Content uploaded by Sanford M Jacoby
Author content
All content in this area was uploaded by Sanford M Jacoby on May 04, 2019
Content may be subject to copyright.
Measurement
of
Compensation:
Union and Nonunion
Sanford M. Jacoby and Daniel J.B. Mitchell*
For
many purposes, the economic impact of unions
is
better measured by
the proportion of union wages in total payrolls rather than by the proportion
of unionized employees in the overall workforce.
We
use recently available
Current Population Survey data to generate estimates of the former.
We
also show that published data
from
the Survey on median union and
nonunion wages produce substantially larger estimates of the union1
nonunion wage differential than figures based on mean wages. Finally,
we note that the Bureau of Labor Statistics’ Employment Cost Index
gives undue weight to the union sector because of
its
Laspeyres methodology.
THE
PROPORTION
OF
UNIONREPRESENTED
EMPLOYEES
in the
workforce is often used as
a
key index of union strength and influence. For
many years, estimates of the unionization rate were based on membership
estimates obtained from unions by the
U.S.
Bureau of Labor Statistics
(BLS). More recently, the BLS has shifted to representation estimates from
the Current Population Survey (CPS) as its primary data source
on
unionization. The
CPS
has also been the source of widely used estimates of
union versus nonunion wage levels based on annual estimates of union and
nonunion median “usual weekly earnings” for fulltime employees.
For some purposes, the proportion
of
employees who are unionrepresented
is an appropriate measure of union influence. If, for example, union political
power is
at
issue, a count of union workers is an important piece
of
information. But in other cases, the proportion of total payrolls (or total
compensation) covered by collective bargaining is the more relevant statistic.
In particular, when investigating the influence
of
union wage settlements in
*
The authors’ affiliations are, respectively, Anderson Graduate School of Management, University of
California at
Los
Angeles and Institute of Industrial Relations and Anderson Graduate School of
Management, University of California at
Los
Angeles.
INDUSTRIAL RELATIONS,
Vol.
27,
No.
2 (Spring 1988).
0
1988 Regents of the University
of
California
0019/8676/88/527/2
15/$10.00
215
216
/
SANFORD
M.
JACOBY
AND
DANIEL
J.
B.
MITCHELL
an aggregate wage index, the correct and appropriate union weights are
based on payrolls or compensation, not on employment. Despite the
importance of payrollbased weights for assessing the economic impact of
unions, we know of no previously published calculations of such weights.
In this paper, we use data now available from the
CPS
to calculate these
estimated payroll weights for commonly used indexes, such as average hourly
earnings and compensation per hour. Union importance in these indexes
is shown to be understated by traditional employment weights. The
understatement is caused by two factors: union workers earn more than
nonunion workers and they work longer hours. We also show that for some
purposes, the published
CPS
data on union/nonunion pay differentials are
misleading, owing to the
BLS’
use of median rather than mean pay measures.
The
BLS’
muchquoted estimate of a roughly onethird wage gap between
union and nonunion pay in the mideighties is substantially overstated
because the use
of
medians reduces the effect on the ratio of highly paid
nonunion employees.
Finally, we note that the unionsector weights in the Employment Cost
Index, an index intended by the
BLS
eventually to become the key indicator
of wage change, are too large.
Union
Impact
on
Aggregate
Wages
When aggregate wage changes occur, some of the change stems from the
unionized sector of the economy and some from the nonunion. Determining
the source of wage change can be important in an overall assessment of the
economy, particularly with regard to inflation. In periods when unions are
perceived as strong, it often is assumed that wage inflation stems from
bargaining demands. Thus, the federal antiinflation wage controls mounted
in the early seventies frequently were rationalized by, and focused on,
seemingly excessive union settlements. Similarly, during periods when unions
are perceived as weakas in the eightieswage moderation may be attributed
to union concession bargaining.
A
full assessment of the union impact on aggregate wages must take
account
of
the possibility of uniontononunion wage influences, such as
spillovers, patterns, and threat effects. However, before such assessments
can be attempted,
a
simple question must be asked: “If union wages were
to rise by
1
per cent
without
affecting nonunion wages, how much would
!The authors would like to thank Maury Pearl, who acted as research assistant on this project, and
A1
Schwenk and others at the
U.S.
Bureau
of
Labor Statistics who assisted in providing necessary data
and technical information. They also wish to thank two anonymous referees.
Measurement
of
Compensation
I
217
aggregate wages rise?” What is involved here is the basic arithmetic of the
direct union effect on the wage index, i.e., the weight of the union sector
in the index. Indirect effects, such as spillovers, must be added to the direct
effect before the full impact of union settlements on aggregate wage change
will emerge. But the indirect effects are secondary, and common sense
suggests that the direct impacts will dominate, especially in the short run.
Thus, it is the direct effects we seek.
Aggregate wage indexes, such as the monthly average hourly earnings
series and the quarterly compensation per hour index, are computed by
dividing the total payroll (or total compensation bill) by the number of
employee hours. If union wages were to rise by
1
per cent, the direct effect
on these indexes would be determined by the union weight in payrolls
or
compensation, not by the union weight in employment. The payroll or
compensation weights will be identical to the employment weights only if,
on average, union and nonunion workers receive the same level of hourly
pay and work the same number of hours per period. In general, neither of
these assumptions is valid. Union workers tend to earn higher pay than
nonunion workers and they tend to work longer hours.
The empirical generalizations of higher pay and longer hours for
union workers do not involve any complex standardizations for worker
characteristics. It is the gross (unstandardized) observations that are relevant.
Even if unions have no causal impact on pay and hours, the fact that
unionized workers are paid more and work longer hours than do nonunion
workers gives the union sector added weight in aggregate wage indexes.
Data
Sources
Beginning in the eighties, the Current Population Survey has been used
regularly to obtain data on earnings and unionization. This information is
now available
to
researchers in published form and on computer tapes. In
this study, we take advantage of these CPS data to estimate the appropriate
unionsector weights in the indexes of compensation per hour and average
hourly earnings.
Sample.
Our calculations combine data from the CPS earnings files for
1983 and 1984. These files were compiled by the BLS for outgoing rotation
groups within the CPS sample. They are the same sources the BLS uses to
produce its annual tables on union membership, representation, and union
versus nonunion earnings (BLS, 1982a; Flaim, 1986; Adams, 1986).
Data reported below have been restricted to private, nonagricultural
employees; all selfemployed persons have been removed. In order to enhance
218
/
SANFORD
M.
JACOBY AND DANIEL
J.
B.
MITCHELL
comparability to the establishmentbased data reported in the average hourly
earnings
(AHE)
series, however,
our
estimates include 14 and
1
5yearolds;
parttime workers are also included except where indicated. Neither 1415
year olds nor parttimers appear in the published BLS tables on union/
nonunion earnings differences. Thus, the definitions adopted in this paper
are better suited for calculating union payroll or compensation weights than
those underlying the published tables. Finally, for purposes of the
computations described below, unionized employees are defined as those
workers
represented
by unions; nonmembers who are in collective bargaining
units are counted as unionized employees.
In the discussion that follows, we refer to two samples drawn from the
CPS tapes: the “global” sample (meant to simulate compensation per hour)
and the
“AHE’
sample (meant to simulate average hourly earnings). The
global sample includes all wage and salary earners remaining after application
of
the restrictions and definitions noted above. The
AHE
sample was derived
by taking the global sample and excluding from it those occupations that
are not included in the payroll data used to generate the
AHE
series, namely,
managers, supervisors, and certain other classifications.2 For
19831984,
there were
283,989
observations in the global sample, and
206,316
in the
AHE sample.
Calculating
the
Union/Nonunion Pay Ratio: Means
vs.
Medians
When the BLS codes the
CPS,
it
censors some data values. In particular,
wage and salary workers reporting weekly earnings of
$1,000
or more are
assigned
a
value of earnings of
$999.
Because the BLS only reports median
figures for usual weekly earnings, its published tables are unaffected by this
truncation of the upper tail of the earnings distribution. However, since it
is essential to use means rather than medians to calculate payroll weightsthe
objective of this studyestimates of the upper tails had to be reconstructed.
The upper tails
of
the earnings distribution were assumed to follow the
form
of
a
Pareto di~tribution.~ Methodology following BLS practice was
used for the estimations of the upper tails.4 Pareto factors were calculated
The AHE series includes only production workers in mining and manufacturing, construction workers
in the construction industry, and nonsupervisory workers in all other industries
(BLS,
1986).
*,
where
A
is
the Pareto constant and d is the lower limit of the interval to which the Pareto distribution is
assumed valid. When x is not
less
than d, the conditional mean of the openended intervalthe Pareto
factoris then (MAl)(x*), where x*
is
the lower
limit
of the openended interval.
See
Theil (1967).
“he
BLS
has developed a maximum likelihood estimator for
A
(defined
in
footnote
3)
which was
applied in this study. See West (n.d.).
’In a Pareto distribution, the probability of
finding
an
individual whose income exceeds
x
is
(dd);
Measurement
of
Compensation
I
219
for each industry/unionstatus distribution presented below, and these factors
were used to compute the needed earnings means and payroll weights.
The upper panel of Table 1 compares the median weekly earnings figures
published
by
the BLS with the median and mean earnings figures computed
directly from the merged 19831984 files.5 There is a noteworthy discrepancy
between mean and median earnings, especially for nonunion workers. A
number
of
highearning employees fall into the nonunion category. Not
surprisingly, a lesser proportion of unionized workers are very high earners.
The extreme nonunion values have little impact on medians but carry larger
weight in means.
Labor market analysts who rely on the published BLS tables for estimates
of union/nonunion wage differentials should do
so
with caution. The median
differential is substantially larger than the estimated mean differential (38
per cent vs. 12 per cent). Thus, users cannot assume that calculated union/
nonunion differentials are insensitive to the measure of central tendency
employed. To the contrary, Table 1 demonstrates that medians and means
are not interchangeable.
Analysts who use the BLS tables probably want to know what the average
union worker earns compared with the average nonunion worker. The
BLS
medianbased differential
is
clearly a substantial overstatement
of
the answer
to that question. Generally, medians will be relatively insensitive to changes
in the distribution profile.
It
is possiblethough we cannot verify itthat
this insensitivity is responsible for the failure of the BLS published
differentials to reflect the union wage concessions of the eighties.6
Table
1
shows that the unionlnonunion wage differential
is
quite sensitive
to the sample’s occupational specifications. When occupational definitions
are changed to match those used in the AHE serieschiefly by eliminating
managers and supervisorsthe median earnings differential is raised to
56
per cent, and the mean is boosted to 29 per cent.
It
could be argued that
the AHE figures in Table 1 are derived from a sample that approximates
the potential union membership base and thus constitutes a more realistic
measure of uniodnonunion earnings differences. However, even for this
restricted sample, a significant meadmedian gap remains for nonunion
The
BLS
divides the earnings distribution into ranges and applies an interpolation technique in the
range in which the median is located to produce its median estimate. In contrast, the
19831984
estimates
of
the median presented
in
Table
1
are calculated directly from the twoyear distribution vithout
application
of
an interpolation technique. There appears to be little practical difference between direct
estimation and estimation by interpolation.
61f
wage concessions were concentrated among abovemedian union workers, they would not pull down
the union median. Freeman
(1986)
reports an absence
of
an apparent concession impact using
CPS
data.
His analysis was based on regression coefficients estimated across individual workers and not on medians,
however, which may suggest other difficulties with the
CPS.
Freeman’s paper does not report how he
dealt with the truncated ends of the union and nonunion earnings distributions.
220
/
SANFORD
M.
JACOBY AND DANIEL
J.
B. MITCHELL
TABLE 1
MEAN
AND
MEDIAN WEEKLY EARNINGS
FOR
PRIVATE, NONAGRICULTURAL WAGE
&
SALARY
EARNERS, 19831984
Fulltime employees
BLSpublished Medians and means calculated from
medians 19831984
CPS
earnings files
Global sample AHE sample
1983 1984 Median Mean Median Mean
Union $391 $404 $400 $410 $390 $399
Nonunion 287 302 290 367 250 310
Ratio:
unionl
nonunion 1.36 1.34 1.38 1.12 1.56 1.29
Fulltime and parttime employees
combined, 1983 1984
Global sample AHE sample
Median Mean Median Mean
Union
$380 $391 $365 $378
Nonunion 236 304 200 252
Ratio:
unionl
nonunion 1.61 1.28 1.83 1.50
Source:
BLSpublished data
from
Employment
and
Earnings,
XXXII
(January,
1985),
p.
211;
other
data
from
19831984
earnings
files
from
the
Current
Population Survey (see description in text).
workers, a discrepancy which again reflects the weight of the upper tail
of
the nonunion earnings distribution.
The lower panel of Table
1
includes parttime workers in both the global
and
AHE
samples. Thus, the lower panel is more comprehensive than the
tables on union vs. nonunion earnings published by the
BLS.
Parttimers
are paid less than fulltime workers, often on an hourly, and certainly on a
weekly basis. They are also heavily nonunion.
As
a
result of these differences, absolute earnings are reduced for both
the union and nonunion groups (relative to the upper panel), but the
reduction
is
comparatively greater for nonunion workers. This comparative
effect raises the weekly unionhonunion earnings differential. For example,
Measurement of Compensation
i
221
the global median differential increases from
38
per cent for fulltime workers
to
61
per cent for full and parttime workers combined. The same type of
meadmedian discrepancies that characterized fulltime workers are again
observed when parttimers are included in the analysis.
