Loglikelihood ratio test with MI data PreviousNext
Mplus Discussion > Missing Data Modeling >
Message/Author
 Thomas Klausch posted on Sunday, November 06, 2011 - 4:09 pm
Dear discussion list, Linda and Bengt,

A statistical question, semi-related to Mplus, maybe you can help. I am trying to do compute an LRT (loglikelihood ratio test) between two nested models using a multiply imputed data set. The model is logistic regression.

The type=imputation command line produces a mean loglikelihood with a standard deviation. How do I compute the LRT test from this accounting for the SE of the log likelihood? Can Mplus compute an LRT test between two or multiple nested models for me?

The latter questions is also more general - I would not know how to have Mplus perform an LRT it in a single data set case.

Thank you very much.

Thomas
 Tihomir Asparouhov posted on Monday, November 07, 2011 - 7:13 pm
You should take a look at
http://statmodel.com/download/MI7.pdf

This article describes the LRT-imputation method of Meng and Rubin.

In Mplus this method is implemented for the test of model fit (unrestricted model v.s. structural model). For general pair of nested models the method is not yet implemented in Mplus. Alternatively you may be able to utilize the Wald test (see the above note for description) available with the "Model test" command.

Without imputations the LRT is is simply double the log-likelihood difference between the two nested models.
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: