Unexpected Costs of Performance Incentives with Production Uncertainty : Theory and Evidence from a Real-Effort Laboratory Experiment

Policy advocates often push for market-like performance incentives for publicly provided services such as in education or healthcare (as with Medicare in the U.S.). The evidence supporting output-based, performance incentives for such jobs is mixed at best. One possible explanation is that the production process for these jobs is characterized by production uncertainty, mathematically represented as uncertainty in the marginal product of inputs. In a stylized principal-agent model, I show how production uncertainty can make output-based incentives ineffective or even detrimental to employee productivity. If high-impact inputs also have highly variable marginal productivity, employees will favor ineffective inputs as piece-rate bonuses grow. In such cases, input-based incentives may prove more effective. This paper provides preliminary evidence from a real-effort laboratory experiment that workers do indeed shift away from high-variance inputs even at the expense of their average pay. Participants are paid based on the number of easy or hard math questions they successfully answer. During treatment, the marginal payoff for each question type is randomized. I show participants shift away from their efficient question type as production uncertainty grows for that type