Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend
access to Journal of Political Economy
POLITICAL ECONOMY
Volume LXVI AUGUST 1958 Number 4
JACOB MINCER
281
Personal income distributions are not come variance increases with time for
normally or symmetrically distributed, each age cohort but, given a stable popu-
but the distribution of logarithms of in- lation, the aggregate variance remains
come is rather symmetric and in a rough unchanged.
way approximates normality. The proc- Unless we assign specific interpreta-
ess of "random shock" just described tions to the "chance" factor, it is difficult
generates a log-normal distribution if ap- to see how the stochastic models increase
plied to the logarithms of income rather our understanding of the processes un-
than to income itself. Thus the proper derlying the formation of personal in-
assumption to be made is that the ran- come distributions. If the "chance" fac-
dom shock consists of relative or per- tor is to be understood as a net effect of
centage, rather than absolute, income all kinds of causes, this approach is an
changes, which are independent of in- admission of defeat in the efforts to gain
come levels. This is Gibrat's "law of insight into systematic factors affecting
proportionate effect." the distribution of income. Moreover, the
Kalecki has pointed out a serious de- operational scope of the stochastic models
fect in Gibrat's approach.6 The model has not kept pace with the increasing
implies that, as time goes by, aggregate empirical knowledge about the multi-
income inequality increases because each dimensional structure of the personal
subsequent random shock adds a term income distribution. With few excep-
to the sum on the right side of the ex- tions,9 the sole purpose of the models is
pression
to rationalize a presumed mathematical
n
form of the aggregate.
Vn= V0+ E Vt.
t=l
From the economist's point of view,
perhaps the most unsatisfactory feature
This, however, is empirically false. of the stochastic models, which they
Subsequent models correct this defect share with most other models of personal
in either of two ways. One is to postulate income distribution, is that they shed no
a negative correlation between the size light on the economics of the distribu-
of the random shock and the level of in- tion process. Non-economic factors un-
come, to be interpreted as a decreasing
doubtedly play an important role in the
likelihood of large negative changes with
distribution of incomes. Yet, unless one
a decreasing level of income.7 This re-
denies the relevance of rational optimiz-
striction assures constancy of the vari-
ing behavior to economic activity in
ance of the distribution. Another way is
general, it is difficult to see how the
to apply the random shock, without re-
factor of individual choice can be disre-
striction, separately to age cohorts
throughout their life-histories.8 The in- garded in analyzing personal income
distribution, which can scarcely be inde-
6 M. Kalecki, "On the Gibrat Distribution," pendent of economic activity.
Econometrica, XIII (April, 1945), 161-70.
The starting point of an economic
7 Ibid.; see also D. G. Champernowne, "A Model
of Income Distribution," Economic Journal, LXIII analysis of personal income distribution
(June, 1953), 318-51. must be an exploration of the implica-
8 R. S. G. Rutherford, "Income Distributions: A tions of the theory of rational choice. In
New Model," Econometrica, XVIII (July, 1955),
425-40. 9Ibid.
a recent article" Friedman has pointed that a year of training reduces earning
out two ways in which individual choice life by exactly one year.l2 If individuals
can affect the personal income distribu- with different amounts of training are to
tion. One, around which Friedman built be compensated for the costs of training,
his model, is related to differences in the present values of life-earnings must
tastes for risk and hence to choices be equalized at the time a choice of occu-
among alternatives differing in the prob- pation is made. If we add a provisional
ability distribution of income they prom- assumption that the flow of income
ise. Friedman has shown that such a receipts is steady during the working life,
model is, no less than the others, capable it is possible to estimate the extent of
of reproducing the more outstanding compensatory income differences due to
features of the aggregative distribution differences in the cost of training.'3
of income. The other, and more familiar, The cost of training depends upon the
implication of rational choice is the length of the training period in two ways.
formation of income differences that are First and foremost is the deferral of
required to compensate for various ad- earnings for the period of training;
vantages and disadvantages attached to second is the cost of educational services
the receipt of the incomes. This prin- and equipment, such as tuition and
ciple, so eloquently stated by Adam books, but not living expenses.
Smith, has become a "commonplace of For simplicity, consider the case in
economics. "I' which expenses for educational services
XWhat follows is an attempt to cast are zero. Let
one important aspect of this compensa- I = length of working life plus length of
tion principle into an operational model training, for all persons = length of
working life of persons without training,
that provides insights into some features
of the aggregative personal income dis- 12 According to a recent study, the average length
of working life in eight broad occupational groups is
tribution and into a number of decom- as follows:
positions of it which recent empirical re- Mean No.
Years in
search has made accessible. The aspect Occupation Labor Force
Professional and technical workers.... 40
chosen concerns differences in training Managers and officials .............. 41
Craftsmen and foremen ............. 44
among members of the labor force. Operatives and kindred workers ...... 45
Clerical and sales workers ........... 47
Non-farm laborers ......... ........ 51
Service workers .................... 52
II. A SIMPLE MODEL
Similar patterns were observed in 1930,
Assume that all individuals have 1950. Commenting on the findings, the
identical abilities and equal opportuni- the study observe: "In general men spend, on the
average, fewer years in what may be termed the
ties to enter any occupation. Occupations better jobs. The three occupations with the shortest
differ, however, in the amount of train- working life are those in which greater training,
education and experience are required. These are
ing they require. Training takes time,
also the jobs which in general afford larger earnings.
and each additional year of it postpones Clearly the men in the better jobs-as measured by
the individual's earnings for another earnings and education-spend a shorter period of
their lives in the working force" (A. J. Jaffe and
year, generally reducing the span of his
R. 0. Carleton, Occupational Mobility in the United
earning life. For convenience, assume States, 1930-1960 [New York: King's Crown Press,
1954], p. 50).
10 M. Friedman, "Choice, Chance, and the Per- 13 With minor exceptions, the procedure is
sonal Distribution of Income," Journal of Political basically a generalization of the one used by Fried-
Economy, LXI (August, 1953), 277-90. man and Kuznets in Income from Independent Pro-
11 Thus termed by J. R. Hicks in The Theory of fessional Practice (New York: National Bureau of
Wages (New York: P. Smith Co., 1941), p. 3. Economic Research, 1945), pp. 142-51.
an= annual earnings of individuals with n Less obvious is the finding that kn, n-d is
years of training,
a positive function of n (d fixed); that is,
Un = the present value of their life-earnings at
start of training,
the relative income differences between,
r = the rate at which future earnings are for example, persons with 10 years and
discounted, 8 years of training are larger than those
t = 0, 1, 2, . .. , I-time, in years, between individuals with 4 and 2 years
d = difference in the amount of training, in of training, respectively. Hence the ratio
years, and
of annual earnings of persons differing
e = base of natural logarithms.
by a fixed amount of schooling (d) is at
Then least as great as
kd, v
0=XrlI 1
er(1-d) -
Vn = a,
the ratio of earnings of persons with d
when the discounting process is discrete.
years of training to those of persons with
And, more conveniently, when the
no training. However, since the change
process is continuous,
in ki, n-d with a change in n is neg-
Vn = an (e-rt) dt = - ( e-rn - e-rl) ligible,14 it can be, for all practical pur-
r
poses, treated as a constant k.
Similarly, the present value of life- This result can be summarized in the
earnings of individuals with (n - d) following statement: Annual earnings
years of training is corresponding to various levels of train-
a~n-d ing differing by the same amount (d)
Vn-d = a- ( e-r(n-d) - e-rl) differ, not by an additive constant, but
r
by a multiplicative factor (k).
The ratio, kn, n-dy of annual earnings of
This important conclusion remains
persons differing by d years of training is
basically unchanged when, in addition to
found by equating Vn Vn= d:
the cost of income postponement, ex-
an e-r(n-d) - e-rl penses of training are taken into account.
Enn-d = = -
an, n dde-r - e-rl The additional cost element naturally
er(l+d-n) - 1
widens the compensatory differences in
er(1-n) 1
earnings particularly at the upper edu-
It is easily seen that kn, n-d is (a) cational levels, where such costs are
larger than unity, (b) a positive function sizable.15
of r, and (c) a negative function of 1. In 14 Assuming the values of r and I to be in a rather
other words, as would be expected, (a) wide neighborhood of 0.04 and 50, respectively.
