Hi there l-dawg,
I wonder if you had answers to your questions, coz I am now in the same state u were a year ago (or so). I, too, have a logistic regression problem trying to "predict" obsessionality from various clinical and epidemiological parameters, but it's hard to 1) choose which ones to include in my model, 2) which ones to chuck out while pruning. Did you get any further to what you describe above?
I've tried to compare different models, so I formed some 20 different models and then run R square F-tests between them to determine the "best" one. Unfortunately, almost invariably that turned out to be the model with the largest R-squared
Any help will be hugely appreciated, please.