Setting reference class PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Isaac posted on Thursday, March 17, 2011 - 9:24 am
Hi,
I'm running a logistic regression in an LGMM. When I use a binary variable it sets one class as the reference class. When I use all continuous variables it gives me alternative ref classes.
Try as I might, I can't get the ref class to change with with addition of the binary variable. I tried starting values but that doesn't seem to work. Any advice as to what I can do?
The ref class that I want is the largest class if that helps at all.
Thanks,
Isaac
 Bengt O. Muthen posted on Thursday, March 17, 2011 - 4:46 pm
Starting values should work. I assume the model with the binary vble is the largest model, including the cont's vbles, so use those estimates for starting values.

If you don't succeed, send to support.
 Aimee Palumbo posted on Tuesday, November 05, 2013 - 1:29 pm
Hello!
I'm trying to run a model with 3 latent class variables as predictors. I'm having trouble changing the reference class. The program automatically selects the combination of the last class of the 3 variables as the reference class, but based on previous output of those classes, I would like the first class of each variable (and their combination) to be the reference class. Below is the code for the class statement so you can see the variables and the number of classes, and then the code I was using for the code where MPlus decides the reference class.
CLASSES = disadv (2) advtg (2) biol (3);
...
MODEL:
%OVERALL%
monthssu on racer;
biol on advtg disadv racer;
advtg disadv on racer;

MODEL disadv:
%disadv#1%
[persngl-nophone];
biol on racer;
monthssu on racer;

%disadv#2%
[persngl-nophone];
biol on racer;
monthssu on racer;

MODEL advtg:
%advtg#1%
[maleprof-malemgmt];
biol on racer;
monthssu on racer;

%advtg#2%
[maleprof-malemgmt];
biol on racer;
monthssu on racer;

MODEL biol:
%biol#1%
[ki_67r$1 p53r$1 prr$1 err$1 her2r$1];
monthssu on racer;

%biol#2%
[ki_67r$1 p53r$1 prr$1 err$1 her2r$1];
monthssu on racer;

%biol#3%
[ki_67r$1 p53r$1 prr$1 err$1 her2r$1];
monthssu on racer;

Thank you!
 Bengt O. Muthen posted on Tuesday, November 05, 2013 - 5:45 pm
You can try giving good starting values for class indicators such as

[persngl-nophone];

and combine that with Starts=0, hoping that you get the same best loglikelihood as in your original run.

Otherwise you would have to use Model constraint to reparameterize the model and I would think that is a bit cumbersome.
 Aimee Palumbo posted on Tuesday, November 05, 2013 - 7:45 pm
Thank you for the response. I have some follow-up questions. You mentioned giving good starting values and I noticed in other responses you recommend making the starting values for the class I want to be the reference class equal to the class that was the reference class. Is that what you mean by giving good starting values? How would that work with the combination of classes? Or, how else would I determine what the good starting values are? (I was going to use the input provided from the SVALUES statement, but I was having trouble manipulating the starting values in the right way).
Thank you again!
 Bengt O. Muthen posted on Wednesday, November 06, 2013 - 11:07 am
Good starting values are obtained from the solution that you got. So for instance if 2 classes have estimated means for the observed variable y:

%c#1%
[y*-2];
%c#2%
[y*+3];

you give the starting values

%c#1%
[y*+3];
%c#2%
[y*-2];
 Aimee Palumbo posted on Thursday, November 07, 2013 - 6:22 am
Sorry, I have another follow up. When I paste in the estimated means into the input, the code (annotated with what I changed) looks like what I pasted below. But when I run it it doesn't work (I get no output at all). Am I changing the write things? Thanks!

%OVERALL%
monthssu ON racer;
biol#3 ON advtg#1*-0.22257; !Orig biol#1...
disadv#2 ON racer*-2.73738; !Orig disadv#1

[ disadv#2*1.41498 ]; !Orig disadv#1 ...
[ biol#3*0.46229 ]; !Orig biol#1

%DISADV#1.ADVTG#1.BIOL#1% !Orig 2/2/3
monthssu ON racer*-0.02290 (40);
[ monthssu@0 ];
persngl*0.00393 (18);
...

%DISADV#2.ADVTG#2.BIOL#3 %!Orig 1/1/1
monthssu ON racer*0.87204 (36);
[ monthssu*-0.51634 ];
...

MODEL DISADV:
%DISADV#2% !Orig disadv#1
biol#3 ON racer*0.73992; !Orig biol#1
biol#2 ON racer*-2.12795;
[ persngl*0.29310 ] (1);
...
%DISADV#1% !Orig disadv#2
biol#3 ON racer*0.06563; !Orig biol#1
...

MODEL ADVTG: (changed the same way as DISADV)

MODEL BIOL:
%BIOL#3% !Changed from biol#1
[ ki_67r$1*-1.73823 ] (66);
...
%BIOL#1% !Changed from biol#3
[ ki_67r$1*-1.43263 ] (76);
 Linda K. Muthen posted on Thursday, November 07, 2013 - 6:28 am
Please send your files and license number to support@statmodel.com.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: