On Wed, 29 Aug 2007, Mary wrote:

> Kevin, > > Thanks for your reply. > > Yes, 3 additional observations were dropped in the analysis- the first > model used 876 of 890 observations, and the second with the additional > variable used 873 of 890. > > Still, wouldn't one compare models using the AIC statistic? If the > Type 3 statistic for the variable added is also significant, shouldn't I > still conclude that the second model is the better model?

To compare AIC's, you need to omit those three from model one. Simply use a dataset WHERE option:

proc reg data = test ( where = ( additional_var ne . )) ;

Several issues may cloud their loss. For instance, are they representative or leverage points?

The AIC is calculated from the likelihood, which depends on, among other things, the number of observations.

Once you established the model, then you can report your results with the additional observation re-included, if the first model was more appropriate.

I would look from additional posts from David or Peter, as they may have suggestions or corrections.

HTH,

Kevin

Kevin Viel, PhD Post-doctoral fellow Department of Genetics Southwest Foundation for Biomedical Research San Antonio, TX 78227