Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access to The Quarterly Journal of
Economics.
http://www.jstor.org
Equalizing Differences in the Labor Market
Author(s): Charles Brown
Source: The Quarterly Journal of Economics, Vol. 94, No. 1 (Feb., 1980), pp. 113-134
Published by: Oxford University Press
Stable URL: http://www.jstor.org/stable/1884607
Accessed: 17-03-2015 13:02 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp
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.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
EQUALIZINGDIFFERENCES IN THE LABOR MARKET*
CHARLES BROWN
The theory of equalizing differences asserts that workers receive compensating
wage premiums when they accept jobs with undesirable nonwage characteristics,
holding the worker's characteristics constant. Previous research provides only inconsistent
support for the theory, with wrong-signed or insignificant estimates of these
wage premiums fairly common. An oft-cited reason for these anomalies is that important
characteristics of the worker remain unmeasured, biasing the estimates.
In this paper, longitudinal data are used to test this conjecture. Although such
data improve the control for worker characteristics, the plausibility of the estimates
is not markedly improved. Alternative explanations for these results are considered.
"It's indoor work and no heavy lifting."
-Senator Robert Dole, explaining why he wanted to be Vice President.
The theory of equalizing differences-that individuals are induced
to accept less attractive jobs by compensating differences in
their wage rates-is an important tool in economists' attempts to
understand the labor market. Both as a test of the theory and in order
to improve the measurement of compensation, researchers have attempted
to estimate "prices" for nonwage characteristics. Despite
evidence from studies of the internal wage policies of firms that
equalizing differences are present [Doeringer and Piore, 1971, pp.
66-68; and Reynolds 1974,p. 210] recent researchon the determinants
of individual earnings has provided rather limited support for the
theory (see Table I).1
* I have benefited from comments on earlierversions of this paper by C. Clotfelter,
G. Duncan, R. Freeman, E. Lazear, J. Medoff, J. Quinn, W. Vroman, R. Weiss, an
anonymous referee, and seminar participants at Boston College, Harvard University,
the University of Maryland, and the U.S. Department of Labor. I am indebted to the
General Research Board and the Computer Science Center at the University of
Maryland for financial support and computer time. Ollie Ballard provided skillful
research assistance. The paper was completed while the author was an economist at
the Office of Research Methods and Standards, Bureau of Labor Statistics. The views
expressed are those of the author and do not reflect the policies of the BLS or the views
of other BLS staff members.
1. Studies included in Table I were those using micro-data to explain the wages
of broad groups of workers. Interesting studies excluded by this criterion were Antos
and Rosen [1975], and Toder [1972] (both dealt only with teachers) and Mixon [19751
(who used time series data on 3-digit SIC industries). Each of these studies focused
on job characteristics quite different from those in Table I. Antos and Rosen found
that "equalizing wage differentials on working conditions are important sources of
teacher wage variation." Toder found that communities with high proportions of
nonwhites must pay a wage premium to attract teachers of given quality, but there was
no analogous differential for teaching children in poor communities. Mixon reported
that increases in the minimum wage led to statistically significant revision in at least
one (out of five) nonwage characteristics in fifteen out of twenty low-wage industries
studied.
i? 1980 by the President and Fellows of Harvard College. Published by John Wiley & Sons, Inc.
The Quarterly Journal of Economics, February 1980 0033-5533/80/0094-0113$01.00
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
Z S 3 a t-=u~~~~04
Q o - - *- *- * -
.
0
noXc tc o*_ o :o~_u
> S X E t E E 0 E|~~~~~~~~~~~~~~~~~~~0z
1- CY:? 0 00 00 ,,,CI 00 L )L > C- Lo t t
t~~~~~~ O C)E>- > )CoE 0 r-X O Oo- ?
a) co co w
?> r ,D_
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
E X> @ ?; 8 3 9 X;SNO @ C 'C
_~~~~~~~~~~~~~~~~~~~~~~cc *_o CD
C CS t CS O CS ~ CD e O O O 00 C ~ H 00 C CO CS4 CD O u C.)
O O O U o O CO ~~~~~~O O ~~~~OO~~~OCSeCSCSCS~NO 4
. . . . . . . . . . . . . . . . . . . . ..~~~c 4. . .
O O O CO CS O O O O O O O O O O O O O O O O O O O OO!
--.
_ _ _ _ _ I _ l l l l l l l l l l lI~$..,4
+ + + + + I I + 1 1 1 + I I I I I IIC) I-0
C U C4 P~~~~a = $ = Y v 4 ? U .? r = +~~~~~b7D OEE- E m
E = XE X E o E 3 E <,, = ,, ,?,tc c = 3; ;4 ,?CZ> 34? cd > CE b-O3
Q2 Q Q H^ Q ~~~~~~~~~;~~~vH= g v X v v=~~~~~~~n
LO
E --- C> :<
E- CN Cl E)cq ELOC) ?) C) 80 -1 E0LY co
C)C O -4, C) C) CY -4 E)C) r- r- = ~~r. = qc q
CYci 6 6 6 6 6 6 6 E E; 6 C
> a + '? 3+
Ef E-oEOg qOHzH~~~~~~~~~~~~~~~~C44) 9
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
CO b.0
b.0 Cq cr)
biD I
= Lc-
co
Cd
0 Cd
t- 00
m co CT CT cq
Q) C.0 oo CYDM .c CYDLo M M M co c"I
C)
cu
a) bjD
Cd
Cd Cd
C4-4 C4-4
4-4 0 0 f. 0
4
;-4 $-4
(O 0 b.0
Cd 4.-IE = 1 1-4
Cd-4 . E -4_4j,
b-0 In4 Cd
4--4
CO
la a) :
Cd b.0 b.0 u b.0.- 0 Cd$-A
Cd cd(u $-4
cd4 -4 --4 --4 cu
4 cq
co
$-4
Cd --4
64 CYD (D cnm -w CYD
cn NO
$-4
_0
b.0
co
cn
cu
>$-4
: 4-4
C4-4 b-CCIS
11 (1) Q)
Q) Cd cn
> co
-4
cd
>
y
cl
--4 7= --4
Cd
a la 9-, la
4, cd cd cd (M Cd
ck E--q l E. I-- E.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
EQUALIZINGDIFFERENCESIN THE LABORMARKET 117
Lucas [1977, pp. 554-55] found evidence of significant compensation
for repetitive work and somewhat smaller (though statistically
significant) compensation for jobs with bad working conditions
(hazards, extreme temperatures, etc.). Jobs requiring physical
strength appeared to command lower wages, other things equal, and
these differences were "significant" statistically. Neither Bluestone
[1974, pp. 132-222] nor Quinn [1975, pp. 112, 115] nor Hamermesh
[1977, p. 65] found consistent evidence of wage compensation forjobs
requiring physical strength or involving hazards or extreme temper-
atures.
