The Multitask Problem and the Agency Dilemma

Holmstrom and Milgromi48 identified the multitask problem, which can arise in the context of the “principal-agent problem” of economic theory.

You are an “agent” when you are commissioned to act in someone else’s interests. That other person is the “principal.” You are an “agent” for your employer, for example. A tort lawyer is an “agent” for his or her client; a factory worker is an agent for the factory owner. Economists have long recognized a “principal-agent problem.” The canonical model is that of Ross49 who framed the problem from the perspective of the principal. How do you get the agent to do what you want when you cannot observe everything about the agent’s choices and actions? You might “monitor” the agent. The boss, for example, may pop in to see if the employee is working or napping. But it may be costly or even impossible to observe all the agent’s choices. Effort is hard to observe. How do you know, for example, whether your tort lawyer really worked as many hours as you were billed for?

Instead of trying to monitor inputs such as hours worked or intensity of effort, the principal might monitor outputs such as the number of parts produced or the size of a tort award. If outcomes are clearly observable, then you might be able to induce the right outcome by giving the agent a cut of the action. That is, you can link the agent’s pay to outcomes. Thus, tort lawyers typically charge a contingent fee; they work on commission. The same solution to the principal-agent problem is applied when factory workers are paid at piece rates. More generally, the principal should often link the agent’s pay or fees to measurable outcomes. Doing so may align the agent’s incentives with the principal’s desires. There are some difficulties even when outcomes are observable, given possibilities such as different risk preferences between agent and principal. Because such issues do not seem to affect our argument, we will ignore them here.

The standard solution looks to the principal-agent problem looks for “high-powered incentives” to leverage a good outcome from the self-interest of the agent. Paying the worker per piece will inspire a greater effort. Paying the tort lawyer a contingency fee ensures that he or she has the same desire for a big win as you do. This standard result of the economist’s benchmark principal-agent model -- cut the agent in on the action -- is fine as long as outcomes are observable. The multitask problem arises when outcomes are not fully observable. The multitask problem arises when the principal can observe some outcomes, but not others. In that case, the use of high-powered incentives can backfire. The agent focuses on observable dimensions and ignores the rest of the job. When tenure committees reward “research” in the form of published articles, for example, junior faculty have an incentive to skimp on teaching in order to publish more. If outcomes have measured and unmeasured dimensions and if the measured outcomes are rewarded and unmeasured ignored, then agents will have an incentive to improve measured outcomes and neglect unmeasured outcomes. If the unmeasured dimension matters to the principal, the use of high-powered incentives to motivate performance will backfire. Singers who are paid by the decibel will sing loudly and off key.

Holmstrom and Milgromi50 provide an example that is still timely today. They note the “current controversy over the issue of incentive pay for teachers based on their students’ test scores.” Proponents hope “these incentives will lead teachers to work harder at teaching and to take greater interest in their students’ success. Opponents counter that . . . teachers would sacrifice such activities as . . . refining students’ oral and written communication skills in order to teach the narrowly defined skills that are tested on standardized exams.” Citing Putkaii51, they note a case in which “a ninth-grade teacher . . . was caught having passed answers . . . to students . . . in order to improve her performance rating”.52

More recently, a major cheating scandal in Atlanta has emerged. [...] Atlanta Public Schools (APS) personnel from high levels possibly including the superintendent down to principals and teachers were involved in activities such as erasing and correcting mistakes on student answer sheets53. The cheating seems to have been driven by high-powered incentives imposed from without. The federal No Child Left Behind Act, which required states to establish objective outcomes measures, may have added to the pressure. But the cheating began at least as far back as 2001, shortly before the federal Act came into effect.

The schools were under pressure to achieve performance targets as measured by student test results. “Because the targets rose each time a school attained them, the pressure ratcheted up in classrooms each year. Cheating one year created a need for more cheating the next”.54 By the time the scandal broke in 2011 the cheating had grown into a vigorously enforced system. At least some teachers were afraid to resist the pressure to cheat. “‘APS is run like the mob,’ one teacher told investigators, saying she cheated because she feared retaliation if she didn’t”.55 According to one state report on the scandal said, “APS became such a ‘data-driven’ system, with unreasonable and excessive pressure to meet targets, that [the system’s superintendent] Beverly Hall and her senior cabinet lost sight of conducting tests with integrity.”56


48.      Bengt Holmstrom and Paul Milgrom, Multitask principal-agent Analyses: Incentive Contracts, Asset Ownership, and Job Design, 7 J. L. ECON. & ORG. 24 (1991, special issue).

49.      Stephen A. Ross, The Economic Theory of Agency: The Principal’s Problem, 63 AMER. ECON. REV. 134 (1973).

50.      Holmstrom and Milgrom supra note 49, at 25.

51.      Gary Putka, Classroom Scandal: Cheaters in Schools May not be Students, but their Teachers, at 1, Wall Street Journal, Nov. 2, 1991, available at Proquest (last visited Oct. 7, 2011).

52.      Holmstrom and Milgrom supra note 49, at 25.

53.      See Heather Vogell, Investigators into APS Cheating Finds Unethical Behavior Across Every Level, Atlanta Journal-Constitution, July 6, 2011, available at http://www.ajc.com/news/investigation-into-aps-cheating-1001375.html (last visited Oct. 7, 2011).

54.      Id.

55.      Id.

56.      Id.

-- Roger Koppl , Meghan Sacks

from "The Criminal Justice System Creates Incentives for False Convictions"

Quoted on Sun Sep 22nd, 2013