> My point was that the selection bias proposed will not necessarily obscure a biological effect (or at least, correlation).

I'd argue that the sample was not representative of the general population, so "it will not necessarily obscure" isn't relevant IMO because the results aren't even remotely close to being applicable to any other group of men.

> looking for effects using statistical analyses of medical records is an interesting way to direct at least a small amount of research

I agree that it's interesting. And since I can't read the paper at the moment (I'll download a copy through my university when I get a chance) I can't really comment on the statistical methods they used (I'm not even sure what they asked on the questionnaire). Statistical analysis is only useful if you take into account all of the parameter space (or at least all of the significant parameter space) otherwise you'll get a biased result.

Even then, correlation does not imply causation, so a statistical analysis would only be useful as a survey of "what could be investigated to do actual science on". People put far too much faith into statistics, and it

> saying that such questions can only be asked when there is solid control of every factor is probably not a win

Anyone can ask a question. I'm saying that the paper published should've actually been an email (or maybe just an article) sent around a faculty so they could do real research on the topic. "Asking a question" and "publishing a paper on a topic" are very different classes of activities.