FIRMS' CHOICE OF METHOD OF PAY

CHARLES BROWN*

Usingdata fromtheBLS Industry Wage Survey,theauthorteststhe

theorythatfirmschoosetheirmethodsof pay by balancingthe gains
from more precise links between performanceand pay against
monitoring costs.The resultsconfirm mostof thepredictionsfromthe
generaltheory.For example,largefirmsmakesignificantly greateruse
of standard-rate pay thando smallfirms,and incentivepay (such as
of dutiesthaninjobs with
piecerates)is lesslikelyinjobs witha variety
a narrowsetof routines.

THE importanceof differencesin meth- cally the cost of monitoring workers in

ods of pay is not a new topic,' but it is
one that has received increasingattention
fromacademic researchersand compensa-
tion specialists. Its claim to academic
attentionrestslargelyon the hypothesized
relationship between methods of wage
payment and the cost of monitoring
workers in differentcontexts-a theme
that, in turn, lies at the heart of recent
theoreticalinsightsabout the structureof
wages. Improvementsin technology,par-
ticularlythe growinguse of computers in
the office and the factory,and even in
long-haul trucking,are reducing dramati-
* Charles Brown is Professorof Economicsand Re-
search Scientistat the Institutefor Social Research at
the Universityof Michigan,and a Research Associate
of the National Bureau of Economic Research. This
researchwas supportedbythe National Science Foun-
dation (Grant 8708370). The author thanks Nancy
Lemrow, Thomas Hungerford, and Marsha Silver-
berg for researchassistance.For helpfulsuggestions,
he also thanksseminarparticipantsat the NBER and
these universities:Chicago, Cornell, Georgetown,Il-
linois, Kentucky,Michigan,and Toronto.
Copies of the data and computerprogramsused to
generate the resultspresentedin thispaper are avail-
able from the author at the Institutefor Social Re-
search,University of Michigan,Ann Arbor,MI 48109.
1Julius Caesar is said to have enhanced the
some industries.2Compensation specialists
have noted a decline in the link between
pay and performance,and many predict
the crisis of competitivenessfacing U.S.
firmswill force a re-strengthening of this
link for both managerial and nonmanage-
rial employees (McLaughlin 1986; Morse
1986).3 Meanwhile, on the other side of
the Pacific,"there is startingto be a shift
away from pay by seniorityand towards
pay determinedby merit"in Japan as well
(Maidment 1987:S-18).
But problemsin implementingcontracts
linking pay and performance run into
monitoringproblems that are stressed in
the academic literature.One example of
general interest is the contract of New
York Jets quarterback Ken O'Brien,
2 The Washington Post reported that a large hotel
now requires each maid to punch her ID number on
the room telephone when she entersand leaves each
room. Similar opportunitiesare available for moni-
toring typistsusing word processing packages that
can record the typist'swork level while performing
traditional word processing. For evidence of the
potential of modern technology for more precise
monitoringof workers,see National Research Coun-
cil (1986) and Shaiken (1987).
3 A survey of compensation committee members

effectivenessof the Roman army by implementinga of 350 large U.S. corporations found this was "the
performance-basedsalary system in lieu of booty most important objective in the next five years"
(McLaughlin 1986:8). (McLaughlin 1986:7).

Industrialand LaborRelationsReview,Vol. 43, Special Issue (February 1990). X by Cornell University.

0019-7939/90/43SP$01.00

165-S
Sage Publications, Inc.
is collaborating with JSTOR to digitize, preserve, and extend access to
Industrial and Labor Relations Review
Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015 ®

www.jstor.org
166-S
under whichpart of O'Brien's pay is keyed
to his NFL quarterbacks' rating, which
penalizes incomplete passes but not sacks.
Said formerJetsquarterbackJoe Namath,
"I'm amazed at [O'Brien's] accuracy,but I
see him hold the ball more than he should.
I always thought it was better to have
second and 10 than second and 18. I don't
like incentive contracts that pertain to
numbers" (Anderson 1988).
The existingtheoryof the determinants
of method of pay is fairlystraightforward:
in essence, it asserts that differentmeth-
ods of pay offerdifferentapproximations
to an idealized pay-for-performance incen-
tive for workers,but carrydifferentcosts.
In general, the link between pay and
performancewill be strongerwhere per-
formanceis more accuratelyobserved (for
example, where there are piece rates),but
such accuracy is likely to be expensive.
Hence, across firmsor jobs withina firm,
variationsin costs and benefitsof accurate
monitoring are hypothesized to explain
differencesin method of pay.
This theory receives considerable sup-
port in the prescriptive treatments in
personnel texts (though the topic appears
to command less attention in such texts
now than it once did) and in a quite
limited set of empirical analyses. Perhaps
little empirical work in the spirit of the
new personnel economics has been de-
voted to these issues because data in
household surveysand mostestablishment
surveysare not veryhelpful.The purpose
of this paper is to help fillthis gap.
ExistingTheoretical Models
A recent paper by Lazear (1986), which
presents several models of the choice of
method of pay,4 provides a convenient
4 An alternativeapproach would be to emphasize
the somewhatdifferentprincipal-agentliterature,as
developed by Harris and Raviv (1979), Holmstrom
(1979 and 1982), or Green and Stokey (1983). A
fundamentaldifferencebetween these papers and
Lazear's is the assumed source of randomness.
Lazear assumes that unless the cost of an accurate
piece rate systemis undertaken, the firmdoes not
know how much the worker has produced. The
principal-agentpapers, on the other hand, assume
Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
CHARLES BROWN
point of departure. Existingmodels gener-
ally assume the firm has two choices:
paying piece rates (earnings tied to an
objective measure of value of output) or
paying time rates or salaries (earnings not
linked to how much is produced). To
simplify,I will assume the work-weekis
fixed, which means that earnings per
period are fixed under the lattersystem.
In the simplestmodel of this sort,each
worker'sproductivityis fixed (effortis not
an issue),and themethodsof pay sortwork-
ers among firms.Workersknow theirown
productivity, but firmsdo not, unless they
pay a monitoringcost 0. The firmcan ei-
ther pay 0 to measure a worker'sproduc-
tivityq directly,in which case it pays the
worker q - 0, or not pay the cost of the
piece-ratemonitoringsystemand pay sal-
ary S. The piece rate structureobviously
satisfiesthe zero-profitrequirement.Firms
using salaries mustin equilibriumchoose S
so they,too, earn zero profits.Since work-
ers knowtheirown q's,theychoose the type
of firmin whichtheirearningsare highest:
workers with q>S+0 choose piece-rate
firmsand those with q<S+0 choose the
firmspaying salaries. (See Figure 1.) If q
has densityand c.d.f. f(q) and F(q) respec-
tively,the zero-profitcondition is
( fS +0
S= qf(q)dq
(1) F(S + 0) Jo
Lazear shows that an S>O satisfyingthis
conditionexists so long as 0>0. Thus, the
that firms do observe each worker's output (an
exception is part of Holmstrom's 1982 paper) but
this output depends on environmentalfactors the
worker cannot control as well as on effort,so that
paying piece rates introducesrisk,whichagents find
undesirable. The extra informationsought by the
firmis some measure of effort,so that the worker
can be partiallyinsured in cases where output is low
due to environmental randomness rather than
worker shirking. This approach is very likely the
right one for thinking about corporate CEOs or
salespersons,forwhom output is easilymeasured but
environmental randomness is important. Lazear's
approach seems better suited to the blue-collar
workers in the IWS, for whom environmental
randomness is relatively unimportant (the most
importantrandomness, breakdown of machines or
material unavailability,being easily measured) but
measuringoutput (performance) is costly.
 FIRMS' CHOICE OF METHOD OF PAY 167-S

