This is exactly the line of reasoning I wanted to suggest; operant conditioning, as with all forms of learning, has nonlinear results below a certain threshold

This is exactly the line of reasoning I wanted to suggest; operant conditioning, as with all forms of learning, has nonlinear results below a certain threshold of involvement. We're not good at expectation values, especially in these cases: 1. Rare events, like receiving a speeding ticket, are relatively underweighted. Every time you drive without receiving a ticket reinforces speeding behavior. 2. Extremely salient events, like winning the lottery, tend to be overweighted--possibly because they're so important to our cognition, and we tell ourselves plausible stories about winning. There's also a difference between negative and positive reinforcement and punishment. Receiving a speeding ticket is positive punishment; the lottery is positive reinforcement. Couple that with differential media coverage, like you mentioned, and you get significantly different behavior.