The Toxicity of Pay for Performance

by Donald M. Berwick

Despite their superficial logic, systems of merit pay or pay for performance have features that are toxic to systemic improvement. Contingent rewards doled out by supervisors cause decreased focus on customer needs, loss of accurate information about defects and improvement opportunities, avoidance of stretch goals, and decreased innovation. They may also erode teamwork. Pay for performance may mark a naive understanding of the complexity of human motivation.

She sat across the table from me, all ears–one of my very best employees. It was time for her
annual merit review, and, according tothe organization’s policies, she was to be rated as “unsatisfactory,” “satisfactory,” “superior,” or “outstanding” and receive a salary boost of zero, 4 percent, 5 percent, or 6 percent,

I had prepared carefully. Her work was promising indeed. Although she lacked formal training, she
had clear leadership potential and enormous native talent for planning and organizing projects. To
excel further, she would need to work on her writing and presentation skills, and some further
study of simple statistical methods would strengthen her technical analyses. Tactfully,
respectfully, pausing regularly for her questions, I explained my review and offered her guidance.
She asked no questions but only nodded agreement with a friendly smile on her face.

I ended with a review of my advice and a handshake of congratulations for her fine efforts. I
paused for her reflections.

” So,” she asked, beaming–her first words since I had begun, “which is
it … 4 percent, 5 percent, or 6 percent?”

It took me a few moments to absorb the full import of her question. But, finally, I saw the truth:
she had not heard a word that I had said. “Four percent, 5 percent, or 6 percent?” That was her question,
and that–and that alone–was what
she was waiting to hear from me.

So embedded in our culture is the idea that “you get what you pay for,” so familiar are the
assumptions behind “pay for performance,” so fair and obvious does it seem that people should be paid for their worth in American culture, that it may take a sledgehammer to ring a note of question. Indeed, in many
corporate cultures, and most that I have worked in, to raise fundamental questions about these
assumptions is inevitably to invite accusations of naivete and inexperience.

Linking pay to merit is an absolutely obvious instrument of proper management. Because it is
absolutely obvious, it is difficult in the extreme to see that it is very nearly absolutely wrong.
“Pay for performance” is as toxic to true organizational performance as any of the perfidious
tactics of outmoded control-based management that enlightened organizations have long since, and
much more readily, abandoned.

It is not necessary to recount here the classical arguments in favor of merit pay. They fall
generally into two categories: arguments of fairness (good performance deserves its reward) and
arguments of incentive (pay contingent on good performance generates more good performance).

In passing, we might note how thin is the empirical evidence for either of these arguments. There
is little but logic or anecdote behind either assertion; in the entire volume of “classic” Harvard
Business School papers assembled in a collection called “Appraising Performance Appraisal,” only a
single experimental study appears. 1 But, as it happens, little real evidence can be found on
either side of the “pay for performance” debate. (Alfie Kohn, a popular author on management
systems of reward and incentive, claims that the evidence that does exist weighs heavily against
contingent pay as a support for organizational or group effectiveness.2 ) Indeed, it is remarkable
that an issue of such consequence for the guidance of organizations and so hotly contested from
time to time has not benefited more from proper social-scientific experimentation. We who debate
pay for performance on either side must fall back on belief and logic for now for the bulk of our

Some clear definitions, first, may help. Our topic here is “pay for performance,” a
contingent relationship, enforced and implemented in an organizational hierarchy, in which
supervisors judge the merit of the work of those below them in the hierarchy and, based on those
judgments, give out variable and contingent financial rewards. “Merit” can refer to meeting any
goals or standards, whether negotiated in advance with the employee or announced arbitrarily,
whether financial or nonfinancial, whether specific or ambiguous, whether quantitative or
qualitative, whether measured in terms of throughput (e.g., “patients per day”) or internal process
(e.g., “reliable”). “Rewards” (and their opposite, “punishments”) can be purely financial (such as
salary increases or bonuses), or they may consist of other forms of organizational currency (such
as promotion or perquisites). A key and essential component of “pay for performance” is the notion
of contingency, the “merit” and the “reward” are linked explicitly, and the ultimate judge of both
is the supervisor. Paying assembly-line workers on the basis of their productivity is “pay for
performance”; so is paying individual doctors on the basis of their “clinical performance,” however

No dispute exists here about the value and appropriateness of dialogue between workers and
supervisors about goals and performance. It is useful–indeed essential–in an effective
organization for information to flow freely about the extent to which needs are met, customers are
satisfied, plans are implemented successfully, and goals are achieved. For many purposes,
quantitative information is clearer and more instructive than mere narrative, and, accordingly, no
dispute exists here about the great value of measuring performance in many dimensions and at many
locations in the sequence of production.

