Anization (A J) 1976

The Optimal Structure of Incentives and Authority within an Organization
Author(s): James A. Mirrlees

Source: The Bell Journal of Economics, Vol. 7, No. 1 (Spring, 1976), pp. 105-131
Published by: RAND Corporation
Stable URL: http://www.jstor.org/stable/3003192 .
Accessed: 05/09/2014 12:55
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
RAND Corporation is collaborating with JSTOR to digitize, preserve and extend access to The Bell Journal of
Economics.
http://www.jstor.org
This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

All use subject to JSTOR Terms and Conditions
The optimalstructure
of incentives
within
and authority
an organization
James A. Mirrlees
Professorof Economics
NuffieldCollege, Oxford
Two kinds of models for a productiveorganizationare presented. In

thefirst,bothproductionand rewards are based on theperformance
of individuals, which is perfectlyobserved. Their abilities are not
observable. Despite this, theoremsare proved givingstronggrounds
for the equality of wages and marginal products unless there is
monopsonyin the labor market. This latter case is also discussed.
The second model, whichfocuses on the imperfectobservation of
performance,allows interestingdeductions about optimal payment
schedules and organizational structure.
* The usual idea of an organizationis thatit is a group of people (or 1. Introduction

roles) withinwhicha structureof authorityis defined.In otherwords,
the actions of each member of the organizationare constrainedby
certain of the decisions made by other members. Simon (1957) has
developed this view, and it has been taken up more recently by
Arrow (1974). Simon and Arrow, in common with many political
theorists,emphasize the existence and desirabilityof limitson author-
ity,which are termed"responsibility." More generally,it is clear that
relationsof authoritycan, and usually do, operate in both directions
between any two membersor subgroupswithinan organization.But,
even allowingforthis, the possibilitiesof organizationalstructureare
perhaps ratherricherthan those suggested by the term"authority,"
as Marschak and Radner (1972, p. 313) have indicated. In theirwork
on the "theory of teams," theychoose to concentrateon a different,
thoughrelated, aspect of organizations-the diversityof information
available to the members. At the same time,theynarrowtheiratten-
James A. Mirrleesreceived the M.A. fromthe Universityof Edinburgh(1957), the

B.A. fromthe Universityof Cambridge(1959), and the Ph.D. fromthe Universityof
Cambridge (1963). His research centers on systems of incentive and control under
uncertainty.
I am most gratefulto Oliver Williamsonfor suggestingthe subject and invitingme
to present the paper at the Symposium on The Economics of Internal Organization,
which was held at the Universityof Pennsylvania,September 19-21, 1974. The Sym-
posium was supported by grants from the National Science Foundation (NSF-GS-
35889X) and the General Electric Foundation. I am gratefulto the participantsin that
conference for their comments on the firstdraftof this paper, and particularlyto
Michael Spence, whose perception of a serious error, and constructivesuggestions,
stimulateda complete revision of Section 2 of the paper. MIRRLEES / 105

tion, and propose a correspondinglynarrowdefinitionof "team," to
organizationswhose members have common preferences.Some au-
thors have talked more loosely of teams as groups of people who
together can achieve more than if they act separately [Mirrlees
(1972a), Alchian and Demsetz (1972)]. Where such a coalition is
possible, one can consider the possibilityof the group's agreeingto
work togetherwithoutsettingup a systemof authority.Alchian and
Demsetz propose a model of the firmin which the functionof the
managementhierarchyis simplyto measure the labor input of mem-
bers of the group, payments being in accordance with contractual
agreement. They even seem to claim that this is the only model
appropriateto firms,and thatauthoritativerelationshipsdo not occur.
This last claim cannot be accepted, but it may be agreed thatthere
are interestingpossibilities of organization which one would find
difficultto describe in termsof authority,and that these possibilities
can be applied to the same production possibilities as can more
authority-basedmodes of organization. The common featureis per-
sonal relationshipbetween membersof the group,withan established
pattern of interaction. Thus conceived, organizations may include
sharecroppingtenancy of a man's land, or even bank lending; for in
these cases, the contractgoverningthe use of the lender's asset is
based on personal information,and may specifyaspects of individual
behavior [cf. Cheung (1968), Stiglitz(1974)]. Indeed, thereis no sharp
line to be drawn between perfect-market relationshipsand intraor-
ganization relationships: arrangementsmay vary in a continuous
mannerfrompure tradingat an exogenouslyestablishedprice to near
perfectobedience by one party to the command of the other. One
wants to consider the whole of this spectrum,so as to explain actual
organizationalstructures,and prescribe better ones.
It is hard to specifythe spectrumof possible economic relation-
ships in a way that begins to be adequate. In this paper, I make the
task easier by completelyignoringall bargaining.It mightbe thought
thatthisignorestoo much; but thereare interestingsuggestionsin the
literature,including some already mentioned, which avoid game
theory.For example, Williamson(1970) advances a model of organi-
zations, based on imperfectcommunication,fromwhich is derivedan
optimumsize of firmeven when the technologyexhibitsconstantor
increasingreturnsto scale. The models to be developed in thepresent
paper provide a similartheory,based on a more detailed theoryof the
relationshipsbetween the parties involved; but with ratherdifferent
consequences.
In these models, the members of the organizationhave different
interests,and behave in accordance with theirown interests.We are
thereforenot dealing with a team in the Marschak-Radner sense.
Members of the organizationtake independent,but related,decisions.
Thus, thoughrelated to Wilson's theoryof syndicates(1968) (where
diverse individuals share in the consequences of a single decision),
our theoryis essentiallydifferent.The models are closely related to
the theoryof land tenure systems (mentionedabove), the theoryof
agency [Ross (1973)], and models of behavior subject to moral hazard
[Pauly (1968), Zeckhauser (1970), Spence and Zeckhauser (1971)].
This being so, it is as well to repeat the point that Arrow has made
THE BELL JOURNAL about the theory of moral hazard (1971, p. 220), that in situations
106 / OF ECONOMICS where moral hazard arises, thereis, potentially,general advantage in

moral behavior, i.e., behavior not motivatedby narrow self-interest;
and thatsuch behavioroccurs. The models used in thispaper assume
that contracts,explicit and implicit,are exploited by the parties in
theirown interest,so that promises and claims about unobservable
behavior, for instance, are not admissable. A theory that overem-
phasizes self-interested behavior in this way deserves to fail in pre-
dictingvarious featuresof actual organizations;but it would be sur-
prisingif it were wholly irrelevant.
The aim of the models in the paper is to explain the distributionof
incomes within the firm,to explain the existence of authoritative
relationships,and to derive a hierarchicalstructureand investigateits
effectson production. Naturallymuch of this programis imperfectly
worked out, and importantelementsare missing. Throughout,uncer-
taintyplays an essential role. There may be uncertaintyabout the
tasks (or the value of the tasks) thatany memberof the organization
will be undertaking,about his capabilityto undertakethese tasks, and
about the way in which he will or has undertakenthese tasks. Uncer-
taintiesof these kinds suggest an analogy with the theoryof public
finance,where the governmentis assumed to have limitedinformation
about the characteristicsof the populationit rules. Section 2 develops
the theoryof a profit-maximizing organization,which has imperfect
information about the qualities of its workerswhen theyare recruited,
and uses the model to elucidate the relevance of incentiveconsidera-
tions to payment schedules. The model assumes, however, that
workersare perfectlyable to predicttheirown activityand rewards
withinthe organization. It is thereforeimpossible to discuss many
interestingquestions, particularlythe structureof authority.
These issues are takenup in Sections 3-5, where the models allow
some scope forimperfectrewardadministration and monitoring.Sec-
tion 3 deals withthe organizationthat,fromthe point of view of this
paper, seems to be the simplest: an organizationconsistingof two
members,one of whom is subordinate(in a sense to be made precise)
to the other. This example is helpfulin allowing one to model and
explore the various kinds of uncertaintymentionedabove. The model
is thenextended to become a two-levelorganizationin Section 4, and
a multilevelorganizationin Section 5. In Section 6-where we return
to a less technical level afterthe mathematicalcomplicationsof the
previous three sections-features of the models are brieflyrelated to
the concepts of control-lossand control-spanand theirconsequences.
Section 7 summarizes the argument.
* It is commonlyasserted that one reason for the hierarchicalpay 2. The pay

structurewithinorganizationsis to be found in the incentives they structure for
provide. The person whose work is foundgood is promoted,and the unknown
managersand owners of the organizationare thus enabled to discover abilities
which members of the organization do good work. None of this
makes sense for the firmin perfectcompetition.But that is because
perfect competition assumes that the properties, including the
abilities, of each potentialworkerare public knowledge. For incen-
tives to have a role, it is necessary that the managementhave only
imperfectinformationabout the abilities and willingnessof men and
womento work: thentheyare interestedin the way thata structureof MIRRLEES / 107

wages and salaries elicitsmuch valuable activityfromthe able and the
energetic,while allowing others to choose a more moderate work
level and position. It may suit to pay more than the marketrequires.
At any rate, it appears thatconsiderationof the incentivescreated by
the pay structureshould be an importantelement in explainingit.
To explore these matters,a reasonably simple and extrememodel
is suggested. In it, workers are completely aware of their own
abilities,and choose how hard to work on that basis. Employers,on
the other hand, cannot distinguishamongjob applicants, and there-
fore set pay schedules which are applied to all comers. There are
many firmsin the economy, all wanting to maximize profits.The
behavior of workers,which theydeterminethemselvesin the lightof
the pay structure,is assumed to determinethe firm'soutput. Thus
there is no explicit role for authoritativerelations, and the model
looks much like the orthodoxmonopsonymodel withthe firmsetting
prices and the workerschoosing labor supply. The case of a perfectly
elastic supply of workersto the firmwill, however, be discussed, as
well as the monopsonisticcase. The question to be chieflyconsidered
is the relation,in equilibrium,between the wages which workersof
differentskills receive and theirmarginalproductivities.
C] Technologies.Production possibilities will be assumed to take a

form which considerably generalizes the special model which has
been used in recent years to analyze the correspondingproblem in
public finance,that of optimal income taxation. The generalization
does not greatly affect the details of the analysis, except at one
importantpoint; but it is importantto establish that the analysis
applies to a model whereina hierarchicalstructureof employmentis
natural, since that is the object of study in the present paper.
Each firmin this model produces a single kind of output, in
amount y, with fixed factors (which are ignored notationally)and
differentkinds of labor. The work done by a workeris denoted by z:
differentworkerschoose to provide different z. z is a measure of the
quantityand quality of work, one-dimensionalfor simplicity.Output
is taken to be a functionof the distributionof z. This is best illus-
trated by some examples, where z is distributedas a continuous
variable with densityfunctionf:
y = H(fzf(z)dz), (1)
y = h exp (fa(z) log f(z)dz), (2)
y = H(fa(z)H-'(f(z))dz), (3)
or
y = G(f 4A3(z)f(z)dz, f,y (z)f(z)dz). (4)

(1) is the function that has usually been used, for simplicity,in
income-taxtheory,where the contributionsof differentworkers are
perfectlysubstitutiblefor one another. (2) generalizes the Cobb-
Douglas productionfunctionto a continuumof inputs. (3) is a more
general class, including(1), (2), and the generalized CES functionas
special cases. (4) captures the notion that workers may be used in
eitherof two different kinds of activity(e.g., manual or supervisory),
THE BELL JOURNAL the levels of each contributing to total outputthroughthe functionG.
108 / OF ECONOMICS This last form is particularlyworthy of attentionbecause it can

capture the idea that some workersare promotedaftera time when
the employerhas acquired sufficientinformationabout theirworking
performance;and promotionis to a differentkind of activity. This
kind of arrangementis frequentlymentionedas a reason for paying
supervisoryworkers more: it can provide an incentivefor those in
less senioroccupations to workhard. One of our aims is to check this
argument.Perhaps a moreprecise statementof the situationwould be
y = G(b + Xc, (1 - X)c), b = fZEAf3fdZ, C = fzAvfdZ, (4a)
where Xis the (discounted)proportionof career spentin the "lower"
occupation.
One wants to extend these definitionsto generaldistributionsof z,
where, say, a nonzero proportionof the labor force all supply the
same value of z. The extensionis done by continuity.For example, in
case (2), a group of workersconcentratedat a singlevalue of z has no
effecton output.
The assumptionthat output depends only on work levels, z, and
not on ability,n, distinguishesthis model fromthatof Spence (1973),
and others, where a worker's productivityalso depends on n. This
dependence is crucial for Spence's results. Work by Rothschildand
Stiglitzon Spence's model shows thatthe n-independentproductivity
assumption is importantfor the results to be proved below.
In each of the examples above (with suitable differentiability of
the unspecifiedfunctions)outputis a differentiable functionof labor
inputs. This means, in the presentcontext,thatwhen the distribution
f depends differentiably on a parameter E, there exists a functionp
such that, if y is the output resultingfromf
de y = fp(z) a f(z,E)dz. (5)
p(z) is, in a naturalsense, the marginalproductof a workerproviding

work z. Note that in general (i.e., apart fromthe simple case (1)) p
depends on f( ), the distributionof z withinthe labor force. It will be
assumed in what follows thatp exists.
O Workers.Members of the labor force are characterizedby a single

parametern. When faced witha paymentschedule w(z), a workerof
type n chooses z(n) so as to maximize a function,strictlyconcave in
w and z ,
u(w(z),z,n). (6)
w(z) is to be thoughtof as the presentvalue of earningsin the firm,
possibly througha series of promotions,if the workerbehaves in the
way described by z. It is assumed that uw > 0, uz < 0, un > 0.
Though otherinterpretations are possible, n will here be interpreted
as "skill." The maximizationof (6) implies,if everythingis differenti-
able, and z nonzero, that
uww'(z) + uz = 0. (7)
This can be writtenin the form
w'(z) = s(w,z,n), (8)
where s = -uzl/u, is the marginalrate of substitution.The interpre-
tationofn as skill suggeststhatwe assume Sn < 0. The distributionof MIRRLEES / 109

skills, n, withinthe labor force of the firmis described by a density
function+(n).
An alterationof the pay schedule w changes the functionz(n). We
must see how this change affectsoutput,the labor force being given.
Let F be the distributionfunctionof z. Then F(z(n)) is unchanged,so
that variations in F and z are related by
SF(z) + F'(z)Sz(n) = 0,
i.e.
SF = -f(z)Sz(n). (9)
From (5), Sy = fp(z)SF'(z)dz = -fp'(z)SF(z)dz, integratingby
parts. Therefore,using (9),
ay = fp'(z)Sz(n)f(z)dz
fp'(z(n))5z(n)O(n)dn.
= (10)
The specificationof the model is now complete, except for the
conditionsof supply of workers to the firm.The whole spectrumof
possibilitieswill be considered,fromthe firmwitha given labor force
to the firmfacinga perfectlyelastic supply of workers. The case of
perfectlyelastic supply is thatwhere workersof type n are available
to the firmprovided they are assured of some thresholdutilitylevel
(their supply price). Formally, the firmwill be said to operate in a
competitivelabor market when it can recruitand keep workers of
type n if and only if
Max
z
u(w(z),z,n) ? ui(n). (11)
u(n) will be the utilityavailable to the workerof type n in alternative

employment.If all inequalities (11) hold, the firmknows that each
recruitis drawnat randomfroma populationwitha givendistribution
described by ?(n). If some are violated, the recruitis drawn from
thatpart of thepopulationwhere (11) is satisfied.The extensionof the
model to an intermediatecase, to be called the monopsonisticcase,
will be explained later.
The analysis proceeds by means of a series of propositions,the
thirdof which is followed by an importantremark.
Proposition 1. For a firmoperatingin a competitivelabor market,
with a pay schedule which satisfies(11), in equilibrium
ffp(z) -
w(z)}s4(n)dn = 0. (12)
Proof. Consider the effectof recruitingone more worker into the
firm.The expected change in output per recruitis, by (5),
fpp(z)f(z)dz = fp(z(n))p(n)dn,
where z(n) is the work-level chosen by a worker of type n. The
average labor cost per recruitis
fw(z)f(z)dz = fw(z(n))4(n)dn.
It is clear fromthese two equations thatequilibriumis possible onlyif
(12) holds.
In a competitiveequilibrium,all firms(at least if identical)pay the
THE BELL JOURNAL same wage schedules, and the constraints(11) hold with equality.
110 / OF ECONOMICS Rather than attemptto approach the question of the shape of the
equilibriumpay schedule under these conditionsdirectly,I look first

at what is, in effect,the opposite extreme. This is a firmthat has a
given labor force; withthe supplementaryconsiderationthatworkers
would leave if wages fell too low, a constraintthat will not apply to
most of them in equilibrium.
Proposition 2. If a firmmaximizes profitsfor a given labor force,
subject to the constraints(11); and if for all n,
u(p(z(n)),z(n),n) ::- (n), (13)
then for all n
w(z(n)) c< p(z(n)). (14)
Proof. For simplicity,proof is confinedto the case where a density
f(z) exists. Suppose, in order to obtain a contradiction,that the firm
has a wage schedule satisfying
w(z) > p(z), for a < z < b, (15)
with w = p at z = a, b. Constructa familyof schedules w(z,E)
dependingdifferentiably
on E, with
= w(z)
w(z,0) 1
w(z,E) = w(z) when z c a and b c (16)
WE(Z,E) < 0 when a < z < b
w,(z,E) = w,(z) when z = a, b.
This variationof the pay schedule is illustratedin Figure 1. (13) and
(16) implythat, for all small enough E,
u(w(z(n),E),z(n),n) > u7(n) (a < z(n) < b).
FIGURE 1
VARIATION OF THE PAY SCHEDULE

p,w
P(Z)
w(z)
< E
W
(Z,f)--<,
H
LU /
0~
MIRRLEES / III
a b
WORK

Therefore,the proposed variation is consistentwith the constraints
(11), and has no effecton the membershipof the labor force.
The variationis to be chosen in such a way thata densityfunction
f(z,E) exists forthe distributionof work levels, and so thatf(z,E) is a
differentiable functionof E. That is why we mustinsist,in (15), that w
remains a smooth functionof z as E increases.
Applyingequation (5), we have
d = -
d {y - fw(z,E)f(z,E)dz} f[(p - w)fE WEf]dz
= d -
w)f(z,E)dz
dE
d
d f
J{p(z(n,E)) - W(z(n,E),E)}O(n)dn,
where z(n,E) is the work-levelchosen by an n-workerwhen faced

withw(, E). Therefore,the firmwould like to increase E ifforeach n
aE {p(z(n,E)) - w(z(n,E),E)} > 0. (17)
In order to calculate this expression, we need to know the partial

derivativeof z(n,E) withrespect to E. A workerof type n chooses z
so that w' = s (equation (8)). Differentiatingthisrelationshippartially
with respect to E, one obtains
wIE + w" Z E = SW (WE + W Z E) + SZ ZE,
where the E-subscriptdenotes differentiation:w' E, for example,

means a2w/(aEaz). From this it follows that
Ze = W'Ew SW WE (18)
SW*- W' + SZ - W
Using this result, we have
de (w-p) = wE + (w -P') ZE
(w - + + Sz -
p')w'E (SWp' W")WE (19)
SWW, + Sz - W"
using (18) to substitutefor Z E We want to be able to choose WE

satisfying(16) in such a way that this expression is negative.
Utilitymaximizationimplies that the denominatorin (19) is non-
negative, for this is the condition that the indifferencecurve lie
(locally) above the budget constraintw(z). Let us assume slightly
more, viz.,
sww' + sz - w" > 0 (a c z c b). (20)

Other(exceptional) cases are presumablyeasily dealt with. If we now
try definingw(, E) by
WE = - (w -p)r (a c z ? b) (21)
withr > 1, it is readilychecked that(16) is satisfied;and substitution
in (19) yields
d (w + - p)rl
r(w' p
p)2
dE (w -p SWW' + S - W )
THE BELL JOURNAL
112 / OF ECONOMICS
s(w
SW - p)W' -p')+ (w - P)(SwW' + sZ -
"J

The expression withinbraces is a quadratic formin w' - p', and
thereforepositive definiteif
4r(s w' + s - w") > s 2(w - p). (22)
Because of our assumption(20), we can findr to satisfy(22) for all z
in the interval.Therefore,the variationdefinedby (21) does have the
desired propertywhen r is chosen large enough.
Thus inequality(17) is established, and it follows thatthe original
payment schedule cannot have been profit-maximizing for the firm.
The propositionis proved.
This propositionis best appreciated if we consider one firmin an
environmentof otherfirmsall payingless than the marginalproducts
for laborers. For this situation,the propositionasserts that the firm
will not choose to pay any of its workers more than his marginal
product. Thus we may say thatincentiveconsiderationsin themselves
give no reason forpaying anyone more than his marginalproduct. If
any workeris to be paid more than his marginalproduct,it must be
by reason of some constraintimposed on the firm.It is not easy to see
where such a constraintcould come fromin an industryof like firms.
Indeed, in a special case, a much strongerresult can be proved.
Proposition 3. Under competitiveconditions, with productionpos-
sibilitiesavailable to each firmdescribed by
y = H(fzf(z)dz), (1)
in equilibrium,
w(z) = p(z) = H'(fz(n)4(n)dn)z. (23)
Proof. In equilibrium with a competitive labor market, all firms
have the same pay schedule, and u-(n) = max, u(w(z),z,n) =
u(w(z(n)),z(n),n). It will firstbe shown that,for this pay schedule,
w(z) c p(z) for all z. The reason is essentially simple: any firm
constrained(by competition)to pay more than the marginalproduct
forsome worklevel zo can increase its profitsby offeringless thanthe
marginalproduct and doing withoutwork-levelsin the neighborhood
of z0.
Formally, let us suppose that
w(zo) > p(Zo). (24)
zo is the work-levelsupplied workersof type no, and we can choose
zo satisfying(24) so that no otherworkerssupply zo, if necessary by
slightlychangingit. Furthermore,we can take it that zo is the unique
work-levelworkersof type no are prepared to choose, because of the
followinglemma.
Lemma. For the case described by (1), the equilibriumwage schedule
has the propertythatforeach n, u(w(z),z,n) is maximizedby a single
value of z.
Proof of lemma. Suppose, contraryto the statedresult,that z1 and Z2
both maximize u(w(z),z,no). If W(Z2) - W(Z1) 7 (Z2 - zj)H', sup-
pose withoutloss of generalitythat z1H' - w(z1) > z2H' - w(z2). If
w is slightlyincreased in the neighborhood,a small groupwithn c nO
change fromapproximatelyZ2 to approximatelyz1, and a small group
withn < nO change, but remainclose to z1. The effecton profitsof E MIRRLEES / 113
workerschangingfrom Z2 to z1 is

E(Z, - Z2)H' - E(W(Z1) -W(Z2)),
a positive numberof order E. The othereffectsare of order E2. Thus,

profitscan be increased by the proposed perturbation.See Figure 2.
FIGURE 2
INCREASING PROFITS WHEN W(z2) - W(zl) #(Z2 - z1 )H'

w
nO-INDIFFERENCE
CURVE \
p(z) = H'(z)
w(z)
(n0+E)-INDIFFERENCE
Lu CURVE \
(no-- E) -INDI FFERENCE

CURVE ~
NEW W(z)
_ Z
i
Zl Z2
WORK
If W(Z2) W(Z1)-
(Z2 - z1)H', profitscan be increased by
=
raisingthe pay schedule slightlyabove the n0-indifferencecurvejoin-

ing (z1,w(z1)) and (z2,w(z2)), as shown in Figure 3. This completes
the proof of the lemma.
Returningto the proofof the proposition,we considerthe effectof
reducingthe pay schedule forthe firmin the neighborhoodof zo (Fig-
ure 4). This can be done in such a way thatpreciselythose n between
n - E and no + E are now unable to attainu(n). For all othern, z re-
mains unchanged. fzfdz is reduced by "n+Ez z4dn and the wage bill
is reduced by {nO+E w(z)4dn. Thus, profitsare increased by
fnO+E wqdn - {fo+E z4dn .

H'(fz4dn). (25)
Now in the case of the productionfunction(1), it is easily seen that

p(z) = zH'.
Therefore,(25) can be writtenas
{O+E -
{w(z(n)) p(z(n))}4dn,
THE BELL JOURNAL which is positive, by (24) and the continuityof w and p.
114 / OF ECONOMICS This proves that whenever (24) holds, profitscan be increased.
Therefore,in equilibrium,

FIGURE 3
INCREASING PROFITS WHEN w(z2) - w(zl) = (Z2 - z1 )H
w
nO-INDIFFERENCE
CURVE w' w(z)
p(z) H'z
14es
LuI
0~~~~~~~~~~~~~~~~~~~~~~-0
z
Z1 Z2
WORK
FIGURE 4
LOCAL REDUCTION OF THE PAY SCHEDULE

W
(nO+ E)-INDI FFERENCE
/
4 /
~~~~~~~~~~CURVE
nO-INDIFFERENCE
CU RVE /,
\-i / ~~P(z) H'z
(no- E) INDI F F ER ENC
EW w(z)
MIRRLEES / 115
Z? WORK

w(z) c p(z) (26)
forall relevantz. Combining(26) withthe result(12) of Proposition1,
we have in fact
w(z) =P(Z)
as was to be proved.
The proof of this proposition-apart from the lemma which is
merelydisposing of a special case-does not consider the incentive
effectsof the pay schedule, but only the effectson the supply of
laborers. Indeed, it is littlemore than the usual argumentforequality
between wages and marginalproducts, with special care taken to
check thatno incentiveeffectsoccur when a wage is reduced. But the
productionfunctionto which the argumentapplies is a very special
one, which, by assuming perfectsubstitutibility among the different
work-activitiesin the firm,excludes just those featuresof industrial
productionthat one associates with the existence of job hierarchies.
Probablytheproofof thepropositioncould be extendedto cover such
cases as are suggestedin (4) and (4a) above. The argumentshould fail
only when a large change in work-levelby a small numberof workers
has a large effecton marginalproductivities.But that is, surely, an
interestingand large class of cases.
For the general case, a less precise, but almost equally compel-
ling, argumentcan be offered.Suppose that, for a range of work-
levels, wage-rates are above marginalproducts. The firmdoes not
reduce these wage-rates if that would lose all the corresponding
workers, because it needs such workers. But, to be a little more
realistic, a small reductionin wage rates will not suddenly lose all
such workersto the firm.A temporaryreductionwould simplyreduce
recruitingfor a time. Then the change in other marginalproducts is
small and can properlybe neglected: the wage reductionincreases
profits,so long as it does not, and is not expected to, go on too long.
There is, then, a general weakness in wage levels if they ever get
above marginalproducts, so that we can expect them to crumble.
When this argumentis coupled withthe implicationof Proposition 1,
thatincentiveconsiderationswill not in themselvespush wages above
marginalproducts, it is hard to resist the conclusion that, under
competitiveconditions, with unidentifiablelabor skills, wages and
marginalproducts are equal in equilibrium.
This conclusion has been based on an industrywith identical
firms.We should also discuss the equilibriumof a group of firmswith
differentproduction functionstaking their workers from the same
labor market.If all were recruitingthe same mixtureof abilities,they
would, in general,have different impliedmarginalproductivities,and
it would followfrom(23) thatdifferent firmswere payingaccordingto
differentschedules. If workers are well informedabout payment
schedules, this is impossible: differentwage schedules would sort
worker-typesamong firms. A disequilibriumprocess of this kind
could provide different firmswith labor forces of differentcomposi-
tion until p( ) is the same for all, and the wage-paymentschedule
uniform.Thus, equilibriumfor an industryis characterized by the
orthodoxuniformity of marginalproductivitiesand wages. The result,
THE BELL JOURNAL it should be reiterated,has been obtained withoutthe usual assump-
116 / OF ECONOMICS tion that employersknow the abilities of those they are hiring.

Since the resultis not, despite appearances, the standardone, it is
subject to anomalies. The propositiontells us what equilibriummust
be like if the firmis recruitingin a perfectlabor market,but not that
equilibriumexists. I have in mind the possibilitythat a firmwith an
establishedlabor forcemay be able to increase its profitsby changing
to a paymentschedule which will no longerattractnew recruits.This
will be a possibilityif employees cannot expect the same present
value of earningsin anotherfirmas those who occupy similarposi-
tions to theirown. The firmthey are workingfor has acquired infor-
mationabout them(representedby the z they supply) and has placed
them in the organization appropriately. They may have spent an
initialperiod withthe firmat lower pay, establishingtheircredentials
forpromotion.It is not in the interestsof theiremployerto tell a new
employerwhat grades or positions they had reached.
Thus, at least in the case of employees with some seniority,the
employerhas a monopolyadvantage. He may, of course, be unable to
exercise it in the face of collective action by his employees. If he
succeeds in exercisingit, he will cut out recruitmentof workerswho
have sufficient abilityto aspire to higherpaid positions. The policy is
appropriate,therefore,only fora decliningfirm.It may not be a very
realistic possibility. Where it occurs, it means that declining firms
have a less steep wage payment schedule than growingfirms.
O Imperfectlabor markets. The question is thus raised how flat a

paymentschedule it could profitthe firmto adopt bearingin mindthe
disincentiveeffects.I examine this in a more general setting,where
the labor marketis imperfect.Imperfectionwill be captured by ex-
tendingassumption (11): the utilitylevel required to attractanother
worker of ability n is assumed to be an increasingfunctionof the
numberalready employed. Formally, the firmis constrainedby
max u(w(z),z,n) 2 ui(n,O(n)), (27)
ii is an increasingfunctionof 4 (as well as

where the utility-threshold
of n). As before, the firmwishes to maximize
y - fw(z)c(n)dn,
where z = z(n). is the value chosen by a worker of ability n.

The techniques appropriateto this problem are those developed
for the theory of optimal income taxation [Mirrlees (1971)]. The
followingnonrigorousdevelopmentmakes it reasonably easy to see
where the various conditions come from. First, I express the con-
straintsthat, for each n, z(n) maximizes u(w(z),z,n), in a way
convenientforthe problem. Assuming-as one has no rightto do, but
u,tw'(z) + uz = 0.
we are not beingrigorous-that w is differentiable,
(Note that w(-) will be so chosen that z(n) > 0.) Therefore,
dd u(w(z),z,n) = un(w,z,n), (28)
thepartial derivativewithrespect to n. This "envelope condition" is

plainly equivalent to the first-order
condition; and it is convenient
because utilityalso appears in the constraint(27). Define
v(n) = u(w(z(n)),z(n),n), (29) MIRRLEES / 117

and express w and u,, as functionsof v, z, and n, definedby u(w,z,n)
-v:
w = w(v,z,n), Un = qi(v,z,n). (30)

Differentiation
of u(w,z,n) = v with respect to v and z shows that
WV= 1/u., wz = -uz/uw = S. (31)
Using these results, we then calculate
=
v = Unw/uw qz = UnwS + Unz -UwSn (32)
Afterthese preliminaries,we can express the problem as

maximize y - fwc (n)dn,
subject to
v 2 ii(n,o(n)), v'(n) = q(v,z,n),
and introduce Lagrange multipliersX(n), ,u(n) for the two sets of
constraints.Thus,
L = y - fw4dn + fX{v - W(n,4)}dn + fl(n){v'(n) - +}dn
y + f{- w4 + X(v- i) - '(n)v- qj}dn + /i(oo)v(oo)
- A (O)v(0)
is to be stationarywhen v(*), z(*), and ,u(.) are varied. Differentiat-
ing with respect to z(n), we have
p'(z) ) - w z4 - Az = 0,
or
(p'(Z) - S) ) = -puwsn. (33)
with respect to v(n) yields

Differentiation
-wwv + X - ,' - /AIJv= 0
or
uwp. ' + unw,u, = uwX - 4; (34)
with respect to 4 gives

and, finally,differentiation
p - w=XA,+. (35)
(If 4 were zero, we could have p - w < XAu<.)The way in which p'
and p came out in these differentiations may not be quite clear,
althoughit is readilyjustifiedby the argumentsgiven earlier. Heuris-
changes), y is a
tically,we can take it thatlocally (i.e., forfirst-order
constantplus fp(z)0(n)dn. Once thisis granted,it is clear where (33)
and (35) come from.
There are two furtherconditionsto note, arisingbecause L must
be stationaryas v(0) and v(oo) are varied:
A(0) = (oo) = 0. (36)
Also it should be noted that, since the multiplierX is associated
with the inequality v : ii, X 2 0, and vanishes if v > -u. Since by
assumption, iu, > 0, (35) implies that, 4 being positive,
p : w, withequalityif v > u-. (37)
THE BELL JOURNAL
118 / OF ECONOMICS (It is prettyclear that equality would be very exceptional.) This

confirmsand generalizes the earlier result. Intuitivelyone expects
that the firmfeels constrainedby the need to have v'(n) at least as
great as qp.If it were so, ,u would also be nonnegative. But this
depends of course on the way in whichthe otherconstraintoperates.
To see what is going on, let us adopt a definitionof the elasticity
of supply of workers of type n:
T)(n) = wu./(4iiO). (38)
This is the percentage increase in +(n) obtained for a one percent
increase in w; thatincrease beingpaid onlyforz(n). (We ignorein the
definitionthe consequent effecton supplies of other labor-types.)
Using (35) and (38), (34) becomes
Uwa ' + Unw W - i) (39)
The firm'soptimal payment schedule is definedby (39) with (33) in

the form
p'(Z) - w'(z) = -Iuwsn, (40)
the worker's maximization condition w '(z) = s, the labor supply
condition u(w,z,n) = ii(n,4(n)), and the boundaryconditions u(0)
= 0 = A (OO).
Since n describes a worker's capabilities, we are assuming that
Sn < 0, (41)
meaningthata man withgreatern is more able (or willing)to substi-

tute labor for consumption.Under this assumption,it is clear that w
and z must be increasingfunctionsof n. It may not be possible to
measure n in such a way that(41) holds; but forthe remainderof this
section, I assume thatit is. Then (40) shows that ,u > 0 is equivalent
to the wage schedule's being less steep at z(n) than the marginal
productivityschedule.
Solving (39) for A, we obtain
P - 1) dm (42)
n() ( W
where
, = exp (Unw/uw)dv . > 0. (43)

Jn uw
T
,u(O) = 0 is used in deriving (42). A(oo) = 0 implies that
( p -
1)_
ldm = 0. (44)
It follows thatforsome n, - (p - w) /w exceeds 1, while forothersit

is less than 1. It cannot, in interestingcases, equal 1 forall n because
that would imply ,u = 0, and therefore p' = w', i.e., w(z) - p(z) =
constant. This would mean w = constant x 7), which we may reject
since it is not likelythat7-increases with n, whereas w certainlydoes
(grantedassumption (41)).
Let us assume-as seems plausible-that
-qis a decreasingfunctionof n. (45)

Is it possible that ,u is negative for small n? If it were, MIRRLEES / 119

a - (46)
#I1
would be negativefor small n. Now, since , is negative,p' - w' <

0, and, therefore,p - w is decreasingin n; while w is increasingin n,
and 7) decreasing. Therefore,a is decreasing, and remains negative.
The argumentwould continueto apply, and a would be negativefor
all n. This is, because of (44), impossible. The argumentused actually
tellsus that,once , is negativeforsome no, itremainsnegativeforall
n ? nO. Thus we have the following:
Proposition 4. Assuming (41) and (45), there exists no such that (a
being definedby (46))
a0O (n n0) (47)
a O (n no)J (
Furthermore,for all n,
y 20; (48)
and, for all z,
w' (z) p' (z). (49)
The second part of the propositionfollows at once fromthe first,for
(by (44)) we can write , either as j8a,f3dm or -ra,3dm. (47) has
obtained by writing(46) in the form
an interestinginterpretation,
pW 1 (l + a). (50)
w 7
If incentive considerations were ignored, the optimal mark-up of

marginalproduct above wage would be 1II- for each n. (47) and (50)
tell us that the incentiveconsiderationsimplya largermark-upthan
monopsonytheorysuggestsfor the less skilled,and a lower mark-up
for the more skilled. At the same time, the absolute differencebe-
tween marginalproductand wage is higherfor the more skilled: this
is the result of assumption (45).
This section has been addressed to the importanceof incentive
considerationsin determiningthe pay structuresof profit-maximizing
firms.It has been shown that the firmmust have the usual kind of
monopsonypower in the labor marketbefore it has much reason to
pay attentionto the incentive effectsof its pay structure.The one
exception to this is the stagnantfirm,and that can be regarded as a
case of monopsony where there is some inelasticityin response to
wage reductionsbut not to wage increases. In the monopsonycase,
which I take to be in some degree realistic,no workerreceives more
than his marginalproduct, but workersof highercapabilityreceive a
wage which is less close to theirmarginalproductthanthose of lower
ability.The last conclusion depends on assumingthathigherabilityis
associated with lower elasticityof supply; which should be true at
least because of the investmentin reputation,and knowledge of the
firm,whicha workerof higherabilitynormallymakes. It is interesting
that,if firmshave imperfectknowledge of theirrecruits'capabilities,
and recruitin imperfectmarkets,they should in theoryapply a pro-
THE BELL JOURNAL gressive tax to marginalproducts beforepaying wages, and thus do
120./ OF ECONOMICS some of an egalitariangovernment'swork for it.

One can use the equations presented to compute optimumpay-
ment schedules in particularcases, but I do not pursue the model
further here. It leaves out a verygreatdeal of what interestsus about
organizations.Presumablythis model ought to be developed in the
directionof allowingthe firmsome limitedinformation about the men
it hires;but it is unlikelythatthe principalpropositionwould be much
affected.In any case, it would take us furtheraway fromthe internal
organizationof the firm.The chief shortcomingof the model is the
lack of an explicit reason for workers to work together,leaving
informationconsiderations highly implicit: therefore,there are no
reasons forthe hierarchicalorganizationthatone observes, and would
like to explain, for itself,and as an importantinfluenceon the pay
structureof firms.In fact, when theyjoin a firm,workersare uncer-
tain what they will be doing, and what reward theywill receive; but
theirrelationshipsare quite mechanical. In the remainingsections of
the paper, I studymodels thatdo not deal explicitlywiththe distribu-
tion of skills, but do deal withthe uncertaintiesof productionactivity
and supervision.
* Consider a man with von Neumann-Morgenstern utilityfunction 3. Principal and

u(x,z), where x is the payment received as a result of the work he agent
does and z is the work done. Payment is taken to be an uncertain
consequence of the work done. It is a functionof the man's perfor-
mance as observed by his principal. Observed performancemay or
may not accuratelyrepresentthe value to the principalof the agent's'
work; and the value of the work, which I shall call outputand denote
by y, may itselfbe a randomvariable conditionedby z. The accuracy
of the principal's observationof y depends upon the time devoted to
makingthe observation.(It should also depend upon the agent's own
effortsto affectthe observation; but I shall ignore this, important
though it may be. I am not sure how best to model the effectof
possibly competingeffortsto influenceobservationalprecision.)
The formalmodel of the agent is that
z maximizes U(z) = EEEYIZU Y + +E) Z1

= ffu{(Y + 6E),z}f(y,z)g(E)dydE, (51)
where 0 is the schedule governingpaymentto the agent,and 02 iS the

time spent (by the principal) on observation-the idea being that he
samples repeatedly. Observationalerrorsand outputuncertaintyare
assumed independent.It should also be emphasized thatall functions
and random variables occurringare supposed to be nice: points of
rigorwill be ignored, although they are very important.Necessary
conditionsfor maximizationin (51) are
U'(z) =0 (52)
and
U"(z) 0. (53)
The principalhandles the outputof the agent and makes payment
to him. It is convenientto assume that the two are commensurable,
and to give the principal a utilityfunctionv(y - p 0 2). Thus, he
would like to choose 0( ) and 0 to maximize MIRRLEES / 121

EEEylZv{Y-Y (y + Eo2
= fv{Y - y + E),02}f(y,z)g(E)dYdE, (54)

given that z is determinedby (51).
One of the odd featuresof this assumptionis thata principalwho
benefitsfromy cannot observe it, although one mightthinkthat a
man can benefitonly fromwhat he observes. But notice that the as-
sumptionwould look much more reasonable ifthe principalhad many
agents all contributingto the outputhe receives-and we shall come
to that case. In any event, it is common for benefitsto accrue long
afterpayments have been made irreversibly;and Y + E should
properlybe regardedas the agreed basis forpayment,whichhas to be
adhered to whatever output may actually be.
Anotherodd featureis thatthereis no veryevidentreason forthe
principal-agentrelationshipin the model. But it is implicitthat the
agent uses assets owned by the principal. There seems to be no
advantage in making that explicit unless one wants to consider the
decision of how much to let the agent use. Interestingthough that
may sometimesbe, it adds nothingto the pictureof the relationship
that I want to convey.
Returningto the mathematicalproblem, we must include another
constrainton the principal's maximization-that the agent gets ex-
to induce him to accept the contract.In other
pected utilitysufficient
words, there should be a supply price for the kind of labor we are
considering.(In the case of slavery,utilitywould not be the relevant
variable constrained;and indeed morale, motivation,and health are
considerations that should perhaps be broughtin, but they would
complicate mattersfurther.)The constraintis
U(z) ? A, (55)
A being a fixed number.

The analysis now proceeds along lines I have used elsewhere.1
Consider the Lagrangean formobtained by assigningundetermined
multipliersX and ,u to (52) and (55):
L = ff{v(y - p,02) + Xu(4,z) + A (U2 + uf,/f)}fgdydE.
(Here and below, numericalsubscriptsto u and v denote differentia-

tion with respect to the indicated argument.) Differentiating with
respect to the scalars z, and the function0( ), we obtain first-order
conditions:
aL = ffvfigdydE + AU'(z) + .U"(z) = 0,
i.e.,
,vh
= - U"(z)-ffvf gdydE. (56)
Before taking the other derivatives,it is advantageous to change a
variable of integrationand write
L = Off{v(y - +(X),02) + Xu(k(x),z)
+ (U2 + uf/f)}f(y,z)g(0(x - y))dydx. (57)
THE BELL JOURNAL
122 / OF ECONOMICS 1 Mirrlees(1972b,1974).See also Spence and Zeckhauser(1971).

Then
L= -o (58)
ao
(Since I shall not be computing solutions, I do not evaluate this
derivativeexplicitly:it is clearly rathera complicated expression.)
aL
a4(x)
= Of{-vI + XuI + W(u12+ uj1f/f)}f(y,z)g(O(x - y))dy =0.
Since u and its derivativesare independentof y, thismay be rewritten
in the form
I fvlfgdy/ffgdy
= A+ , {u12 + ffzgdy/ffgdy}
or equivalently,usingthe notationE( Ix) forexpectationsconditional

upon a given value x of observed (apparent) output,
E(v, +,x{u1 +E(Lfx)j (59)
(56), (58), and (59), along with(52), effectivelydetermineequilibrium,

whichis characterizedby numbersOand z, and thefunction0. Given
6 and z, (59) determines0. Thus, althoughexplicitsolutionof actual
cases is difficult,one can use (59) to findout somethingabout the
shape of 0.
Before discussing that,we need some assumptionsaboutf. Since
larger z is supposed to mean increased effort,and therefore,on
average, output, it should decrease f for small y and increase it for
large. I shall furthermore assume that
fz/fis an increasingfunctionof y. (60)
Since, for all z, ffdy = 1, ffzdy = 0; thus (60) implies thatfz/f is
negativefor small y, positive for large. A similarassumptionis made
for g, modelled on the case where E is a normal random variable:
log g is a concave functionofe. (61)
Lemma. E(fz1flx) is a nondecreasingfunctionof x.
Proof.
=f
xE(fJx) d {ffzg(6(x - y))dy/ffgdy}
= 6{fzg'dy/ffgdy- ffzgdyffg'dy/(ffgdy)2}
= 5
{E( fz x)-E( x) E g )} (62)
Now by assumptions(60) and (61), fz/fand g'/g are both increasing

functionsof y, and, therefore,positivelycorrelated,given x. There-
fore, by (62)
E (fz xl 2: 0 9 (63)
as was claimed. Q.E.D.

This allows us to deduce that, if (i) u > 0, (ii) u12/u1is a MIRRLEES / 123

nonincreasingfunctionof 0, and (iii) u and v are concave in theirfirst
arguments,then
+(x) > 0. (64)
To prove, we consider E(v,lx)/u, as a functionof 0 and x, a(o,x).
An argumentexactly similarto thatof the lemma shows that a, < 0,
while concavityof u and v impliesthat a,,,> 0. From (59), we have
{a4 -
ax, (65)
a? (u12/u)lb,'(x) = E(f/x)
-
d
which, with all our assumptions, implies as desired that f5' 2 0.
Assumption(iii) is fairlyunexceptionable; and somethinglike (ii) is
bound to be required. But (i) is not the kind of thing we should
assume: it should be deduced. From (56)-bearing in mindthat U" s
0 (and ignoringthe exceptional possibilitythat U" = 0)-we see that
, > 0 ifand only if the principalwould like z to be furtherincreased,
if he could controlit directly,the paymentschedule being fixed. This
is plausible, certainly:one expects the payments to be so arranged
that the agent does not do all the principalwould like. But I have
been unable to findgeneral assumptionsthat exclude the possibility
L< O.
Certain special cases are of interest. Since it is necessary to
simplifydrasticallyin order to get muchfurther,I shall fromnow on
assume that
U12= 0. (66)
Case I: Onlyoutputis uncertain.
In this case, E is equal to zero with probabilityone, and (59)
becomes, since x and y are identical,
vi(y - +(Y)) = A + gf,(Y' ) (67)
In this case, if , < 0, + must be a decreasing functionof y; and

ffvfZgdyE= fv(y - cp(y))f (y,z)dy
= f{v(y - q) - v(yo- q(yo))}fffdy(since ffzdy= 0),
where yois a value of y forwhichf, = 0. Thus by (60), ffvfzgdydE>
0 if 4 is decreasing. But thisby (56) impliesthat ,u > 0-a contradic-
tion. Thus, in this case 4 necessarily increases in y. Indeed, since
, > 0, we can say more. Differentiating,
U (1 - ' 2- V U1 ' > O.

Therefore,
-v111 1/v1
-Vii/Vi - Uii/U
absolute riskaversion of principal

sum of absolute riskaversions
The marginalshare going to the agent willbe lowonly ifhis absolute
riskaversion is significantly
greaterthan thatof theprincipal. (70), of
course, allows us to look at the schedule in more detail. But the
TFIE BEILL JOURNAL second case is perhaps more interestingforthe internalorganization
124 / OF ECONOMICS of firms.

Case II: Only performanceobservation is uncertain.
In this case, y = z withprobabilityone. This is a limitingcase of
the generalanalysis, and one musteithergo back to the beginning,or
confidentlysubstitutea delta-functionforf in (59). In any case the
result is
VI(Z - + (x), 02) = A g ((x Z- (69)
By the same kindof argumentas in Case I, but usingassumption(61),

we can show again that A > 0.
This is the model I shall use in the rest of the paper, so it is
time I attended to one importantfeature of the analysis that has
been neglected. If one considers a normal distributionfor E, g(E) =
(2r)->e'26 and
- =- E. (70)
g
Since E varies from -00 to +00, the right-hand side of (69) is some-
times negative. This is a trifleupsetting,since the left-handside is
always positive (assuming u and v are increasing functionsof in-
come). It mightbe thoughtthat this trouble arises because, with a
normal error, observed output can be negative; but the same thing
happens if x = zeEl". What we have to do is allow forthe possibility(
= 0 explicitly. Maximization of the Lagrangean with respect to (
yields an inequalityif ( = 0:
-v1 + Xu1 - Au1g'/g ? 0. (71)
If X - Ag' /g < 0, (71) holds with strict inequality, and consequently
( = 0.
We can see this explicitlyif we consider an example where both
principaland agent have constant absolute risk aversion:
u, = e-ac? , v1 = e-(Z-). (72)
Using normal g, we findthat
(x) = a az
+/
+ I
a +/3 log
(X + AO(x-z))
(x ? x0 = z + (eat- A)/(,u)) (73)
= 0 (x xo)
The formof this schedule is shown in Figure 5. In passing, it should

be noted that the case where u is -oo when + = 0, which I have
discussed elsewherein the contextof insuranceand incentives(1974),
is even more peculiar: the risk of receivingno paymentis made to do
all the work of providingincentives. But this is not the appropriate
assumptionin the present context.
To show that the upper part of the payment schedule is not
necessarily concave, I note two other special cases:
(i) Principal risk-neutral;agent constant relative risk aversion p:
+(x) = [(B + Cx)"1P]+. (74)
B and C are positive constants,determinedby other aspects of the
model.
(ii) Principal and agent have same constantrelativeriskaversion p: MIRRLEES / 125

+(x) = [z{1 + (B + Cx)-1IP}'-]+. (75)
Both (74) and (75) are concave in theirpositive sections if p 2 1.
FIGURE 5
AN OPTIMAL PAYMENT SCHEDULE
0~~~~~~~~~~~~~~~~~~~~~
xoxO~~~~~~~~~~~~~~~~~~~~~~~
WORK
One wants to know how the schedules vary as the chosen 0, and
the agent's elasticityof substitutionbetween income and effort,vary.
Unfortunatelythat is complicated in all the examples I have looked
at, because of the awkward formof condition(56). From (73), it can
be seen that the payment schedule takes a fairlysharp "logistic"
formas in Figure 6 if St eaz and eaz are both large. (These are
FIGURE 6
A PAYMENT SCHEDULE APPROXIMATING AN INSTRUCTION
z
LL
xO~~~~~~~~~~~~~~~~~~~~
WORK
the absolute values of the firstand second derivativesof q at x0.) This

suggests that large risk aversion on the part of the agent combined
withaccurate observationby the principal(reflectedin large 0) give a
THE BELL JOURNAL schedulethatis a bitliketheeffectof an instruction. An instruction
126 / OF ECONOMICS is, in effect,a promisethat rewardswill vary rapidlyin the neighbor-

hood of a particularoutputlevel, but not elsewhere. Unfortunately,
this suggestioncould be confirmedonly if we knew how A depended
on a, ,3, and 0: that is the difficultmathematicalproblem.
In any case, it is clear that the schedule will only exceptionally
take the really sharp formof an instruction.It is not surprisingthat
such policies are generallysuboptimalwhere men are uncertainabout
what they may achieve and about what they may be seen to have
achieved. What is notable is that the optimal payment schedule is
usually as "unfair" as thatshownin Figure 5, in circumstanceswhere
one mighthave expected liberal allowance for unfavorableobserva-
tional inaccuracies.
In the remainingsections of the paper I shall outline how this
simpletwo-personmodel can be used as a buildingblock in construct-
ing a model of a complicated organization.
* Suppose now that the principal has n men to supervise, all 4. Two-level
identical, who as before choose z to maximize U(z); and suppose organ izations
outputsadd (i.e., constantreturns).The principal(followingCase II
n
of the previous model) receives nz - , p(z + Ei/0), there being
independentobservationof the n agents. He would like to choose +
and 0 so as to maximize
V = Ev(nz -
Y4,nO2). (76)
A small change in +(x) changes V by an amount proportionalto
n
-nE2 . . . Env1(nz - 4(x) - +(X1) ), (77)
Ei being the expectation operator for the random variable Xi = z

+ Ei/0. Thereforethe first-order conditionfor ( is (cf. (59))
nE2. . .EnVl +(U12 - g) (X >) (78)
The effectof having numerous subordinateworkersis that the pay-

ment schedule is governed by an average of v1.
If the principalhas constantabsolute risk aversion, the averaging
drops out conveniently:
n{EeX)}n-1e -(nz(x)) 1 = X + u ( U12 - (79)
In this case the optimal paymentschedule has the same formas that
discussed at the end of the previous section. There is no reason to
thinkit would get flatteror steeper as the number of subordinates
increases. But if absolute risk aversion is decreasing, it is a different
matter,because the relative aggregate riskiness of the principal's
receipts diminishesas n increases. In that case, for large n, we can
regardthe principalas effectively risk-neutral.He chooses ( and 0 to
maximize
v(nz - nEO(z + E/0),n02), (80)
and the first-order
conditionforthe paymentschedule is of the form MIRRLEES / 127

I = xi + U12 -g9) (x>O). (81)
The effectof this averagingis, roughly,to reduce the concavityof

the paymentschedule. With our earlier examples,
U12 = 0, g = (2 r) eie2,
relative risk aversions p, o-
ff=-__2_P(P
0
02
- 1) + O(- + 1)
~~(Z 4) 2
+ 2po-
44 Z-4)
j (82)
(2
Thus p < 1 implies convexity if o- = 0, but concavity (except for

small x) if o- >0.
Withoutdoing the mathematics,we can see thatthe model in this
formprovides a reason fornot increasingthe size of the organization
beyond a certainpoint. If the principalchooses n as well as 0, he will
not be willingto increase n withoutlimit. If n is given, theremay be
no solution that gives the principalsufficientutilityto undertakethe
task. The same would be true withincreasingreturnsto the number
of agents (workers) cooperating. A model of this kind would also
enable one to consider under what circumstances it would pay a
group of workers to have one of their number undertake all the
performanceobservation, and when it would pay instead to have a
symmetricsolutionin which each workerdevotes some of his timeto
"monitoring"(one of the others,presumably). It is not obvious that
the asymmetricsolution outlined here, and assumed optimal by Al-
chian and Demsetz (1972), is in fact optimal when the means of
productionare owned in common.
5. Hierarchical * When the numberof workersbecomes large,it presumablypays to

organizations establish an inspection hierarchy.It can be modelled as follows. I
begin with the worker level, and then continue throughsuccessive
supervisorylevels:
(i) z max Eul(XI),z),
where
X- Z + E1 1.
(ii) q),,O max EU2(02(X2),n1O12),

where
.nl ~1 1
X2= n,z - E lI(Xi') +-E2 = Z2 +-E2.
i=l 02 02
(iii) +k2,02 max Eu3()3(X3), n2022),

where
n2 n21
X3 =l, zi2 - Z)2(X.2) + E3

THE BELL JOURNAL
ad s=1 03
128 / OF ECONOMICS and so on, until the last level

Ot-9 ot-, max Eut(Zt,nt_,0t_
2)
where
nt-I nt-I
Zt= zit-I - X 1t I(Xit-1) X
i=l i=l
At each stage, the maximizationis carried out subject to the maximi-

zations at earlier stages, and subject to the supply price (in utility
terms)of the men at each level being satisfied.
In this specific model, it is assumed that aggregate output is
correctlymeasured at each level, and that efficientaccountants set
the wage bill offagainst outputin every department.The ni may also
be chosen, either at each level, or at the top.
It is plain withoutundertakingmathematicsthat a larger viable
organizationcan be created in this way than is possible when re-
strictedto two levels, because, by fixingthe numberssupervised by
any individual,and the supervisiontime per subordinate,and setting
up some simple payment rule (such as fixed proportionsof the de-
partmentalnet income), we should be able to satisfythe utilitycon-
straints.There is then no reason why the organizationshould not be
increased in size withoutlimit.
It requires more carefulanalysis to see whether,despite the obvi-
ous disadvantage of increasingsupervisorystaff,a large organization
of the kind under discussion may have compensatingadvantages. It
can be shown that if (i) each supervisor has the same number of
immediatesubordinates,(ii) errorsof observationare proportionalto
the mean size of the observed variable, (iii) the payment rule is
proportionalat each level, and (iv) all individualsare identical,then
there is a paymentsystem such that each supervisorgets at least as
much expected utilityas a (firstlevel) workerand the income of the
proprietor(at the top) is approximately
Zt - (AN + B + Nl E) z (z = value of outputper worker),
where N is the numberof workers,A and B are constants,and E is a

random variable that is, likewise, independent of N. Optimization
would, of course, achieve more. (I hope to present proof of this
result, and furtheranalysis of the model elsewhere.)
Ignoring errors, this result implies that profitper worker di-
minishes as N increases. But if the cost of other inputs (such as
capital) is less thanzA per man, the entrepreneurwill prefera larger
organization, provided that his relative risk aversion does not de-
crease too rapidly. Perfectcompetitioncould (if B is large enough)
force zA and othercosts into equalityand encourage small firms,but
very little weakeningof competitiveforces is required to renderthe
largerorganizationnot only viable but more profitable.Any increas-
ing returnswould strengthenthe tendency.
The model suggests another interestingconclusion. If we sup-
pose that the membersof the organizationhave constantor decreas-
ing relativerisk aversion (not that that is so very likely), it seems to
follow,since relativeincome riskinessis smallerat higherlevels, that
paymentschedules should be less concave at the higherlevels-less
like instructions,and more like profit-sharing. The reader will not
need to be reminded that, considering the conjectural basis of the
reasoning,and the restrictivenature of the model, this is not a very MIRRLEES / 129

well-foundedconclusion. But it is the kind of conclusion that one
could hope to obtain from such a model as this one.
6. The scale of * It has been suggested that the scale of firmsis limited by the
firms capabilities of managers,by the intrinsicdifficulty of controllinglarge
organizations, and by the communicationloss resultingfrom long
chains of authority. The model of the previous section made no
explicit allowance for the managerial skills that may be required of
men in charge of large departments.In fact the model was motivated
by a desire to see what kinds of complexitymightarise in a large
organization.It gives a fairlyroutinerole to everyonein the firm:men
eitherwork or observe the results of what others have done; and of
course they choose what to do. In this way informationabout the
value of work-which may be interpretedas the value of different
kinds of work-gets conveyed throughthe organization,fromapex to
workers. There is no apparent reason why the top man's job should
be any harder than that of anyone else; no reason why, in the op-
timum policy, middle managers should get more than workers. If
there are reasons why higher-leveljobs in a hierarchyrequire more
abilitythan lower-level ones, they have largely escaped the model.
