occupancy as response variable in logistic regression

[jlaufenb]

Thanks, Evan. That's very good to know.

[dhewitt]

Evan,

I came to this thread late, but as I read through I could not tell which approach you were advocating. You seemed to indicate that stdized reg coeff are as good as it gets, but then suggested that the specific approach to doing that on the data set in question was a no-no. Did I get that right?

Anyway, B&A spend a fair amount of time on methods for relative importance and it has been dealt with more in the literature since. Personally, I thought M&C was not helpful and mixed the purpose of AICc model selection with other methods with narrower purposes (not to mention some odd choices of simulation specs).

- Dave

[cooch]

Evan,

I came to this thread late, but as I read through I could not tell which approach you were advocating. You seemed to indicate that stdized reg coeff are as good as it gets, but then suggested that the specific approach to doing that on the data set in question was a no-no. Did I get that right?

1\ I think that as long as the model set is balanced with respect to the factors you're interested in, then the 'cumulative evidence' approach B&A talk about is fairly robust.

2\ I don't favor doing 'statistics on statistics', for any number of reasons. If you want to consider a putative relationship between a covariate an occupancy (or anything else), build the corresponding ultrastructural models reflecting that relationship, and see 'how well they do' relative to models without the relationship.

[dhewitt]

Then I find myself in full agreement with you, sir. :)