Nor is there dispute here about the probability of and need for variation in compensation among
individuals. There is a market for talents and experience, and an organization that tries to ignore
the conclusions of that market will have trouble recruiting and retaining the people who can help
it most. Changes in the cost of living, issues about the sharing of organizational profits, and
employee demands that their compensation increase in accord with their skills will always be facts
of organizational life, and compensation will therefore vary among people and over time.

No dispute exists here either about decisive remedial managerial action for either egregious
misconduct or ineptitude, or, for that
matter, about the festival celebration of real heroes. Those who cannot or should not do a job
should leave that job–and be removed by the organization if they do not leave of their own accord;
the occasional gold medalist deserves her medals and our applause.

But “pay for performance” goes beyond any of these. It is necessarily contingent and usually
one-directional (top down), while dialogue about performance is not. It involves, not variation in
pay that is a fact of life enforced by the environment, but variation that senior leaders enforce
by choice–and could change if they wished. It is not just a process applied to the occasional
miscreant or the rare hero but one maintained in the very fabric of the organization, affecting all
by design.

Under this definition, “pay for performance” is guaranteed to be toxic in any organization in at
least the following ways.

1. “Pay for performance” makes the supervisor the customer. Organizations accountable to society or
to markets must meet the needs of their customers in order to thrive–this is the central strategic
message in modern quality management. Forget your customer, and, sooner or later, your customer
will forget you. Modern management systems seek ways to help their employees at all levels to
inquire about external needs and demands and to take initiative to meet those needs. A
“customer-focused” organization has inverted the pyramid of accountability; employees who meet and
know the external customer are the internal customers of the management system. In such an
organization, every supervisor needs to know how to improve his or her own ability to help others
meet customers’ needs.

“Pay for performance” distorts this focus; it changes the direction of concern. Not in theory but
almost always in fact the employee facing a supervisor about to dispense “4 percent, 5 percent, or
6 percent” has one question foremost in mind, namely, “How can I please the supervisor?” This is,
simply stated, not the key question upon which organizational survival depends. It diverts energy
away from the true interests of the organization. The interests of the organization lie outside its
walls; “pay for performance” turns people inward.

2. “Pay for performance” deprives the organization of essential information. Because of this
inversion of customer-supplier relationships within the organization, valuable information decays.
Two losses are the most costly ones. First, supervisors learn less than they otherwise could about
their own opportunities for improvement. Few employees facing merit reviews muster the courage to
correct their supervisors or to ask for better help in meeting customer needs. The person best able
to help the supervisor gain knowledge about needed improvements in management is the employee, who
is placed by “pay for performance” in the worst possible circumstance for giving that help.

Second, in the setting of performance review, the organization loses valuable information about
defects. Suppose, for example, the supervisor credits the employee with a successful result on the
basis of which the merit pay will increase, while the employee knows that the apparent “success” is
not at all what it is being cracked up to be. What the organization needs is the information that
the result was not good and that the information system is flawed; what the organization will
probably get is a silent smile, warm thanks, and an employee now trapped between honesty and

3. “Pay for performance” increases internal competitiveness and barriers. In the pursuit of quality, good fences make bad neighbors. A great organization today seeks constantly to cut windows between its functional areas and to help all employees feel more and more part of one team, with common pursuits and shared self-interest. “Pay for performance” never seems to hit this theme properly. Either the contingency is individualized and people come to worry that they must cling to the credit so they will not lose the pay or the contingency is set at such an extreme level of aggregation that the “congratulations for great performance” rings hollow to individuals, no one of whom ever can believe that he or she, alone, had much at all to do with the performance for which they are rewarded. The contingency upon which such variable pay is based tends therefore to be either fragmenting, at one extreme, or irrelevant, at the other. Either is damaging to the relationships within the organization. If merit pay is individual, and especially if it is distributed down functional lines of  hierarchy, then employees will decrease the extent to which they share information and efforts across boundaries. One hospital CEO described to me his system of profit-center management, in which middle management bonuses depended on local budget performance. I asked him if one of his managers would transfer resources from his department to another’s if it would help the organization as a whole. “Yes,” the CEO answered honestly, “if he were crazy.”

