAINT, a standard error is estimated for the parameter. You do not give it. 

Back to top 