15 The percentage increase in relative income
people with more training command
differences between persons with (n) and (n - d)
higher annual pay; (b) the difference be- years of training resulting from the introduction of
tween earnings of persons differing by d schooling expenses can be measured by the ratio of
years of training is larger, the higher the annual schooling expenses to annual earnings in
groups with (n - d) years of training.
rate at which future income is dis- For example, if average earnings of high-school
counted, that is, the greater the sacrifice graduates are $4,000 and the annual expenses of a
involved in the act of income postpone- college education are $1,000, then to the compensa-
tory income differences due to the deferral of in-
ment; (c) the difference is larger, the come for 4 years (k4 -1) we must add (1,000/4,000)
shorter the general span of working life, (k4 - 1) to compensate for the cost of tuition. This
since the costs of training must be re- increases the differences by 25 per cent (see my un-
published Ph.D. dissertation, "A Study of Personal
couped over a relatively shorter period. Income Distribution" [Columbia University, 19571.
These conclusions are quite obvious. chap. ii, Note 2).
It is, of course, the purpose of the the factor k increases somewhat with the
model to make the distribution of annual level of training, so that even the loga-
earnings a sole function of the distribu- rithms of income would be slightly
tion of training among members of the skewed.
labor force. It follows from what we have Formally, the existence of positive
just shown that this function is of a very skewness introduced by compensatory
simple form: given the distribution of income differences due to differences in
training, the multiplicative constant k training can be shown, and its extent can
serves as a "conversion factor" which be estimated, in a rather simple way.
translates it into a distribution of earn- Let the quartile deviation in the sym-
ings. In order, therefore, to make state- metric distribution of training be d
ments about the theoretical distribution years, and let Y be the first quartile
of earnings, we must first consider the income, Q1. It follows from the previous
distribution of training within the uni- argument that the median, Q2, equals
verse of this model. Y times kd and the third quartile, Q3,
Under the most stringent assumptions equals Y times k2d.
of identical abilities and equal access to Hence Bowley's measure of skewness16
training, the distribution of occupational is
choice, defined as choice of particular
Sk- _(Q3-Q2)-(Q2-Q1) kd2d-2 kd+ 1
lengths of training, would become a
Q3 -Q1 k2d-1
matter of tastes, specifically those con-
cerning the different activities in the dif- (k E- 1 ) 2 kd - 1
= = - o O
if one group consists of members with OCCUPATION 25-35 35-45 45-55 55-65 TOTAL
Professional and
very similar ages and in another there is managerial ..... 30.6 31.1 23.8 14.5 100.0
Clerical and sales. 35.8 28.0 21.8 14.4 100.0
a wide range of ages, there may be less Craftsmen and
foremen ....... 29.9 30.0 24.1 16.0 100.0
income dispersion in the first group, evenOperatives and
service. 33.8 29.8 21.6 14.8 100.0
though its life-path of earnings may be Non-farm laborers. 32.4 28.2 22. 7 16.7 100.0
members. The income differences cor- For convenience, divide the labor force
responding to those occupational cate- into two broad groups of "occupations,"
gories, however, increase with age. Life- those requiring very little training, char-
patterns of earnings are not parallel; acterized by a practically flat life-pattern
their divergence becomes more pro- of earnings (ABU in Fig. 1), and those
nounced with added years of experience, requiring a considerable amount of train-
so that income dispersion increases as we ing, with a pronounced positive slope of
move from younger to older age groups. life path (CBT in Fig. 1). First, we may
Annual
Earnings
H __ __ __ _
O~~~~~~~~~L /
FIG. 1.-Hypothetical life-paths of earnings in occupations differing in the amount of training they
require.
