Bengt Holmström Prize Lecture: Pay for Performance and Beyond

432 

The Nobel Prizes

total surplus is increased in that way. Given a fixed commission, if a private task 

is shut down, the agent will spend the time freed up on the principal’s task. So, 

whether the task should be shut down is just a question of whether the agent’s 

chosen amount of a private activity, given the commission rate set by the princi-

pal, will be worth more when spent on the principal’s task.

The lower the commission, the less time the agent will spend on the princi-

pal’s task and the more on his private tasks. As a result, the marginal value of the 

principal’s task goes up while the marginal values on the agent’s private tasks go 

down as the commission rate goes down—the agent is spending too much time 

on private tasks and too little on the principal’s. So, excluding private activities 

becomes more important, the lower the commission rate. Constraints on private 

activities act as substitutes for monetary incentives by increasing the amount of 

time the agent spends on the principal’s task. In this example, the restrictions 

are placed in decreasing order of importance of the tasks (as measured by the 

average private value of tasks).

The agent’s incentive will be weaker if the performance in the main task is 

measured with lower precision (the variance of ε is higher) and this will imply 

that more tasks will be excluded in order to substitute for the weaker incen-

tives.

19

 Responsibility and freedom to choose co-move as the precision of mea-

surement varies.

20

Beyond task restrictions, other instruments can also be important. For exam-

ple, bureaucratic rules in organizations are as prevalent as the complaints about 

them. Bureaucracy is detested because it discourages initiative and innovation. 

Too often people fail to see that there is a purpose for most bureaucratic rules. 

In the first instance, these rules are not symptoms of an organizational disease, 

but rather part of the solution to complicated incentive problems. Because pay-

for-performance tends to be a costly incentive instrument in the firm, due to 

problems with performance measurement, bureaucratic rules and restrictions 

will often serve as effective substitutes (Holmström 1989).

Firms also place a number of other restrictions on its employees that limit 

what they can do. One of the most obvious and universal rules (until recently, 

perhaps) is the requirement that an employee cannot work for other firms. 

Another common requirement is that employees have to come to the office for 

set hours. Naturally, these constraints are driven by other considerations, too, 

but their effect on employee incentives is nontrivial.

B. Job Design

Another important dimension of job design is the distribution of tasks 

between workers.

---

Pay For Performance and Beyond 

433

Go back to the two-task case with one agent. One task is perfectly measured 

(a routine task) and the other task is measured with considerable noise (an inno-

vative task). The agent’s cost function depends only on total time spent. I con-

cluded that providing low-powered incentives on the routine task is appropriate 

as an incentive for innovation.

What if one could split the two tasks between two agents? One agent takes 

care of the routine task while the other specializes on the innovative task. This 

avoids the tension between routine and innovative tasks. The routine task can be 

provided with high-powered incentives while the innovative task has low-pow-

ered incentives. More generally, one can show (Holmström and Milgrom 1991) 

that if there is a continuum of tasks (to avoid integer problems) with varying 

measurement costs, then one should give all the hard-to-measure tasks (defined 

by a cutoff) to one agent and the easy-to-measure tasks to the other agent. The 

hard-to-measure tasks are provided with equal low-powered incentives, while 

the easy-to-measure tasks are provided with equal high-powered incentives. The 

more risk-averse agent takes on the low-risk tasks.

More specialization may be one way to deal with the teaching-to-the-test 

dilemma that is so hotly debated. The idea has been promoted by Neal (2011) 

based on the multitasking model. There is of course already a significant degree 

of specialization in teaching, but one may want to increase as well as shift the 

responsibility of soft skills to teachers that specialize in these. These teachers 

should not be held responsible for narrow test results.

There is an important additional lesson hidden in the logic of specialization. 

In the model discussed, no task is shared because that would require incurring 

risk for both agents in the given task. If one agent is already incentivized to do 

half of the project, incurring the risk cost associated with adequate incentives, 

there is no additional risk cost to have him do the whole task. While this result 

depends partly on our additive production function and the continuum of tasks, 

the logic is clear. There is a fixed cost to having a second agent join the project, 

because both have to be incentivized. This feature is of more general validity and 

may be one reason for individual responsibility (unity of command).

The same principle is relevant for job rotation. One benefit of job rotation is 

that it offers a form of relative performance evaluation: having two agents work 

in the same task gives information about the difficulty of the task and this is 

valuable for evaluating each agent’s performance in line with the informativeness 

principle. But there is also a downside from an incentive point of view. If perfor-

mance information comes in with a lag, one wants to make the incentive of the 

first agent depend on the results coming in during the second agent’s shift. This 

is similar to two agents sharing the same task as discussed above. The first agent 

---