If merit pay appears in organization-wide bonuses, then employees tend to feet helpless in
responding to the contingency. I recall one hard day seeing patients at one stage in my
professional life, a day that had been a constant uphill battle against missed appointments,
unannounced patients with extremely complex problems, battles for approval with an outside
utilization review office, and a frustrating search for speech therapy services for a mute
three-year-old. The world seemed arrayed in opposition to my effectiveness, and I was exhausted as
I began to open my mail late in the afternoon. Therein was a check for a “productivity bonus” of
$297 and a note of organizational congratulations for my fine efforts–as it were, “Keep it up,
fella.” The first thought I had is not publishable. My second was, “Somebody doesn’t understand at

4. “Pay for performance” costs a great deal to administer. I know of no specific studies of the
proportion of organizational energy that a “pay for performance” system consumes, but experience
suggests that it is substantial. Elements of cost include the following: (1) supervisory training;
(2) creating and managing forms and records; (3) supervisory meetings and upward reporting; (4)
making decisions; (5) justification, revision, grievance, and reply; (6) administration of raises,
bonuses, and rewards; (7) goal setting and goal negotiation; and (8) collecting and analyzing
performance information. Greater still are the opportunity costs. Even at its best, “pay for
performance” is still a system of “inspection” in a technical sense. Quality management theory
counsels that the bulk of management energy should be devoted not to inspection of quality but to
planning and improvement. Whatever time the management system is devoting to the inspection
inherent in “pay for performance” is time denied to the much more significant management enterprise
of quality planning and improvement.

Even if “pay for performance” produced benefit for the organization (and I dispute that it
typically does produce benefit), the benefit would have to be great enough to outweigh the high
costs of maintaining the system.

5. “Pay for performance” is inescapably unfair. Statistical specialists working in quality
management study the proportions of variation in quality and occurrence of defect that turn out
empirically to be due to people in a system of production compared with other causes in the system
(such as the rules of procedure, the equipment, the raw materials, measurement error, and so on).3
Even in service industries, and, to the extent it has been studied, even in health care, the
preponderance of variation is not due to the people but is due instead to other sources. (The
relevant mental experiment would be to substitute a new work force, randomly chosen from qualified
candidates, in the current system and then to ask if basic performance levels would change.)

At a deeper level, even that proportion of variation in performance that is attributable to “human”
attributes of the system of production is itself complex in structure. A worker brings many
attributes to the work: skills, knowledge, attitude, mood, experience, ethnicity, nonwork
constraints, and ambitions, to name a few. A “pay for performance” system, especially if maintained
for purposes of incentive, can reasonably attach a contingent reward for an individual only to that
portion of variable performance that is not only attributable to the individual but that is also,
at least in principle, under the control of that individual. An “incentive” to make me a
competitive downhill skier would have no chance of success; I have bad knees.

