It should be simple to develop a predictive model of programmer effectiveness. We can measure the number of lines of codes, number of defects committed, number of defects fixed, etc. against the average developer on the average software project and then we have a way of approximately evaluating a programmer's contributions to a software project.
Good point. Would you agree that: even if a phenomenon (for example performance of some sort of worker) is hard to model, we should still try, especially if that phenomenon provides enormous value to society?
Science is really hard.