Colin Camerer (Cal Tech) & Robin Hogarth (Univ. of Chicago) said:
We reviewed 74 experimental papers in which the level of financial performance-based incentive given to subjects was varied. Our primary interest is in advancing the simmering debate in experimental methodology about when subjects should be paid, and why. The data show that incentives sometimes improve performance, but often don’t.
This unsurprising conclusion implies that we should immediately push beyond debating the caricatured positions that incentives always help or never help. Adopting either position, or pretending that others do, is empirically misguided and scientifically counterproductive.
In our view, the data show that higher levels of incentives have the largest effects in judgment and decision tasks. Incentives improve performance in easy tasks that are effort-responsive, like judgment, prediction, problem-solving, recalling items from memory, or clerical tasks. Incentives sometimes hurt when problems are too difficult or when simple intuition or habit provides an optimal answer and thinking harder makes things worse.
In games, auctions, and risky choices the most typical result is that incentives do not affect mean performance, but incentives often reduce variance in responses. In situations where there is no clear standard of performance, incentives often cause subjects to move away from favorable `self-presentation’ behavior toward more realistic choices. (For example, when they are actually paid, subjects who dictate allocations of money to others are less generous and subjects choosing among gambles take less risk.)
One way to comprehend these results is a "capital-labor-production theory" of cognition (extending Smith and Walker, 1993). The capital-labor-production framework assumes that the 18 `labor’ or mental effort subjects exert depends upon their intrinsic motivation and financial incentives.
But the effect of extra effort on performance also depends on their level of cognitive `capital’-- know-how, heuristics, analytical skills, previous experience in the task, and so forth-- and its productive value for a specified task. Capital and labor can substitute: For example, a few experiments suggest that one session of experimental experience has an effect roughly comparable to (at least) tripling incentives.