BLRT PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Janne Petersen posted on Sunday, August 30, 2009 - 7:06 am
Hi

I have applied the BLRT to help me deciding the number of classes in a traditional latent class model. But what happens when this test is applied to a model with restrictions? More concrete then I have assumed the conditional probability for a certain answer to an item to be equal among some classes. M-plus do not give any errors, but I am not sure what the alternative model is. Is it the model with k-1 classes without any restrictions or … ?

Thank you, Janne
 Bengt O. Muthen posted on Sunday, August 30, 2009 - 3:11 pm
I think the k and k-1 model have to have the same parameter restrictions because the BLRT procedure generates data according to a k-1 model using the same parameter restrictions.
 Janne Petersen posted on Monday, September 07, 2009 - 4:43 am
Thank you for your answer. It gave me a couple of ideas that I have now tried out. My problem is that maybe it is not clear which restriction there is left on a k-1 class model, when there are parameter restrictions across classes. One of my k class models looks like this:

%c#1%
[perifer$1] (1);
[vener$1] (2);
[facial$1] (3);
[abdom$1] (4);
[buffalo$1] (5);
%c#2%
[perifer$1];
[vener$1];
[facial$1];
[abdom$1](4);
[buffalo$1](5);
%c#3%
[perifer$1](1);
[vener$1](2);
[facial$1](3);
[abdom$1];
[buffalo$1];

I have tried fitting this model switching around the classes, and the BLRT and it’s DF changes. In this example the BLRT only has one degree of freedom. My guess is that this is because the two-class model with class 2 and class 3 does not have any restrictions when class 1 is erased. Is it correct that Mplus always try to remove the first class? If so it is very nice, because then I can control the restrictions in my k-1 alternative model.

Alternatively, is there a way to directly specify the k-1 class model? This is relevant for me, as I have more complicated parameter restrictions across classes for the pi’s.

Thank you, Janne
 Linda K. Muthen posted on Monday, September 07, 2009 - 9:39 am
Mplus always drops the first class. In your case, dropping the first class also removes some constraints.

There is no way to specifically specify the k-1 class model.
 Thomas Olino posted on Friday, November 13, 2009 - 11:30 am
I wanted to clarify on the implementation of tech14 in the newest version of Mplus. Is it still necessary to have the last class (ie., the k-th class) as the largest class?

Many thanks!
 Linda K. Muthen posted on Friday, November 13, 2009 - 12:06 pm
It is still recommended to have the largest class as the last class.
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: