Microsoft (Research?) published some stuff in the last year or two that looked really interesting. I don't have a link at the moment and I'm on my way out the door, unfortunately. But I remember thinking that it had promise because it drew conclusions from a huge base of data from actual Microsoft developers working on long-term projects.
Another paper I like is Little's report on data from projects at Landmark Graphics [1]. It's not a controlled study, and shouldn't be treated like one, but it also looks at a large base of data from real projects.
My biggest complaint about academic papers is that they're often conducted with students, using practices out of context, working on projects of trivial size and depth. I like both of the above not because they're perfect, but because they get beyond those problems.
Another paper I like is Little's report on data from projects at Landmark Graphics [1]. It's not a controlled study, and shouldn't be treated like one, but it also looks at a large base of data from real projects.
My biggest complaint about academic papers is that they're often conducted with students, using practices out of context, working on projects of trivial size and depth. I like both of the above not because they're perfect, but because they get beyond those problems.
[1] http://www.toddlittleweb.com/Papers/Little%20Cone%20of%20Unc...