A scientific approach would be to have a large number of program requirement documents and a large number of programmers, and assign a methodology to half the p

A scientific approach would be to have a large number of program requirement documents and a large number of programmers, and assign a methodology to half the programmers, forbid the method to to the other half, and then measure the output from the exercise along important axes like time to completion, correctness, and compute resource consumption. This has been approximated in studies a small number of times, but with too small sample sizes. Achieving robust results would be enormously expensive.