An Interplant Test of the Efficiency Wage Hypothesis
Author(s): Peter Cappelli and Keith Chauvin
Source: The Quarterly Journal of Economics, Vol. 106, No. 3 (Aug., 1991), pp. 769-787
Published by: Oxford University Press
Stable URL: http://www.jstor.org/stable/2937926
Accessed: 06-04-2018 11:07 UTC
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.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access
to The Quarterly Journal of Economics
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
AN INTERPLANT TEST OF THE EFFICIENCY WAGE
HYPOTHESIS*
PETER CAPPELLI AND KEITH CHAUVIN
The analysis that follows tests the shirking model of efficiency wages by
examining the relationship between rates of employee discipline and relative wage
premiums across plants within the same firm. The structure of this data set controls
for many of the problems that confound other tests of efficiency wage arguments,
and the results suggest that greater wage premiums are associated with lower levels
of shirking as measured by disciplinary dismissals. Shirking and discipline are also
lower where conditions in the labor market raise the costs associated with shirking
by making it more difficult to find alternative employment. It is less clear, however,
whether the wage in this case is necessarily efficient in the sense of generating
reductions in discipline sufficient to offset the costs of the wage premium.
I. INTRODUCTION
Efficiency wage models are based on the notion that there is a
relationship between relative wage levels and worker productivity,
broadly defined, which in turn explains a variety of otherwise
puzzling behavior such as the presence of involuntary unemployment.
Perhaps the most popular of these arguments have been
those suggesting that wage premiums and the threat of losing them
create incentives for workers to reduce unproductive behavior or
"shirking." The structure and implications of such efficiency wage
arguments have been debated at length, but there have been few
attempts to test empirically the hypothesis that wage premiums
can reduce shirking. This paper provides a direct test of the main
implications of the shirking efficiency wage model using plant-level
data from the auto industry. The results suggest that wage
premiums are in fact associated with lower levels of disciplinary
problems, as the shirking models of efficiency wages imply. But it is
less clear whether the overall benefits of the reduction in shirking
exceed the cost of the wage premium.
II. SHIRKING AND EFFICIENCY WAGES
The argument for a positive relationship between worker
productivity and wage levels began in development economics
*Thanks to Franklin Allen, Francine Blau, William Dickens, Wallace Hendricks,
and Lawrence Kahn for helpful comments. The first author thanks the
Center for Human Resources at the Wharton School for supporting this research.
o 1991 by the President and Fellows of Harvard College and the Massachusetts Institute of
Technology.
The Quarterly Journal of Economics, August 1991
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
770 QUARTERLY JOURNAL OF ECONOMICS
where the connection was physiological: higher wages provided a
better diet which in turn permitted greater effort and output (e.g.,
Leibenstein [1957], Bliss and Stern [1978]). Contemporary versions
of the efficiency wage argument rely on the costs associated
with dismissal to provide an incentive for workers to be productive
and avoid shirking. Shapiro and Stiglitz [1984], for example, note
that in the absence of jobs suited to piece rates, it is difficult for
employers to monitor job performance continuously and therefore
to adjust wages downward when performance falls. Instead, they
argue that employers rely on occasional monitoring and the threat
of dismissal if workers are caught shirking to provide the incentive
for them to maintain their performance. A worker would have no
incentive to avoid being fired at full employment and marketclearing
wages, however, because they could immediately find an
equivalent job at the same rate of pay. So employers pay a premium
above market rates to create an incentive for workers to keep their
current job. When all employers raise their wages, there is an
excess supply of labor at that wage and involuntary unemployment:
some workers with a reservation wage below the new market
level remain unemployed.'
Theoretical arguments aside, the important empirical question
for efficiency wage arguments is whether there is any evidence
of a relationship between wage premiums and performance, and
then whether the overall benefits from reduced shirking outweigh
the costs of the wage premium (they should be equal at the
margin). There is anecdotal evidence for some kind of an efficiency
wage effect, such as the fact that many firms pay above market
rates as a matter of policy and presumably do so because of the
gains such a policy offers (see, e.g., Milkovich and Newman [1987,
p. 210]). But there have been few attempts to test even a weak form
of the efficiency model-are there productivity gains at all from
wage premiums-because it is difficult to identify and control for
exogenous, nonwage factors that might affect worker productivity,
such as differences in the quality of workers and in the nature of
their jobs. It is also difficult econometrically to avoid the identification
problem: are higher wages the cause or the result of greater
worker productivity?2
1. See Katz [1986], Stiglitz [1986], and Akerlof and Yellen [1986] for reviews of
the efficiency literature.
2. Raff and Summers [1987] provide a historically based examination of Henry
Ford's $5 day wage plan and find some support for efficiency wage effects. Leonard
[1987] examines levels of supervision across firms and finds for some occupations
that it is higher at firms where wage premiums are higher, the opposite of what
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 771
III. PREVIOUS RESEARCH
Although attempts to test efficiency wage models as such have
been rare, previous research has examined the relationships
between wage premiums and some aspects of productivity. The
conclusions from this research suggest that if the efficiency wage
effects exist, they may be offset empirically by other effects. For
example, a wage premium of the kind associated with an efficiency
wage may produce income effects that could reduce any positive
relationship with productivity. One way to purchase leisure where
workers cannot control their work schedule is by shirking: absenteeism,
reducing work effort on the job, etc. The price of that
leisure includes the expected cost associated with being caught.