salaryfirmssave on monitoringcosts and must assume that there is some minimum

pay lowerwages, but these advantages are level of effortE that firmscan observe
exactly counterbalanced by their having cheaply;5workerswho do not work at this
less productiveworkers. level of effortare discharged by time-rate
The most obvious prediction of this firms.
simple model is thatwithinan occupation, A worker who puts forth effort E
workerspaid piece rates will have higher produces E units of output, but at psychic
earnings than workers paid time rates. cost C(E). The piece rate firmpays a wage
Available evidence supports this predic- equal to E -0 and, if the worker works
tion (King 1975; Pencavel 1977; Seiler under a piece-ratesystem,his or her utility
1984). is E -0- C(E). Utilityis maximized at E*,
A second predictionis that as the costs where C'(E*)= 1, and utilityis U*. Alter-
of directly monitoring output increase, natively, if the individual works in a
fewer workers will work in piece-rate time-rate firm, an effort level E is re-
firms.This patternis clear fromFigure 1. quired, and a salaryS is paid. The worker
An increase in 0 shifts down the w(q) achieves utility U = -QC(E) =E- E),
functionfor piece rate firms.The initial since S must equal E to satisfy the
effect of this change is to increase the zero-profitcondition. The worker com-
number of workers receiving time rates, pares U* to U and chooses the firm
and to allow time-rate firms to earn offeringthe greater utility.
positiveprofits(since the additional work- To strengthenthe comparison with the
ers have qo'>q>qo but theyare paid S, the earlier model, assume that workersdiffer
average of q from zero to qo). This effect in their cost-of-effortfunctions. Let
in turn raises the wages offered by C(E) = E2/N, where N represents the en-
time-ratefirms, furtherincreasing their ergy (or ability) of the worker: workers
share of the work force. withhigherN produce a given (increment
The above model emphasizes sortingof to) E at lower (marginal) cost. Then
workerswith differentabilities. An alter- U=E-E21N is the utility offered by
native is to emphasize workereffort.The time-rate firms. Under piece rates, the
key distinctionbetween abilityand effort workerchooses E* = N12 and attainsutility
is that the latter is controlled by the level U* = (N/4)- 0. Workers are sorted
worker, and can be manipulated by the across firmsby theirvalue of N, as shown
firm by offering appropriate incentives. in Figure2.6
Because workers are assumed to find
greater effort distasteful, however, we

5 If there were no minimum level of effortthe

ZIV
w(q) firmcould easily verify,workerswould presumably
setE equal to zero and the firmcould not survive.An
alternativeapproach is to assume there is a minimal
level of effortE >0 below which less effortdoes not
raise utility.
6 Figure 2 is drawn on the assumptionthatE is low
enough relativeto the minimumvalue of N that all
workerswithlow N work for the salarL firm.As the
S above algebra makes clear, U*(N) and U (N) willcross
twice.Call the two values No and No. For N>NO, the
workerprefersthe piece rate firmand exerts E>E.
For N?<N<No, the worker prefers time rates. If
thereare any workerswithN<NJ, however,theywill
q-0 prefer piece rates because piece rates yermit a
leisurelypace of workbelow E. Given thatE typically
qo q corresponds to a very low level of effort(just going
*- Salaried - - Piece Rate throughthe motions) and that previous studies find
piece rate workers earning more, this possibility
Figure 1. Salary and Piece Rate Firms. seems unlikelyto be importantempirically.

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
168-S CHARLES BROWN

U
will be clearer when the data are de-
U(N) scribed.)
Suppose there are three intensitiesof
monitoring,and for any monitoringstrat-
egyj the expected wage offereda worker
of qualityq is a linear functionof q:

(2) w, = aj, + bq.

- 2N
The statistical discrimination literature
(Aigner and Cain 1977; Lundberg and
Startz 1983; Garen 1985) shows that the
NG
relationship between expected wage and
*- Salaried Piece Rate -
worker quality will be steeper (that is, b1
- -

Figure 2. Salary and Piece Rate Firms. will be larger) the more accurate the
available productivity indicator. If all
three monitoringintensitiesare used, and
The implicationsof this model, appar- bj<b2<b3, it must be thataj>a2>a3. Since
ent in Figure 2, are similarto those of the the cost of monitoringworkersis presum-
previous model: piece-rate workers earn ablylower ifthe monitoringstrategyis less
higher wages, and an increase in the cost accurate,and the cost O0is subtractedfrom
of running a piece-rate systemincreases a worker's output in setting the wage,
the fractionof workers(range of N values) differencesin 0 contributeto this ranking
who prefertime rates. of the a's.
As in Lazear's model, assume that piece
Three-Choice Models rates correspond to a precise but expen-
sive measuring of physical output. Merit
pay offers a less expensive but less
The assumption that pay of time-rated
accurate alternative.8Part of the weaker
workers is unrelated to productivityis
link between pay and performanceunder
overly restrictive-at least for some time-
rated blue-collar workers and a clear
majorityof clerical and technicalworkers
(Cox 1971; BNA 1981; Personnick 1987). 8 One
mightmake two objectionsto thisclassifica-
Withinthe general categoryof time-rated tion. First, it might be argued that, in some cases,
workers,some receive wages that depend piece rates are less accurate than supervisorratings
as an indicator of productivity(for example, where
on job categoryand perhaps senioritybut qualityof output is veryimportant).Such a situation,
not performance,whereas others' wages however, consistentwith the frameworkdeveloped
are set individuallybased on supervisors' here, can be described as one in which the cost of a
perceptionsof theirproductivity. A gener- precise piece-rate systemis very high. Second, one
alization that allows the firm to choose might wonder whether merit ratings are very
expensive, given that supervisorshave a reasonably
among three wage-settingmethods-one accurate estimateof workers'productivity thatcomes
in whichwages depend only marginallyon "free" from the act of supervising. A merit pay
performance ("standard rates"), one in systemin whichconsiderable weightis placed on the
whichtheydepend on supervisors'evalua- evaluations, however (see next paragraph). imposes
not only the cost of the supervisors writingdown
tions ("merit pay"),7 and a piece rate what theyalready know; morale considerationsseem
system-is developed in this section. (The to demand that a serious merit-pay system be
empirical motivation for this way of formalizedso that workerswill accept it as fair, and
definingthe boundaries between methods coordinated so that otherwiseidenticalworkerswith
differentsupervisorsare not treatedverydifferently.
Finally,the personnel literaturesuggests it is diffi-
7 Merit pay may take the form of "contests" in cult-perplexingly so to an economist-for top
whicheach worker'sratingdepends on his measured managementto enforcesizable meritdifferentials for
performancerelativeto everyoneelse's, as described workersat lower levels. See Hamner (1983); Strauss
in Nalebuffand Stiglitz(1983). and Sayles (1980).

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
 FIRMS' CHOICE OF METHOD OF PAY 169-S

w
monitoring intensityor the benefits of
W(q) betterperformanceindicatorschange, the
use of the threemonitoringintensitiesand
wM(q)
their associated methods of pay will also
change. An increase in Oj leads wj(q) to
shift downward, and the number of
ws(q)
workerswho opt for that method falls.
The benefits of better performance
indicators are presumably larger where
Wp(q) differencesin worker productivenessare
larger. In the limitingcase in which all
workers are equally productive,expendi-
q1 q2
q tures to assess their performance would
qo q3a
Standard Merit Piece
-> <_ <.
not be profitable.
Rates Pay Rates
To be more precise, suppose wage
Figure 3. Standard Rate, Merit Pay, and Piece offersare
Rate Firms.

(3) wj = aj + bjvq
merit pay comes from errors in supervi-
sors' ratingsof performance;the peculiar
where v depends on the sensitivityof
feature of most merit-pay systems that output to worker quality. Let v initially
incrementsfeed intothebase wage contrib-
equal one, as in Figure 3. Now let v
utes as well (Schwab and Olson, thisissue). increase. The boundary between standard
Firms using standard rates presumably
rate and meritpay workers,qj, is defined
spend little on measuring performance,
by
since theydo not use it in wage-setting.9
Let S, M, and P denote standard rates,
merit pay, and piece rates, respectively. -as - am
(4)
Figure 3 shows a three-methodgeneraliza- (bM- b)v
tion of Figure 1. Although bs may be
greaterthan zero and bp may be less than
one, the essential differencehere is that An increase in v thus reduces qi (and q2),
merit pay is added as an intermediate reducing the use of standard rates and
method of pay. increasingthe use of piece rates.'0
If the costs of using a particular How does the three-alternativemodel
work when workers vary their effortin
9 My focus is on the relationshipbetween perfor-
response to wage incentives? A worker
mance and pay in the shortrun. Even if it is granted
who exerts effortE produces E worth of
that those who performbetterunder standard-rate output and earns an expected wage of a1
regimes have a greater probabilityof promotion in + bjE if he or she worksunder method of
the future,the relationshipbetween pay and perfor- pay j. A worker with energy level N has
mance would stillbe "less than that associated with utilityU- = a1 + b E - E2IN. To maximize
the other forms of compensation that are more
closelytied to currentoutput" (Barron and Loewen- utility,the worker chooses effortlevel E*
stein 1986:604). Future compensationis neglected in = b1N/2,earns W. = aj + bVN/2,and
the textbecause it is not measured in the IWS. attainsutilitylevel U* = aj + b/N/4.If we
It is worth noting that future promotions would now plot UA(N),we have a diagram like
probably depend on measured performanceunder
any of the systems,so neglecting them may be a
Figure 3, except that the axes are U* and
defensible simplification when the object is to N rather than wj and q. Recognizing that
compare systems. Also, to the extent that less
accurate monitoringleaves standard-ratefirmsless
able to identifythe "right" workers to promote to
10I am assuming that these "first-round"effects
higher positions,there is presumablya productivity
loss that should be counted as part of the cost of are not completelyundone by subsequent changes in
using thissystem. response to positiveprofits.

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
170-S
the workerchooses the method of pay that
offersthe highest lU, we again have the
workers sorting according to N, and the
merit-pay workers receiving wages be-
tween the low-wagestandard-rateworkers
and the high-wagepiece-rateworkers.
Thus far,I have takenthe supplyof labor
to the occupationor industryin questionas
given.Althoughitis clearthatthisis a strong
assumption,it is not clear how best to relax
it. One possibilityis to assume thatworkers
face an alternativewage w, independentof
q. This assumptionis reasonable if workers
do notknowtheirabilities," I butwhenwork-
ers do know theirabilitiesit is equivalentto
assuming that abilityin the occupation in
question is uncorrelatedwith abilityin al-
ternativeoccupations.
A less restrictiveoption is to assume that
there is an alternativewage functionWa(q)
thatmustbe met ifworkersof qualityq are
to be attracted.Although the details de-
pend on the shape and positionof the Wa(q)
one chooses, the basic impact of including
wa is to eliminatepart of the q distribution
from Figure 3. For example, in Figure 4
workerswith qo<q<ql stillwork for stan-
dard rate firms,those withql<q<q4 work
formerit-pay firms, thosewithq4<q<q5 ac-
ceptalternativeemployment, and onlythose
withq>q5 work for piece-ratefirms.It re-
mains truethata smallincreasein Op(shift-
ing wp(q)downward)reduces the incidence
of piece rates.The differenceis thathere a
small increase in Op has no effecton the
number of merit-payor standard-ratework-
ers, and theirshareof employmentin this
marketriseswhentheworkerswithq slightly
above q5 leave for alternativeemployment.
Measurable Determinantsof
Method of Pay
The extension of the standard model
leads to the predictionthatthe prevalence
of each of the threemethodsof pay should
be inverselyrelated to the cost of using it,
and that piece rates should be more com-
1 Lazear (1986) introduces w when workers do
not know their own abilities, but reverts to the
fixed-supplymodel when the workerknowsq.
Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
CHARLES BROWN
w
wVq(q
--I ~ ~~~~~~~~~~~~
I
' I ~~~~~~~~~I
i
I I I I
I ,,...I |.,
. I
q0 qI q4 q2 q5 q3
Standard Merit Work Elsewhere Piece
Rates Pay Rates
Figure 4. Standard Rate, Merit Pay, and Piece
Rate Fi rms.
mon and standard rates less common where
the differences in productivityamong work-
ers are greatest. Factors related to these
determinants (and how they can be mea-
sured in the data described below) are sum-
marized in Table 1. These factors include
those mentioned in the method-of-pay lit-
erature and in personnel texts.
Perhaps the most commonly cited indi-
cator of monitoring costs in the economics
literature is the size of establishment
or firm.
The classic reference here is Stigler (I 962),
who argued that large employers have a
significant disadvantage in monitoring
workers. (More recent studies using this
premise are Oi 1983 and Garenl 1985.)
A closer look at the Stigler argument
makes it clear that the greater monitoring
costs correspond to an increase in Om -
greater difficultyin monitoring through su-
pervisors' ratings -rather than an increase
in Op. Indeed, since piece-rate systems en-
tail fixed costs that can be spread over more
workers in larger establishments (Cleland
1955; ILO 1984), Op should be lower in
such workplaces. This consideration dem-
onstrates the practical importance of the
three-way choice of method of pay. We ex-
pect larger establishments to make greater
use of piece rates, less use of merit pay,
and perhaps greater use of standard rates.
The argument that larger establish-
ments have an advantage in implementing
piece rate systems has a subtler implication
 FIRMS' CHOICE OF METHOD OF PAY 171-S

Table 1. Impacton Methodof Pay,and Data.

Variables,Hypothesized
Hypothesized b
Variable Impacta Data
Size of Establishment T P, K M Estab. Size (IWS)
Occupational Concentration TP X (share of occ k)2 (IWS)
k
DiversifiedDuties ,P DOT Variables
Skill (Variation in vmp) mS, T P DOT Variables
Accuracy& Quality Important tP DOT Variables
but Hard to Measure
Growthin Q/L ,P A In VA/L (COM)
Capital Intensity , P?, I M? VA-Payroll (COM)
Shipments
Importanceof Teamwork IP Holidays (IWS)
Total Time Off
ProportionFemale S5?, T P Female Empl. Share (IWS)
Union Coverage (U) T S, M,?P WhetherUnionized (IWS)
Threat of Unionization T S, M,?P (1-U) U-hat (IWS)
a p = piece rates; M = merit pay; S = standard rates.
b IWS = IndustryWage Survey; DOT = Dictionary ofOccupationalTitles;COM = Departmentof Commerce,
CountyBusinessPatternsfor 1977. VA = value added; L = number of workers;U-hat = predictedvalue of U.

that has not been discussed in the litera- An interrelatedset of job characteris-
ture. The more separate jobs within an tics-skill level and the importance of
establishmentof given size, the greaterthe accuracyand qualityof output-have con-
cost of settingup and maintainingpiece flictingeffectson method of pay. Almost
rates. Greater occupationaldispersion raises by definition,high-skilljobs are jobs in
Op and should, other things equal, be which workeroutput is sensitiveto differ-
associated with less use of piece rates. It ences in worker quality. Thus, high-skill
may also increase the cost of a given level jobs should have greater benefit from
of accuracyfor meritpay ratingsin which precise monitoring, and greater use of
each worker'sperformanceis compared to piece rates and less use of standard rates
the group average, because the group (Beach 1975:670). On the other hand,
average would be based on fewerobserva- when accuracy and quality of work are
tions and contain more noise.
important-characteristicsthat are often
A relatedthemeis thatestablishments with
but not necessarily associated with skill
long,standardizedproductionrunsor those
in whichindividualemployeesperform the level-Op is likelyto be high and the use of
same tasksrepetitively are amenable to piece piece rates less common (Cornell 1936:
rates.Conversely,where (due to shortpro- 537; Lansburgh and Spriegel 1940:416;
duction runs or the nature of the jobs in- Pencavel 1977:232). More precisely,since
volved) individual workers perform a wide piece rate systems can penalize workers
range of duties and there is considerable for defective pieces, situations in which
day-to-dayvariationin the importanceof accuracyand qualityare importantbut not
various duties,it is difficult(that is, costly) easily verified pose an obstacle to using
to devise a piece-ratesystemthat correctly piece rates.
priceseach of the tasks(Cleland 1955; Cor- A frequently mentioned disadvantage
nell 1936; Carlson 1982).12 of a piece rate systemis thatrapidgrowth of
output per workerdue to technological
12
change or increased capital intensityneces-
Cleland (1955) and Oi (1983) suggest that long
standardizedproductionruns are more characteristic
sitatesrevisionsin the piece rate structure,
of large than small establishments,reinforcingthe raising the cost of piece rates. If the piece
predicted effectof establishmentsize. rate is adjusted downward,morale is likely

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
172-S
to suffer, particularly if the workers
believe that management used piece rates
to get them to work hard and then
adjusted the rate downward. In principle,
adjustments to reflect the fact that new
machineryhas made the old piece rate too
generous mightbe accepted, but in prac-
tice,determiningwhetherthe reductionis
fair is difficult.Lansburgh and Spriegel
(1940:419)'3 stated this problem quite
clearly:
Reallyradicalchangesin productionmethod,
whichso changethejob as to make the past
rateabsurd,havebeenfrequentlylookedupon
by workersmerelyas an excusefor cuttingrates.
... This confusionbetweenlogicalpiece rate
re-adjustments resultsin nu-
and rate-cutting
merousborderlinecaseswhichit is difficult
to
becausethereare no real data,
settleamicably,
convincingto bothsides,whichmaybe used as
a basis.
Piece rates as understood in the theoret-
ical literaturelink individual performance
and pay. Consequently,where teamproduc-
tion is important, measuring individual
performanceis more difficultand the cost
of an accurate piece-rate system will be
greater.Firmswillavoid piece rates "if the
cost of determining how much each
individualemployee has produced at each
stage in a production process is excessive"
(Beach 1975:681).
The relationship between capital intensity
and method of pay is complicated by the
fact that the literaturehas made use of
capital intensityas an indicatorof several
forces. Lazear (1986) shows that an in-
crease in the required value of capital per
worker is like a simultaneous increase in
all 0's-the cost of capital per worker is
subtractedfromthe output of each worker
under each payment system,and the use
of particular payment schemes is not
affected.Others have used capital inten-
sityas a proxyfor machine-paced produc-
tion (for example, McKersie et al. 1964
and Carlson 1982; but see Pryor 1984:41
for evidence challenging this link), which
reduces the attractivenessof piece rates
13 See also Cornell (1936:711); Pencavel (1977:
233); and Carlson (1982:20).
Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
CHARLES BROWN
(Cornell 1936:537; Cox 1971:55) and
increases the use of standard rates (Beach
1975:680). Alternatively,greater capital
intensity may be interpreted as more
machineryentrustedto the worker. Since
maintenance of machineryis, like quality,
something that may sufferunless special
measures are undertaken in a piece rate
system, I expect less use of piece rates
where capital intensityis greater. Finally,
capital intensity may have an indirect
effectby changing the skill level of the
work. To the extent that the occupation
mix is affected, the skill measures of
individual occupations would capture this
effect.To the extentthatthe skillrequired
withinoccupational titleschanges, a capi-
tal intensitymeasure will pick up that
effect as well. On balance, previous re-
search suggeststhatcapital intensityprob-
ably should reduce use of piece rates,but
this predictiondepends on hard-to-verify
links between capital intensityand more
basic determinantsof method of pay, and
in having controlled adequately for the
skilllevel of the work involved.
Goldin (1986) has argued that the
proportion of women in the firm'swork
force may have an importantbearing on
the method of pay. Her analysiscompares
a piece rate systemto a timerate system,in
which better performance is rewarded
through futurepromotion,and less care-
ful (less expensive) monitoring is em-
ployed. She argues that the latter system
can be efficientfor those planning stable
attachmentto the firm.Piece rates would
be more common for female work forces,
however, because the promise of future
promotion is too bland a carrot for those
planning short tenure with the firm.
Goldin found that women were indeed
more likely to be piece-rate workers,
whereas men tended to be paid piece rates
in integratedestablishmentsbut timerates
in segregated male workplaces.14
A three-waymethod of pay categoriza-
tion allows us to put Goldin's hypothesisto
14
Pencavel (1977) found only a very weak
tendency for women to be piece-rate workers,in a
relatively small sample of Chicago punch press
operators.
 FIRMS' CHOICE OF METHOD OF PAY
a more precise test. Not only should
women be more likely to be paid piece
rates, theyshould be less likelyto be paid
standard rates,since presumablythe rela-
tiveimportanceof promotionas an incen-
tive is greatest where few within-
occupation distinctions are made. An
alternative view is that women avoid
systems where supervisors have great
discretion(namely,meritpay systems)and
concentrate in systems with the formal
protectionofferedby eitherstandard rates
or piece rates.Both models predictwomen
are more likelyto work under piece rates
(as Goldin demonstrated); they differ in
their prediction about the distributionof
women between merit pay and standard
rate systems (which Goldin's data could
not distinguish).
Thus far, the discussion has implicitly
focused on nonunion establishments,or
situationsin whichthe presence of a union
makes no differencefor method of pay.
Freeman (1982:4-5) has argued, however,
that unions opt for standard-rate wage
systems,both as protection against arbi-
trariness by low-level supervisors and
because eliminating wage disparities
strengthensa political sense of solidarity.
Although Freeman's analysis focuses on
the choice between standard-rate and
merit-paysystems,a more explicitconsid-
eration of piece rates would be illuminat-
ing. Piece rates provide protectionagainst
supervisorydiscretion'5but not solidarity-
enhancing wage equality. Thus, the rela-
tionshipsbetween unionization and piece
rates allows us to distinguish between
these motivations. The distinction has
potentiallyimportantpolicy implications.
In an environmentwhere heavily union-
ized industries need improvements in
productivityto remain competitive,tech-
nological monitoringcould provide incen-
tivesforbetterperformancewithoutsacri-
ficing protection from arbitrarysupervi-
sory discretion.
15
Under an imperfectpiece rate system,supervi-
sors can still assign favored employees to jobs with
"loose" rates-that is,jobs thatare overpriced(Beach
1975:699). The union would need to restrictsuch
in assignment.
flexibility
Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
173-S
Finally,the historyof firms'response to
the threatof unionizationsuggests another
union-related hypothesis. Jacoby (1984)
found that nonunion firms threatened
with unionization often adopted more
formal, standardized wage structuresto
fend off unionization. This finding sug-
gests that merit-paysystemsmay be less
common among nonunion firms threat-
ened by unionization than among non-
union firmsfacingless serious threats.
Following the formal models on the
topic, Table I emphasizes differencesin
"objective" conditions in the workplace.
An alternative, more elusive factor is
differences in management philosophy
(Lewis 1960:462; Carlson 1982:20): some
managers believe it is essential to reward
productive employees, whereas others
worry more about the harm done by
distinctions that are perceived to be
unjustified.Although there is no precise
measure of managerial philosophy, it is
possible to extend an idea recentlyintro-
duced by Dickens and Katz (1986) in a
related context.Given thatofficeworkhas
no obvious technologicalsimilarityto plant
work across manufacturing establish-
ments,it would be useful to know whether
thereis an importantrelationshipbetween
methods of pay of office and plant
workersin an establishment.' A finding,
for example, that establishments using
standard rates for officeworkershave no
particular tendencyto use standard rates
for their plant workers would tend to
support the importance of factors like
those in Table 1 rather than managerial
stylein determiningmethods of pay.'7
Data
The IndustryWage Survey (IWS) pro-
gram of the Bureau of Labor Statistics
provides data on methods of wage deter-
minationand other establishmentcharac-
16
Of course, officeworkersare almost never paid
piece rates. But there is variationin the importance
given to individual differences in productivity,
particularlyin the importance of seniorityversus
meritin range-of-ratesystems.
17 Methods of pay of officeworkersare available
for only a minorityof the IWS industries.
174-S CHARLES BROWN
teristics.The Survey itselfcollects,froma
sample of establishmentsin approximately
60 industries (four-fifthsof them in
manufacturing),data on seven categories
of establishmentcharacteristics:propor-
tion of production (and, in a few indus-
tries, office) workers paid by various
"methods of pay"; fringe benefit policies
(for example, number of holidays and
vacation days at differentsenioritylevels);
number of workers employed; union
coverage; region; metropolitan/non-
metropolitan location; and details about
the product and production technology,
which vary fromindustryto industry.
Ten methods of pay are distinguished,
five for time rates and five for incentive
pay:18
Timerates.(1) In a single rate system,all
those in a job category receive the same
wage. When a range of rates is used for a
job category, progression through the
range is governed by (2) merit, (3)
seniority,or (4) a combination of merit
and seniority.Finally, (5) wages may be
individuallydetermined.
Incentivepay. Incentives may take the
form of (1) individual piece rates, (2)
individualbonus pay (pay for exceeding a
production quota), (3) group piece rates,
(4) group bonus pay, or (5) commissions.
18
A more precise definitionof the IWS categories
is provided in each report: Formal rate structuresfor
time-ratedworkersprovide single rates or a range of
ratesforindividualjob categories.In the absence of a
formal rate structure, pay rates are determined
primarily by the qualifications of the individual
worker. A single rate structureis one in which the
same rate is paid to all experienced workersin the
same job classification. (Learners, apprentices, or
probationaryworkersmay be paid according to rate
schedules that startbelow the single rate and permit
the workersto achieve the fulljob rate over a period
of time.)An experienced workeroccasionallymay be
paid above or below the single rate for special
reasons, but such paymentsare exceptions. Range-
of-rate plans are those that specify minimum or
maximumrates,or both,for experienced workersin
the same job. Specific rates of individual workers
withinthe range maybe determinedby merit,length
of service, or a combination of the two. Incentive
workers are classified under piecework or bonus
plans. Pieceworkis work for which a predetermined
rate is paid for each unit of output. Production
bonuses are based on productionover a quota or for
completionof a task in less than standard time.
Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
The "single rate" and "range of rates-
seniority" categories are standard-rate
methods. The "range of rates-merit"and
"individual determination"categories are
merit-pay methods.19 "Individual piece
rates,""individual bonus pay," and (negli-
gible in these data) "commissions"corre-
spond closely to the piece rate systemas
characterized by the theory described
above. "Range of rates-combination"strad-
dles the boundary between meritpay and
standard rates. "Group piece rates" and
"group bonus pay" may act like individual
piece rates if groups are small, but Weiss
(1987) offersstrongevidence thatin large
groups these incentiveeffectsmay be lost.
Unfortunately,the data do not include the
group size forthose paid group incentives;
but these categories are fairlyrare in the
10 industriesstudied here.
In addition,foreach productionworker
in "studied occupations,"a subset of occu-
pations that includes the major occupa-
tions in the industry,the IndustryWage
Survey provides the hourly wage;20 gen-
der; whetherpaid byincentiveor timerates;
and occupation(roughly30 occupationsare
distinguishedfor each industry).21
Table 2 shows the distributionof estab-
lishmentsby method of pay22 for blue-
collar workers in 10 industries.23Each
'9 This usage differsslightlyfromthatin compen-
sation textbooks,where merit pay means range of
rate systemsin whicha worker'spositionin the range
depends on merit reviews (and perhaps seniority),
and thus usually would not include a less formal
"individual determination" system. For a more
detailed descriptionof what is meant by meritpay in
thatcontext,see Schwab and Olson, thisissue.
20 The hourly wage includes piece rates and
production bonuses but excludes annual non-
productionbonuses and premium pay for overtime,
holidays,and shiftwork.
21 A sense of the fineness of detail of the IWS's
breakdown of occupations can be gained from a
sampling of occupations it lists for the wood
household furnitureindustry:router operators (dis-
tinguishedby whethertheydo set up work or not),
rip saw operators, furnituresanders (3 types), and
furniturepackers.
22 Establishments thatused more than one method
are assigned to the three methods in Table 1 in
proportion to the fractionof workers paid by each
method.
23 These are the same industries analyzed by
Freeman (1982), withthe addition of men's and boy's
 FIRMS' CHOICE OF METHOD OF PAY 175-S

Table 2. Method of Pay of Establishments, by Industry.

% ofEstablishments
Using:a
Numberof Standard Merit Piece % Coveredby
Industry Establishments Rates Pay Rates Unions
NonferrousFoundries 364 38 47 15 49
Paints 292 58 42 0 62
Textile Dyeing & Finishing 149 72 20 7 49
IndustrialChemicals 270 79 20 1 74
Cotton Textiles 342 66 9 26 16
Wool Textiles 57 61 17 22 33
Shirts 220 3 23 75 28
Plastics 876 43 52 5 45
Household Furniture 331 18 63 19 41
Steel 332 54 42 4 71
a Apart fromrounding error,these three columns sum to 100%.

Source:IndustryWage Survey.

industrymakes serious use of two or three uniformlydistributedover N occupations,

methods,even thoughindividualestablish-
ments typicallyuse one or two. There is,
therefore, considerable within-industry
variation to explain. Although union sta-
tus is related to method of pay-in
particular, unionized establishmentsare
more likely to use standard rates (Free-
man 1982)-the variationin Table 2 is not
just a reflectionof differencesin collective
bargaining coverage. The distributionof
piece-rate shares in Table 2 is similar to
that in the broader set of IWS industries
(Seiler 1984:365) except that industriesin
which 60% of more of the workers are
paid piece rates are underrepresented.
The IWS data alone allow the measure-
ment of several of the determinantsof
method of pay listed in Table 1. Size of
establishmentis coded in eight employ-
ment classes. These were converted to a
continuous variable by assigning to each
categorythe mean establishmentsize (for
the establishment's4-digitindustry)using
data fromCountyBusinessPatternsfor 1977
(U.S. Department of Commerce 1979,
Table 1B). The measure of occupational
concentrationused here is the sum (over
IWS occupations) of the squared share of
the occupation in the establishment'sem-
ployment(based on workersin the subset
of "studied" occupations). If all workers
workin a singleoccupation,the concentra-
tion index equals one; when workers are
shirts (which Freeman deleted because he was
focusingon non-piece-ratepay).
it is 1/Nor close to zero.
A measure of the importance of team-
work thatis admittedlycrude and indirect
can be constructedfromthe IWS data on
holidays and vacations,if we assume that
"teamwork"implies a need to coordinate
the work schedules of team members (in
otherwords,a cost to having some but not
all members of the team present on any
given workday). Thus, where teamworkis
important,we should expect time off to
take the form of coordinated leisure
(holidays)ratherthan uncoordinatedvaca-
tions. The share of holidays in total time
off (holidays plus vacation days) is there-
fore an indirect measure of the impor-
tance of teamwork.24Unfortunately,this
variablewas strikingly unrelatedto method
of pay and, given the indirectnessof its
link to teamworkin the firstplace, it was
deleted fromthe final regressions.
The female share of employmentin the
studied occupations can be computed
directly from the IWS worker records.
Moreover,when method of pay of individ-
ual workers is considered, a dummy
variable for female workers is based
directlyon the worker'sIWS record.
The IWS also determines whether a
majorityof the production workersin an
establishment are covered by collective
bargaining. The measure of threat of
24
The IWS vacation data are presentedseparately
at differentlevels of seniority.I convertthese data to
an overall average using distributionsof workersby
tenure by industryfromSekscenski(1980).

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
176-S CHARLES BROWN

unionization used here is based on the requiring the precise attainment of set
predicted probabilityof unionization, U- limits, tolerances, or standards"-which
hat, from a regression of a union-status seems to be a natural indicator of the
dummyon the other exogenous variables. importance of accuracy and quality. The
Clearly,the threatof unionizationapplies precision described in this variable, how-
only to nonunion firms,so this variable ever, may well be an easilychecked sort,so
equals U-hat for nonunion establishments penalties for substandard pieces (and
and zero for establishments that are hence piece rates) would be feasible after
already unionized. all. Alternative DOT variables indicate
Because IWS gives the 4-digit SIC whether the job requires aptitude for
industryand employmentsize class of its making "generalizations, evaluations or
establishments,the establishmentscan be decisions" based on "sensoryor judgmen-
matchedto Census of Manufacturers'data tal criteria" or based on "measurable or
by industry-by-sizecell. Specifically,the verifiablecriteria."The sum of these two
growthin output per workeris measured dummyvariables is an index of the extent
by the change in the logarithmof value to which thejob requires the sort of work
added per worker between 1967 and in which care is important but haste is
1977,25and capitalintensity by 1 - (payroll+ difficultto penalize. Therefore, less use of
materials)/shipments in 1977. piece rates would be expected where such
The IWS also provides detailed occupa- judgments are important.
tion coding forworkersin studied occupa-
tions-sufficientlydetailed occupations to Establishment-LevelRegressions
correspond closely (often with identical
titles)to those in the Dictionary of Occupa- Table 3 presents a set of equations
tionalTitles.From the DOT file,I matched showing the relationship between the
three typesof variables. First,the primary proportionsof workerspaid incentivepay
skill measures are the three indicatorsof and standard rates (with merit pay the
General Educational Development(reason- omitted category),using differentcombi-
ing, mathematics,and language) and the nations of the IWS methods of pay for
required level of SpecificVocational Prep- these categories. Table 4 presents similar
aration. In addition, the DOT rates occu- equations, but with the sample divided
pations on a 5-pointscale for 11 aptitudes, into union and nonunion establishments.
ranging from math ability to foot-eye The results for (In-) establishmentsize
coordination; the maximumof these apti- are as predicted by models in which
tude ratingsis used as an additional skill difficultyin monitoring workers leads
indicator.26Second, the DOT indicates large employersto avoid meritpay and to
whether an occupation requires a toler- use either standard rates or incentivepay
ance forchangingduties or a tolerancefor instead. Larger establishmentsare more
repetitivework. The differencebetween likely than smaller firmsto use incentive
these variables(whichtakes the values - 1, pay in both union and nonunion settings;
0, 1) is used as a measure of diversityof theyare more likelyto use standard rates
duties. Third, the DOT indicates whether in nonunion establishments and across
thejob requires "adaptabilityto situations nonunion and union establishmentscom-
bined, but not in union workplaces. The
25
related hypothesisthat occupational con-
Although growth due to greater capital per centration encourages piece rates and
workercannot be distinguishedfrom growthdue to
technologicalchange, either cause of "overly"rapid perhaps discourages standard rates is not
growthshould reduce the incidence of piece rates. supported by the data. The coefficients
26 Using the maximum rather than the mean are wrong-signed,though not very pre-
reflectsa beliefthat "skill"means that unusual levels ciselyestimated.
of something are required; one would not down-rate
the "skill" of mathematicians(or football players)
As noted previously, "skill," which is
because littlehand-eye coordination (language apti- measured here by a wage-weightedoccu-
tude) is required for thejob. pation index (based on workersin studied

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
 FIRMS' CHOICE OF METHOD OF PAY 177-S

Table 3. Effectsof Various Factors on Method of Pay: 3,211 Establishments.

of Workers
DependentVariable= Proportion Paid by:
Individual SingleRateso0
Independent Mean All Incentive SingleRatesor Range:Seniorityor
Variable (Std. Dev.) Incentives or Bonus Range:Seniority Range:Combination
in (Employment) 4.84 .022* .019* .034* .051*
(1.15) (.004) (.004) (.008) (.008)
Occupational .29 - .029 - .034 .089 .017
Concentration (.20) (.023) (.020) (.047) (.046)
Wage-Weighted 1.34 .092 .050 - .091 - .269*
Occupation Index (.24) (.068) (.057) (.136) (.133)
Prop. Change in .72 .080* .064* -.063 -.056
Value Added/Worker (.13) (.035) (.030) (.071) (.069)
Capital's Share .25 - .217 - .195 .392 .246
of Costs (.05) (.161) (.136) (.321) (.314)
Prop. Female .29 .038 .042* .020 .099*
(.32) (.021) (.018) (.042) (.041)
Union .47 .004 .003 .303* .296*
(.50) (.018) (.015) (.036) (.035)
Union Threat .19 - .001 - .019 - .146* - .007
(.24) (.033) (.028) (.065) (.064)
Mean of .14 .12 .47 .61
Dependent Variable
All equations have dummy variables for 4-digit industry,region, and metropolitanlocation, not shown
separately.
*ItI 1.96.

occupations), has theoreticallyindetermi-
nate effectson methods of pay. In the
data, there is littleconsistentrelationship
between the skill index and method of
pay.
The hypothesis that rapid growth in
value added per worker is a deterrentto
use of piece rates does not fare well in this
empirical analysis: the relevant coeffi-
cients are wrong-signed and sometimes
significant.One technicalexplanation for
this failure is that the variable is matched
to IWS establishmentsbased on industry
and establishmentsize, and both industry
dummies and establishmentsize are also
entered in the equations. A considerable
amount of variationin the industryby size
cells, however,is not accounted for by the
dummies and ln(establishmentsize): when
the change in value added per worker is
regressed on these variables,only 34% of
the variationis explained. A more substan-
tive explanation may be that large jumps
in output per workerare characterizedby
dramaticchanges in the typeof machinery
in use. Of so, worker opposition to
changes in piece ratesmaybe muted,since
the complaint that the change was unfair
would be less credible.
The relationshipbetween capital inten-
sityand method of pay was theoretically
ambiguous, depending on what one be-
lievescapitalintensity is measuring,though
it is most commonlyconjectured thathigh
capital intensitydiscourages the use of
piece rates. The estimates suggest that
more capital-intensiveestablishmentsare
less likely to use piece rates and more
likely to use standard rates, though the
association is not even close to statistical
significance.The "technical"problemcited
above for growingoutput per workermay
have more force here: industrydummies
and ln(establishmentsize) account for75%
of the variationin our measure of capital
intensity.Moreover, when industrydum-
mies are excluded (so that the cross-
industry relationship is not netted out),
capital intensityis significantly
(negatively)
related to use of incentivepay.
As Goldin's model predicts, establish-
ments with larger proportions of female
workersare more likelyto use piece rates.
The effect on use of standard rates,

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
178-S CHARLES BROWN

Table 4. Effects of Various Factors on Method of Pay: 1,523 Union and 1,688 Nonunion
Establishments.
UnionEstablishments NonunionEstablishments
DependentVariable DependentVariable
Independent Mean All SingleRatesor Mean All SingleRatesor
Variable (Std Dev) Incentive Range:Seniority (Std Dev) Incentive Range: Seniority
ln(Employment) 4.90 .023* .007 4.79 .020* .043*
(1.13) (.006) (.012) (1.17) (.006) (.010)
Occupational .26 - .070 - .003 .32 - .007 .176*
Concentration (.18) (.036) (.076) (.21) (.031) (.057)
Wage-Weighted 1.41 .073 .276 1.27 .099 -.387*
Occupation Index (.25) (.103) (.217) (.22) (.091) (.166)
Prop. Change in .73 .080 -.075 .72 .051 .058
Value Added/Worker (.13) (.050) (.105) (.13) (.053) (.096)
Capital's Share .26 - .087 .791 .24 - .199 - .246
of Costs (.05) (.211) (.444) (.05) (.259) (.473)
Prop. Female .21 .031 .028 .37 .045 .045
(.29) (.032) (.067) (.33) (.027) (.055)
Mean of .11 .66 .17 .29
Dependent Variable
All equations have dummy variables for 4-digit industry,region, and metropolitanlocation, not shown
separately.
* ItI 1.96.

however, is either zero or positive. Gol-
din's model (whichemphasizes the impor-
tance of providingimmediateincentivesto
femaleemployees) seems inconsistentwith
the latterfinding.27It is temptingto argue
simply that the labor force participation
pattern assumed in Goldin's model is no
longer strong enough to produce such
differences. That argument, however,
leaves unexplained the strong association
between proportion female and piece
rates.
As expected from others' research,
unionized establishments make signifi-
cantly greater use of standard-rate pay
than do nonunion establishments.The
effectson incentivepay are insignificant,
both statisticallyand practically. Unions
neitheravoid incentivepay, as the egalitar-
ian model of unions would suggest, nor
use it more often,as the block-supervisory-
discretion model would have it. Rather,
both forcesappear to be at work,more or
less canceling each other out.
27
When detailed methods of pay are used as
dependent variables,the proportionfemale is associ-
ated with a significantincreasein the proportion of
workerspaid by range of rates-seniority.
The final explanatory variable is the
threat of unionization, defined so that
high values represent establishmentsthat
are nonunion but have high "predicted"
values of being union, and are therefore
plausiblyregarded as threatenedbyunion-
ization. I hypothesizedthatsuch establish-
ments would be more likely to use
standard-ratepay systemsin an attemptto
head off unionization. Table 3 shows no
evidence of such an effect.An admittedly
after-the-fact explanation for this result
may be that establishments that have
remainednonunion despitebeing in indus-
triesthatare unionized may be exactlythe
ones thathave the mostto lose fromunion
practices like standard rates, and hence
they devote more resources to remaining
nonunion.
One furtherexperimentis not reported
in Table 3. The IWS file for paints
contained methods of pay for office
workers as well as production workers.
Unfortunately,this is the one industryin
which there are no incentive-payworkers
(see Table 2). Moreover,80% of the office
workers had merit pay systems(and the
rest had standard rates), so there was less
variation than one would like in office

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
 FIRMS' CHOICE OF METHOD OF PAY 179-S

methods of pay. When the proportion of Table 5. Effects

of VariousFactorson
office workers paid standard rates was Methodof Pay:Workersin Selected
added to an equation similar to the OccupationswithDOT Matches.
equations for standard-ratepay in Table DependentVariable=
3, but for paints alone, its coefficientwas Worker Paid by:
positivebut veryimpreciselyestimated.
Single
All Ratesor
Independent Mean Incen- Range:
Worker-LevelRegressions Variable (Std. Dev.) tives Seniority
The results in Table 5 are based on a ln(Employ- 4.79 .025* .026*
sample of individual workers,one drawn ment) (1.14) (.007) (.009)
at random from each establishment'sset Occupational .29 - .025 .064
of workers in studied occupations.28Be- Concentration (.20) (.038) (.054
fore turning to the establishment-level Wage-Weighted 1.35 .137 - .130
variablesin Table 5, it is worthnotingthe Occupation Index (.24) (.114) (.164)
ways in which these regressions differ Prop. Change in .72 .094 -.008
Value Added/ (.12) (.058) (.083)
fromthose in Table 3. First,the establish- Worker
ment-levelmethods of pay referredto all
Capital's Share .25 -.269 .449
nonsupervisory production workers, of Costs (.04) (.260) (.373)
whereas Table 5 is limited to individual Prop. Female .29 .001 .072
workersin "studied" occupations. Second, (.33) (.037) (.053)
about a tenth of the potential worker Union .49 - .012 .364*
sample was lost due to a failure to match (.50) (.029) (.041)
workers' IWS occupations to any DOT Union Threat .19 - .019 - .055
occupation. Finally, the coefficients re- (.24) (.052) (.075)
ported in Table 5 control for the DOT Gen Educ Dev: 2.69 .050* -.032
characteristics(and include a variable for Reasoning (.81) (.020) (.029)
the gender of the individual worker), Gen Educ Dev: 1.92 .010 -.023
whereas those in Table 3 do not. Fortu- Mathematics (1.07) (.014) (.020)
nately,the results for establishment-level Gen Educ Dev: 1.86 - .051* .037
variables in Table 5 are not verydifferent Language (.80) (.014) (.021)
from those in Table 3. The most obvious SpecificVoc 3.90 .005 .016
Preparation (1.78) (.008) (.011)
change is that the proportion female no
Maximum 3.04 -.005 -.030
longer increases use of incentivepay-but Aptitude Score (.52) (.016) (.023)
the individualworker'sbeing female does. Diversified -.46 - .046* .026
(There is stilllittleevidence of employers' Duties (.40) (.017) (.024)
havinga comparable aversionto women in Precise .79 .087* - .093*
standard-ratejobs.) Standards (.40) (.017) (.024)
The most noteworthyresultsin Table 5 Generalize, .51 - .073* .061*
are those concerning the worker-level Evaluate, Decide (.57) (.016) (.024)
Female .29 .070* - .030
28
One drawbackof the worker-level data is that (.45) (.019) (.027)
the characterizationof the methodof pay is less Mean of .16 .44
detailed-onlytime-rated and incentive-paidworkers Dependent Variable
are distinguished.Thus, for time-rated workersin N = 2,888.
establishmentsthatuse bothstandardratesand merit All equations have dummy variables for 4-digit
pay,itis notcertainin whichcategory theindividual industry, region, and metropolitan location, not
belongs.Sinceonlyabout 10% of theestablishments shown separately.
use both standardrates and meritpay for their * tl? 1.96.
production workers, however, thisproblemis notan
overwhelming one. It was finessedin the obvious variables.The various skillmeasures show
way:fortime-rated workers, the "dummy"variable
forstandard-rateworkers is in facttheproportion of littleoverall relationshipto method of pay.
the establishment's time-ratedworkers(standard Incentivepay is positivelyrelated to a high
ratesplusmeritpay)whoreceivedstandardrates. requirement for general reasoning skills,

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
180-S CHARLES BROWN

and negatively related to a need for unionizationon the use of incentivepay or

general language skills. The remaining the effectof skilllevel on methods of pay
coefficientsare tinyand statisticallyfrag- in general, there was no strong a priori
ile. Given the theoretically ambiguous basis for predicting the relationship be-
relationshipbetween skill and method of tween observable measures and methods
pay detailed above, these results should of pay. In most of the other cases, bearing
not be over-interpreted. in mind the undeniable difficultyof
As predicted, there is less use of measuring the factors that researchers
incentivepay (and greateruse of standard have cited as likely to be important,the
rates) in jobs with diverse duties than in resultsof the empiricalanalysisseem to be
jobs with unchanging duties repetitively generally in line with theoretical predic-
performed. Moreover, since the variable tions. Specifically,the resultsconfirmthat
runs from - 1 to + 1, the estimatedeffect larger establishmentare less likelyto use
on use of incentivepay is practicallyquite piece rates and predominantly female
significant. establishmentsmore likelyto do so; union-
The variables labeled "precision" and ized establishments make significantly
"generalize, evaluate, decide" were in- greater use of standard-ratepay than do
tended to capture, respectively,situations nonunion establishments;incentivepay is
in which accuracy was important but less likelyin jobs with a varietyof duties
relativelyeasily verifiedand situationsin than in jobs with a narrow set of routin-
which judgments that would be hard to ized duties; and the ease of monitoring
evaluate numericallywere important.The work quality is correlated with the use of
former should encourage, and the latter incentivepay. Other hypotheses-that oc-
discourage, use of piece rates (with per- cupational dispersion, capital intensity,
haps opposite-signed impacts on use of and rapid growth in value added per
standard rates). The data stronglyconfirm employee should all be negativelyassoci-
thisprediction.Again, the practicalsignif- ated with the use of piece rates, and the
icance of the coefficientsis noteworthy. threat of unionization should encourage
greater use of standard rates-are much
less well supported by the results. In my
Conclusions
view, the failure to show the last three
The goal of this studywas to determine relationships may reflect problems in
whether the use of differentmethods of measuring variables the effectsof which
paying production workers is related in may not be terriblylarge in the firstplace.
predictableways to the costs of undertak- I hope thissuggestiondoes notseem a mere
ing the monitoring that the methods excuse-the tired refrainof measurement
require. For incentive pay, these are the problems,the firstrefuge of scoundrels-
costs of measuring output and settingup because the problemshere are genuine.
(and updating) the relationship between One methodological point is notewor-
pieces and pay. For meritpay, theyare the thy. The matching of IWS and DOT
costs of obtaining sufficientlycareful rat- occupation codes was undertakenwiththe
ings from supervisors and of convincing knowledge that the IWS occupations were
the workers that these ratings should be both much more detailed and much more
taken seriously.The costs of standard rate DOT-like in concept than, say, 3-digit
pay systems are presumably negligible, Census occupations-and the hope that
although such systemsforfeitthe benefit IWS-DOT matcheswould amount to more
of correctincentivesthat a well-conceived than previous attemptsto matchDOT and
piece rate (or merit pay) system can Census occupations (including my own
provide. An attemptwas also made, where work) had managed. That hope seems to
possible, to account for institutionalfac- have been confirmed.It is a pitythere are
tors (specifically, unionization, or the not more worker-levelvariables on the
threatof unionization). IWS for the IWS-DOT matched job
In some cases, such as the effect of characteristicsto explain!

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
 FIRMS' CHOICE OF METHOD OF PAY
REFERENCES
Aigner, Dennis, and Glen Cain. 1977. "Statistical
Theories of Discrimination in Labor Markets."
Industrialand LaborRelationsReview,Vol. 30, No. 2
(January),pp. 175-87.
Anderson, Dave. "'Super' Jets Question Today's
Jets."N.Y. Times,Aug. 7, 1988.
Barron, John, and Mark Lowenstein. 1986. "On
ImperfectEvaluation and Earnings Differentials."
EconomicInquiry,Vol. 24, No. 4 (October), pp.
595-614.
Beach, Dale. 1975. Personnel: The Managementof
Peopleat Work.3d Edition. New York: MacMillan.
Bureau of National Affairs.1981. "Wage and Salary
Administration."Personnel Policies Forum No.
131.
Carlson, Norma. 1982. "Time Rates Tighten Their
Grip on Manufacturing." MonthlyLabor Review,
Vol. 102, No. 5 (May), pp. 16-22.
Cleland, Sherill. 1955. The Influenceof Plant Size on
Industrial Relations. Princeton, N.J.: Industrial
Relations Section, PrincetonUniversity.
Cornell, William. 1936. Organization and Management
in Industryand Business.New York: Ronald Press.
Cox, John. 1971. "Time and Incentive Pay Practices
in Urban Areas." MonthlyLabor Review,Vol. 91,
No. 12 (December), pp. 53-55.
Dickens,Williams,and Lawrence Katz. 1986. "Indus-
trial and Occupational Wage Patternsand Theo-
ries of Wage Determination." NBER Working
Paper 2014.
Freeman, Richard. 1982. "Union Wage Policies and
Wage DispersionWithinEstablishments."Industrial
and Labor RelationsReview,Vol. 36, No. 1 (Octo-
ber), pp. 3-21.
Garen, John. 1985. "Worker Heterogeneity,Job
Screening, and Firm Size." Journal of Political
Economy, Vol. 93, No. 4 (August), pp. 715-39.
Goldin, Claudia. 1986. "MonitoringCosts and Occu-
pational Segregation by Sex." Journal of Labor
Economics, Vol. 4, No. 1 (January),pp. 1-27.
Green,Jerry,and Nancy Stokey. 1983. "A Compari-
son of Tournaments and Contracts."Journal of
Political Economy,Vol. 91, No. 3 (June), pp.
349-64.
Hamner, W. Clay. 1983. "How to Ruin Motivation
with Pay." In Richard Steers and Lyman Porter,
eds., Motivationand WorkBehavior.3d Edition. New
York: McGraw-Hill,pp. 264-276.
Harris, Milton, and Arthur Raviv. 1979. "Optimal
Incentive Contractswith ImperfectInformation."
Journalof EconomicTheory,Vol. 20, No. 2 (April),
pp. 231-59.
Holmstrom,Bengt. 1979. "Moral Hazard and Observ-
ability."Bell Journalof Economics,Vol. 10, No. 1
(Spring), pp. 74-91.
. 1982. "Moral Hazard in Teams." BellJournal
ofEconomics, Vol. 13, No. 2 (Autumn),pp. 324-40.
InternationalLabour Office. 1984. Payment byResults.
Geneva: ILO.
Jacoby,Sanford. 1984. "The Development of Inter-
nal Labor Markets." In Paul Osterman, ed.,
Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015
181-S
Internal Labor Markets.Cambridge, Mass.: MIT
Press.
King, Sandra. 1975. "Incentive Pay in Auto Repair
Shops." MonthlyLabor Review, Vol. 98, No. 9
(September),pp. 45-48.
Lansburgh, Richard, and William Spriegel. 1940.
IndustrialManagement. 3d Edition. New York: John
Wiley.
Lazear, Edward. 1986. "Salaries and Piece Rates."
JournalofBusiness,Vol. 59, No. 3, pp. 405-31.
Lewis, L. Earl. 1960. "Extent of Incentive Pay in
Manufacturing."Monthly LaborReview,Vol. 83, No.
5 (May), pp. 460-63.
Lundberg, Shelly,and Richard Startz. 1983. "Private
Discriminationand Social Interventionin Compet-
itive Labor Markets." AmericanEconomicReview,
Vol. 73, No. 3 (June), pp. 340-47.
Maidment,Paul. 1987. "No Small Change: A Survey
of Japan." The Economist, December 5, pp. S1-S34.
McKersie, Robert, Caroll Miller, and William Quar-
terman. 1964. "Some Indicators of Incentive Plan
Prevalence." MonthlyLabor Review,Vol. 84, No. 3
(May), pp. 271-76.
McLaughlin, David. 1986. "Pay for Performance:A
Perspective." Topicsin Total Compensation, Vol. 1,
No. 1 (Fall), pp. 7-14.
Morse, Edward. 1986. "ProductivityRewards for
Nonmanagement Employees." Topicsin Total Com-
pensation,Vol. 1, No. 1 (Fall), pp. 85-100.
Nalebuff, Barry, and Joseph Stiglitz. 1983. "Prizes
and Incentives: Towards a General Theory of
Compensation and Competition." Bell Journal of
Economics, Vol. 14, No. 1 (Spring), pp. 21-43.
National Research Council, Panel on Technology
and Women's Employment. 1986. ComputerChips
and Paper Clips.Vol. 1. Washington,D.C.: National
Academy Press.
Oi, Walter. 1983. "Heterogeneous Firms and the
Organizationof Production."EconomicInquiry,Vol.
21, No. 2 (April), pp. 147-71.
Pencavel, John. 1977. "Work Effort,On the Job
Screening,and AlternativeMethods of Remunera-
tion." In Ronald Ehrenberg,ed., Researchin Labor
Economics,Vol. 1. Greenwich, Conn.: JAI Press,
pp. 225-58.
Personick,Martin.1987 . "White-CollarPay Determi-
nation Under Range-of-Rate Systems." Monthly
Labor Review,Vol. 107, No. 12 (December), pp.
25-30.
Pryor,Frederic. 1984. "Incentivesin Manufacturing:
The Carrot and the Stick." MonthlyLabor Review,
Vol. 104, No. 7 (July),pp. 40-43.
Seiler, Eric. 1984. "Piece Rate vs. Time Rate: The
Effect of Incentives on Earnings." Review of
Economics Vol. 66, No. 3 (August), pp.
and Statistics,
363-76.
Sekscenski,Edward. 1980. "Job Tenure Declines as
Workforce Changes." BLS Special Labor Force
Report 235.
Shaiken, Harley. 1987. "When the Computer Runs
the Office."New YorkTimes,March 22, sec. 3, p. 3.
Stigler,George, and Leonard Sayles. 1980. Personnel:
182-S CHARLES BROWN

The Human Problemsof Management.3d Edition.
Englewood Cliffs,N.J.: PrenticeHall.
U.S. Departmentof Commerce. 1979. County
Business
Patterns1977. Washington,D.C.: GPO.
Weiss, Andrew. 1987. "Incentives and Worker
Behavior: Some Evidence." In Haig Nalbantian,
ed., Incentives,Cooperation,and Risk Sharing. To-
towa, N.J.: Rowman & Littlefield,pp. 137-50.

Downloaded from ilr.sagepub.com at Univ of Connecticut / Health Center / Library on July 1, 2015