A principle stating that a high-probability behavior can be used to reinforce a low-probability behavior.