Association for Public Policy Analysis & Management

Handling the educational needs of a diverse student body in preparation for an increasingly globalized labor market will require creating a robust teacher labor force that can sustain the specific needs of at-risk subpopulations. Yet, how to compensate teachers to reflect productivity differences and provide incentives that reward performance remains a practical challenge, as many performance incentive schemes have not demonstrated consistent efficacy. Recent large-scale U.S. initiatives, including Race to the Top and the Teacher Incentive Fund have placed substantial emphasis on the development of compensation mechanisms to link pay and performance, as these programs have awarded a combined $6.4 billion since 2010 to 92 districts in 32 states for proposals to fund new teacher performance incentives or improve teacher accountability. Internationally, the United Kingdom, India, Chile, Mexico, Israel, Australia, and Portugal are considering or have implemented teacher incentive programs.

The results of teacher performance incentives have been mixed. While large scale field experiments and policy innovations are surely needed in the teacher labor market, the design challenge resides in the more general space of models for incentive contracts. A necessary theoretical and empirical problem is how to design an incentive to induce the optimal allocation of effort among multiple tasks, which is usually modeled by assuming agents know the production function. Unlike some production processes in which output relies solely on worker skill and effort, teaching is distinguished by its complexity and its dependence on the reciprocal effort of students. The result is a context in which individual teachers are uncertain about the net marginal productivity of inputs. The innovation of this paper is to develop a model in which uncertainty about the production process in student learning (“production uncertainty”) is incorporated explicitly in the model and to assess agent responses to different incentive schemes in a laboratory experiment.

In the model, the presence of production uncertainty reduces the effect of an outcome-based incentive on a teacher’s overall effort level due to risk aversion, an effect I label “futility.” Furthermore, an outcome-based incentive can induce production friction, which predicts that teachers will redistribute their effort to inputs with lower variance in marginal productivity. This implies teachers could maintain or even increase their level of effort in response to a performance incentive without improving otucomes.

The initial empirical test of these predictions is in a laboratory experiment. The experimental design itself is an innovation, allowing me to test a complex, multi-input real-effort task with different types of uncertainty. Preliminary results confirm my model’s basic predictions, though more experimentation is planned.

Optimal contract design in the principal-agent problem influences how policy makers have tried to improve education. My theory, confirmed by preliminary lab results, has several policy implications. (1) Teacher incentive schemes should focus on rewarding measured behaviors that are within the control of teachers. (2) Incentives based only on student test scores can reduce average output even if teachers are not reducing overall effort. (3) Policy makers should focus on creating reliable processes for conducting in-class evaluations.