Employers are unlikely to set a wage unilaterally that leads to a
reduction in effort, but such wage premiums might be established
through collective bargaining, for example. Taylor [1961] and
Walters [1977] argue that when wages for British miners were
raised above market rates, the miners used the increases to
purchase additional leisure through absenteeism and shirking.
Hirsch and Hausman [1983] find econometric support for this
argument and an overall negative relationship between productivity
and the miner's wage increase.
Industrial psychologists have also examined the relationship
between pay levels and performance, and while none of their
studies provides a rigorous test of efficiency wage arguments per se,
some of the conclusions are relevant. The stream of research most
relevant to the efficiency wage position in psychology is known as
"equity theory" [Adams 1963], which examines the relationship
between relative pay levels and productivity-related job behavior.
Equity theory borrows heavily from Adam Smith's [1776] notions
about the role of comparisons in job choice. Stated simply, workers
compare the inputs they bring to a job (skill, effort, etc.) and the
outcomes from the job (pay and nonpecuniary benefits) with inputs
and outcomes for jobs they choose as comparisons. If workers feel
undercompensated relative to their comparisons, they may redress
this perceived inequity or cognitive dissonance by adjusting their
inputs downward (shirking), leaving their job (turnover),
efficiency wage arguments would suggest if supervision is thought of as providing
monitoring that reduces shirking (i.e., supervision costs are a substitute for
shirking costs). Whether this is a good test of shirking depends on whether
supervision is a good proxy for shirking. The fact that supervisors may also perform
some higher skill aspects of production work, in addition to monitoring performance,
suggests that the relationship may be a complicated one.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
772 QUARTERLY JOURNAL OF ECONOMICS
"rationalizing" the discrepancy-reevaluating their inputs and
outcomes, choosing new comparisons, etc.-or some combination
of the three that will reestablish a sense of equity. (One problem
with this literature is the difficulty in explaining which option will
dominate.) Similarly, if they feel overcompensated, in the spirit of
efficient wage arguments, they may in theory improve their
performance, reduce turnover, or again rationalize the discrepancy
away.
As Akerlof [1984] notes, laboratory studies based on models of
equity comparisons found that subjects who were made to feel
overpaid relative to a control group reduced this inequity by
increasing their inputs and outperforming their more equitably
paid colleagues, as efficiency wage arguments would predict [Adams
and Rosenbaum 1962; Pritchard et al. 1972].3 When these experiments
were continued for more than one day, however, the
overpaid workers soon dealt with the perceived inequity by rationalizing
away any sense of overpayment-reevaluating their contributions
and choosing new comparisons. One way to think of this
rationalization process is that the workers come to see themselves
as being in a different, higher-wage labor market. Productivity
then fell back to the level of the equitably paid subjects [Vroom,
1964; Lawler et al., 1968]).
One problem with these laboratory studies is that the costs
associated with dismissal, which drive the efficiency wage hypothesis,
have been insignificant, especially when compared with fulltime
employment. But if respondents do rationalize away any sense
of overpayment even where a very obvious market premium is
paid, then perhaps the importance of the wage premium as an
incentive not to shirk is reduced.4
3. These same studies, however, found that workers who were overpaid on a
piecework basis dealt with the inequity by reducing their inputs and productivity;
with inequitable piece rates, the argument goes, the perceived inequity in earnings
gets worse with increased output, so reducing output reduces the rate of inequity.
4. A related stream of research developed by March and Simon [1958]
addresses this concern by examining the relationship between alternative job
opportunities and productivity in actual jobs (e.g., Smith, Kendall, and Hulin
[1969]). But with the exception of weak relationships with turnover [Hulin,
Roznowski, and Hachiya, 1985], they find no relationship between opportunities in
the labor market, broadly defined, and productivity, also broadly defined. Unfortunately,
these studies also may not accurately assess the efficiency wage hypothesis
because they do not attach wages to current jobs or to measures ofjob opportunities
making it difficult to assess accurately the costs associated with dismissal. The test
we outline below addresses both problems by examining the efficiency wage
hypothesis in an actual industrial setting where labor markets and alternative
wages can be accurately identified.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 773
Of course, the loss of a wage premium may not be the onl
associated with dismissal. These other costs, such as the possib
of a period of unemployment if caught shirking and dismisse
also deter shirking independent of any concern about losing the
wage premium. But the possibility that wage premiums may
generate income effects or that workers may rationalize them away
suggests that the relationship between wage premiums and a
reduction in shirking may in fact be ambiguous and needs to be
tested empirically.
IV. THE MODEL AND DATA
The test of the relationship between shirking and wage
premiums that we propose uses a unique set of plant-level data for
1982 taken from the internal records of a large manufacturing
company.5 The characteristics of these data address many of the
problems noted above that plague econometric tests of the efficiency
wage model. First, many of the exogenous factors that might
affect worker performance are standardized across plants. For
example, all of the plants included in the study are represented by
the same union, the United Auto Workers, and are covered by the
same labor agreement which standardizes virtually all of the terms
and conditions of employment across plants. All of the data are for
production workers, and production jobs are virtually identical
within categories of plants (e.g., assembly versus component
plants); the characteristics of the jobs and of the workers needed to
fill them do not differ. Second, management's personnel policies on
issues such as shirking and discipline are centrally controlled and
are generally identical across all plants as are the union's policies
for dealing with management on these issues.6 Plant managers
might be expected to differ in where and how they exercise
discipline, but the union may appeal all discipline cases through the
contractual grievance procedure, and the collective bargaining
agreement states that a common, companywide Appeal Committee
will rule on all grievances unresolved at the plants. The fact that a
central body has de facto oversight for discipline cases across all
plants serves to standardize discipline practices across those
5. The company requests anonymity as a condition of using the data.
6. Local collective bargaining agreements exist at the plant level, but they
cannot alter the national agreements and are generally limited to work rules issues
such as assigning job classifications.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
774 QUARTERLY JOURNAL OF ECONOMICS
plants. It would therefore be difficult for disciplinary practices and
standards to vary across plants (e.g., with economic circumstances)
and to create problems of simultaneity. In short, not only are
measures of performance and shirking comparable across plants,
but virtually all of the exogenous factors that could affect shirking
and discipline are common across plants.
Finally, because wages and other aspects of compensation are
established through a companywide collective bargaining agreement,
we know that they cannot suffer from the identification
problem outlined above: wages in each plant obviously are not
affected by differences in productivity across plants because they
are set centrally and are identical across all plants.7
Wages can, however, be a cause of differences in productivity
across plants. These workers certainly appear to earn a wage
premium above market rates;8 and although the wage rate for
production workers is the same across all plants, the size of their
wage premium varies considerably across plants. The labor market
for the unskilled and semi-skilled jobs that make up production
work in this manufacturing industry is limited geographically to a
reasonably small area, and unskilled workers focus their job
searches within that area.9 By comparing the hourly wage paid in
each plant to the prevailing wage in the plant's labor market,
defined here as the average hourly wage for production work in
each Standard Metropolitan Statistical Area (SMSA), we can assess
the extent of the wage premium above the market rate at each
plant-what a worker has to lose if dismissed.'0 That premium
7. In addition, wage rates are an accurate proxy for comparing total compensation
across plants relative to their labor markets because the terms and conditions
of employment cannot vary across plants. For example, plants cannot offset higher
wage rates relative to their labor markets with lower benefits and worse working
conditions.
8. Production workers in this industry in 1982 earned approximately 150
percent of the average hourly wage for similar production work in the economy as a
whole [Employment and Earnings, 19821, and workers in this company earned
more than the industry average. Wages do not vary with seniority in this contract
(other than for probationary employees) or with other characteristics that might
vary systematically across plants.
9. For example, we calculate from the 1984 Current Population Displaced
Worker Survey that only 5 percent of those who lost their jobs between 1979 and
1983 moved to take a job or to look for work. This is consistent with estimates that
only about 6 percent of the displaced workers who received Trade Adjustment
Assistance even took the relocation resources to which they were entitled [Swigart,
1984].
10. Even if one believes that differences in average wage levels in each
community simply reflect cost-of-living differences, the gap between plant wages
and area wages still varies across plants and represents the potential loss associated
with dismissal, in this case varying with living costs. See Topel [1986] for arguments
about the causes of differences in market wages across local labor markets.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 775
varies across the plants in this company from zero (where plant
wages are no higher than the area rate) to as much as 100 percent
above the area rate. We recognize that because these premiums are
not set unilaterally by the employer, there is no reason to believe
that they will necessarily generate cost-effective reductions in
shirking. Nevertheless, if the underlying theory is correct, the
premiums should lead to some reduction in shirking regardless of
what produced them, and the reductions in shirking should vary
directly with the size of the premium.
Wage profiles that increase the returns associated with seniority
and, in turn, provide incentives to avoid shirking and dismissal
in order to capture those returns [Lazear, 1979] may provide an
alternative to the use of efficiency wages. The presence of these
seniority-based wage profiles can complicate tests of efficiency
arguments because of the need to sort out the effects associated
with current wage premiums from those associated with seniority
profiles. This is especially so where data are taken across employers
because these wage profiles may differ substantially across employers.
The analysis in this case is simplified because wages in this
company do not differ by seniority, eliminating that complication."
The absence of seniority-based wages, however, makes it
more difficult to generalize conclusions from this case to firms that
do have such wage profiles.
In addition to the lost wage premium, the costs associated with
dismissal include the search costs associated with finding a new
job, and these also vary across geographic areas and plants. Where
unemployment is higher, dismissed workers may have to search
longer or may have to move to other communities before finding a
job. Labor performance, broadly defined, should be higher, and
shirking lower where the expected costs associated with dismissal
are greatest: where wages in the outside labor market are lower
and the difficulty associated with finding a new job is greater.
There may be more than one mechanism through which a
wage premium affects performance and labor costs. For example, a
wage premium may reduce voluntary turnover [Pencavel, 1970;
Phelps, 1970; Salop, 1979]; Leonard [1987] and Krueger and
Summers [1987] find evidence for this effect. Reductions in
voluntary turnover may be a less important issue empirically in
this case because voluntary turnover is less of a cost to firms in
11. We attempt to control for nonwage returns from seniority by including a
variable to measure plant seniority; the relationship between seniority and any
nonwage returns is covered by the contract and should be constant within the firm.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
776 QUARTERLY JOURNAL OF ECONOMICS
unskilled jobs like most in this industry where workers can easily
bereplaced. Efficiency wages may also lead to the selection of better
employees with higher productivity [Malcomson, 1981; Weiss,
1980], but the selection nexus appears to be a very different type of
effect because it operates not on current employees but through the
selection of new ones. The mechanism most central to the basic
concern of efficiency arguments-providing incentives for current
workers to maintain peformance-is through a reduction in shirking
[Stiglitz, 1976; Solow, 1979; Shapiro and Stiglitz, 1984]. The
empirical problem is to find direct measures of shirking at the
workplace. Overall labor productivity measures, for example, may
not be good proxies for shirking even across plants in the same firm
because they can be dominated by factors other than individual
worker performance, such as short-run problems with equipment
or with supplies, changes in techniques and in products, unionmanagement
conflicts, etc.
