Unionism and the Dispersion of Wages
Author(s): Richard B. Freeman
Source: ILR Review, Vol. 34, No. 1 (Oct., 1980), pp. 3-23
Published by: Sage Publications, Inc.
Stable URL: http://www.jstor.org/stable/2522631
Accessed: 29-03-2018 11:26 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
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to
ILR Review
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION
OF WAGES
RICHARD B. FREEMAN*
This study examines the effect of trade unionism on the dispersion of wages
among male wage and salary workers in the private sector in the United
States. It finds that the application of union wage policies designed to
standardize rates within and across establishments significantly reduces wage
dispersion among workers covered by union contracts and that unions
further reduce wage dispersion by narrowing the white-collar/blue-collar
differential within establishments. These effects dominate the more widely
studied impact of unionism on the dispersion of average wages across industries,
so that on net unionism appears to reduce rather than increase wage
dispersion or inequality in the United States.
T RADE unionism alters the distribution of
wages in several ways. First, by raising
the wages of organized workers relative to
others, unionism changes the dispersion of
wages in the economy, increasing inequality
when highly paid workers are organized
and reducing inequality when low-paid
workers are organized. On the basis of estimates
of the wage effect and its correlation
with wage levels, Lewis concluded that by
raising wages unionism has raised the relative
inequality of average wages among industries,
as measured by the standard deviation
of relative wages, by two to three percentage
points.' In addition, simply by creating
differentials between otherwise comparable
workers (regardless of their level of
*Richard Freeman is a professor of economics at Harvard
University. He has benefitted from the research
assistance of Casey Ichniowski.
1H. Gregg Lewis, Unionism and Relative Wages in
the United States: An Empirical Inquiry (Chicago:
University of Chicago Press, 1963), p. 282.
pay), unionism also increases inequality.
Alternately, however, unions also affect the
dispersion of wages within the organized
sector through the "standard rate" policies
stressed in the institutional literature.2
While most economists accept the notion
that standardization of rates reduces dispersion
among union members, quantitative
estimates of this effect are lacking.
The within-sector effect could be large,
offsetting or more than offsetting the increase
in inequality due to the impact on
dispersion across groups, or it could be
small.
2Lloyd G. Reynolds and Cynthia H. Taft, The Evolution
of Wage Structure (New Haven, Conn.: Yale
University Press, 1956); Sumner H. Slichter, James J.
Healy, and E. Robert Livernash, The Impact of Collective
Bargaining on Management (Washington, D.C.:
The Brookings Institution, 1960). Sidney and Beatrice
Webb, Industrial Democracy (London: Longmans,
Green and Co., 1902). David A. McCabe, The
Standard Rate in American Trade Unions (Baltimore,
Md.: The Johns Hopkins Press, 1912).
Industrial and Labor Relations Review, Vol. 34, No. 1 (October 1980). ? 1980 by Cornell University.
0019-7939/80/3401-0003$01 .00
3
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
4 INDUSTRIAL AND LABOR RELATIONS REVIEW
The purpose of this paper is to estimate
the magnitude of the effect of unionism on
the dispersion of wages in the organized
sector and to use the estimates to examine
the potential contribution of the institution
to overall wage dispersion. The first
section of the paper reviews the evidence on
trade union standard rate policies and considers
the economic rationale underlying
them. The second section contrasts the dispersion
of earnings among blue-collar
workers in the organized sector with that in
the unorganized sector. Section three examines
the effect of unionism on the differential
between production and nonproduction
workers within establishments. Section
four then compares the dispersion-increasing
and dispersion-reducing effects of unionism
to obtain an estimate of the overall
effect of the institution on wage inequality.
A final section reports conclusions.
Union Wage Policy andIntrasectoral
Dispersion
Unionism is expected to reduce the dispersion
of wages among organized workers
because of long-standing union wage policies
in favor of the "standard rate," defined
as uniform piece or time rates among comparable
workers across establishments and
impersonal rates or ranges of rates in a given
occupational class within establishments.
That unions strive to standardize rates
across establishments has long been recognized
by institutional labor economists.
Indeed, according to Slichter, Healy and
Livernash, "wage standardization within
an industry or local product market is the
most widely heralded union wage policy."3
Sufficient examples exist of major collective
bargaining agreements that achieved standardization
of rates to suggest, moreover,
that the goal of uniformity across firms has
influenced the wage structure. The development
of the Comprehensive Wage Study
in steel in 1946 - 47 appears to have increased
uniformity among plants in that
industry.4 Successive steel contracts from
3Slichter, Healy, and Livernash, The Impact of Collective
Bargaining, p. 606.
4Jack Stieber, The Steel Industry Wage Structure
(Cambridge, Mass.: Harvard University Press, 1959).
1947 to 1954 eliminated the longstanding
southern "Birmingham" geographic differential.
The ILGWU and Amalgamated
Clothing Workers have established uniform
piece rates in their contracts in broad geographic
areas. The Teamsters reduced regional
differentials for over-the-road drivers
in 1964 when the National Master Freight
Agreement was signed. In most instances of
multi-employer bargaining (which in 1974
constituted 42 percent of major collective
agreements in the United States), or multiplant
bargaining (an additional 42 percent5),
uniform or near uniform rates are
established across firms.
The policy of standardization of rates
across plants has not been adhered to blindly,
of course. "Exceptions" are often granted
to take account of specific competitive
situations, such as the danger of a plant
closing,6 and the relevant sector or wage
"contour" for standardization changes as
market conditions change. There is no denying,
however, that union policies operate
toward uniformity of rates among similar
plants and less dispersion within the organized
sector.
The economic rationale and strength of
policies toward standardization of rates
across establishments will depend on market
conditions. When firms compete in the
same market, both employer and worker
interest can be expected to favor standard
rates. On the firm side, no enterprise wants
union contracts that are more expensive
than those of its competitors.7 On the
worker side, as long as markets cannot be
differentiated to permit price discrimina'U.S.
Department of Labor, Bureau of Labor Statistics,
Characteristics of Major Collective Bargaining
Agreements (Washington, D.C.: G.P.O., July 1, 1974),
Table 18, p. 11.
6David H. Greenberg, "Deviations from WageFringe
Standards," Industrial and Labor Relations
Review, Vol. 22, No. 2 (January 1968), pp. 197- 209.
Morris A. Horowitz, The New York Hotel Industry:
A Labor Relations Study (Cambridge, Mass.: Harvard
University Press, 1960), pp. 165- 66. Kenneth Alexander,
"Market Practices and Collective Bargaining
in Automotive Parts," Journal of Political Economy,
Vol. 69, No. 1 (February 1961), pp. 15 - 29.
7Thomas Kennedy, The Significance of Wage Uniformity
(Philadelphia: University of Pennsylvania
Press, 1949), p. 2.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 5
tion, the desire for a single rate makes good
sense in terms of collective behavior. Without
a common rate across firms competing
in the same market, the monopolistic price
would come under severe pressures in economic
downturns when some union members
could be expected to undercut others.
Uniformity across firms "takes wages out of
competition."
When market conditions differ among
firms, so that price discrimination by the
union is possible, the desire for standardization
of rates is weaker. Even here, however,
the amalgamation of locals into a
single national and the lack of adequate
internal redistributive mechanisms within
the union limits the extent of price discrimination.
Union solidarity is difficult to
maintain if some workers are paid markedly
more than others, and such a pattern invites
division within the organization and
loss of certain common advantages, such as
joint strike funds and interrelated policies
toward major employers. Policing an agreement
to maintain monopolistic rates is also
likely to be difficult in this case. On the
employer side, firms in low-wage local
markets have often fought standardization
of rates on the grounds that standardization
deprives them of the advantage of lower cost.
Despite some opposition to standard rates
from high-wage union locals8 and low-wage
firms and the granting of exceptions (which
can be viewed as a step toward price discrimination),
however, the overall pressures
appear to operate toward standardization
of rates. The balance struck between standardizing
rates and granting exceptions will
be influenced by such factors as market conditions,
union coverage, and elasticities of
demand.
The second major component of union
standardization policies is for equalization
of pay and reduction of "personal differences"
among similarly skilled workers
within establishments. Prior to unionization
many industries are plagued by what
have been called "inequity" problems, with
different wages paid to individuals depending
not on the jobs held but on the workers'
8See, for example, McCabe, The Standard Rate in
American Trade Unions.
characteristics as perceived by foremen.
Under unionism, however, the process of
wage-setting within firms is quite different,
with job rates rather than personal rates
the major determinant of pay. The number
of job categories is often relatively small
(only 36 classifications in steel, for example),
gathering diverse activities in single
categories and thus narrowing the potential
dispersion; and the range of rates within job
categories tends to be narrow. While many
large nonunion enterprises employ similar
formal wage-setting practices today, the
option for personal differentials based on
ability (or favoritism, or any other factor)
within a job category is generally larger
than in the union sector. Merit increases
appear, for example, to be less prevalent in
the union than nonunion sector. In the
1970s, 43 percent of companies gave plant
employees "wage adjustments based on a
merit plan,"9 whereas just 12.5 percent of
major union contracts had merit progression
plans.'0 Overall, according to Slichter,
Healy, and Livernash, "the influence of
unions has clearly been one of minimizing
and eliminating j udgement-based differences
in pay for individuals employed on the
same job" and of "removing ability and performance
judgements as a factor in individual
pay for job performance.""
Several factors appear to explain union
policies favoring reduction of within-establishment
wage variation. Within a union,
when the mean wage exceeds the median, a
majority of members can presumably be
expected to favor redistribution in favor of
the lower paid. In a simple median voter
model of union behavior, the 50+ percent of
members who earn less than the mean would
favor a policy of greater gains for the lower
paid. Union opposition to personal rates
probably also reflects worker solidarity and
preference for obj ective standards as opposed
to the subj ective decisions of foremen.
9Bureau of National Affairs, Wage and Salary Administration
Survey, Bulletin 97 (Washington, D.C.:
G.P.O., July 1972), Table 7, p. 14.
10U.S. Department of Labor, Bureau of Labor Statistics,
Characteristics of Major Collective Bargaining
Agreements (July 1, 1974).
11Slichter, Healy, and Livernash, The Impact of
Collective Bargaining, p. 602.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
6 INDUSTRIAL AND LABOR RELATIONS REVIEW
It is difficult to see how the union would be
able to maintain its organizational strength
and monopolistic prices, in fact, with significant
personal differentials within firms.
Moreover, since presumably all workers obtain
higher wages in the presence of the union,
there are no "losers" from the policy but
simply differential gainers.
Not all union wage policies, it should be
noted, operate toward reduction of dispersion.
The effect of unionism on blue-collar
occupational differentials is unclear, apparently
varying by the type of union and
workers organized. Craft unions may widen
the wage structure by raising the pay of the
highly skilled. Industrial unions, previously
expected to reduce differentials by
negotiating constant cents per hour increases,
in recent years have sometimes operated
to maintain or increase percentage
skill differentials in response to pressures
from skilled workers. After reviewing the
evidence, Reynolds and Taft concluded that
"any net effect on occupational differentials
. . . [is] in the direction of narrowing."'I2
Whatever unions do to the skill differential
among blue-collar workers, however, they
tend to raise the pay of production workers
relative to higher paid nonproduction
workers within firms and thus narrow that
component of occupational wage differen-
tials.
Finally, note that standardization of piece
rates, as opposed to time rates, has no clear
effect on dispersion. If in the absence of unionism
piece rates would be higher in less
productive plants, standardization would
increase dispersion in hourly pay; if, conversely,
piece rates would be higher in more
productive plants, standardization would
decrease dispersion.
These complications notwithstanding,
the institutional evidence of wage policies
under collective bargaining suggests that
trade unionism can be expected to reduce
inequality of wages within the union sector,
largely by equalizing rates across establishments
and by replacing personal rates by
formal job rates within establishments. The
key issue addressed in this paper is the mag-
12Reynolds and Taft, The Evolution of Wage Structure,
p. 185.
nitude of this reduction. Do unions reduce
inequality in wage dispersion within the
union sector by a sizeable amount? And
how does this reduction compare to the increase
in inequality due to the potential increase
in dispersion of wages between the
union and nonunion sectors?
Dispersion Among Blue-Collar Workers
To evaluate the quantitative impact of
standardization policies on workers in the
union sector, this section compares the dispersion
of wages among otherwise similar
organized and unorganized blue-collar
workers, using data from the Current Population
Survey and Expenditures for Employee
Compensation survey. Dispersion is
measured by the standard deviation of the I n
of earnings, an appropriate metric when
earnings are set by the in earnings function
widely used in modern labor economics;
when union wage differentials are measured
in relative rather than absolute terms; and
when earnings are lognormally distributed.
In comparison to other widely used measures
of wage dispersion, the standard deviation
of in wages weights inequality more
heavily at the lower end of the distribution
than at the upper end.'3 Since this will attach
less significance to the narrowing of the
white-collar/blue-collar gap in the upper
part of the distribution than to potential
increased inequality between union and
nonunion blue-collar workers in the lower
part, the standard deviation of in metric is
likely to "understate" the equalization of
wages under unionism relative to other
widely used metrics (such as Gini coefficients,
for example).
The principal problem in comparing dispersion
of wages between organized and unorganized
blue-collar or production workers
is to differentiate between the effect of unionism
and the effect of other factors correlated
(for whatever reason) with unionism.
If union workers were more alike in
personal characteristics or in their distributions
among industries or occupations than
13For a discussion of the properties of diverse measures
of inequality see Anthony B. Atkinson, "On the
Measurement of Inequality," Journal of Economic
Theory, Vol. 9, No. 1 (1970), pp. 244- 63.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 7
nonunion workers, the variance in 1 n wages
would be lower in the union sector for
reasons extraneous to standardization poli-
cies.
This study employs several techniques to
deal with this problem. One is to compare
dispersion within narrowly defined groups,
such as workers in the same three-digit industry
with the same age, sex, and education.
As comparison cells become increasingly
narrow, the possibility of differences
due to omitted factors that are correlated
with unionism is reduced.
Multiple regression analysis provides
another tool for "correcting" for the effect
of different characteristics and identifying
those due to union wage policies. Let Xi be a
determinant of in wages (W), bi be its coefficient,
and e be the residual. Then, with the
superscript u for union workers and n for
nonunion workers, we have union and nonunion
wage equations:
(1) W a + buXu +e and
A
(2) an =~a II + E b nsn + eArz(2) W a b
The regression decomposes the variance in
Wu and Wn into a part explained by the
wage determinants and a residual.
Next, if union workers have different
characteristics than nonunion workers, the
variance and covariance of the Xs will differ
between the two groups, with resultant
differences in the variances of W "and WI'.
Variance decomposition of the equations
will be used to eliminate the effect of differences
in characteristics by estimating the
impact of different dispersions of the Xs on
the dispersion of earnings, given either the
union or nonunion regression weights.
For example, the extent to which 02 (Wu)
differs from a2 (Wan) as a result of differences
in the characteristics in the samples
can be gauged by:'4
(3) [0(bi )2 [2 (Xt) - a2 (XI,]
A A ( n r
+ bb[a(XX) - a (
"4The two variances differ only by the regression coefficients
and thus represent the approximate standardization
for differences in characteristics. This is not,
of course, a complete decomposition.
where a 2 (XU) is the variance in characteristic
i among union members; o(X `X ') is
the covariance in characteristics among
union members and a 2(X7) and a2(X3X'1)
are the relevant variances and covariances
for nonunion workers and where bs can be
taken from either the union or nonunion
regressions. Any variation not attributable
to Equation 3 represents the effect of unionism
on dispersion among similar workers.
The decomposition of differences in
dispersion can be pursued further by comparing
the variance explained by the regressions,
conditional on similar characteristics,
to the residual variation. With variances
and covariances of the Xs fixed, unionism
can change dispersion by altering
the effect of wage-determining variables on
earnings, reducing (or increasing) the regression
coefficient in earnings equations;
or by altering the dispersion of earnings
among workers with the same wage-determining
characteristics. The first effect can
be estimated by comparing the dispersion
of wages of a group of workers with given
characteristics that would result from Equation
1, the union wage equation, with dispersion
that would result from Equation 2,
the nonunion wage equation:
(4) E [(bw)2 _ (bU)2] o2(Xi)
+ E(b. gb - b U) a(X X.)
where o2(Xi) and o(XiX1) refer to the dispersion
of characteristics among either
union or nonunion workers.
Differences in the residual variances
themselves, as reflected in the standard
errors of estimates of the equations ( a =
E/e 21N where N = degrees of freedom), provide
one possible measure of the impact of
unionism on the wages of workers with
identical characteristics within the separate
sectors. The residual variances reflect the
variation remaining after the coefficients
and variances and covariances of independent
variables in the union equation and
in the nonunion equation have been taken
account of in the separate regressions.
Finally, since the sum of squares due to
the regression and the sum of squared reThis
content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
8 INDUSTRIAL AND LABOR RELATIONS REVIEW
siduals in each equation are independent,
differences between the various variances
can be tested by the standard F statistics with
degrees of freedom dependent on the observations
and number of control variables
in the regressions: F = ( a V An) 2 for the residual
variation and F = R 2 a2(Wn)/R 2
&2 (WU) for the explained variation.
Current Population Survey data set. Estimates
of the dispersion of wages among
union and nonunion blue-collar workers
and of the contributions of differences in
characteristics, differences in the coefficients
in wage equations, and of residual
variance to differences in dispersion were
made using the Current Population Surveys
(CPS) of the U.S. Bureau of the Census for
May 1973, May 1974, and May 1975.1' The
May surveys of the CPS contain data on
union membership, usual weekly earnings,
usual hours worked, and other characteristics
of workers that permit analysis of differences
in dispersion in relatively narrow
groupings. To obtain a large sample, the
three surveys were amalgamated into a
single sample, with earnings in 1974 and
1975 deflated to 1973 levels to maintain
comparability. To reduce differences in the
characteristics of workers, the analysis focuses
on male private wage and salary workers,
exclusive of students, and treats manufacturing
and the rest of the economy sep-
arately.
The overall dispersion of wages among
union and nonunion blue-collar workers in
the sample is summarized in Table 1 in
terms of the standard deviation in usual
hourly and weekly earnings for manufacturing
and nonmanufacturing, respectively;
the difference in standard deviations; and
the F test for the difference. While lack of
controls makes interpretation of the results
subject to question, the pattern is clear: in
both manufacturing and nonmanufactur"5The
Current Population Survey is a monthly
survey of about 50,000 households. The May Survey
asks questions about usual weekly earnings and unionism
that provide, perhaps, the best data available on
dispersion of wages differentiated by union status. See
the U.S. Department of Labor, Special Labor Force
Report, Bulletin 195, for a detailed discussion of the
earnings data.
ing, the dispersion of wages among unionized
male blue-collar workers is considerably
lower than among nonunion workers.
The differences in standard deviations range
from - .10 to - .14 or from 22 to 30 percent
of the standard deviation in the nonunion
sector. By the F test, all of the differences in
the table are significant at better than the
one percent level.
The distribution of in wages in the two
sectors themselves is examined in the figure,
which presents the frequency distribution of
1 n usual hourly earnings for union and nonunion
workers. The figure permits some
evaluation of the possibility that wages are
less dispersed among unionists largely because
of some "peculiarity" in the tails of
the distribution, such as the absence of
either very low or very highly paid workers
from the union sector.'6 The figure shows no
striking aberrations in the distributions.
In both manufacturing and nonmanufacturing,
the upper and lower parts of the
earnings distribution are more compressed
about the median in the union sector, resulting
in more "peaked" frequencies.
Measured by the percentage difference between
quintiles and the median, the top
quintile is twice as far above the median in
the nonunion than in the union distribution
in both manufacturing and nonmanufacturing
while the bottom quintile is twice
as far below the median in the nonunion
than union distribution in nonmanufacturing
and 60 percent further below in manufacturing.
Overall, the difference between
the top and bottom quintiles is markedly
less in union distributions, with in differences
in manufacturing of .519 (nonunion)
and .298 (union) and differences in nonmanufacturing
of .653 (nonunion) and.326
(union).17
16This would occur if either low wage firms were
driven out of the unorganized sector by union wage
gains or if high wage workers eschewed unionism because
of standardization policies.
"7The precise quintile deviations of the distributions
of log wages were: percentage deviation of the first
quintile from median: manufacturing, - 0.162 (union),
- 0.259 (nonunion); nonmanufacturing, - 0.164
(union) and - 0.328 (nonunion); percentage deviation
of the fifth quintile from the median: manufacturing,
0.136 (union) and 0.267 (nonunion); nonmanufacturing,
0. 162 (union) and 0.325 (nonunion).
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 9
Table 1. Comparison of the StandardDeviation of theLog of Usual Hourly and Weekly Earnings
Among Blue-collar Private Wage and Salary Male Workers, 1973 - 75, by Union
Membership.
Standard Deviation of Log Earnings
Usual Usual
Sector Number of Hourly Weekly
Observations Earnings Earnings
Manufacturing
Union 8339 .288 .302
Nonunion 6835 .398 .436
Difference - - .110 - .134
F test of difference
in 0 - 1.91 2.08
Nonmanufacturing
Union 6253 .350 .366
Nonunion 9227 .451 .508
Difference - -.101 -.142
F test of difference
in a - 1.66 1.93
Source: Tabulated from May 1973, 1974, and 1975 Current Population Survey data tapes. Usual hourly earnings
obtained by division of usual weekly earnings by usual hours worked. To eliminate effect of inflation on wage
differences among the 3 years, the wages of 1974 were divided by 1.0765; those in 1975 by 1.1782 to put them on a 1973
basis, using average hourly earnings of workers on private payrolls [U.S. Department of Labor, Bureau of Labor
Statistics, Employment and Training Report of the President (Washington, D.C.: G.P.O., 1977) Table C-3, p. 296].
Students and persons working fewer than 20 hours per week were deleted from the samples, and samples were limited
to private wage and salary workers.
As a first step toward determining whether
the markedly lower dispersion of earnings
in the union sector can be attributed to union
wage policies as opposed to the possibly
greater similarity of union than nonunion
workers, the standard deviation of the log of
hourly earnings was calculated within more
narrowly defined industry, occupation,
geographic, education, and age groups.
Table 2 summarizes the results in terms of
the differences in the standard deviations
( Yu- nu) and the number of differences
that are statistically significant at the 5 percent
level. The data show clearly that inequality
is smaller in the union sector
within detailed categories. In manufacturing,
the dispersion is lower among organized
workers in all two-digit industries,
in 68 of 75 three-digit industries, in all but 2
state groupings, and in all other categories.
Figure 1. Comparison of the Distribution of Hourly Earnings Among Union and Nonunion
Male Blue-Collar Workers by Sector.
Manufacturing Sector Nonmanufacturing Sector
0.16 0.16-
0.14 - Union 0.1 Union
0.12- 0.12-
0.10 Nonunion \ 0.10 Nonunion
0.8 - 0.8bL0.6
- b 0.6
0.4- 0.4-
0.2 - 0.2 -
0.05 0.10 0.15 0.20 0.25 0.30 ; 0.05 0.40 0.15 0.20 0.25 0.30
Log (Hourly Pay) Log (Hourly Pay)
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
10 INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 2. Differences in the Standard Deviation of the Log of Usual Hourly Earnings
of Male Blue-Collar Workers by Selected Characteristics.
Number of
Differences
Statistically
Total Significant at
Number 5% level or
of Distribution of Differences better by
Detailed Groups Groups ( u nonunion - a union) F test
Manufacturing >.00 .00-.05 .05-.10 .11-.15 'z..15
Two-digit
industry 22 0 6 7 6 3 16
Three-digit
industry 75 7 14 20 18 15 50
"Two-digit"
occupation 16 0 5 9 2 0 13
State groups 28 2 3 10 10 3 22
Schooling
categories 6 0 1 2 2 1 6
Age
categories 5 0 1 1 3 0 5
SMSA-size
categories 5 0 0 1 4 0 5
Nonmanufacturing
Two-digit
industry 22 3 3 4 7 4 12
Three-digit
industry 84 16 10 22 15 21 31
"Two-digit"
occupation 19 2 2 7 6 2 15
State groups 29 0 3 11 10 5 25
Schooling
categories 6 0 0 5 1 0 6
Age categories 5 0 1 2 1 1 5
SMSA-size
categories 5 0 0 3 2 0 5
Source: Calculated from Current Population Survey tapes, May 1973- May 1975.
In most of these cases, moreover, the differential
is significant at the 5 percent level by
the F test, whereas in no case where a, >
o , is the differential significant or
large.
In the nonmanufacturing sector the pattern
is less striking but still clear-cut.
Among three-digit industries, a,, is less
than o ,n in 68 of 84 cases; it is significantly
lower in 31 cases and not significantly
higher in any, It is also lower by significant
amounts in most of the other comparisons.
On the basis of these comparisons, it appears
that the lower dispersion of earnings
among union workers cannot be attributed
to such patterns as, say, a greater concentration
of organized workers in certain industries,
occupations, or age groups.
The effect of several characteristics on
dispersion is estimated next by regressing
the log of hourly and weekly earnings of
union and nonunion workers, taken sepThis
content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 11
arately, on years of schooling, experience
(which is calculated as age - years of
schooling - 5), experience squared, race,
marital status and number of dependents,
and dummy variables for geographic locale
(state, SMSA), industry, and occupation.
The equations provide estimates of the bu
and bV coefficients needed for the various
standardizations in Equations 3 and 4 and
provide estimates of residual variances as
well.
Table 3 presents the results of the regressions.
Columns 1 and 2 record the mean
and standard deviation of selected variables
for union and nonunion workers and list the
other control variables. The remaining
columns give regression coefficients on four
major determinants of earnings and other
information about the calculations.
According to the standardization hypothesis,
the regression coefficients on the principal
wage-determining variables should be
lower in the union sector. This expectation
is borne out in the data. Lines 1-3 show
noticeably smaller effects for schooling and
experience in the union than nonunion
equations. Given the small standard errors,
the differences are highly significant. For
example, in the fifth and sixth columns of
coefficients, where an extensive set of control
variables are used, schooling has a .034
impact on the hourly earnings of nonunion
workers compared to .020 on the hourly
earnings of union workers in manufacturing
and a comparable differential effect
of .028 versus .015 in the nonmanufacturing
sector. The experience differentials are
also markedly lower among union workers,
suggesting flatter life cycle earnings profiles.
The exclusion of union-negotiated
fringe benefits, which accrue largely to
older workers,'8 however, leaves open the
impact on total compensation, as opposed
to straight-time pay. Only the coefficient
on race shows any divergence from this
pattern, with a smaller impact in the union
equation in manufacturing but not in nonmanufacturing,
a result possibly due to the
"8Richard B. Freeman, "The Effect of Trade Unions
on Fringe Benefits," NBER Working Paper No. 292
(Cambridge, Mass.: National Bureau of Economic
Research, October 1978).
historic pattern of discrimination by craft
unions.
The coefficients on the diverse dummy
control variables listed in line 5 of the table
can be compared in terms of a measure not
shown in the table: the standard deviations
of these coefficients for the union and nonunion
groups. In manufacturing, the standard
deviation of the estimated coefficients
of the dummy variables among the 21 industry
dummies used in columns 1-4 was
0.51 in the union sector compared to 0.55 in
the nonunion sector. The equivalent standard
deviations among occupations were
0.28 (union) and 0.33 (nonunion); among
regions 0.08 (union) and 0.10 (nonunion);
and among SMSA groups, 0.07 (union) and
0.08 (nonunion). Thus in each case the
dummy variables reveal greater differentiation
in the nonunion sector. In nonmanufacturing,
the results are stronger: the
standard deviation of coefficients on the
dummy variables on 2-digit industry dummies
is 0.56 for union and 0.92 for nonunion
workers; the standard deviation on the coefficients
for regions is 0.07 (union) versus
0.09 (nonunion); while the standard deviation
on the coefficients by occupations are
0.51 (union) and 0.56 (nonunion). Comparable
calculations based on the regressions
with the more detailed controls in
columns 7 - 10 give similar results. With
states rather than regions as independent
variables, the standard deviations of the
dummy variable coefficients on geographic
areas are: manufacturing, union (0.39),
nonunion (0.57); nonmanufacturing, union
(0.29), nonunion (0.57). By contrast, the
standard deviations of coefficients on the
three-digit industry dummies are smaller
than those at the two-digit industry level.'9
Overall, the evidence suggests a diminution
of the effect of the diverse wage-determining
factors in the union sector, as predicted by
the standardization policy.
To what extent is the lower standard
19With the same detailed controls the standard deviation
of the dummy variable coefficients among industries
were: manufacturing, 0.91 (union) and 0.94 (nonunion);
nonmanufacturing, 2.11 (union) and 2.14
(nonunion), suggesting smaller differences as we
further develop the industry variables.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
12 INDUSTRIAL AND LABOR RELATIONS REVIEW
o' I~ , o o Coo -C ? o - CI tn
0~~~~~~~~~~~~~~0 C
; 1 1 *0 I ~~~~~C:,J tO :>. >- Cow CT t
U~~~~~~~~~~~C _) C> 0 Oj tn C oc 0 o *z t:CD 4 C9[? s
t D ) I z .0 I : >~~C) C Co C) g- cQ-4t CD 1 cq C ) + C
AV)~~~~~~~~~~~~~~C 01 ,O 00 .. .. .. .
O~~~~~~~~~~~C ,y C) I ) 0 cI a n
CS~~~~~~~~~~f CT "Z. .o go Cgom 'N CIO
c~~n Q_) W I a I *t 1 Cc0 C C CO C0) GM O m N CI CI
*_s, XV F t 00 00 00 -O - 1 t "
04 ~ ~ ~ ~ 4
0
X : -q 0 (= t .* I
CT C C) ) G - r- 000
bI n
0_ vD m3 m > _ N m t t
~~~~~~~~~~~~~~~C 1 O ,t Cl _j _ _b
0 0 an an
0-4 - I C 1 --
u pl 000000
00
c 00O C0
t-o0~~~~~~~~t
_ o ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~ 4-
o~~~~~~~~~~~~~~~~~~~~~~~ZC
S ~ ~ ~ ~ 4_ 000~~144
cj~~~~~~~~~~~
o~~~~~~~~~~~:Iuc
Ej4
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 13
CM
an - > Qo r
C* *C C Ct
bX
C0
o oC o ? ? X- t (S 00 Cl ?
00 CM C) 't, C 00 00Cj4~0
QQ -- - 04 1 gM
C>~~~~~~~~~~~~~~~~~~Can C 00 C> C 00 00 MCt
C> 8~~
C)('lC 4 C>z ',C
< N04
Ci M
CM Q " - 04
CM in~~~~~~~~~~~~n
00
04~~~~~~~~~~
0 ZZ~~~~~~~~~~~~
cli . z
M 0 M -o O 0
oooo?CulXKtNCgt* * I * * 0
b
-04 rN ^ O
O O O O O O < O N K N H~~~c~I
0 00~ 00
-~ ' ) C- .
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
14 INDUSTRIAL AND LABOR RELATIONS REVIEW
deviation of earnings in the union sector
due to the reduction in the coefficients on
wage-determining factors as opposed to
differences in the variation of characteristics
among workers? How much of the
difference in dispersion is due to reduction
in dispersion among workers with similar
characteristics?
Calculations designed to answer these
important questions are given in Table 4,
which uses Equations 2 and 3 to evaluate the
differential impact of the forces at work.
Line 1 gives the differences in the standard
deviation of the log of hourly earnings between
union and nonunion workers. Line 2
estimates the difference that could result if
union (nonunion) workers had the characteristics
of nonunion (union) workers. Because
workers in the union sample evince, in
general, less dispersed characteristics, the
figures in line 2 show a diminished differential
in dispersion, though one that is not
large enough to alter substantially the basic
finding. According to the estimates, if union
workers had the same dispersion of characteristics
as nonunion workers, the variance
of their wages would be substantially lower
than the variance of the wages of nonunion
workers. Adjustment for differences in
characteristics has only a "moderate" effect
on the differences in dispersion.
Line 3 of the table calculates the effect of
the difference in regression coefficients on
dispersion, using the characteristics of union
and nonunion workers as weights. It
shows that because of the smaller impact of
measured factors on union than nonunion
earnings, dispersion was lower by 0.050 to
0.078 points in manufacturing and by 0.041
to 0.044 points in nonmanufacturing.
Finally, line 4 of the table compares the
residual variation in the union and nonunion
wage equations, as given by the standard
errors of estimate at the bottom of Table
3. The standard errors of estimate from the
regressions show that the union sector has
less dispersion in both manufacturing and
nonmanufacturing. Comparing the differences
in dispersion due to the smaller regression
coefficients in line 3 with the differences
due to smaller residual variation in
line 4 reveals an interesting difference between
the manufacturing and nonmanufacturing
sectors. According to the table,
unionism lowers dispersion in manufacturing
largely by reducing the effect of measured
characteristics while it lowers dispersion
in nonmanufacturing largely by
reducing variation within narrowly defined
groups. What is important, however, is not
the precise route of the impact but the fact
that, corrected for differences in characteristics,
union workers have a markedly lower
dispersion of wages than nonunion work-
ers.
Expenditures for Employee Compensation
data set. The effect of unionism on dispersion
can also be examined with establishment
data on the compensation of production
workers from the Expenditures for
Employee Compensation (EEC) survey of
the Bureau of Labor Statistics. The establishment
data have both advantages and
disadvantages for examining the standardization
hypothesis. On the plus side, the
data relate to total compensation (including
fringes), giving a more comprehensive
measure of compensation than earnings on
the CPS tapes. In addition, unionization is
measured by a 0- 1 collective bargaining
coverage variable, which is 1 if 50 percent or
more of the production workers are organized
and 0 otherwise. This is presumably
better than the union membership variable
on the CPS, since wages are set by contracts
for nonunion as well as union workers in
the unit. One major disadvantage of the
EEC tape is that the data do not contain information
on worker characteristics and are
limited to establishment averages for all
production or nonproduction workers. To
rectify the lack of information on personal
characteristics, the mean education, percent
above fifty and below thirty years of age,
percent male, and percent black of union or
nonunion production workers in the threedigit
industry in which the establishment
was located were added to the data tape,
using estimates based on the CPS files. Detailed
industry dummy variables and these
measures partly remedy the lack of information
on personal characteristics. Another
problem, for which there is no remedy, is
that for reasons of confidentiality the BLS
deleted certain large establishments from
the data set, with unclear effects on disThis
content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 15
Table 4. Estimates of the Effect of Differences in the Dispersion of Characteristics, of
Differences in Regression Coefficients, and Residual Variation on the Difference in the Standard
Deviation of the In of Hourly Earnings Between Nonunion and Union Production
Workers.
Differences in
Standard Deviations Manufacturing Nonmahufacturing
1. Initial difference in standard deviations .110 .101
2. Difference, after correcting for
different characteristics
a. Using union wage equation .070 .084
b. Using nonunion wage equation .088 .082
3. Difference attributable to different
earnings equations
a. Using characteristics of union
workers as weights .050 .041
b. Using characteristics of nonunion
workers as weights .078 .044
4. Difference attributable to differences
in residual variation .049 .069
Source: Calculated from Current Population Survey tapes, May 1973- 75.
Line 1 from Table 1.
Line 2a: Estimated as -(Wn) A (WU) where A(WU) is calculated from
a2 (Wu) = (bi>)2 a2 (Xi) + EX bu bu U(Xi X]) (the variance that would be found among worker
with the characteristics of nonunion workers paid by the union wage equation.)
Line 2b: Estimated as a (Wn) - a (Wu) where a (Wn) is calculated from
2(Wn)=~ X(bin) a2(Xi) + i b i. b n (X i' X lu) (the variance that would be found among worke
the characteristics of union workers paid by the nonunion wage equation.)
Line 3a: Estimated using Equation 3 with 02(Xt) and Ur(Xu Xu) as weights.
Line 3b: Estimated using Equation 3 with o2(Xn) and a (Xn Xn) as weights.
Line 4: Calculated from columns 3 and 4 of Table 3, line 7.
persion.20
To obtain as large a sample as possible,
the EEC surveys for 1967 - 68, 1969 - 70, and
1971 - 72 were amalgamated into a single
sample. Despite its problems, the EEC
survey is an especially valuable source of
data on the impact of collective organization
on dispersion because it permits us to
check on one of the presumed routes of the
standardization of wages, equalization of
rates across establishments, which cannot
be studied with data tapes on individuals.
Estimates of the link of trade unionism to
20See Richard B. Freeman and James L. Medoff,
"New Estimates of Private Sector Unionism in the
United States," Industrial and Labor Relations Review,
Vol. 32, No. 2 (January 1979) for discussion of
this problem.
the standard deviation of the log of total
hourly compensation for production workers
in manufacturing industries in the EEC
data file are given in Table 5. Line 1 records
the number of union and nonunion establishments
in the sample. Line 2 gives the
standard deviations in the log of total compensation
while line 3 records the standard
error of estimate obtained by regressing
the log of compensation on the variables
listed in line 4. Line 5 gives the R2 from the
regression and line 6 compares standard
deviations within specified narrow cate-
gories.
The principal conclusion to be drawn
from the table is that, consistent with the
standardization hypothesis, unionism is
associated with markedly lower variations
in compensation across establishments.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
16 INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 5. Estimates of Dispersion of Total Compensation Per Manhour of
Nonoffice Workers Among Manufacturing Establishments, by Union Status.
Union Nonun ion
1. Number of observations 2580 1494
2. Standard deviation of log of total
compensation per manhour .258 .302
3. Standard error of estimates of equation,
after regression onrcontrol variables .194 .221
4. Regression variables
20 industry dummies a
5 year dummies
3 region dummies
SMSA dummy
Size of establishment I V
Characteristics of workers in 3 digit sector V V
(age, education, sex, race)
Ratio production/nonproduction workers V V
5. R2 .442 .476
6. Number of cases in which standard deviations
in groupings are smaller by statistically significant
amounts (5% level) by group
a. Within 21 industries 8 1
b. Within 9 size of establishment groupings 6 0
c. Within 4 regions 3 1
aCharacteristics of workers
collar union and nonunion workers separately in each three digit industry from the May 1973 - 1975 Current Population
Survey tapes.
Source: Expenditures for Employee Compensation data tapes, 1967 - 72.
The "raw" differences in standard deviations
in line 2 differ by 0.044 or 15 percent,
yielding an F statistic of 1.37, significant at
better than one percent. The residuals from
the regression differ by 0.033, also 15 percent,
and have an F statistic of 1.30, also
significant at better than one percent. The
detailed comparisons in line 6 reveal significantly
lower dispersions among union
firms in eight of twenty-one industries compared
to a significantly higher dispersion in
just one industry; and significantly lower
dispersions in the union sector in six of nine
size categories and three of four regions.
The results in nonmanufacturing, summarized
in Table 6, tell a more complex
story, in part because of the peculiar distribution
of firms among sectors. In the EEC
union sample 54 percent of firms are in contract
construction compared to 24 percent
of nonunion firms. Given this differential
pattern and the high wages in construction,
the standard deviation of the log of hourly
compensation turns out to be higher in the
union sector (line 1), by a modest but statistically
significant amount, contrary to
our hypothesis. This result appears, however,
to be due entirely to the sample differences
noted above: when we look within industries,
the results are reversed. This is clear
in the regression of the log of hourly compensation
on the industry controls and
other variables. Because of the "peculiar"
cross-industry distribution of establishments,
the regression equation accounts for
71 percent of the variance in the log of earnings
in the union sector (largely via the industry
dummies) compared to 48 percent
of the variance in the nonunion sector.
"Corrected" for differences in characteristics,
the direction of the difference in
standard deviation is reversed. In line 3, the
residual dispersion of wages in the union
sector is a sizeable 0.062 or 22 percent less
than the residual dispersion in the nonunion
sector. Although not shown, the F
statistic between the variances is 1.63, which
is considerably larger than the F of 1.11 in
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 17
Table 6. Estimates of Dispersion of Total Compensation Per Manhour of
Nonoffice Workers Among Nonmanufacturing Establishments, by Union Status.
Union Nonunion
1. Number of Observations 2393 3621
2. Uncorrected for industry and other controls .416 .394
3. Corrected for industry and other controls .224 .286
4. Number of cases in which standard deviations
within groupings are smaller by statistically
significant amounts (5% level) within industries
(number of establishments in parentheses)
a. Contract Construction .240 .291
(1295 union, 865 nonunion)
b. Communication .173 .277
(40 union, 11 nonunion)
c. Electricity, Gas and Sanitary Service .134 .704
(109 union, 41 nonunion)
d. Wholesale Trade .289 .361
(129 union, 383 nonunion)
e. Banking, Credit and Security Services .149 .401
(2 union, 96 nonunion)
f. Rooming, Personnel and Miscellaneous Services .290 .353
(214 union, 495 nonunion)
g. Amusement and Recreation .683 .394
(9 union, 49 nonunion)
h. Educational .130 .350
(10 union, 75 nonunion)
Source: Expenditures for Employee Compensation data tapes, 1967 - 1972.
line 1. More importantly, comparisons of
standard deviations within 27 industry
groupings in line 4 show seven with significantly
lower dispersions among union
plants compared to one with significantly
higher dispersion in the union sector. Thus,
although somewhat more equivocal due to
a greater raw standard deviation for the
union sample, the EEC data on nonmanufacturing
tend to support the conclusion
that dispersion is lower in the union sector.
Impact on White-Collar/Blue-Collar
Differences
Thus far we have found that dispersion of
wages among organized blue-collar workers
is lower than dispersion of wages among
unorganized blue-collar workers for two
reasons: smaller impacts of wage-determining
factors on earnings and smaller
"residual variation" among similar workers.
Since we included occupation and skill
variables in our analysis and found smaller
coefficients on those variables, we have
dealt with the impact of unionism on bluecollar
skill differences as well as on the
union effect on workers within skill cate-
gories.
There is, however, one important effect
of unionism on within-establishment inequality
that has been ignored in our analysis,
namely, the effect of union wage gains
on the white-collar/blue-collar differential.
Since unionization occurs almost exclusively
among blue-collar workers and since
white-collar workers tend to be paid more,
on average, than blue-collar workers, the
union wage effect tends to reduce inequality
in this case.
The magnitude of the impact of trade
unionism on the white-collar/blue-collar
differential within the union sector can be
estimated by regression analysis using the
EEC data set. Let W = wage of blue-collar
workers; X = characteristics of blue-collar
workers; UN = 0- 1 dummy variable for
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
18 INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 7. Estimates of the Impact of Unionism on the Office Worker (White-Collar)/
Nonoffice Worker (Blue-Collar) Differential in Total Compensation Per Manhour,
by Sector.a
Manufacturing Nonmantufacturing
1. Number of establishments 4074 6014
2. Mean differential (standard deviation) .397(.356) .266(.506)
3. Regression coefficient (standard error)
on union dummy variable -.12 (.018) -.14 (.024)
4. Additional controls
a. Age dummies, blue-collar 2 2
b. Percent male, blue-collar
c. Mean education, blue-collar
d. Percent white, blue-collar V V
e. Age dummies, white-collar 2 2
f. Percent male, white-collar
g. Mean education, white-collar V V
h. Percent white, white-collar a I
i. Region dummies 3 3
j. Industry dummies 20 29
k. Ratio white- to blue-collar workers V V
1. Number of workers V V
5. R2 .152 .251
6. S.E.E. .329 .440
aThe dependent variable is in (total compensation per manhour of office workers/total compensation per
manhour of nonoffice workers).
Source: Expenditures for Employee Compensation data tapes, 1967 - 72.
collective bargaining coverage of bluecollar
workers; Ww = wage of white-collar
workers; and Xw characteristics of whitecollar
workers. Then the following regression
model can be used to estimate the
change in the differential within an establishment
due to unionism:
(5) ln(W, /W)j = ao
+ aXwj - a2X, - a3UNi + EI
where c is a random variable with mean 0
and variance cU2. The distinctive feature of
Equation 5 is that it exploits the existence of
data on blue- and white-collar workers
within the establishment to determine the
effect of unionism on intra-establishment
inequality. Because the dependent variable
refers to within-establishment differences,
any "omitted firm factor" that affects both
blue- and white-collar workers has similarly
been differenced away in the calculation.
The coefficient on unionism reflects the
impact of collective bargaining on bluecollar
wages relative to its (positive, negative,
or zero) "spillover" effect on whitecollar
wages.
Table 7 presents estimates of Equation 5
with the following set of control variables:
region and SMSA and industry dummies
comparable to those in Table 3, the mean
characteristics of white- and blue-collar
workers in the organized and unorganized
three-digit industry on which the establishment
is located (from the May CPS tapes, as
described earlier); unionism of blue- and
white-collar workers; and the ratio of whiteto
blue-collar workers in the establishment.
Line 1 of the table records the number of establishments
in the analysis; line 2 gives the
average white-collar premium in the establishments;
line 3 gives the coefficient on
the collective bargaining coverage dummy
variable; line 4 gives the additional controls
in the experiment while the remaining
lines give the summary statistics.
The calculations show that the whitecollar/blue-collar
differential is significantly
reduced by collective bargaining of
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 19
blue-collar workers in both manufacturing
and nonmanufacturing industries. In manufacturing
an average differential of .40
(line 2) is reduced by .12 points as a result of
collective bargaining of plant workers. In
nonmanufacturing, an average differential
of .27 is reduced by .14 points. By lowering
the income advantage of white-collar workers
in an establishment, trade union wage
gains reduce inequality in the organized
sector.
Total Effects of Unionism on
Wage Dispersion
Given the preceding estimates of the impact
of trade unionism on the dispersion of
wages among organized blue-collar workers
and on the white-collar/blue-collar differential,
what is the total effect of unionism on
dispersion in the organized sector? And how
does this effect compare to possible increases
in dispersion due to the union/nonunion
blue-collar wage differential?
The impact of unionism on the dispersion
of wages in the organized sector as a
whole can be evaluated by the standard conditional
variance formula. Let cy be the
variance in in wages for all workers in a sector
and Cy 2be the variance for production
(blue-collar) workers and a2 the variance
for white-collar workers and let a be the
production (blue-collar) worker share of
workers. Then if (WW - Wb) is the differential
between the mean in wages of whitecollar
and blue-collar workers,
(6) C2= dab + (I- a) C2,
+ a(1- a)(Wz - Wb).
Table 8 presents estimates of the total
effect of trade unionism on inequality in the
organized sector, using Equation 6. Line 1
records the standard deviation in the log
hourly earnings of production workers with
the nonunion dispersion of characteristics
from Table 1 and the estimated dispersion of
in hourly earnings if those workers were
organized, using the estimates in Table 4.
Line 2 records the contribution of the standard
deviations to the variance in in earnings
using Equation 6, with a estimated as described
in the table note.2' The next two
21Evidence given in Richard B. Freeman and James
L. Medoff, "Substitution Between Production Labor
lines estimate the contribution of the whitecollar/blue-collar
differential to the dispersion.
The estimate of the differential in
the absence of unionism is the actual in
differential between male white-collar and
blue-collar workers on the CPS file, uncorrected
for any differences in characteristics
on the assumption that these are, in fact,
different forms of productive labor. The
estimate of the differential in the presence
of unionism is obtained by deducting the
estimated effect of unionism on the differential,
using the regressions in Table 7.22 The
contribution of the differential to the variance
is estimated as a(1- a) (W,, - Wb)2
according to Equation 6. Line 5 records
estimates of the standard deviation of the
hourly earnings of white-collar workers,
using data from the May 1973 - 75 CPS file.
In this calculation we assume that the variance
of earnings among white-collar workers
is the same in the nonunion and union
sectors and can be approximated by the dispersion
among all white-collar workers.23
If there is a spillover of union standardization
policies on white-collar labor, we have
underestimated the overall impact of unions
and Other Inputs in Unionized and Nonunionized
Manufacturing," Review of Economics and Statistics
(forthcoming), Table 2, shows that the division of
production and nonproduction workers differs somewhat
between the union and nonunion sectors in
manufacturing in the EEC. In the union sector, production
workers constitute 65 percent of total manhours;
in nonunion manufacturing, production manhours
constitute 72 percent. This difference is too
small to merit analysis with different proportions.
22This is an approximation as the regression estimates
relate to the effect of unionism on the mean In
blue-collar/white-collar differential rather than to the
difference between the mean In blue-collar wage and
the mean In white-collar wage.
23 While this assumption seems sufficiently plausible
that any difference in dispersion of earnings among
white-collar workers could be attributed to "other
factors," it is important to make sure that there are
no enormous disparities that would raise the overall
dispersion in the union sector. Accordingly, I calculated
the standard deviation of total compensation per
manhour of office workers in the EEC sample, finding
figures of 0.29 in union manufacturing. 0.40 in nonunion
manufacturing, and 0.42 in union nonmanufacturing,
and 0.46 in nonunion nonmanufacturing.
Whatever interpretation is placed on these differences,
they do not run counter to the assertion that
overall dispersion is lower in the union sector.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
20 INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 8. Estimated Effects of Unionism on Within-Sector Dispersion of
Hourly Earnings.
Manufacturing Nonmanufacturing
Nonunion Union Nonunion Union
1. Standard deviation of earnings,
male blue-collar workers .398 .310 .451 .369
2. Contribution of line 1 to
variance in in earnings
( al )02 .116 .070 .114 .076
3. White-collar/blue-collar
differential .46 .34 .37 .23
4. Contribution of line 3 to
variance in earnings [(a) (1 - a) X
squared differential, line 3] .042 .023 .034 .013
5. Standard deviation of earnings,
male white-collar workers .442 .442 .554 .554
6. Contribution of line 5 to
variance in earnings [(1 - a ) X
square of line 5] .086 .086 .136 .136
7. Total variance of earnings
(lines 2 + 4 + 6) .244 .179 .284 .225
8. Standard deviation of earnings
(square root of line 7) .494 .423 .533 .474
Sources: Line 1: figures for absence of unionism from Table 1; figures for presence of unionism obtained by subtracting
estimated effect of unionism on comparable workers from Table 4, line 2b.
Line 2: estimated a , blue-collar share of male workers taken from CPS files: 0.73 for manufacturing and 0.56 for
nonmanufacturing. Note that CPS estimates outside manufacturing differ from production worker figures in
establishment surveys, in part because of different classifications and in part because of the large number of female
blue-collar workers outside manufacturing.
Line 3: figure for absence of unionism obtained by taking in differential between usual weekly earnings/usual
hours worked of all male white-collar workers and of nonunion blue-collar workers in the sample. Figure for presence
of unionism obtained by deducting estimated union effect from Table 7, line 3.
Line 5: calculated from May 1973- 75 CPS tapes, based on samples of 5568 manufacturing and 13,044 nonmanufacturing
workers.
in lowering inequality. Line 6 uses Equation
6 to estimate the contribution of dispersion
among white-collar workers to the total
variance.
The payoff to the calculations is in the
final lines of the table, which show the estimated
variance and standard deviation of
earnings in the absence of unionism and
estimated variance and standard deviation
of earnings in the presence of unionism. In
both the manufacturing and nonmanufacturing
sectors unionism is estimated to
reduce dispersion significantly. Unionism
lowers the standard deviation by .071 points
or 14 percent in manufacturing, and by .059
points or 11 percent in nonmanufacturing;
unionism lowers the variances by .065
points (27 percent) in manufacturing and by
.059 points (21 percent) in nonmanufactur-
ing.
The final issue to consider is how the estimated
dispersion-reducing effects of unionism
compare to the dispersion-increasing
effects of unionism due to the union wage
effect. Are the dispersion-reducing or the
dispersion-increasing effects of unionism
larger? Calculations designed to answer this
question are presented in Table 9. It uses the
conditional variance equation (Equation 6),
applied to union and nonunion male bluecollar
workers, to estimate the effect of
unionism on the dispersion of wages of all
male blue-collar workers and then considers
the contribution to wage dispersion of the
effect of unionism on the male white-collar/
blue-collar differential. Line 1 records estiThis
content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 21
Table 9. Comparison of the Estimated Dispersion-Decreasing and
Dispersion-Increasing Effects of Unions on the Earnings of Male Workers.
Manufacturing Nonmanufacturing
1. Estimated contribution of standardization of
wages to variance in In earnings of blue-collar
male workers -.048 - .033
2. Estimated change in variance of in earnings of
blue-collar male workers due to union wage effects
a. Comparable workers .004 .025
b. All male workers .008 .054
3. Estimated contribution of variance of in earnings
of blue-collar workers to total variance
a. Comparable workers -.044 - .008
b. All male workers - .040 .021
4. Estimated contribution of union-induced
reduction in white-collar/blue-collar differential
to variance of male earnings - .011 - .009 to - .021
5. Net effect of unions on variance of male earnings
a. Comparable workers -.043 - .013 to - .025
b. All workers -.039 .002 to - .010
Source:
Line 1: Multiplicand of estimated reduction in variance from Table 4, line 2b with union share of blue-collar male
labor from Table 1. Union share in manufacturing = 0.55; union share in nonmanufacturing = 0.40.
Line 2a: Multiplicand of union share of blue-collar male workers, nonunion share, and the square of the estimated
union wage effect, with wage effect of 0.12 in manufacturing and 0.32 in nonmanufacturing.
Line 2b: Multiplicand of union share of blue-collar male workers, nonunion share and the difference between the
square of the estimated differential between union and nonunion workers in the presence of unionism and the square
of the estimated differential between union and nonunion workers in the absence of unionism. The estimated
differentials are 0.07 in manufacturing and 0.19 in nonmanufacturing in the absence of unions; 0.19 in manufacturing
and 0.51 in nonmanufacturing in the presence of unions.
Line 3: Sum of line 1 and lines 2a or 2b.
Line 4: The differential in the presence of unionism is the actual differential on the CPS files: manufacturing 0.36,
nonmanufacturing 0.26. The differential in the absence of unions in manufacturing is 0.43, the sum of 0.36 and the
multiplicand of the estimated effect of unionism on the white-collar/blue-collar differential from Table 7 and the
union share of blue-collar labor. The first estimated differential in the absence of unions in nonmanufacturing is
0.32, the sum of 0.26 and the multiplicand of the estimated effect of unionism on the white-collar/blue-collar
differential from Table 7 and the union share of blue-collar labor. The second estimated differential in the absence of
unionism in nonmanufacturing is 0.39, the sum of 0.26 and the multiplicand of the estimated union wage effect on
the CPS file and the union share of labor.
Line 5: Sum of line 4 and blue-collar share of male workers multiplied by lines 3a or 3b.
mates of the effect of the reduced dispersion
of wages among blue-collar workers, by
sector. It is obtained by multiplying the
estimated effect of unions on comparable
workers (Table 4, line 2b) by the proprotion
of male blue-collar workers who are organized,
as given in the figures in Table 1.
Since roughly half of male workers are union
members, the figure in line 1 is about
half of the estimated effect of unions on the
variance of blue-collar workers.
Line 2 presents two separate estimates of
the dispersion-increasing effect of the union
wage differential on the variance. Line 2a
records the effect of union wage gains on the
dispersion among comparable workers, obtained
by estimating the union wage effect
in the CPS data. Controlling for all other
characteristics included in Table 2, the
union premium among male blue-collar
workers was estimated to be 0.12 in manufacturing
and 0.32 in nonmanufacturing.24
24 This analysis is based on a regression of log hourly
earnings on all of the control variables listed in Table
3. In manufacturing, with the smaller set of controls,
the union coefficient and standard error were 0.12
(.005), with the larger set of controls, the coefficient
and standard error were 0.11 (.005). In nonmanufacturing
the estimates were 0.32 (.006) with smaller set of
controls, and 0.311 (.006) with larger set of controls.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
22 INDUSTRIAL AND LABOR RELATIONS REVIEW
Following Equation 6, the premium is
squared and multiplied by the relevant
shares of union and nonunion workers
among male blue-collar workers. Line 2b
uses the estimates of the union premium and
the differential between union and nonunion
blue-collar men, uncorrected for differences
in characteristics, to evaluate the
dispersion-increasing effect of unions on
male blue-collar labor. The estimates in 2b
take account of the fact that unionized labor
tends to be more skilled and therefore higher
paid than nonunion labor, even in the absence
of organization. The estimates are obtained
in accordance with Equation 6 by
subtracting the square of the differential between
union and nonunion workers in the
presence of unionism and the estimated
differential in the absence of unionism. The
difference in the presence of unionism is the
"raw" uncorrected difference while the estimated
differential in the absence of unionism
is the raw differential minus the estimated
union wage effect.
The results of these calculations are clear.
Among manufacturing workers, the dispersion-reducing
effects of unions dominate
the dispersion-increasing effects, both
among all blue-collar workers and among
comparable blue-collar workers. Among
nonmanufacturing workers, on the other
hand, the net effects are small, with a slight
reduction in dispersion in the comparable
workers' calculation and an increase in the
all workers' calculation.
Line 4 contains estimates of the change
in the variance of earnings due to the impact
of unionism on the white-collar/blue-collar
differential. In both manufacturing and
nonmanufacturing, the differential in the
presence of unionism is taken to be the actual
differential in the CPS files for the relevant
sector. In manufacturing, the differential
in the absence of unionism is estimated
as the actual differential plus the multiplicand
of the union share of blue-collar labor
and the estimated effect of unions on the
white-collar/blue-collar differential in
Table 7. In nonmanufacturing, due to the
There were 15,480 observations in nonmanufacturing
and 15,174 in manufacturing.
significant difference between the estimated
effect of unionism on the white-collar/bluecollar
differential in the EEC file (0.14) and
the estimated union wage effect in the CPS
file (0.32) (a pattern that is consistent only
if unions have a major "spillover" effect on
white-collar labor), two estimates are given.
The first (lower) estimate uses the EEC
estimates of the effect of unionism on the
white-collar/blue-collar differential (multiplied
by the union share of blue-collar
labor) while the second estimate assumes
that the differential is lowered by the CPS
estimated union wage effect (multiplied by
the union share of blue-collar labor). Because
in manufacturing the estimated effect
of unionism on the white-collar/blue-collar
differential in the EEC file and the estimate
of the union wage effect in the CPS file are of
the same magnitude, only one estimate is
necessary on that sector.
The final line gives our estimates of the
impact of unions on dispersion of male
workers in both sectors. In manufacturing,
the effect is clearly negative and reasonably
large. In nonmanufacturing, where the
union share of employment is lower and the
estimated union/nonunion wage differential
larger, the direction of the effect is
smaller and dependent on whether EEC or
CPS estimates of the effect of unionism on
white-collar/blue-collar differentials are
used.
Finally, what about the effect of unionism
on the manufacturing/nonmanufacturing
differential? Focusing once more on production
workers, the contribution of unionism
to the differential wages in manufacturing
versus nonmanufacturing can be estimated
from the evidence on proportions of
workers organized in the two sectors, the
size of the union wage effect, and the basic
differential between the sectors. Since unionism,
by our estimates, raised the wages
of 55 percent of male production workers
in manufacturing by 0.12 points compared
to an effect of 0.32 points on 40 percent of
male production workers in nonmanufacturing,
the overall wage of blue-collar manufacturing
workers may have risen by 0.06
points compared to an increase of 0.13
points in nonmanufacturing. The differential
between nonunion workers in the two
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms
UNIONISM AND THE DISPERSION OF WAGES 23
sectors was 0.11 in favor of manufacturing
workers. If, in fact, unions raised the earnings
of nonmanufacturing workers by 0.07
points relative to manufacturing workers,
the net effect would be to reduce inequality
by lowering the manufacturing advantage
from 0.11 to 0.04, reducing dispersion by a
modest amount, according to the relevant
part of Equation 6.
In sum, the dominant effect of unionism
on dispersion is through the reduction in
intrasectoral dispersion, with the sizeable
reduction in dispersion in manufacturing
leading to an overall reduction in inequal-
ity.25
Conclusions, Provisos, and Implications
The major findings of this study can be
summarized briefly as follows:
(1) Trade unions have adopted wage
policies designed to reduce dispersion of
earnings within and across establishments,
for what can be rationalized as plausible
economic reasons.
(2) Other things equal, the dispersion of
earnings is significantly lower among organized
blue-collar workers, in part because
of a reduced effect of standard wage-determining
factors on earnings and in part because
of smaller dispersion within categories
of workers having the same wage-determining
characteristics.
(3) Dispersion of average compensation
is also lower among establishments that are
unionized than among those that are not.
(4) Unionism reduces white-collar/blue-
25These calculations are, it should be stressed, consistent
with Lewis's estimates of the enlarged dispersion
of the interindustry wage structure due to unionism.
According to Lewis (p. 292), unionism may have
increased dispersion of wages among industries by
0.022 to 0.026 in points in 1958. Assuming a similar
impact in the 1970s, we see that the dispersion-reducing
effects in line 1 of Table 9 exceed these figures.
Note that Lewis's estimated 8 percent increase in
dispersion (p. 295) takes as its base the dispersion of
wages among industries, not the dispersion among
people. It is thus appropriate to use his estimates of the
in point effect of unions rather than of the percentage
effect of union wage gains on dispersion.
collar earnings differentials in the organized
sector, further contributing to within-sector
reductions in dispersion.
(5) Overall, the within-sector effect of
unionism on dispersion appears to more
than offset the increase in dispersion of
earnings across industries, so that on net
unionism reduces inequality.
The major weakness with these findings
is that, despite our efforts, at least some of
the lower dispersion in the union sector may
be due to inadequately controlled characteristics
of organized workers. While our controls
have been at least as complete as those
in other studies, the possibility that more
narrow groupings would reduce the difference
in variances cannot be denied. In addition,
to minimize problems of comparability
we have dealt exclusively with male
workers, which leaves open the possibility
that if (for some unknown reason) unionism
has a different effect on the dispersion of
wages of women, our results may not be
generalizable. To the extent that reduction
in dispersion of earnings leads to reduced
dispersion of abilities in the union sector,
there may be little or no economywide effect
on inequality or efficiency.
Assuming that the result is correct, however,
the welfare implication of a reduction
in wage inequality due to unionism is by no
means clear. To the extent that the dispersion
of wages reflects disequilibrium, the
influence of peculiar nonmarket forces, and
the failure of the market to bring about
"equal pay for equal work," the reduction in
dispersion among comparable workers may
have desirable efficiency implications. To
the extent that the dispersion in the absence
of unionism reflects dispersion of marginal
products perfectly, however, the reduction
among comparable workers may cause inefficiencies.
Finally, of course, the distributional
effects of standard wage policies must
also be weighed in any assessment of their
effect on welfare. Normative issues aside,
what is important is that unions have a sizeable,
generally neglected, impact on withinsector
dispersion, which may be attributable
to their standardization-of-rate policies.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:22 UTC
All use subject to http://about.jstor.org/terms