Union Wage Practices and Wage Dispersion within Establishments
Author(s): Richard B. Freeman
Source: ILR Review, Vol. 36, No. 1 (Oct., 1982), pp. 3-21
Published by: Sage Publications, Inc.
Stable URL: http://www.jstor.org/stable/2522290
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 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION
WITHIN ESTABLISHMENTS
RICHARD B. FREEMAN*
This study analyzes establishment-level data primarily to examine the
effect of unionism on the wage structure within establishments. The major
finding is that within-establishment dispersion of wages is significantly
narrower in unionized than in nonunionized establishments, a pattern the
author attributes in large part to unions' wage practices, such as single rate or
automatic-progression modes of wage payment as opposed to merit reviews
and individual wage determination. The data also show that dispersion in
average wages is narrower among organized plants, but by more modest
amounts than the within-establishment differential. Overall, the evidence
suggests a major role for explicit union wage policies in explaining the dispersion
of wages within firms and in the economy as a whole.
ONE of the principal goals of trade
union wage policies has been to reduce
dispersion of wages through standard rate
policies. These policies seek to obtain
"equal pay for equal work" across establishments
and to reduce "inequities" and
differentials based on perceived personal
characteristics rather than on specific job
tasks. Recent work on dispersion of wages
among union workers and among nonunion
workers has suggested that these policies
have produced markedly lower dispersion
in the union sector.' Because of a pau*The
author is a professor of economics at Harvard
University. He has benefited from the research
assistance of John Ballen, Joyce Jacobsen, and Mark
Hindal. This work was supported by a National
Science Foundation grant and is part of the National
Bureau of Economic Research's labor studies program.
'Richard B. Freeman, "Unionism and the Dispersion
of Wages," Industrial and Labor Relations Review,
Vol. 34, No. 1 (October 1980), pp. 3- 23.
city of data on establishments, as opposed
to individuals, and on establishment wage
practices, however, this work did not document
the effect of unions on wage dispersion
within establishments, nor did it show
the link between wage practices favored by
unions and dispersion.2 As a result, the literature
currently lacks any estimates of the
impact of union wage policies on the wage
structure within establishments.
The present paper seeks to remedy this
gap in our knowledge by using data from the
Bureau of Labor Statistics' Industry Wage
Survey3 on wages for workers within estab-
2While Slichter, Healy, and Livernash, among
others, suggest that unions reduce wage dispersion
through single-rate policies, no quantitative relationship
has yet been established. See Sumner H. Slichter,
James J. Healy, and E. Robert Livernash, The Impact
of Collective Bargaining on Management (Washington,
D.C.: The Brookings Institution, 1960), pp.
592- 623.
31n the Industry Wage Survey, the Bureau of Labor
Industrial and Labor Relations Review, Vol. 36, No. 1 (October 1982). ? 1982 by Cornell University.
0019-7939/82/3601-0003$01.00
3
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
4 INDUSTRIAL AND LABOR RELATIONS REVIEW
lishments and on the explicit wage practices
of organized and unorganized establishments.
The data on within-establishment
wages are used to calculate measures
of dispersion of wages by establishment.
These measures of establishment-level dispersion
are then related to the union status
of the establishment and to its explicit wage
practices.
Union Wage Policies
Within Establishments
With rare exception, unions have sought
to reduce differentials among workers with
nominally similar skills and job tasks
within establishments. They normally seek
such reduction through two types of wage
policies: a single rate of pay for each occupational
group and a seniority-based progression
of rates up to a maximum. Single-rate
policies, with one level of pay for all workers
in a specified job category, reduce dispersion
more than do other wage policies, but even
seniority plans with progression to a maximum
level tend to exert substantial control
over dispersion by requiring similar treatment
of workers who have the same seniority.
Because of union desires for reduction of
differentials, many progression plans have
tended, moreover, effectively to meld into
single rate maxima.4 Union pressures to reduce
the range of rates within job categories
and to expand the number of job titles included
within categories further narrow
the dispersion. Overall, Slichter, Healy, and
Livernash conclude that because of these
two avenues of influence-single rates and
automatic progression plans-the role of
unions in within-firm wages has "clearly
Statistics provides information on most workers in
establishments sorted into specific four-digit industries.
The only workers who are not in the sample are
those lost because the BLS is unable to categorize them
into a detailed occupation. The survey also includes
data on establishments.
4Slichter, Healy, and Livernash found, for example,
that at the time of their study, in a major unionized
company with a wage plan providing automatic
progression within a range of wages, 87 percent of
the workers had reached the top of the range and thus
were effectively paid a single rate. See Slichter, Healy,
and Livernash, The Impact of Collective Bargaining
on Management, p. 605.
been one of minimizing and eliminating
discriminatory judgement-based differences
in pay for individuals employed in
the same job."5
At the other end of the spectrum, unions
have generally opposed merit review and
individual-determination payment plans,
with the result that in the early 1970s, when
43 percent of all major companies used
merit review for blue-collar workers, just
12.5 percent of major union contracts contained
a merit progression plan.6 Moreover,
since unionized workers under contracts
with merit plans who are not given increases
can file and win grievances, many union
merit plans also resemble automatic progression
or single-rate plans. All told,
unions have been very successful in removing
performance judgments as a factor in
determining individual workers' pay.
Three basic factors appear to explain
union preference for reduced differentials
within firms. First, there is the belief of
many workers that existing rate differentials
reflect favoritism and discrimination rather
than relative job duties and responsibilities.7
"Inequity problems" have historically
plagued numerous industries, leading
workers and their organizations to prefer
"objective" standards to evaluation of individuals
based on the subjective decisions
of foremen.8 As long as supervisors are imperfect
and make decisions based on criteria
other than "true" contributions to the
firm (which are exceedingly difficult to
measure), it is reasonable to expect employee
preference for narrow job-related
rates. In a world characterized by the Rawls
51bid., p. 602.
6Bureau of National Affairs, Wage and Salary Administration
Survey 97 (Washington, D.C.: B.N.A.,
July 1972), p. 14, Table u; and U.S. Department of
Labor, Bureau of Labor Statistics, Characteristics of
Major Collective Bargaining Agreements (Washington,
D.C.: G.P.O., July 7, 1974), p. 33, Table 3.5.
'Richard Lester and E. Robie, Wages Under National
and Regional Collective Bargaining (Princeton,
N.J.: Princeton University, 1946).
8Prior to the comprehensive wage study in the steel
industry, for example, pay within a company was often
fixed by department supervisors without relation to
rates elsewhere. See Lloyd Reynolds and Cynthia
Taft, The Evolution of the Wage Structure (New
Haven, Conn.: Yale University Press, 1956), pp.
45-46.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION 5
" veil of ignorance" where workers do not
know whether they will benefit or lose from
seemingly arbitrary supervisory decisions,
simple "maxi-min" behavior will dictate
preference for narrow ranges of rates.9
Finally, workers with risk-averse preference
functions or with preferences for a
narrow distribution of relative wages will
favor standard rate policies rather than a
wide range of wage rates.
Second, there are organizational reasons
for expecting unions to favor narrow ranges
of rates. Viewing the union as a political
organization dependent on average (median)
worker preference, one can see that
when the median wage is less than the mean
wage, a majority of workers will favor redistribution
to the lower paid, which can
lead to wage policies reducing inequality.'0
In other words, the median worker who
receives less than the mean wage will favor
redistribution; hence, 50+ percent will prefer
standard-rate policies. Finally, worker
solidarity and organizational strength are
also more likely to be greater when workers
receive roughly the same pay than when
they receive very different levels of pay.
Whatever the reasons for union pressures
for reducing wage inequality within establishments,
such policies are a fact of
economic life. Do they have a significant
effect on the typical organized establishment?
Is dispersion lower within organized
rather than unorganized plants and, if so,
by how much? To what extend can any observed
differences in wage dispersion within
establishments be attributed to specific
wage practices associated with collective
bargaining? In short, how important are
union policies for reducing dispersion
within establishments in the labor market?
Data
To answer these questions, it is necessary
9Rawls's "veil of ignorance" refers to the notion
that, given complete lack of knowledge of how he or
she will be treated, a worker will tend to be conservative
in accordance with the analogue of maxi-min
principle. See Jack Rawls, A Theory of Justice (Cambridge,
Mass.: Harvard University Press, 1971).
"Richard B. Freeman, "Individual Mobility and
Union Voice in the Labor Market," American Economic
Review, Vol. 66, No. 2 (May 1976), pp. 361-68.
to have data on (1) wages of workers within
establishments, (2) establishment wage
practices, and (3) the union status of the
establishments. In contrast to widely used
data files on individuals," establishment
data sets with information on individuals
within establishments are exceedingly rare.
One of the few such data files is provided by
the Industry Wage Surveys of the Bureau of
Labor Statistics. These surveys, conducted
since the 1940s by the Industry Wage Division
of the Bureau of Labor Statistics, are
designed to obtain information on wage
levels and practices within firms for the
purpose of aiding establishments in understanding
their labor market environment.
The surveys cover a random sample of establishments
in major four- (or in some
cases, three-) digit SIC industries and provide
data on distribution of wage payments
unavailable elsewhere.
For purposes of this study, the Industry
Wage Surveys have several major advantages.
They obtain data on the wages of
individual workers within establishments,
information that allows for the calculation
of within-establishment dispersion of
wages. They also obtain data on how many
production workers in each establishment
are paid under the following types of payment
plans for time rates: (1) individual
determination, (2) range of rates with merit
review, (3) range of rates with automatic
progression, (4) combination of range of
rates with both merit review and automatic
progression, (5) single rate, and (6) incentive
rates of pay.12 They categorize each produc"1Such
as the Current Population Survey and the
National Longitudinal Survey.
'2These methods of wage payment are defined by
the BLS survey as follows:
Formal rate structures for time-rated workers provide single
rates or a range of rates for individual job categories. In the
absence of a formal rate structure, pay rates are determined
primarily by the qualifications of the individual worker. A
single rate structure is one in which the same rate is paid to all
experienced workers in the same job classification. (Learners,
apprentices, or probationary workers may be paid according to
rate schedules which start below the single rate and permit the
workers to achieve full job rate over a period of time.) An
experienced worker occasionally may be paid above or below
the single rate for special reasons, but such payments are exceptions.
Range-of-rate plans are those in which the minimum,
maximum, or both of these rates paid experienced workers for
the same job are specified. Specific rates of individual workers
within the range may be determined by merit, length of service,
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
6 INDUSTRIAL AND LABOR RELATIONS REVIEW
tion worker by an industry-specific detailed
occupational code, such as card grinder
in the wool textile industry, which permits
more precise controls on occupation and
skill than in most data sets widely used by
economists.
At the same time, the Industry Wage
Surveys have some disadvantages. Except
for sex, they lack information on the human-capital
(age, education, and training)
and personal characteristics (race and marital
status) of all workers and on those production
workers whose occupations are not
classified in the survey. Because of the narrow
occupational categories, however, it is
unlikely that the absence of these data is a
tremendous drawback.
This study will consider nine four-digit
industries in which time rates are the usual
structure of payment. It focuses on time rates
because the standardization of piece rates
has no clear effect upon wage dispersion,
the dispersion of the effective hourly rate
for piece-rate workers depending also on
the dispersion of productivity among workers
within establishments.
The characteristics of the nine-industry
sample under study are listed in Table 1.
The sample was chosen so as to include a
sufficient number of both union and nonunion
establishments to permit comparisons.
In total the data include information
on nearly 3,000 establishments, 49 percent
of which are organized, and on 500,000 individual
workers, 45 percent of whom are
organized-an exceedingly large number
of observations even by modern labor economic
standards. Selecting a set of industries
in this manner does not provide a random
sample. Picking industries with
stronger or weaker union organization
might have produced somewhat different
results.'3
or a combination of these. Incentive workers are classified under
piecework or bonus plans.
See U.S. Department of Labor, Bureau of Labor
Statistics, Industry Wage Survey:Industrial Chemicals,
Bulletin 1978 (Washington, D.C.: G.P.O., June 1976),
p. 48.
"5The Industry Wage Surveys cover about 40 private
sector manufacturing industries. Note that while we
selected industries so as to have a reasonable number
of both union and nonunion establishments, there is
nothing artificial or biased in our selection.
A surprisingly large number of unions
are represented among the organized workers
within the industries chosen. In the industrial
chemicals industry, for example,
organized firms in the survey are covered
by the International Chemical Workers
Union; the Oil, Chemical, and Atomic
Workers Union; the United Steelworkers
of America; and single company and local
unions not associated with national or
international unions.
The Dispersion of Wages
The main dependent variable in the study
is the dispersion of wages within establishments.
Dispersion is measured by the variance
and standard deviation of the natural
log (ln) of wages, metrics that are appropriate
if wages follow the lognormal distribution
or the In earnings function widely used
in empirical work. All of the analyses are
also performed using the variance of wages
in dollar units. Use of natural rather than
In units strengthens all of the findings reported
in this paper.
To obtain the variance of the In of wages,
we took the In of wages of workers in each
establishment and then calculated the mean
In wage and its properly weighted variance.
This statistic was calculated for all
production workers in an establishment,
for male and female workers separately,
and for workers in the major industryspecific
occupations.
Since the dispersion of wages within an
establishment is a variable not widely examined
in labor market analysis, it is of
some importance to examine its distribution
in the sample. On average, the within-establishment
variance has a mean of .017, which
compares with a variance of .028 for wages
across establishments.
Table 2 decomposes the total mean sum
of squares of wages into the between-establishment
and within-establishment components
and then further decomposes the
within-establishment sum of squares into
between-occupation and within-occupation
components. While there are notable
differences across industries, the table shows
substantial dispersion both within and between
establishments and within and beThis
content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION 7
Table 1. Characteristics of the Sample.
Number of
Industry Union Number of Number of Workers per
Status Workers Establishments Establishment
Paints & Total 10,941 291 37.6
Varnishes Union 7,734 179 43.2
Nonunion 3,207 112 28.6
Textile Dyeing Total 19,739 148 133.4
& Finishing Union 8,875 73 121.6
Nonunion 10,864 75 144.9
Cotton, Man-made Total 151,150 306 494.0
Fiber Textiles Union 28,524 42 679.1
Nonunion 122,626 264 464.5
Total 10,651 56 190.2
Wool Textiles Union 2,678 19 140.9
Nonunion 7,973 37 215.5
Industrial Total 71,659 269 266.4
Chemicals Union 57,330 199 288.1
Nonunion 14,329 70 204.7
Wood Household Total 37,079 330 112.4
Furniture Union 13,306 137 97.1
Nonunion 23,773 193 123.2
Miscellaneous Total 70,354 875 80.4
Plastic Products Union 36,749 397 92.6
Nonunion 33,605 478 70.3
Fabricated Total 23,077 331 69.7
Structural Steel Union 17,700 235 75.3
Nonunion 5,377 96 56.0
Nonferrous Total 18,199 363 50.1
Foundries Union 11,629 178 65.3
Nonunion 6,570 185 35.5
Source: Computed from the Bureau of Labor Statistics Industry Wage Survey.
tween occupations within establishments.
On average, 40 percent of the variance in in
wages of production workers in a detailed
industry is due to the within-establishment
differentials of concern here. Of the withinestablishment
variance in In wages, on
average, 66 percent is due to within-establishment
differences in occupational
means.14
Now that we have some notion of the
magnitude and nature of within-establishment
dispersion of wages, we turn to
the question of concern in this paper: differences
in dispersion between organized and
14Both of these figures are based on unweighted
averages of the fractions of variance for each industry
from Table 2.
unorganized establishments.
Unionism and In variance of wages within
establishments. The results of the first
stage of our analysis are presented in Table
3. It contrasts the variance in In wages in
union and nonunion establishments in
several ways, each of which strongly supports
the conclusion that within-establishment
dispersion of wages is much smaller in
union than in nonunion settings. Columns
1 through 3 contrast the mean variance of In
wages for organized and unorganized establishments;
the differences in means show
unionized plants with a lower variance and
the t-tests show these differences to be significant
in six of the nine cases. To make
sure that these results are not due to different
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
8 INDUSTRIAL AND LABOR RELATIONS REVIEW
04 G 4 ..I kn 04 0= 0 00 (0t-, 00 00 00 r ~ (
WS t N <, Q t : s 1n bc' cr ' > o o > o
O0 04 < t X 00 i 04 04 C 04 00 I
0 ~ ~ ~ ~ ~~~~~~~~~~0
~~~~~0 0 ~ ~ ~ ~ ~ ~ ~ 00
C (00 -0 00 0 " ? 0 V
t - 0 - 0 0 0 0 oo 0t - I
00 0 0
oo
(I) 0~~~~~~~~~~~~n 0c
(/): dqKt 0 D o 00 K> I~ - . (
-m - 04 -(0 K> '0 00 0-:
ts- 0 0 0 - 04 - ? sy
0 0 00 0 C
0. bf~~~~~~~~~b
0~~~~~~~~~C
0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~:
N 4 00 u~kf 004 0 00 0
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION 9
characteristics of union and nonunion establishments,
we ran regressions for each of
the nine industries using the variance of the
in wage in each establishment as the dependent
variable and independent variables
including a 0- 1 dummy variable for
unionism and several control variablessize
of establishment, region, and the percentage
of workers in each detailed occupational
category (to eliminate the possibility
that differences are due to concentration of
workers in occupations with low variances
of wages in the union sector). The resultant
union coefficients and standard errors, recorded
in column 4, show that the differences
in means are not attributable to differences
in characteristics. In seven of the nine
industries, the union coefficient is significantly
negative at a 95 percent confidence
level, and in six of nine the coefficient is
significant at a 99.5 percent confidence
level.
For each industry, we also ran regressions
using the within-establishment variance of
In wages for a particular occupation as the
dependent variable. The number of significant
union coefficients in these regressions
is displayed in columns 5 through 9. The
evidence confirms that unions reduce wage
dispersion within occupations within establishments.
Unions decrease the variance of
In wages in 124 of the 176 detailed occupations,
significantly so at 10 percent in 68
occupations and at 5 percent in 51 detailed
occupational groups. Correspondingly, the
union coefficient increased the variance in
only 52 of the 176 occupations, significantly
at the 10 percent level in 14 and at the 5 percent
level in only 8.
Corresponding estimates for the standard
deviation of In wages, rather than for the
variance of In wages, are presented in Table
A in the appendix. Because many analyses
of income distributions focus on standard
deviations, it is useful to examine those
results as well, in large measure to evaluate
the magnitude of the estimated union effects.
What stands out in the calculations is
the fact that the union effect is not only
highly significant in all but one case, but
also large in absolute magnitude. These
calculations show standard deviations of
wages in the union sector averaging .0268
units-or 22 percent-below the standard
deviation in the nonunion sector. We conclude
that in the sample under study, unionism
is associated with markedly lower dispersion
of wages within establishments.
Unionism and dispersion-reducing wage
practices. We consider next the routes by
which unionism reduces within-establishment
wage dispersion. The greater our ability
to relate the union effect to specific wage
practices favored by unions, the greater are
our understanding of the nature of the results
and our willingness to attribute them
to unions as economic institutions. According
to the standard-rate hypothesis set out
earlier, we expect unionism to increase the
proportion of workers covered by the most
egalitarian wage systems-single rate and
automatic progression-and to reduce the
proportion of workers covered by those
systems allowing greater dispersion and
managerial discretion, merit review, and
individual determination.
Table 4 presents the results of analyzing
the effect of unionism on the different methods
of wage payment used within establishments.
It records the mean percentage of
workers in union and nonunion establishments
enrolled in five time-rate payment
plans presented more or less in order of their
likely impact on dispersion, from the plan
with potentially the least dispersion to the
plan with potentially the most dispersion.
It also records the union coefficient and its
standard error from a regression of the percentage
of workers in each firm in the payment
plan on unionism and the average
worker's wage, the ratio of male to female
production workers, the ratio of office to
production workers, region size, and occupational
independent variables.
In all nine industries, unions increase the
percentage of workers paid by single-rate
plans and decrease the percentage paid by
individual determination. In all but cotton,
man-made fiber textiles, the effects are large
and statistically significant. In miscellaneous
plastics, for example, an average of 63
percent of workers in union plants are covered
by single rates compared to 12 percent
of workers in nonunion plants, whereas at
the other end of the spectrum, just 4 percent
of union compared to 49 percent nonunion
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
10 INDUSTRIAL AND LABOR RELATIONS REVIEW
0
c OC D - N N 0 0? tC - 0
g ~ ~~~~~~~ G| .O 0t t' 0. | O - - l > Z
0 C3 E 0 0 0 0 O~
:0~~~~~~~~~~~~~~~~~~~~~
C,)~ ~ ~ ~ ~ ~ ~ ~ ~ ~~~~~C
W T o t o - _ to _ 0
K L A D D ?.') ? ?) 1 CCl) ~ ~ 0C - C ) C ~>0
~~~ t~~~~~- ~~~ C-1 0 1~0 C-1
S .< e f > : z ? F r S 2 E 3; >; n 2 i~~~-e
ho ts zI G) G) 0- 0 C.
4-j ~ ~ ~ - - - 0 4-QJ
S CZ
C~~~~ ~~~~ ~-1 C(1 C)J - 0 , [O - 0
C-4 C) C) C) C) ~ ~ ~ -~ J
Q) Ct~~tCJ ~ ( G)~ ~~ ~ cCJ - cCj~~
0 0) 00 0 - 0 n 0 - 0 00 0 0 00 0) 04 0 0
Q) u ct ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~~ c ~
0 0 0 0~~~ 00 0 --
C~ 0 0 0-C~ (
~S m 0 0 0 0 0
0 0 ~~~~ ) O 0 -
00 O) -0 0O t- c O~sC
0 i 0o 0 0 0
_ _ _ _ _ _ _ _ _ _ _ _ _ o -~ c~ -
C) CIA~~~~~~~~~~~~~.
m
CZ
M
Ct ~ ~
~~ c~~~e~~ QJQJ~~
-~~~~ ~ ~ ~~ ~ ~ ~ ~ HH~~~~~~ci~~~mb
~~~~~~~~~~~~~~Cs-.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION
-~0
-0 ~~~~~~~~~~~~~~~~~
0~
0
~~~~~~~~~~~~~~~~~~~~
CS~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~C
0.) 0 &~~~~~~~~~~~~~~~~~C4 ) C
.2 0 ~~~~~~~~~~~~~~~~~~~~
~~~ 0 0 CO N 0~~~~~~~~~~~~~~~C CO CO 0
0
Q.) 0~~~~~~~~~~~~~~~~~~~~~~~
0I~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~CZ
00 m~~~~~~~
-0 . ~~~~~~~~~~~~~~~0
A4 C4~~~ 0 $-4
~~~~ ~ ~ ~ ~ ~ ~ 0~0- 0, 00 -c -= N
0.) . CZC
v v- 00 t- Noc~c ~c r-, to tO 00 Gf) '1 0-',
.0 C) C-4 C> (=> 0~~~~~~~~~~~~~~.
-0 ~~~~~~~~~~~~~~~~~~. -.~~~~~~~~~~~~~b) 0.
00 00 0) 00 O 0Z
ct .~0- 000 Q
W 01
C~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~C
CO
Ln bo m C-4 u~~~~~~~~~~~~~00 bJ0
CO Q)
0 -~ CO 00 CO N 04- 0 o
0~~~~~~~~~~~~~~)Q 0
ctCO
C-4 "r 0 in 00 0T 02
00! c H.Ou o
-~~~~~~ ~ ~ ~ ~ ~ 00~~~~~~~~~~0~~Q) (
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
12 INDUSTRIAL AND LABOR RELATIONS REVIEW
workers are paid by individual determination
of rates. Moreover, only when the efforts
of unions are totally directed toward single
rates of pay, such as in the industrial chemicals
industry, do unions appear to adversely
affect automatic progression plans. The
conclusion that unionism is associated
with establishment wage practices likely to
reduce dispersion is inescapable.
Method of payment as an interven ing variable.
To determine the extent to which
method of wage payment, particularly the
adoption of single-rate wage practices,
accounts for the lower dispersion in organized
establishments shown in Table 2, we
have added to the regressions of the variance
of In wages on unionism and other controls
of Table 3 the fraction of workers under all
wage systems, with the single-rate factor
deleted to prevent singularity. If the union
coefficient is substantially reduced by the
addition of the variables measuring wagepractices,
then we can conclude that explicit
wage practices are a major intervening variable
in the relationship between unions
and within-establishment dispersion.
The results of the calculations, given in
Table 5, confirm that in all but one case
the addition of the methods of payment variable
reduces the negative effect of unionism
on the variance of wages. The union coefficients
are reduced in eight of the nine sectors
and by significant amounts in three.
In four of the industries over 50 percent of
the difference in variances between union
and nonunion firms is explained by wage
practices, and in another two cases, 20- 30
percent of the union coefficient is so explained.
In two industries, addition of the
methods of wage payment actually changes
the wage coefficient from negative to positive.
Thus, for a majority of the industries
studied, a significant portion of the union
effect is explained by the method of wage
payment.
As for the effect of the methods of wage
payments themselves, the coefficients for
virtually all of the time-rate methods are
positive, indicating that all of the payment
methods raise the variance of the firm wage
relative to the omitted single-rate method
promoted by unions. (The one exception is
the automatic progression payment plan in
the wool textile industry, which has a negative
coefficient). Moreover, as might be expected,
the individual-determination coefficients
tend to be sizable (in all but the cotton,
man-made fiber textiles and fabricated
structural steel industries) coefficients
(seven of nine cases) are larger than the combination
merit review and automatic progression
coefficients and all are larger than
just the automatic progression coefficients.
Finally, eight of the nine combination merit
review and automatic progression coefficients
are larger than just their automatic
progression plans. In sum, the contribution
of payment plans to variance of wages
within firms is as expected, and the differences
in use of plans is a major component
of the observed differences in inequality.
Interpreting the Results
The finding that unionism is associated
with markedly lower dispersion of withinestablishment
wages and with explicit wage
practices that have a significant effect on
the dispersion of earnings is consistent with
the hypothesis that standard-rate policies
have a sizable impact on establishment
wages and, more generally, with one of the
major contentions of institutional labor
economics: that explicit policies of market
organizations (unions and firms) can affect
market outcomes. Surely the most immediate
interpretation of our results is that union
(and firm) policies greatly affect the pattern
of wages.
One may, however, object to reading
causality into the statistical analysis because
of the possible endogeneity of the
union organization or the wage practices.
With respect to the union effects, perhaps
the inverse relation between unionism and
dispersion simply reflects the greater likelihood
that unions organize low-dispersion
firms. For instance, some may argue that
workers in such firms are likely to be more
homogeneous and thus more easily organ-
izable.
The objection to the line of causality
stressed here finds no support in fact. Extant
institutional and statistical evidence
suggests, if anything, that workers in plants
with greater dispersion of wages, not those
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION 13
_vr I 0 0 0 o
E ? ' t I o. ? > I ? ? o o? I I
o 0 4 St 1 a - 0
~~~~~~ 0 . nCfOC - 0S i t- 0 CI 0 C o o 0 9 0 I
CI - C I I I |
: ~ ~ ~ ~ ~ C 0 | | -3 ,|o o R o 0 0o - 0 0-
aWWW~I004,I._s,
O0 0 0 0 1
X a m oo 0 0 c o Z
sHt<>txs)s)t~~~~~~~~~~-cxXtr-O~CtsCtCtt;4 t t B t o~~ - > - 0 c(i 0 m - oo- oC
I e 9 0 0 0 =
0 C = _ . C ? ; _ 3 e t ~ U 4 Q : X 4 U sC ,
C) CIA C) -T C
10 00 nXn
Q~~~~~~~~~~~~~~~~~~TQ
~~ 0 0 0 0 9 0~~~~~~~~G 00 0 0 .
CZ
0) 0 Go 0) 0 04 00 C
~~O O 0 0 0~~~~~m 0 00 0
cc .
0 .
0 u~~~0 ~ 1
bJ)~ 0
000000 00 00 0 0 -000 00 00 00 00 00 0 00>
k~~~~~~~~~~~~~C ~>C
0 p z; CZ ct
cr~~i~~
.5 ~~~~~~~~~~~~~~ ~~~~~~~~ ~ ~ ~ ~ ~ ~ C
ct ctC
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
14 INDUSTRIAL AND LABOR RELATIONS REVIEW
with narrower dispersion, are more favorably
attuned to unionism. Foulkes reports
from his interviews with nonunion employers
that several eschewed rewarding workers
by merit pay for fear that such practices
would lead to unionization.' In the analysis
of NLRB elections by Farber and Saks, the
standard deviation of wages in the firm was
estimated to have a positive but insignificant
effect on the vote for unionism, but
their inclusion of a second term in which
the standard deviation appears suggests
that this underestimates the positive effect
of inequality on the vote for unions.'6 In
addition, Farber and Saks found that individuals
with earnings below their firm
mean were significantly more likely to vote
for unions than those with earnings above
their firm mean. This suggests, as they note,
that "workers at the lower end of the interfirm
earnings distribution ... expect a larger
increase in earnings from unionization,"''7
consistent with the causal link in which
unionism reduces dispersion.
Further, evidence from the 1977 Quality
of Employment Survey shows no discernible
difference between the dispersion of
earnings of nonunion blue-collar workers
who would vote for having a union to represent
them and the dispersion of those who
would vote against a union. Those for the
union had a mean log wage of 1.26 with a
standard deviation of .45; those against had
a mean log wage of 1.42 with a standard
15See Fred K. Foulkes, Personnel Policies in Large
Nonunion Companies, (Englewood Cliffs, N.J.:
Prentice-Hall, 1980), p. 340- 41.
16Specifically, Farber and Saks report calculations
with the inverse of the standard deviation of wages in a
firm and with the wage of an individual minus the
wage of the establishment divided by the same standard
deviation. The first term obtains a coefficient -.207
with an asymptotic standard error of .273, the second a
coefficient - .161 (.049). The derivative of votes with
respect to the inverse of the standard deviation of wages
is the sum of the two, - .368-which is much larger, of
course, than the coefficient on the inverse itself. Note
that we report results with respect to the standard
deviation, not its inverse, and thus we have reversed
the sign. See Henry S. Farber and Daniel H. Saks,
"Why Workers Want Unions: The Role of Relative
Wages and Job Characteristics," Journal of Political
Economy, Vol. 88, No. 2 (April 1980), pp. 349- 69.
"lIbid., p. 363.
deviation of .46.18 Finally, while I believe
that attempts to use systems equations to
discern lines of causality from cross-section
data are of little value,'9 the one effort to use
such procedures reported thus far shows
that the systems calculation yields results
confirming our conclusions.20
The reader who remains skeptical that
one can infer causality from cross-sectional
comparisons and institutional evidence
may ask for longitudinal data that would
indicate whether the reduction in dispersion
occurred before or after the advent of
unionism. Because of a present lack of longitudinal
data sets on establishments, this is
not possible at the establishment level. It is,
however, possible to compare the dispersion
of the log wages of individual workers
before and after unionism in widely used
data sets of individuals. While such an analysis
does not permit differentiation of the
effects of unionism on within-establishment
and among-establishment dispersion,
it does provide a test of the overall impact of
unionism on dispersion. If union wage
policies are the cause of the observed differences
in dispersion, the dispersion of wages
would be expected to fall among workers
moving from nonunion to union status and
to rise among workers moving from union
to nonunion status.
Analysis of the dispersions of log wages
of workers in two longitudinal data sets confirms
this expectation, as can be seen in
Table 6. This table shows the standard deviation
of the earnings of persons changing
'8These figures are based on responses to the Quality
of Employment Survey question: "If an election were
held with secret ballots would you vote for or against
having a union or employees' association represent
you?" In our tabulation of dispersion, 126 persons
answered yes and 182 answered no. See Robert Quinn
and Graham Steiner, Quality of Employment Survey
(Ann Arbor, Mich: Inter-University Consortium for
Political and Social Research, 1977).
'9Richard B. Freeman and James L. Medoff, "The
Impact of Collective Bargaining: Illusion or Reality?"
in Jack Steiber, Robert B. McKersie, and D. Quinn
Mills, eds., U.S. Industrial Relations, 1950-1980: A
Critical Assessment (Madison, Wis.: Industrial Relations
Research Association, 1981), pp. 47- 97.
20Barry Hirsch, "The Interindustry Structure of
Unionism, Earnings, and Earnings Dispersion,"
Industrial and Labor Relations Review, Vol. 35, No. 1
(October 1982), pp. 22- 39.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION 15
Table 6. Longitudinal Analysis of Dispersion in in Wages of
Workers Who Change Union Status.
Standard Deviation
of In Wages Long tudlnal
Survey a Sample Before After of Change
Size Change Change on Dispersion
National Longitudinal Survey
of Young Men (NLS)
Workers Who Changed Status,
1969- 1976
Nonunion to Union 237 .42 .33 -.09
Union to Nonunion 231 .39 .42 .03
Michigan Panel Survey of
Income Dynamics (PSID)
Workers Who Changed Status,
1968- 1977
Nonunion to Union 47 .38 .30 - .08
Union to Nonunion 59 .32 .46 .14
a Samples include all workers in th
Richard B. Freeman, "Longitudinal Analysis of Trade Union Economic Effects" (in process).
union status in two widely used longitudinal
data sets on individuals: the Michigan
Panel Survey of Income Dynamics and
the National Longitudinal Survey of Young
Men. Both data sets show a substantial reduction
in dispersion for workers moving
from nonunion to union status and an increase
in dispersion for workers going in the
other direction. A more detailed analysis of
these data requires a separate study, but it
suffices here to note the consistency of the
results in Table 6 with our interpretation
of lines of causality.
In sum, whereas there is evidence supporting
the argument that unions choose
wage policies that reduce dispersion, there
is neither institutional nor econometric
support for the reverse causality.
Further Empirical Tests
Unionism and among-establishment
dispersion. In addition to reducing the
dispersion of wages within establishments,
unionism can be expected to affect the dispersion
of wages among establishments.
Under the banner of "equal pay for equal
work," unions have long pressed for standardization
of rates across establishments
in the same sector. This goal conflicts, however,
with another primary union goal, the
desire to achieve monopoly wage gains. The
achievement of such gains can be expected
to increase dispersion to the extent that
different establishments have different
elasticities of labor demand.
We find in our data set, controlling for
the regional location of establishments
and their distribution of employees by occupation,
that dispersion of wages among
unionized establishments is smaller than
among nonunionized establishments in
six of nine industries. The empirical results
are shown in Table 7. Columns 1 and 2
display the cross-establishment dispersion
of wages in the sample as a whole. Column 1
records the mean square error in the sample,
while column 2 records the mean square
error calculated from a regression of the
establishment wage on occupation and
region independent variables (weighted
by establishment size). Columns 3 and 4
record the mean square error in union and
nonunion sectors obtained from separate
regressions for the two groups, while column
5 gives the difference in mean square
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
16 INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 7. Comparison of Variance of in Wages Between Firms.
Total Sample
Residual Sum
of Squares
from Regression
Mean with Occupation Union a Nonunion a Difference
Sum of Squares & Region Residual Residual in Residual
Industry Between Firms Dummiesa Sum of Squares Sum of Squares Sum of Squares
Paints 8c
Varnishes .0343 .0085 .0053 .0148 - .0095
Textile Dyeing
8c Finishing .0238 .0023 .0027 .0017 .0010
Cotton, Man-made
Fiber Textiles .0044 .0002 .0002 .0002 -.0000
Wool Textiles .0153 .0013 .0017 .0010 .0007
Industrial
Chemicals .0209 .0004 .0004 .0005 - .0001
Wood Household
Furniture .0387 .0037 .0046 .0029 .0017
Miscellaneous
Plastic Products .0352 .0026 .0023 .0028 -.0005
Fabricated
Structural Steel .0386 .0039 .0034 .0044 - .0010
Nonferrous
Foundries .0431 .0069 .0050 .0079 - .0029
a These numbers were taken from a weighted regression controlling for region and occupation (weighted by
number of workers per firm).
errors. In five of nine industries, we find
lower dispersion in the organized sector and
in one, essentially no difference.
Unionism and within-industry dispersion.
The analysis thus far has studied the
effect of unionism on wage dispersion by
analyzing differences in the dispersion of
wages between organized and unorganized
production workers. The impact of unionism
on the dispersion of wages as a whole
depends not only on the impact on organized
labor, however, but also on the union
wage effect. By raising the wages of organized
production workers compared to those
of otherwise comparable unorganized production
workers, unionism increases dispersion.
By raising the wages of production
workers relative to higher-paid nonproduction
workers, unionism reduces dispersion
of wages within an industry.
How do these conflicting effects balance
out? To answer this question, we decompose
the variance of the In of wages of all workers
in a sector as follows:2'
(1) 2 = aBU( r2BU) + aBN( a2BN)
+ a W( a2 W) + a BU aBN(ln u/n)2
+ a BU a W(ln u/w)2 + a W a BN (In w/n)2,
2lThis formula is based on the usual conditional
variance formula. It differs from that in Freeman,
"Unionism and the Dispersion of Wages," by explicitly
relating the variances and wage differences of
blue-collar union and blue-collar nonunion workers
and of white-collar workers to the total variance in
one formula, whereas the previous decomposition
looked separately at blue-collar and white-collar
workers in the union part of the industry and at bluecollar
and white-collar workers in the nonunion part
of the industry. The formulas can, of course, be derived
from one to the other.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION 17
where
a2 = variance of in wages in the in-
dustry;
a BU = share of all workers classified as
union and blue-collar;
OLBN = share of all workers classified as
nonunion and blue-collar;
a w = share of all workers classified as
white-collar;
a 2BU = variance of In wages of bluecollar
union workers;
a 2BN = variance of In wages of bluecollar
nonunion workers;
a2w = variance of in wages of whitecollar
workers;
u/n = ratio of union to nonunion
blue-collar wages;
u/w = ratio of union blue-collar to
white-collar wages; and
w/n = ratio of white-collar to nonunion
blue-collar wages.
To determine the effect of unionism on
the variance of wages in the industry, we
difference Equation 1 with respect to unionism.
This yields:
rA
(2) LA2 = aBULA [L2BU]
+ aBu aBN' I4n u/n]2
+ aBUaw A [Inu/w] 2,
where the effect of unions on the dispersion
of wages of white-collar workers, A a 2 W.
and on the differential between nonunion
blue-collar and white-collar workersA
(In n/w )-are assumed zero.
The first term on the right-hand side of
Equation 2 is just the sum of the union effect
on the dispersion of wages within organized
establishments and between organized
establishments. The second term depends
on the differential between union
blue-collar and nonunion blue-collar workers,
while the third term depends on the
differential between union blue-collar
workers and nonunion white-collar
workers.
The information needed to calculate
Equation 2 is presented in Table 8. Column
1 records the proportions needed for the
analysis: the fraction of workers who are
unionized blue-collar, nonunionized bluecollar,
and white-collar in each industry.
Column 2 gives the estimated effect of
unionism on the variance of In wages in an
establishment, obtained by regressing the
within-establishment variance on unionism
and the relevant control variables, as in
Table 3, but weighting the regressions by
number of workers: the exact calculations
are given in Appendix Table A. Column 3
gives estimates of the effect of unions on the
variance of wages across establishments by
taking differences in mean squared errors, as
in Table 7. Because we are adding together
variances to get a total for the work force as a
whole, both of these figures are based on
calculations in which firms are weighted
by their number of employees. Hence, the
figures differ slightly from those in Tables
3 and 7.
The contribution of the within and between
effects to the overall industry variance
in wages is given in column 4. It is simply
the sum of columns 2 and 3 multiplied by
the unionized blue-collar share of labor
(aBU) reported in column 1. In six of the nine
industries, the contribution is negative, indicating-as
noted earlier-that through
the within- and among-firm effects, unionism
lowers dispersion.
Column 5 presents an estimate of the
union wage effect in each sector. This estimate
is obtained from regressions of the
mean In wage in each establishment on'its
union status and the full set of controls
used in Table 3. These results are shown in
Appendix Table B. The dependent variable
is weighted by the number of workers in
each firm. Consistent with previous work,
union wage effects are positive in the majority
of cases. The unweighted average of effects,
however, is just .06, which is somewhat
smaller than the result obtained in
most studies. In one industry, unionism
is estimated to have a modest negative effect
on wages.
Assuming no differential between union
and nonunion establishments in the absence
of unionism, the effect of the union
wage differential on the variance according
to Equation 2 is just aBUaBN multiplied
by the differential squared. This is
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
18 INDUSTRIAL AND LABOR RELATIONS REVIEW
br 3 33
V 01
t~~~~ I ZiZL C v
Cf C'f 0 C VC C!o/o 91 z7 C4 - c8 o r
tt t ya-X; s~~~~~~~0 o4 ol cs t4 i
00 C~~~~~'f 0 ~~~0 0 0l - 0 r0C
O0 0 0 0 00 0e 0
?- I' I' c- I' I' N t Cl |
A I t 1 Q U U D / c | c) zc zc c 0
cli ~ ~ ~ ~ ~ ~ c
0~ 0 0 0 C~l C 00 c
Cf 0 Cf ~~~ r0 i- C0 r
C~ CD CD 0D CD 04 C~
410II I . CD ; | ? V
ak, %l |_____._O___|__________N ___________|____________a_
/> ? ,-i * i 00 r- 5 C- 0 C5l cW C
~~~ 0 0 0~~~~~Ll 00 0- 0 0 0
- 41~~~~~~~~~~0 Cl r ~l ClcC, ~ g C0,17 C 00
4,~~~~~~~~~~~~~~~~ 0 0
024 0 0 0 0C 0 0 bCn
o' o o0 0 0 00 00 0- -~ C14
C,[~ ~~~0 0 - Cf 1- fC
~~~ 00 Cf t'- Cf Cf ~~~~~~~~~~~~~~~~~~~~ ~ ~ C
4(~~~~~~~~ Q
V 0
0~~~~~~~~~~~~~~~~~~0
* 4, 4 c~~~~~~~~~r an 0f 0 O
024, ~ ~ 00 ~a x7- 00 00 7- C
Q) ~~~~~ r~~~ - - ~~~~~~ 0 C~~~~~ 0
0 0 0 0 0 0 0~C 0 -
4,9, 0 0 0 0 0 0
9~~~ 0 ~~~ 0 0 0 0 0 0 ~~~~~~~O 0
,k) :Z0 0 0 0 bi)Co
o 0~ 0n 00 0 , M C
* 0~~~~~~~~~~~~~
CM ~~~~~~~~~~~
C,[ C~~~l ho
C,[~~~~~~~~
O C
0 C
-~~~~~ - ~ ~
ct m
- -~~~ ~~~ ~ C v
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION 19
recorded in column 6. Note that, with the
exception of three industries, the effect is
negligible to the third decimal place, and
thus dwarfed by the effects in column 4.
Because the Industry Wage Survey tapes
lack information on the pay of white-collar
workers, we have been forced to rely on another
data set to obtain estimates of the effect
of unionism on the blue-collar/whitecollar
pay differential. We have estimated
the blue-collar/white-collar differential in
the absence of unionism using the relevant
three-digit industry nonunion observations
from the Expenditures for Employee Compensation
Survey (EEC) and then used our
estimates of union wage effects to calculate
the potential impact of unionism on the
differential. If, consistent with the results
reported by Freeman and Medoff, there is
relatively little spillover of wages from
union to nonunion blue-collar workers,
and if union wages do not affect whitecollar
wages in a firm, this is the appropriate
impact.23 If the presence of unionism raises
wages of nonunion blue-collar labor, it is an
underestimate; if the presence of unionism
raises the wages of nonunion white-collar
labor in organized firms, it is an overestimate.
Following Equation 2, the union
effect is calculated as the difference between
the squared In blue-collar/white-collar
differential from the EEC tape and that
differential adjusted for the union wage
effect on blue-collar workers.
[n blue-collar W ~i2
white-collarW '
[In ( blue-collar ~W+ union wage effect)] 2white-collar W
As can be seen in the final column in
Table 8, consistent with the somewhat different
calculations for the entire economy by
Freeman,24 the dispersion-reducing effects
22The regression estimates are available in an earlier
version of this paper: see Richard B. Freeman, Working
Paper No. 752 (Cambridge, Mass.: National
Bureau of Economic Research, September 1981).
23Richard B. Freeman and James L. Medoff, "The
Impact of the Percent Organized on Union and Nonunion
Wages," Review of Economics and Statistics,
Vol. 63, No. 4 (May 1982), pp. 561-72.
24In Freeman, "Unionism and the Dispersion of
of unionism dominate the dispersion-increasing
effects in all of the industries cov-
ered.
Conclusion
This study has used establishment-level
data to examine the effect of unionism on
the wage structure within establishments,
focusing on the impact of unionism on
wages in an establishment rather than
merely on wages of an individual. It reports
five principal findings:
(1) Organized establishments have much
lower dispersion of wages than otherwise
comparable unorganized establishments in
the same four-digit industry. Using the
standard deviation of the In of wages as a
measure of dispersion, unionized establishments
have levels of dispersion that
range from 5 percent to 50 percent below
those of nonunionized establishments in the
industries studied, with an unweighted
differential of 22 percent.25 Diverse variables
controlling for size of establishment, region,
and detailed occupation of workers do not
greatly affect the magnitude of these results.
(2) Organized establishments have
adopted explicit wage practices that tend to
reduce wage inequality. They favor singlerate
or automatic progression modes of
wage payment as opposed to merit reviews
and individual determination. A sizable
part of the union-induced reduction in
within-establishment dispersion is attributable
to the explicit wage practices in such
plants. By favoring wage practices that
Wages," I did not have proper data to decompose the
union effect on the variance of the wages of blue-collar
workers into within-establishment and among-establishment
dispersion-reducing effects. In addition,
that study was limited to male workers, whereas this
one looks at all workers.
25The unweighted differential of the standard
deviation of In wages is a simple mean over all nine
industries of the percentage reduction from the numerous
variances in the union sector. The unweighted differential
of the variances of In wages over all industries,
computed as the percentage reduction in variance
in the union sector from the nonunion sector,
is .33 percent. The weighted differential of the variance
of In wages is the mean differential for each
industry weighted by the number of firms in the industry.
It is computed to be .27 percent.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
20 INDUSTRIAL AND LABOR RELATIONS REVIEW
narrow ranges of rates among workers and
limit managerial discretion, unions reduce
inequality within firms. Even taking account
of wage practices, however, unionized
firms tend to have lower inequality among
workers, indicative of an influence on the
operation of specific wage-setting mechanisms
as well as on the choice of the practices
themselves.
(3) Dispersion of wages among organized
plants in the same four-digit industry also
tends to be lower than dispersion of wages
among nonorganized establishments,
taking account of differences in the regional
and occupational distribution of the two
sets of plants. This result is found in five of
nine industries.
(4) The total effect of unions on the dispersion
of wages among blue-collar workers
depends upon three components: the effect
of unionism on dispersion within organized
firms, the effect of unionism on dispersion
across organized firms, and the union wage
effect. In six of nine industries studied the
net effect of unions on dispersion, taking
account of all three effects, is negative, indicating
that in the majority of cases studied
unionism lowers wage inequality for bluecollar
workers.
(5) Because the union wage gains bring
blue-collar wages closer to white-collar
wages in the unionized sector, unionism
tends to have a greater negative impact on
the dispersion of wages among all workers
in an industry than on the dispersion of
blue-collar workers alone. Assuming that
unionism does not influence the wage dispersion
among white-collar workers, we
find that in all nine industries, unionism
reduces dispersion of wages among all
workers by sizable amounts.
Put broadly, these results indicate that
explicit union wage policies have a major
impact on the dispersion of wages within
firms and in the economy as a whole.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms
UNION WAGE PRACTICES AND WAGE DISPERSION 21
Appendix
Table A. Differences between Union and Nonunion Establishments in
Standard Deviations of Within-Establishment In Wages.
(standard errors are in parentheses)
Mean of Within Establishment . ..
Industry Differences in Means Union Coefficient
Union Nonunion
Paints & .0887 .1276 -.0389* -.0251
Varnishes (.0075)
Textile Dyeing .1122 .1275 -.0153 -.0098
& Finishing (.0118)
Cotton, Manmade .1374 .1385 - .0011 - .0106
Fiber Textiles (.0044)
Wool .1031 .1301 - .0270* - .0220
Textiles (.0101)
Industrial .0727 .1087 - .0360* - .0360
Chemicals (.0057)
Wood Household .1098 .1251 - .0153** -.0178
Furniture (.0074)
Miscellaneous .1402 .1776 - .0374* - .0340
Plastic Products (.0049)
Fabricated .0851 .1407 - .0556* - .0561
Structural Steel (.0062)
Nonferrous .1270 .1686 - .0416* - .0492
Foundries (.0064)
*Significant at be
**Significant at b
Source: The union and nonunion means for the standard deviations of wages are calculated from the means of the
standard deviations of the 1 n wages for each firm. The t-test is the standard test of the difference between two mean
assuming unequal variances of the union and nonunion distribution. Also included in the regressions are size and
region and, for the overall firm equation, occupation controls. In the detailed occupations, firms with only one
worker listed under one occupation are deleted.
Table B. Union Wage Coefficients from Regressions Weighted by the Number of Workers per Firm
with the Dependent Variable the Firm's Average In Wage.a
Union Coefficient Union Coefficient
Industry (Standard Error) Industry (Standard Error)
Paints & .0566 Wood Household .0460
Varnishes (.0220) Furniture (.0216)
Textile Dyeing .0220 Miscellaneous .0511
& Finishing (.0314) Plastic Products (.0095)
Cotton, Manmade .0157 Fabricated .1665
Fiber Textiles (.0088) Structural Steel (.0221)
Wool .0510 Nonferrous .1102
Textiles (.0393) Foundries (.0205)
Industrial - .0045
Chemicals (.0190)
aAlso included as independent variables are occupation, region, and size controls.
This content downloaded from 147.251.55.24 on Thu, 29 Mar 2018 11:26:29 UTC
All use subject to http://about.jstor.org/terms