Although not central to his work, one of my favorite papers of today’s Nobel prize winner, Jean Tirole, is Extrinsic and Intrinsic Motivation (written with Roland Benabou). In this paper, Tirole and Benabou try to resolve the economist’s intuition that incentives motivate with the idea from psychology that incentive schemes can sometimes demotivate. The psychologists argue that extrinsic motivation can reduce intrinsic motivation (but they are not at all clear on why this should be the case). Tirole and Benabou try to produce a similar finding by arguing that in addition to providing motivation an incentive scheme gives the agent, the one being incentivized, some information and the information may undermine the motivation.
For example, if I tell my son. “If you get an A in math, I will give you $1000.” What does my son conclude?
- My father must think math is very important for my future to offer me $1000. My father is smart. I will work hard.
This is the message that I hope to send. But my son knows that I know something about math and also that I know something about him and he may use this knowledge to make a very different inference.
- If my father thinks I need $1000 to get an A, math must be very hard or I must lack talent. I will work for an A this year but next year I should probably not sign up for advanced math classes.
Or perhaps he infers
- If my father is offering me $1000 to do the right thing , he must not trust my judgment.
- My father is trying to use his money to control me. I rebel!
Thus reward has two effects a pure incentive effect (holding information constant) and an inference effect. Notice that the inference effect depends on the context. Thus, without knowing the context–how the father gets along with the son and their history of interaction–we can’t know what the effect of the “incentive” will be. Thus I have argued that “an incentive is not an objective fact but a subjective interpretation.”
I’m not convinced that Tirole and Benabou have the right answer on intrinsic and extrinsic motivation but thus and other papers indicate Tirole’s broadness of thought and his characteristic approach to issues.
By the way, working out the equilibrium in these games is not at all easy because the principal knows the agent will infer information about the characteristic from the reward structure and the agent knows the principal knows that the agent will infer information about the characteristic from the reward structure and so on – thus we have a Moriarty problem and must look for conditions such that there can be an equilibrium in which everything is common knowledge. But heh, if these problem were easy you wouldn’t get a Nobel prize for solving them!
See Tyler’s post and my other posts below for much more on Tirole.