An all-knowing supervisor could adjust for this problem and make the contingencies apply only to
the worker-controllable variation in performance. In practice, that is impossible. A
back-of-the-envelope calculation shows the magnitude of the issue. Imagine, for argument’s sake,
that 30 percent of the variation in the productivity of physicians in an HMO is associated with
individual characteristics (a proportion far greater than in most industries and unlikely to
actually occur). Imagine, further, that half of that proportion is attributable to “controllable”
characteristics (like effort level and learnable skills), with the other half being associated with relatively immutable traits (like the use of language, speed of calculation, and cautiousness in the face of risk). Thus, of variation in performance, 15 percent could be said to be susceptible to the motivation associated with merit pay and 85 percent not.
Now, imagine that a supervisor who observes variation in performance that is (in actual fact) not alterable by motivation reaches the correct conclusion (i.e., “Incentives cannot help”), say, four out of five times. Imagine, also, that the same supervisor, when observing variation that is (in actual fact) alterable through motivation reaches the correct conclusion (i.e., “Incentives can help”) fully half the time. A simple calculation shows that, of all the instances in which the supervisor thinks that motivating the workers will produce better performance, fewer than one-third actually are of that type. [Here is the calculation: The supervisor attributes 0.2 x 85% = 17% of the observed variation to the worker when, in fact, the worker could not control the cause and the supervisor attributes 0.5 x 15% = 7.5% of the observed variation correctly to the worker’s motivation. Thus, when the supervisor says, “This was controllable by worker motivation, and I expect that incentives can help,” the statement is correct only 7.5/(7.5 +17) = 31% of the time.] From the worker’s point of view, this produces extraordinary irrationality in the reward structure. More than half the time that the supervisor “rewards” or “punishes” performance, for example, the worker is not in fact in control of that performance. The sense develops that the reward structure is blunt, arbitrary, and often unfair, because, statistically, it is, despite the best efforts of the supervisor to be fair. The unfairness comes from the statistical hazards of attributing cause in a complex causal system. Further, in the calculations used here in the example, I presume that the discrimination abilities of the supervisor (80 percent “specificity” and 50 percent “sensitivity” for detecting variation that is reachable by incentives) are considerably better than I believe these abilities to be in most performance appraisal systems.
6. “Pay for performance” reduces intrinsic motivation. Many tasks, especially in health care, are potentially intrinsically satisfying. Relieving pain, answering questions, exercising manual dexterity, being confided in, working on a professional team, solving puzzles, and experiencing the role of a trusted authority–these are not at all bad ways to spend part of one’s day at work. Pride and joy in the work of caring is among the many motivations that do result in “performance” among health care professionals.

In the rancorous debates about compensation, fees, and reimbursement that so occupy the time of health care leaders and clinicians today, it is all too easy to neglect, or even to doubt, the fact that nonfinancial and intrinsic rewards are important in the work of medical care. Unfortunately, neglecting intrinsic satisfiers in work can inadvertently diminish them. Indeed, it has been possible in experimental settings to demonstrate a reduction in satisfaction from work by introducing extrinsic motivational factors. Students who will gladly work on a puzzle spontaneously when an experimental psychologist leaves them alone in a room will cease such spontaneous effort when the psychologist first offers to pay them to solve the puzzle.
It has been possible in experimental settings to demonstrate a reduction in satisfaction from work by introducing extrinsic motivational factors.
It is too much to say that pay for work does not support work, but it is psychologically tenable to assert that contingent pay for better work may decrease the joy one feels in that work. W. Edwards Deming called this phenomenon “overjustification” and believed that paying people to achieve what they would want to achieve anyway tends to reduce their satisfaction in the achievement.5 My seven-year-old daughter read book after book until her teacher began giving reading assignments and “stars” for completion, at which time completing reading “homework” became a nightly crisis.
7. “Pay for performance” “slows change. Especially in health care, breakthroughs in performance will require substantial changes in the way we do our work. We require an unprecedented level of innovation if we are to produce better outcomes at substantially lower cost. Innovation does not come without risk.

Logically, I see no reason why contingent pay should decrease risk taking; in theory, one ought to
be able to rig it to support change. In actual practice, however, “pay for performance” seems
almost always to exert a highly conservative, “antichange” influence. When goals that will be the
bases for variable compensation are set in advance, employees argue not for higher aspirations but
for lower ones. The conversation about performance is a debate about what is possible, not about
how to make something unprecedented possible. Long arguments take shape about how, exactly,
performance will be measured–arguments that do not focus on the processes of work but rather on
the processes of counting work. Usually, failed experiments (the inevitable result of the
willingness to take risks) result in deductions from “peformance” scores, even if the risks were
undertaken with the interests of the organization firmly in mind. Efforts in innovation, learning,
and the satisfaction of curiosity–enormous assets in any organization that wants to accelerate
improvement–are rarely counted as “performance,” and even training and education, being nowhere
reflected on the balance sheet, tend to be treated as “benefits” instead of as forms of
“performance.” Thus, “pay for performance,” often introduced to assist an organization whose
overall performance is unsatisfactory, tends to impede exactly those forms of systemic innovation
and learning that are, in the long run, most likely to dig it out of trouble.