Both statements can be made stronger note that absolute differences in earnings
by specifying that they apply not only are small within untrained groups (life-
to absolute but also to relative disper- path A B U), but they become pro-
sion. The former follows directly from nounced in groups with higher levels of
the model. If we now include in the as- training (CBT). These differences (abso-
sumption about the slopes of life-paths lute dispersion) may be measured by the
of earnings in various "occupations" the slopes of the paths or by the segments
observation that these slopes are neg- UL and TL, respectively. Income levels
ligible or even negative in occupations are represented by the heights US and
requiring little or no formal training, as TS, respectively. Clearly UL/US <
in many manual jobs, then the two TL/TS. That is, relative dispersion in-
propositions must also apply to relative creases with "occupational rank. " Second,
dispersion. the ratio TS/US increases with age:
The argument underlying these propo- TSjUS > T'S'/U'S'. That is, percent-
sitions can be presented quite simply age differences and hence the relative dis-
with the help of a geometric illustration. persion of earnings increase as we move
from a younger to an older "occupation- marily with respect to level and dis-
mix." persion.""2 He found that there is con-
We may now return to the aggregative siderably more symmetry in component
income distribution and explore the im- income distributions than in the aggre-
plications for the total of the hypothesis gate. For example, the distributions for
about patterns of income in component sets of the three broad occupational
groups. groups of employed males, "blue-collar
First, it is obvious that the addition of workers," "white-collar workers," and
"intra-occupational" differences to the professionals, managers, and proprietors,
"interoccupational" differences increases had less skewness when considered sepa-
aggregate inequality. Moreover, "inter- rately than in the aggregate.
occupational" differences themselves Pigou's hunch about the anatomy of
must increase: The present value of a personal income distributions, even when
life-flow of income of given size is smaller, confirmed by Miller's empirical investi-
the steeper the positive slope of the age- gations, however, cannot explain the
income relation. Hence the equalization phenomenon of aggregative positive
of present values requires larger "inter- skewness. It leaves unanswered the basic
occupational" differences in income than question why a merger of relatively
those derived on the assumption of symmetric distributions should result in a
horizontal income flows in all "occupa- positively skewed aggregate. Clearly,
tions." without further specifications, a merger
Finally, it can be shown that the con- of component symmetric distributions
clusions about "intra-occupational" pat- could very well produce a negatively
terns of income dispersion reinforce the skewed or a symmetric aggregate.
implication of aggregative positive skew- My model provides specifications
ness reached on the basis of "interoccu- which insure that a merger produces
pational" differences alone (in the simple positive skewness in the aggregate. The
model) This finding provides an answer aggregative skewness was already im-
to an important question frequently plied by the simple model. In that form,
raised in discussions of personal income however, income dispersion within occu-
distribution. pations was implicitly assumed to be
Before invoking the distribution of zero. But if its existence is admitted, pat-
property as a decisive explanation of terns of "intra-occupational" dispersion
positive income skewness, Pigou con- might easily affect the aggregative, posi-
sidered the possibility that positive tive skewness previously derived. This
skewness of the income distribution may would be the case, for example, if dis-
arise from merging of a number of persion within less trained groups were
homogeneous, non-skewed subgroups systematically and considerably greater
into a non-homogeneous, positively than that within more highly trained
skewed total.20 Recently, H. P. Miller groups. Geometrically, this would mean
has offered evidence to suggest that "the shortening the right tail of the aggrega-
skewness of income distributions is 21 H. P. Miller, "Elements of Symmetry in the
largely due to merging several symmet- Skewed Income Curve," Journal of the American
Statistical Association (March, 1955), pp. 55-71.
rical distributions which differ pri-
22 For a formal statement and discussion of the
20 Pigou, op. cit., p. 246. Other writers have fol-necessary and sufficient conditions for positive
lowed Pigou along these lines. skewness see Mincer, op. cit., chap. ii, Note 4.
tive income distribution and extending is, however, not so serious as it would
the left tail-a change in the direction of seem. Abstraction from secular trends in
negative skewness. However, this con- income imparts a downward bias to the
tingency is ruled out by my findings slopes of life-paths of income. On the
about component groups. In fact, I have other hand, cyclical and seasonal forces
derived patterns of "intra-occupational" impinging on the economy are largely
dispersion that are exactly the opposite eliminated by the cross-sectional "life-
of those given in the preceding extreme path." In this respect, it fits the theo-
example. This positive relation between retical construct in which these disturb-
income levels and income dispersion in ances are removed by assumption. More-
component groups reinforces the effect of over, the relevant income expectations
intergroup ("interoccupational") skew- are likely to be shaped by the contempo-
ness to produce an even greater positive rary cross-sectional picture.
skewness in the aggregate. Another translation problem that
bears directly on the selection of data
THEORETICAL CONCEPTS AND THEIR
is presented by the concept of training.