Payroll
Weights
Table
2
presents employment and payroll weights for full and parttime
workers. Aggregate weights are shown together with weights for industrial,
occupational, and demographic groups. The reported payroll weights were
calculated by multiplying the estimated mean weekly earnings figures by
their corresponding employment levels and then dividing the union payroll
by the total (union plus nonunion) payroll for each classification shown.
Employment weights are simply the proportion of individuals in a specific
category who were represented by unions.
The top line of Table
2
shows that unions represented
18
per cent of
private, nonagricultural wage and salary workers in
19831984,
but that
these unionized workers earned about
22
per cent of total payrolls in those
years. The discrepancy between employment weights and payroll weights
reflects the earnings differential between unionized and nonunion workers.
As
could be anticipated from the exclusion of higherearning occupations,
the gap between employment and payroll weights is larger for the
AHE
sample
(28
per cent vs.
20
per cent) than for the global sample
(22
per cent
vs.
18
per cent).
Disaggregated weights and reverse pay differentials.
When finer breakdowns
are examined, it can be seen that the employment weight is usually less
than the payroll weight. Reverse cases occur only when mean earnings for
nonunion workers in a grouping exceed mean union earnings. Since the
earnings data are
gross,
i.e., not standardized for employee characteristics,
such reverse differentials can occur even if unions raise the wages of their
members. Cases in which the union payroll weight is
smaZZer
than the
employment weight can be found among workers in the finance, insurance,
and real estate industry, the mining industry, and among managers and
profe~sionals.~
Gender differences.
Table
2
indicates that the gap between global
employment and payroll weights for males is relatively small when compared
'For
example, in the case
of
managers and professionals, unionized workers might be nurses and
engineers, while nonunion workers might be doctors, lawyers, and executives.
222
/
SANFORD M. JACOBY AND DANIEL
J.
B.
MITCHELL
TABLE
2
EMPLOYMENT
AND
PAYROLL
WEIGHTS
FOR
19831984,
PRIVATE,
NONAGRICULTURAL
WAGE
AND
SALARY
EARNERS
Employment weight Payroll weight
All
industries
Full
time only
Indusrrial sector
Construction
Finance, insurance,
real estate
Manufacturing
Mining
Services
Transportation
&
public utilities
Wholesale
&
retail
Occupational
group
Managers
&
professionals
Operators, fabricators,
&
laborers
Precision production,
craft
&
repair
Service workers
Technical, sales,
administrative
Age
Under
40
years
40
and over
Black
White
Other
Female
Male
Race
Sex
Full
time only
Full time only
18
21 20
24 22
22 28
29
27
4 30
4 37
3 43
4
29
21
9
44
42
26
10
49
30
18
11
47
50
24
12
56
9
10 12 14
8
36 11
35 7
47 12
46
33 35 40 44
10
10
11
10 17
12 18
12
15
23 17
27 19
23 24
34
26
18
17
28
20
21
34
21
19
37
27
26
11
13
23
25
13
16
27
30
14
15
24
24
17
19
33
34
~~~
Source:
See
Table
1.
with the corresponding gap for females. In fact, for fulltime workers only,
the male gap
is
negative in the global sample.
This
finding for males is a
case of the “tail wagging the distribution”; wellpaid, unorganized, male
managers and supervisors dominate the global male payroll weights. When
these workers are removed, as in the
AHE
sample, a positive gap appears.
Measurement
of
Compensation
/
223
The female earnings distribution is less skewed; there are proportionately
fewer highly paid nonunion women than men. Nevertheless, the global
unionhonunion wage differential does not exist only because of the union/
nonunion wage gap among female workers.
At
the aggregate level, unionized
men count heavily in total payroll weights; there are more of them, and they
earn more (on average) than unionized women. Thus, the disproportionate
presence of males contributes to the aggregate unionhonunion wage gap
which reflects male/female wage differentials as well as union/nonunion
differences within the two sex groups.
Age
characteristics.
There is a parallel situation with regard to the two age
groups presented in Table
2.
Taken separately, the gap between the
employment weight and the payroll weight for the global sample is evident
only for younger workers (those less than
40
years old). But, on average,
younger workers earn substantially less than older workers and are less likely
to be unionized. The gap between employment weights and payroll weights
for the overall global sample reflects both unionhonunion age differentials
across age groups and agerelated wage differences.
Racial characteristics.
Perhaps the most striking feature of Table
2
is the
high proportion
of
blacks’ pay which comes from the union sector: Over a
third of the black payroll in
19831984,
even for the global sample, was set
by collective bargaining. This confirms the widespread conclusion in the
unionimpact literature that union wagesetting is very important for black
workers (see Freeman and Medoff,
1984).
Impact
of
parttime workers.
The second row of Table
2
shows that adding
parttime workers to the sample has little effect on union payroll weights.
This result may seem surprising since Table
1
demonstrated that adding
parttime workers raises the union/nonunion earnings differential substan
tially. Why isn’t that boost in earnings differentials reflected in the payroll
weights?
This seeming paradox is easily explained. When payroll weights are
calculated, the addition of parttime workers greatly increases the total
number of nonunion individuals in the denominator of the weight. The
“body count” effect of parttimers on nonunion payrolls offsets their wage
depressing effect. In contrast, within the union sector, parttimers add few
bodies but have
a
sufficient wagedepressing effect to drag down the union
payroll relative to the nonunion.
224
/
SANFORD M. JACOBY AND DANIEL J.
B.
MITCHELL
Union Impact on Fringe Benefits
Previous studies have found that, in addition to raising wages, unions
increase expenditures on most kinds of fringe benefits. The union “fringe
effect” often is larger than the union wage effect (Freeman and Medoff,
1984).
Unfortunately, fringe benefits such as pension plans, health and life
insurance, and similar programs are not included in the
CPS
data on usual
weekly earnings. Employerpaid payroll taxes, e.g., for Social Security, are
also omitted. Yet, a complete account of the union weight in total
Compensation must include fringe benefits and payroll taxes.
Limited sensitivity
of
the results.
Although an adjustment for fringe benefits
is needed, the actual results obtained are not likely to be highly sensitive to
the precise technique used. National income account data suggest that the
average ratio (union plus nonunion) of total compensation to wages in the
private sector was about
1.2
in
19831984.
Of course, if both union and
nonunion workers had the same
1.2
ratio, the union payroll weightadjusted
to
a
total compensation basiscould be unaffected; the weight would remain
at the
22
per cent global estimate of Table
2.
It is simple arithmetic,
however, to
try
alternative assumptions about the union compensationto
wage ratio, given the
1.2
total ratio and the
22
per cent union payroll weight.
The payroll weight adjusted for unionhonunion compensation differences
will increase by only about
1.8
percentage points for each added
10
percentage
point increase in assumed union compensation impact. That
is,
if the union
compensationtowage ratio were
1.3
rather than
1.2,
the
22
per cent union
payroll weight would rise to
23.8
per cent on a compensation basis.
If
the
ratio were
1.4
rather than
1.3,
the compensationadjusted union payroll
weight would be
25.7
per cent.8 Thus, even in the absence of good data on
union versus nonunion total compensation, substantial errors in estimating
global compensation (as opposed to payroll) weights are not likely.
Alternative sources
of
compensation information.
One possibility for adjusting
payroll data to a total compensation basis, given the relative insensitivity of
the results, is to use an “average” effect found in the unionimpact literature
(e.g., Lewis,
1986).
The difficulty with that approach is that the literature
typically involves methodology specifically designed to extract the “pure”
union effect from the gross unionhonunion differential. We seek the gross
differential, not the pure effect.
A second possibility
is
to use information from the biennial
BLS
surveys
sThe
global payroll weight
of
22
per
cent means that there
are
$22
of
union
payroll
for
every
$100
of
total
payroll.
($22
x
1.3)/($100
x
1.2)
=
23.8
per
cent.
($22
x
1.4)/($100
x
1.2)
=
25.7
per
cent.
Measurement
of
Compensation
I
225
on expenditures for employee compensation. These surveys were discontinued
in the midseventies, however, and the outdated data raise obvious difficulties.
Moreover, the unionlnonunion breakdown in the published surveys applied
only to “nonoffice” workers rather than to all employees. Still, since the
surveys purport to measure approximately what we need, it is worth
examining their data.
In 1968, according to the
BLS
survey, the ratio of total compensation
(wages, fringes, and payroll taxes) to wages was 1.23 in the union
sector and 1.15 in the nonunion ~ector.~ Subsequently, the ratio pairs
(union:nonunion) were 1.25:1.16 in 1970, 1.28:1.18 in 1972, and 1.32:1.20
in 1974 (the final year for which unionhonunion data were published). If
ratios of this magnitude prevailed in 19831984, the global payroll weights
shown on Table
2
would have been boosted by about
1
percentage point on
a total compensation basis, i.e., from
22
per cent to 23 per cent,
a
rather
mild adjustment.
It is possible that the union impact on compensation (relative to wages)
grew from the midseventies until the 19831984 period on which Table
2
is based. That union wages relative to nonunion wages rose during those
years is well known. But there is no direct measurement available.
As
an
alternative to direct measurement, we can combine the national income data
cited above and a crude regression technique to factor out the union/
nonunion differential.
The national income accounts include data on wages and salaries
(corresponding roughly to earnings in the
CPS)
and total compensation on
a
detailed industry basis. Thus, it is possible to calculate the ratio
(R,)
of
total annual employee compensation to total annual payrolls for each two
digit industry in 19831984.
lo
Unfortunately, national income data do not
differentiate between union and nonunion compensation. However,
a
cross
industry regression with
R,
as the dependent variable and the industrylevel
union payroll weights as the independent variable will yield a roughand
probably upward biasedestimate of the union compensation impact.
l1
”Source:
U.S.
Bureau
of
Labor Statistics (1971, p.
20).
Wages are defined as “pay
for
working me.”
Related data cited below are from
U.S.
Bureau of Labor Statistics (1973, 1975, 1977).
“’National income data on compensation, wages and salaries, and employment appear annually in the
Survey
of
Current
Business,
usually in the July issue, and in various supplements.
The issues regarding interpretation of the coefficient are analogous to those involved in crosssectional
union wage impact studies, as discussed extensively in Lewis (1963) and in Mitchell (1980). If union
and/or nonunion
R,
values are infleenced by the unionization measure, biases will result. Since union
bargaining strength is likely to increase as the unionization measure rises, there probably will be an
upward bias in the coefficient. We did try various regression specifications using R, as the dependent
variable and union payroll weights
(PI)
as the unionization measure; none yielded impressive results.
For example, across 57 industries drawn from the national income accounts and matched with
CPS
data, the following regression was estimated: log(R,)
=
.14
+
.17P,, where both coefficients are
significant at the 5 per cent level but
RZ
=
.08.
This
regression suggests that for nonunion workers,
R,
=
1.16
and for union workers, 1.37.
226
/
SANFORD
M.
JACOBY
AND
DANIEL
J.
B.
MITCHELL
When applied to the national income data for
19831984,
this technique
yields an estimated union compensationtowage ratio of almost
1.4
and a
nonunion ratio in the
1.11.2
range. Thus,
if
this (undoubtedly too high)
estimate is used to adjust the
22
per cent global union weight of Table
2,
it raises that weight to about
25
per cent. It seems reasonable to assume,
therefore, that the global weight on a total compensation basis would be
somewhere between
23
per cent and
25
per cent.
A
TimeSeries
Perspective
The CPS did not collect data on union vs. nonunion wages during much
of the period of declining unionization, i.e., the period from the midfifties
to the present. However, it is possible to use alternative data sources to
make rough estimates of the time path of the union weight in payrolls.
Specifically, we used national income account data on wages and salaries per
employee for industries with aboveaverage and belowaverage unionization
rates.
Pay
trends for these two industry groupings were assumed to track
general union and nonunion pay trends, respectively. These data were
benchmarked to the
19831984
CPSbased payroll estimates (discussed
earlier) to produce comparable estimates for selected twoyear periods
beginning in the midfifties. The results are graphed in Figure
1.'*
The Figure permits comparison of the union employment and payroll weights
over the estimation.
It
shows that union payroll weights typically have been
35
percentage points above employment weights. And, based on the discussion
in the previous section, another roughly
2
percentage points should be added
to adjust the data to
a
total compensation basis. Thus, the importance of the
union sector in aggregate pay setting has been chronically understated by the
traditional employmentbased unionization rate.
On the other hand, although union wages generally rose relative to nonunion
wages during the period from the midfifties through the seventies, the widening
union pay advantage only partially offset the downward pull
of
the declining
unionization rate. And, in the eighties, the widening uniodnonunion pay trend
reflected in the national income account data has reversed, intensifying the
decline of the union payroll and compensation weights.
I2Unionization rates used to divide industries into the two groupings were taken from
U.S.
Bureau
of
Labor Statistics (1972, Table
2)
and applied to 1970. Forty industries were used (see Appendix). The
benchmarking technique involved linking the nationalincomebased percentage pay changes between
each twoyear period appearing in Figure
1
and the 19831984 period for which we had estimates of
union and nonunion pay from the
CPS.
Thus,
if in
an
earlier period union pay was estimated to
be
75
per cent of union pay in 19831984,
an
implicit union wage of .75 times the 19831984 level of pay
was assumed. Union and nonunion employment estimates of pay levels in each period could be
transformed
into
union payroll weights using BLS estimates of private union membership.
Measurement
of
Compensation
I
227
16
!
I
FIGURE
1
Union weights: Employment and Payroll
I
19531954 19581959 19631964 19681969 19731974 19781979 19831984
TWOYEAR
PERIODS
(7
EMPLOYMENT
+
PAYROLL
The
Employment Cost
Index
Beginning in the midseventies, the
BLS
has made available the
Employment Cost Index (ECI) as a measure of wage rate change. Initially,
the ECI was confined to private wages and salaries. During the eighties,
however, the index was expanded to include benefit and payroll taxes and
state and local government. Unlike the older compensation per hour and
average hourly earnings series, the ECI is published with a uniodnonunion
breakdown. The availability
of
these comparative data suggest that this index
might be a good source of information on union payroll and compensation
weightsa possibility reinforced by the
BLS’
touting of the ECI as the best
measure
of
wage and compensation
rate
change.13
I3According to the
BLS
(1982a, p.
78),
the
ECI
was developed because alternative wage indexes
“...were found to
be
fragmented, limited in industrial and occupational coverage, insufficiently timely
or
detailed,
or
subject to influences unrelated to the basic trend in employee compensation.”
228
/
SANFORD
M.
JACOBY
AND
DANIEL
J.
B.
MITCHELL
Unfortunately,
ECI
methodology precludes an accurate tracking of payroll
and compensation weights. To obtain information on wage change for the
ECI,
the
BLS
tracks ajixed sample ofestablishments. The
ECI
is
a
Laspeyres
index,
so
it
is
based on fixed employment weights, t00.l~ Until the mid
eighties, the base period for the
ECI
was 1970. Thus, the relative shrinkage
in employment in industries
in
which unions were concentrated after 1970
was not reflected in the
ECI.
The shift toward nonunion production
within
industries was also not reflected because of the fixed sample of
ECI
establishments. Most of this shift occurred as new establishments, which
were nonunion, replaced older union establishments. The newer establish
ments were not permitted to enter the sample.
Until the mideighties, the proportion
of
union employees in the
ECI
was
about onethird, well above estimates from other sources, and reflecting the
upward biases just cited. Given the overestimate of union employment,
payroll and compensation weights for the union sector in the
ECI
also were
too high.15
In
the mideighties, the base year for the
ECI
was moved to
1980 and new establishments were included in the sample. These alterations
in the
ECI
dropped the unionization rate embedded in the index to about
onefourth, still too high, because
of
post1980 developments.
As
long as
the trends of the eighties persist, i.e.,
a
shift toward industries which have
historically had low unionization rates and a shift within industries toward
nonunion production, the
ECI
will continue to place excessive weight on
the union sector.16
Conclusions
The proportion of workers represented by unions has traditionally been
used as a measure
of
the importance
of
unions in wage setting. However,
a
preferable measure is the percentage of the total national payroll located in
the union sector.
If
union workers receive identical rates of pay and worked
the same number of hours, there would be no discrepancy between these
measures. But, in fact, the pay and hours differences produce payroll or
compensation weights for union workers which are higher than their
employment weights.
I4For information
on
ECI methodology, see
BLS
(1982a, 1986) and Wood (1982).
I5The
BLS
will make available the payroll and compensation weights for the ECI but prefers that
they not be published. Hence, we cite
only
the employment weights in the text.
IhIt
should be noted that if the
ECI’s
union
and nonunion components are used separately
to
analyze
changes
in
uniodnonunion pay differentials;
for
example, the overweighting of the
union
sector
in
the
overall index has
no
effect. Moreover,
in
periods when union and
nonunion
wagechange trends diverge,
the overweighting will have only a small effect on the overall ECI.
Measurement
of‘
Compensation
I
229
When presenting these findings to labor market analysts, we have often
been told that “everyone knows” that payroll weights

rather than
employment weightsshould be used for judging the union sector impact
on wage indexes. Perhaps that is
so.
Specific statements in the literature
recognizing this fact are hard to find, however.17 Moreover, until the
CPS
data became available, estimates of the type we have presented could not
have been made readily. Now that data are available, analysts of the union
effect on general wage movements ought to be using payrollbased weights.
It would be helpful if the CPS methodology omitted the truncation of
reported usual weekly earnings at the $1,000 level. Median estimates would
not be effected, but means (which we have shown differ substantially from
medians) would become available directly. Similarly, if the establishment
surveyon which average hourly earnings are basedincluded information
on the union status of the sampled establishments, union payroll weights
could be calculated and reported. Finally,
a
revival of the survey of union
versus nonunion benefits practices would assist in estimating union weights
in total compensation.
Appendix
Industries Used to Calculate Pay Trends
in
Sectors with
Above and Below Average Unionization Rates
Industries with belowaverage unionization had unionization rates (union
members as a percentage
of
wage and salary workers)
S
20.8 per cent,
where 20.8 per cent is the private, nonagricultural average unionization rate
in 1970. These industries are marked with a “(B)” in the listing below. All
other industries had aboveaverage unionization.
Forestry and fishing
(B)
Mining
Construction
Lumber and wood products
Furniture and fixtures
Stone, clay, and glass products
Primary metals
Fabricated metals
Machinery except electrical
Electrical equipment
TO
the extent that researchers have devoted their efforts to developing union weights, those efforts
have focused
on
estimates
of
employment weights, not payroll weights. See,
for
example, Freeman and
Medoff
(1979).
230
/
SANFORD
M.
JACOBY
AND
DANIEL
J.
B.
MITCHELL
Motor vehicles and parts
Other transportation equipment
Instruments (B)
Miscellaneous manufacturing industries
Food and kindred products
Tobacco manufactures
Textile
mill
products (B)
Apparel and other textile products
Paper and allied products
Printing and publishing
Chemicals and allied products
Petroleum and coal products
Rubber and miscellaneous plastics products
Leather and leather products
Railroad transportation
Other transportation
Telephone communications
Other public utilities
Wholesale trade
(B)
Retail
trade (B)
Banking and other financial services (B)
Insurance and real estate
(B)
Business services
(B)
Repair services
(B)
Personal services
(B)
Entertainment (B)
Health services (B)
Welfare and religious services
(B)
Educational services
(B)
Other professional services
(B)
REFERENCES
Adams,
Larry
T.
“Union Membership
of
Employed Wage
and
Salary Workers,”
Current Wage
Flaim,
Paul
0.
“New Data on Union Members and Their Earnings,”
Employmenr and Earnings,
XXXII
Freeman, Richard
B.
“In Search
of
Union Wage Concessions in Standard Data Sets,”
Industrial Relations,

and
James
L.
Medoff. “New Estimates of Private Sector Unionism
in
the United States,”
Industrial
.
What Do Unions Do?
New
York:
Basic
Books,
1984.
Lewis,
H.
Gregg.
Union Relative Wage Effects: A
Sumey.
Chicago, IL: University
of
Chicago Press,
.
Unionism and Relative Wages in the United States: An Empirical Inquiry.
Chicago,
IL:
University
of
Developments,
XXXVIII
(March,
1986),
4550.
(January,
1985), 1314.
XXV
(Spring,
1986), 131145.
and Labor Relations Review,
XXXII
(January,
1979), 143174.
1986.
Chicago Press,
1963.
Measurement
of
Compensation
I
231
Mitchell, Daniel J.B.
Unions, Wages, and Inflation.
Washington, DC: Brookings Institution, 1980.
Theil, Henri.
Economics and Informarion Theory.
Amsterdam: North Holland Publishing, 1967.
U.S.
Bureau of Labor Statistics.
BLS
Measure
of
Compensation.
Bulletin 2239. Washington, DC: GPO,
.
BLS
Handbook of Methods,
Volume
I.
Bulletin 21341. Washington, DC: GPO, 1982a.

.
Technical Description of the Quarten) Data on Weekly Earnings from the Current Population
Survey.
.
Selected Earnings and Demographic Characteristics of Union Members,
1970. Report 417. Washington,
.
Employee Compensation in the Pnvare Nonfarm Economy.
Bulletins 1722, 1770, 1873, 1963, for 1968,
West, Sandra
A.
“Estimation
of
the Mean from Censored Income Data,” unpublished paper,
U.S.
Wood, G. Donald
Jr.
“Estimation Procedures for the Employment
Cost
Index,”
Monthly Labor Review,
1986.
Bulletin 2113. Washington, DC: GPO, 1982b.
DC: GPO, 1972.
1970, 1972, and 1974, respectively. Washington, DC: GPO, 1971, 1973, 1975, 1977.
Bureau of Labor Statistics, Office of Research and Evaluation, n.d.
CV (May, 1982), 4042.