Psychologists have long argued about the power of priming, i.e the power of subtle cues and reminders to influence behavior. For instance, there are a number of academic papers that find that if you make a woman write down her name and circle her gender before taking a math test, she will do substantially worse than if she just writes her name. The idea is that women perceive that they are not good at math, and circling their gender reminds them that they are women and therefore should be bad at math. I’ve always been skeptical of these results (and indeed failed to replicate them in one study I did with Roland Fryer and John List) because gender is such a powerful part of our identities that it’s hard for me to believe that we need to remind women that they are women!
In an interesting new study, “Bad Boys: The Effect of Criminal Identity on Dishonesty,” Alain Cohn, Michel Andre Marechal, and Thomas Noll find some fascinating priming effects. They went into a maximum security prison and had prisoners privately flip coins and then report how many times the coin came up “heads.” The more “heads” they got, the more money they received. While the authors can’t tell if any one prisoner is honest or not, they know that on average “heads” comes up half the time, so they can measure in aggregate how much lying there is. Before the study, they had half the prisoners answer the question “What were you convicted for?” and the other half “How many hours per week do you watch television on average?” The result: 66 percent “heads” in the treatment where they ask about convictions and “only” 60 percent “heads” in the TV treatment.
How dishonest are prisoners versus everyday people? When they play the same game with regular citizens, the coin comes up “heads” 56% of the time.
So how powerful is that one question on conviction? The behavior of the prisoners who are asked the TV question is actually closer to that of regular citizens than it is to the behavior of the prisoners who were primed.
As an economist, I hate the idea that priming might work. As an empiricist, I guess I better get used to it.