Fortunately, we have a very precise measure of shirking taken
from the company's internal records: The rate at which workers
were dismissed at each plant for disciplinary reasons (DISL).12 The
costs associated with dismissals are exactly the mechanism behind
the shirking model of efficiency wages. Further, we can be reasonably
certain that these disciplinary actions in fact result from poor
performance and productivity, broadly defined (e.g., low performance
levels, tardiness, absenteeism, breaking safety procedures,
bad relations with supervisors or fellow workers, etc.). The United
Auto Workers' contract with the employer limits management's
ability to discipline employees to such performance-related issues,
and as enforced by a strong union and grievance procedures, these
restrictions effectively eliminate arbitrary and capricious actions.
These disciplinary actions are expressed as a rate per plant
calculated as the number of dismissals/the number of workers for
the year 1982.
Given standardized discipline policies within the firm, the rate
of disciplinary dismissals should reflect the incidence of shirking at
each plant which should in turn be a function of the utility
associated with shirking for workers at each plant. Following
Shapiro and Stiglitz [1984], that utility Us should be lower where
the expected value of alternative employment E (w') is lower and
12. The measure of dismissals may include cases where the union has filed a
grievance and secured reinstatement of a dismissed worker through arbitration.
The number of workers ultimately dismissed may therefore be somewhat lower.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 777
the utility associated with being dismissed Ub is also lower, give
constant probability of being caught shirking and dismissed P an
constant rates of voluntary turnover across workers:
(1) rU,, = (1 -P)w -P[E(wa) -Ubb
where
w = current wage
wa = alternative wage,
r = discount rate used to evaluate the utility.
We assume that r is distributed across individuals = N(urr2)
and can be thought of as determining the preferences for shirking;
higher values of r are associated with higher utility from shirking
(greater weight placed on the gains from reduced effort until
caught). We also assume that dUs/dw > 0, d2U,/d2w < 0. Each
worker will adjust their work behavior such that aU/A (P[E(Wa) Ub]
= aU8/aw. At the efficiency wage w*, aU8/w/U8/w = 1. Some
workers may shirk at the efficiency wage equilibrium because of
the heterogeneity of discount rates.
The utility associated with being dismissed and unemployed
(Ub) is measured by the maximum amount of unemployment
benefits payable to dismissed workers (level x duration, UIBEN)
which vary by state and therefore across plants. (Workers discharged
for willful or deliberate misconduct-above and beyond
simple shirking-may lose some period of eligibility for benefits.)13
The expected value of alternative employment E(Wa) is determined
in part by the costs associated with finding a new job. The extent of
unemployment across plants affects the probability and costs of
finding alternative employment. It is measured first by the rate of
13. The fact that an employer says that a discharge was for cause does not
necessarily constitute dismissal for misconduct for the purposes of unemployment
insurance. State unemployment agencies review discharge claims, and in general,
discipline problems must go beyond ordinary negligence, inefficiency, and poor
performance to include willful, deliberate misconduct in order to be classified as
dismissed for misconduct. Workers whose dismissals meet the unemployment
insurance misconduct test lose some weeks of eligibility for benefits (the level of
benefits is unaffected) with the number of weeks lost rising according to the severity
of the misconduct-up to 26 weeks for felony crimes, e.g. See U. S. Department of
Labor [1990] for details. Workers dismissed for poor performance (shirking) are
generally eligible for full benefits while even those dismissed for gross misconduct,
such as felony convictions, lose only some of their eligibility. Because the basic
framework for eligibility is dictated by the Federal Social Security Act, the
variations in administrative practices across states are relatively minor. The
potential loss of some eligibility creates incentives for workers to keep any
misconduct on the more moderate side of the discipline distribution.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
778 QUARTERLY JOURNAL OF ECONOMICS
unemployment at each plant-the percentage of workers currently
on layoff (LAIDOFF), many of whom are looking for jobs (perhaps
temporary jobs given that historically, most laid-off workers have
eventually been recalled). In larger labor markets the number of
workers laid off at the plant may have only a marginal effect on the
prospect of finding jobs, but it can have a disproportionate effect on
the perceptions of workers concerning those prospects. We also
include the unemployment rate in the SMSA (UE) to help capture
the state of the labor market for other jobs. A measure of average
seniority (SEN) at the plant-percent of workers < 10 yearsprovides
a rough proxy for age-related marketability; other things
equal, workers with more seniority are older and have more
difficulty in securing new employment. Workers with more seniority
also lose seniority-related benefits if dismissed, and this may
offset the reduced disincentive to shirk associated with having less
time remaining on the job (given mandatory retirement) and
therefore a shorter period in which to benefit from wage premiums
Plants with more senior workers may also experience less shirking
because they have already identified and dismissed workers with a
disposition toward discipline problems, in contrast to plants with
more new and untested employees where workers with those
dispositions may not yet have been identified.
Finally, the wage for alternative jobs (wa) is measured by
average wages paid in each SMSA for production work. We define
the wage premium, WPREM, as w - wa, and the effect of the wage
premium on shirking, Ua/(WPREM) < 0 if the efficiency argument
dominates; it could be > 0 if the income effect dominates or = 0 if
the rationalization associated with equity theory holds (also if
efficiency and income effects both have zero effects or cancel out).
