Testing Nested Models

[NickHeeks]

Hi all

My statistics knowledge is fairly basic so forgive me if this is a simple question. I am using a cumulative logit model and am trying to test whether the addition of a 2nd level improves the fit of the model. Previously I have used the likelihood ratio test, but MLwiN doesn't produce this for logistic regression.

What test should I use to work this out and how do I do this in MLwiN?

Thanks

[billb]

Hi Nick,
That's quite tricky. If you are using classical estimation then you can do an approximate Wald test though this isn't so great as assumes normality for the variance parameter. Alternatively you can use the DIC diagnostic with MCMC estimation although again this is not universally liked but would be my route.
Hope that helps.
Bill.

[garrygeorge]

well explained