EMPIRICAL COUNTERPARTS
It will be recalled that I have subdivided
Ultimately, it is the degree of conform- this concept into "formal training," de-
ity of empirical observations with the fined by the time spent primarily in
conclusions suggested by the model preparation for the job, and informal
that establishes its usefulness. We must training or experience on the job. Given
note, however, that properties of the in- the former, the latter is conveniently
come structure specified by the model measured by age. The identification of
do not in themselves constitute a "pre- experience with age should not create
diction" about the empirical income dis- much difficulty, since the existence of
tribution. The validity of such an inter- central tendencies is to be expected in the
pretation depends on the way in which timing of both training and entry into
theoretical concepts like training, in- the labor force.
come, and life-paths of income are trans- More difficulty arises in measuring
lated into empirically identifiable, meas- "formal training." Years of school com-
urable counterparts. pleted as reported by the census do not,
The translation is necessarily imper- unfortunately, include time spent in
fect, in the sense that an exact empirical vocational, trade, and business schools,
representation of theoretical concepts is not to speak of apprenticeships and vari-
seldom possible, available, or even desir- ous forms of on-the-job training pro-
able. For example, the relevant income grams.23 Moreover, the schooling classi-
differences in a given annual income dis- fication is of limited usefulness, since it is
tribution are those among individuals rarely cross-classified with other relevant
differing in age and not those due to the characteristics of the population.
aging of the same individuals. While the A meaningful, though not easily
theoretical concepts of training and quantifiable, indicator of "formal train-
compensation thus involve a longitudinal ing" is occupational status. We can
view of individual income, they must be think of the set of occupations among
brought to bear on cross-sectional data.
23 U.S. Census of Population 1950, Vol. II: Char-
The discrepancy between the dynamic acteristics of the Population, Part 1: U.S. Summary,
concept and the cross-sectional measure Introduction, pp. 44-46.
which the labor force is divided as con- cate rather clearly that earnings are not
stituting a hierarchy ranging from occu- only higher but also increase more
pations requiring little training up to rapidly with age (or decline more slowly
highly specialized occupations whose after the peak of earnings is reached) in
practice presupposes a great deal of in- the more highly trained groups than in
vestment in human capital. If we can the less trained ones. A statistical study
order occupational groups in such a of the 1939 and 1949 income data pro-
"vertical" way, we can use their ranks as vided by the decennial censuses of popula-
indexes of the amount of formal training. tion reveals that the income differential
Despite the shortcomings of the edu- between young men and those who have
cational classification and the difficulties reached the age of peak income is much
in occupational ranking, I have used greater for college graduates than for
both education and occupation to meas- men with less schooling.24 According to
ure the amount of formal training. the same study, similar differentials are
For defining units of income and in- found when racial groups or sex groups
come recipients, it is clear that earnings are viewed separately. They exist in in-
rather than total incomes and persons comes from all sources as well as in wage
rather than families correspond to the and salary earnings. The same differen-
theoretical concepts. It is also desirable tials persist in incomes of spending units
to restrict the income recipients to per- classified by the schooling of the head.25
sons between the ages of twenty-five When occupation, rather than educa-
and sixty-five years, so as to include all tion, is used as a classificatory principle,
training groups after most have entered the occupational groups must be ranked
the labor force and before a sizable num- with respect to the amount of training
ber have retired. Furthermore, in order they presuppose. With broad occupa-
to avoid variations in income introduced tional groups, the vertical ordering from
by variation in man-hours or weeks of unskilled to highly skilled groups as
work during the year-a factor about shown in census tabulations is reason-
which the model is silent-either earn- ably appropriate for our purposes. By
ings of full-year workers or hourly rates and large, skill is an end-product of
should be studied. training, and the occupational ranks
Unfortunately, data fulfilling all these roughly follow the levels of education
requirements are practically non-exist- and of earnings in the groups.
ent. I was forced, therefore, to use data The positive association between oc-
with varying definitions of income and cupational rank, level of earnings, and
income recipient. The extent to which the amount of age change in earnings
the measures deviate from the require- stands out very clearly in figures of earn-
ments must be kept in mind as possible ings of employed males in the United
sources of discrepancy between theory States.26 Similar differences in occupa-
and fact. tional life-paths of income can be found
LIFE-PATHS OF INCOME
24 H. P. Miller, Income of the American People
Available data on the variation of (New York: John Wiley & Sons, 1955), pp. 65-68.
earnings and incomes by age within 25 Federal Reserve Bulletin, XLI (June, 1955),
broad population groups classified by 615, Supplementary Table 3.
educational and occupational status indi- 26 Miller, op. cit., Table 24, p. 54.
TABLE 1 *
TABLE 2
form flow of earnings during the working tion of earnings is abstracted from (row
life, in row b under the assumption that a), the training factor "explains" about a
the shapes of the 1949 cross-sectional life- third of the existing dispersion and
paths (as shown in Fig. 2) constitute the skewness; when age variation is intro-
expectation which is discounted.29 These duced (row b), the theoretical dispersion
theoretical coefficients were then com- is increased only slightly, but the extent
pared with corresponding coefficients of skewness "accounted for" by the
calculated from the actual distribution theory is increased considerably.
6000
z3000 - 12
0
z
5-7
2000 /
15 25 35 45 55 65
AGE
Source: U.S. Census of Population (1950), Ser. P-E, No. 5-B: Education, Tables 12, 13
1949 (Table 1). From the comparison in A striking demonstration of the in-
Table 2 it appears that when age varia-
crease in income dispersion with age
29 The dashed lines in Fig. 2 were obtained by when the aggregate of spending units is
shifting the upper two solid lines downward to the
partitioned into age groups is provided in
level at which all present values are equalized (at
age fourteen). For purposes of discounting, the low- a recent study based on 1948 data com-
est income path was extrapolated back to age four- piled by the Federal Reserve Board's
teen, the middle one to age nineteen. The use of
annual Survey of Consumer Finances.
income rather than earnings figures may bias the
slopes upward; abstraction from similar trends in The systematic positive relation between
education and earnings may impart a counteracting age and family income inequality is re-
bias. The procedure is clearly to be viewed as grop-
flected in a consistent and pronounced
ing for some orders of magnitude rather than as
estimating in any more rigorous sense of the word. drift of Lorenz curves away from the line
cording to the author, "this relation be- When income recipients are classified
tween the degree of income concentra- by educational background, that is, by
tion and age is one of the most interest- years of schooling as defined in the cen-
ing and perhaps important findings of the sus, the expected increase in income dis-
study." That this relation between in- persion with level of training appears as
come inequality is just as pronounced in shown in Figure 4.32
100
90 Age Groups
25-314
80 35-44
45-54 I
70' 55-64
:60
50
$40
to
05-,
30 a
20
10
0 10 20 30 40 50 60 70 80 90 100
Percentage of' income recipients
Source: U.S. Census Bureau, Current Population Reports, Consumer Income, Ser. P-60, No. 16, Table 3, p. 13
100
70
4) /
5 60 -9
40 -
a. 30-
20
10
C
0 10 20 30 40 50 60 70 30 90 100
Percentage of income recipients
Source: U.S. Census, Current Population Reports, Ser. P-60, No. 3, Table 13, p. 22
FIG. 4.-Education and income inequality, United States, 1946 (urban males, 25-65 years old)
than the other groups, particularly at the the groups under study, the more closely
lower ranges of the distribution. This do the results conform to theoretical
partial discrepancy between theory and predictions.
fact appears in most data that include all Some doubts about the meaningful-
workers regardless of the length of their ness of the findings for occupational
employment during the year. I have groups presented in Figure 5 are raised
shown elsewhere34 that the variation in by the heterogeneity of these broad occu-
man-hours introduced by the inclusion pational groups. In each such group in-
come dispersion might be largely a
33For comparable information about intra-
occupational income dispersion for various dates, product of concealed interoccupational
places, and definitions see ibid., Table 111-8, p. 87. income differences among more detailed
34 Ibid., pp. 66-70, and passim. and homogeneous occupations included
in the broad classification. The problem they separate wage and salary incomes
may be rephrased in the following terms: of full-period36 workers from those of all
Does the hypothesized relation between workers.
occupational rank and intra-occupation- Using income shares of top quintiles
al dispersion, which seems to exist in of workers in each detailed occupation
broad occupational groups, also hold as a measure of intra-group inequality,
true for more detailed occupations? we can examine the existence of a rela-
V4
0 50
040-
30-
20-
10
0 10 20 30 40 50 60 70 8o 90 100
Per cent of incane recipients
Source: U.S. Census Bureau, Current Population Reports, Consumer Income, Ser. P-60, No. 16, Table
FIG. 5.-Wage and salary incomes of employed males, non-farm, United States, 1953 (1
and over).