There are several potentially important influences on discipline
rates across plants that are exogenous to the efficiency wage
model. The first relates to characteristics of the plant and of the
type of work performed in it. Jobs in assembly plants are also
thought to be more oppressive than those in other plants because
the work is more heavily dominated by the assembly line and
workers have less ability to influence their work. Discipline problems
may be more common in these plants as a result. We include a
variable to identify assembly plants (ASSMBLY) and expect it to
vary positively with discipline rates. However, jobs in assembly
operations where work is highly automated may on average be less
skilled than those in component manufacturing, and wages in the
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 779
outside market should therefore be lower for jobs in the former
than in the latter. If this is so, then the wage premium might be
greater for assembly workers (because wages are the same across
types of plants), reducing the incentive for these workers to shirk
and working against the effect hypothesized above.
In addition, some industry observers believe that worker
attitudes and behavior such as militancy and worker discipline
vary by region of the country. A variable is included to identify
plants in the Michigan area which are older and often have
histories of conflict (MICHIGAN); plants in the south (SOUTH)
are newer and, many believe have more compliant, disciplined
work forces. Plants in other areas-mainly in the Midwestconstitute
the omitted category.
An additional factor affecting the variance in shirking behavior
across plants may be associated with Hirschman's [1970] and
Freeman's [1980a] "voice" argument. If shirking, as psychologists
argue, is in part a reaction against problems at the workplace (i.e.,
absenteeism as a form of withdrawal from an unpleasant situation),
arrangements that give workers an alternative means for
addressing those problems could reduce the incidence of shirking.
All plants have the same mechanisms-union representation,
grievance procedures, etc.-but they may not function equally
well. Fortunately, we have a unique measure of the effectiveness of
these voice arrangements across plants. Corporate-level management
rated the problem-solving ability of local unions and management
at each plant (VOICE) using a ten-point scale (1 = confrontational
relations; 10 = cooperative, problem-solving relations). Where
shirking behavior is a reaction against problems at the workplace,
we expect less shirking at plants where union-management relations
provide a more effective mechanism for solving workplace
problems. These variables, their means, and standard deviations
are presented in Table I.
One potentially important omitted variable may be a comparison
of the level of fringe benefits between the plants and their
outside labor markets. These benefits represent a significant aspect
of compensation, and as Freeman [1980b] notes, fringe benefits are
correlated with wages that could bias a wage measure upward.
Given that fringe benefits vary with wages, it may be best to think
of the wage premium as measuring compensation: both wages and
benefits. It is not obvious what other kinds of variables are omitted:
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
780 QUARTERLY JOURNAL OF ECONOMICS
TABLE I
VARIABLE DEFINITIONS, MEAN VALUES, AND STANDARD DEVIATIONS
Variable Definition Mean S.D.
DISL
Rate of worker dismissals for disciplin- 0.0968 0.076
ary reasons
W.
Average hourly wage in 1982 for pro- $ 10.58 1.5
duction workers in each plant's
SMSA
WPREM
The wage premium for each plant de- $ 2.92 1.5
fined as the average hourly company
wage--W
WPREM'70
The wage premium for each plant in 1.03 0.4
1970 as defined above
UE
Unemployment rate in the SMSA in 13.06 4.0
which the plant is located
LAIDOFF
Percentage of workers on layoff 25.47 14.7
SEN
Percentage of workers with less than 45.08 18.8
ten years of seniority
VOICE
Corporate assessment of cooperative re- 6.96 1.4
lations/problem-solving at plant
(1 = least cooperative; 10 = most)
ASSMBLY
Dummy variable for assembly plants 0.23 0.4
MICHIGAN
Dummy variable for plants in Michigan 0.60 0.5
SOUTH
Dummy variable for plants in the South 0.06 0.2
UIBEN
Maximum state unemployment insur- $4,928.31 1,044.3
ance benefits by plant location
SMSA data from U. S. Bureau of Labor Statistics 1982 Area Wage Surveys. Unemployment data are from
Employment and Earnings [1982] and from Highlights of State Unemployment Compensation Laws [1983]. All
other data are from internal company records for 1982.
while it may be that some workers may be more inclined to shirk,
ceteris paribus, there is certainly no consensus as to what characteristics
determine that inclination and no a priori reason to believe
that they would be correlated with the right-hand-side variables.
The incidence of shirking as measured by disciplinary layoffs
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 781
across plants can be estimated as follows:
(2) In [DISL/1 - DISL] = a - bNWPREM - b2UE
- b3LAIDOFF + b4SEN - b5VOICE + b6ASSMBLY
+ b7MICHIGAN - b8SOUTH + bgUIBEN + e.
Plant-level data based on the average characteristics of the
individual workers in each plant are especially susceptible to
heteroskedasticity because the error term varies inversely with the
number of workers employed in the plant: where N = workers in
each of k plants, E (e) = 1/N2 E1[{elk 12] = S2/Nk. As Amemiya and
Nold [1975] note, the weighting used to correct for heteroskedasticity
should be modified in log odds regressions to include an
equation error as a measure of omitted variables. Otherwise, the
standard errors of the estimates will be underestimated, and the
efficiency of those estimates reduced. Each observation was
weighted by [SOLS + {EMPLOYMENTk* DISLk(1 - DISLk) 1`1`2,
where S2LS is an estimate of the variance calculated from the
parameters of an OLS estimation of the above log odds equation,
and EMPLOYMENTk is the number of workers at each of k plants.
The log odds equation was estimated again using the weighted
observations.
V. RESULTS
The results in Table II provide empirical support for the
efficiency wage hypothesis as applied to shirking. The wage premium
(WPREM) suggests that there are fewer shirking-related
discipline problems where wage premiums are higher. This relationship
dominates any income effects or any of the rationalization
associated with equity theory that may be nested in the model and
that would operate in the opposite direction.
There is also support for the argument that the costs associated
with dismissal in addition to the loss of the wage premium
influence the incidence of shirking. Those costs include the prospects
for reemployment as measured by layoff rates across plants
(LAIDOFF) and seniority/age-related job prospects (SEN).14 The
fact that the costs associated with unemployment could provide
such discipline is an old argument in economics that goes back at
14. To examine whether the variance in layoff rates across plants is driven by a
variance in management practices which, in turn, could also account for the
difference in discipline rates across plants, we ran the above regression without the
LAIDOFF variable. The results (see Appendix) are virtually unchanged.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
782 QUARTERLY JOURNAL OF ECONOMICS
TABLE II
REGRESSION RESULTS FOR RATES OF DISCIPLINARY LAYOFFS ACROSS PLAN
[(Dependent variable = In DISL/1 - DISL)]
Weighted least squares Elasticitya
(1) (2) (3)
Intercpt -1.071 -1.20
(0.96) (1.16)
WPREM -0.21** -0.24* 0.547
(0.10) (0.13)
UE -0.03 -0.012 0.330
(0.04) (0.045)
LAIDOFF -0.009* -0.008* 0.207
(0.005) (0.005)
SEN 0.007* 0.008* 0.285
(0.004) (0.004)
VOICE -0.14*** -0.14*** 0.908
(0.054) (0.05)
ASSMBLY 1.008*** 1.02*** 0.210
(0.182) (0.18)
MICHIGAN 0.09 0.289 0.053
(0.238) (0.24)
SOUTH 0.054 -0.057 0.003
(0.336) (0.36)
UIBEN -0.00004 -0.00006 0.178
(0.00009) (0.00009)
WPREM'70 0.37 0.291
(0.54)
S.E.E.= 0.62 0.61
F = 6.43 6.32
n = 78 78
* = significant at 10 pe
** = significant at 5 pe
= significant at 1 perc
All of the values for D
the variance in DISL exp
a. Evaluated at the mea
least to Marx's [1976] notion of an "industrial reserve army." In
addition, the quality of the relationship between plant management
and the local union (VOICE) has a significant effect on the
incidence of shirking, perhaps providing a means for solving
problems that might otherwise contribute to discipline rates.
Finally, assembly plants (ASSMBLY) are associated with significantly
higher discipline levels, perhaps because workers there have
more to lose from dismissal or perhaps because the work there may
be more difficult. The other variables were not significant.
The wage premium result (although not the unemployment
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 783
result) is also consistent with a different mechanism: the selectio
of better workers. Higher wage premiums may lead to larger
queues of workers and possibly to the selection of better workers
who, for example, may have less initial preference for shirking. But
this mechanism relies on a relative wage premium not at the
current date but when the present work force was hired. While the
premiums in the two periods are probably not identical, it is
possible that the variable for the 1982 premium is positively
correlated with the hiring premium and could be picking up some
of the effects associated with the latter.
VI. EXAMINING AN ALTERNATIVE HYPOTHESIS
We attempt to examine the potential effect of a wage premium
on selection first by calculating the wage premium across plants for
the year 1970. Given median seniority of twelve years in our 1982
data, this corresponds to the year the median employee was hired.
The wage premium in 1970 (WPREM'70) was calculated in the
same way as was the 1982 premium, and we enter it along with the
1982 premium into equation (2). The results are reported in Table
II and suggest that the wage premium at the point of hire did not
have a significant effect on discipline rates. Indeed, the effect is in
the wrong direction. The current premium, however, retains its
significance and the expected sign, suggesting that it is not driven
by selection effects.
These results are consistent with the basic shirking argument
that the costs of dismissal, including the loss of a wage premium,
create incentives that reduce shirking. The more difficult question
is whether the wage premium is in fact a cost-effective way of
reducing shirking. In other words, does the optimality condition for
efficiency wages hold? As noted above, because wage levels and the
premiums that result are not set unilaterally by the firm, we would
not necessarily expect the premiums to optimize the reduction in
shirking. It might still be interesting, though, to try to estimate the
value of the reduction in shirking.
The best way to estimate the benefits of the reduction in
shirking associated with wage premiums would be to estimate a
production function that includes the discipline rate as a right-handside
variable. Ideally, such tests might also include measures of any
reductions in turnover associated with wage premiums in order to
assess the relative importance of these different efficiency wage
mechanisms. Unfortunately, we do not have the data to estimate a
production function consistent with the relationships examined
above or to draw comparisons with turnover.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
784 QUARTERLY JOURNAL OF ECONOMICS
Instead, we must rely on less direct measures. As
the elasticities reported in Table II, the relationship with the
market wage is second only to voice in its effect on discipline rates.
But the dollar comparisons tell a more complicated story. A one
dollar increase in the wage premium over market rates at a
representative plant of 1,000 workers would lead to 16.5 fewer
disciplinary actions per year.15 The wage costs alone of that dollar
increase, however, total well over $2 million per year, or approximately
$121,000 per disciplinary action.
While this may appear to be an expensive means of reducing
shirking, it is difficult to know the total benefits associated with a
reduction in discipline cases. A company representative suggested
that the administrative costs alone associated with a dismissal can
run as high as $75,000.16 The costs of the incidents leading to
discipline vary depending on the cause but could be high, for
example, if the cause was sabotage. More important, wage premiums
may lead to a reduction in unobserved shirking-unobserved
because of the imperfect monitoring of employee behavior-and if
we assume that only a relatively small proportion of shirkers are
caught, this reduction could be substantial. The uncertainty
associated with these benefits makes their value vary with the risk
preferences of the employer. Finally, there may also be benefits
from the wage premium in addition to the reduction in shirking,
such as reductions in turnover, which are not measured here.
VI. CONCLUSIONS
Efficiency wage arguments rely on a largely untested relationship
between wage premiums and worker productivity. The arguments
above find evidence supporting the assertion that wage
premiums are associated with reductions in shirking as measured
by discipline rates. The wage premium appears to provide incentives
to avoid dismissal rather than leading to the selection of
workers less inclined to shirk. Factors such as the level of
unemployment in the local labor market and the seniority of the
work force also influence the costs associated with dismissal and
15. Part of the reason is that unions invariably file grievances against all
dismissals, and the administrative costs of a full grievance procedure are high.
Further, management loses a nontrivial share of these grievances and must pay
back wages for the period since the discharge, often a year or more. Dismissals may
also raise the experience rating costs associated with unemployment insurance. All
of these costs must be included in the expected costs of a discharge.
16. The new disciplinary rate = e /1 + e', where d = in (DISL/1 - DISL) +
(b.)(WPREM). The change in discipline rates varies with the initial level of the rate
and is evaluated here at the mean.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 785
are similarly related to the level of shirking. Finally, discipline
rates are also lower in plants where the union and management are
better able to resolve their problems. It is difficult to identify the
value of the reduction in shirking associated with a given wage
premium, but it does seem that there could be nontrivial returns
associated with a wage premium.
APPENDIX: SUPPLEMENTARY REGRESSION RESULTS FOR RATES OF DISCIPLINARY
LAYOFFS ACROSS PLANTS
Dependent variable: In (disciplinary rate/1 - disciplinary rate)
Weighted least squares
(1) (2) (3) (4) (5)
Intercpt - 1.48** -1.67 -2.23** -1.59 -1.331
(0.50) (0.54) (0.88) (1.16) (1.24)
WPREM -0.14** -0.25** -0.20* -0.25* -0.23*
(0.06) (0.12) (0.11) (0.13) (0.13)
UE -0.02 -0.006 -0.01
(0.04) (0.05) (0.047)
LAIDOFF -0.009* -0.009* -0.008* -0.01*
(0.005) (0.005) (0.005) (0.005)
SEN 0.007* 0.008** 0.005 0.01** 0.004*
(0.004) (0.003) (0.004) (0.004) (0.0075)
VOICE -0.14*** -0.15*** -0.14*** -0.14***
(0.52) (0.052) (0.053) (0.054)
ASSMBLY 1.03*** 1.015*** 1.09*** 1.08*** 0.94***
(0.17) (0.18) (0.19) (0.18) (0.23)
MICHIGAN 0.122 0.18 0.06 0.31 0.28
(0.18) (0.19) (0.25) (0.24) (0.25)
SOUTH 0.02 0.05 -0.17 -0.01 -0.067
(0.32) (0.32) (0.33) (0.36) (0.36)
UIBEN 0.00006 -0.00006 -0.00006
(0.00009) (0.00009) (0.00009)
WPREM'70 0.45 0.44
(0.46) (0.56)
SKILLa -0.005
(0.009)
SEN15b -0.005
(0.008)
SEE 0.62 0.62 0.62 0.62 0.61
n = 78 78 78 78 78
a. SKILL = percentage of plan
b. SEN15 = percentage of plan
Equation (5) contains all of th
* = significant at 10 percent l
** = significant at 5 percent
= significant at 1 percent le
THE WHARTON SCHOOL, UNIVERSITY OF PENNSYLVANIA
SCHOOL OF BUSINESS, UNIVERSITY OF KANSAS
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
786 QUARTERLY JOURNAL OF ECONOMICS
REFERENCES
Adams, J. S., "Toward an Understanding of Inequity," Journal of Abnor
Social Psychology, LXVII (1963), 422-36.
, and W. B. Rosenbaum, "The Relationship of Worker Productivity to Cognitive
Dissonance about Wage Inequities," Journal of Applied Psychology, XLVI
(1962), 161-64.
Akerlof, G., "Labor Contracts as Partial Gift Exchanges," Quarterly Journal of
Economics, XCIV (1982), 79-83.
_ "Gift Exchange and Efficiency Wage Theory: Four Views," American
Economic Review Proceedings, LXXIV (1984), 79-83.
, and J. Yellen, Efficiency Wage Models of the Labor Market (Cambridge:
Cambridge University Press, 1986).
Amemiya, T., and F. Nold, "A Modified Logit Model," Review of Economics and
Statistics, LVII (1975), 255-57.
Bliss, C. J., and N. Stern, "Productivity, Wages and Nutrition," Journal of
Development Economics, V (1978), 331-97.
Employment and Earnings (Washington, DC: Bureau of Labor Statistics, May
1982), Table C.
