Linear Regression PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Carola Viehoever posted on Friday, May 14, 2004 - 7:11 am
There might be a mistake in the following Syntax, because the Output contains the same regressionkoefficients for all classes. We have no clue, where the mistake is! Can anybody find the mistake?

SYNTAX
Data:
File is o:\LMH\Manuskripte\Greipl\Mappe1.dat ;
Variable:
Names are
f4l f6e f9wba ;
Usevariables are
f6e f9wba ;
classes = c(10);
Analysis:
Type=mixture;
MODEL:

%overall%
f9wba on f6e ;

%C#1%
[f9wba f6e *-2.1973] ;

%C#2%
[f9wba f6e *-2.1973] ;

%C#3%
[f9wba f6e *-2.1973] ;

%C#4%
[f9wba f6e *-2.1973] ;

%C#5%
[f9wba f6e *-2.1973] ;

%C#6%
[f9wba f6e *-2.1973] ;

%C#7%
[f9wba f6e *-2.1973] ;

%C#8%
[f9wba f6e *-2.1973] ;

%C#9%
[f9wba f6e *-2.1973] ;

%C#10%
[f9wba f6e *-2.1973] ;


OUTPUT:
TECH10;
SAVEDATA:
file is lca_ex1.txt ;
save is cprob;
format is free;
 Linda K. Muthen posted on Friday, May 14, 2004 - 7:54 am
They are held equal across classes as the default. You need to mention them in the class-specific MODEL commands to relax the equality constraint. I would do this one class at a time.
 Anonymous posted on Monday, May 17, 2004 - 2:18 am
Hi Linda,
thank you for your fast answer. How can i relax the constraint? I thought that regression coeff's specified by the ON command are free, these by BY command not.
 Linda K. Muthen posted on Monday, May 17, 2004 - 6:09 am
You are referring to defaults for multiple group analysis. The defaults differ for mixture models. You relax the constraint by putting the ON statement in the class-specific MODEL commands, for example,

%C#1%
[f9wba f6e *-2.1973] ;
f9wba on f6e ;

I think this should work.
 Anonymous posted on Monday, May 17, 2004 - 8:06 am
Hi Linda,

thanks a lot for your answer! It works!!! :-)

Now we´re only looking for the R squares for each class. Are they calculated in mplus?
 Linda K. Muthen posted on Monday, May 17, 2004 - 8:35 am
Ask for STANDARDIZED in the OUTPUT command and you should get them.
 Anonymous posted on Friday, July 22, 2005 - 10:18 am
Professors Muthen,

I would like to report adjusted R-square statistics for two classes in a mixture regression model with Mplus 3.12. I am able to get the R-square statistics for the separate classes with Standardized output but would like to adjust these for the the number of IVs and cases in each class.

Is this appropriate for this form of model?

If so, should I just use the traditional formula for adjusting R-square using the n's for the two classes? Does the latent class variable require any different treatment in the formula?

Thanks.
 bmuthen posted on Friday, July 22, 2005 - 11:10 am
I would think the same approach applies as in regular regression, but I have not looked into this.
 Anonymous posted on Friday, July 22, 2005 - 3:00 pm
(I tried to post this follow-up question earlier but don't see it here, so I apologize if this is a duplicate posting.)


Thanks, Professor Muthen.

Because I found substantial variation in the two class sizes, I think it is important that I report adjusted R-squares for the separate class regressions to reflect those class size differences in relation to the number of independent variables in my model.

If you will allow me, I have a follow-up question about that calculation. I am using this formula to make the adjustments:

AdjRSQ = RSQ - ((1 - RSQ) * numIVs)/(n - numIVs -1)

I am using the class size reported by Mplus as the n in each adjustment, but I'm not sure whether the latent class variable should be counted in numIVs. Since the latent class is a constant within each class regression model, I would think it should not be included in the numIVs in the RSQ adjustments for each class.

Would you advise me on this?

Also, since you do not report adjusted R-square in Mplus, I'm wondering if you would suggest I just ignore this whole issue and simply report the unadjusted R-squares. I have had reviewers insist on adjusted R-squares in the past.

Thanks again for your help, patience and wisdom on this.
 bmuthen posted on Friday, July 22, 2005 - 5:17 pm
I would think the latent class variable should not be counted in numIVs. I ignore adjusting R-square, but then again I haven't been up against reviewers commenting on my regressions.
 Anonymous posted on Friday, July 22, 2005 - 5:33 pm
Thanks so much! Have a great weekend.
 Anonymous posted on Wednesday, October 11, 2006 - 5:00 am
Good morning,

Is there a way to generate information regarding sums of squares, total sums of squares, etc... using Mplus output commands? If there is not, could you suggest a method for accomplishing this using Mplus output? Thank you.
 Linda K. Muthen posted on Wednesday, October 11, 2006 - 10:25 am
There is no straighforward way to do this using the Mplus output.
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: