Journal of Economic Literature

Vol. XXXVII (March 1999), pp. 7-63

The Provision of Incentives in Firms

CANICE PRENDERGAST'

1. Introduction literature from two perspectives. First,

an underlying assumption of this litera-
I NCENTIVES ARE the essence of eco- ture is that individuals respond to con-
nomics. Despite many wide-ranging tracts that reward performance. Accord-
claims about their supposed importance, ingly, I consider whether agents behave in
there has been little empirical assess- this way, and whether these responses are
ment of incentive provision for workers. always in the firm's interest. Second, I ad-
The purpose of this paper is to critically dress whether firms write contracts with
overview existing work on the provision these responses in mind. In other words,
of incentives. Since the interests of do contracts look like the predictions of
workers and their employers are not al- the theory?
ways aligned, a large theoretical litera- Incentives are provided to workers
ture has emphasized how firms design through the compensation practices of
compensation contracts to induce em- firms, encompassing monitoring, evalu-
ployees to operate in the firm's interest. ation, and contracting, and firms use
This literature has reached into many ar- many different mechanisms to align in-
eas of compensation and has pointed to a terests. Some workers, such as sales-
multitude of different mechanisms that force employees, are predominantly re-
can be used to induce workers to act in warded for their efforts through explicit
the interests of their employers. These contracts that relate pay to observed
include piece rates, options, discretion- measures of performance. Others are
ary bonuses, promotions, profit sharing, rewarded not on individual measures of
efficiency wages, deferred compensa- performance but on more aggregate
tion, and so on. My objective here is to measures, such as profit-sharing ar-
evaluate this literature in the light of a rangements. However, many employers
growing empirical literature on compen- eschew the use of explicit contracts,
sation. Where possible, I will address the preferring to reward individuals based
on a discretionary subjective measure of
1 University of Chicago and NBER. This yaper performance. Finally, some employers
has developed from discussions with many differ-
ent people, too many to thank individually. How- prefer to avoid pay-for-performance al-
ever, I would like to especially thank Robert together. The objective here is to un-
Gibbons, Bengt Holmstrom, Edward Lazear, Lars
derstand these diverse means of com-
Stole, and Robert Topel for helping to shape these
ideas. I have additionally received very helpful pensation and their implications for the
comments on this paper from Judy Chevalier, provision of incentives for agents to ex-
Bentley MacLeod, Brigitte Madrian, Derek Neal,
ert effort and allocate their time in the
Michael Raith, Mike Waldman, the editors (John
McMillan and John Pencavel), and two anonymous appropriate way. Of course, the answer
referees. All errors are my own. to how contracts affect performance is
7

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
8 lournal of Economic Literature- Vol. XXXVII (March 1999)

not going to be universal; instead, the approach, which takes the answer to the
purpose will be to illustrate situations first question as given, is to identify
where it appears to improve incentives whether observed contracts vary in the
but also to point out the pitfalls of such way that the theories suggest they
a reliance on contractual outcomes. should. For instance, if risk is a con-
Such prescriptions will be tempered by straint to offering incentives, does the
the nature of the job carried out by strength of the relationship between
workers, the extent to which they have pay and performance fall as the mea-
discretion in their jobs, and the extent sures on which contracts are condi-
to which the measures used to pay workers tioned become more noisy? This is a
truly reflect the inputs of effort. truer test of recent contributions to
The paper is organized along the cen- agency theory, which largely hold that
tral themes of the literature. Section 2 contracts are designed with the re-
considers static contracts, in which in- sponses of agents in mind. Here the
centives are offered in a single-shot set- evidence is more mixed, with some
ting. I begin in Section 2.1 by setting up work finding evidence in favor of the
the basic theoretical apparatus that will theories, while others find little.
be used throughout the paper. A single An alternative reason why it may be
model is provided at this stage which difficult to provide incentives is that
can encompass the main themes of the contracts cannot completely specify all
literature, though I initially address the relevant aspects of worker behavior. As
trade-off of risk and incentives. Here a result, contracts offering incentives
the provision of incentives is aided by can give rise to dysfunctional behav-
the use of pay-for-performance, but the ioral responses, where agents empha-
primary constraint on incentives is that size only those aspects of performance
their provision imposes additional risk that are rewarded. For example, con-
on workers, which is costly to firms sider a baseball player who receives a
through higher wages. From this per- contract with a reward for hitting home
spective, pay-for-performance is con- runs. The danger here is that the player
strained by the noisiness of the mea- will attempt to hit home runs even in
sures used to reward agents, and the situations where it is not warranted. Or
ability of agents to handle risk. teachers who are rewarded on test
There is a substantial empirical litera- scores may teach "for the test." Such
ture testing the trade-off between risk behavioral responses arise because con-
and incentives. The premise of this lit- tracts often cannot rely upon a holistic
erature is that relating pay to perfor- measure of the worker's contribution at
mance increases output, but at the cost every moment in time. Because of this,
of imposing risk on workers, which is agents can "game" the compensation
reflected in higher wages. Two basic system when they have multiple instru-
themes have been taken. First, a series- ments at their control. Following Bengt
of papers considers "Do Incentives Holmstrom and Paul Milgrom (1990)
Matter?"; in other words, do employees and George Baker (1992), this incentive
perform better when placed on com- problem has become known as multi-
pensation schemes where pay is more tasking, where compensation on any
closely related to performance? Recent subset of tasks will result in a realloca-
evidence suggests that there are strong tion of activities toward those that are
responses of output to the use of pay- directly compensated and away from
for-performance contracts. The second the uncompensated activities. Recent

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 9

empirical work has illustrated such be- tainted by supervisor bias or workers
havioral responses to incentive con- currying favor.
tracts. As a result of these concerns, it Perhaps the most common means of
is predicted that in those positions rewarding white-collar workers for ef-
where there are significant oppor- fort is by promotion. Workers are pri-
tunities for reallocation of activities, marily allocated to positions in firms on
there will be an absence of pay-for- the basis of their talents. However, an
performance; in essence, complex jobs implication of job allocation and promo-
will typically not be evaluated through tion is that a job hierarchy provides in-
explicit contracts. centives to workers. Section 2.3 assesses
Since it is difficult to specify all as- the contribution of tournament theory,
pects of workers' jobs in an explicit con- where promotions are modeled as a
tract, a common way of providing incen- tournament among a group of agents
tives is to use subjective performance competing for a fixed set of prizes,
evaluation, perhaps in addition to some rather like a sports setting. Tournament
objective assessments. The typical theory makes a number of predictions,
worker operates in a setting where ef- which have found support in empirical
forts are exerted in the hope of a pro- work. First, larger prizes should result
motion, salary revision, or bonus, which in more effort. Second, in a contest
are typically at the discretion of supe- where there is a single winner, the prize
riors. Such subjective assessments have should be increasing in the number of
the benefit that they can be a more contestants; finally, workers who fall be-
fully rounded measure of performance; hind in contests should be likely to take
for instance, the baseball player could risky strategies to "catch up." An addi-
be rewarded for hitting a home run only tional implication is that when dysfunc-
if attempting to do so was warranted at tional behavioral responses arise, firms
the time. However, there is consider- may use bureaucratic rules to allocate
able evidence that subjective assess- promotions. For instance, many firms
ments also give rise to biases. For in- heavily weigh seniority in promotion de-
stance, when evaluations are subjective, cisions after controlling for produc-
workers are likely to waste valuable re- tivity, where sometimes the "wrong"
sources (work time, for example) curry- worker is promoted on seniority
ing favor with their bosses. In addition grounds. This section shows how such
to the incentive to engage in such ac- bureaucratic rules are the optimal re-
tivities, a number of other problems sponse to incentives for workers to en-
have been highlighted, ranging from gage in dysfunctional responses to
"leniency bias," where supervisors are evaluations procedures.
reluctant to give bad ratings to workers, Many workers are employed in set-
and "centrality bias," where supervisors tings where output measures are not
compress ratings around some norm the outcome of the inputs of a single
rather than truly distinguishing good individual, but rather derive from the
from bad performance. Section 2.2 con- joint contributions of many individuals.
siders the provision of incentives as a Such team production concerns are the
trade-off of the distortions implicit in subject of Section 2.4, where the focus
evaluation; explicit contracts result in is on two incentive aspects of team pro-
agents optimizing relative to the con- duction. First, there is considerable
tract ("you get what you pay for"), while evidence of free riding in teams. De-
subjectively assessed benefits may be spite this, the data suggest productivity

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
10 Journal of Economic Literature, Vol. XXXVII (March 1999)

improvements of the order of 4 to 5 effort as a younger worker is not just

percent from the introduction of com- the contemporaneous return but the
pany-wide profit-sharing schemes, prospect of receiving the returns of an
where the benefits of increased effort older worker in the future. Consider-
are shared with often thousands of oth- able empirical evidence suggests that
ers. Since this constitutes an apparent firms do indeed follow such compensa-
violation of standard agency theory tion practices, though there are often
(why exert effort if I gain only 1other I ofplausible interpretations of these
the benefits?), I address this issue in results. This section also considers how
some depth. Second, one of the reasons the return to promotion changes as a
often suggested for the success of worker ascends a firm's hierarchy.
such team compensation schemes is Another important feature of dy-
that it gives workers an incentive to namic agency contracts is that they can
monitor one another via peer pressure. be renegotiated over time based on pre-
Available evidence, though scant, sug- vious performance. Such renegotiation
gests outcomes rather different from opportunities have been termed career
those predicted by the theoretical concerns, which are the topic of Section
literature. 3.2. Once again, consider the baseball
Efficiency wage theory argues that player example, but where the player is
the provision of incentives causes work- offered a fixed salary in a given season.
ers to receive rents above their market Despite the fact that there is no imme-
wages. In effect, workers are offered diate relation between pay and perfor-
rents because they are less likely to mance, he is likely to have incentives to
shirk if their jobs are valuable; hence exert effort because good performance'
high wages induce effort. I provide a will improve future contracts. In other
brief review of this literature in Section words, the market "settles up." Such re-
2.5, and argue that some of the tests putational concerns imply that effort ex-
used here are of lower power than one ertion can occur without explicit pay-
would like. for-performance contracts, though
All the effects described above ignore rarely at the efficient level. This career
the fact that workers remain with em- concerns model has testable implica-
ployers and in the labor market for long tions for the behavior of workers and
periods of time. The fact that workers contracts which have been borne out in
have careers (rather than one-time rela- the small existing literature on this
tionships) allows workers and firms dis- subject.
cretion over pay, which results in some A final role for repeated relationships
different implications from the static in this environment is that they allow
models described above. Section 3 con- for honest behavior in settings where
siders such dynamic linkages in the pro- cheating would occur in a static rela-
vision of incentives, where the optimal tionship. The ability of workers to pun-
contract offered today depends on ish firms that renege on their obliga-
either the contract offered yesterday or tions implies that repetition of the
behavior yesterday. Two aspects are relation can imply better outcomes
emphasized. Section 3.1 considers de- when performance measures are sub-
ferred compensation, where firms sys- jectively determined. In addition, re-
tematically "overpay" older workers and peated observations on the performance
"underpay" their younger counterparts. of workers can allow more precise
Then part of the return to exerting inferences on their performance, thus

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 11

providing another role for repeated stand the evaluation and compensation
relationships. Such dynamic contracting of those with noncontracted output.
issues are dealt with in Section 3.3. Before beginning the substantive
The literature has provided an impor- parts of the paper, let me make clear
tant organizing tool for understanding what this overview does not cover.
the wage policies of firms, and enor- First, as mentioned above, its purpose
mous advances have been made in some is not to be a comprehensive overview
areas. Most of this paper is designed to of the large theoretical literature on
illuminate such advances. However, the compensation. Instead, a theoretical ap-
paper concludes with three observations paratus is provided as a guiding device
in areas where the literature has not for understanding the main influences
been so successful. First, there is con- in the theoretical literature. References
siderable evidence that incentives mat- to the relevant theoretical work are pro-
ter; paying individuals to do X causes vided for the interested reader. Second,
them to do X. However, it is much less a large literature on compensation con-
clear that the theories that assume such siders aspects other than incentive pro-
responses have great predictive power. vision, since obviously compensation re-
Second, the literature has suffered from flects other considerations, such as
identification difficulties. Such identifi- human capital acquisition and learning.
cation concerns arise from two separate This literature is not considered here;
sources. First, researchers must over- the emphasis is solely on the provision
come the standard empirical identifica- of incentives. Another important re-
tion problem (which realizes that the striction is that I do not address how
choice of contracts is endogenous). ownership of assets alleviates incentive
However, even in situations where there and bargaining concerns.2
is evidence consistent with agency the- Perhaps the most important omission
ory, the literature has been plagued with from the paper is that it concerns only
a second (theoretical) identification how effort can be induced by incentive
problem where outcomes are often contracts. As a result, I do not consider
equally consistent with other plausible such issues as job enrichment, as in J.
theories. Richard Hackman and Greg Oldham
Third, I believe that there has been (1980). More generally, I do not ad-
insufficient focus on workers whose dress in much depth the noneconomic
outputs are hard to observe, in particu- literature on the effect of compensation
lar those where subjective assessments schemes on performance, though where
are used. Instead, the understandable the predictions of the economic models
focus of the literature has been on oc- appear particularly at odds with those of
cupations (such as CEOs, mutual fund other disciplines, I briefly consider the
managers, professional golfers, etc.) noneconomic evidence.
for which measures of output are avail- Finally, it should be noted that this
able. However, the majority of workers is not the only overview of the issues
do not satisfy these criteria. Instead, covered in this paper; for alternative
most workers are evaluated on subjec-
2 Following Sanford Grossman and Oliver Hart
tive criteria, where firms choose how to (1986) and Hart and John Moore (1989), there is
evaluate and how to pay based on those now a large literature on the theory of the firm
evaluations. Consequently, it seems to taking the perspective of incomplete contracting,
where hold-up or bargaining problems constrain
me that a critical avenue for future efficiency, which has implications for vertical inte-
research should be to better under- gration or, more generally, control rights.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
12 Journal of Economic Literature, Vol. XXXVII (March 1999)
perspectives on some similar issues, see Although the principal cannot ob-
Robert Gibbons (1996), Gibbons and Mi- serve the actions of the agent, she can
chael Waldman (forthcoming), Edward potentially condition payments on a set
Lazear (1995), and James Malcomson of signals that are correlated with the
(forthcoming). agent's actions. For illustrative pur-
poses, I consider two such signals, an
2. Static Contracts objective measure of performance, y,
and a subjective measure of perfor-
The first part of the paper considers
mance, s. Objective measures are char-
the use of static contracts, where the re-
acterized by the fact that they can be
lationship between the worker (agent)
verified for contractual purposes, while
and firm (principal) is onetime. I begin
a subjective measure is anything that is
by setting out the basic theoretical
not verifiable to a third party. I assume
structure that will be used throughout
that the principal maximizes expected
the paper, though in this section the
profits (output minus wage costs),
primary focus is on the trade-off of risk
where expected output is given by the
and incentives.
sum of the effort of the worker, e, and
2.1 The Basic Theory and the his ability, cc. I assume that the signals y
Trade-off of Risk and Incentives and s are characterized by

The premise of agency theory is that y = e + CC + y, (2)

a principal designs contracts in order to and
guide appropriate actions by an agent. s = e + CC + Es, (3)
The agent is assumed to take some ac-
where ei- N(O,62). Thus, ay is the mea-
tion e 2 0, which is unobserved by the surement error of the objective signal
principal. Throughout the paper I will
and a2 is its counterpart on the subjec-
refer to e as effort, though there are
tive signal. Although s is subjectively de-
other plausible interpretations as will termined, in this section I assume that
become clear below. The agent is effort contracts can credibly be written on that
averse. The purpose of this survey is to measure. Problems associated with sub-
identify the major themes of the theo- jectivity are addressed in more detail be-
retical literature in a simple way. To do
low. The term cc refers to the ability of
so, I choose a simple parameterization the agent and for the moment is symmet-
of the agent's utility function, where rically unknown to all agents (the case
the agent cares about wages w and where it is privately known is considered
effort e; I assume that the agent has below). I assume that cc - N(0,02). All
exponential utility random variables are uncorrelated with
V = - exp[-r(w-C(e))], (1) each other.
where w is the worker's wage, r 2 0 is Perhaps the most important observa-
the constant rate of absolute risk aver- tion of the early contributions to agency
sion, and the worker's cost of supplying theory (Holmstrom 1979) is what has
effort is C(e). Purely for tractability, become known as the Informativeness
the cost function is assumed to be Principle, which (loosely) implies that
quadratic, where C(e) = c. The princi- any measure of performance that (on
pal is assumed to be risk neutral, and the the margin) reveals information on the
worker has reservation utility U*.3
the worker. This could be a position in another
3 This reservation utility is simply the utility firm or the value of leisure if the agent chooses
arising from the best outside option available to not to work.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 13

effort level chosen by the agent should then f, + f3 = I so that the first best level
be included in the compensation con- of effortis exerted. However, if r > 0 and
tract. In this context, this implies that there is measurement error on both per-
both the objective and subjective mea- formance indicators, effort is always be-
sures should be used to reward the low the first best level (f3 + f3 < 1). A
worker. The attraction of the exponen- particular measure's weight is decreasing
tial utility function with normally dis- in the variance of the measure's signal,
tributed errors is that, following Holm- so noisy measures should be less used.
strom and Milgrom (1987), the optimal However, the weight attached to any
contract relating wages to these ob- measure is increasing in the noisiness of
served signals is linear (which makes the other measure, though total incen-
comparative statics simple to illus- tives, P* + .3r are decreasing in the noisi-
trate).4 Thus, the optimal means of ness of any measure. Similarly, a high
compensation is given by degree of risk aversion mutes incentives,
and incentives rise as the return to effort
tv = PO + PyY + sS, (4)
(I) rises.
where 13o is the worker's salary and Pi is
the "piece rate" on signal i. (The salary 2.1.1 An Alternative Interpretation
plays little role here, being chosen
A useful way of interpreting this result
simply to ensure that the worker earns
on incentives is that the principal follows
his reservation utility. As a result, unless
a two-step procedure for providing in-
relevant it is ignored in what follows.)
centives, where the principal separates
The worker chooses effort to maximize
performance evaluation from the provi-
V. which implies that optimal effort e* is
sion of incentives. First, she optimally
given bv' e* = 13t 1S Note that the first- aggregates information on the worker's
1~~~~ c~~~ effort, assuming a diffuse prior. This es-
best level of effort is 1-, which only oc-timate, which minimizes the mean-
curs if f35 t .s= 1. Optimizing over the cy2y + Cy2S
choice of compensation contract, it is square error, is given by z = s +
simple to show that the firm chooses
The firm then optimally compensates on
piece rates of
this performance measure by discounting
the piece rate from 1 for risk sharing rea-

PY ,2. =
+ 6y2 ~ sons, by rewarding via w = Po + jz, where
~~s
4-rc((.2. y+ O.s2Cy2)
+ Cy2)CF (5) (5
1
and
1 + rcca? (7)

y 2 + y2 + rC((y2 + Cy2)(y& + 2Cy2 (6)

where cy= =2 + y is the regression

This simple model illustrates the trade- error. Note that a tends to 1 as r tends to
off between risk and incentives that has zero, so piece rates on this optimal ag-
been a central early theme of the litera- gregation converge to unity as the
ture. If the worker is risk neutral (r = 0), worker becomes more able to handle
4 Holmstrom and Milgrom illustrate this point in
risk.
a dynamic environment, where agents control ef- The importance of this interpretation
forts continuously where in any period a binary is that in the standard model of incen-
realization of success is achieved. In that setting,
compensation that is linear in the score (number
tives, the firm optimally aggregates in-
of successes) is optimal. formation on performance and then

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
14 Journal of Economic Literature, Vol. XXX VII (March 1999)

discounts this optimal aggregator for tive performance evaluation has two
risk sharing reasons. One point that will testable implications. First, in environ-
become clear below is that this rule of ments where there are common factors
optimal aggregation will not be fol- affecting compensation, agents should
lowed when agents take distortionary be partially rewarded on how well they
actions in response to contracts. do relative to others, and second, the
degree of relative performance evalu-
2.1.2 Implications of the Basic Theory ation should increase in the correlation
between the two signals.
Relative Performance Evaluation. The Selection Effects of Contracts.
The most important implication of the The second outcome of this simple
analysis above is that errors in mea- model is that compensation contracts
suring performance constrain the provi- have selection effects, with higher piece
sion of incentives. As a result, any signal rates being relatively more attractive to
that is informative about performance better workers, as in Lazear (1986). An
should be used in compensation pack- implication of this is that firms now de-
ages (the Informativeness Principle). sign contracts not only to induce effort
The most common example of the use but also to affect the type of workers
of this principle has been application that they hire. To see this, adapt the ba-
of Relative Performance Evaluation, sic set-up above by assuming that work-
where the performance of one agent is ers privately know their own ability, cx,
compared to another when choosing where for simplicity I assume that the
compensation. reservation utility of the worker does
Relative performance evaluation is not depend on ox. Let M be the mone-
used as a means of filtering out com- tary certainty equivalent of the reserva-
mon risk from compensation packages. tion utility UP. Since the optimal con-
To give a concrete example, consider tract will then reflect selection
two salesforce workers who carry out concerns, it will differ from (5) and (6)
similar jobs. Demand for the products above. Then for any linear contract w,
in the area in which they both work var- as in (4), only those workers whose abil-
ies for common reasons beyond their ity exceeds W will choose to work for
control. If agents are compensated the firm, where
solely on their own productivity, they
are exposed to the risk inherent in the Po + (3y + 3s)[o* + e*] (8)
common fluctuations in demand. A so-
- p2 (y,2 + P2(y2+ ,Psay) - C(e*
lution to this problem is to (at least par-
tially) reward the workers on how well By substitution, this implies that
they do relative to each other; in this
way they are not penalized so much for i (Ps2R2 + y2 + pscp2cs2yy) - Po + M
marketwide changes in demand.5 Rela-
py + P(s

5To see a role for relative performance evalu- _ I3 + Ps3 C

ation, consider the model above where (i) there 2c (9)
are no subjective signals ( c2=o), (ii) ability is un-
important (y2 = 0), but (iii) there is another workerThen it is straightforward to show that the optimal
whose performance is correlated with that of the contract for the worker is w = Po + ,yy - ,y, where
agent. Let that worker's output be given by
y= e + , where the "tilde" refers to the other
1 + rcy2(1 _ p2) and py
agent, and the error terms are distributed as
above. Assume that there are some common Thus, one worker is penalized when the other
shocks that hit both agents so that cov(?, E) = p. does better, all other things equal.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firmns 15

Then any case where a firm shifts from a combination of normal errors and ab-
fixed wage scheme to one where piece sence of income effects yields the linear
rates are used will increase the average contracts above. However, the reader
quality of worker. should be aware that this is a special
The primary focus of the agency lit- case and that in general the sharing rule
erature has been on how contracts in- will not be linear. See Holmstrom
duce certain behaviors from agents. (1979) and Grossman and Hart (1983)
However, compensation contracts also for details. The attraction of the struc-
play a central role in recruiting workers ture above is therefore its tractibility.
to firms.6 By offering greater pay-for- Yet many observed contracts are non-
performance, firms may hire a better linear, where, for example, discrete bo-
distribution of workers, since the more nuses are offered for exceeding some
able will benefit more from these con- performance threshold. For instance,
tracts than will be the case for the less Kevin J. Murphy (1998) highlights the
able. Of course, this changes the design importance of such bonus contracts for
of optimal contracts, since compensa- executives. Perhaps the most important
tion contracts now fulfill a dual role of form of nonlinearity concerns the threat
both inducing effort and aiding the se- of being fired, where wages vary little
lection of appropriate workers. As a re- with performance but where poor per-
sult, firms may distort the effort deci- formance is punished by dismissal.
sions from the choices in (5) and (6) in Rather remarkably, the theoretical lit-
order to select certain types of workers. erature has made little progress in un-
The Shape of Compensation Con- derstanding the observed (nonlinear)
tracts. The specification of preferences shape of compensation contracts, despite
and measurement errors above is not in- costs associated with nonlinearities,
nocuous. First, effort is one-dimen- which are described below.
sional. A more general setting would al-
2.1.3 Empirical Tests of the
low the agent to carry out multiple
Basic Agency Model
activities, a point that is returned to be-
low. Second, the actions of the agent Empirical work has taken two con-
can affect only the mean of the distri- ceptual approaches. First, a body of
bution of output; no actions can affect work considers "Do Incentives Matter?"
the higher moments of the distribution. In other words, does worker perfor-
For instance, agents cannot vary the mance improve when pay is more sensi-
riskiness of the performance measures. tive to performance, and if so, by how
Finally, the exponential specification of much? It should be remembered that
preferences ignores income effects. The this is not a test of the agency theory
above. Instead, it tests a necessary in-
6 There are many other aspects of compensation gredient for the theory, namely, that
that have selection effects. One mechanism which agents respond to incentives, not neces-
has been emphasized (George Akerlof, 1976) is
where agents reveal their willingness to work by
sarily that contracts are designed to re-
working harder than is efficient. Law partnerships flect the trade-off above. Consequently,
are the typical example used, where associates a second approach considers the con-
"burn the midnight oil," largely to appear moti-
vated, though the productivity effect of the final
tracts offered to agents to identify
hours is low. Renee Landers, James Rebitzer, and whether the concerns mentioned above
Lowell Taylor (1996) empirically address this issue (risk, the return to effort, etc.) are
using survey evidence from law partnerships,
which suggests a preference for hours reductions
reflected in observed compensation
if no inferences are drawn from doing so. practices. I address each in turn.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
16 Journal of Economic Literature, Vol. XXX VII (March 1999)

Do Incentives Matter? Until recently, by Harry Paarsch and Bruce Shearer

there was remarkably little work in eco- (1996). In this case, climatic and soil
nomics documenting the effect of com- conditions determine the use of piece
pensation policies on performance.7 The rates or salaries. Their data are less ex-
paucity of such work probably arose tensive than Lazear's, so their estimates
from the absence of the necessary infor- are less precisely measured. They carry
mation. An adequate test of the effect out a number of useful bounds tests
of pay on performance needs data on that constrain the effects of pay on pro-
contracts offered to workers, measures ductivity. First, wages rise by 6 percent
of performance, and an understanding when workers operate under piece rates
of why the contracts vary across work- relative to salaries. This constitutes a
ers. Despite these constraints, a num- lower bound on the effect of pay on
ber of recent papers have illustrated productivity; otherwise the firms would
quite substantial incentive and worker prefer to simply retain workers on fixed
selection effects. It is important to bear salaries. A plausible upper bound on the
in mind here in evaluating these studies effect of pay on performance is the raw
that in each of the cases documented productivity difference, which is 35 per-
below, the nature of the job carried out cent.8 The authors use a structural form
by the workers is "simple," in the sense of estimation to control for contract se-
that an aggregate measure of the lection effects and estimate that the in-
worker's performance is easily available. centives from piece rates for a given
Lazear (1996) considers the impact of worker are about 10 percent.9
piece rates on the performance of work- The attraction of these two pieces of
ers who install auto windshields. Man- work is that both have individual data
agement changes provided the impetus on performance and contracts. A series
for changes in compensation from fixed of other papers has been more con-
salaries to piece rates, and Lazear illus- strained by data limitations, but none-
trates that productivity rose by approxi- theless has provided useful informa-
mately 35 percent from this change, tion on the effect of compensation poli-
with wages increasing by 12 percent. cies on performance. First, Rajiv
Lazear also uses turnover data to illus- Banker, Seok-Young Lee, and Gordon
trate the selection effects above, where Potter (1996) consider the effect of
approximately a third of the improved piece rates on sales in retail department
performance can be attributed to selec- stores. Data are collected at the store
tion effects; the less able left the firm level rather than for individuals, and
and more talented workers replaced they show that store productivity rises
them. Similar evidence is presented by between 9 and 14 percent from the
from a study of Canadian tree planters change, though the authors cannot dis-
tinguish between true incentive effects
and worker selection. Sue Fernie and
7 At a general level, this section is concerned
with understanding the effect of prices on the David Metcalf (1996) address the com-
market for leisure; when the price of on-the-job pensation of British jockeys, where
leisure rises, do agents consume less of it? The
some jockeys are employed on fixed
premise of this section is that the alternative to
exerting effort is laziness, but a little-understood
aspect of this literature concerns quite how agents' 8 This constitutes an upper bound because piece
incentives differ from those of the principa For rates were used in favoralle conditions.
recent work pointing out incentives for agents 9 See Chris Ferrall and Shearer (1998) for
other than to be "lazy," see James Heckman, Jeff another structural approach to identifying the
Smith, and Chris Taber (1996). parameters of the agency problem.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 17

retainers and others are offered prizes and pay, once again suggesting the effi-
for winning races. Their results again ciency of relating pay to measures of
suggest significant incentive effects, performance. Finally, each of the pa-
though their sample size is small. Fi- pers mentioned above considers the ef-
nally, John McMillan, John Whalley, fect of contracts on outputs rather than
and Lijing Zhu (1989) and Theodore measuring effort itself. An exception to
Groves et al. (1994) address how Chi- this is Andrew Foster and Mark Rosen-
nese economic reforms have affected zweig (1994), who collect data on effort
performance levels through changed exerted by agricultural workers in the
compensation practices. McMillan, Philippines. They do so by examining
Whalley, and Zhu estimate a production weight changes for workers on piece
function for Chinese agriculture using rates and salaries (time wages), with the
aggregate data, and show that perhaps inference being that weight reduction
75 percent of the increases in agricul- reflects greater effort. They note, first,
tural productivity from 1978 to 1984 that conditional on calorie intake, those
can be attributed to the introduction of on piece rates lose more weight than
the responsibility system, which allows those on fixed salaries. This suggests
local communes to retain a share of that the exertion of effort under piece
their profits. Groves et al. use survey rates causes weight loss. Second, with-
data on 800 enterprises in the Chinese out conditioning on calorie intake,
industrial sector, where information re- weight gain is higher for those on
ported by managers suggests a strong piece rates than salaries, illustrating
link between industrial performance that those who operate on piece rates
and the use of bonuses and contract ultimately put on more weight, i.e.,
labor. 10 surplus created from the use of piece
Larry Kahn and Peter Sherer (1990) rates.
use the personnel files of a large com- In summary, this new literature
pany to identify the effects of subjective points to considerable effects of com-
performance evaluation on the perfor- pensation on performance. Studies that
mance of white-collar office workers. allowed the effects of incentives to be
They show that better evaluations were separated from worker selection issues
achieved by those employees who have suggest that perhaps one-third of the
a steeper relation between evaluations increase in performance arises from at-
tracting better workers. It is worth em-
10 At a more aggregate level, Louis Putterman
(1990) illustrates large increases in the perfor-
phasizing two points here. First, in each
mance of Chinese township and village enterprises of the cases considered above, workers
when they were allowed to keep larger shares of carry out "simple" jobs, in the sense
output. By contrast, Nicholas Barberis, Maxim
Boycko, and Andrei Shleifer (1996) show little in-
that aggregate measures of performance
centive effects of equity holdings in the Russian are available; it is for these jobs that
retail sector, instead emphasizing human capital piece rates are most likely to work. Sec-
aspects of success. See a so John Abowd (1990),
who uses event study methodology to determine
ond, while it is important to show that
whether the stock market responds favorably to incentives matter, these studies are not
the introduction of sensitive pay-for-performance truly a test of agency theory. They are
schemes. He finds evidence in favor of this,
though, as he admits, the results are not conclu- merely a test of an input to the theory,
sive. Note, however, that the interpretation of where a more precise test is to address
these results depends on why firms introduced whether contracts are structured as pre-
more sensitive compensation schemes; only if they
were originally "too low" should we expect this dicted by the theory. I now turn to this
response. issue.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
18 Journal of Economic Literature, Vol. XXXVII (March 1999)

Before doing so, however, I should (1971) and Mark Lepper, David
mention a noneconomic literature that Greene, and Robert Nisbett (1973) for
holds that offering explicit incentives example, and Barry Staw (1989) for
can reduce productivity by eliminating other interpretations of these findings.
the intrinsic desire to carry out some Perhaps the most cautious caveat that
activity. In other words, pay-for-perfor- we can apply to the results above based
mance harms incentives, unlike the sug- on these findings is that they may be
gestions above. The premise of this re- most plausible for activities where little
view is that "effort" refers to some intrinsic motivation is evident without
activity that the individual would rather explicit incentives. See David Kreps
avoid. Yet sociologists and psychologists (1997) for more observations on this
take the perspective that individuals issue.
often have pride in their work and enjoy Do Contracts Reflect Agency Con-
carrying out required tasks. This, of cerns? The theoretical apparatus set up
course, is not a problem for the theory above suggests not only that compensa-
above as long as such intrinsic interest tion should change with measures of
is not adversely affected by pay-for-per- performance, but also that the size of
formance. Yet it is sometimes argued this relationship depends on such fac-
that such a link exists, so that paying tors as the noisiness of these measures,
people on the margin to carry out the marginal return to effort, and the
some activity reduces their intrinsic risk tolerance of the agents. Accord-
enjoyment of the task."I ingly, a second theme of the literature
While this idea holds some intuitive has concerned identifying the relation-
appeal, it should be noted that there is ship between compensation schemes
little conclusive empirical evidence and proxies of these measures.
(particularly in workplace settings) of Perhaps the most celebrated example
these influences.12 See Edward Deci of empirically estimating compensation
schemes has been a series of papers
11 One version of this is that when an individual
performs an act, he must justify the action. If he is that estimate pay-for-performance for
not directly paid for the act, he will rationalize his executives and, particularly, chief ex-
efforts by perceiving that he enjoyed the task. By ecutive officers. More specifically, the :
contrast, if he is rewarded for carrying out the
task, this rationalization is no longer necessary and coefficients above are estimated for a
will attribute the reason for doing the task to the series of performance measures. Here
monetary rewards, which will lead him to dislike
the typical paper has estimated the rela-
the activity. This dislike could result in worse
performance under piece rates. tionship between performance (stock
12 The methodology typically used in this litera- price return, earnings, etc.) and some
ture is to consider two groups carrying out some
measure of the agent's welfare (pay,
interesting activity. For instance, some experi-
ments have allowed children to draw pictures or propensity to be fired, etc.). See Mur-
play with toys. One group is placed on pay-for- phy (1985), Michael Jensen and Murphy
performance while the other is not. Intrinsic moti- (1990), and Stephan Kaplan (1992), for
vation is then tested by considering the behavior
of the individuals after the supposed period of the example. 13 Using data for U.S. chief
experiment is over. If those who are on pay-for-
performance are less willing to continue the activ- those who have operated at a more leisurely pace
ity than those who are not on such schemes, it is without pay for performance. Thus they may be
argued that intrinsic motivation falls from the use less likely to continue'the activity for reasons other
of explicit incentives. While this logic may indeed than intrinsic motivation; instead, diminishing
be correct, an alternative which seems plausible is marginal returns to the activity will suffice.
that if those who operate on piece rates perform 13 Also see Richard Lambert and David Larker
better during the experiment period, they are sim- (1987), Ann Coughlin and Ronald Schmidt (1985),
ply more tired of carrying out that activity than and Martin Conyon and Simon Peck (1996) for

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 19

executive officers, Jensen and Murphy vary with relevant parameters. In other
estimate that a $1,000 increase in the words, are signals used less heavily
value of a typical large U.S. company when they are noisier, or when agents
increases pay by approximately $3.25, are less able to handle risk? An early
most of this coming from stock owner- attempt to test the predictions of
ship. On the basis of this number, the agency theory is Seiichi Kawasaki and
authors argue that too few incentives McMillan (1987), who are concerned
are provided to executives. While the with the relationship between Japanese
conclusions taken from this literature firms and their subcontractors. Since
could be correct, this seems a poor firms would like subcontractors to con-
method of testing agency theory.14 This strain costs, they are reluctant to write
is because many of the factors relevant cost-plus contracts. Instead, a sharing
for choosing the level of compensation rule is specified, where a fraction of
are unobserved; the optimal piece rate costs can be passed on. On average,
depends on risk aversion and the re- about 60 percent of cost overruns are
turns to effort, both of which are un- passed on, but this figure varies with
known to the econometrician.15 As a re- the environment. First, subcontractors
sult, it is difficult to determine whether who face very volatile costs can pass on
compensation schemes are set opti- more costs than those for whom there is
mally, or to claim that the relationship little volatility. Second, smaller subcon-
between pay and performance is too tractors, who are less able to handle
low or too high.16 It is a little like claim- risk, can pass on more costs. Finally,
ing that prices are too high without the authors use a measure of the mar-
knowing costs. ginal product of the effort of the sub-
A second approach to understanding contractors (whether they order and ne-
the impact of agency theory is not to gotiate over the prices of their
consider the level of pay-for-perfor- materials) to show that when the mar-
mance but to address how coefficients ginal product of effort is higher, fewer
costs can be passed on. These predic-
other examples. Another strand of this literature tions are supportive of the agency
addresses the monitoring power of boards of di- concerns above.17
rectors. See Michael Weisbach (1988) and Ben
Hermalin and Weisbach (1988) for example. Many of the other studies testing for the
14 One interesting conclusion of this paper is trade-off between risk and incentives in
that the flow of payments of stock and options in a
contracts concern executive compensa-
given year is largely independent of the level that
a CEO has inherited entering that year. The fact tion, largely because of data availabil-
that the inherited level does not crowd out new ity.18 The results here are rather mixed.
issues of stock suggests that there is no common
First, John Garen (1994) finds little evi-
level of desired incentives.
15 See Jean-Jacques Laffont and Mohamed Ma- dence that the noisiness of performance
toussi (1995) and Chris Ferrall and Anthony Smith measures has any effect on contracts,
(1997) for a structural approach to identifying though Rajesh Aggerwal and Andrew
these unknowns.
16 Some simple calculations show that a 1 per- Samwick (1998) find evidence for such a
cent change in the stock price of a large U.S. com-
pany changes pay by over a quarter of a million 17 Banri Asanuma and Tetsuya Kikutani (1992)
dollars; although $3.25 per $1,000 may seem carry out a similar study on the Japanese auto
small, Fortune 500 firms are so large that this industry, and also find results supporting the
translates into large dollar sums. I have no way of trade-off of risk and incentives.
evaluating whether this is a large sum of money '8Though see Lee Alston (1981) and Alston and
for a CEO relative to the private benefits they get Robert Higgs (1982) for data on share-cropping
from "shirking." See Brian Hall and Jeff Leibman contracts, and Charlie Brown (1990) for more
(1996). aggregate data.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
20 Journal of Economic Literature, Vol. XXXVII (March 1999)

trade-off. Lambert and Larker (1987), Available evidence on relative perfor-

Robert Bushman, Raffi Indejikian, and mance evaluation has also focussed on
Abbie Smith (1996), and Chris Ittner, the compensation of executives.22
David Larker, and Madhav Rajan (1996) First, Richard Antle and Abbie Smith
test whether the weights placed on ob- (1986) find weak evidence that the com-
jective and subjective measures respond pensation of executives falls as other
to the noisiness of the objective mea- firms do better, holding own perfor-
sures. Straightforward manipulation of mance fixed, although their data set is
(5) and (6) illustrates that the weight small. Using a more comprehensive sur-
placed on subjective measures of perfor- vey of firms, Gibbons and Murphy
mance should increase in the noisiness of (1990) find that executives are indeed
the objective measures, while the weight penalized when a competitor group
on objective measures obviously falls.19 fares better, as predicted by the theory.
Lambert and Larker (1987) and Ittner, However, somewhat surprisingly, the
Larker, and Rajan (1977) find evidence relevant peer group seems to be the en-
in favor of this, though the results are tire stock market rather than companies
rarely resounding. For instance, Ittner, in the same industry. (One would imag-
ine that there would be more correla-
Larker, and Rajan find that the ratio tion in shocks within the same indus-
try.) Finally, they illustrate that the
is significantly increasing in y2 , but B3 is
degree of correlation between the mar-
not.20 By contrast, Bushman, Indijikian,
ket and the firms (i.e., the extent to
and Smith (1996) find little effect of
which there is a common shock) pre-
variability of objective measures on sub-
dicts the use of relative performance
jective contracts. Testing across many
evaluation. Murphy (1998) also notes
occupations, Brown (1990) also finds lit-
that direct observation of contracts
tle relation between the existence of
illustrates more extensive use of
piece-rate compensation schemes and
such evaluation than when inferring
the noisiness of those measures.21
contracts as above.23
19 Marianne Bertrand and Sendhil Mullanaithan
(1997) also consider how various means of incen- (1994) and Rachel Hayes and Shaefer (1997).
tive provision can act as substitutes for one an- Shaefer illustrates how pay-for-performance varies
other by showing that direct contractual incentives with firm size, where larger firms have lower f
for CEOs are increased in situations where take- coefficients due to risk aversion or liquidity con-
overs are less likely. In particular, they use state- straints. Hayes and Shaefer provide a useful con-
level variation in takeover laws to show that when tribution to understanding the effect of subjective
states pass legislation that makes hostile takeovers performance evaluation by showing that future
more difficu[t, firms respond by making their performance measures (such as earnings) can be
executives more financially liable for the returns predicted by previous discretionary compensation
of the firm. changes to chief executives. Their interpretation
20 It should not be surprising that the authors of this is that these agents are rewarded for taking
are more likely to find a stronger relationship from
the "right" actions even in settings where the
the ratio of the levels than with the level of a sin-
immediate objective returns do not arise.
gle measure. However, the absence of a statisti- 22 Though see Edward Fee and Charles Hadlock
cally significant PI suggests that the size of (1997),
the who note that managers in major news-
effect of noise on incentives is not huge. papers are more likely to be replaced when com-
21- One prediction of agency theory which is petitor newspapers increase circulation. This can
borne out in the data is that those workers on clearly be interpreted as relative performance
piece rates will typically earn more than those on evaluation, though since papers are substitutes in
fixed wages. Agency theory would predict this as a the product market, the circulation of other
return to risk (or rents to ability in the case of papers may simply be another measure of the poor
worker selection). John Pencavel (1977), Trond performance of the newspaper manager.
Peterson (1992), and Daniel Parent (1998) illus- 23 It should not be assumed that there is univer-
trate such differences. Also see Scott Shaefer sal agreement on the frequency of relative perfor-

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 21

To summarize this section, there is 2.2 Other Behavioral Responses

some evidence that contracts are de- to Compensation Schemes
signed to optimally trade off risk against
incentives. However, the' evidence is So far, we have considered only an
hardly overwhelming, with some studies agent's incentive to exert "effort." How-
showing the effect of noise on piece ever, compensation schemes often have
rates while others show little. Thus, unintended consequences caused by
while agents do appear to respond to agents changing their activities in other
incentives, it would not appear that ways that are beneficial to them but
on the margin, the risk measures that not to their employer. Therefore, a
have been considered are the true potential cost of pay-for-performance
constraining factors on the provision schemes is not only that they impose
of incentives. At one level, of course, risk on agents, but also that the agents
risk aversion (or at least liquidity can "game" the evaluation procedure to
constraints) must play a role, since their advantage. This arises because
otherwise agents could be offered piece m'any jobs are complex, in the sense
rates of 100 percent (i.e., sell the firm that many aspects of those jobs are hard
to the agent), but on the relevant to contract over. As a result, the use of
margin, the effect of risk appears to be explicit contracts could cause agents to
limited. Instead, perhaps the most focus too much on those aspects of the
striking aspect of observed contracts is job included in the contract to the det-
that the Informativeness Principle, i.e., riment of those that are excluded. A
that all factors correlated with perfor- couple of examples should illustrate the
mance should be included in a compen- nature of the problem. Consider the
sation contract, seems to be violated contract offered to Ken O'Brien, a foot-
in many occupations. For instance, ball quarterback, in the mid-1980s.
there are many measures of the produc- Early in his career, he had a tendency
tivity of an academic (such as publica- to throw interceptions. As a result, he
tions, teaching ratings, etc.) or a base- received a contract that penalized him
ball player (batting average, home every time he threw the ball to a mem-
runs, etc.), yet explicit contracts are ber of the opposition. However, while it
rarely written on those measures. The was the case that he subsequently threw
reason for this is not because these fewer interceptions, this was largely be-
measures are infinitely risky (as the cause he refused to throw the ball, even
previous section would require), but in cases where he should have done so.
rather that contracts can typically be As Joe Namath put it, "I see him hold
written only on a subset of activities onto the ball more than he should . . .
carried out by an agent, and rewarding I don't like incentive contracts that per-
agents on a subset of all things tain to numbers" (quoted in Brown
that they do can cause dysfunctional 1990). Or the practice used at AT&T,
behavioral r esponses, to which I now where computer programmers were re-
turn. warded on the number of lines of code
that they produced in their programs.
mance evaluation. See Aggarwal and Samwick Not surprisingly, this resulted in longer
(1998); Jason Barro and Robert Barro (1990); programs than was necessary. These ex-
David Blackwell, James Brickley, and Michael amples have the same conceptual fea-
Weisbach (1994); and Janakiraman, Lambert, and
Larker (1992) for empirical work finding little ture; agents can change the nature of
evidence on this incentive device. their activities in response to objective

This content downloaded from

13ffff:ffff:ffff:ffff:ffff on Thu, 01 Jan 1976 12:34:56 UTC
All use subject to https://about.jstor.org/terms
22 Journal of Economic Literature, Vol. XXXVII (March 1999)

contracts in a way that is privately their jobs are multi-dimensional, and

beneficial to the agent but harmful to opportunities for reallocation of tasks to
his employer. Following Holmstrom increase rewards are certainly possible.
and Milgrom (1991), this distortion has Despite this, most incentives for these
become known as multi-tasking. jobs are provided by explicit incentives
As a result of the danger of agents (primarily through stock holdings). The
overemphasizing objective criteria, it is reason is that aggregate measures of
typically argued that firms should not performance are available through, say,
pay based on objectively measured cri- the stock price return, which is rela-
teria, but instead should use subjective tively exempt from multi-tasking con-
performance evaluation. The attraction cerns. In situations where executives
of subjectively determined measures of are assessed on non-holistic measures,
performance is that they allow a more I provide evidence below that they
holistic picture of performance to be at- also behave in ways that are privately
tained, not possible with objective con- beneficial.
tracts. For instance, in the AT&T exam- Subjective assessments, however,
ple, a subjective assessment could also induce inefficient behavioral re-
reward for long programs only in cases sponses. The literature in both econom-
where those programs are warranted. As ics and more particularly in human re-
a result, for jobs without clear aggre- sources management has emphasized
gate measures of performance, rewards how incentives provided through sub-
tend to be allocated in a discretionary jective assessments cause agents to
fashion. change their behavior, and cause
Two examples are apposite here. supervisors to distort their reports, in
First consider the case of a baseball such a way that efficiency is harmed.
player. It is difficult to imagine an occu- The purpose of this section is to address
pation for which there are more mea- how objective and subjective signals
sures of performance. Despite this, it is should be used in situations where both
not common for players to have con- potentially induce inefficient responses
tracts where pay is directly related to in behavior. In order to highlight the
specific performance measures. Part of distinctive features of this section, I
the reason for this is that teams are re- restrict attention to the case of risk-
luctant to offer a contract that rewards neutral agents, so that any effects that
a player for home runs, say, because the arise are due to behavioral responses.
player may have an incentive to hit The effect of risk aversion in this set-
home runs even when it is not in the ting is largely additive, in the sense that
interest of the team for him to do so. By higher risk aversion reduces incentives;
contrast, the more common cases where since there are no interesting interac-
players are offered explicit bonuses are tions between behavioral responses
for aggregate measures of performance, and risk aversion, risk neutrality is
such as making the All Star Team or be- assumed.
ing the league's Most Valuable Player.
2.2.1 Multi-Tasking
Since these are more holistic measures
of performance, they suffer less from The essence of this section is that at
the multi-tasking dilemma. The second times agents will take actions other than
relevant example concerns chief execu- those the principal would like to
tive officers. No one could claim that induce. Since contracts are an imper-
their jobs are not complicated; clearly fect representation of the worker's

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 23

contribution to the firm, workers can Rather than model this approach to
"game" the compensation scheme to "gaming," I use a simpler structure that
their benefit. Holmstrom and Milgrom makes comparison to the basic model
(1991) address this issue by assuming easier and derives from Baker (1992),
that agents carry out multiple activities, where an agent chooses only a single ef-
and choose the allocation of activities fort decision, but there is a divergence
based on offered contracts. Consider a between the privately and socially opti-
simple setting where an agent chooses mal effort level. For example, an agent
between activities a and b, where the who is rewarded on quantity produced
cost of effort for the worker is such that may know that his effort is worthless if
he is indifferent to how they are allo- there is no demand for the goods,
cated between tasks (i.e., only the sum though it is in his private interest to
of effort matters).24 Then with the lin- produce. To model this, assume that
ear contract above, if the principal of- the expected marginal product of the
fers a contract with a higher return to agent is, as above, e + cX, but the ob-
one activity than the other, the agent jective measure on which the worker is
allocates all his effort to one task, even rewarded, y, is given by
if the principal would prefer the agent
q = ge + cc + F-9, (10)
to allocate his time to both tasks. In or-
where all variables are distributed as in
der to induce the agent to allocate time
Section 2.1. Throughout the paper a
to both tasks, the same incentives must
"tilde" over a variable refers to a cor-
be offered on both. But there need be
rupted version of the appropriate signal.
no reason why this is otherwise optimal;
The only difference from the basic set-
for instance, the measurement error on
ting is through ~i. Here the marginal ef-
one (e.g., quantity produced) may be
fect of effort on the indicator depends
much lower than for the other (e.g.,
on g, while true productivity is indepen-
quality produced). As a result, multi-
dent of that measure. Assume that g is
tasking imposes constraints on the
privately known by the agent so that the
trade-off between risk and incentives.25
marginal return to effort on surplus is
24 An important implication of this theory has
unity, but the expected private return
been implications for ownership of assets. To take depends on g because contracts are writ-
a simple example, consider a worker who is em- ten on y. Assume that g N(1,2), where
ployed on a piece rate. If he is employed by a
Firm, he may have an incentive not to takeoR is a direct measure of the extent to
due
which the agent can "game" the compen-
care of his machinery, as his incentives are simply
to produce as much as possible over a short period sation system. (In the previous section,
of time. By contrast, an agent who owns his Or = 0 )26
machinery has better incentives to exercise due
care. As a result, ownership can solve some multi-
tasking problems, and this result is also consistent
with the observation that the self-employed have on one task likely leads to less on others. Such
high piece rates, while those employed by firms substitution opportunities constrain the ability of
tend to have compensation that is less sensitive the firm to ofJer piece rate contracts.
to performance. This observation is based on 26 This section is not meant to be a detailed
Holmstrom (1996), and empirical work by Eric description of the theoretical implications of
Andersen and David Schmittlin (1984) is consis- dysfunctional behavioral responses; instead the
tent with this. Recent empirical work on franchis- results are meant to be illustrative. For example,
ing, such as Francine Lafontaine (1992) and in this hidden action game, the firm could design
Margaret Slade (1996), also emphasizes this revelation mechanisms where the agent reveals g
approach. to the firm, rather than restrict attention to linear
25 Agents reallocate efforts between activities sharing rules. As in Baker (1992), it is assumed
when efforts on tasks are substitutes or comple- that the firm cannot design such revelation mecha-
ments; for instance, if time is limited, more time nisms.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 24 Journal of Economic Literature, Vol. XXXVII (March 1999)

Number of Recruits Per Month

6 Evaluations carried out at times TI and T2

TI T2 Tenure

Figure 1. The Be

The problem with basing compensa- convex). The principal responds to

tion on y is that the marginal return toby muting incentives. Given this in-
this
effort depends on R. One solution iscentive, to the firm's expected surplus
condition compensation on a subjective maximization problem is equivalent to
measure of performance, which is ex- maximizing
empt from this problem. We defer
yq(l + R
2
this possibility until the next subsection
by assuming thatG 2=o, SO that only ob-
The optimal piece rate is then trivially
jective measures are used. Compensa-
given by
tion is based on a linear signal of the
performance measure,
PY 2 <1 (12)
1 + y2
w = PO+ py Y(1 if aj > 0. Thus, even with risk neutral
agents, incentives are below unity in or-
The agent optimally chooses effort
der to constrain inefficient behavioral re-
equal to e* = . (If ay2 = 0, this becomes sponses.27 In other words, firms mute

27This simple set-up implies that firms should

e* as in the previous section.) If
always offer some incentives to workers. However,
c > 0 note that the agent bases effort on in the multi-task setting in Holmstrom and Mil-
grom it is straightforward to show situations where
a measure uncorrelated with social sur- the firm is better off offering no incentives. Essen-
plus (which is costly as effort costs are tially, this requires that agents be willing to supply

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 25

incentives when workers can game the (ii) how close he is to the quota; this is
system.28 given by e + cx above. However, private
and social returns are unlikely to co-
2.2.2 Empirical Evidence
incide with a quota system of compen-
on Dysfunctional Responses
sation, where agents are rewarded peri-
to Compensation Schemes
odically. A number of distortions can
Many of the empirical studies illus- arise. First, if individuals are rewarded
trating inefficient behavioral responses on aggregates over long time periods,
to objective compensation schemes they may have an incentive to wait "un-
arise from the fact that many compensa- til the last minute" simply for discount-
tion schemes are nonlinear in perfor- ing reasons. Asch (1990) considers such
mance measures or use aggregates over incentives for Navy recruiters to reallo-
long time periods, such as a year. Paul cate effort over time. The recruiters are
Healy (1985), Beth Asch (1990), Paul offered pay-for-performance (based on
Oyer (1998), and Pascal Courty and number of recruits) through the pros-
Jerry Marschke (1996) consider various pect of either speedier promotion or in-
forms of agency behavior under such creased probability of further tours of
compensation systems. Consider a situ- duty. Her primary interest is in identi-
ation where an agent gets a discrete fying how the recruiters allocate their
prize for reaching a quota by some date, efforts across time in response to this
say December 31. The agent must incentive scheme. Figure 1 illustrates
choose effort throughout the year based the average number of recruits over the
on the probability of reaching that evaluation period.
quota by the end of the year. The social Incentives are provided at the pre-
surplus of his efforts is independent of specified dates T1 and T2. Prizes vary by
(i) when he carries out the activity and the number of recruits, but contract
termination dates are specified in ad-
vance so that the recruiters know by
some effort even when there are no explicit
contracts. when they must reach their targets. Fig-
28 Firms must typically determine not only how ure 1 shows performance gradually in-
to pay their employees but also what they let them creasing until the evaluation date and
do, i.e., the must choose task assignments. Holm-
strom and Milgrom (1991) use multi-task logic to then discretely falling after the evalu-
identify the optimal allocation of workers to safe ation period. This suggests intertempo-
and risky jobs. In order to reduce the possibility of ral effort reallocation in response to the
the worker misallocating activities across tasks,
Holmstrom and Milgrom show that the optimal compensation scheme. Oyer (1998) pro-
solution to this problem may be to allocate only vides similar evidence on the effect of
risky tasks to some agents and only safe tasks to intermittent rewards based on quotas
others. Another aspect of job design concerns the
amount of authority offered to workers. Philippe on the average levels of effort for
Aghion and Jean Tirole (1997) address this by salesforce workers.29
noting that incentive concerns may induce "exces- Similar effort reallocation would, of
sive" delegation of authority to workers. Consider
the problem of a firm that would like its subordi- course, occur in settings where there
nates to exert effort, but where rewards from are linear contracts with infrequent
doing so are not through wages, but through doing evaluation; people simply prefer to wait
desirable tasks. Workers may fear that their deci-
sions will be overruled by a superior, so that the
agent does not get the returns to exerting effort. A 29 0yer also illustrates that the "threshold" ef-
solution to this problem is to give the worker fect of rewards (i.e., that they are nonlinear) can
authority over the decisions made, so that he real- either increase or decrease average effort levels
izes that if he exerts effort, he is likely to see the over the evaluation period, depending on the
return from doing so. difficulty of attaining the quota.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
26 Journal of Economic Literature, Vol. XXXVII (March 1999)

Optimal Number to Terminate

: ~~~~~~~~~~I.

# of Candidates
S S(1+niN)

Figure 2. Terminating

to exert effort. However, another char- a bonus if they attain certain standards
acteristic of quota systems is that incen-by June 1 of each year. For example,
tives vary by whether the agent is close the agency could be rewarded if 40 per-
to the evaluation quota. In particular, cent of its trainees attain jobs. Criti-
an agent who is close to winning the cally, the Department of Labor (which
prize will have greater incentive (~t highadministers this system) offers these in-
in the terminology of the previous centives as a function of "graduated"
model) than one who has either ex- employees (i.e., those clients who have
ceeded the quota or is unlikely to reach finished the training program). But the
that quota (,u low in both cases). Evi- agencies can decide when to graduate
dence on this is provided in Healy them. As a result, the agencies have an
(1985), Oyer (1997), Courty and Mar- incentive to strategically graduate em-
schke (1997), and Andrew Leventis ployees. Consider a case where at June
(1997). Courty and Marschke consider 1, the agency must decide how many of
the effect 'of incentive contracts offeredn unemployed candidates to graduate
to agencies that provide job training for where it has already graduated N dur-
individuals on welfare. Job training for ing the year, and must place a percent-
welfare recipients is typically carried age of s in employment in order to
out by private agencies, which are of- achieve a bonus. Figure 2 illustrates the
fered incentives for desirable outcomes. strategic incentives.
In particular, these agencies are offered If the agency has not reached its

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 27

target or has exceeded 1 + N of its tar- efforts (or activities) in response to the
N
get, it will graduate all n to maximize compensation scheme that they face.
returns (assuming that these are less Keith Brown, W. Harlow, and Laura
employable than new trainees, as seems Starks, (1996) and Judith Chevalier
reasonable). However, if the agency and Glenn Ellison (1997a) consider
slightly exceeds its standards, it will the likelihood that agents will change
be more reluctant to graduate, since the riskiness of their activities in re-
graduating all no longer means getting sponse to incentives. They consider in-
the bonus. Thus graduations close to efficient risk taking by mutual fund
June 1 will be lower for success rates managers.30 Mutual fund managers are
close to the target, as illustrated in typically rewarded as a linear function
Figure 2. This plots the non-monotonic of the assets that they control. How-
relationship between privately optimal ever, the flow of funds into mutual
number of graduations and candidates. funds is not a linear function of perfor-
Courty and Marschke empirically illus- mance (particularly for younger funds,
trate that the agencies do exhibit this shown here). Instead, as illustrated in
behavior. Chevalier and Ellison, the flow of funds
Healy (1985) illustrates that execu- is given in Figure 3.
tives exhibit similar strategic behavior The important aspect of this figure
in reporting earnings when they are re- for our purposes is its nonlinearity.
warded as a nonlinear function of earn- Over some ranges, the shape of the re-
ings. Consider compensation schemes turns function is concave, while over
that reward for earnings only over some others it is convex. These data reflect
range, where there is a floor below the propensity for individuals to place
which compensation cannot fall and a funds in mutual funds based on pre-
ceiling above which it cannot rise. vious performance. In those regions
Healy shows that executives who have where the payoffs are concave, there is
already reached their ceiling do not re- an incentive to take inefficiently few
port all their earnings. Similarly, they risks, while convexity implies an incen-
do not report all earnings if they have tive to take too much risk. For example,
earnings so low that they are unlikely to for those funds that perform 20 percent
reach the region where they earn posi- worse than the risk adjusted market re-
tive marginal returns. Leventis (1997) turn, there are incentives to avoid
considers the response of surgeons to (local) risk, while for those performing
incentives. In New York, surgeons are 10 to 15 percent better than the mar-
penalized if their mortality rates exceed ket, there is an incentive to increase
a threshold. They respond by taking less risk. Chevalier and Ellison estimate
risky cases as they approach that thresh- the risk taking of mutual fund managers
hold. Finally, Oyer (1997) finds empiri- at the end of the year (as measured
cal evidence on another implication of by the standard deviation of the differ-
quota compensation schemes; namely, ence between fund return and the mar-
that there will be greater variability in ket) as a function of performance until
sales at the end of the financial year
than at other times. 30 Note that allowing agents to choose the riski-
Behavioral responses are not specific ness of their outcomes violates the assumptions of
to quota systems. A number of other the model above. In the context of risk taking,
nonlinear incentives are generally optimal. See
authors have directly considered the in- Holmstrom and Joan Ricart i Costa (1986) for a
centive for agents to reallocate their model on contracting in the context of risk taking.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
28 Journal of Economic Literature, Vol. XXXVII (March 1999)

Expected Flow in the Following Year

I I I I W Performance
-20% -10% 10% 20%

Figure 3. The Flow of F

September. For example, those funds described above, Anderson, Burk-

that are 20 percent worse than the hauser, and Raymond (1993), Heckman,
January-September risk-adjusted mar- Heinrich, and Smith (1997), and Courty
ket return are predicted to take too lit- and Marschke (1997) show that when
tle risk in the October-December pe- the training agencies are rewarded on
riod, while those that perform 15 their success in placing trainees in jobs,
percent better have an incentive to take they "cream skim," i.e., they recruit
excessive risks. They show that the only the most qualified candidates
agents do allocate assets in this way.3' rather than the most needy. Marschke
This risk taking occurs at the cost of (1996) additionally shows that when
lower risk-adjusted returns, suggesting these agencies are rewarded on certain
a divergence between social and private criteria, they focus more on the types of
returns. training that induce these outcomes,
Other work has suggested different though at the cost of other types of de-
dimensions on which agents respond to sired training. In a sports setting, Brian
compensation schemes. For instance, in Becker and Mark Huselid (1992) show
the context of the job training setting that increases in prize money among
professional auto drivers result in more
31 Excessive risk taking by the high performers risky driving, as witnessed by more cau-
is statistically significant only using portfolio time tion flags. Finally, Robert Drago and
series data. When examining the set of funds for
which portfolios themselves can be observed, this Gerald Garvey (1997) use Australian
effect becomes insignificant. survey data to illustrate that when

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 29

agents are placed on individual pay-for- profits, any wages offered to the agent
performance schemes, they are less come from his pocket. Thus, even
likely to help their coworkers. though an agent exerts effort and per-
In summary, this work suggests that forms well, the supervisor may claim
firms get what they pay for; by empha- otherwise to keep costs down.32 An ex-
sizing certain outcomes in pay, they ample where measures may be manipu-
make those outcomes more likely to oc- lated in order to reduce costs is the
cur. Urnder some circumstances, this in- movie industry, where actors are some-
creases productivity (as the studies in times paid on the "net profits" of a film.
Section 2.1 would suggest), yet in more As a result, there have been numerous
complex settings, this can result in a court cases regarding "creative account-
reallocation of activities that is not ing" designed to keep net profits low
obviously efficient. even for apparently successful films.33
See Carole Cheatham, Dorothy Davis,
2.2.3 Subjective Performance Evaluation
and Leo Cheatham (1996) for more de-
Because of such multi-tasking con- tails on this. Yet theft is not solely the
cerns, firms are often reluctant to offer privilege of principals. Consider the
rewards based on objective measures of case of a cab driver who leases his cab
performance. Instead, they prefer to
use subjective measures, where pay is at 32 The theoretical literature has suggested a
the discretion of the impressions of a number of solutions to the danger that a principal
superior. The attraction of such means will underreport to save on wage costs. First, Clive
Bull. (1987), Bentley MacLeod and Malcomson
of payment is that they offer a more ho- (1989), and Baker et al. (1994) consider a role for
listic view of performance; the agent repeated interaction between the principal and
can be rewarded for a particular activity agent as a means of reducing incentives to renege.
Here the principal remains -onest because a fail-
only if that activity was warranted at ure to do so involves retaliatory action through a
the time. For example, in the case of future failure to provide effort. Second, in a static
the football quarterback who stopped setting, firms can at times be induced to act hon-
estly by imposing other contractual costs on them
tlhrowing the ball when confronted with from claiming performance is poor. Charles Kahn
a payment scheme that penalized him and Gur Huberman (1988) show how the use of
for doing so, suitable adjustments could up-or-out contracts, where a worker is either
retained at a high wage or fired, can solve this
be mnade with a subjectively determinedproblem in settingis where efforts involve training.
performance measure. However, the es- In that case, if t e principal claims poor perfor-
sential feature of subjective assessmnentsmance, she must fire the agent, which may be
more costly than retaining him at the high wage.
is that they cannot be verified by out- (See Dominique Demougin and Aloysious Siow
siders (or at least it is costly for third (1994) for another interpretation of up-or-out
parties to determine performance), rules.) Canice Prendergast (1993) makes a similar
point in the context of promotions. Finally, Jan
which gives rise to the possibility that Zabojnik (1997) argues that firms can mitigate the
performance measures will be manipu- incentive of supervisors to steal deserved wage
lated or distorted from their true val- payments by designing contracts that emphasize
revenues rather than profits.
ues. Such distortions can arise for a 33 For instance, "Forrest Gump" supposedly
niumber of different reasons, which I earned negative net profits despite being the
now describe. fourth highest grossing movie of all time. It is
claimed that manipulation of the net profit figures
Theft, One danger of assessments occurred because the writer had been promised 3
that are subject to manipulation is that percent of net profits. James Garner successfully
a principal will underreport perfor- sued Universal Studios when his 37.5 percent of
net profits from "The Rockford Files" turned out
maiice in order to save on wages. If a to be worth zero. These cases are described in
supervisor is also residual claimant on Cheatham, Davis, and Cheatham (1996).

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
30 Journal of Economic Literature, Vol. XXXVII (March 1999)

on a daily basis. In most US cities, the Landy and J. Farr (1980), A. Mohrman
cab driver has a piece rate of 100 per- and Edward Lawler (1983), Kevin R.
cent; he pays a fixed fee for the cab and Murphy and Jeannette Cleveland
keeps all revenues. This compensation (1991), and Patrick Larkey and
scheme is not used because driving a Jonathon Caulkins (1992) document
cab has little risk; the demand for cabs negligible difference in ratings and
clearly depends on such variable factors compensation across workers.35 This re-
as weather. Instead, cab drivers typi- duces the value of subjective assess-
cally keep all their revenues because ments as a means of providing incen-
they can manipulate output, as true out- tives, since the relationship between
put cannot be observed. More specifi- effort and pay is clouded by other
cally, they can turn off the meter and influences.
negotiate a fare with the passenger, as This literature also points out that
occurred in situations where piece rates such compression is more severe in
less than unity were used. The most ef- situations where ratings are important
ficient (static) solution to this problem for pay setting: supervisors are reluc-
is simply to let the driver keep all reve- tant to impart bad news to workers if it
nues, as he no longer has an incentive means salary adjustments. Ironically, an
to privately contract. implication of this is that many firms
Compression of Ratings. There is now explicitly separate pay setting from
considerable evidence in the personnel subjective evaluations. According to
literature that supervisors distort sub- George Milkovich and Alexandra Wig-
jective performance ratings by not suffi- dor (1991, p.109), "A traditional rule of
ciently differentiating good from bad thumb among managers has also sug-
performance in their ratings. In this gested the wisdom of decoupling the
scenario, the supervisors are themselves appraisal process from merit pay . . .
agents, who have incentives to treat [The] concern has been that managers
workers in ways not desirable to the will deliberately inflate performance
principal when offering evaluations. appraisal rating to distribute merit pay,
Two relevant forms of compression are thus decreasing the chances that
noted in this literature: "centrality bias" employees with real training needs will
and "leniency bias." Centrality bias re- be identified or increasing the chances
fers to a practice where supervisors of- that overrated employees will be pro-
fer all workers ratings that differ little moted beyond their capacities."36 From
from a norm. Leniency bias implies that
supervisors simply overstate the perfor- 35 Direct evidence on leniency in ratings is pro-
mance of the poor performers.34 Such vided in H. F. Rothe (1949) and E.A. Rundquist
and R. H. Bittner (1948), while Leonard Ferguson
compression is well documented in the (1949) and Lee Stockford and H.W. Bissel (1949)
personnel literature, where Frank illustrate that such leniency is exacerbated when
the supervisor knows the subordinate for a long
time.
34An obvious reason for this is that it is simply 36 There is almost no empirical work in the eco-
unpleasant for supervisors to offer poor ratings to nomics literature on this topic (though see James
workers, so they avoid this pain. It is also worth Medoff and Katherine Abraham 1981 and Baker,
pointing out that such compression need not be Gibbs, and Holmstrom 1994a,b for indicative evi-
inefficient in a dynamic setting. For instance, sup- dence). The only example I know of is a Harvard
pose that a worker performs poorly. Telling the Business School case by Kevin J. Murphy on the
worker that their performance was poor can easily compensation practices of Merck in the mid-
result in discouragement, say because they feel 1980s. During this time at Merck, supervisors
that their promotion prospects are low. As a result, were required to rate on a 1-5 scale, yet 97 per-
firms may prefer to reveal little information. cent of workers were offered ratings of 3 or 4,

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 31

this perspective, appraisals are largely how Navy supervisors admitted distort-
used for training purposes rather than ing performance ratings in order to in-
to allocate rewards to those exerting crease the prospects of their preferred
most effort. subordinates. Such activities have two
Rent-Seeking Activities. In most large possible distortions. First, agents de-
organizations, supervisors are not resid- vote time and energy to "sucking up"
ual claimants on output, and so the in- that would be better spent on produc-
centive to underreport performance to tive tasks. Second, information may be
keep costs low may not be critical. A inefficiently collected on individuals, so
more pertinent problem with subjec- it may be difficult to determine whether
tive assessments in large firms may be good performance ratings derive from
the danger of "rent-seeking activities," favoritism or from genuinely good per-
which refers to any actions that agents formance.38 As a result, firms may not
carry out that are designed to increase know who to promote or reward.39
the likelihood of better ratings from su- Contracts under Subjective Perfor-
pervisors, but that have less value on mance Evaluation. In this section, I
surplus than some other activity that consider the rent-seeking effects of
they could carry out. This has been the
primary focus of the economics litera- 38 In many situations there are simply no objec-
tive measures of output that can used to reward
ture on subjective performance evalu- the agents and supervisors. Other problems arise
ation. The relevant theoretical work on here. Consider the case of a figure skating judge
this issue includes Holmstrom (1982), How would one give incentives for a judge to offer
honest assessments of the performance of a
Milgrom (1988), Milgrom and Roberts
skater? There are essentially no objective mea-
(1988), Tirole (1992), and Franklin Al- sures of output; instead, the only available
len and David Gale (1992). At this level measures are the subjective assessments of other
judges. Yet agency concerns abound in these set-
of generality, there is nothing that tings, and a solution used to evaluate judges is to
makes influence activities specific to compare assessments with the assessments of
situations where evaluations are subjec- other judges. A judge is then penalized if she re-
peate y differs from the average assessments of
tive.37 However, a central theme of this other judges. This is not an iso[ated case; similar
literature, which arises with subjec- problems arise with evaluating academic candi-
tively evaluated schemes, is that super- dates, art and film critics, and so on. This gives
rise to the problem of "yes men," whose objective
visors will misreport when evaluating is simply not to look different from anyone else,
workers. For instance, workers may curry leading to less efficient evaluations. See Prender-
favor in such a way that supervisors ex- gast (1993) for details. Another example of the
absence of output measures concerns situations
aggerate performance, or, as in Tirole where diagnosis is important, as in Curtis Taylor
(1992), simply bribe their superior for a (1995). Consider the market for auto repair. There
better evaluation. As a result, supervi- are few measures of output in an auto repair shop
if the car is not broken; all that is known is that
sors exhibit favoritism to those who the car leaves in working order, not that the mar-
spend the most time currying favor. For ginal product of the auto shop was high. As a re-
instance, D. Bjerke et al. (1987) note sult, an ineffiency can arise where the auto shop
will claim repairs are needed even in cases where
they are not.
where there were negligible differences between 39 It is not necessarily the case that trades be-
the two groups. Management felt that this discre- tween supervisors and employees are inefficient.
tion offered to supervisors offered few incentives For instance, Prendergast and Robert Topel
to their workers. (1996) consider a model of favoritism that illus-
37 For instance, a worker could exert effort trates the costs considered above; also, these
on visible tasks (Jonathon Paul 1992) at the ex- trades have benefits in that they offers power to
pense of those that are truly productive. Or he supervisors, which they value. See Holmstrom and
could work "too hard" in order to make a good Milgrom (1990) and Hideshi Itoh (1992, 1993) for
impression (Holmstrom 1982). other work on efficient side-trades.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
32 Journal of Economic Literature, Vol. XXXVII (March 1999)

subjectivity. Suppose that in addition to and

the corrupted objective measure y, 2

agents can also be rewarded on a subjec-

tive assessment s made by a supervisor.
(1 + a2 )(1 + c)-1 (15)
In keeping with the rent-seeking litera-
ture, it is assumed that s need not equal These piece rates illustrate the trade-off
s, the true evaluation. Instead, the agerntof gaming and influence. Here risk neu-
can ingratiate himself to his supervisors trality no longer guarantees efficient ef-
by carrying out a bias activity b, where fort unless either the agent cannot exert
he can induce the supervisor to make a influence (K = o) or there is no incentive
report which is b higher than s at a per- to "game" the objective scheme (G2 = 0).
Subjective assessments rise with K; i.e., as
sonal cost of K(b)= -2Kb 40 Therefore, on the cost of influence activities increases,
the margin "sucking up" has a payoff, firms will rely more on the subjective
though the firm realizes that the agent is measures. As above, the objective mea-
carrying out such activities in equilib- sure's use falls with ay2. In the case
rium. The report made by the supervisor where there are no objective signals that
on the worker is s = s + b , so s is the cor- can be used (6, = oo), the optimal choice
rupted version of s. Let the compensa-
of gBs= _ c This view of compensation
tion contract offered to the worker be
+ K
given by
contracting shows how pay-for-perfor-
w = ,Bo + pBq y + P,3. (13)mance is constrained not by the risk-
The firm now must choose compensa- sharing considerations of Section 2.1, but
tion weights for an objective measure rather by the behavioral responses of
that is subject to gaming opportunities agents. The use of objective measures
and a subjective measure subject to has the drawback that agents allocate
influence activities. The worker now their efforts at the wrong time (i.e.,
makes two choices, e and b, both of based on R), while subjective assessments
which depend on the contract offered. waste resources on ingratiation.41
He optimally chooses Empirical Evidence on Subjective
Contracts. A primary focus of the per-
e*(040y>S)= c and b*(g43Kfs)= 4+ - sonnel literature is on the design and
Therefore, increases in K make influence implementation of contracts for workers
more costly and hence less prevalent. whose output is not easily observed.
The returns to the supervisor from the The issues that arise in this empirical
rent-seeking are assumed to be negli- literature concern optimal discretion
gible, and routine calculations show that offered to supervisors, the use of
the optimal contract is characterized by bureaucratic rules (such as maximum

(1 + )(1 + K)-
the "piece rates" pay increases allowable within job

41 This view of influence incentives considers

c
how changes in effective piece rates on subjective
measures induce influence. However, other vari-
py= (14) ables may play a similar role. Margaret Meyer,
Milgrom, and Roberts (1992) and Shaefer (1994)
argue that the financial performance of firms may
also induce changes in influence activities. For ex-
ample, firms in decline may be laying off workers,
40 The costs of effort, e, and bias, b, are inde- so that the returns to influence could rise if there
pendent purely for simplicity. are considerable rents from retaining a job.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 33

grade), and the costs and benefits of ties carried out by the agent reduces
different evaluation schemes. A disap- the likelihood of piece rate pay, consis-
pointment of the economics literature tent with the theory above. MacLeod
has been the paucity of information col- and Parent also show that piece rate
lected on the evaluation of workers with jobs typically involve few tasks, while
poorly measured output. Despite the those with many tasks are character-
fact that most workers in the economy ized by subjective assessments of
are evaluated subjectively, the econom- performance.
ics literature has largely focused on the
2.3 Tournaments
aggregation of observed objective sig-
nals.42 While we have learned much So far, attention has been restricted
from this literature, the set of workers to individual compensation schemes
with easily observed output is a small where the level of pay varies with the
fraction of the population. performance of the agent. However, in
Of course, many of the insights that many situations agents exert effort in
govern compensation under objective order to get promoted to a better paid
measures also hold when workers are position, where the reward associated
subjectively assessed. Despite this, the with that position is fixed and where
observations above make clear that there is competition between agents for
there are many other influences at play those positions. For instance, Gibbs and
with subjective assessments. For in- Hendricks (1996) use the personnel
stance, if supervisors refuse to differen- files of a large firm to illustrate little
tiate between the good and bad per- variation in pay within job grades based
formers, the insights of Section 2.1 on performance; instead, most pay in-
regarding trading off risk against incen- creases occur through job (or at least
tives are probably of less importance job title) changes. Since a large propor-
than might be the case, as there is little tion of wage changes are associated
variance in performance measures any- with such promotions, a central theme
way. There is a need to collect more in- of the economics literature, following
formation about the evaluation and Lazear and Sherwin Rosen (1981), has
compensation of the worker for whom been to examine the incentive effects of
output measures are hard to obtain. Re- promotion schemes via tournament
markably, the only studies in economics theory. This section briefly describes
that I am aware of that address pay and the main themes of this literature.
evaluation for such workers are Brown Promotions are used for many differ-
(1990) and MacLeod and Daniel Parent ent reasons, perhaps the most impor-
(1998). Brown considers the determi- tant of which is to sort workers on the
nants of standard rate pay, subjective basis of their talents. For instance,
merit pay, and piece rates for a large Rosen (1982) illustrates how a competi-
sample of workers. His most robust tive labor market allocates workers to
finding is that a greater diversity of du- different positions based on their tal-
ents, and rewards them accordingly. An
42This is not to say we know little about how implication of such allocation decisions
wages change within firms. There is considerable
work using personnel files (such as Baker, Gibbs, is that promotions also have incentive
and Holmstrom 1994a,b; Michael Gibbs 1995; and effects, and a common theme of the lit-
Gibbs and Wallace Hendricks 1996) illustrating erature has been to address the incen-
how job tenure and promotions affect wages. What
is not known is how these compensation decisions tive effects of promotions, through the
are made. lens of tournament theory. Tournament

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
34 Journal of Economic Literature. Vol. XXXVII (March 1999)

theory considers a group of agents com- obvious, this aggregation rule turns out
peting for a fixed set of prizes. The not to be efficient when dysfunctional
prizes are specified in advance and behavioral responses arise; this will
agents exert effort to increase the likeli- become clear below.
hood of winning a better prize. Rather All that matters for rewards and
like a sports game, all that matters for hence effort decisions is relative perfor-
winning is not the absolute level of mance. Accordingly, note that the dis-
performance, but how well one does tribution of zI - Z2 N(el - e2, (2 + 2?2).
relative to others. Assume that the agents are risk neutral.
I begin by considering the simple Then each agent exerts equilibrium
analytics of tournament theory. To do effort until
so, consider two agents 1 and 2, who
i* W- L a[zi zjle, e2] (17)
exert efforts e1 and e2 respectively un-
c aei
der exactly the same circumstances as
Since each is perceived to be equally
in the section on risk sharing, where
likely to win, the marginal change in the
signals y and s are observed on each
probability of winning is the density of
agent, which I call yi and Si for agent i.
the distribution of ZI - Z2 evaluated at
(None of the distortions associated with
zero. This implies equilibrium effort of
multi-tasking or subjective performance
evaluation are initially considered.) W-L
They compete for two fixed prizes. To CA2nt(62+ 2+2) (18)
simplify further, I assume that the two
Therefore, the agent's effort is in-
signals are equally valuable so that
creasing in the size of the prize and
s= y= 2. The principal designs a in the efficiency of monitoring. Because
tournament in this setting by choosing
(i) a prize to be given to the winner, W, the optimal level of effort is 1, the
firm sets the optimal prize W* - =
(ii) a prize given to the loser, L, and
i22t(y2 +272) to induce the first best
(iii) a rule that determines who the win-
level of effort. Thus, as illustrated in
ner should be. Since both signals are
Lazear and Rosen (1982), the principal
equally valuable, the optimal rule for
has induced the first best level of effort
determining who wins the prize is
through the use of a tournament.
simply that agent 1 wins if
Empirical Tests of Tournament The-
y I+ Sl y2 +S2 ory. Tournament theory offers a num-
Zi = 2 - 2 Z2. (16) ber of testable implications. First,
Otherwise, agent 2 should be awarded greater prizes lead to more effort. A
the winner's prize. As in the section on number of authors have verified this
risk-sharing, this rule is nothing more prediction, typically from the sporting
than optimally aggregating information arena. First, Ron Ehrenberg and Mi-
on performance and then awarding the chael Bognanno (1990) illustrate that
prize to the worker who has highest ex- professional golfers on the European
pected effort.43 While this may appear circuit have lower scores when the prize
money for which they compete in-
43 Where workers are risk neutral and there are creases. They illustrate this both by
no allocation effects of promotion, it actually
doesn't matter which (symmetric non-degenerate)
aggregation rule is used, as the wage spread can agents are risk averse, or the firm is allocating
be changed to counter any inefficiencies in the ag- the most able workers to more responsible jobs, the
gregation rule at no cost to wages. But this result firm strictly prefers this aggregation rule to any
is special; it occurs only in this case. If either the other.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 PrenderLiast: The Provision of Incentives in Firms 35

looking across tournaments (where dif- number of individuals competing at the

ferent tournaments have different prize next rank below. Brian Main, Charles
money) and by observing the incentives O'Reilly, and James Wade (1993), Tor
of players who start in different posi- Eriksson (1996), and Martin Conyon
tions beginning the final round.44 and Simon Peck (1997) illustrate that
Becker and Huselid (1992) show that more competitors do indeed increase
higher prizes result in faster (though the prize for becoming CEO. Another
riskier) driving by professional NASCAR prediction of agency theory is that bi-
drivers. One of the few tests of tourna- ased tournaments (where one agent has
ment theory outside a sports setting is a greater chance of winning than the
Charles Knoeber (1989) and Knoeber other in a two-person setting) result in
and Walter Thurman (1994), who study lower effort, and for those who fall be-
the broiler chicken industry. Large hind, excessively risky behavior.47 I
broiler companies reward farmers on a know of no empirical work that illus-
relative performance metric rather like trates that biased tournaments reduce
tournaments, in order to filter out im- incentives, other than casual observa-
portant common risk due to such fac- tion in sports tournaments of teams
tors as disease. As predicted by the "giving up" when they are behind at the
theory, higher prizes result in better end of the game and where both win-
performance, here measured by the ning and losing teams replace their best
weight of chickens. players with substitutes. Knoeber and
Of course, these tests are simply Thurman (1995) illustrate that broiler
more evidence on whether incentives farmers who are unlikely to win the
matter, not whether contracts are de- tournament do indeed take riskier ac-
signed with these responses in mind. tions in order to improve the prospects
However, a prediction of tournament of winning.
theory is that in settings where there is A problem with tournaments is that
a single winning prize,45 the prize for since individuals are evaluated on how
victory is increasing in the number of well they do relative to others, they are
competitors.46 Empirical tests of this unlikely to help their competitors in
proposition have been carried out on need. This point is theoretically illus-
executive data, where the return to trated in Lazear (1989) and Raffi Rob
becoming CEO is increasing in the and Peter Zemsky (1997).48 Drago and

47In the context of the model above, suppose

44 The payoff in such tournaments is convex, that one agent is A more able than the other.
where for instance the payoff for coming first Then the marginal incentive to exert effort is
rather than second is much higher than from com- (W - L)f(A) < (W - L)f(O), where f is the density of
ing 34th rather than 35th. Thus position starting relative luck.
the final round offers different incentives. 48 A related point arises in Lorne Carmichael
45 This is the case that has been studied empiri- (1988) and Guido Freibel and Michael Raith
cally, where the interest has been on one agent (1997). They describe a situation where agents
acceding from senior executive to CEO. worry about hiring good colleagues where there
46 This result relies on the distribution of the could be competition for available slots. In the
measurement errors being single-peaked at zero. absence of some constraint on competition or the
The marginal return to effort can be parameter- allocation of rents, those who choose new workers
ized by the density of the distribution of the will be likely to prefer the less able, as they do not
measurement errors evaluated at the equilibrium constitute such strenuous competition. Carmichael
probability of promotion. Then as the number of argues that tenure for academics is a solution to
competitors rises, the marginal effect of effort this problem, as the positions of the insiders are
falls, since the density is lower as promotion be- already guaranteed. Freibel and Raith instead
comes less likely. As a result, the prize must rise to focus on restrictions on the communication of in-
compensate. formation as the optimal solution. In both cases,

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
36 Journal of Economic Literature, Vol. XXXVII (March 1999)

Garvey (1997) find evidence consistent which must assign workers to tasks in
with this using survey data from the firms based on comparative skills, inter-
Australian manufacturing sector. They act with the provision of incentives for
show that when agents report promo- workers.
tion incentives to be strong, they are In the context of the standard model
less likely to let others use their with risk aversion in Section 2.1, there
equipment, tools, or machinery. is little reason why the firm should pay
Why Are Tournaments Used? The solely on the basis of relative output, as
available evidence suggests that to a occurs in tournaments.49 While agency
large extent, firms primarily provide in- theory suggests that relative perfor-
centives through the prospect of promo- mance should be used in situations
tion (Baker, Gibbs, and Holmstrom where there is common risk, it is only in
1994a,b; and Gibbs and Hendricks very special cases that the optimal
1996), where higher wages can only be means of compensation involves only
attained through changing ranks. relative performance evaluation (Dilip
Rather surprisingly, there is very little Mookherjee 1984), as occurs in tourna-
work devoted to understanding why this ments. Intuitively, there is information
is the case, i.e., why the optimal means on effort from the worker's absolute
of providing incentives within large performance, independent of his rank,
firms (at least for white-collar workers) which is all that matters for tourna-
seems to be tournaments rather than ments. Given this, why are they so
the other means suggested in the popular?50
previous sections. A related reason to filtering out com-
An important function of promotions mon shocks is that evaluators often can-
is in sorting workers to jobs. Promotion not place a number on the performance
in many firms takes the form of a job of a worker, but are capable of making
change, in the sense that responsibili- rank order comparisons. Thus, all that is
ties increase with ability. While the is- necessary to carry out evaluations of
sue of sorting workers to jobs has been workers is to determine which worker is
studied at some length (Rosen 1982; better. In addition, since prizes are
Michael Sattinger 1993), the interaction fixed, it is not necessary to determine
between incentives and sorting remains how much better one worker is than an-
little understood. At a very general other; all that is needed is rank order
level, it appears that promotion can "kill information. While this answer seems to
two birds with one stone," as it both im- have some plausibility, it is hardly com-
proves the allocation of talented work- plete. For instance, firms frequently
ers to jobs and provides incentives have to make decisions based on the ab-
(Baker, Jensen, and Murphy 1988), but solute performance of workers: for ex-
the exact mechanics of this remain un- ample, should they respond to a wage
clear (though see Prendergast 1993; and
Dan Bernhardt 1995). To phrase this 49 It is of course true that with risk neutral
another way, we know relatively little workers who carry out one activity, the contract
about how internal labor markets, above gives rise to the first best. However, so do
many other contracts, so why are tournaments
typically chosen?
these distortions arise because there are restric- 50 In the description above, tournaments are ef-
tions on the ability of agents to make efficient fectively com petitions between agents. However,
monetary transfers. For instance, if the insiders an equally valid interpretation of promotion has
could sell their positions to the newcomers, effi- agents competing against a fixed exogenous
cient allocations would arise. threshold, such as a tenure standard.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendereast: The Provision of Incentives in Firms 37

offer from a competing firm? Since they tinguish the performance of the able
can make such absolute comparisons in from the less able. Instead, perfor-
this setting, it remains a mystery why mance ratings are compressed around
they are ignored in the provision of some norm. One advantage of tourna-
incentives. ments is that they force managers to
Another reason provided by the lit- make decisions; they no longer have the
erature for the use of tournaments, "luxury" of paying everyone the same
where prize structures are fixed, is that (or close to the same). Second, as men-
it avoids the possibility of the firm re- tioned above, promotions are linked to
neging on paying wages. In the previous changes in responsibilities. An advan-
subsection, I mentioned that when per- tage of tying wages to responsibilities,
formance is subjectively assessed, it pointed out in James Fairburn and Mal-
cannot be verified to a third party. As a comson (1996), is that it may cut down
result, there is a danger of the firm on influence activities. Assume that a
underreporting performance in order supervsior could simply allocate a fixed
to save wages. As pointed out in pool of bonus money in any way she
Carmichael (1983), firms can avoid this chooses, independent of job assign-
by committing to a prize structure, as in ment. Then she is particularly suscepti-
the model above, where the prizes W ble to rent-seeking activities or outright
and L are prespecified. The workers re- bribery to obtain those bonuses. By
alize that the distribution of prizes is contrast, consider a case where the su-
fixed, yet retain incentives through the pervisor can offer a high wage only if
prospect of improving their particular the worker is reassigned to a more re-
prize. While theoretically correct, it is sponsible position, for whose output the
unclear how important this is in reality supervisor is responsible. In this set-
for two reasons. First, firms rarely com- ting, the supervisor is less likely to re-
mit to the size of prizes. Second, it spond to the rent-seeking activities of a
would seem that in many cases they can less able employee because promoting
easily renege, by claiming bad business that employee results in lower produc-
conditions or whatever.51 Finally, an- tivity than if the most qualified person
other solution is simply that the firm is promnoted. Thus, tying wages to job
commits to a wage bill and allows the responsibilities can reduce inefficient
supervisor to assign wages as he sees fit; influence activities.
there is no need to specify that wages
2.3.1 Bureaucracy
be attached to ranks.
Two more speculative explanations A central feature of organizations is
may also be offered for the use of tour- the use of bureaucracy, where rules are
naments. First, the psychology litera- used to allocate resources rather than
ture described above illustrates a allowing individuals discretion over re-
marked reluctance of supervisors to dis- source allocation. In the context of
human resources, many examples come
51 Perhaps the solution to this problem is that to mind.
in First, Richard Freeman and
situations where labor markets reward talent, pro-
Medoff (1984) illustrate the importance
motions become an effective form of commitment
for the reason that if a worker is promoted to a of seniority in promotion and layoff de-
more responsible position, his wage will rise sim- cisions, independent of profitability
ply because the labor market values these skills, as
considerations. They note that among
in Waldman 1984. Thus workers may be willing to
exert effort in the hope of achieving this endoge- nonunion firms, almost 42 percent lay
nously determined prize. off solely on the basis of seniority

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
38 Journal of Economic Literature, Vol. XXXVII (March 1999)

considerations, while a mere 14 percent setting where agent can respond to

ignore seniority, only considering prof- compensation schemes in inefficient
itability considerations. Second, Sey- ways. To illustrate the incentive to act
mour Spilerman (1986) notes that su- bureaucratically, two ingredients are
pervisors are often constrained in necessary. First, some measures of per-
the raises that they can offer to their formance must be corruptable. I illus-
subordinates, as job grades typically trate this by considering a situation
carry ranges (minimum and maximum) where the subjective signal is subject to
that cannot be exceeded. This feature influence activities. Second, bureauc-
is considered at some length in Spiler- racy has the connotation that informa-
man and Hiroshi Ishida (1994), Baker, tion is not effectively collected. To
Gibbs, and Holmstrom (1994b), and model this, I consider a situation where
especially Gibbs and Hendrick (1996), promotion involves the allocation of the
who address the provision of incentives worker to a new position, where there is
to workers who are "maxed out" (i.e., a higher return to ability. As a result,
are at the top of their pay ranges). the firm would like to aggregate infor-
In each case cited, it appears that mation efficiently to minimize worker
these pay restrictions have real effects. misallocation.
Spilerman also notes that positions In particular, the winner of the tour-
are often characterized by minimum ex- nament is now assigned to another job,
perience requirements, where workers which is identical to the previous job ex-
must stay in a particular position for a cept that the (linear) marginal return to
certain amount of time before they ability is higher.52 Thus the winner of the
can be promoted. This occurs indepen- tournament is reallocated to a new posi-
dent of the ability level of -the agent in- tion. Since the winner is assigned to a
volved. In each of these cases, dis- job with higher return to ability, there is
cretion is taken from the hands of a return to identifying which worker is
supervisors. more talented; this reduces the prob-
The essence of bureacratic rules is ability of inefficient allocation. The other
that resources are allocated in an ex distinction from the set-up in the pre-
post inefficient fashion. For instance, a vious subsection is that that the subjec-
worker is promoted on seniority even tive signal can be distorted, as in Section
though a better candidate exists. Recent 2.2. Thus, the agent is evaluated on a
developments in agency theory, follow- non-corruptible objective measure y' as
ing Milgrom and Roberts (1988) and Ti- above, but also on S = si + bi. The cost of
role (1992), provide a simple reason for bias is as in the previous subsection,
. bi2
such rules: while rules harm ex post ef-
K(bi)= 2 and the two noise terms are
ficient allocations, they improve the in-
centive for agents to allocate their ac- equal, y2 = 6y2 = y2
tivities correctly, by avoiding influence Consider the ex post optimal alloca-
activities. For instance, although pro- tion rule. Since the productivity of the
motion by seniority has allocative ineffi- most talented worker is higher in the
ciencies, at least there is little lobbying. promoted position, the ex post optimal
I address this issue here in the context rule places the "best" worker in that po-
of the tournament model, because most sition. This means that agent 1 should be
of the prominent examples of bureauc-
racy involve such decisions. However, 52 So expected output is giveln by e + yx, where
bureaucracy will typically occur in any 7> 1.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 39

promoted if workers exert influence activities

yI+ SI y2 +S2 dbi
Zi = 2 =Z2 (K<ooSO e>0),
sodO
exactly as in the previous section. Other- the optimal choice of 0* is less than 1
wise, agent 2 wins the tournament. This and bureaucracy is used. In other words,
is the non-bureaucratic allocation, as firms commit to bureaucratic rules in or-
decisions are made in an ex post optimal der to reduce the incentive to engage in
fashion. influence activities, even though this
Now consider an alternative decision sometimes involves the misallocation of
rule, where the measure used to reward resources.54 Hence the optimal provision
and promote is of incentives involves bureaucracy.

Ai(O, yi, ) i+e

-= + S 2.4 Team Production

for 0 < 0 < 1. Thus, worker i is promoted

Most workers hold jobs that involve
if zi(e, yi,s) >(0, y,), which deviates productive interactions with their col-
from the ex post optimal rule if 0 < 1. Let
leagues, where output reflects the con-
Q(0) be the gain from allocating workers tribution of many individuals. Team pro-
using an aggregation rule with weight 0,
duction problems potentially arise in
for 0 < 0 < 1, relative to not using the s
situations where individual contributions
signal.53 Bureaucracy arises if 0 < 1. The
to output cannot be easily identified and
reason bureaucratic rules are used here
compensation must be based on team
is that the rent-seeking activity is in-
production. In that setting, the classic
creasing not only in the prize for promo-
free-riding problem arises, where agents
tion W - L, but also in 0, the weight
fail to internalize the benefits that ac-
placed on the signal. It is straightforward
crue to other members of the team when
to show that the level of bias chosen by making effort decisions. This effect,
both agents is given by which has also been referred to as the 1
problem (since each agent receives this
P 0(W-L) (19) share of output in a partnership with N
E2i 2(0)K
members), prevails in situations where
where X2(0)= V(2(O)). Thus agents exert rewards cannot exceed the revenues of
more bias when either the monetary re- the group (Holmstrom 1982; Kenneth
turn is high, or when the effect of the McLaughlin 1994). The available tests of
signal on decisions is high. Critically, in- free riding in teams largely come from
fluence activities are increasing in 0. As a the observation of partnerships in law
result, the firm commits to underweight firms or medical practices. First, Joseph
the corruptable signal. It is simple to Newhouse (1973) considers the effect of
show that the optimal choice of 0* is
54 At an anecdotal level, the economics depart-
given by the first order condition ment of an esteemed U.S. university uses a rule
where once a faculty member has published a cer-
- 0(W* - L*) dbi>, (20) tain number of papers, he is offered tenure. This
rule, which takes away important discretion from
E2it (0) dO
the hands of evaluators, was introduced because of
previous accusations of favoritism, when deans
where W* and L* are the optimally cho- and senior faculty had discretion over promotion
sen prizes. The key point here is that if decisions. This is used to reduce the incentives of
the junior faculty to curry favor with their senior
53 It is easy to show that Q(O) = O,Q'(O) > O for colleagues, even though it sometimes induces
0 < 1, Q'(1) = O, and Q"(0) < O. inefficient promotions.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
40 Journal of Economic Literature, Vol. XXXVII (March 1999)

Change in Calls per Hour

Calls Per Hour Before Change

8 10 12 16 18 20

-2

Figure 4. Response of Telephone Operators to Team-Based Compensation (Hansen 1997).

group incentives in a medical practice, a measure used to exogenously identify

and notes that when the fraction of reve- variation in practice size.55 They illus-
nues that are shared with others rises, (i) trate that poorer measures of performance
overhead costs rise, and (ii) doctors work arise when more revenues are shared
fewer hours. Richard Bailey (1970) finds with others, once again endorsing the
qualitatively similar results, while Arlene importance of the free-rider problem.56
Leibowitz and Robert Tollison (1980) There are many possible solutions to
find that larger law partnerships typically
result in worse cost containment. These 55 The idea here is that more risk-averse doctors
studies simply compare productivity will operate in larger practices, as they value in-
come-smoothing opportunities more. Once these
measures of partnerships on different measures of risk aversion are shown to be inde-
sharing rules without addressing why pendent of other productivity measures (and that
contracts vary, and so are subject to obvi- the instrument has some explanatory power), this
is a legitimate identification strategy.
ous selection criticisms. For instance, it 56However, this is not to say that team-based
could be that the less able work in teams rewards cannot generate incentives relative to no
since they have less to share, which incentive pay, as illustrated for the steel industry
by Boning, Ichniowski, and Shaw (1998), who also
could explain the low performance mea- effectively control for heterogeneity in contract
sures, independent of any behavioral ef- choice through variation in the manufacturing
fect of teams. Martin Gaynor and Pauly environment. See also Encinosa, Gaynor, and
Rebitzer (1997) for an interesting attempt to dis-
(1990) use survey evidence on medical tinguish between economic and noneconomic
practices, where reported risk aversion is notions of team production and compensation.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 PrenderLyast: The Provision of Incentives in Firms 41

the team production problem.57 One able (with the more able preferring in-
solution that has attracted some atten- dividual based schemes), Weiss identi-
tion has been the use of peer pressure, fies a U-shaped relationship between
where agents can possibly rnonitor one worker turnover and prior productivity.
another and mete out punishments to In his sample, the medium-ability work-
those who fail to perform adequately. If ers are more likely to remain than
the cost of such monitoring is suffi- either the more able or the less able
ciently low, this can negate the free- when placed on team-based compensa-
rider problem, as illustrated in Eugene tion. One interpretation of this is that
Kandel and Lazear (1992). However, the most able leave as they prefer indi-
empirical evidence on peer pressure re- vidual-based schemes elsewhere, while
veals behavioral responses different the least able also leave as the peer
from those posited in the theory. pressure makes their jobs too unpleas-
Andrew Weiss (1987) and Daniel Han- ant. Thus the (admittedly scant) evi-
sen (1997) consider the effect of team- dence suggests that team-based com-
based compensation on the individual pensation gives rise to problems when
productivity of agents. (In both cases, workers vary in their ability.
the employers choose to pay on the pro- Most of the work on team compensa-
ductivity of the team despite the avail- tion concerns profit-sharing schemes, as
ability of measures of individual pro- in Derek Jones and Takeo Kato (1995),
duction.) Weiss studies the productivity Douglas Kruse (1993), and Marc Knez
of blue-collar workers in a pharmaceuti- and Duncan Simester (1997). These
cal company, while Hansen addresses studies are carried out on large firms,
the incentives of telephone operators where the wages (or often pensions) of
for a large financial company. Both employees are based on the profits of the
authors illustrate that the use of team- entire firm, either through ESOPs or bo-
based compensation schemes improves nuses. Standard reasoning of the N prob-
the performance of those who were less lem suggests that there should be a neg-
productive on individual schemes but ligible response by agents to these
decreases that of the more productive. incentives, since, for example, a worker
Hansen's results are summarized in who gets to keep I of the returns to
Figure 4. He studies the performance effort in a 1,000-worker firm should have
of telephone operators, measured by few incentives. Despite this, studies con-
the number of calls they handle in an sistently show that the productivity of
hour. Figure 4 plots the change in the firms using profit-sharing plans exceeds
number of calls dealt with after the in- that of those that do not, with available
troduction of team-based compensation estimates suggesting improvements in
schemes as a function of the number of the range of 4-5 percent from these
calls that were handled under fixed schemes.
wages. The negative slope shows that Since these results appear to be such
the more able agents reduce the num- a violation of standard agency theory,
ber of calls made while the less able im- alternative explanations have been
prove. A related point concerns the se- sought. The most popular versions in-
lection effect of teams. While one's first volve some notion of either peer pres-
impression is that team production is sure58 (as described above), mutual
likely to be more attractive to the less
58 It should be emphasized that merely alluding
to peer pressure hardly suffices here. For profit
57 See Holmstrom (1982). sharing to have effects on peer pressure in a large

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
42 Journal of Economic Literature, Vol. XXXVII (March 1999)

% in Productivity: Relative to Non-Adoption

Within Firm Estimate of Profit-Sharing

6

XX -2 / -1 0 1 2 3

Year Rela

L2 \ /Year of Change

-4

Figure 5. Profit Sharing and Productivity (Kruse 1993).

monitoring, or "belonging," where the cross section, this could simply reflect
employees feel as if they are "in this to- the possibility that firms with no profits
gether." Without meaning to dismiss rarely introduce such schemes, so
these potential motivations, there are a higher profitability could have little to
couple of reasons to be skeptical about do with the effect of such schemes. Re-
the validity of these results as a test of searchers have solved this by looking
team production incentives. These "within firm." In other words, does pro-
doubts arise because (i) the data may ductivity rise in those firms with profit
not really illustrate productivity in- sharing more than in those without such
creases due to the compensation schemes?
scheme, or (ii) the observed increases, Using this methodology, a large-scale
though related to the compensation study by Kruse (1993) finds that this is
changes, may have little to do with the the case, where productivity rises by 3
team production problem. First, the percent more in firms with profit shar-
cross-sectional data illustrate that firms ing than in those without. While this is
that use profit sharing have higher pro- an interesting approach to under-
ductivity than those that do not. In the standing the effect of pay on perfor-
mance, and a considerable improve-
firm, it must be the case that the costs of enforce- ment over existing work, it constitutes a
ment through peer pressure (pointing out errors legitimate identification strategy only if
or slacking to the relevant person) must be negli-
gible, since the monitor equally receives only aN
the trend in productivity changes is
share of any improvements herself. identical between the two sets of firms.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 43

Suppose, for example, that some firms ply that one should be wary of simple
have declining productivity and do not "fixed effect" estimates as a way of
use profit sharing, while others have ris- eliminating unobserved heterogeneity.
ing productivity and use profit sharing. Assume for the moment that these
Then even if there were no effects of problems were solved, and that the ef-
the compensation scheme on produc- fects of the compensation schemes on
tivity, this methodology would suggest productivity were robust after control-
such a relationship since there are ling for this issue. Could we then con-
unobserved differences in changes in clude that agents are willing to exert ef-
productivity correlated with the intro- fort despite the - problem, so that the
duction of the contracts. Relevant free-rider problem loses some of its po-
empirical evidence taken from Kruse is tency? A second problem with the litera-
presented in Figure 5. ture is that these studies do not generally
This measures productivity changes test for free riding in a team setting, in
(value added per employee) for the the sense laid out in the theoretical lit-
adoptors of profit sharing compared to erature, because the theory considers the
those firms that do not implement effect on incentives holding utility con-
profit sharing, measured from three stant. But this is not the case with profit
years before ( - 3 ) to three years after sharing; compensation rises in most
( + 3 ) the scheme was initiated. From firms that use profit sharing; Knez and
the point of adoption (date 0) until 3 Simester (1997) for one example.61
years later, productivity rose by 3 per- Could the empirical results simply re-
cent, suggesting an effect caused by the flect the effect of giving workers more
compensation scheme. This is the money, and not the effect of team pro-
"within firm" estimate used in the lit- duction? To take an extreme example,
erature. However, a comparison of the suppose that profit sharing increased the
firms before adoption suggests that pro- pay of a worker by 25 percent. There are
ductivity may have been rising faster in a multitude of reasons to expect that
those firms anyhow.59 Perhaps the 3 such an increase in wages will improve
percent is merely a continuation of that productivity. An obvious implication is
trend, so that the productivity effects that the firm will attract better workers,
are not caused by the compensation or existing workers will work harder as
plans.60 This is not, of course, conclu- their jobs are more valuable; this is the
sive evidence that profit sharing does premise of the efficiency wage literature
not work; instead, my objective is sim- described below. In either case, produc-
tivity will rise in a way that has little to
do with the fact that pay depends on
59 This effect is not specific to the use of profit-
profits; instead, incentives and selection
sharing schemes. Hassan Tehranian and James
effects arise simply from more pay, not
Waegelein (1985) illustrate that executive com-
pensation plans relating pay to performance are
pay conditional on firm behavior. This is
generally introduced after a number of years of
abnormally positive stock returns. not a problem, of course, if all the
60 This depends on the time-series correlation in authors are interested in is the effect of
productivity changes. If there is no correlation in
changes, then the Kruse interpretation is clearly
right, as those firms with productivity growth are 61 The theory suggests that wages should rise,
no more likely to have future productivity growth but only by the increased cost of effort plus any
than those without. However, to the extent that risk premium associate with the variability in wage
productivity changes are positively correlated, payments. However, there remains the possibility
these results overestimate the effect of profit that profit sharing gives rise to rents earned by
sharing on productivity. workers.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
44 Journal of Economic Literature, Vol. XXXVII (March 1999)

profit sharing; however, it does preclude number of papers have directly tested
testing the classical team production for the importance of efficiency wages
problem. using firm-level data by examining the
relationship between supervision and
2.5 Efficiency Wages wage rents. A reasonable conjecture is
that the probability of being caught
So far, it has been assumed that work -
shirking is increasing in the supervisor-
ers earn their reservation utilities, where
worker ratio. It immediately follows
a firm does not offer rents to its workers
that firms face an isoquant in (wage
to induce effort exertion. Efficiency
rent, supervisor-worker ratio) space,
wage theory concerns situations where
where they can trade off higher wages
firms offer workers such rents in order
against more supervisors. Thus, wage
to induce effort exertion. In the context
rents and supervisors are substitutes.
of incentive provision, firms overpay
Erica Groshen and Alan Krueger (1990)
workers in order to make their jobs valu-
address this issue using hospital em-
able, which makes them less likely to
ployee data, and find evidence in favor
shirk. In this way, the cost of job loss
of the theory. By contrast, Derek Neal
(which ensues if agents are caught shirk-
(1992) uses more aggregate data and
ing) is large, so they exert effort at the
finds little relationship between these
efficient level (Carl Shapiro and Joseph variables.
Stiglitz 1984; Dan Raff 1992; Daron
It is difficult to test for the existence
Acemoglu and Andrew Newman 1997). A
of efficiency wages, where workers earn
simple way to interpret the shirking ver- rents to induce effort exertion. First,
sion of efficiency wage theory is to con-
while finding that wages and supervi-
sider a situation where the agent's wage
sors are substitutes along an isoquant
cannot be reduced below 0, which is as-
of fixed effort is consistent with effi-
sumed to be the reservation utility. In
ciency wages, exactly the same conclu-
other words, even if the agent is caught
sion is true in the basic agency model
shirking, he cannot be penalized by of-
with no rents.62 Thus, this is a test of
fering him a wage less than the reserva-
incentive theory, not necessarily a test
tion utility. To simplify matters, assume
of workers earning rents. In order to
that the effort decision e is binary, set
test for rents, one would need to see,
equal to either 0 or 1, so effort of 1 has a
for example, whether higher levels of
marginal cost of 2. Monitoring is such
supervision within a job increase worker
that the worker who shirks is caught with
turnover (since more supervisors re-
probability p. Since the worker cannot
duce wages).
be penalized below 0 for shirking, the
A second possible problem with this
firm must offer a wage of at least w* = c
methodology concerns the prospect that
to induce effort exertion, which implis
the observed variation in supervisors
that the worker earns rents of (lP)c
2p 62 To see this, remember that holding effort
from the relationship. Thus, inefficient fixed, wages in the basic model exceed reservation
monitoring (p < 1) yields rents for the wages both by effort costs and by the riskiness of
the evaluation procedure. But if supervisors can
worker. be hired to provide more accurate reads of perfor-
This theory has spawned a large mance, wages fall for a fixed level of effort as the
literature, ranging from studies of riskiness of the compensation scheme falls. Thus,
once again wages and supervisors are substitutes,
unemployment to examinations of inter- though without any implications for the existence
industry wage differentials. A small of rents.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 45

and wages across firms may not in- tracts and realizations. This section
volve moving along a common isoquant addresses the primary themes in this
in wage-supervisor space. More spe- literature.
cifically, in available data, would we
3.1 Deferred Comnpensation
expect to see wages and supervisors as
substitutes or complements? The prob- When agents remain with an em-
lem is that either may easily arise in a ployer for a long period of time, there is
world of efficiency wages and depends no necessary reason why the employer
critically on the source of variation should pay the worker his expected
across firms. On the one hand, if the marginal product in all periods; instead,
source of variation across firms is the workers could be paid better in some
cost of supervisors, then the two in- periods than in others. One aspect of
struments are likely to be substi- this that has attracted both theoretical
tutes, where firms substitute away from and empirical interest has been "de-
high-cost supervisors into wages. On ferred compensation," where workers
the other hand, if the source of vari- are overpaid when old, at the cost of
ation across firms is in the return to being underpaid when young. From
effort (so some firnms value effort exer- this perspective, part of the reason
tion more than others), those firms why older workers are better paid
that want more effort will use more of than younger workers is not that they
bothl instruments relative to those are more productive, but simply that
that do not value such high effort. they have accumulated enough tenure
This effect, which relies on the mar- to garner these contractual returns.
ginal cost of each instrument to be To make matters clear, the purpose of
increas ing in its quantity, implies that this subsection is to address whether
supervisors and wages will be comple- and why firms use compensation
mnents in the data.63 As a result, it is schemes like Figure 6, where wage ex-
hard to see how one can refute the ex- ceeds productivity for older workers,
istence of efficiency wages with this but is less than productivity for younger
methodology.64 workers.
Many explanations have been offered
3. The Dynamics of Agency Contracts for the use of such compensation pro-
files. Stephan Salop and Joanne Salop
The focus of the paper so far has
(1976) argue that delayed compensation
been on static contracts, where the
aids the selection of desirable workers.
contracts and behavior in one year
For example, firms often incur signifi-
have no effect on future contracts.
cant turnover costs when workers
Yet employees and firms make matches
leave, and one way of attracting those
that typically last for long periods of
who are less likely to leave is to offer
time, and a considerable literature
(quasi-)rents only to those who re-
now exists that addresses intertemporal
main at the firm for long periods of
linlks in the contracts offered to
time. An alternative possibility is that
w-^, orkers, where the contract offered
deferred compensation is useful be-
tlhis year depends on last year's con-
cause there may be a significant delay
See Susani Athey and Scott Stern (1997). in observing the effects of efforts. As
64 Groshen and Krueger attempt to control for a result, firms may prefer to wait to
tlls by arguing that rates of supervision are largely
set by regulation, though it is unclear how regula- generate a better inference on the
tory authorities set these levels. worker's performance before rewarding

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
46 Journal of Economic Literature, Vol. XXXVII (March 1999)

Productivity-
N. -1"

, '/
t Compensation
*"Compe Value of Leisure
nsationl

Age

AF A. D

Figure 6. Deferred Compensation.

him.65 However, the primary focus on the firm long enough to attain those
deferred compensation has been as a rents. As a result, younger workers can
means of providing incentives to work- be offered lower current compensation
ers, as in Lazear (1981). The idea here than older workers (relative to market
is simple. Consider a firm that offers options), while maintaining incentives
rents to its older workers for the for all (Akerlof and Lawrence Katz
efficiency wage reason described 1989).
above. For large enough rents, older To understand the mechanics of this
workers are willing to exert effort problem, consider the efficiency wage
rather than be fired. But rents to older model above, where there are two peri-
workers are also attractive to youlnger ods of the worker's career, "young" and
workers, because exerting effort in- "old." (In this section, I will typically
creases the likelihood of surviving in consider two-period settings for simplic-
ity.) In the single-period setting, it was
65 Finally, it may be that wages are deferred
simply because workers have preferences for
shown that the firm must offer the agent
wages that increase with age. This is interpreted
either as a preference for thinking that we are a wage of w* = c to induce effort exer-
doing better from year to year, or as a means of
forced savings, which agents do not trust them- tion of e = 1. Since "old" workers have
selves to do. See George Loewenstein and only a single period of employment re-
Nachum Sicherman (1991) and Robert Frank and
Robert Hutchens (1993) for empirical evidence on
maining, the firm will offer that wage
such preferences. when workers are old. Remember, how-

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 47

One possibility is to compare wages to

ever, that this entails rents of (- productivity
P)c for in those occupations. How-
' ~~~~~~2p
the worker at this point in life. The firm ever, it is in those occupations that we
can take advantage of this when design- expect that deferring compensation is
ing the younger worker's contract, be- unlikely to be used. For instance, con-
cause if the younger worker is caught sider the compensation of salesforce
shirking, he loses not only the rents asso- workers. Because their productivity is
ciated with the "young" wage, w,' but easy to observe, compensation can be
based directly on those measures; there
also the future rents (- P) because
2p
the is no need to use deferred compensa-
tion.69 For this reason, one may not be
firm fires him if he is caught shirking.66
able to consider those occupations with
Let 6 be the discount factor between the
easily observed output to discover the
two periods. Then it is trivial to show
extent of deferred compensation.
that the agent will exert effort of e = 1 so Despite these constraints, a number
long as
of papers have shown wage changes for
older workers that appear to have little
Wy2 6(-p)c (21)
to do with productivity effects. Each of
which is less than w* if p < 1. In other these studies has some problems, in
words, the wage offered to the young is that there are other interpretations, but
strictly less than that offered to the old. the aggregate picture suggests the de-
Thus, the firm defers compensation as ferring of compensation. First, Richard
part of an optimal payment package.67 Freeman and James Medoff (1984) and
Do Firms Defer Compensation? De- Spilerman (1986) illustrate that firms
termining whether firms defer compen- often build seniority provisions into
sation is conceptually easy; simply com- pay, promotion, and retention deci-
pare wages to productivity. However, sions, even when not warranted by pro-
productivity is typically difficult to mea- ductivity considerations.70 For example,
sure. But there are some occupations rules that promote workers on the basis
for which productivity is observed.68 of seniority rather than productivity
offer such workers tenure-related
66 It should be noted that perhaps the most advantages. Second, Medoff and Kath-
common way in which firms provide incentives to erine Abraham (1980) illustrate that
workers is through the threat of being fired if
their performance is not satisfactory. However,
the paper has had little to say about this means of
incentive provision. Firing can be seen as a form surgeons. He illustrates that surgeons' pay rises
of nonlinear incentive contract where the worker with age, although their performance as a surgeon
is paid a fixed wage for performance above some (risk adjusted mortality rates) becomes worse. Of
critical level, but is terminated otherwise. This course, it could be that older surgeons have more
simple model provides a reason for such incentive unmeasured aspects of production, such as train-
schemes. In particular, the firm uses firing as a ing, which is reflected in higher wages.
way of excluding agents from future benefits 69 In fact, one could argue that if we find evi-
which would accrue if they retain their jobs. In the dence of backloading for those occupations, it
presence of liquidity constraints, this becomes an must be that backloading occurs for reasons other
efficient means of incentive provision. than incentives, since incentives can be provided
67 The central feature of deferred compensation in simpler ways.
is that good performance in one period yields the 70 Robert Frank and Robert Hutchens (1984)
opportunity for benefits in future periods. Similaralso consider the wage profiles of two occupations
results in the context of tournaments arise in (bus drivers and airline pilots) where wages con-
Meyer (1992). tinue to rise with seniority, even though there is
68 An interesting recent study on this is Leventis little reason to expect productivity to increase
(1997b), who considers the productivity and pay of after some initial period.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
48 Journal of Economic Literature, Vol. XXXVII (March 1999)

Productivity -

/,,

t / Compensation

Age

45 50 55 60 65

Figure 7. Deferred Co

the performance evaluations of senior proxy for the productivity of the em-
workers differ little from those of their ployed person, since there is no one to
less senior counterparts, yet their wages shield the self-employed worker from
are higher. They interpret this as fur- changes in his productivity. Lazear and
ther evidence of the use of deferred Robert Moore (1986) show that the
compensation.71 wage profiles of the self-employed are
Another approach to addressing the indeed flatter than those of the em-
importance of deferred compensation is ployed. Hence if the wage profile of
to compare the wage profiles of the the self-employed maps the produc-
self-employed to those in similar posi- tivity of the employed, this suggests
tions who are employed by firms. Con- the "overpayment" of older employed
sider two workers who, say, are consult- workers.72
ants, where one is self-employed and In a similar vein, Lawrence Kotlikoff
the other is an employee of a firm. If and Jagadeesh Gokhale (1992) use the
they both carry out the same job with wages of newcomers to a large firm to
equal efficiency, the wages of the self- identify the returns to seniority within
employed consultant should be a good

71 An alternative interpretation of these data is 72 Of course, there are other interpretations.

that assessment standards depend on seniority, For instance, it could be that more training is pro-
i.e., workers could be assessed relative to their vided to employed workers, which they pay for
potential, in which case senior workers could be early in their careers, but garner the returns later
better despite similar evaluations. in life.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firmns 49

firms. Consider two 65-year-old work- to readily available measures of produc-

ers in the same position, one of whom tivity, so that there is little need to use
has been with the firm since leaving deferred compensation. It should be
high school and the other who has just remembered here, however, that an
joined. The firm has little obligation to alternative interpretation of Figure 7
the newcomer and will pay him his mar- is that specific skills are important, so
ginal product. But if specific human that newcomers earn less because
capital is of limited importance, new- they are less productive, even within
comers in the same position as those positions .73
with more tenure should have similar
3.1.1 The Returns to Promotion
productivity. If so, any differences in
in Hierarchies
wages between the newcomer and the
worker with longer tenure could be The essence of the previous section is
attributed to a difference between that contracts in one period depend on
wages and productivity. Thus, they contracts in previous periods. Another
proxy the productivity of workers with application of this idea concerns how
long tenure by the wages of newcomers the returns to promotion vary as work-
in the same position. Using this meth- ers progress through a firm's hierarchy.
odology, the authors can roll back the It is well known that the returns to pro-
productivity profile of a worker based motion increase at higher ranks in a
on age of arrival. Their results for two firm. See Gibbs (1992), Main et al.
occupations (office workers and sales- (1991), Conyon and Peck (1997), and
force workers) are replicated in Figures Richard Lambert et al. (1993) for de-
7 and 8 below, where the effect of a tails.74 Since workers typically progress
pension vesting at 55 is ignored to sim- through the ranks of firms over time,
plify the picture. this has an obvious relation to deferred
Figure 7 provides evidence on the compensation. More generally, this sug-
compensation of office workers; the gests inter-temporal linkages in con-
authors note a discernable difference tracts, where prizes at one level depend
between pay and productivity, with on previous prizes. A number of reasons
younger workers being less well paid have been proposed for this behavior.
than their alternatives, and older work- First, ignoring incentive issues, Rosen
ers earning more. These data are consis-
73 An auxiliary implication of deferring compen-
tent with a view that for occupations
sation is that workers will be reluctant to retire,
where it is difficult to provide objective i.e., workers who are paid "too much" will stay too
measures of performance, as would be long. Following Lazear (1981), Figure 6 illustrates
the case for these white-collar clerical that workers will retire when their wage equals the
value of their leisure, i.e., at time A**. On effi-
workers, the optimal means of provid- ciency grounds, they should retire at A*, when
ing incentives is to offer "carrots" in productivity equals the value of time. Mandatory
retirement at A* is efficient (though there are
the future. By contrast, consider the
other possible mechanisms to replicate this out-
compensation of salesforce workers in come, such as the use of pensions). Hutchens
Figure 8. (1988) provides some empirical evidence consis-
Here there is little difference be- tent with this model, by noting that US firms
whose wage profiles rose rapidly as workers aged
tween wages and imputed produc- were also those which tended to use mandatory
tivity. A plausible interpretation of retirement.
74 None of these empirical papers has attempted
the difference between this and Figure
to distinguish between the competing hypotheses
7 is that for salesforce workers, in- below, though they are often framed in terms of
centives can be provided by tying pay the Rosen (1986) work.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
50 Journal of Economic Literature, Vol. XXXVII (March 1999)

- ~~~~~~~~~~Productivity

Compensation

Age

45 50 55 60 65

Figure 8. Deferred Com

(1982) illustrates that in hierarchies also play a role. A number of possible

where the decisions of superiors have explanations for convex wage structures
implications for the marginal produc- generated by incentive considerations
tivity of those in lower positions, there can be imagined. First, income effects
is a large return to ability. Due to what may cause wage increases on promotion
has become known as the "magnifica- to rise as workers ascend the hierarchy.
tion effect" (where the decisions of se- Quite simply, it may take more money
nior workers are magnified many to induce effort from the rich than from
times), the returns to ability are convex, the less well off.75 Second, raises upon
so on simple marginal productivity promotion may increase because the op-
grounds, more able workers will earn timal level of effort is higher at more
many times the wages of their less able elevated ranks, as decisions made at
counterparts. higher ranks have more wide-reaching
In the neoclassical model, the wage effects; it is more important for the
earned by a worker is the supply price CEO to work hard than for a shop floor
of labor. Despite this, the wages of se-
75This relies 'on the marginal rate of substitu-
nior executives often triple overnight tion between income and leisure varying with the
when they accede to the position of level of income, unlike the exponential utility
CEO, so it is doubtful that this is the function described above. For instance, a utility
function of the form V(w, e) = U(w) - C(e), where
only influence generating wages. As a U(w) has the usual properties of risk aversion, will
result, it is generally felt that incentives suffice.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 51

worker to do so. If the marginal return 3.2 Dynamic Renegotiation of

to effort is increasing in rank, convex Contracts: Career Concerns
wage profiles will arise.
Performance measures reveal infor-
Neither of these reasons i:mplies that
mation not only on the efforts exerted
rents are awarded to workers at higher
by an individual but also on his innate
ranks. However, a final reason proposed
ability. Career concerns arise in situ-
by Rosen (1986) argues that such con-
ations where agents exert effort not just
vex wage schedules may offer rents to
to maximize current pay but also to af-
workers in senior positions as a means
fect the perceptions of others. Consider
of providing incentives to all workers.
the case of a baseball player on a fixed
Rosen points out that part of the return
salary. The analysis of the previous sec-
for promotion from rank a to b is not
tions would suggest that the player
simply the pay difference between the
would have little incentive to exert ef-
two ranks, but also the increased pros-
fort, as there is no immediate relation-
pect of further promotion to ranks c, d,
ship between pay and performance.
and so on. A suitable analogy is that
This, of course, is patently false, since
part of the return to winning the first
players exert effort in order to affect fu-
round of a tennis tournament is not just
ture contracts, which depend on current
the prize money for reaching the sec-
performance. In essence, contracts can
ond round, but the deferred prospect of
be renegotiated on the basis of perfor-
the prize money for future rounds. This
mance as the market settles up. Follow-
idea, which has been labeled the "op-
ing Eugene Fama (1980) and Holm-
tion" value of promotion, implies that
strom (1982), such career concerns
wage increases from promotion are de-
have been proposed as a means of
creasing in the prospect of future pro-
motions. Thus the winner of the final prizes. However, this wage distribution can be
round must be offered greater incen- seen as the outcome of the optimal static tourna-
ment being repeated, with no inter-rank links in
tives than the winner of the first round,
the design of the optimal tournament. To see this,
for whom the deferred benefit is larger. assume that the winner of the optimal static tour-
This idea is formally similar to the nament gets a prize of P. Then with reservation
utilities normalized to zero, the winner of any
Lazear (1979) model of deferred com-
given round has a net utility increase of 7 and the
pensation described above, where work- loser has a net utility change of - < 0. Then con-
ers value the future rents associated sider the wage distribution that arises from an
with acceptable behavior today when elimination tournament, i.e., where one loss ex-
cludes one from future tournaments. Consider an
making their effort decisions. As with agent who loses in the first round; his net utility is
Lazear's work, a critical component of - P For a loser in the second round, net utility is
both models is that rents must be 0, while a loser in the third round earns net utility
of P, and so on. Thus, the net utility from winning
earned in the static agency setting; oth- successive rounds is linear, with the gain being 2.
erwise there is no need for dynamic But this is not true for the final round, because
the winner of the final must win two more net
links between contracts.76
tournaments than the loser of the final. As a re-
sult, the final jump in utility in the elimination
76 Perhaps the most striking observation in tournament is P, not 2'1
P for the reason that the
Rosen (1986) is that in an elimination tournament ultimate winner does not finish the contest with a
with risk-neutral workers, the optimal wage struc- loss, unlike all other agents. Therefore, the utility
ture consists of constant prizes for "promotion" (and wage) distribution is linear in rank until the
until one reaches the last rank, at which point final level, at which point it "jumps." Hence the
there is a discretely higher prize. This final prize distribution of wages generated in Rosen could be
is often interpreted as necessary to give incentives seen as arising from the arithmetic of repeated
at the end of the competition, since there are no static tournaments rather than the generation of
longer incentives generated by possible future an option in repeated tournaments.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
52 Journal of Economic Literature, Vol. XXXVII (March 1999)

inducing efficient effort exertion even in d2 = (1 - I3h)[ed + E(cx 1 yi)], where e! is

the absence of explicit contracts. One at- the equilibrium level of effort in period
traction of this literature is that it places 2 and E(oc I y') is the perceived level of
the agency issues more firmly in the ability of the worker.77 It is through the
context of a labor market that values second-period salary that career con-
employees and affects the behavior of cerns arise; the reservation utility of the
the firm. From the career concerns per- worker depends on first-period perfor-
spective, because outside options matter mance. To understand how this affects
(in the sense that other firms will bid incentives, note that
for workers), incentives are provided
even in circumstances where explicit E(cx I y')= -2 + (yi - e*), (22)
pay-for-performance is not offered.
The following changes are made to where ej is the expected level of first-
the basic model outlined above along period effort (thus, the market is not
the lines of Holmstrom (1982) and fooled in equilibrium). As in the pre-
Gibbons and Murphy (1992). First, vious section, let 6 be the discount factor
assume that the agent works for two between periods 1 and 2. Then for any
periods, t = 1,2, rather than the single first period contract, WI = f8ol + PyI31, the
period of Section 2. Further, assume agent will exert effort of
that the worker gains fromn being
perceived as talented. In particular, p3rl + P6F2) ;y2) + 2 (3
let the labor market be competitive,
e* ~ ~ ~ +~ (23)
where the worker earns his expected
Consequently, for >2 < 1, period 1 incen-
productivity in period 2. The worker is
tives are greater than in the static set-
assumed to be evaluated on a common
ting, because future contracts depend on
subjective (i.e., non-contractible) signal
perceptions. Even in cases where there
of st=et +xO+ st in period t,t= 1,2,
where all variables are distributed as in are no explicit contracts (K = 0), the
Section 2.1 and where the time-sub-
agents will exert effort of c( a +a)
scripted error terms are independently
distributed across the two periods. solely to affect perceptions. In additior
Some of the implication of the career however, there are implications for ob-
concerns model is on observed con- served compensation contracts. In par-
tracts. As a result, it is also assumed ticular, straightforward calculations show
that the firms can base compensa- that the optimal choice of f3sl implies less
tion on a measure St = St + bt, where bt explicit incentives in period one than in
refers to bias activities exerted in
the second period, i.e., P3-h < P12, which
period t, which have the same costs has been tested.
1CbP
K(bt) = 2 as above. This simple model offers a nuniber of
2
implications of career concerns.78 First,
Consider the second period, where the
agents will exert positive levels of effort
worker receives a contract W2 = f02 + 322S 2.
Then by analysis identical to that 77 The worker earns j32(ee + E(:ulyu1)) in expecta-
tion from the p iece rate, so this is the salary at
above, the optimal choice of f3-2 is given
which the firm breaks even.
78 It should be emphasized that some of these
by f 2 = 1C, as in the static setting.
results are specific to the case where the speed of
learning is independent of effort exerted. See
Importantly, since wages equal expected Mathias Dewatripont, Ian Jewitt, and Jean Tirole
productivity, the salary component is (1997) for details.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prender,ast: The Provision of Incentives in Firms 53

in the absence of explicit contracts. sider the likelihood of mutual fund man-
Second, agents will generally exert agers being fired on the basis of current
inefficient levels of effort. In this 2-pe- and previous performance. As predicted
riod setting, in the absence of contracts, by the theory, this relationship is greater
there is always underprovision of effort for younger managers, about whom there
in both periods because the career con- is little information, than for older man-
cerns can never be sufficiently impor- agers. Gibbons and Murphy (1992) con-
tant in the first period as there is only a sider a final implication of the theory
single period in which to "cash in" from alluded to above; that contracts offered
a good reputation. This is not necessar- to workers will have more explicit pay-
ily true in a general T-period model, as for-performance provisions as workers
Holmstrom (1982) has shown that get closer to retirement. Gibbons and
sometimes agents will exert effort above Murphy generalize this simple model to
the efficient level to affect percep- illustrate that optimal sensitivity of con-
tions.79 A third effect of career con- tracted pay to performance will mono-
cerns models is that effort exertion de- tonically increase as workers accumu-
pends on the length of time "on the late tenure. They test this prediction on
job." When workers begin in their posi- US executives and find evidence consis-
tions, little is known about them, so that tent with the prediction. Paul Gompers
productivity realizations have signifi- and Josh Lerner (1994) illustrate similar
cant effects on perceptions of ability. In effects on the contracts offered to ven-
addition, workers who are young have a ture capital managers.
long time over which to garner the re- Before concluding, it is worth making
turns to a good reputation, so that this a couple of other observations on the
model predicts high effort among the implications of career concern settings.
young (or those with low tenure). This First, many of the comparative statics
is not true after tenure is accumulated, alluded to in the static setting need no
because much is already known about longer hold. For instance, consider the
workers, and there is less time over optimal choice of incentives in the
which to generate a reputation. Thus static model of Section 2.1, where noisy
effort levels fall over time. measures imply reduced incentives. Un-
A corollary of this is that worker wel- like the static setting with risk aversion,
fare will be more sensitive to early re- this model predicts that first period in-
alizations of output than those that ar- centives are increasing in c2 so that in-
rive for workers who have been "on the centives are increasing in the noisiness
job" for a long period of time. A test- of the measures of performance. This is
able implication of this is that current because as measures of performance
performance should be more predictive become noisier, career concerns fall in
of rewards for younger workers (or those the sense that there is little updating on
with less tenure) than for older workers. ability. Thus, explicit measures of per-
To my knowledge, the only test of this formance must rise to compensate, and
prediction has been carried out by so the basic trade-off between risk and
Chevalier and Ellison (1997b) who con- incentive no longer holds.80 In a similar
vein, Meyer and John Vickers (1997)
79 The reason for this is that in a T-period
model, a reputation gained in period s has value 80This effect arises most clearly in this setting
over the remaining T - s 2 1 periods, so that career when the worker is risk neutral. The effect would
concerns effects can be large enough to induce be tempered and possibly reversed in a setting
more than efficient effort. where workers are risk averse.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
54 Journal of Economic Literature, Vol. XXXVII (March 1999)

illustrate that relative performance 3.3 Dynamic Enforcement of Contracts

evaluation may not be desirable when
The literature has emphasized a cou-
career concerns are present. This arises
ple of other ways in which repetition
because reducing measurement error
can improve the agency relationship.
through relative performance evalu-
First, throughout the paper I have
ation, while good in a static agency
stressed the importance of subjective
model, can be harmful in the context
measures of performance. But if these
of career concerns, because the more
measures cannot be verified to third
that is known about ability, the less rea-
parties, why would a principal ever hon-
son to exert effort for career concern
estly reveal these measures? Here the
reasons.
literature has stressed an important role
Second, it should not necessarily be
for repeated relationships. Consider a
assumed that career concerns always in-
setting where a principal must choose
crease effort. A simple reinterpretation
whether to reward a worker for good
of the model, following Gibbons (1987)
performance that is unverifiable.
and Meyer and Vickers (1997), formal-
Though it may be part of an optimal
izes ratchet effects as a career concern
contract to reward the worker for good
problem. Ratchet effects arise when
performance, in a static setting the
firms react to information that costs of
principal will generally renege on the
production are lower by reducing the
contract in order to save on the extra
pay of agents. For example, firms could
wage costs. However, standard repeated
require workers to produce higher quo-
game logic can imply that the principal
tas when they illustrate that high per-
will compensate the worker in the ap-
formance levels are possible. To formal-
propriate way if the worker can
ize this, assume that in the rnodel above
threaten to withhold future effort if he
ox now refers to ability in the firm,
fails to do so. In that setting, Clive Bull
which has no value outside, so higher
(1987) and MacLeod and Malcomson
ability means higher productivity. In
(1989) illustrate that with sufficiently
this case the renegotiation of the
high discount factors, repetition can
contracts imply that better agents re-
generate efficient outcomes that would
ceive lower salaries, since able agents
not arise in the static setting. See
will earn more from any fixed piece
MacLeod (1993) and Malcomson (1998)
rates. As a result, agents now have an
for surveys of this literature. Baker,
incentive to restrict output (to avoid
Gibbons, and Murphy (1994) extend
such downward revision in salaries), so
this logic to show that such implicit
that career concerns can harm incen-
contracts interact with explicit contracts
tives.81
in interesting ways, so that the exist-
ence of explicit contracts can either re-
81 In this section, only situations where the inforce implicit contracts or crowd
worker exerted "effort" were considered. How- them out.
ever, career concerns have been shown to affect
The common feature of these models
many dimensions of performance. For example,
Holmstrom and Ricart I Costa (1986) and Her- of incentive provision is that firms
malin (1993) consider career concern problems
when agents choose the riskiness of the projects ning with David Sharfstein and Stein (1992), have
they take. Jeremy Stein (1990) and Paul (1992) ad- addressed how career concerns can induce agents
dress how career concern models can induce myo- to become either conservative or impulsive. See
pia, where agents care excessively about short- Prendergast and Lars Stole (1996), and Jeffrey
term returns to projects rather than their net Zweibel (1995) for details of that literature, and
present value. Finally, a series of papers, begin- Owen Lamont (1996) for empirical work.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 55

sometimes value the future enough to 4. Conclusion

induce them to act honestly, even in
situations where there is an incentive to This survey has covered a wide range
contemporaneously cheat. An implica- of issues associated with compensation
tion of this is that when future rents are and incentives, ranging from the behav-
high, firms will be less likely to renege ior of professional bowlers to chicken
on their implicit obligations than in farmers to chief executive officers. Let
situations where the firms expect to me conclude here by pointing to what I
gain little in the future from their repu-feel have been the major contribu-
tations. Bertrand (1997) empirically tions and drawbacks of the theoretical
considers this by estimating the likeli- and empirical literature, which could
hood of firms reneging on implicit risk- be used to suggest future research
sharing agreements with their workers. directions.
She shows that firms that are subject to First, from the evidence collected
more competition (or which are in fi- above, it does appear that agents re-
nancial distress) are more likely to re- spond to incentives. Groves et al.
nege on these wage contracting agree- (1994), Lazear (1996), Paarsch and
ments than those firms that earn higher Scherer (1996), Banker, Lee, and Potter
rents. (1996), and Boning, Ichniowski, and
A second use for repeated relation- Shaw (1998) all point to strong effects
ships in agency models is that repetition of pay-for-performance on output, ad-
may allow better inferences to be drawn mittedly in settings where measures of
on performance. For example, there is overall performance were available.
likely to be considerable period-to- Foster and Rosenzweig (1994) provide
period variation in how well a salesforce direct measures of effort costs to sug-
worker fares, and rewarding the agent gest that piece rates are associated with
on performance in a single period may trying harder. Similarly, Knoeber
expose him to considerable risk. How- (1989), Knoeber and Thurman (1994),
ever, the principal can often do better and Ehrenberg and Bognanno (1990)
to consider performance over a longer point to strong effects of prize money
period of time, which may improve in- on the behavior of individuals in tourna-
ferences on whether the agent has ments. Finally, the empirical evidence
shirked in the past or not. This is surely on teams (such as Gaynor and Pauly
important in reality, and there has been 1990) suggests the importance of free
some work on this issue in the litera- riding in teams, though Weiss (1987)
ture, where the principal typically uses and Hansen (1997) suggest effects of
law-of-large-number arguments to gen- peer pressure that are different from
erate better inferences on the perfor- those predicted in the theoretical litera-
mance of the worker, which allows bet- ture. Yet it should not be implied that
ter risk sharing between the principal such responses to incentives are neces-
and the agent. However, these papers sarily beneficial. Evidence from Healy
typically use limiting argument to gen- (1985), Asch (1990), Brown, Harlow,
erate the benefits of repetition (as in and Starks (1996), Chevalier and El-
Roy Radner 1985, for example) where lison (1997), and Courty and Marschke
the agent and workers interact for a (1995) among others, suggests that
large number of periods. As a result, agents are also capable of actions that
this literature has generated few are privately beneficial at the cost of
empirically testable predictions. overall efficiency.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
56 Journal of Economic Literature, Vol. XXXVII (March 1999)

The empirical evidence has also signed for workers in complex jobs, a
pointed to significant selection effects point I will return to below. Finally, the
of contracts. Lazear (1996) illustrates section on deferred compensation
positive selection from the use of piece seems to suggest that firms do indeed
rates; better employees prefer pay for overpay older workers at the expense of
performance. Also interesting is that their younger counterparts. However,
the selection effects appear to be of in those cases, there are typically other
roughly equal size to the incentive ef- plausible interpretations of the data.
fects, despite the overwhelming focus All in all, the available empirical evi-
on incentive effects in the theoretical dence on contracts does not yet provide
literature. Finally, Weiss (1987) has il- a ringing endorsement of the theory.
lustrated the attraction of team-based This could be because the tests consid-
pay, not to the worst workers as pre- ered are weak, or because the theory is
dicted by simple theory, but to those of not capturing all the relevant features
medium ability, where the best and of compensation contracts. Many of the
worst find the constraints of team constraints on the literature have been
production unattractive. imposed by data limitations; there are
The available evidence suggests that simply no easily accessible databases
incentives do matter, for better or for with personnel data. Seen in this light,
worse. It is much less clear, however, it is unsurprising that much of the work
whether the theoretical models based on incentives has been on executives,
on this premise have been validated in for whom there are publicly available
the data. The true test of agency theory data. In addition, it seems clear that an-
is not simply that agents respond to in- other limitation of the literature has
centives, but that the contracts pre- been the fact that contracts are often
dicted by the theory are confirmed by unobserved, where they must be in-
observed data. Here the literature has ferred from the empirical relationship
been less successful. The literature on between pay and performance, which is
the trade-off between risk and incen- tainted with many confounding effects.
tives has had mixed results. Some This is not meant as a criticism of the
authors, such as Kawasaki and McMil- literature; the best work is being done
lan (1987) and Ittner, Larker, and Rajan with the available data. But it is not sur-
(1996), find evidence of such a trade- prising that recent successes in estimat-
off, while Garen (1993) and Bushman, ing the effect of agency contracts con-
Indejikian, and Smith (1996) find little. sider settings where data on contracts
Even in cases where the effects are have been observed, and a critical com-
present, the results are sometimes ponent of future research will surely be
brittle or explain very little of the vari- the collection of such data.
ation in observed contracts. Similarly, A second problem that pervades this
there is mixed evidence on the impor- literature is identification, which comes
tance of relative performance evalu- in two guises. The first is the standard
ation. This is not to say that these theo- empirical identification problem, where
ries are not correct, merely that the jury the researchers need to understand why
is still out. It is difficult to know contracts vary across environments. It
whether the theoretical predictions on is not enough to simply compare the
subjective contracts stand up to empiri- productivity of workers on piece rates
cal scrutiny, because there is so little to those on salaries to estimate the
literature on how contracts are de- effect of pay on performance. Various

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 57

selection problems have been consid- measures are easily observed. Largely
ered in the literature above, and a cen- because of data availability, there has
tral task of recent contributions to the been considerable work done on such