Freeman, R. B., "The Exit-Voice Tradeoff in the Labor Market: Unionism, Job
Tenure, Quits, and Separations," Quarterly Journal of Economics, XCIV
(1980a), 643-73.
, "Unionism and the Dispersion of Wages," Industrial and Labor Relations
Review, XXIV (October 1980b), 3-20.
"Highlights of State Unemployment Compensation Laws" (Washington, DC:
National Foundation for Unemployment Compensation and Workers' Compensation,
June 1983).
Hirsch, B. T., and W. J. Hausman, "Labour Productivity in the British and South
Wales Coal Industry," Economica, L (1983), 145-57.
Hirschman, Albert O., Exit, Voice, and Loyalty: Response to Decline in Firms,
Organizations, and States (Cambridge, MA: Harvard University Press, 1970).
Hulin, C. H., M. Roznowski, and D. Hachiya, "Alternative Opportunities and
Withdrawal Decisions: Empirical and Theoretical Discrepancies and an
Integration," Psychological Bulletin, XCVII (1985), 233-50.
Katz, L. F., "Efficiency Wage Theories: A Partial Evaluation," in S. Fischer, ed.,
NBER Macroeconomic Annual (Cambridge, MA: MIT Press, 1986).
Krueger, A., and L. H. Summers, "Reflections on the Inter-Industry Wage
Structure," in K. Lang and J. S. Leonard, eds., Unemployment and the
Structure of Labor Markets (Oxford: Blackwell, 1987).
Lawler, E. E. III, C. A. Koplin, T. F. Young, and J. A. Fadem, "Inequity Reduction
over Time in an Induced Overpayment Situation," Organizational Behavior
and Human Performance, III (1968), 253-68.
Leibenstein, H., Economic Backwardness and Growth (New York: Wiley, 1957).
Leonard, J. S., "Carrots and Sticks: Pay, Supervision, and Turnover," Journal of
Labor Economics, V (October 1987), s136-53.
Malcomson, J., "Unemployment and the Efficiency Wage Hypothesis," Economic
Journal, XCI (1981), 848-66.
March, James G., and Herbert A. Simon with the collaboration of Harold Guetzkow
Organizations (New York: Wiley, 1958).
Marx, K., Capital, Part VI, Chapter XXV, Section 3, "The Industrial Reserve
Army" (New York: Penguin Books, 1976).
Milkovich, George T., and Jerry M. Neuman, Compensation (Plano, TX: Business
Publishing, Inc., 1987).
Pencavel, John H., An Analysis of the Quit Rate in American Manufacturing
Industry (Princeton, NJ: Industrial Relations Section of Princeton University,
1970).
Phelps, Edmund S., Microeconomic Foundations of Employment and Inflation
Theory (New York: Norton, 1970).
Pritchard, R. D., M. D. Dunnette, and Jorgenson, D. O., "Effects of Perceptions of
Equity and Inequity on Worker Performance and Satisfaction," Journal of
Applied Psychology Monograph, LVI (1972), 75-94.
Raff, D. M. G., and L. H. Summers, "Did Henry Ford Pay Efficiency Wages?"
Journal of Labor Economics, V (October 1987), s57-s87.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms
INTERPLANT TEST OF THE EFFICIENCY WAGE HYPOTHESIS 787
Salop, S., "A Model of the Natural Rate of Unemployment," American Economic
Review, LXIX (1959), 117-25.
Shapiro, C., and J. E. Stiglitz, "Equilibrium Unemployment as a Worker Discipline
Device," American Economic Review, LXXIV (June 1984), 433-44.
Smith, A., An Inquiry into the Nature and Causes of the Wealth of Nations (Dublin:
Whiteshire, 1776).
Smith, P. C., L. M. Kendall, and C. L. Hulin, The Measurement of Satisfaction in
Work and Retirement (Chicago: Rand-McNally, 1969).
Solow, R., "Another Possible Source of Wage Stickiness," Journal of Macroeconomics,
1(1959), 79-82.
Stiglitz, J. E., "Efficiency Wage Hypothesis, Surplus Labour and the Distribution o
Income in L.D.C.'s, " Oxford Economic Papers, XXXVIII (1976), 185-207.
,and A. Weiss, "Incentive Effects of Terminations: Applications to the Credit
and Labor Markets," American Economic Review, LXXIII (December 1983),
912-27.
Swigart, R. P., ed., Managing Plant Closings and Occupational Readjustments: An
Employer's Guidebook (Washington, DC: National Center on Occupational
Readjustment, 1984).
Taylor, A. J., "Labor Productivity and Technological Innovation in the British Coal
Industry," Economic History Review, 2nd Series XIV (1961), 48-70.
Topel, R. H., "Local Labor Markets," Journal of Political Economy, XCIV (1986),
S111-S43.
U. S. Department of Labor, Employment and Training Administration, "Com
son of State Unemployment Insurance Laws" (Washington, DC: January
1990).
Vroom, V. H., Work and Motivation (New York: John Wiley and Sons, 1964).
Walters, R. H., The Economic and Business History of the South Wales Steam Coal
Industry, 1840-1914 (New York: Arno Press, 1977).
Weiss, A., "Job Queues and Layoffs in Labor Markets with Flexible Wages,"
Journal of Political Economy, LXXXVIII (1980), 526-38.
This content downloaded from 147.251.55.59 on Fri, 06 Apr 2018 11:07:49 UTC
All use subject to http://about.jstor.org/terms