Smith [1973] found that "the probability of [work-related]death
may be fully reflected in wage rates, but evidence of compensating
differentials related to nonfatal injuries is scant." Thaler and Rosen
[1975, pp. 289-94] concluded that, for workers in the most hazardous
occupations, occupation-specific mortality rates do exert a positive
influence on wage levels. Even here, the point estimates of this premium
and their statistical significance are sensitive to the other
variables included in the estimating equation and the functional form
(linear or log-linear) employed.
Taubman [1975, pp. 51L52] analyzed the relationship between
wages and the reasons that individuals in the NBER-TH sample of
above-average ability males gave for choosing their occupation. Interpreting
the latter as reflecting differences in their actual job
characteristics, he found that most had statistically significant,
plausibly signed, and often substantial coefficients. Duncan [1976,
p. 472] found substantial compensating differentials for some job
characteristics (freedom to control hours worked, safe working conditions,
and employment and income stability);2 however, the probability
of observing such estimated differences by chance was not
explored. Duncan and Stafford [1977, p 15] report positive premiums
for work effort and for jobs that restrict "opportunities to choose an
individual or flexible work schedule and an individual work pace"but
these differences become statistically insignificant when a theoretically
preferable wage measure is employed. Lazear [1977, p. 175]
reports significantly lower wages for young men enrolled in school,
"consistent with an equalizing-difference explanation which argues
2. Because Duncan reported standardized weights from canonical correlation
rather than unstandardized regression coefficients, his results were adjusted to make
them more comparable to others in Table I. The unstandardized weight as a fraction
of the mean wage equals the standardized weight times the coefficient of variation of
the wage divided by the standard deviation of the job characteristics. However, the
latter was adjusted so that the range of the job characteristic was zero to one in all cases,
making them more nearly comparable to the dummy variables used by other authors
in Table I.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
118 QUARTERLYJOURNAL OF ECONOMICS
that students optimally choose more flexible and easier jobs at the
cost of lower wages."
Evidence on other important job characteristics is more limited.
For example, human capital theorists maintain that individuals gain
entry into occupations with prospects of higher future wages only by
accepting lower current wages. Empirical support for this proposition
has been rather indirect (e.g., inferences from the pattern of the
variance of earnings by age [Mincer, 1974, pp. 58-59], or differences
in age-earnings profiles between groups [Lillard, 1977, p. 521].3
Schiller and Weiss [1977,p. 17] investigated the relationship between
pension benefits and wages in a sample of workersin large firms. They
found support for the equalizing-differences hypothesis among
younger workers but not among those nearing retirement. The relationship
between wages and other fringe benefits (vacations, health
insurance, etc.) has not been examined, although the existence of a
tradeoff is often asserted in collective bargaining contexts [Reynolds,
1974, p. 217].
The overall pattern that emerges from Table I is one of mixed
results: some clear support for the theory but an uncomfortable
number of exceptions. Among the studies that fail to find equalizing
differences, the most common explanation is the omission of
important worker abilities, biasing the coefficients of the job
characteristics.
The purpose of this paper is to provide a more appropriate test
of the theory. In Section I a model of labor-market choice that emphasizes
equalizing differences is presented. The model formalizes
the omitted variable argument and suggests that even "favorable"
results could underestimate the magnitude of these differences.
Moreover, the analysis in Section II suggests that a more appropriate
test of the model can be conducted when longitudinal data are
available.
In Section III the data are described. The National Longitudinal
Survey (NLS) Young Men's sample provides seven years' data on the
labor market experiences of males age 14-24 in 1966. Data on occupational
characteristics are taken from several sources and matched
to individuals on the basis of their occupation or industry. The results
are presented and discussed in Section IV. Concluding observations
are offered in Section V.
3. Hause's [1973] study of the covariance between earnings at different points
in time (using Swedish data) is an exception.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
EQUALIZING DIFFERENCES IN THE LABOR MARKET 119
I. THE MODEL
The central focus of the theory of equalizing differences is the
choice made by individuals with given personal characteristics (X)
among jobs with different wages (w) and differing nonwage attributes
(Z). In order to attract labor of a given quality, an employer offering
jobs that are hazardous or otherwise undesirable must pay higher
wages than employers offering jobs with more desired nonwage
characteristics. Therefore, an individual faces a set of jobs with differing
combinations of w and Z, and is assumed to choose among these
opportunities in order to maximize utility.
Formally, let Z be the vector of nonwage characteristics, measured
so that larger values of Z represent less desired jobs. ("Less
desired" is understood to reflect the preferences of the marginal individual.)
If f (w,Z;X) is the function relating wages to nonwage job
characteristics for persons with personal characteristics X, the theory
asserts that ow/lZ1 > 0 for all j.
This prediction can be tested once the function f(w,Z;X) is
specified. Unfortunately, theory provides no guidance in the choice
of functional forms, at least for most Z's.4 A convenient specification
is the semi-log equation,
(1) W-=ln(w) = XB + ZA + u,
where u is a random disturbance.5
Equation (1) is the estimating equation used in most of the
studies mentioned earlier, where the estimates of the A's were often
insignificantly different from zero and sometimes "significant" but
wrong-signed.
It is not difficult to explain this result, even if the theory is correct
(A > 0). Consider what happens to the estimate of A when some of
the X's are omitted from equation (1). For simplicity, suppose that
(2) W = Bo+ X113 + X2B2+ ZA + u,
4. For a discussion of the relationship of f to production relationships and workers'
tastes, see Thaler and Rosen [1975, pp. 268-86].
5. Equation (1) has been estimated by ordinary least squares in previous studies.
Several readers of earlier drafts of this paper have questioned the appropriateness of
such estimation, arguing that u and Z are inevitably correlated. Ordinary least squares
might be appropriate if all the systematic determinants of earning capacity were included
among the X's; one might then argue that transitory variation (due, for example,
to measurement error in reported wages) was independent of Z. Of course, X is never
fully specified; but the resulting difficulties are more easily understood from the
"omitted variable" discussion in the text than from a simultaneous-equations-bias
perspective. In any case, finding instruments for the Z's that are not themselves X's
would be extremely difficult.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
120 QUARTERLY JOURNAL OF ECONOMICS
where BI, B2, and A are scalars, XI and X2 are orthogonal, and data
on X2 are unavailable. The bias in the estimate of A will have the same
sign as the correlation between X2 and Z. If the people with more X2
''spend" some of their greater earning capacity on reducing Z, this
correlation will be negative,6 and the estimate of A will be biased
toward (or even beyond) zero. The omitted-variable argument is the
most frequently encountered explanation for insignificant or
wrong-signed coefficients [e.g.,Lucas 1977, p. 555;Toder, 1972, p. 440;
Quinn, 1975, p. 103;and Duncan and Stafford, 1977, p. 3]. Moreover,
it implies that even right-signed estimates may be too small [Antos
and Rosen, 1975, p. 137].
Rigorous generalization of this argument to cover more than one
Z does not appear possible.7 However, in general, the bias in estimating
Ah will have the same sign as the coefficient of Zh in a hypothetical
regression of the omitted X of the included X and all of the
Z's. If that hypothetical coefficient is negative, the estimate of Ah is
biased toward (or beyond) zero.8
II. THE IMPORTANCE OF LONGITUDINAL DATA
Clearly, it is important that the determinants of earning capacity
be specified as completely as possible. However, the data usually
6. Evidence in Lucas [1974]on the relationship between observable determinants
of earning capacity (schooling and age) and nine job characteristics is consistent with
this hypothesis. Bowles [1972, p. S238] argues that this positive relationship holds for
social class and nonmonetary job characteristics generally. Bailey and Schwenck [1972,
p. 15] report "an increasing emphasis [onemployer-financed retirement and insurance
plans] accompanying higher levels of earning power." Duncan [1976], however, found
that, controlling for education, other determinants of earning capacity (e.g., experience,
SES, tenure, test score) were not significantly related to nonpecuniary job character-
istics.
7. With two Z's, equation (2) would become
(2') W = Bo+ B1X + B2X2 + Z1A, + Z2A2 + u.
The bias in estimating, say, A2 has the same sign as
[COv(X1,Zi) CoV(X1,Z2) - var(XI) COV(ZI,Z2)] COV(X2,Z1)
+ [var(XI) var(Z1) - COV(Xi,ZI)2] COV(X2,Z2).
Assuming that cov(X1,Zj) is negative for all i and j, the sign of the bias is still indeterminate,
since we cannot rule out the possibility that the first term in brackets is
negative and large enough to make the bias positive. A moderately helpful assumption
is that r(X1,Zi) = r(X2,Z1), i = 1,2, where r is the simple correlation. In this case, the
bias has the same sign as r(X1,Z2) - r(X1,Zl)r(Z1,Z2), which will be negative unless
r(Z1,Z2) is positive and both r(Z1,Z2) and r(X1,Zl) are substantially larger than
r(X ,Z2) in absolute value.
8. Given two individuals with identical observed X's and identical Z's (except
for Zk ) and even-money odds, would you bet that the individual with greater Zk had
more or less of the unmeasured X? This question elicits one's subjective expectation
of the hypothetical coefficient: betting that the individual with greater Zk will have
less unmeasured X is equivalent to expecting that Ak is biased toward zero.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
EQUALIZINGDIFFERENCESIN THE LABORMARKET 121
available are deficient in that they provide no information on many
potentially important personal characteristics. For example, the
Survey of Economic Opportunity (the source of wage and personal
characteristics data used by Lucas, Bluestone, and Thaler and Rosen)
contain no measures of intelligence orfluency in oral communication,
and its measures of social background are far from ideal.9
To the extent that most of the omitted dimensions are individual-specific
(i.e., do not change over time), they can be summarized
by an individual-specific intercept. With cross-sectional data, including
individual-specific intercepts among the X's would be impossible,
since there would be one such "variable" for each observation.
Given data on wages and jobs of individuals at several points in
time, the use of individual-specific intercepts is a feasible strategy for
controlling for individual characteristics that do not change over
time.
Of course, some determinants of earningscapacity do change over
time (e.g., an individual's age, work experience, formal job training,
and marital status). To the extent that such dimensions remain unmeasured,
the omitted-variable problem persists, albeit (hopefully)
reduced. Fortunately, the NLS data include fairly detailed information
on many of these characteristics.
Including several hundred individual-specific intercepts in the
X-matrix would exceed the capacity of almost any computer program
that calculates regressions. Fortunately, there is a computationally
feasible alternative.
Define
Wt= W-it, 1,2I , (numberof individuals)
wi wi LWt,
(3) ijt Xijt - - jt, = 1, 2, ..., (numberof X's)
T
t=(
Zikt = Zikt - Zikt, k = 1,2, .. ., (number of Z's).
T t=1
It can be shown10that regressing W on X and Z gives the same estimates
of B and A as regressing W on X, Z, and the set of individualspecific
intercepts.
9. See Bowles [1972, section III].
10. The key is to partition the independent variable matrix so that the set of individual
intercepts is in one block and the X's and Z's are in the other. Note that the
standard errors for each coefficient calculated by standard computer programs from
the transformed data must be corrected to reflect the loss of degrees of freedom due
to the (swept-out) intercepts. See Pacific Consultants [1976, pp. 202-15].
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
122 QUARTERLYJOURNAL OF ECONOMICS
Equation (3) emphasizes that, with individual-specific intercepts,
it is changes in W, X, and Z over time for each individual that identify
B and A. This would cause difficulties if individuals remained in the
same occupation, since the changes in Z would presumably be negligible.
Fortunately, occupation-changing is the rule rather than the
exception for young men. In the sample analyzed below, 85 percent
of the individuals changed 3-digit occupations at least once between
1966 and 1973, and 60 percent moved to a new broad (one-digit) occupational
group.11
III. THE DATA
The NLS Young Men's sample provided data in each of the seven
years 1966-1971 and 1973. Eliminating individuals who were college
graduates, were in school at any of the survey dates, or did not provide
usable data on wages, industry, occupation, or the other variables
described below reduced the sample from 5,225 (roughly 2,000 of
whom were out of school in 1966) to 470. The subsample studied here
was older, had more stable labor force attachment, and came fromless
"advantaged" backgrounds than the larger sample.
Given the individual-intercept strategy, only determinants of
earning capacity that change over time need be (orcan be)12included
explicitly in equation (1). The effects of those determinants of earning
capacity that do not change over time (e.g., schooling or race) are reflected
implicitly in the intercepts.
The Young Men's surveys provide information on seven determinants
of earning capacity that change over time:
1. Six dummy variables (which equal one for observations from
the 1967, 1968, 1969, 1970, 1971, and 1973 surveys, respectively, and
zero otherwise) were included. These dummies capture the general
growth in wages due to technical progress, price inflation, etc., and
the effects of time spent not working.
2. Three types of variables reflect human capital investments.
The first is the cumulative time spent in civilian formal training
programs, measured in hours of training/(40 X 52). Three types of
training programs could be distinguished: company training, parttime
school courses, and "'other."'13The second is cumulative work
11. Individuals who do not change jobs are still "useful"observations, in that they
help to determine the coefficients of the individual characteristics.
12. Including a characteristic that did not change over time would make that
characteristic an exact linear combination of the individual-specific intercepts.
13. Part-time school included business and technical school, regular school, and
correspondence courses. Because formal apprenticeships are coded separately only
in 1966 and 1973, they were included in the "other" category.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
EQUALIZINGDIFFERENCES IN THE LABORMARKET 123
experience since 1965, measured in weeks employed/52. Presumably,
this variable measures the productive effect of past on-the-job investments
associated with informal "learning by doing" rather than
formal training programs. Finally, tenure with current employer,
measured in months/12, is included to capture differences in firmspecific
human capital investments.
3. Two measures of unionization were employed. First, the
fraction of workers unionized in the individual's (3-digit Census) industry
was coded separately for office and production workers, based
on data in Freeman and Medoff [1980]. Second, the probability that
the individual worker was himself covered by a collective bargaining
agreement was included. For 1969-1971, this was a dummy variable
(1 = covered, 0 = not covered) from the NLS file. Whether the worker
was covered by a collective bargaining agreement was not ascertained
in the remaining years. If the worker was working with the same employer
in, say, 1968 as in 1969, the 1969 dummy was coded for 1968.
If the worker was with a different employer, the probability of his
being covered in 1968 was taken to be the fraction of (office or production)
workers in his industry who were covered.
4. Marital status was represented by a dummy variable that
equals one whenever the respondent is currently married and zero
otherwise.
5. Geographicvariation in wage rates is reflected in three dummy
variables, which equal one whenever an individual worksin an SMSA,
lives in the South, or lives in the West, respectively, and zero other-
wise.
6. The effect of involuntary job separation was represented
(crudely) by the cumulative number of times an individual has been
fired or laid off.
7. The effect of health problems on earning capacity was represented
by a dummy variable that equals one whenever the respondent
reported a health problem that interfered with work activi-
ties.
In order to compare the results presented here to those of other
researchers, conventional determinants of workers'wages that do not
change over time were included instead of the individual-specific
intercepts in some regressions.These variableswereyearsof schooling,
months of military training programs, labor-market experience in
1965, knowledge of the world of work, an index of the socioeconomic
status in which the individual grew up, and race.
An important source of data on job characteristics was the Dictionary
of Occupational Titles (DOT) file. Originally, each of over
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
124 QUARTERLYJOURNAL OF ECONOMICS
13,000 DOT occupations was assigned a dummy variable for each
characteristic (1 = present, 0 = absent). These 13,000 occupations
were then aggregated into the nearly 300 3-digit Census occupations
using a conversion matrix based on the October 1966 Current Population
Survey. Thus, the "score" for each occupation represents the
probability that a randomly selected individual in that 3-digit occupation
has the given job characteristic. For details see Lucas [1974].
Both "laborers, not elsewhere classified," and "operatives, not elsewhere
classified," are subdivided according to Census industry, in
order to reflect the heterogeneity of these important "residual"
categories.
Four characteristics were selected for study here:
1. the job requires performing repetitive functions;
2. the job requires working under stress;
3. the job requires physical strength;
4. the job involves bad working conditions (extremes of cold or heat,
humidity, vibrations, or hazards).
An alternative, more specific measure of bad working conditions,
the increase in the actuarial probability of death associated with
hazardous occupations [Society of Actuaries, 1967] was also used.
Since these data measure the extra risk from working in such occupations,
occupations not covered were assigned a value of zero. This
made it imperative that Society of Actuaries' occupations be matched
to Census occupations (or occupation-industry pairs) whenever
possible. These annual extra-mortality probabilities have been
multiplied by 1,000 as a scaling factor;thus, we have the extra deaths
per thousand man-years in each occupation. (Among all males age
twenty-five, about two out of 1,000 die each year;thus, the coefficient
of this variable may be interpreted as the compensation for increasing
the normal risk of death by one-half.)
Data wereavailable fromvarioussourceson fourother potentially
important job characteristics:
1. The number of hours usually worked by the individual was
recorded each year by the NLS. The logarithm of hours worked and
a separate dummy variable for part-time work (less than thirty-five
hours per week) were used.
2. A dummy variable that equals one whenever the individual
worked for a federal, state, or local government, and zero otherwise,
was constructed. It was intended to reflect the greater job security of
government employment [Blechman, Gramlich, and Hartman, 1975,
p. 68], although other interpretations are possible.
3. When an individual was currentlyenrolled in a formaltraining
program (e.g., company training program), the NLS file gives the
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
EQUALIZINGDIFFERENCES IN THE LABORMARKET 125
number of hours per week the individual spends in training. The ratio
of training hours to usual hours worked was coded as a measure of
self-investment opportunities of the job. The NLS did not determine
whether the firm paid the employee for time spent in training, or for
tuition, or other expenses. This seems almost certain for company
training, unlikely for part-time schooling, with "other" training in
between. Given Lazear's argument that students select more flexible,
less demanding jobs while enrolled, one would expect a negative sign
even for part-time schooling, but the interpretation of this coefficient
is unclear. Finally, workers who reported their occupations as being
in Census "apprentice" categories often did not report themselves
to be enrolled in formal training.14Whatever the "formality" of apprenticeship,
there is strong reason to believe that substantial onthe-job
training occurs in these occupations. Therefore, a dummy
variable that equals one whenever the respondent's occupation is one
of the Census apprentice categories, and zero otherwise, was created.
Human-capital theory leads one to expect a negative sign for this
variable.
4. Ideally, total compensation (including fringe benefits) would
be used as the dependent variable. Lack of appropriate data has
precluded this in the past, and the NLS data are no exception. However,
data on total expenditures for labor compensation, divided into
wage and nonwage components, are available for 2-digit SIC industries.15Annual
data from 1965 to 1969 were averaged, and the values
assigned to each Census industry were those of the 2-digit SIC industry
to which it belonged. For manufacturing industries these estimates
were refined using data on a 3-digit SIC level for 1967 from
the Annual Survey of Manufacturers [1973]. From these data the
ratio of nonwage compensation to wages was calculated, and matched
to each individual according to his Census industry. If the logarithm
of total compensation is some function of X and the other job characteristics
(i.e., h(X,Z)), then
ln(wage) = h(X,Z) - ln(1 + nonwage compensation/wages)
h(XZ) - (nonwage compensation/wages),
so that the coefficient of our measure of nonwage compensation to
wages would be -1.0.
Two other "job characteristics" taken from the DOT file were
also included in some regressions: GED (general educational devel-
14. For example, in 1966 and 1973 (the only years in which formal apprenticeship
programs were recorded separately), only half of those who were in apprentice occupations
were recorded as being in apprenticeship programs.
15. Unpublished data, Bureau of Economic Affairs.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
126 QUARTERLYJOURNAL OF ECONOMICS
opment) and SVP (specific vocational preparation). Their inclusion,
in addition to the other determinants of earning capacity described
above, can be justified in several ways: as measuring omitted productive
characteristics of individuals; as reflecting perceived productivity
differences among individuals that are not explained by the
other X's; as reflecting wage differentials that workers in some jobs
receive over identical workers in other jobs.
IV. RESULTS
Estimates of the parameters of equation (1) appear in Table II.
Columns 5 to 8 include individual-specific intercepts, while columns
1 to 4 do not.
In columns 1-4, coefficients of the individual characteristics are
generally in line with expectations and previous research using these
data [e.g.,Griliches, 1977;Kohen, 1972]. The human capital variables
are generally significant. The exceptions are cumulative company
training and tenure; the near-zero coefficient for tenure probably
reflects the fact that job search and consequent job changes are important
for workers in their twenties. If anything, the cumulative
experience variable has an implausibly large effect. The unionization
variables, marital status, and geographic variables have the expected
impacts. The (crudely constructed) layoff-discharge variable has no
effect on wages; health problems have a small negative impact, but
the coefficient is about the same size as its standard error. Race,
knowledge of the world of work, early experience, and years of
schooling have the expected effects on wages, while military experience
has little impact.
The individual characteristics in columns 1 to 4 are more comprehensive
than those typically included in the studies in Table I.
However, the coefficients of the job characteristics in columns 1-4
display the same inconsistent relationship to theoretical predictions
one observes in Table I. The government worker coefficient is significantly
negative and reasonable in magnitude. The supplements
variable is wrong-signed and very significantly different from -1.0.
Time spent in school or "other" training programs has a fairly large
negative impact on current wages, but time in company training or
being an apprentice does not. The four DOT variables are either
"significant" and wrong-signed (repetitive work in columns 1 and 2,
and bad working conditions) or insignificant. A zero coefficient for
jobs requiring physical strength is sensible for workers in their
twenties, but the other three characteristics should generate positive
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
- * - - - - - *~~~*~-~ct' .c C 00 U- 'It 00 r- r- 00 Cn CD H .H .mDC CYD Cq 00 .cD CS X
C) r-- CYD r- C cq r- CY C'l co -t LC C r-- cq r-- <= r-- <= (= C) C
00 C C C ) CC Cl C~CC Cl )) C'
CO C) Cl C) C)C ) )
_ _ _ _I _ I _I _ _ I'-_ _ _'-
0
"t Ce CD 00 Lo I'l sm 00 Cy:)O cCD A 00 CYD Cq 0 CD C
C C) )C)~C)
LO
C~C~ C C ) CC (=C C) C) C) C) C- C)C ) Cl
-
CC
CC
Cl
-CCl 6 Cl C
-* - - - - - - - -~~* ~-*-
CC
CC O 0C dz 4N .
m00 LOC ) r 4 CT C4 C) CC Cl C0 CD Cl
CC C C CC C
)
l Cl
- * - - - - - / - - - ~~~~~~*,-.*
CC CC LO 00 O d 00C O 00 Cl C H. CD C
t
- CC . 00 CD Cl
Cy: U.) o o 11 o oooo ooo4
) C MCC C l MCC C C LO r-4 Cl r rCC
Cl C>lCl C)C) l ) C) C Cl N C C> C> C> C)C) C)C )C )C
oo cq o) cq cq CYD Cq "I o= U- CII r ,t M C: r- CO r- 1t r- M r- o= o
- ~ ~~~~~~~~~~~~cql . cl L- (m )- t-
U.
r-4MIl M r- -. " -.4c
-~~~~~~~~~~~r-L C
- - - - -- * - * - *
- H
r C r C C M = >" C
C
L M M = r M r " r =
C~~~~~~~~~~~~e m r-. m 0 mooooooooo
CQ
?- CD C i cqCQ -O Y r-O 't r- LO "It r- L- m C> r- CY r "t r-- o: H o o
C ) O) O1 LO O LO O LO O LO O LO O LO O- O- O O O LO OO .O)
6666 6 86666 66 6 6 66 6i 6 6
H _ _
CQ
C C lCO O ~00 CHH 0-O LO s ~ CC t O Cl
i Cl i C0l mCC d *' t Ol N N Om O O
LO o o 6 66 66O~O O 666OOOOOO
> _
= LnO n LO
n O n LO
n O n LO N
> O
C% CQ LO _N
_
ct~~~~ C, o' Co t CC t CC t : CC e t ccs
_
o o ~ co co co co ~ n H n o t co o t m o o
. . . . . . . . . . . . . . . . . . ..o
Ki w H O O o o o o o o oo ooooooc snc
_~~~~~~~~~~~~~~~~
n ~ ~ ~ 0 .z = .
ct z O ezz t:) Qf w Q PaD~~~~~~~~~~~~~C
>~~~~~~~~ -l 1 l l 1 l
A Y U U U U Y E E E E
N 00Cs , Q CC= b Ci2: : >
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
* - * - * - * - * - - *
O) CD OC C O L D Cm O Cl Lo N I C N C Om
CC.'. N OI Cl C lC) C C)C )C
oo M
C
cg
C
o
C)
o
Co
O O . . . . O. . . . . . . . .O . . O.
