Second the endorsement, but it's important to understand that the relevant average is the particular animal's current baseline (not, for example, the average behavior of similar individuals). That is, you look at the behavior you're getting, and you reward the top N% of behavior along whatever dimension you want to reinforce. Over time the cluster of behaviors will shift in that direction. Keep rewarding the top N% and it will keep shifting in that direction until other factors make that impossible.