It's not easy research, but there have been some papers published on the topic. The problem of course is that there are a lot of confounding factors.
Besides, that my point, isn't it? Promotors of TDD argue that their personal intuition suffices, despite having little evidence to back their claims.
You would need to compare teams of equal skill level who were working on very similar software over quite a long period of time.
What would the benchmarks be? Number of defects in production code? Time to ship? Severity of defects? Time spent debugging?