An attempt to answer such a question tion between such inequality and the
involves breaking down the broad occu- occupational rank of the groups. One
pational groups into more detailed com- way of doing this is to inquire whether
ponent occupations and studying their most detailed occupations in the "top"
income distributions. Fortunately, a broad groups (professional, managerial)
recent census monograph breaks down tend to have larger quintile shares than
the large groups into 118 more detailed occupations in the intermediate groups,
occupations and gives income distribu- and so on. Table 3 indicates a positive
tions and measures of income dispersion answer in terms of median shares and
for each of the component groups.35 The mid-ranges of shares for component oc-
data are suitable for our purposes, since cupations within the broad groups. Thus
35 Miller, op. cit., Table C-5, pp. 193-96. 36 Persons working 50 weeks or more in 1949.
when the shares of top quintiles are ar- ences in levels of earnings in the hier-
rayed in order of increasing size for the archy of occupations are due to differ-
28 subgroups of "craftsmen and fore- ences in training required by them. Thus,
men," the median quintile share is 30.2, within the framework of the model, the
and the 14 central subgroups have quin- statement that the amount of intra-oc-
tile shares running from 28.0 to 31.4 per cupational dispersion is positively re-
cent of the aggregate income of the sub- lated to occupational rank can be re-
group. As is seen, the medians follow placed by an equivalent one, namely,
the occupational rank. However, the that dispersion is positively related to
wide mid-range for clerical workers de- levels of earnings in the occupational
prives the median of that group of much groups.
significance, since it indicates an absence Thus we could, in effect, use average
of central tendency. earnings of groups as indicators of their
TABLE 3*
sense, is a factor that cannot be dis- From the point of view of organization
missed in studies of income inequality of the production process, an industry
within component population groups. is a particular combination of factors of
production. When labor is subdivided by
MIXED COMPONENT GROUPS
occupations differing in training and
When the aggregate income distribu- skill, it can be viewed as a set of distinct
tion is broken down by any criterion factors of production differing in the ex-
other than occupation or age, it yields tent of capital accumulated in them. As
component groups which generally differ with non-human capital, some indus-
in occupational composition and age tries have high "capital ratios," other
distribution. When such differences in low ones, the state of the arts in each
composition are pronounced, the findings industry being what it is. Given full
that the more highly trained'workers are utilization of labor (that is, abstracting
characterized by both higher income from variations due to part-period earn-
levels and greater income inequality can ings), we should expect a greater dis-
be utilized to predict the rank order of persion of earnings in industries with
income inequalities of the component relatively many professional and man-
groups. agerial workers than in those with rela-
Let each component group consist of tively few.
several occupational strata. It can be That the occupational composition,
shown that inequality in a component or what may be called the average level
group is a "weighted" sum of the in- of human capital, is an important factor
equalities within strata, with "weights" in determining the extent of intra-
reflecting the relative sizes and relative industry income dispersion is evident on
income levels of the strata.38 Therefore, several levels of aggregation:
the larger the proportion39 of "top" oc- Table 4 presents data on income in-
cupational strata, such as professionals equality and the occupational composi-
and managers, in a group, the larger the tion within ten major industrial groups.
income dispersion in the group. Geo- The occupational composition is that of
metrically speaking, a greater weight at- 1950 as reported in the census. The in-
tached to upper occupational groups ex- equality measure, the income share of
tends the right-hand tail of the income the top quintile of workers, refers to
distribution of the occupation-mix. wages and salaries in 1949 and total
DISTRIBUTIONS BY INDUSTRY
money incomes in 1953 and 1954. In all
three years inequality is positively as-
When the distribution of wages and
sociated with the proportion of highly
salaries is disaggregated by industrial
trained occupations (professional, tech-
origin, the component distributions of
nical, and managerial): the coefficient of
earnings must, to some extent, reflect dif-
rank correlation is +0.85 in 1949, +0.93
ferences in the occupational compositions
in 1953, and +0.80 in 1954.
of the various industries. Such differ-
At a more detailed level of aggrega-
ences exist, no matter what boundaries
tion,40 the role played by occupational
are imposed on the concept of industry.