literature has been to explicitly model occupations as chief executive officers,
the source of that variation.82 The sec- golfers, mutual fund managers, tree cut-
ond problem of identification occurs at ters, windshield installers, and so on,
a theoretical level. The typical theoreti- for whom it is possible to construct ob-
cal paper addresses how a certain in- jective measures of output. Work on
stitution may be optimal. Comparative these occupations has provided impor-
statics, when offered, are usually of the tant insights into how incentives oper-
form that institutions or contracts are ate and how they translate into con-
likely to vary with certain parameters. tracts. However, to put it simply, most
However, almost no theoretical work people don't work in jobs like these. In-
has distinguished among plausible theo- stead, most workers are evaluated on
ries there. For many observed phenom- subjective criteria, where firms choose
ena, a multiplicity of theories are con- how to evaluate and how to pay based
sistent with the facts.83 Consequently, a on those evaluations. The literature on
second necessary ingredient for future personnel and human resources man-
empirical research is that theoretical agement has long understood that a dif-
work be aimed at better distinguishing ficult aspect of compensation is the
among theories.84 evaluation of such workers. For in-
The literature has been successful in stance, how do firms get supervisors to
providing an important organizing tool tell the truth about their subordinates?
for understanding the compensation Contracts surely reflect these concerns,
practices of firms, and the empirical yet the economics literature has had
work has cast light on those aspects that relatively little to say, beyond the obser-
appear to be most important. However, vations in Section 2.2. I believe that sig-
a final problem with the literature thus nificant progress could be made by em-
far has been an excessive focus on the pirically understanding how subjective
contracts of workers for whom output assessments are made. How are deci-
sions made on performance and how do
82 See Prendergast (1995) for a discussion of evaluations translate into pay, training,
various identification strategies. and promotion decisions? With what
83 For example, Main, O'Reilly, and Wade
factors do such decisions vary? This is a
(1990), among others, have tested for the impor-
tance of tournament theory by considering (i) difficult exercise, since there is no obvi-
whether wages rise in a convex fashion as one ous taxonomy to categorize types of
moves up the hierarchy, and (ii) whether the prize
subjective performance evaluation, but
for becoming CEO is increasing in the number of
contestants for the job. These outcomes have gen- I believe it would be useful.
erally been found to be true, which is consistent To conclude, agency theory has pro-
with tournament theory. However, it is equally
vided an important framework for un-
true of a hierarchy as in Rosen (1982), where
workers are allocated to jobs on the basis Qf com- derstanding compensation issues. Not
parative advantage without incentives being rele- surprisingly, there has been a lag in
vant. (Wages rise in a convex fashion due to the
testing some of the empirical predic-
magnification effect. Wages increase in the num-
ber of competitors as the best of N workers is on tions, though the last couple of years
average better than the best of N-1 workers.) Con- have seen considerable advances made.
sequently, it is difficult to determine which theory The available empirical evidence ap-
best predicts the data.
84 See Gibbons and Waldman (1998) for one pears to be supportive of the theory,
approach to this problem. though not resoundingly so in some

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
58 Journal of Economic Literature, Vol. XXXVII (March 1999)

settings. As mentioned above, the em- Asanuma, Banri and Tatsuya Kikutani. 1992. "Risk
Absorption in Japanese Subcontracting: A Mi-
pirical work has been restricted partly croeconomic Study of the Automobile Indus-
through the unavailability of data on try," J. Japanese Int. Econ., 6, pp. 1-29.

contracts, which is being rectified. Sec- Asch, Beth. 1990. "Do Incentives Matter? The
Case of Navy Recruiters," Ind. Lab. Relat. Rev.,
ond, empirical and theoretical work 43, pp. 89-107.
needs to continue to address the impor- Athey, Susan and Scott Stern. 1997. "An Empirical
tant identification issues that plague the Framework for Testing Theories About Com-
plementarity in Organizational Design,"
literature. Finally, there is a lot left to mimeo., MIT.
learn about the evaluation of workers Bailey, Richard M. 1970. "Economies of Scale in
whose output is hard to see, where ob- Medical Practice," in Empirical Studies in
Health Economics. H. Klarman, ed. Baltimore:
jectives and outcomes are determined ohns Hopkins Press.
by superiors. Since this constitutes most Baker, George. 1992. "Incentive Contracts and
of us, this seems a large hole to fill in Performance Measurement," J. Polit. Econ.,
100, pp. 598-614.
the literature. Baker, George; Robert Gibbons, and Kevin J.
Murphy. 1994. "Subjective Performance Mea-
REFERENCES
sures in Optimal Incentive Contracts," Quart. J.
Abowd, John. 1990. "Does Performance-Based Econ., 109, pp. 1125-56.
Managerial Compensation Affect Corporate Baker, George; Michael Gibbs, and Bengt Holm-
Performance?" Ind. Lab. Relat. Rev., 43:3, pp. strom.. 1994a. "The Wage Policy of a Firm,"
s52-74. Quart. J. Econ., 109:4, pp. 921-55.
Acemoglu, Daron and Andrew Newman. 1997. . 1994b. "The Internal Economics of a
"The Labor Market and Corporate Structure," Firm: Evidence from Personnel Data," Quart.
mimeo., MIT.
Aggarwal, Raj and Andrew Samwick. Forthcoming.
aJ. Econ., 109:4 pp. 881-919.
Ba er, George and Bengt Holmstrom. 1995. "In-
"The Other Side of the Tradeoff: The Impact of ternal Labor Markets: Too Many Theories, Too
Risk on Executive Compensation," J. Polit. Few Facts," Amer. Econ. Rev., 85:2, pp. 255-
Econ. 59.
Aghion, Philippe and Jean Tirole. 1997. "Formal Baker, George; Michael Jensen, and Kevin J. Mur-
and Real Authority in Organizations," J. Polit. phy. 1988. "Compensation and Incentives: Prac-
Econ., 105, pp. 1-29. tice vs. Theory," J. Finance, 43, pp. 593-616.
Akerlof, George. 1976. "The Economics of Caste Banker, Rajiv; Seok-Young Lee, and Gordon Pot-
and the Rat Race and Other Woeful Tales," ter. 1996. "A Field Study of the Impact of a
Quart. J. Econ., 90, pp. 599-617. Performance-Based Incentive Plan," J. Acc.
Akerlof, George and Lawrence Katz. 1989. "Work- Econ., 21, pp. 195-226.
ers Trust Funds and the Logic of Wage Pro- Barberis, Nicholas; Maxim Boycko, and Andrei
files," Quart. J. Econ., 104, pp. 525-36. Shleifer. 1996. "How Does Privatization Work?
Allen, Franklin and Douglas Gale. 1992. "Mea- Evidence from Russian Shops," J. Polit. Econ.,
surement Distortion and Missing Contingencies 104, pp. 764-91.
in Optimal Contracts," Econ. Theory, 2:1, pp. Barro, Jason and Robert Barro. 1990. "Pay, Perfor-
1-26. mance and the Turnover of Bank CEOs," J. La-
Alston, Lee. 1981. "Tenure Choice in Southern bor Econ., 8, pp. 448-81.
Agriculture," Explor. Econ. Hist., 18, pp. 211- Becker, Brian and Mark Huselid. 1992. "The
32. Incentive Effects of Tournament Compensa-
Alston, Lee and Robert Higgs. 1982. "Contractual tion Systems," Admin. Sci. Quart., 37, pp. 336-
Mix in Southern Agriculture since the Civil 50.
War: Facts, Hypotheses and Tests," J. Econ. Bernhardt, Dan. 1995. "Strategic Promotion and
Hist., 42:2, pp. 327-53. Compensation," Rev. Econ. Stud., 62, pp. 315-
Anderson, Eric and David Schmittlein. 1984. "In- 39.
tegration of the Sales Force: An Empirical Bertrand, Marianne. 1997. "From the Invisible
Examination," RAND J. Econ., 15, pp. 385-95. Handshake to the Invisible Hand? How Product
Anderson, Kathryn; Richard Burkhauser, and Jen- Market Competition Changes the Employment
nie Raymond. 1993. "The Effect of Creaming Relationship," mimeo., Harvard U.
on Placement Rates Under the Job Training Bertrand, Marianne and Sendhil Mullainathan.
Partnership Act," Ind. Lab. Relat. Rev., 46, pp. 1997. "Executive Compensation and Takeover
613-25. Legislation," mimeo., Harvard U.
Antle, Richard and Abbie Smith. 1986. "An Em- Bishop, John. 1987. "The Recognition and Reward
pirical Investigation of the Relative Perfor- of Employee Performance," J. Lab. Econ., 5,
mance Evaluation of Corporate Executives," J. pp 36-56.
Acc. Res., 24. pp. 1-39. Bjerke, D.; J. Cleveland, R. Morrison, and W. Wil-

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firms 59

son. 1987. "Officer Fitness Report Evaluation mance: Lessons from a Federal Job-Training
Study," Report TR88-4, Navy Personnel Re- Program," Amer. Econ. Rev., 87, pp. 383-88.
search and Development Center. Deci, Edward. 1971. "The Effects of Externally
Blackwell, David; James Brickley, and Michael Mediated Rewards on Intrinsic Motivation," J.
Weisbach. 1994. "Incentives of Divisional Man- Personality Soc. Psych., 18 p. 105-15.
agers: Evidence of Texas Banks," J. Acc. Econ., Demougin, Dominique and Acops'ous Siow. 1994.
17, pp. 331-58. "Careers in Ongoing Hierarchies," Amer. Econ.
Boning, Brett; Casey Ichniowski, and Kathryn Rev., 84, pp. 1261-77.
Shaw. 1998. "Incentive Pay for Production Dewatripont, Mathias; Ian Jewitt, and Jean Tirole.
Workers: An Empirical Assessment," mimeo., 1997. "Information Structures and Career Con-
Carnegie Mellon. cerns," mimeo., ECARE.
Bretz, R. and George Milkovich. 1989. "Perfor- Drago, Robert and Gerald Garvey. 1998. "Incen-
mance Appraisals in Large Organizations: Prac- tives for Helping on the Job: Theory and Evi-
tice and Research Implications," mimeo., Cor- dence," J. Labor Econ., 16: 1, pp. 1-25.
nell U. Ehrenberg, Ron and Michael Bognanno. 1990.
Brown, Charles. 1990. "Firms Choice of Method "The Incentive Effects of Tournaments Revis-
of Pay," Ind. Lab. Relat. Rev., 43:3, pp. 165-82. ited: Evidence from the European PGA Tour,"
Brown, Keith, W.V. Harlow, and Laura Starks. Ind. Lab. Relat. Rev., 43, pp. 74-89.
1996. "Of Tournaments and Temptations: An Encinosa, William; Martin Gaynor, and Jim Re-
Analysis of Managerial Incentives in the Mutual bitzer. 1997. "The Sociology of Groups and the
Fund Industry," J. Finance, 51, pp. 85-110. Economics of Incentives," NBER Working Pa-
Bull, Clive. 1987. "The Existence of Self-Enforc- per 5933.
ing Wage Contracts," Quart. J. Econ., 102, pp.Enikson, Tor. 1996. "Executive Compensation
147-59. and Tournament Theory: Empirical Tests on
Bushman, Robert; Raffi Indejikian, and Abbie Danish Data," mimeo., Aarhus U.
Smith. 1996. "CEO Compensation: The Role of Fairburn, James and James Malcomson. 1996.
Individual Performance Evaluation," J. Acc. "Performance, Promotion, and the Peter Princi-
Econ., 21, p. 161-93. ple," mimeo., U. Sussex.
Camerer, CoIin and Robert Hogarth. 1997. "The Fama, Eugene. 1980. "Agency Problems and the
Effects of Financial Incentives in Experiments: Theory of the Firm," J. Polit. Econ., 88, pp.
A Review and Capital-Labor-Production The- 288-307.
ory," mimeo., U. Chicago. Fee, Edward and Charles Hadlock. 1997. "Man-
Cappelli, Peter and Keith Chauvin. 1991. "An In- agement Turnover and Product Market Compe-
terplant Test of the Efficiency Wage Hypothe- tition: Empirical Evidence from the US News-
sis," Quart. J. Econ., 106, pp. 769-87. paper Industry," mimeo., U. Florida.
Carmichael, Lorne. 1983.."Firm Specific Human Ferguson, Leonard. 1949. "The Value of Acquain-
Capital and Promotion Ladders," Bell J. Econ., tance Ratings in Criterion Research," Personnel
14, pp. 251-58. Psych., 2, pp. 93-102.
. 1988. "Incentives in Academics: Why is Fernie, Sue and David Metcalf. 1996. "It's Not
There Tenure?" J. Polit. Econ., 96, pp. 453-72. What You Pay It's the Way That You Pay It And
Cheatham, Carole; Dorothy Davis, and Leo That's What Gets Results: Jockeys Pay and Per-
Cheatham. 1996. "Hollywood Profits: Gone formance," London School Econ. Discus. Paper
With the Wind?" CPA J., pp. 32-34. 295.
Chevalier, Judith and Glenn Ellison. 1997a. "Risk Ferrall, Chris and Bruce Shearer. Forthcoming.
Taking by Mutual Funds as a Response to In- "Incentives and Transactions Costs Within the
centives," J. Polit. Econ., 105:6, pp. 1167-200. Firm: Estimating an Agency Model Using Pay-
. 1997b. "Are Some Mutual Fund Managers roll Records," Rev. Econ. Stud.
Better than Others? Cross-Sectional Patterns in Ferrall, Chris and Anthony Smith. 1997. "A Se-
Behavior and Performance," mimeo., U. Chi- quential Game Model of Sports Championship
cago. Series: Theory and Estimation," mimeo.,
Conyon, Martin and Simon Peck. 1997. "Corpo- Queens U.
rate Tournaments and Executive Compensa- Foster, Andrew and Mark Rosenzweig. 1994. "A
tion: UK Evidence," mimeo., U. Warwick. Test for Moral Hazard in the Labor Market:
Coughlin, Ann and Chakravarthi Narasimhan. Contractual Arrangements, Effort, and Health,"
1992. "An Empirical Analysis of Sales-Force Rev. Econ. Statist., 76, pp. 213-27.
Compensation Plans," J. Bus., 65, pp. 93-123.Freeman, Richard and James Medoff. 1984. What
Coughlin, Ann and Ronald Schmidt. 1985. "Execu- Do Unions Do? New York: Basic Books.
tive Compensation, Management Turnover Friebel, Guido and Michael Raith. 1997. "Strate-
and Firm Performance," J. Acc. Econ., 7, pp. gic Recruiting and Chains of Command,"
43-66. mimeo., U. Chicago.
Courty, Pascal and Jerry Marschke. 1995. "Moral Garen, John. 1994. "Executive Compensation and
Hazard Under Incentive Systems: The Case of a Principal-Agent Theory," J. Polit. Econ., 102,
Federal Bureaucracy," mimeo., U. Chicago. pp. 1175-99.
-. 1997. "Measuring Government Perfor- Gaynor, Martin and M. Pauly. 1990. "Compensa-

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
60 Journal of Economic Literature, Vol. XXXVII (March 1999)
tion and Productive Efficiency in Partnerships; Executive Compensation for Future Perfor-
Evidence from Medical Group Practice," J. mance," mimeo., Northwestern U.
Polit. Econ., 98, pp. 544-74. Hackman, J. Richard and Greg Oldham. 1980.
Gibbons, Robert. 1987. "Piece-Rate Incentive Work Redesign. Reading, MA: Addison Wesley.
Schemes," J. Labor Econ., 5, pp. 413-29. Healy, Paul M. 1985. "The Effect of Bonus
. 1996. "Incentives and Careers in Organi- Schemes on Accounting Decisions," J. Acc.
zations," in Advances in Economics and Econo- Econ., 7:1-3, pp. 85-107.
metrics: Theory and Applications. David Kreps Heckman, James; Carolyn Heinrich, and Jeff
and Ken Wallis, eds. Cambridge: Cambridge U. Smith. 1997. "Assessing the Performance of
Press. Performance Standards in Public Bureaucra-
Gibbons, Robert and Kevin J. Murphy. 1990. cies," Amer. Econ. Rev., 87, pp. 389-96.
"Relative Performance Evaluation for Chief Ex- Heckman, James; Jeff Smith, and Chris Taber.
ecutive Officers," Ind. Lab. Relat. Rev., 43, pp. 1996. "What do Bureaucrats Do? The Effects of
30-52. Performance Standards and Bureaucratic Pref-
. 1992. "Optimal Incentive Contracts in the erences on Acceptance into the TPA Program,"
Presence of Career Concerns," J. Polit. Econ., in Reinventing Government and the Problem of
100, pp. 468-506. Bureaucracy. G. Libecap, ed. Stamford, CT:
Gibbons, Robert and Michael Waldman. 1998. "A JAI Press.
Theory of Wages and Promotion Dynamics In- Hermalin, Ben. 1993. "Managerial Preferences
side a Firm," mimeo., Cornell U. Towards Risky Behavior," J. Law, Econ., Or-
. Forthcoming. "Careers in Organizations: gan., 9, pp. 127-36.
Theory and Evidence," in Handbook of Labor Hermalin, Ben and Michael Weisbach. 1988. "The
Economics. Determinants of Board Composition," RAND J.
Gibbs, Michael. 1995. "Incentive Compensation in Econ., 15, pp. 589-606.
a Corporate Hierarchy," J. Acc. Econ., 19, pp. Holmstrom, Bengt. 1979. "Moral Hazard and Ob-
247-77. servability," Bell J. Econ., 10, pp. 74-91.
Gibbs, Michael and Wallace Hendricks. 1996. - . 1982a. "Managerial Incentive Problems: A
"Are Administrative Pay Systems a Veil?" Dynamic Perspective," mimeo., Kellogg School
mimeo., U. Illinois. of Management.
Gompers, Paul and Josh Lerner. 1994. "An Analy- . 1982b. "Moral Hazard in Teams," Bell J.
sis of Compensation in the US Venture Econ., 13, pp. 324-40.
Capital Partnership," mimeo., Harvard Business . 1996. "The Firm as a Subeconomy,"
School. mimeo., MIT.
Green, Jerry and Nancy Stokey. 1983. "A Com- Holmstrom, Bengt and Paul Milgrom. 1990.
parison of Tournaments and Contracts," J. Polit. "Regulating Trade Among Agents," J. Inst.
Econ., 91, pp. 349-65. Theor. Econ., 146:1, pp. 85-105.
Groshen, Erica and Alan Krueger. 1990. "The . 1991. "Multitask Principal Agent Analyses:
Structure of Supervision and Pay in Hospitals," Incentive Contracts, Asset Ownership and Job
Ind. Lab. Relat. Rev., 43, pp. S134-47. Design," J. Law, Econ., Organ., 7:special issue,
Grossman, Sanford and Oliver Hart. 1983. "An pp. 24-52.
Analysis of the Principal-Agent Problem," Holmstrom, Bengt and Joan Ricart I Costa. 1986.
Econometrica, 51, pp. 7-45. "Managerial Incentives and Capital Manage-
. 1986. "The Costs and Benefits of Owner- ment," Quart. J. Econ., 101:4, pp. 835-60.
ship: A Theory of Vertical and Lateral Organiza- Holmstrom, Bengt and Jean Tirole. 1989. "The
tion," J. Polit. Econ., 94, pp. 691-719. Theory of the Firm," in Handbook of Industrial
Groves, Theodore; Yongmiao Hong, John McMil- Organization. Vol. 1. Richard Schmalansee and
lan, and Barry Naughton. 1994. "Autonomy and Robert Willig, eds. North-Holland, pp. 61-133.
Incentives in Chinese State Enterprises," Hutchens, Robert. 1986. "Delayed Payment Con-
Quart. J. Econ., 109:1, pp. 183-209. tracts and a Firm's Propensity to Hire Older
Hall, Brian and Jeff Liebman. 1997. "Are CEOs Workers," J. Lab. Econ., 4, pp. 439-57.
Really Paid Like Bureaucrats?" NBER Working Itoh, Hideshi. 1992. "Incentives to Help in Multi-
Paper 6213. Agent Situations," Econometrica, 59, pp. 611-
Hansen, Daniel. 1997. "Worker Performance and 36.
Group Incentives: A Case Study," Ind. Lab. Re- . 1993. "Coalitions, Incentives and Risk
lat. Rev., 51, pp. 37-49. Sharing," J. Econ. Theory, 60, pp. 410-27.
Hart, Oliver and Bengt Holmstrom. 1987. "The Ittner, Christopher; David Larker, and Madhav
Theory of Contracts," in Advances in Economic Rajan. 1997. l"The Choice of Performance Mea-
Theory. T. Bewley, ed. Cambridge: Cambridge sures in Annual Bonus Contracts," Account.
U. Press. Rev., 72:2, pp. 231-55.
Hart, Oliver and John Moore. 1989. "Property Janakiraman, Surya; Richard Lambert, and David
Rights and the Nature of the Firm," J. Polit. Larker. 1992. "An Empirical Investigation of
Econ., 98, pp. 1119-58. the Relative Performance Evaluation Hypothe-
Hayes, Rachel and Scott Schaefer. 1997. "Implicit sis," J. Acc. Res., 30, pp. 53-69.
Contracts and the Explanatory Power of Top Jensen, Michael and Kevin J. Murphy. 1990. "Per-

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
 Prendergast: The Provision of Incentives in Firnms 61

formance Pay and Top Management Incen- and Microeconomic Forecasters," NBER Work-
tives," J. Polit. Econ., 98:2, pp. 225-64. ing Paper 5284.
John, George and Bart Weitz. 1989. "Salesforce Landers, Renee, James Rebitzer, and Lowell Tay-
Compensation: An Empirical Investigation of lor. 1996. "Rat Race Redux: Adverse Selection
Factors Related to the Use of Salary versus In- in the Determination of Work Hours in Law
centive Compensation," J. Market. Res., 26, pp. Firms," Amer. Econ. Rev., 86:3, pp. 329-49.
1-14. Landy, Frank and J. Farr. 1980. "Performance
Jones, Derek and Takeo Kato. 1995. "The Produc- Rating," Psych. Butl., 87, pp. 72-107.
tivity Effects of Employee Stock-Ownership Larkey, Patrick and Jonathon Caulkins. 1992. "All
Plans and Bonuses," Amer. Econ. Rev., 85:3, pp. Above Average,"' mimeo., Carnegie Mellon U.
391-415. Lazear, Edward. 1981. "Agency, Earnings Profiles,
Kahn, Charles and Gur Huberman. 1988. 'Two- Productivity, and Hours Restrictions," Amer.
Sided Uncertainty and Up-or-Out Contracts," J. Econ. Rev., 71:4, pp. 606-20.
Lab. Econ., 6, pp. 423-44. . 1986. "Salaries and Piece Rates," J. Bus.,
Kahn, Lawrence and Peter Sherer. 1990. "Contin- 59:3, pp. 405-31.
gent Pay and Managerial Performance," Ind. . 1989. "Pay Equality and Industrial Poli-
Lab. Relat. Rev., 43, pp. 107-21. tics," J. Polit. Econ., 97:3, pp. 561-80.
Kandel, Eugene and Edward Lazear. 1992. "Peer . 1995. Personnel Economics. Cambridge
Pressure and Partnerships," J. Polit. Econ., and London: MIT Press.
100:4, pp. 801-17. .- 1996. "Performance Pay and Produc-
Kaplan, Steven. 1994. "Top Executive Rewards tivity," NBER Working Paper 5672.
and Firm Performance: A Comparison of Japan Lazear, Edward and Robert Moore. 1984. "Incen-
and the United States," J. Polit. Econ., 102:3, tives, Productivity and Labor Contracts," Quart.
pp. 510-46. J. Econ., 99:2, pp. 275-96.
Kawasaki, Seiichi and John McMillan. 1987. "The Lazear, Edward and Sherwin Rosen. 1981. "Rank
Design of Contracts: Evidence from Japanese Order Tournaments as Optimal Labor Con-
Subcontracting," J. Japanese Int. Econ., 1:3, pp. tracts," J. Polit. Econ., 89:5, pp. 841-64.
327-49. Leibowitz, Arleen and Robert Tollison. 1980.
Knez, Marc and Duncan Simester. 1992. "Firm "F'ree Riding, Shirking and Team Production in
Wide Incentives and Mutual Monitoring at Legal Partnerships," Econ. Inquiry, 18, pp.
Continental Airlines," mimeo., U. Chicago. 380-94.
Knoeber, Charles. 1989. "A Real Game of Leonard, Jonathon. 1987. "Carrots and Sticks: Pay,
Chicken: Contracts, Tournaments and the Pro- Supervision, and Turnover," J. Labor Econ., 5,
duction of Broilers," J. Law, Econ., Organ., 5:2, pp. s136-52.
pp. 271-92. Lepper, Mark; David Greene, and Richard Nis-
Knoeber, Charles and Walter Thurman. 1994. bett. 1973. "Undermining Children's Intrinsic
"Testing the Theory of Tournaments: An Em- Interest with Extrinsic Reward," J. Personality
pirical Analysis of Broiler Production," J. Lab.
Social. Psych., 28, pp. 129-37.
Econ., 12:2, pp. 155-79. Leventis, Andrew. 1997a. "Cardiac Surgeons Un-
Kotlikoff, Lawrence and Jagadeesh Gokhale. 1992. der the Knife," mimeo., Princeton U.
"Estimating a Firm's Age-Productivity Profile .- 1997b. "The Relation Between Surgeon
Using the Present Value of a Worker's Earn- Age and Outcomes," mimeo., Princeton U.
ings," Quart. J. Ecort., 107, pp. 1215-42. Loewenstein, George and Nachum Sicherman.
Kreps, David. 1997. "Intrinsic Motivation and Ex- 1991. '-Do Workers Prefer Increasing Wage
trinsic Incentives," Amer. Econ. Rev., 87, pp. Profiles?" J. Labor Econ., 9, pp. 67-84.
359--65. MacLeod, Bentley. 1994. "Incentives in Organiza-
Kruse, Douglas. 1993. "Does Profit Sharing Affect tions: An Overview of Some of the Evidence
Productivity?" mimeo., Rutgers U. and Theory," in Trends in Business Organiza-
Laffont, Jean-Jacques and Mohamed Matoussi. tion. Horst Siebert, ed. Kiel: J.C.B. Mohr.
1995. "Moral Hazard, Financial Constraints, MacLeod, Bentley and James Malcomson. 1989.
and Sharecropping in El Oulja;' Rev. Econ. "Implicit Contracts, Incentive Compatibility,
Stud., 62, pp. 381-99. and Involuntary Unemployment," Econo-
Lafontaine, Francine. 1992. "Agency Theory and metrica, 57, pp. 447-80.
Franchising: Some Empirical ResulAts,' RAN.D MacLeod,
J. Bentley and Daniel Parent. 1998. "Job
Econ., 23:2, pp. 263-83. Characteristics and the Form of Compensa-
Lambert, Richard and David Larker. 1987. "An tion," mimeo., USC.
Analvsis of the Use of Accounting and Market Main, Brian; Charles O'Reilly, and James Wade.
Measures of Performance in Executive Com- 1993. "Top Executive Pay: Tournament or
pensation Contracts," J, Acc. Res., 25, pp. 85-" Teamwork," J. Labor Econ., 11, pp. 606-28.
125. Malcomson, James. Forthcoming. "New Develop-
Lambert, Richard, David Larker, and Keith ments in the Study of Contracts in Labor Mar-
Weigelt. 1993. The Structure of Organizational kets," in Handbook of Labor Economics.
Incentives," Admiin. Sci. Quart., 38, pp. 438-61. Marschke, Jerry. 1996. "Incentives in a Govern-
Lamont, Owen. 1995. "Macroeconomic Forecasts ment Bureaucracy," mimeo., U. Chicago.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms
62 Journal of Economic Literature, Vol. XXXVII (March 1999)

McLaughlin, Ken. 1994. "Team-Based Compensa- on Business Seasonality," Quart. J. Econ.,

tion Schemes," mimeo., Hunter College. 113:1, pp. 149-86.
McMillan, John; John Whalley, and Lijing Zhu. Paarsch, Harry and Bruce Shearer. 1996. "Fixed
1989. "The Impact of China's Economic Re- Wages, Piece Rates, and Incentive Effects,"
forms on Agricultural Productivity Growth," J. mimeo., U. Laval.
Polit. Econ., 97:4, pp. 781-807. Parent, Daniel. 1996. "Methods of Pay and Earn-
Medoff, James and Katherine Abraham. 1980. ings: A Longitudinal Analysis," mimeo., U.
"Experience, Performance, and Earnings," Sherbrooke.
Quart. J. Econ., 95:4, pp. 703-36. Paul, Jonathon. 1992. "On the Efficiency of Stock-
. 1981. "Are Those Paid More Really More Based Compensation," Rev. Fin. Stud., 5, pp.
Productive: The Case of Experience," J. Human 471-502.
Res., 16:2, pp. 186-216. Pencavel, John. 1977. "Work Effort, On-the-Job
Meyer, Margaret. 1992. "Biased Contests and Screening, and Alternative Methods of Remu-
Moral Hazard: Implications for Career Pro- neration," in Research in Labor Econonomics,
files," Annales d'Econ. Statist., 25-26, pp. 165- Vol. 1, R. G. Ehrenberg, ed. JAI Press, pp. 225-
87. 58.
Meyer, Margaret; Paul Milgrom, and John Petersen, Trond. 1992. "Payment Systems and the
Roberts. 1992. "Organizational Prospects, Influ- Structure of Inequality: Conceptual Issues and
ence Costs, and Ownership Changes," J. Econ. an Analysis of Salespersons in Department
Manage. Strat., 1:1, pp. 9-35. Stores," Amer. J. Soc., 98, pp. 67-104.
Meyer, Margaret and John Vickers. 1997. "Perfor- Prendergast, Canice. 1993a. "The Role of Promo-
mance Comparisons and Dynamic Incentives," tion in Inducing Specific Human Capital Acqui-
J. Polit. Econ., 105:3, pp. 547-81. sition," Quart. J. Econ., 108:2, pp. 523-34.
Milgrom, Paul. 1988. "Employment Contracts, In- . 1993b. "A Theory of Yes Men," Amer.
fluence Activity and Efficient Organization," Econ. Rev., 83:4, pp. 757-70.
J. Polit. Econ., 96, pp. 42-60. . Forthcoming. "What Happens Within
Milgrom, Paul and John Roberts. 1988. "An Eco- Firms?" in Labor Statistics and Measurement.
nomic Approach to Influence in Organizations," John Haltiwanger, Marilyn Manser, and Robert
Amer. J. Soc., 94, pp. s154-79. Topel, eds. NBER; Chicago: U. Chicago Press.
Milkovich, George and Alexandra Wigdor. 1983. Prendergast, Canice and Lars Stole. 1996. "Im-
"Motivation and Performance Appraisal Behav- petuous Youngsters and Jaded Oldtimers: Ac-
ior," in Performance Measurement and Theory, quiring a Reputation for Learning," J. Polit.
Frank Landy, S. Zedeck and J. Cleveland, eds. Econ., 104:6, pp. 1105-34.
Hillsdale, N.J.: Erlbaum. Prendergast, Canice and Robert Topel. 1996. "Fa-
Mohrman, A. and Edward Lawler. 1991. Pay for voritism in Organizations," J. Polit. Econ.,
Performance: Evaluating Performance Ap- 104:5, pp. 958-78.
praisal and Merit Pay. Washington: National Putterman, Louis. 1990. "Effort, Productivity, and
Academy Press. Incentives in a 1970s Chinese Peoples Com-
Mookherjee, Dilip. 1984. "Optimal Incentive mune,"J. Compar. Econ., 14, pp. 88-104.
Schemes With Many Agents," Rev. Econ. Stud., Radner, Roy. 1985. "Repeated Principal-Agent
51, pp. 433-46. Games With Discounting," Econometrica, 53,
Murphy, Kevin J. 1986. "Incentives, Learning and pp. 1173-98.
Compensation: A Theoretical and Empirical In- Raff, Daniel. 1988. "Wage Determination and the
vestigation of Managerial Labor Contracts," Five-Dollar Day at Ford," J. Econ. Hist., 48, pp.
RandcJ. Econ., 17, pp. 59-76. 387-99.
. 1998. "Performance Standards in Incen- Rob, Raffi and Peter Zemsky. 1997. "Cooperation,
tive Contracts," mimeo., USC. Corporate Culture and Incentive Intensity,"
. Forthcoming. "Pay, Performance, and Ex- mimeo., U. Pennsylvania.
ecutive Compensation," in Handbook of Labor Rosen, Sherwin. 1982. "Authority, Control, and
Economics. the Distribution of Earnings," Bell J. Econ., 13,
Murphy, Kevin R. and Jeannette Cleveland. 1991. pp. 311-23.
Performance Appraisal: An Organizational Per- . 1986. "Prizes and Incentives in Elimina-
spective. Boston: Allyn and Bacon. tion Tournaments," Amer. Econ. Rev., 76, pp.
Nalebuff, Barry and Joseph Stiglitz. 1984. "Prizes 921-39.
and Incentives: Towards a General Theory of Rothe, H.F. 1949. "The Relation of Merit Ratings
Compensation and Competition," Bell J. Econ., to Length of Service," Personnel Psych., 2, pp.
2, pp. 21-43. 237-42.
Newhouse, Joseph. 1973. "The Economics of Rundquist, E.A. and R. H. Bittner. 1948. "Using
Group Practice," J. Human Res., 8, pp. 37-56. Ratings to Validate Personnel Instruments: A
O'Reilly, Charles; Brian Main, and Graef Crystal. Study in Method," Personnel Psych., 1, pp. 163-
1988. "CEO Compensation as Tournament and 84.
Social Comparison: A Tale of Two Theories," Salop, Steven and Joanne Salop. 1976. "Self-Selec-
Admin. Sci. Quart., 33, pp. 257-74. tion and Turnover in the La or Market," Quart.
Oyer, Paul. 1998. "The Effect of Sales Incentives J. Econ., 90:4, pp. 619-27.

This content downloaded from

131.130.169.5 on Wed, 06 Jan 2021 09:03:29 UTC
All use subject to https://about.jstor.org/terms