C C) C)N) N) CCl ClC)C)C C)C) C)C )CO LO ) t-O LOO O LO O OLO LO O O- O O
* _- * _ ~* _~* _~*_ _*_ _ _
o c cscs e to cC to tCl to C HCC r s
ooO O N Cf Ol oCO O C rO O
* _-* _-* _-* _-* _ _
LO O C> r-- C) O7 CS) CO CO r- O) C> C r-- C C) C) O
C CCO > NCto C) toO C0 to Co CC NO C
C9 O N O ~ Ol O O l ~O CO
* ~- * ,- * ,- * ,- * ,-* ,-* ,-- ,-CC C ,
Cl C : CC Cl - Cl to r 0)M C CC CC Cl (0 MC) 1 to Cl zp
CC CN N N Clo Co C - C C) ~ CCC. C. C
* _-*_
* _ * _ * _ ~* _ * _ * _ * _ _* _ CtoCS) CC* _~*_
CS <=>CC CC 4 CC) Cl to (: N Cl < CC: Cl CC C) _ to > Cl Cl C Ct CC
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Co CC CC CC Cl CC Cl to o N Cl : C C CC Cl ) C)Co o t o o o Cl Cl Co Cc
* _ ~* _ ~* _ ~* _ ~* _ ~* _ -__ * _ ~Coo C* _- *_ _
,t m cm "t r-c ( cq LO r- L- oN oo cs Cq Co : a) <= ., Lo (= o-- <c Cq Cq IL OX
C CC C- CC C ClClO to ,: CC C O CC CC - rtO O tO O : Cl Cl: O to CC
CC> Nq N: C=l ClCC>CC C >C C C : CC C:
CO ~~~O C>1 0 t1 0 O 1 0 1 0 0 0 0 000~~~~~~~~~~OOOOOOOOOOO~q) C
* _ _ * _ * _ * _ * _ _ _ _ * _ e~~~~~~~~~~~~~~~~~~~~~L00 L- MO* _ * _
Ics cm cs cm 0 cs1 Co c1 Lo r- r- cq -. 0o o. qcM 0o t L o (= o-- (= Cq cq t co 00
C C M C) CCCl CCCN O CC C C) CO OC C) C) -)N'C C t OO O CO
* _ * _ * _ * _ * _ * _ _ _ _ * _ CO~~~~~~~~~~~~~~~~~~~~~00 CDcCy:) _*
O CyD 1 14 CyD CYD Cq Cq Lo r Co Cj 0 0 IH CC ) U.) t U.) H Cq Cq ,,I LoU:
ce tr C CN Mc- mc q q m rc = =o o Cq r-cs Co o o o o oo
> x x L ssOm nC ) m OL- C CI Lo m m x Om _) _- _ LO
cfs
-r
, f- M
r Lo CeD L-c or ,c
M M M
Cs D 0 Co Lo Co7s C)o -"Ir t o LO cm
M .,I
. . .
.,I
.
.,I
., . = .
.II
. .= .q .- . . Lo . . .= . .= .= .- ~...
z w O O OO O O O O O O O O O O O O O O O O~~~~~~~~~~~~~~~~~~~~~~Cj 06 r- O Ci C'i ; CS O5 O
c C - - - - - - - - - - CO - <5) C'l - C)~~~~~~~~C
Co t
C)
C) CO)
C ~~~~~~~~~~~C
-. 5 ? | C d) - i8?
4 >
CoC -CIS
CIS.~ ~ bC.
o> S c4
CCC 5
(n
)ccn cis~
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
* * _ _
*
~-* *0m N CS LfC)O00 C 0 LO Im>
00 Lo I >-: C> LOC0 Cm CqC O, LO ? q I
O OCl 1 0 0 C> 0OC> ClC 0 0Csl 00C l
CYD Cq LO r-- LO 00 "t LDeOCO CID
m
1H
L
C t Cy: Cq "t ,t Lo O q 1C- Cq CO CoCM C O > I- Co MC LO Cj r- -t O C>1C C 1 Lo Cq LoCM -t Cq L-CM O= CM?
O> O cq O) O l C: O: O CS OOOCq 00 I
O 't Cy: r- Cy: r- -t r- Cq LO > t I' O ) L- O C) t ( Cq Lo 00 Cm Cq Cq M
00NCS 9CDC)O > DOCCq e Cti CS O H LO) C> LO Cm C)
,C t- o
CcM c
Lo rjCq I, CYD Csq o o cccI~~j <=> Cq C) C) CT C) C)C>~~~~~>n Cq O I
O
Cy: .0 L
fO H O 'O O cO cOo O O LO O- O)
'll
0 0 0qC CS0 m I
C:
C
Clt CC
C0
CC
O
Cl CC
ClCC
o- o-- o o
-
o
oC
o
o o Cy o OC ooo
I~~~ ~ ~ ~~~~~~_0 _0 _0
I _
Iq _q
I I _ I, _ _
Lo I. _
I= _y I _ C
C) C) C) C CIo r- C> oo (= oo C> C) (>C> C cs m c) cn c) c) cs Cl c) C'I ~co
66666~~~~~~~~666566 66 6 66 6 6666 6 6 6 56 6 6) ),
o~ o- Lo
ro=
o-~C o
Hoo oo0 oq ooY oq o-- oY oj ~C o o o o o
Cq
o-- CM C O I
CNI
o- o Lo o- o o o) o o1 or o o oo Lo oq o~ oq o o o Loo o oo oo ^
* _ _ -_* _~* _ _* _ _ _ * _ ~* _ ~* _-~ I
~~~~~~~~~~~.00 X 00 COs=> U ,,I r- C> MC Cq CoCfo- Oo m C) Cm
* _ _ _ * _- * _ _ * _ _ _ * _ * _-* _I
C) C) o) m C)C0 '. Cl Cl CC oo cC CC r Z C> Hc>C l>
ICI C;I CC
00 -0 0 M ~ CCCl-L t- 0 0L r-q L C>CCL
I-r- r- r-
o>
I- Loo oo o ce C:)
M C orcst
cM (> Cq co C)
o t- co o0 o0 c o q Lo o- o o. o o1 o Loo0 o coo~ o1 o- cs Cq
oro r- Loo C) oo
Lo (=> cto -(>H oo Lo LO
> o o o
m r ~~~e O cs O e O cs H co O cs cs) > 00 O ~~~~~~~~~~csLo cq L-c Cqcs 00 o urCq
66666666~~~~C;C;C566C 666666 6 6 6 6 ~6 6 6 66<6
00 D *o E . oE . CCE c C<1 ? > 00 to QtoC Cl O00 00 cnCf
D Cl Ev CC CCO Cl to C CC ZC ; ;to CC C
6666666666666666666666~~~~~~~~~~~~~~~~~~6666666~~c C
En~~~~~~~
0 ~~~~~~~~~~~a C.o()>
C CZ
~~~ S C C C ~~~~~~~~~~~~~~~~~-~~~~~ C ~~~~~~~C) I
CIS ~~Cd cis ~
C
-C
;4a
as as 4-~CC,
cr) ~4 9L
C
U
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
130 QUARTERLYJOURNAL OF ECONOMICS
wage differentials. The risk of death variable has a statistically significant
positive coefficient that is roughly three times as large as that
reported by Thaler and Rosen. Part-time workersreceive lower wages
(in line with analogous results for females; see Rosen [1976]) but
otherwise longer workweeks generate lower hourly wages. Including
GED and SVP makes the coefficients of repetitive work and bad
working conditions less negative, but has little additional impact.
