Anonymous posted on Monday, March 18, 2002 - 10:58 pm
I am running version 2.2 and am wondering if I can use an ordered categorical as a mixture indicator. My variable is on a 0-4 scale and has a case with a score from 0-4 for each latent class. I'm receiving an error message that says "Ordered thresholds 1 and 2 for class indicator U1 are not increasing."
Do you have an example where you have used an ordered cat. as a mix indicator? Any suggestions for start values? Are the values still 0-4 or do they become 0-1?
Any help is greatly appreciated!
My syntax is below:
have the following model: VARIABLE: NAMES are x1 x2 x3 y1 y2 y3 u1; USEVAR =x1 x2 x3 y1 y2 y3 u1; categorical= u1; Classes = c(3);
Analysis: Type = mixture; miteration =403; Model: %overall% int by y1@1; int by y2*; int by y3*; [y1-y3@0]; c#1 on x1; c#2 on x1;
Yes,Version 2 does allow ordered polytomous latent class indicators. However, you must give starting values for each threshold and they must be increasing. It looks like you only gave a starting value for u$1, so the default value of 0 is used for u$2 and this is less than the starting values given for u$1.
Examples 25.9A and 25.11A give examples of assigning starting values for ordered polytomous latent class indicators.
Do you have any guidelines in terms of the maximum number of variables one might enter in a latent class analysis, using ordered polytomous indicators (provided one has a large sample)? Is it it possible to combine both ordered polytomous and continous variables as indicators of a latent class model? Thanks!
You can combine continuous and polytomous latent class indicators.
I don't know of any guidelines. Of course, sample size is an issue. And Mplus has a limit of 500. But that shouldn't be a problem. Perhaps some of the references on the website might address this issue. There is also a Sage publication by McCutcheon on latent class analysis.
Yi-fu Chen posted on Wednesday, March 12, 2003 - 1:55 pm
Hi, I am quite new to the latent class analysis. Now I have a model with 3 ordinal observed indicators (3 categories each) and a three-class latent variable. I have trouble to give starting value for the thresholds. I am thinking to use the observed proportions (say, 35%, 35%, 30% for indicator 1) to give starting value. Is this a reasonable way? If yes, can you provide any formula about the connection between observed proportions and the thresholds? If no, could you provide any guideline to give starting value of thresholds?
Any sugguestion and reference are appreciated.
bmuthen posted on Wednesday, March 12, 2003 - 2:40 pm
That's an ok approach. Regarding translating proportions (probabilities) into thresholds on the logit scale, please see the User's Guide, Technical Appendix 1, page 342. Note that a threshold is the same as an intercept, but with opposite sign. Let us know if this is not sufficient help. If you want to take a simple approach, with 3-category items you can simply use logit threshold starting values of -1, 0, +1.
Yi-fu Chen posted on Thursday, March 13, 2003 - 8:18 pm
Hi, Thank you for your reply. I have tried your suggestion using -1, 0, 1. But I keep getting the error messages as following: THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO A CHANGE IN THE LOGLIKELIHOOD DURING THE LAST E STEP.
AN INSUFFICENT NUMBER OF E STEP ITERATIONS MAY HAVE BEEN USED. INCREASE THE NUMBER OF MITERATIONS OR INCREASE THE MCONVERGENCE VALUE. ESTIMATES CANNOT BE TRUSTED. SLOW CONVERGENCE DUE TO PARAMETER 15. THE LOGLIKELIHOOD DERIVATIVE FOR THIS PARAMETER IS 0.11049487D-02.
IN THE OPTIMIZATION, ONE OR MORE LOGIT THRESHOLDS APPROACHED AND WERE SET AT THE EXTREME VALUES. EXTREME VALUES ARE -15.000 AND 15.000. THE FOLLOWING THRESHOLDS WERE SET AT THESE VALUES: * THRESHOLD 1 OF CLASS INDICATOR BOY3 FOR CLASS C#3 AT ITERATION 75 * THRESHOLD 2 OF CLASS INDICATOR BOY3 FOR CLASS C#3 AT ITERATION 75
Increase the number of MITERATIONS in the ANALYSIS command. Also, check TECH1 to see which parameter 15 is. Then check your output to see what value that parameter gets to before estimation stops. If it is close to 0 or one fix the parameter value to 15 or -15.
I don't have a reference for understanding logit thresholds. Just think of them values that have to be exceeded for the event to occur.