Perhaps the best candidate is the ability to take decisions: for the
choice of paymentschedule-which stands, of course, for advice and
instruction,redeployment,and even encouragementand discipline-
surely has greater effectsat higherlevels than at lower ones. This
consideration may also support the second reason for diminishing
returnsmentionedabove. An adequate model for skill in decision-
takingremains to be built.
Communicationlosses are the basis of Williamson's analysis of
the size of firms(1970). The evidence quoted by him was experimen-
tal; but it seems plausible thatinformation is lost in long chains. Yet,
the model in the previous section gives an example of how informa-
tion may be conveyed where at each link in the chain, observationis
imperfect,everyone is choosing what to do on the basis of self-
interest,and as a result creates an incentive system which conveys
information to the next man down. The transmissionof information is
imperfect,because, for example, no one has an incentive simplyto
obey an instructionor to pass it on. Yet it turnsout thatthe contribu-
tion of the organizational structureto diminishingreturns is very
weak, becoming forlarge organizationsessentiallynegligible.Uncer-
taintyin communicationdoes not necessarily implyincreased losses
in proportionto the size of the organization.
7. Final remarks * Two differentmodels of an organization have been discussed,

complementingotherkindsof models thathave appeared in the litera-
ture. In these models, imperfectinformationbinds the organization
together.They are thereforevery much in the spiritof the Alchian-
Demsetz paper referredto at the beginning.The models-particularly
in the latterpart of the paper-seem at present too cumbersome to
answer many of the questions one would ask of them. For example,
the analysis of optimal payment schedules is seriously incomplete,
and nothinghas been said about the optimumshape of the organiza-
THE BELL JOURNAL tion (how nl, n2, . . , nt-I should compare to one another).
.
130 / OF ECONOMICS Apart fromthe interestthat the models may have as pictures-

which I hope is not negligible-the most importantconclusions (and
suggestions)of this paper may be summarizedas follows:
(1) Imperfectinformationabout employees is not in itselfenough to
explain deviations fromthe elementaryconclusions of the theoryof
the firm,that wage rates are equal to marginalproducts.
(2) Where (because of labor marketmonopsony) a firm'spay struc-
ture is devised to encourage work and the demonstrationof ability,
the marginalproduct of a more highlyskilled man exceeds his wage
by more thanfor less skilled workers,althoughthe paymentschedule
is steeper than pure monopsony theoryimplies.
(3) For thefirmas a whole,a hierarchicalstructuredoes notnecessarily
impose decreasing returnsto scale.
(4) Withinthe firm,optimal payment schedules for individual em-
ployees would normally pay nothing whenever apparent perfor-
mance falls below a certain point (this was a Pareto-optimality
result-it is in the worker's interest,given the supervisor's utility
level).
(5) The optimizingmodel used in the paper suggests some tendency
forthe givingof orders (ratherthan allowing "initiative") to be more
nearlyoptimalat lower levels in a hierarchy,and whereverthe princi-
pal is stronglyrisk averse.
References
AlCHIAN, A. A. AND DEMSETZ, H. "Production, InformationCosts and Economic
Organization." The AmetricanEconiomnic Reviewi,,Vol. 62, No. 5 (December 1972).
ARROW, K. J. Essays in the Theory of Risk-Bearing. London and Amsterdam:
North-Holland, 1972.
. The Limits of Orgalizcationl. New York: W. W. Norton,1974.
CFEUNG, S. "Private Property Rights and Share-Cropping." Joulrnlalof Political
Ecolnomyi),(1968).
MARSCHAK, J. AND RADNER,R. EconomiiicTlheoryr of Teams. New Haven:Yale Univ.
Press, 1972.
MIRR1LEES, J. A. "An Exploration in the Theory of Optimum Income Taxation."
Reviewiof Economic Stdlides(April 1971).
. "On Producer Taxation." Reivieii,of Econiomic Stdlides(January 1972a).
"Population Policy and the Taxation of Family Size." Jouarnzal of Plublic
Economifhcs(Augu1st 1972h).
. "Notes on WelfareEconomics, Infor-mation and Uncertainty"in M. S. Balch,
D. L. McFadden and S. Y. Wu, eds., Conttributionisto Economic Analysis. Oxford
and Amsterdam: North-Holland, 1974.
PAULY, M. "The Economics of Moral Hazard." TlheAmerican Econiomic Reviewl,
Vol. 58, No. 3 (June 1968).
Ross, S. A. "The Economic Theory of Agency: The Principal's Problem." Tlhe
Americail Economic Reviewt,Vol. 63, No. 2 (May 1973).
SIMON, H. A. Adlininistrative Behavior. New York: Free Press, 1957.
SPENCE,A. M. "Job MarketSignalling."QuarterlyJouirnalofEconlomics(August 1973).
AND ZECKHAUSER, R. "Income, Information,and Individual Action."
The American Economic Reviewv,Vol. 61, No. 2 (May 1971), pp. 380-387.
STIGIITZ, J. E. "Incentives and Risk Sharingin Sharecropping,"Reviewlvof Economic
Stuidies(April 1974).
WILLIAMSON, 0. E. "Hierarchical Control and Optimum Firm Size." Joulrnalof
Political Econiom-ly(April 1967). Reprintedin D. Needham, ed., Readings in the
Organization. New York: Holt, Rinehartand Winston,
Econiomticsof Incldustriail
1970.
WILSON, R. "The Theory of Syndicates." Econiomnetrica (January 1968).
ZECKFIAUSER, R. "Mutual Insurance: A Case Study of the Tradeoffbetween Risk
Spreading and Appropriate Incentives," Joulrnail of Economic Theory (March
1970). MIRRLEES / 131


Anization (A J) 1976

Uploaded by

Copyright:

Available Formats

You might also like

Anization (A J) 1976

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Anization (A J) 1976

Uploaded by

Copyright:

Available Formats

The Optimal Structure of Incentives and Authority within an Organization

Author(s): James A. Mirrlees

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

Two kinds of models for a productiveorganizationare presented. In

* The usual idea of an organizationis thatit is a group of people (or 1. Introduction

James A. Mirrleesreceived the M.A. fromthe Universityof Edinburgh(1957), the

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

* It is commonlyasserted that one reason for the hierarchicalpay 2. The pay

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

C] Technologies.Production possibilities will be assumed to take a

y = G(f 4A3(z)f(z)dz, f,y (z)f(z)dz). (4)

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

de y = fp(z) a f(z,E)dz. (5)

p(z) is, in a naturalsense, the marginalproductof a workerproviding

O Workers.Members of the labor force are characterizedby a single

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

u(n) will be the utilityavailable to the workerof type n in alternative

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

VARIATION OF THE PAY SCHEDULE

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

where z(n,E) is the work-levelchosen by an n-workerwhen faced

aE {p(z(n,E)) - w(z(n,E),E)} > 0. (17)

In order to calculate this expression, we need to know the partial

where the E-subscriptdenotes differentiation:w' E, for example,

Using this result, we have

using (18) to substitutefor Z E We want to be able to choose WE

sww' + sz - w" > 0 (a c z c b). (20)

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

a positive numberof order E. The othereffectsare of order E2. Thus,

INCREASING PROFITS WHEN W(z2) - W(zl) #(Z2 - z1 )H'

(no-- E) -INDI FFERENCE

raisingthe pay schedule slightlyabove the n0-indifferencecurvejoin-

fnO+E wqdn - {fo+E z4dn .

Now in the case of the productionfunction(1), it is easily seen that

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

LOCAL REDUCTION OF THE PAY SCHEDULE

\-i / ~~P(z) H'z

(no- E) INDI F F ER ENC

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

O Imperfectlabor markets. The question is thus raised how flat a

ii is an increasingfunctionof 4 (as well as

where z = z(n). is the value chosen by a worker of ability n.

dd u(w(z),z,n) = un(w,z,n), (28)

thepartial derivativewithrespect to n. This "envelope condition" is

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

w = w(v,z,n), Un = qi(v,z,n). (30)

Afterthese preliminaries,we can express the problem as

with respect to v(n) yields

with respect to 4 gives

This content downloaded from 85.244.145.75 on Fri, 5 Sep 2014 12:55:22 PM

Uwa ' + Unw W - i) (39)

The firm'soptimal payment schedule is definedby (39) with (33) in

meaningthata man withgreatern is more able (or willing)to substi-

, = exp (Unw/uw)dv . > 0. (43)

It follows thatforsome n, - (p - w) /w exceeds 1, while forothersit