Bengt Holmström Prize Lecture: Pay for Performance and Beyond

430 

The Nobel Prizes

Several scandals are associated with the introduction of test-based incen-

tives for teachers. One of the most notorious ones occurred in the Atlanta public 

school system, where the results on the Competency Tests showed irregularities 

in test scores (extraordinary gains and losses in a single year). An investigation 

revealed that 44 out of 56 schools cheated on the 2009 Competency Test, and 

178 teachers and principals were found guilty of correcting student answers. In 

2015, 11 teachers were convicted of racketeering and received stiff sentences, 

including jail terms.

The scandal was an unintended consequence of the No Child Left Behind 

Act of 2001 and illustrates how good intentions can go wrong. As in the case of 

Wells Fargo and many similar cases, those guilty of wrong-doing describe their 

acts as the result of feeling excessive pressure to perform.

The debate about the costs and benefits of testing and rewarding teachers 

and schools is ongoing (Neal 2011). The issue is less about cheating and more 

about other costs such as teaching to the test. Like Stephen Kerr (1975), in his 

well-known article “On the Folly of Rewarding A, While Hoping for B,” the mul-

titasking model warns against aggressive use of rewards for easy-to-measure sub-

jects like mathematics at the expense of developing important social and other 

hard-to-measure skills. But economics goes beyond Kerr’s criticism of rewarding 

narrow goals by studying the many substitute instruments that firms can use 

quite effectively. I will return to these after summarizing the main lessons from 

the multitask problem.

17

D. Low-Powered Incentives

A first lesson from multitasking is that there are two ways to provide incentives

for a hard-to-measure activity like supply of quality. The direct way is to pay for 

measured quality, however poor the measure is. The indirect way is to reduce 

the incentive on quantity, because this reduces the opportunity cost of providing 

quality. How effective the indirect way is depends on how poor the quality mea-

sure is and how strongly the two tasks compete with each other for the agent’s 

time. In the pure time-allocation case the tasks compete directly with each other 

for the agent’s attention, giving rise to a maximal substitution effect.

A second lesson is that measurement problems can explain why even per-

fectly measured activities may be left without any incentive. The fact that the agent 

can control a measure well, because the measure is accurate with respect to a 

task or a subset of tasks, does not imply that it should be used. Sometimes the 

exact opposite is true. Employees tend to like accounting measures that they can 

control such as costs, but these measures are often partial and therefore poorly 

---

Pay For Performance and Beyond 

431

aligned with true value creation, in which case low-powered incentives on these 

measures are called for.

The third lesson is the most general. Because the opportunity cost of pro-

viding incentives for any given task depends on all the incentives for the other 

tasks, one should always consider the full portfolio of agent tasks when designing 

incentives.

V. ALTERNATIVE INCENTIVE INSTRUMENTS IN FIRMS

Firms use pay-for-performance schemes rather sparingly. Most employees are on 

fixed pay. Williamson (1985) offers one explanation: the integrity of performance 

measures within the firm is weaker than in the market. Multitasking offers a 

complementary explanation. Firms have access to many substitutes for pay-for-

performance incentives that are not easily accessible through market contract-

ing. Foremost among them is the ability to control work through job assign-

ments, job designs and a variety of implicit and explicit rules that the firm sets.

A. Rules and Constraints

Consider the case of working at home versus coming to the office.

18

 Assume that 

the task that the agent is assigned—one task for simplicity—can be done just as 

easily at home as in the office. The only difference between the two options is 

that the agent has other activities that he can engage in at home, activities that 

provide private benefits. He can work on other projects at home, do sports, watch 

TV, rest, play with his kids and so on. We can think of these privately valuable 

activities as offsets against the cost of working for the principal. The downside is 

that they reduce the time the agent can spend on the principal’s task.

If the agent were paid the value of the output he produces, there would be 

no problem working from home. As a residual claimant, the agent would do 

the socially correct calculation on how much time to spend on private activities 

and how much on the principal’s task. But this contract would also force the 

agent to bear risk. If risk is costly and the incentive power therefore reduced, 

the agent’s commission will not fully reflect the marginal value of his activities. 

The principal’s task will, as a result, become less competitive against the private 

activities. Suppose each of the private activities can either be shut down or the 

agent will be free to pursue the activity as he wishes—a zero-one decision for 

each activity—which activities should be allowed and which ones shut down? 

Note that allowing private activities is a form of compensation for the agent. So, 

the principal will also benefit from letting the agent undertake an activity if the 

---