40 Miller's study quoted in Table 4 contains
38 For a mathematical formulation see Mincer, further breakdowns of the census wage and salary
op. cit., chap. iv, Notes 1-3. statistics into 91 intermediate industry groups.
39 Up to a point, depending on the number and These are subdivided into 117 more detailed indus-
sizes of parameters distinguished in the group (see tries. The study provides frequency distributions of
ibid.). 1949 wages and salaries as well as corresponding
TABLE 4*
means, medians, quartiles, and income shares of ferences in the training-mix produce pre-
quintiles for each industry. The recipient units are dictable patterns of income inequality.
males, fourteen years of age and over, including
part-period and part-time workers (Miller, op. cit., Roughly speaking, the greater the aver-
Appendix Tables B-i and B-3). age amount of training in the group, the
41 Proportions were computed from Occupational greater the inequality in its income dis-
Characteristics (Census Special Rept. P-E, No. IB),
tribution.
Table 134, p. 290.
Thus earnings of full-year employed
42 These contained over 90 per cent of all male
wage and salary workers. The exclusion of sixteen non-white workers show less inequality
industries from the correlation is the result of a than those of similar white workers;
crude, but conservative, attempt to eliminate part-
period workers from the distributions: industries
earnings of female workers are less dis-
with over 20 per cent of workers receiving less than persed than earnings of male workers,
$1,000 during the year were excluded as being too
when part-period earnings are excluded;
strongly affected by the part-period income vari-
able. These were agriculture, forestry, fisheries, earnings of male full-period workers who
lumber and wood, printing and publishing, food- are heads of families are more unequal
stores, five and ten cent stores, gasoline service
stations, drugstores, eating and drinking places, than those of single men; and, finally, in-
retail florists, private households, hotels and lodg-
ings, dress and shoe repair, theaters and motion 43 For statistical details and sources see Mincer,
pictures, and miscellaneous entertainment. op. cit., pp. 109-27.
the empirical evidence is clearly con- choice of longer training, the compensa-
sistent with all the implications of the tory differences are augmented by a dif-
model about the effects of education, oc- ferential ascribable to ability alone. The
cupation, and age on patterns of per- financial outlays incurred in training in-
sonal income distribution. crease the compensatory differences in a
It is useful, in conclusion, to point out way that magnifies dispersion and par-
some limitations of the present study so ticularly skewness in the aggregative dis-
as to avoid possible misinterpretations. tribution. They may, indeed, increase the
In terms of predictive power, the limita- differences beyond the amount necessary
tions are quite obvious. The model pre- for equalization of present values by
dicts the existence of empirical regulari- sharply restricting supply. Finally, when
ties such as aggregative skewness and incomes rather than earnings are con-
ordinal patterns of dispersion in some sidered, the positive association of prop-
broad classifications of income. It was erty incomes with occupational level and
not adapted to predict absolute or even age magnifies income differences in a way
relative magnitudes of parameters in the which is likely to accentuate the em-
component distributions. It does not tell, pirical regularities implied by the train-
therefore, how much of the observed ing factor alone.
sizes of parameters or of the difference It is thus the consensus of all the fac-
between parameters of relevant compo- tors just listed that produces the em-
nents is accounted for by the factors built pirical results. The purely compensatory
into the model. factor around which the model is built
Regarding the distinction between was chosen for its inherent significance,
compensatory and restrictive income dif- its analytical convenience, and the theo-
ferences, it was shown that, even under
retical interest attaching to the result
perfectly free choice, differences in train-
that even perfect equality of ability and
ing would produce a number of features
opportunity implies neither income
of the frequency distributions of income
equality nor symmetry45 in the income
which are actually observed. This does
distribution.
not mean, however, that these features
are a reflection of such compensatory in- 45 In this connection, recall the statement of
come differences exclusively. Allyn Young to the effect that it is not income in-
equality but skewness that is a symptom of the
Interoccupational differences in in- distortion (in a normative sense) of the income
come levels are not entirely compensa- scheme of society (Allyn Young, "Do the Statistics
tory, once a distribution of "abilities" is of the Concentration of Wealth in the United States
Mean What They Are Commonly Assumed To
introduced. If there is some degree of Mean?" Journal of the American Statistical Associa-
positive association between ability and tion, March, 1917, pp. 471-84).