It seems like the confounding factors are enormous to testing this sort of thing. I also think it's problematic that those studies were using students. Strong typing becomes really beneficial after you begin to understand the flow and architecture of code, and your codebases become larger than a simple Project Euler problem.