The Impact of the Percentage Organized on Union and Nonunion Wages
Author(s): Richard B. Freeman and James L. Medoff
Source: The Review of Economics and Statistics, Vol. 63, No. 4 (Nov., 1981), pp. 561-572
Published by: The MIT Press
Stable URL: http://www.jstor.org/stable/1935852
Accessed: 29-03-2018 09:35 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
The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The
Review of Economics and Statistics
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
THE IMPACT OF THE PERCENTAGE ORGANIZED ON UNION
AND NONUNION WAGES
Richard B. Freeman and James L. Medoff*
T HE impact of unions on wages is likely to
depend on the extent to which they organize
workers in the relevant product market.1 As the
organization in a market increases, the opportunity
for substituting nonunion for union products
will be reduced, lowering the elasticity of demand
for organized workers and the potential
loss of employment for a given wage increase. As
a result, the wages of union workers are likely to
be higher, all else the same, the greater the percentage
organized. The wages of nonunion
workers may also be influenced by the extent of
organization, though the direction of the effect is
not clear. On the one hand, union wage gains due
to greater coverage may induce increases in
nonunion wages because of the threat of organization
and/or because of shifts in demand favoring
nonunion producers brought about by the
increased relative cost of union labor. On the
other hand, the supply of labor to nonunion firms
may increase as a result of reduced employment
in the union sector, which would most likely
depress nonunion wages. Whether the threat plus
demand effect or the supply effect dominates is
an empirical issue. The impact of the percentage
organized on the union wage differential (the difference
between the natural logarithms of union
and of nonunion wages) depends on the relative
magnitudes of the likely positive impact on union
wages and the positive or negative impact on
nonunion wages.
This paper seeks to disentangle the relation
between the percentage of workers organized in
a product market and the wages received by
union workers and by nonunion workers. In contrast
to most of the literature on the union wage
effect, which either relates some average of
wages in an industry to the percentage organized
or which relates the wages of individuals to their
membership in a union, our analysis examines
the impact of the percentage organized on the
compensation of union labor and nonunion labor
taken separately.2 By relating the wages of
unionized workers to the percentage covered by
collective bargaining in the relevant product
market, we estimate directly the extent to which
unionized workers in highly organized markets
receive higher wages than unionized workers in
less organized settings. By relating nonunion
wages to the percentage covered, we provide
direct estimates of the extent to which, as a result
of threat, demand, and supply effects, nonunion
workers in highly organized markets receive
higher or lower wages than nonunion workers in
less organized industries or areas.
Two sets of data are used in the study: information
on individuals from the 1973, 1974, and
1975 May Current Population Surveys (CPS),
which contain data on usual weekly earnings,
usual weekly hours, union membership status,
and key personal characteristics; and information
on establishments from the Bureau of Labor
Statistics' 1968, 1970, and 1972 Expenditures for
Employee Compensation Surveys (EEC), which
contain data on the components of compensation,
labor hours, collective bargaining coverage,
and some relevant establishment characteristics.
The availability of both individual and establishReceived
for publication April 24, 1980. Revision accepted
for publication May 27, 1981.
* Harvard University.
We have benefited from the comments of K. Abraham,
C. Brown, H. G. Lewis, and L. Summers. We are most
grateful to G. Bialecki, J. Fay, C. Ichniowski, L. Nelson, M.
Van Denburgh, L. Wilson, and J. Zax for research assistance.
The study has been supported by the National Science Foundation
(Grant APR 77-16279) and the National Bureau of
Economic Research (under its program for research on labor
economics). Any opinions expressed are not necessarily
shared by the individuals who have aided us, NSF, or NBER.
I Throughout our theoretical discussion we refer to a product
market, which is the appropriate unit of observation for
an analysis of the relationship between percentage organized
and wages. However, in the empirical work we focus on
either a 3-digit Standard Industrial Classification or Census
industry (in the manufacturing analysis) or a Current Population
Survey state group (in the construction analysis). Unfortunately,
the data used do not permit a closer correspondence
between the theoretical and empirical parts of our study.
2 For an early attempt to disentangle this relation, using
average wages in an industry and percentage organized, see
Rosen (1969). For more recent related analyses, see (in alphabetical
order) Donsimoni (1978), Hendricks (1975), Kahn
(1978), and Lee (1978). For a trenchant treatment of the
analysis, see Lewis (1980).
[ 561 1
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
562 THE REVIEW OF ECONOMICS AND STATISTICS
ment data, each of which has weaknesses and
strengths, provides a valuable check on our
findings. Because unionized workers are primarily
blue-collar labor, we restrict our attention to
production or nonoffice workers. Because of the
distinct features of wage-setting in the public sector,
we further limit our focus to private wage
and salary workers. Finally, since the analytic
model relates collective bargaining coverage to
wages through the product demand curve, we let
the nature of our analysis depend on the nature of
the relevant product markets. We compare union
organization and wages across industries in manufacturing,
where as a first approximation product
markets can be taken to be national, but
compare organization and wages across geographic
areas in construction where product
markets tend to be local in nature.
The major finding of our research is that in
manufacturing the percentage organized in a
product market has a strong positive association
with the wages of union (production) workers,
but either no association or a weak positive association
with the wages of nonunion (production)
workers. As a result, within manufacturing,
the union-nonunion wage differential in a product
market is positively related to the extent of organization.
For construction, the results appear to
be similar, though sensitive to specification.
I. The POW Relationships
There are three basic reasons for expecting the
percentage organized in a market to be positively
related to the wages of union workers. First, it is
likely that the greater is the union coverage of a
sector, the lower will be the elasticity of demand
for the product of unionized firms (since there
are fewer nonunion competitors) and, as a consequence,
the lower will be the elasticity of derived
demand for labor. If, as is most probable,
unions are concerned with the number and employment
of their members, as well as with their
members wages, they can be expected to press
for higher wages in markets with a relatively high
percentage organized, and hence, with a relatively
low labor demand elasticity. Second, a
positively sloped percentage organized/wvage
(POW) curve for covered workers might also be
observed because unions have located only in
sectors with low elasticities of demand for labor
or because, no matter where they have organized,
they have survived only where the demand
elasticity is low. In these scenarios, a low
wage elasticity of demand causes a high percentage
organized and, at the same time, allows the
union to obtain high wages; the percentage organized
is an indicator of an initially low elasticity
of demand for labor rather than a "cause" of
a low elasticity. Third, to the extent that unions
in heavily organized sectors are able to reduce
the substitutability between production labor and
other factors, along the lines suggested by
Freeman and Medoff (forthcoming(a)), a positive
percentage organized wage relationship may
result. In this case the causality is from percentage
organized to the elasticity of substitution
(rather than to the product demand elasticity)
and then to the elasticity of demand for labor and
to the wage.
The present study focuses primarily on the
path from percentage organized to the elasticity
of product demand to the elasticity of the derived
demand for labor to union wage demands. We
control, albeit imperfectly, for the possibilities
that the union POW schedule is positively sloped
because the demand elasticity for labor is lower
to start with in the union sector or is made lower
by union efforts to restrict input substitution; this
is done either by holding fixed for a set of variables
capturing potentially relevant industry
characteristics or by studying the impact of percentage
organized within a given industry. Despite
our efforts, however, the possibility that the
observed POW relationships for union workers
to some extent reflect the locus of unionism and
the degree of factor substitution in different settings
cannot be dismissed.
Formal Analysis
The relationship between the extent of organization
in a market and potential union wage gains
can be discussed formally as follows.3 Let -q, > 0
be the elasticity of demand for the final product
I While the model to be presented is quite simplistic in that
it abstracts from union-management bargaining and other factors
likely to determine wage rates under unionism, it does
capture the critical fact that, all else the same, unions can
increase wages at a lower price in terms of lost employment
where the product demand elasticity is low. For similar discussions
of the determinants of union wage gains, see the
articles by Segal (1964) and Levinson (1967).
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
UNION ORGANIZATION AND WAGES 563
in the union sector, P be the percentage
organized/100, a be labor's share of total cost in
unionized firms, o- > 0 be the elasticity of substitution
between unionized labor and other factors,
and -ql > 0 be the elasticity of demand for
union members with respect to the wage. It is
well known that, if the supply of capital is
infinitely elastic to a sector,
X,l = a-q, + (I - a) (T. (1)
Under the highly plausible assumption that the
demand elasticity for the products of unionized
firms is a function of the extent of coverage
[d-q,/dP < 0], and holding (T constant for
simplicity, we get
d-,/dP = ad-q/dP = afl'x < 0. (2)
The union is assumed to maximize a standard
convex utility function
U = U(ln W, ln E), (3)
where W is the wage rate for union members and
E is their employment, which together represent
a point on the demand curve for union labor.
Along this curve, d ln E = - nd In W. Thus, the
wage which maximizes (3) must fulfill the condi-
tion
J1 / U2 = (4)
where U1 and U2 are the partial derivatives of the
maximand with respect to ln W and In E, respectively.
Since U1 and U2 depend only on the wage
rate, we can derive the relationship between
ln W and m} by first writing the expression
U,IU, = +i(ln W) with i/' assumed < 0, (5)
and then taking its inverse, which gives
In W = 4F-l( U1/U2) = iIr'().- (6)
Differentiating (6) we see that
d In W/dP = (dip- 1/d )(dTh/dP)
= a-q f'ljf > . (7)
The dependence of the slope of the POW
schedule for union workers on the relation between
the percentage organized and the demand
elasticity for the output of organized firms (fl'.)
brings the product market to the fore of the analysis.
Under reasonably general assumptions
about the extent of product market competition,
it can be shown that ' will, as posited, be
negative. Consider, for example, a product market
in which commodities differ across firms because
of either the location of customers in regional
or local markets or small differences in the
firms' commodities and where there are N
equally sized firms in the industry, each of which
has a cross-elasticity of demand of y with every
other firm. Ignoring, for simplicity, the effect of
changes in wages and prices on total industry
output, the elasticity of demand for the output of
the organized sector will depend on the number
of firms to whom output can be lost, and thus on
the organized share of the market. In the example
under consideration, the relation is a simple
linear curve, with the elasticity of demand for the
output of the organized sector falling proportionately
to the fraction of enterprises that are
organized. For example, if the fraction organized
in the relevant product market increases from 0.3
to 0.8, the elasticity of demand in the organized
sector will decline by 0.5Ny.
A negative effect of coverage on the elasticity
of product demand, and hence a positive effect
on union wage rates, is also to be expected in
oligopolistic situations, although the complexity
of price-setting precludes any definitive analysis.
If all firms in the sector alter prices when the
"leading firms' make changes, the key issue
may be whether or not the union has organized
the leaders. For example, a large change in coverage
which fails to alter the union status of the
leaders might very well have less impact on the
product demand elasticity facing the union sector
than a small change that alters the union status of
the leading firms. Overall, increases in coverage
should increase the union wage rate, but the precise
relation cannot be determined.
The most complex case to consider is that of
homogeneous goods produced for a perfectly
competitive market, defined as one in which
firms with the same U-shaped cost curve can
enter freely. In the long-run static equilibrium,
unless there were 100% coverage (or equivalently
unless nonunion firms paid union wages to
forestall organization), the elasticity of demand
for output of the organized sector would be infinite,
implying that there would be no union
wage effect, in the absence of, for example,
offsetting productivity gains of the type found by
Brown and Medoff (1978), Frantz (1976). and
Clark (1978), or quasi-rents due to differences in
costs not associated with uniofiism. Assuming no
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
564 THE REVIEW OF ECONOMICS AND STATISTICS
threat effects and instantaneous market adjustment,
the POW curve for union workers would
be discontinuous, with wage rates unrelated to P
until complete coverage (including the automatic
coverage of new firms) was achieved, at which
point they would jump sharply to a value dependent
on the demand elasticity for labor in the
sector. However, since going out of business is
not instantaneous and a given wage effect can be
maintained with less than complete coverage by
organizing additional enterprises, the actual relationship
may be much smoother. Moreover, the
greater the percentage organized, the smaller will
be the degree of potential nonunion competition
faced by union enterprises and, hence, the
slower will be the likely speed at which covered
establishments go out of business. For this reason,
the elasticity of demand for union labor is
likely to be smaller the greater the percentage
organized. Hence, we can expect an upward
sloping POW curve for union workers in the
competitive environment posited.
In all of our calculations we take the percentage
organized as the independent variable affecting
wages. While there is almost always some
degree of joint determination of economic variables,
we believe this is the most sensible line of
causality for two reasons. First, there is no basis
in theory or fact (that we know of) for the expectation
that wage levels would be positively associated
with union organization. Second, even
though unions have an incentive to organize sectors
where, for some reason, there are substantial
rents and, hence, the potential of bargaining
for a large wage increase without a reduction in
employment, management in these settings is
likely to have the resources (as well as the will)
to fight unions' efforts.
NotilIilo)n VWageS
The effect of percentage organized on nonunion
wages in a given product market (with a
given standard production function and a given
price of capital set by the overall economy) is
more complicated because there are some factors
operating to produce a negative POW relation
and others operating to produce a positive POW
relation. On the one hand, union wage gains are
likely to lead to a reduction of employment in the
union sector and a potential increase in the supply
of nonunion workers. On the other hand,
union-induced increases in cost will shift product
demand toward nonunion firms, raising the demand
for nonunion workers. Assuming for the
moment that labor must be employed in either
the union or nonunion sector of the product market
under analysis (but that capital can move to
other product markets), the net effect of the two
forces on nonunion wage rates is likely to be
negative, since the increase in the supply of labor
can be expected to exceed the increase in the
demand. This is because substitution both among
inputs and among products operates on the supply
side to displace union workers and augment
the supply of nonunion labor, while only output
substitution operates to raise the demand for
nonunion workers.4
While the result just stated holds when labor is
tied to one product market, it most likely will not
apply when, because of differences in the geographic
locale of organized and nonorganized
production, the supply effect is not operative at a
product market level. Since displaced union
workers generally do not end up producing the
same commodity, the demand effect will usually
dominate, putting an upward pressure on nonunion
wage rates which does not depend on the
"threat" of unionization.
The likelihood that nonunion wages are affected
by unionism independently of the demand
and supply effects outlined above creates a still
I Formally, under our assumptions, the percentage decline
in employment in the union sector of the industry is - (a-ox +
(1 - a)o-) W, where a is labor's share of output in the
union sector, i. is the price elasticity of output demand f
by unionized firms, aT is the elasticity of substitution in the
union sector, and W is the percentage increase in union
wages. Therefore, the percentage increase in the supply of
labor to the nonunion sector will be - ( ar, + ( I
a)o--)(E`/E`)W, where E'l is labor employed in the union sector
and Ell is labor employed in the nonunion sector of the
industry. Assuming that qx- < o and that an increase in the
union sector output price simply shifts demand to the nonunion
sector, the cross-price elasticity of demand for nonunion
output with respect to the union price will be equal to the own
price elasticity of demand for union output, ij\. Thus, the
percentage increase in the demand for labor in the nonunion
sector will be aw(EI'/E") W. Note that this analysis differs
from that offered by Johnson and Mieskowski (1970). In their
2-sector general equilibrium model, the nonunion sector must
employ both the displaced labor and the displaced capital,
which causes the price of capital, as well as the price of labor,
to adjust. In our partial equilibrium analysis, displaced capital
may be absorbed by other industries as well as by the nonunion
sector of the "small" industry under consideration;
hence capital's price is unaffected by changes in that indus-
try.
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
UNION ORGANIZATION AND WAGES 565
more complex situation. When organization is
extensive, it is highly possible that nonunion
workers will observe higher union wages and
seek similar rates of pay. By the "wage relativity"
hypothesis, their supply price will increase
and their effective supply (hours weighted by
productivity) can be expected to fall unless they
are given comparable wages. More importantly,
perhaps, the probability of organization is likely
to increase, which is likely to induce nonunion
employers to pay higher wages.
II. Econometric Specification and Data
To isolate the effect of the extent of organization
on union wages and on nonunion wages it is
necessary in the regression analysis to control for
other potentially important factors, such as age
or experience, skill, years of schooling, residential
location, sex and race. Because the prime
independent variable in the manufacturing analysis,
the percentage organized, relates to the environment
of firms in an industry, it is particularly
important to control for other related characteristics--establishment
size, concentration ratio,
and so on-which may be correlated with both
the percentage organized and wage rates. With
all of these factors in mind, we write the basic
regression model for the manufacturing analysis
as
Wij = aP; + UNITIj + cINDj + Uij, (8)
where Wij is the ln(wage) of the ith worker (establishment)
in industry j; Pj is the percentage organized
in the jth industry/100; UNIT1j is a vector
of worker or establishment characteristics
(depending on the data set); INDj is a vector of
industry characteristics; Uij is an independently,
identically, and normally distributed residual
with mean 0; and where the sample includes either
union or nonunion production (or nonoffice)
workers.
The particular control variables available in
the two data sets under study differ substantively;
the Current Population Surveys contain
detailed information on the characteristics of individuals
but not on the establishments in which
they work; the Expenditures for Employee
Compensation Surveys contain the opposite.
As stated above, a key factor in the estimation
of the effect of P on W is the extent to which the
relevant industry factors are held fixed. We attack
this problem in three ways.
First, in the manufacturing sector regressions
we include some industry-level variables, such as
the four-firm concentration ratio, average firm
size, and the injury rate in the industry. Inclusion
of these variables guarantees that P is not "picking
up" their effect on wages.
Second, in the manufacturing analysis we
allow for more general industry effects by including
2-digit Standard Industrial Classification
(SIC) industry dummies. To the extent that the
2-digit SIC dummies capture some of the differences
between more detailed industries, these
controls reduce the chance that omitted industry
factors bias the estimated coefficient of P.
Third, in the analysis of the construction sector
we control for industry factors by examining
the effect of coverage on wages across geographic
areas within one industry, whose relevant
product markets can be reasonably defined
on a regional basis. While analyzing geographic
areas within a given industry solves the problem
of uncaptured industry factors, this approach
also has a potential problem: organization may
be correlated with omitted geographic variables
that influence wages, leading to biased estimates
of union and nonunion POW effects.
As emphasized in the preceding discussion, an
omitted industry variable which is partially correlated
with both percentage organized and wages
will bias estimated POW associations. However,
it is important to note that (other variables held
constant) if the omitted variable had the same
effect on the wage rates (in ln units) of union and
nonunion workers and the same relationship with
percentage organized in the union and nonunion
samples, the difference between the estimated
effects of coverage on union and nonunion wages
(i.e., the estimated effect of coverage on the
wage differential) would be unbiased. This is because
the omitted factor would bias the estimated
coverage coefficients in the union and nonunion
regressions by the same absolute amount, so that
differencing would eliminate the bias term.
Whether the unobserved variable has the same
effect on the wages of union and nonunion workers
and the same relationship with percentage
organized is an issue which cannot, by definition,
be answered with the data. However, with a
sufficiently large set of industry controls in the
union and nonunion regressions, it is difficult to
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
566 THE REVIEW OF ECONOMICS AND STATISTICS
think of reasons why the two necessary conditions
for unbiased differences in POW effects
would be grossly violated.5
The Data
To estimate the effect of percentage organized
on wages, we have examined two sources of
data: the May CPS for 1973-75 and the EEC for
1968, 1970, and 1972.6 Because the EEC establishment
data lack demographic information, we
used the CPS responses to calculate the mean
values of selected variables for union and nonunion
production (blue-collar plus service)
workers in each 3-digit Census industry, recoded
the variables (as well as possible) to the 3-digit
SIC scheme used on the EEC tapes, and added
the CPS variables for either the organized or
unorganized sector of a 3-digit SIC industry to
each EEC establishment's record in accordance
with the establishment's 3-digit SIC industry and
the collective-bargaining status of its nonoffice
employees. The variables (percentage male, percentage
white, mean schooling completed, mean
of age minus schooling completed minus five,
and the means of some other potentially relevant
characteristics) were derived from weighted
counts of private wage and salary production
workers represented on the 1973-75 May CPS
file.
The principal independent variable in the
study, the percentage of nonoffice workers organized,
was estimated for 3-digit 1967 SIC manufacturing
industries from 1968, 1970, and 1972
EEC information regarding whether a majority of
each responding establishment's nonoffice employees
were covered by union-management
agreements. These estimates are described in detail
in Freeman and Medoff (1979).
Variables measuring other industry characteristics
that might be correlated with both wage
rates and percentage organized were added to
one or both of the union and nonunion manufacturing
data sets. They are the following:
-Average size of firm (added only to the CPS
records, since the EEC provides information on
establishment size), to take account of the wellknown
positive effect of size on wages (Masters,
1969) and the potential positive correlation between
extent of unionism and size. This variable
was measured as the value of shipments in 1972
in each 3-digit SIC industry divided by the number
of firms in the industry in 1972. It was calculated
with data from U.S. Department of Commerce,
table 5.
-Concentration ratio, to take account of the
possible impact of concentration on wages
(Weiss, 1966) and the likelihood that unionization
is higher in more concentrated sectors. This
variable was measured as a weighted average of
the fractions of shipments in 1972 accounted for
by the four leading firms in the 4-digit SIC industries
composing a 3-digit SIC industry. It was
derived with data from U.S. Department of
Commerce, table 5 and with the scaling factors
used by Weiss to correct for the market power of
firms selling in closed local markets.
-Injury and illness rate, to control for the
potential effect of dangerous work conditions on
wages and the possibility that this rate is associated
with the extent of unionism. A variable,
equal to the mean number of lost workdays due
to injury and illness per full-time worker in 1972,
1973, and 1974 in each 3-digit SIC industry, was
calculated with data from U.S. Department of
Labor (1974, 1975, and 1976b).
The industry control variables were bridged as
well as possible to the industrial classifications
on the CPS and EEC tapes. The details of the
construction of variables are available on request,
as are the data employed in our analyses.
A priori one cannot readily determine the possible direction
of the bias. Longitudinal studies of the union wage effect
show smaller differentials than do cross-sectional studies,
which suggests that the cross-sectional effects may be upwardly
biased due to unobserved personal characteristics. It
might be inferred from this fact that our cross-sectional POW
effects are biased upward as well. However, we do not believe
that this problem can be addressed in an acceptable
fashion in the absence of longitudinal data; "correcting" for
sample selectivity with a cross-sectional data set like the one
we employ by adding an inverse Mills ratio or fitting a system
of equations, including a unionization equation, does not
appear to yield useful results for analysis of union effects. See
Freeman and Medoff (forthcoming (c)).
6 The CPS and EEC Surveys are described in detail in U.S.
Department of Labor (1976a), pp. 5-23 and pp. 175-183,
respectively. Note that because of the nature of the CPS and
EEC sampling techniques, pooling across years leads to replication
of persons or establishments, which could bias our
standard errors. However, our work and that of Lewis (1980)
yield results comparable to those in the text for the different
years taken by themselves. The insensitivity of the results
from one year to the next, despite significant changes in
macroeconomic conditions, suggests that the POW effects
are relatively invariant to economic conditions, at least during
the period under analysis.
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
UNION ORGANIZATION AND WAGES 567
III. Empirical Results
Table 1 presents estimates, based on the
1973-75 May CPS, of the impact of percentage
organized (covered by collective bargaining) in
an industry on the usual hourly earnings of manufacturing
production workers who are union
members and on the earnings of comparable
nonmembers. The first and second columns of
numbers give the means and standard deviations
of the relevant variables in the union and nonunion
samples; the third and fourth columns record
for the two samples the regression
coefficients and standard errors on percentage
organized and on three other industry-level variables.
The other controls in the equation are
listed in the table.
The principal result demonstrated in table I is
that percentage organized has a sizable positive
effect on union but not on nonunion wages,
which implies an increasing wage differential
with coverage. With the sample of union members,
the estimated coefficient on organization is
0. 153 compared to 0.004 with the sample of nonmembers.
This implies, for example, that the
union wage effect in a manufacturing industry
with 80% organized is likely to be about 9 percentage
points higher than in an industry with
only 20% organized. At a given set of means,
with 20% organized, the union wage effect is
about 5 percentage points; with 80% organized
the comparable effect is about 14 points.
Table 1 also demonstrates that firm size is positivel.y
and significantly associated with wages in
both union and nonunion settings; the estimated
firm size coefficient is smaller under unionism.
The injury and illness rate has roughly the same
statistically significant positive effect on the earnings
of union members and nonmembers. The
concentration ratio does not appear to have a
TABLE 1.-ESTIMATES OF THE PERCENTAGE ORGANIZED WAGE (POW) RELATION FOR
MANUFACTURING PRODUCTION WORKERS; 1973-75 MAY CPS INDIVIDUAL-LEVEL DATA
Dependent Variable:
ln(Usual Hourly Earnings)
Mean Coefficient
(S.D.) (S.E.)
Union Nonunion Union Nonunion
Sample Sample Sample Sample
Independent Variables:
Industry Level
Percentage of industry's nonoffice workers 0.687 0.546 0.153 0.004
covered by collective bargaining/100 (0.208) (0.199) (0.024) (0.024)
ln(shipments perfirm) 1.526 0.984 0.025 0.044
(1.265) (1.089) (0.006) (0.008)
Four-firm concentration ratio 0.430 0.357 -0.012 0.045
(0.185) (0.151) (0.031) (0.040)
Injury rate (days per worker) 0.729 0.675 0.041 0.037
(0.380) (0.439) (0.014) (0.016)
Individual Levela
Industry dummies (20); state dummies (28);
occupation dummies (4); survey dummies (2);
SMSA-size dummies (4); marital status dummies
(3); sex dummy; race dummy; schooling
completed; age - schooling completed - 5 and its
square; number of dependents; constant yes yes
R2 0.421 0.471
SEE 0.270 0.325
N 10,679 11,204 10,679 11,204
Mean(S.D.) of 1.451 1.190
Dependent Variable (0.354) (0.445)
t The four Standard Metropolitan Statistical Area (SMSA)-size dummies
population as of 1970 was -1..000,000( < 1,0000()( but -500,000; <500,
SMSA not identified on the CPS file; or did not live in an SMSA. The ma
married, spouse absent; widowed or divorced; or never married.
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
568 THE REVIEW OF ECONOMICS AND STATISTICS
systematic impact on the wages of either union or
nonunion manufacturing production workers; the
finding with the union sample runs counter to our
expectation that concentration affects the product
demand elasticity, and, hence, the labor demand
elasticity, leading to higher wage rates in
unionized firms.7
We have also conducted comparable analyses
using a percentage organized variable constructed
from CPS data on individuals' union
membership status. The results obtained were
qualitatively similar to those given in table 1.8
We focus on the results with the EEC percentage
organized measure because we believe coverage
is a better indicator of union strength than mem-
bership.
Table 2 provides estimates of the impact of
percentage organized (covered by collective bargaining)
on the compensation (wages plus all
voluntary fringes) per hour and "direct payments"
(wages plus those voluntary fringes which
are paid directly to workers rather than to a fund)
per hour of nonoffice employees in manufacturing
establishments in which a majority of the
blue-collar workforce was (at the time of the
relevant EEC Survey) covered by unionmanagement
agreements and in comparable establishments
in which a majority was not covered.
In the case of hourly compensation, the
estimated coefficient of proportion unionized is
0.157 for covered establishments and 0.047 for
ones that are not covered. These estimates imply,
for example, that the union compensation
effect for manufacturing nonoffice employees is
about 6 percentage points larger in an industry
with 80% organized than in one with only 20%
organized. In the case of direct payments per
hour worked, the estimated union POW effect is
0.125 and the comparable nonunion effect is
0.038.9 These estimated POW relationships indicate
that the impact of industry organization on
the union-nonunion direct payments differential
is smaller than suggested by the CPS data. This
divergence might reflect the different unit of observation
in the two surveys, the fact that a small
number of very large establishments representing
15% of total manufacturing employment were
excluded from the EEC tapes to preserve
confidentiality, definitional differences in various
key variables, and/or differences in the dates of
the two surveys. Nevertheless, with the EEC
data, the union direct payments effect in percentage
points for production workers in the
average manufacturing industry increases by
about 0.08 of a point for each point increase in
the percentage of the industry's production
workers who are unionized. At a giv,en set of
means, with 20% organized, the estimated
union-nonunion direct payments differential is 5
percentage points; at 80% organized the comparable
differential is 10 points.
The estimated coefficients on the industry control
variables with the EEC samples are for the
most part quite imprecise. The estimates clearly
improve when we turn from the variables derived
at the industry level to those that exist on an
establishment basis. Firm size, measured as ln
(non-office hours worked in the establishment),
has a positive effect on wages (and, similarly,
7 Numerous additional experiments were conducted with
the CPS data. For example, in one set of regressions similar
to those presented, a capital/labor hours variable was added,
in another the number of occupational dummies was increased
from 4 to 57, and in a final one an imports to domestic
consumption ratio was included. While the inclusion of the
imports variable seemed like the most important change,
since it was expected that this variable would be related to an
industry's product demand elasticity, the variable did not
perform in accordance with the theory set out in section I.
However, none of the modifications just described had a
meaningful impact on the main results given in table 1. In
addition, separate regressions, identical to those in table 1,
were fit for white males, black males, white females, and
black females. For each group except black males, the conclusions
concerning the impact of percentage organized on
union wage rates, nonunion wage rates, and the wage differential
were roughly the same as those drawn from the regressions
presented in table 1. In the case of black males, percentage
organized had a small insignificant effect on union as
well as nonunion wage rates, for reasons which are not apparent
to us.
8 Regressions otherwise identical to those in table 1 yielded
the following coefficients (standard errors) on the extent of
membership variable: 0.238 (0.041) with the union sample,
and -0.013 (0.036) with the nonunion sample. The CPS
membership percentages used in these regressions are discussed
at length in Freeman and Medoff (1979).
9 We also estimated models in which the dependent variable
was ln(straight-time pay per straight-time hour) whose
mean (standard deviation) was 1.403 (0.252) with the union
sample and 1.166 (0.296) with the nonunion sample. With this
dependent variable, the estimated POW coefficient (standard
error) was 0.140 (0.028) for unionized establishments and
0.063 (0.039) for those that are nonunion. While these results
are consistent with our primary conclusion that the unionnonunion
compensation differential grows substantially with
the percentage organized, it should be noted that they indicate
a non-trivial positive relationship between the extent of
organization and the nonunion hourly straight-time pay component
of hourly compensation.
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
UNION ORGANIZATION AND WAGES 569
total compensation) with both the uniou and
nonunion samples. A variable equal to the ratio
of hours worked by nonoffice employees to hours
worked by all employees in the establishment
assumes a negative and very significant estimated
coefficient in both the samples.10 This result
suggests either that blue-collar jobs in
nonoffice-employee intensive establishments are
relatively low skill (e.g., assembly-line jobs) or
have relatively desirable nonpecuniary charac"I
With the nonoffice hours/total hours variable excluded
from the table 2 regressions, the estimated coefficient (standard
error) of percentage organized was 0.140 (0.031) in the
union hourly compensation regression and 0.082 (0.041) in
the comparable nonunion regression. For hourly direct payments,
the effect was 0.110 (0.029) with the union sample and
0.069 (0.040) with the nonunion sample. We also experimented
with using a percentage union member variable
from the CPS as our measure of unionism. The coefficients of
this variable, in analogues to the table 2 regressions, were: in
the In(cornpensation) model, 0.192 (0.057) with the union
sample and -0.116 (0.086) with the nonunion sample; and in
the ln(hoiurlv direct payments) model, 0.141 (0.054) with the
union sample and -0. 111 (0.084) with the nonunion sample.
We also fit a number of additional models: one controlled for
the fraction of production employees (in the union or nonunion
sector of the relevant industry) in the five broad occupational
groups used in creating the occupation-group dummy
variables included in the table 1 regressions; another held
constant the fraction of production employees in the five
SMSA-size categories, the fraction in the four marital status
groups used in deriving table 1, and mean number of dependents;
two others included the capital-labor hours ratio or the
ratio of imports to domestic consumption. The estimated
POW relationships under each of these specifications were
roughly the same as those indicated by the table 2 results.
TABLE 2.-EsTiMATES OF THE PERCENTAGE ORGANIZED WAGE (POW) RELATION
FOR MANUFACTURING NONOFFICE WORKERS:
1968, 1970, AND 1972 EEC ESTABLISHMENT-LEVEL DATA
Dependent Variables:
ln(Hourly ln(Hourlv
Compensation&) Direct PaYments )
Mean Coefficient
(S. D.) (S. E.)
Union Nonunion Union Nonunion Union Nonunion
Sample Sample Sample Sample Sample Sample
Independent Variables:
Industry Level
Percentage of industry's nonoffice workers 0.714 0.510 0.157 0.047 0.125 0.038
covered by collective bargaining/100 (0.214) (0.224) (0.030) (0.040) (0.028) (0.039)
Four-firm concentration ratio 0.380 0.339 -0.050 -0.096 -0.058 -0.103
(0.177) (0.143) (0.033) (0.062) (0.031) (0.060)
Injury rate (days per worker) 0.715 0.638 -0.006 0.038 0.018 0.036
(0.366) (0.400) (0.026) (0.040) (0.024) (0.039)
Percentage male; percentage white; mean
schooling completed; mean of age schooling
completed - 5 and its square yes yes yes yes
Establishment Level
ln(nonoffice hours uvorked) 13.364 12.204 0.037 0.044 0.031 0.037
(1.534) (1.749) (0.003) (0.004) (0.003) (0.004)
Nonoffice hours worked/total hours worked 0.776 0.811 -0.405 -0.400 -0.377 -0.358
(0.148) (0.170) (0.033) (0.042) (0.031) (0.041)
Industry dummies (20); region dummies (3);
survey dummies (2); constant yes yes yes yes
R2 0.451 0.498 0.449 0.476
SEE 0.208 0.234 0.195 0.228
N 2,576 1,465 2,576 1,465 2,576 1,465
Mean (S.D.) of Dependent 1.511 1.189 1.427 1.149
Variable (0.279) (0.327) (0.261) (0.312)
In 1967 CPI-deflated dollars. Hourly compensation is defin
sick leave, and civic and personal leave; severance pay; an
and retirement plans, vacation and holiday funds, severanc
plans) to (total hours paid for minus vacation hours min
variable is constructed by substituting total direct paym
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
570 THE REVIEW OF ECONOMICS AND STATISTICS
teristics or that office and nonoffice workers are
substitutes. Since under the substitution interpretation
the nonoffice labor/total labor variable
does not belong in the regression models, we
reestimated the table 2 equations with it excluded.
While the conclusions concerning the effect
of P on union wages and the wage differential
drawn from this estimation are not very different
from those based on table 2, the regressions
indicate that P has a significant positive
effect on the hourly compensation and hourly
wages of nonunion as well as union employees.
In sum, analysis of both CPS and EEC data for
production workers in manufacturing lends general
support to the notion of an upward sloping
POW curve in the union sector of a market, reveals
either no spillover or a small positive association
between coverage and compensation for
nonunion workers, and demonstrates that the
union-nonunion compensation differential grows
substantially with growth in the extent of union
organization.
Construction Industry
Are the observed POW relationships specific
to the manufacturing sector where product markets
are likely to be national, or do similar relationships
exist in nonmanufacturing industries
for which product markets tend not to be nationwide
in scope? To address this question we have
focused on one major industry characterized by
local product markets, construction, examining
the relationships between the percentage of the
industry's workers in a state who are organized
and union and nonunion wages. This experiment
differs from the one for the manufacturing sector
in that it holds technological and market factors
fixed by looking within one industry instead of by
using industry-level controls in (what amounts
to) a cross-industry analysis. While states are not
the optimal unit for defining product markets in
construction, the mobility of construction workers
from site to site over a wide geographic area
makes states a less inappropriate unit for construction
than for other sectors with "local"
markets. Limited sample sizes make unionization
figures for less aggregate geographic units
subject to considerable potential measurement
error.
We did weighted counts with the 1973-75 May
CPS tapes to determine the percentage of employed
private wage and salary construction production
workers in each CPS state group who
were union members. This variable was added to
a construction worker extract from the CPS
tapes. Because the CPS amalgamated states into
29 groups, there are just that number of distinct
union figures.
Table 3 presents estimates of the effect of the
percentage organized (union members) in a state
on the wages of union and nonunion construction
workers. The primary controls employed in this
analysis are the following: four I-digit Census
production worker occupation group dummies;
three region and four SMSA-size dummies; and
three dummy variables for the sector of the construction
industry in which the worker is employed
(General Building Contractors, General
Contractors Except Building, and Special Trade
Contractors, with the Not Specified construction
group deleted). The results lend additional support
to the claim that there is a positive POW
relationship for unionized workers. When
ln(isisal hourly earnings) is the dependent variable,
the effect of coverage with the union sample
is a sizable 0.283; when ln(uslial weekly earnings)
is on the left, the coefficient on the coverage
variable falls sharply to 0.182. This presumably
is due to the effect of high hourly wages on
hours worked. With the nonunion sample, by
contrast, the estimated coefficient of the fraction
organized variable in both regressions, while
positive, is neither substantially nor significantly
greater than zero. The net result is that in construction
as in manufacturing the union-nonunion
wage differential grows with growth in the percentage
organized. At a given set of means, with
20% organized the hourly wage effect is 37 percentage
points: with 80% organized the comparable
effect is 50 points.
The results with the state construction industry
sample are potentially sensitive to inclusion
of other state (group) variables, presumably due
to the smaller number of state unionization
figures (29). In one analysis, we added to the first
two regressions presented in table 3 the mean of
ln(usual real hourly earninigs) for male nonconstruction
production workers in the appropriate
state and found that the estimated coefficient
(standard error) on percentage organized
dropped noticeably, to 0.213 (0.063) in the equation
for union workers and to -0.055 (0.083) in
the equation for nonunion workers. When a
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
UNION ORGANIZATION AND WAGES 571
TABLE 3.-ESTIMATES OF THE PERCENTAGE ORGANIZED WAGE (POW) RELA-riON
FOR CONSTRUCTION PRODUCTION WORKERS;
MAY 1973-75 CPS INDIVIDUAL-LEVEL DATA
Dependent Variables:
ln( Usuial Hourly ln(Usual Weekly
Earnings) Earnings)
Mean Coefficient
(S.D.) (S.E.)
Union Nonunion Union Nonunion Union Nonunion
Sample Sample Sample Sample Sample Sample
Independent Variables:
State Level
Percentage of construction workers in state 0.531 0.386 0.283 0.062 0.182 0.017
who are union members/100 (0.180) (0.172) (0.056) (0.066) (0.056) (0.076)
Individual Levela
Industry dummies (3); region dummies (3);
1-digit occupation dummies (4); survey
dummies (2); SMSA-size dummies (4); marital
status dummies (3); sex dummy; race dummy;
schooling completed; age - schooling
completed - 5 and its square; number of
dependents; constant yes yes yes yes
R2 0.276 0.286 0.263 0.307
SEE - 0.289 0.364 0.291 0.420
N 2,327 2,825 2,327 2,825
ln(Uslual Holurly 1.919 1.333
Earnings) (0.337) (0.428)
ln(Usual Weekly 5.599 5.002
Earnings) (0.337) (0.502)
i See note in table I for description of the SMSA-size and marital status dummies.
similarly-derived state mean of ln(usual real
wveekly earnings) was added to the third and
fourth equations represented in table 3, the union
POW effect fell to 0.147 (0.058) and the nonunion
effect to -0.077 (0.085). Perhaps the safest conclusion
to reach is that the union-nonunion wage
differential in construction depends positively on
the percentage organized and that precise POW
coefficients for union and nonunion workers
cannot be estimated due to the small number of
states in what amounts to weighted cross-state
regression analysis. 1I
Conclusions
The results presented in this study indicate
that in U.S. manufacturing the percentage organized
in a product market has a strong positive
association with the wages of union members but
not with the wages of nonmembers, making the
union wage differential a positive function of the
extent of organization. For industries characterized
by local markets, in particular construction,
the percentage of an industry's workers in a
geographic area who are organized appears also
to raise the union wage differential, but the effects
on the wages of union and nonunion workers
taken separately are less clear. Thus, the
findings suggest that the percentage organized is
an important determinant of "union power," a
conclusion that could not be drawn from standard
union wage studies, which either relate
average earnings in an industry (or area) to the
extent of unionism in the industry (or area) or
relate an individual's or establishment's wage
rate to a union dummy variable but not an interaction
of this dummy with a product market
organization variable.
In our theoretical discussion, we argued that a
positive association between percentage organized
and the hourly compensation of union
" Cross-state analyses were also done for seven other industries
for which product markets are more likely to be local
than national. While the results were mixed, because, we
believe, states (or even SMSAs) do not adequately demarcate
the appropriate product market boundaries, the union wage
effect was positively related to percentage organized in five
of the seven settings.
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms
572 THE REVIEW OF ECONOMICS AND STATISTICS
members was likely to reflect the impact of union
coverage in a product market on the availability
of substitutes produced in nonunion settings and
thus on the derived demand elasticity for union
workers; a lower elasticity of labor demand
would lower the cost in terms of lost membership
for a given wage increase, leading unions to make
larger wage demands. This is not, however, the
sole possible explanation of the observed regression
results. Percentage organized could be positively
related to both the demand elasticity for
labor and union wage gains without being related
to the demand elasticity for union-made products.
This is because the elasticity of labor demand
depends on the elasticity of substitution
(o-) and labor's share of costs (ax) as well as on
the product demand elasticity (-q,), raising the
possibility that our results could reflect the impact
of coverage on o- and a or the fact that
unions seem to locate and survive where the aand
a are such as to produce an innately small mql.
Unfortunately, it is not possible to sort out the
relative importance of the determinants of ql -o-,
a, and -q,.-in bringing about the strong positive
POW relationships we have observed. However,
the reader should be reminded that the construction
analysis was done for a particular industry.
It seems to us much more difficult to explain a
within-industry positive relationship between
percentage organized and the union-nonunion
wage differential in terms of cross-state variation
in the technological parameters o- and a than in
terms of cross-state differences in the demand
elasticity for the relevant locally-traded product.
REFERENCES
Brown, Charles, and James Medoff, "Trade Unions in the
Production Process," Journal of Political Economy 86
(June 1978), 355-378.
Clark, Kim, "Unions and Productivity in the Cement Industry,
unpublished Ph.D. dissertation, Harvard University,
1978.
Donsimoni, Marie-Paule, "An Analysis of Union Power: The
Role of the Structure and Conduct of the Labor
Movement," unpublished Ph.D. dissertation, Harvard
University, 1978.
Frantz, John, "The Impact of Trade Unions on Production in
the Wooden Household Furniture Industry,' unpublished
senior honors thesis, Harvard College, 1976.
Freeman, Richard B., and James L. Medoff, "New Estimates
of Private Sector Unionism in the United States,'
Industrial and Labor Relations Review 32 (Jan. 1979),
143-174.
, "Substitution between Production Labor and Other
Inputs in Unionized and Nonunionized Manufacturing,'
this REVIEW (forthcoming (a)).
I What Do Unionis Do? (New York: Basic Books,
forthcoming (b)).
I "The Impact of Collective Bargaining: Illusion or
Reality?' in Jack Stieber, Robert B. McKersiekand
Daniel Quinn Mills (eds.), U.S. Industrial Relations
1950-1980: A Critical Assessment (Madison, Wisconsin:
Industrial Relations Research Association, forthcoming
(c)).
Hendricks, Wallace, "Labor Market Structure and Union
Wage Levels," Economic Inqiuiry 13 (Sept. 1975),
401-416.
Johnson, Harry, and Peter Mieskowski, "The Effects of
Unionization on the Distribution of Income: A General
Equilibrium Approach," Qubarterlv Journal of Economics
84 (Nov. 1970), 539-547.
Kahn, Lawrence M., "The Effect of Unions on the Earnings
of Nonunion Workers," Industrial and Labor Relations
Reviewi 31 (Jan. 1978), 205-216.
Lee, Lung-Fei, "Unionism and Wage Rates: A Simultaneous
Equations Model with Qualitative and Limited Dependent
Variables," International Economic Review 19
(June 1978), 415-433.
Levinson, Harold M., "Unionism, Concentration, and Wage
Changes: Toward a Unified Theory,'" Industrial and
Labor Relations Revieiw 20 (Jan. 1967), 198-205.
Lewis, H. Gregg, Unionism and Relative Wages in the
United States (Chicago: University of Chicago Press,
1963).
"Interpreting Union Coefficients in Wage Equations,"
Duke University (1980) (mimeographed).
Masters, Stanley, "Wages and Plant Size: An Interindustry
Analysis,' this REVIEW 51 (Aug. 1969), 341-345.
Rosen, Sherwin, "Trade Union Power, Threat Effects, and
the Extent of Organization," Review of Economic
Studies 36 (Apr. 1969), 185-196.
Segal, Martin, "The Relation between Union Wage Impact
and Market Structure," Quarterly Journal of Economics
78 (Feb. 1964), 96-114.
U.S. Department of Commerce, Bureau of the Census, 1972
Census of Manufactures, Special Report Series, Concentration
Ratios in Manufacturing (Oct. 1973).
U.S. Department of Labor, Bureau of Labor Statistics, Occupational
Injuries and Illnesses by Industry, 1972,
Bulletin 1830 (1974).
, Bureau of Labor Statistics, Occupational Injuries by
Industry, 1973, Bulletin 1874 (1975).
, Bureau of Labor Statistics, BLS Handbook of Methods
for Surveys and Studies, Bulletin 19 10 (Washington,
D.C.: U.S. Government Printing Office, 1976a).
I Bureau of Labor Statistics, Occupational Injuries
and Illnesses by Industry, 1974, Bulletin 1932 (1976b).
Weiss, Leonard W., "Concentration and Labor Earnings,'
American Economic Reviewv 56 (Mar. 1966), 96-117.
This content downloaded from 147.251.55.15 on Thu, 29 Mar 2018 09:35:59 UTC
All use subject to http://about.jstor.org/terms