[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: test to compare reduced and full models in xtgee with logit and log links

From   Maarten buis <[email protected]>
To   [email protected]
Subject   Re: st: test to compare reduced and full models in xtgee with logit and log links
Date   Wed, 13 Jan 2010 00:47:09 -0800 (PST)

--- On Wed, 13/1/10, Rosenstock, Summer E. wrote:
> I would like to test whether adding an interaction term to
> my model makes the model an overall better fit. In the past
> with logistic regression I've used the log likelihood ratio
> test to see if there is added benefit by comparing the full
> model to the reduced model, but stata is not so keep to do
> this with xtgee. 

Testing interactions in a non-linear model can be a 
surprisingly complicated, if you are interested in marginal
effects, see e.g. (Norton et al. 2004). However, things 
are much easier if you want to remain within the odds ratio
or risk ratio metric. In your case you can then estimated
the model with interactions and use 
-testnl exp(_b[interaction])=1- to test the interaction 
effect (where "interaction" is the name of your interaction

Here is a recent post about the difference between the two:

> I'm having to use logit in some cases because not
> all of my log-binomial models will converge, which is super
> frustrating...but that's another topic. 

I posted an alternative solution to that problem recently:

Hope this helps,

Edward C. Norton, Hua Wang and Chunrong Ai (2004) Computing 
interaction effects and standard errors in logit and probit 
models. The Stata Journal, 4(2): 154--167.

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index