8. “Pay for performance” is disrespectful of human relationships. In most of adult social human
interaction, contingency rewarding is rude. Who would accept a dinner invitation offered “in return
for your good behavior”? We solve problems together in sports teams, families, clubs, and
neighborhoods, not by explicit, contingent economic exchange, but rather by building on our shared
purposes, our common curiosity, our love, our sense of duty, or even by identifying the same enemy.
“Pay for performance is, with a few minor exceptions, generally reserved for only two settings:
commercial contracts and work.

This segregation of work as different from other forms of human interaction is so common as to seem
inescapable. But any student of the interior life of organizations knows that the social
relationships–the noneconomic forces–endure nonetheless. Our employees help each other; affection
develops and matures; teamwork counts on its own merits; people share their curiosity; and tribes
emerge bound by common rituals and common enemies. When one looks closely, the contingent reward
system–“pay for performance “–is as dissonant and distorting of the real life of the work setting
as it would be at a dinner party. People hate it; it feels wrong; it has little to do with their
valued relationships; it is an unwelcome game. Most of all, it erodes the potential for the true,
interpersonal, adult-to-adult relationship of equals among all of the parties to organizational life.
Fundamentally, as a human being, the CEO is not different in worth, character, or dignity from the
lowest-level employee. In the final analysis, we either believe that or we do not and our actions
reveal our beliefs far better than our words. Contingent pay down the line of hierarchy enforces
the erosive fiction that we are not all of the same stuff.

This point echoes of course in religion, ethics, and values. But it echoes also in organizational
performance. I cannot name a great team I have known in which an internal structure of contingent
pay, doled out by one team member to another, seemed at the root of its greatness. Relationships
mattered, purpose mattered, but “pay for performance” was not in the picture.
If “pay for performance” was not superficially logical, it would not have survived in the face of
its obvious defects and in the face of the dearth of good evidence to support it. It takes courage
in organizations openly to doubt its worth, and even more courage to abandon it in favor of less
well-described approaches to both pay and performance.

Defenders of merit pay will ask what the alternative is. There are no superb answers. Any viable
system of compensation must respect market forces, employee demands for growth, and numerous other
real-world factors. In addition, there are now strong pressures for various forms of gain-sharing,
in which employees as a group benefit as stakeholders in overall organizational performance. Clever
recognition and celebration systems, sometimes including financial reward, can apparently help
support morale and energy for and energy for improvement. Some companies report success with “pay for learning” systems, in which growth in skills is recognized in the pay structure. These are all ideas worth developing further, but, for the present, the answers remain incomplete to the question, “If I do not pay people according to their performance, upon what other basis will I vary their pay?”

I find myself an extremist and therefore suspicious of my answer. But it is, nonetheless, the best
answer I have yet found regarding merit pay for doctors or any group of workers; namely, “Stop it.”
Merit pay, “pay for performance,” and their close relatives are destructive of what we need most in
our health care industry–teamwork, continuous improvement, innovation, learning, pride, joy,
mutual respect, and a focus of all of our energies on meeting the needs of those who come to us for
help. We can find better ways to decide on how we pay each other and better uses for our energies than in the study and management of carrots and sticks.

Donald M. Berwick, M.D., is a practicing pediatrician and is President of the Institute for
Healthcare Improvement, Boston, Massachusetts.

This article appears as a chapter in Measuring Clinical Care – A Guide for Physician Executives.
Tampa, Fla.: American College of Physician Executives, 1995.
Quality Management in Health Care, 1995, 4(1), 27-33 © 1995 Donald M. Berwick.

1. Harvard Business Review. Appraising Performance Appraisal. Boston, Mass.: Harvard Business
School Press, 1993.

2. Kohn, A. “Why Incentive Plans Cannot Work.” Harvard Business Review 71, no. (1993): 54-63.

3. Scholtes, P.R. An Elaboration on Deming’s Teachings on Performance Appraisal. Madison, Wis.:
Joiner Associates, 1987.

4. Condry, J. “Enemies of Exploration: Self-initiated versus Other-initiated Learning.” Journal
of Personality and Social Psychology (July 1977).
5. Deming, W.E. The New Economics. Cambridge, Mass.: Massachusetts Institute of Technology Press,

Recent Posts