Punishment Without Cause: Regression and the Effects of Leader Attribution Errors1

Kahneman and Tversky (1973) observed that failure to understand simple regression to the mean could cause leaders to falsely learn that punishment was more effective than reward in shaping subordinate performance. This fundamental hypothesis and other attribution theory hypotheses were tested in a field experiment in which subordinate performance was entirely random. The basic hypothesis was partially supported, but leader belief about subordinate ability was found to moderate leader reinforcement behavior. Leaders who believed that their subordinates had high ability subjected them to increasing amounts of punishment over time, while reward amounts remained relatively unchanged. The opposite pattern was evident but not statistically significant for leaden who believed that their subordinates were of low ability. Rewards were decreased over time, while punishments were left unchanged.