"Slightly in favor of typechecked (he typed smugly). But that's not really enough. It's not clear there's even a 5% bump in productivity."
One question would be "does it compound?" A "5%" advantage (leaving aside how we could even define something that precisely) that compounds every 1000 lines is hard to detect in any reasonable study but could have an astounding effect by the time you've written a 50,000 line application. Even a 1% advantage compounds quickly to significant differences at scale, yet in the small will be virtually impossible to detect as it will be more than swamped by the noise floor.
Part of the problem with these studies is that yes, programmer skill probably does dominate at the smaller end of the scale. I've come to really not like dynamic typing for the sort of work I do, but I still can and do crank out 500-line Perl scripts at times, because if that really is all I need, static vs. dynamic typing aren't the question. It's about how comfortable I am, how comfortable my coworkers are, what kind of library support I need, and a lot of other considerations long before I worry about typing. By contrast, putting together a multi-dozen-thousand line codebase leads me to entirely different considerations in my language choice.
My summary of the summaries is that we still have virtually no hard evidence to go on that has any utility or relevance to real-world software engineering. At best we have some suggestions about how we might teach students how to do their tiny school assignments and a couple of interesting-but-inconclusive case studies on code of real size.