Individual-specific intercepts are included in columns 5 to 8. The
coefficient of cumulative part-time schooling is reduced, and the effect
of cumulative "other" training is eliminated. The impacts of the regional
dummies are less in line with those in earlier studies. The
coefficients of the unionization variables decline, but this was predictable.
If some firms pay above-market wages, they should attract
better qualified workers, thus offsetting part of the initial wage differential.
If the individual-specific intercepts do in fact provide a
superior control for variation in workerquality, their inclusion should
reduce the impact of unionization.
The impacts of the intercepts on the coefficients of the job
characteristics vary considerably, and there is no marked improvement
in the correspondence between these coefficients and a priori
predictions. The government worker and supplements/wages variables
are nearly unaffected. The time in company training variable
remains negligible, the coefficients of the part-time school and other
training variables are reduced in absolute value, but the apprentice
dummy acquires a significant negative coefficient. The effects of repetitive
work and bad working conditions become less negative (the
former remaining "significant"), while stress and strength are unaffected.
The risk of death variable, however, loses its significant
positive effect. The coefficients of the continuous workweek and the
part-time dummy variables have become less negative.
The lack of consistent improvement in coefficients of the job
characteristics due to the intercepts might be attributed to a lack of
variation in the transformed variables (see equation (3)), leading to
imprecise parameter estimates. In fact, however, the standard errors
of these coefficients are not substantially raised by the addition of the
intercepts. This reflects the fact that occupation-changing among
young workers is common, so that "within-individual" variation in
job characteristics is substantial.
Several experiments with the estimating equation are not reported
in Table II. First, the cumulative training variables were coded
using months ratherthan hours of training (i.e., neglecting hours spent
per week). Second, the extra risk of death variable was replaced by
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
EQUALIZING DIFFERENCES IN THE LABOR MARKET 131
a variable that took the values used by Thaler and Rosen in their
subset of risky occupations, and zero otherwise, to test whether the
coding of the additional occupations was responsible for the differences
from their results. Third, rates of growth in employment in the
individual's industry and occupation from 1960 to 1970 were added
as explanatory variables, to test the idea that employers in growing
industries or occupations might offer both higher wages and better
working conditions to attract more employees. None of these changes
led to results appreciably different from those in Table II.
Various restrictions of the basic sample were also considered. The
sample was divided by race, and also by years of schooling (did/did
not graduate from high school). While some of the job characteristics
(government worker, time currently in training) were more often
significant for whites and high school graduates, there was no clear
pattern to the disaggregated equations. Next, the sample was restricted
to those with scores of at least 28 (out of a possible 56) on the
"knowledge of world of work" test. The motivation was to exclude
those with the least information about the job market. Unfortunately,
the test emphasizes questions that would be verifiable from Census
tabulations rather than reflecting detailed knowledge about the individual's
local labor market. In any case, the results for the resulting
2,639 observations were not markedly different from those in Table
II. Finally, the sample was restricted to individuals who had been out
of school for at least two years at the 1966 survey (2,674 observations),
to check whether individuals finding their way in the labor market
were obscuring the more systematic behavior of other workers.
However, the results were qualitatively similar to those in Table II.
V. CONCLUSIONS
The hypothesis that the inconsistent support for the theory of
equalizing differences that characterized previous studies was due
to the omission of important dimensions of worker quality was not
supported by the data. Despite reasonably adequate measures of those
worker characteristics that change over time and a statistical technique
for holding constant differences that do not, the coefficients
of job characteristics that might be expected to generate equalizing
differences in wage rates were often wrong-signed or insignificant.
One is left with several explanations for this failure, none of which
is entirely convincing.
1. "Labor markets are simply not as competitive as the theory
of equalizing differences assumes." While the assumptions of the
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
132 QUARTERLYJOURNAL OF ECONOMICS
perfect-information profit- and utility-maximizing model most often
used to explain the equalizing difference hypothesis-and relate it
to applied welfare economics [Thaler and Rosen, 1975]-may be too
strong, considerably weaker assumptions still imply such differences.
Suppose that wages and working conditions are determined by collective
bargaining without the threat of extinction compelling these
decisions to conform to cost-minimizing outcomes. Suppose that
workerslack information about workingconditions and underestimate
the differences in workingconditions among firms. As long as workers
prefer better working conditions and higher wages, and employers
hire the applicants they perceive to be most qualified, the relationship
between wages and unpleasant job characteristics holding worker
quality constant should still be positive-though weaker than the
stronger set of assumptions would imply.
2. "The marginal worker's tastes may be different from those
assumed in the a priorisigning of the coefficients." For example, while
some workers abhor physical labor, others prefer it to more sedentary
endeavors; thus, jobs requiring physical strength may not be unpleasant
for the marginal worker, and no equalizing difference would
be required.However, this conjecture is much less convincing for most
of the other job characteristics in Tables I and II.
3. "The job characteristics are not well-measured." Undoubtedly,
there is a large element of truth to this assertion, particularly
for characteristics "matched" on the basis of occupation rather than
being reported directly by the worker. It would be an attractive explanation
for coefficients that fell a little short of plausible magnitude
or statistical significance. But it is difficult to construct a measurement-error
rationale for coefficients that are wrong-signed and significantly
different from hypothesized values (e.g., repetitive work
or supplements/wages).
4. "Omitted variables-both individual characteristics that
change overtime and job characteristics-may be biasing the results."
Admittedly, some determinants of changes in individual productivity
(e.g., intensity of informal on-the-job training) remain unmeasured.
However, the results were little improved when individual-specific
"abilities" were controlled. The omission of some job characteristics
raises a more complicated issue. One might expect good job characteristics
to be positively correlated in general. However, controlling
for all X's, the partial correlation among job characteristics is more
difficult to assess. It may be, for example, that individuals who do
repetitive work have more freedom to work overtime or require less
(costly) job search to find jobs. Lacking data to hold these omitted
characteristics constant, one can only speculate.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
EQUALIZING DIFFERENCES IN THE LABOR MARKET 133
5. "Testing the hypothesis on a sample in their early and
mid-twenties is inappropriate." It is not obvious why workers in this
age range should provide weaker support forthe hypothesis than older
workers.After all, the common stereotype of youth is one who is overly
sensitive to working conditions, insufficiently willing to put up with
repetitive, stressful, or otherwise unpleasant work in order to "make
something" of himself.16 In any case, as reported in Section IV, deleting
the least experienced fifth of the sample failed to provide clearer
support for the hypothesis.
One could undoubtedly construct a more convincing case for each
of these explanations, but it is doubtful that it would be fully satisfactory,
explaining the "successes" in Tables I and II as well as the
failures. While the present paper provides little support for an oftused
explanation, the task of choosing (or combining) the alternatives
remains.
tJNIVERSITY OF MARYLAND
REFERENCES
Antos, Joseph R., and Sherwin Rosen, "Discrimination in the Market for Public School
Teachers," Journal of Econometrics, III (May 1975), 123-50.
Bailey, William R., and Albert E. Schwenk, "Employer Expenditures for Private Retirement
and Insurance Plans," Monthly Labor Review, XCV (July 1972), 15-
19.
Blechman, Barry M., Edward M. Gramlich,and Robert W. Hartman, Setting National
Priorities: The 1976 Budget (Washington: The Brookings Institution, 1975).
Bluestone, Barry, "The Personal Earnings Distribution: Individual and Institutional
Determinants," Ph.D. thesis, University of Michigan, 1974.
Bowles, Samuel, "Schooling and Inequality from Generation to Generation," Journal
of Political Economy, LXXX (supplement) (May/June 1972), S219-S255.
Doeringer, Peter, and Michael Piore, Internal Labor Markets and Manpower Analysis
(Lexington, MA: D. C. Heath, 1971).
Duncan, Gregg,"Earnings Functions and Nonpecuniary Benefits," Journal of Human
Resources, XI (Fall 1976), 462-83.
-, and Frank Stafford, "Pace of Work, Unions, and Earnings in Blue Collar Jobs,"
unpublished manuscript, March 1977.
Freeman, Richard, and James Medoff, What Do Unions Do?, forthcoming. 1980.
Griliches, Zvi, "Wages of Very Young Men," Journal of Political Economy, LXXXIV
(supplement) (Aug. 1976), S69-S86.
Hamermesh, Daniel S., "Economic Aspects of Job Satisfaction," in Orley Ashenfelter
and Wallace Oates, eds., Essays in Labor Market and Population Analysis (New
York:John Wiley and Sons, 1977), pp. 53-72.
Hause, John C., "The Covariance Structure of Earnings and the Job Training Hypothesis,"
unpublished manuscript, December 1973.
Kohen, Andrew I., "Determinants of Early Labor Market Success Among Young Men:
Ability, Quantity, and Quality of Schooling," Ph.D. thesis, Ohio State University,
1973.
Lazear, Edward. "Schooling as a Wage Depressant," Journal of Human Resources,
XII (Spring 1977), 164-76.
16. Moreover, using older workers would make it likely that current wages would
be quite different from the wages anticipated by those workers when they made their
career decisions.
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions
134 QUARTERLYJOURNAL OF ECONOMICS
Lillard, Lee, "Inequality: Earnings vs. Human Wealth," American Economic Review,
LXVII (March 1977), 42-53.
Lucas, Robert E. B., "Working Conditions, Wage-Rates, and Human Capital: A Hedonic
Study," Ph.D. thesis, M.I.T., 1972.
-, "The Distribution of Job Characteristics," Review of Economics and Statistics,
LVI (Nov. 1974), 530-40.
-, "Hedonic Wage Equations and Psychic Wages in the Returns to Schooling,"
American Economic Review, LXVII (Sept. 1977), 549-58.
Mincer, Jacob, "The Distribution of Labor Incomes: A Survey with Special Reference
to the Human Capital Approach," Journal of Economic Literature, VIII (March
1970), 1-26.
-, Schooling, Experience, and Earnings (New York: Columbia University Press,
1974).
Mixon, J. Wilson, "The Minimum Wage and the Job Package," B.L.S. Working Paper
No. 67, January 1977.
Pacific Consultants, The Impact of Win II: A Longitudinal Evaluation, processed,
1976.
Quinn, Joseph, "The Microeconomics of Early Retirement," Ph.D. thesis, M.I.T.,
1975.
Reynolds, Lloyd G., Labor Economics and Labor Relations, sixth edition (Englewood
Cliffs, N.J.: Prentice Hall, 1974).
Rosen, Harvey, "Taxes in a Labor Supply Model with Joint Wage-Hours Determination,"
Econometrica, XLIV (May 1976), 485-508.
Schiller, Bradley, and Randall Weiss, "Pensions and Wages: A Test of the Equalizing
Differences Hypothesis," unpublished manuscript, January 1977.
Smith, R. S., "Compensating Wage Differentials and Hazardous Work," Technical
Analysis Paper No. 5, Office of Evaluation, U. S. Department of Labor, August
1973.
Society of Actuaries, 1967 Occupation Study (Chicago: Society of Actuaries, 1967).
Taubman, Paul, Sources of Inequality in Earnings (New York: American Elsevier
Publishing Company, 1975).
Thaler, Richard, and Sherwin Rosen, "The Value of Saving a Life: Evidence from the
Labor Market," in Nestor E. Terleckyj, ed., Household Production and Consumption
(New York: N.B.E.R., 1975).
Toder, Eric, "The Supply of Public School Teachers to an Urban Metropolitan Area:
A Possible Source of Discrimination in Education," Review of Economics and
Statistics, LIV (Nov. 1972), 439-43.
U. S. Census Bureau, Annual Survey of Manufacturers: 1968 and 1969 (Washington:
U. S. Government Printing Office, 1973).
This content downloaded from 147.251.185.127 on Tue, 17 Mar 2015 13:02:25 UTC
All use subject to JSTOR Terms and Conditions