The rules of thumb specified there contain no comment about estimation with covariates. In my models, I use categorical covariates quite often, one covariate most of the time.

So I wonder, when considering the number of degrees of freedom, should one base it on the total sample size, or on the average number of respondents belonging to a category of the covariate?

For example, with a total sample of N=1200 and 4 nearly equally sized categories of the covariate, is it better to go for

d.f.=50 (N=1200) or

d.f.=5 (N=300)?

P.S. I have many parameters in the model, so it would take a tremendous amount of time for Model Explorer to do the search.

Thanks,

Igor