Message/Author |
|
|
Hello: I'm having some problems getting my head around how MPLUS can be jigged to run a Seemingly Unrelated Oredered Probit. Suppose for simplicity's sake, I have two ordinal dependent variables "red" and "blue" that I wish to regress on co-variates "c1" and "c2" I don't believe a model statement like this gives me what I need as it is the errors in the ordinal dependent variables that necessitate correlation: red on c1 c2; blue on c1 c2; Am I correct in this assumption? What I am thinking might be a workaround is to designate red and blue as latent variables and then proceed with the estimation? f_red by red@1; f_blue by blue@1; f_red on c1 c2; f_blue on c1 c2; Questions: 1) Is this line of reasoning correct? 2) Are there any other workarounds? I get too many integration points with the MLR (my preferred) estimator and non convergence with WLSMV. Thanks, Tom |
|
|
For WLSMV, you would say: red on c1 c2; blue on c1 c2; red WITH blue; For MLR, you would say: red on c1 c2; blue on c1 c2; f BY red@1 blue; f@1; [f@0]; where the residual covariance parameter is found in the free factor loading for blue. |
|
ssp2yssp12 posted on Wednesday, August 08, 2012 - 2:20 am
|
|
|
Hello: I am running a seemingly unrelated ordered probit model, with two ordinal dependent variables. I have run various nonnested models and would like to compare them. Is there a particular model fit/goodness of fit statistic in the output that I should be looking at to compare models? Thank you, S |
|
|
If you use maximum likelihood and the probit link, you could look at BIC. |
|
ssp2yssp12 posted on Wednesday, August 08, 2012 - 10:09 am
|
|
|
Thank you! However, what if I am using the MLMV or WLSMV estimators, rather than ML? |
|
|
All estimators that start with ML are maximum likelihood. You will not get BIC with WLSMV. |
|
Back to top |