Agent-based simulation of work teams: under close monitoring, punishing and correcting managers produce nearly identical error counts (99.5 vs 99.2), but punitive teams are five times likelier to lose at least half their workforce (0.15 vs 0.03)
Simulating what happens when managers punish mistakes rather than correct them (my paper, out today in JASSS) Disclosure up front: this is my own paper, published today and open access.