I don't know anything about raising kids, but I know about dogs and machine learning. The similarities you can draw between the two are pretty funny. Reward leads to learned behavior being more generalizable. With punishment, at best they correctly understand "if I do X in Y situation, I'll be punished for it." More likely, they learn "if I do X in Z situation, I'll be punished for it," where Z might be some incorrect thing. With reward, they learn, "if I do X, I MIGHT be rewarded for it," and the situation is less of a factor. That's the way it seems to play out, anyways.
In ML, it can be sort of thought of a "reinforcement learning" (reward-based) vs. "evolutionary learning" (punishment-based), and RL seems to kick the shit out of evolution.
