Well I have to cop out from this one a bit, it was one of the central points of the course on criminal psychology I took as part of my law degree but I don't have references at hand. My books from that time are in storage because I'm moving.

Of course one can't quantify exactly how much several factors come into play when making a decision, but basically the overall conclusion from experiments was: as long as the severity of the punishment outweighs the benefits of the crime (e.g. a fine of at least 11$ for stealing 10$), that severity doesn't matter very much any more. So punishing all theft by death is only a marginal deterrent. The policy focus should be on increasing the chance of getting caught. (death penalty is an exaggerated and probably silly and untrue example, but the fact that people still steal in places with Shariah law which punishes theft by loss of a hand shows that extreme punishment is not an absolute deterrent).

But yeah - I can't point you to the literature. Coincidentally I read in the news paper a few days ago that my prof from that course is caught up in a scandal about a book on an unsolved murder and hasn't been heard from for 2 weeks, so I can't really expect he'll answer an email about it either ;)