There's the field of empirical computer science (for example http://www.cs.jhu.edu/~nasmith/erm/) - and I've seen a couple of articles recently pushing for a more evidence-based approach to methodology. No links, I'm afraid - can't find any - but it is out there.

Edit: the link I included is more to do with algorithms and the like, mea culpa, but there are also studies cited in books like Code Complete that cover what you mention. How recent, or how good the research is, is another matter...