presented at the 2003 North American Users' Group meeting.
That may answer one of your questions.
I can't answer the first question, though I'm curious about it. I'd
like to know what happens in these models where cases and controls are
assumed to be correlated based on matching criteria, when in fact they
are not (or not very because the matching criteria is skimpy or
unrelated to the disease process). When models are used on correlated
data where the model misses that fact, bad things happen. Do bad things
happen the other way round?
