Professional Documents
Culture Documents
POLITICALECONOMY
VolumeLXVI AUGUST 1958 Number4
be drawn from research into the influ- the "permanent" and the "transitory" components
of income as applied to the analysis of consumption
ence of income distribution on consump- in Milton Friedman, A Theory of the Consumption
tion is that the effects of inequality de- Function (Princeton, N.J.: Princeton University
pend upon its causes.' Thus factors as- Press, 1957).
2 A more detailed review of these theories can be
sociated with observed inequality must
found in H. Staehle, "Ability, Wages, and Income,"
be taken into account before the data Review of Economics and Statistics, XXV (February,
can be put to any use. 1943), 77-87.
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
to apply the random shock, without re- general, it is difficult to see how the
striction, separately to age cohorts factor of individual choice can be disre-
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
XWhatfollows 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
that provides insights into some features workinglife of personswithout training,
12 According to a recent study, the average length
of the aggregative personal income dis-
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, 1940, and
Assume that all individuals have 1950. Commenting on the findings, the authors of
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,
ing they require. Training takes time, education and experience are required. These are
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
year, generally reducing the span of his their lives in the working force" (A. J. Jaffe and
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).
10M. 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-
11Thus 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
the relative income differences between,
start of training,
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=
X rlI
er(1-d) - 1
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
a-
-
ing differing by the same amount (d)
Vn-d = ( 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
found by equating Vn Vn= d: basically unchanged when, in addition to
the cost of income postponement, ex-
an e-r(n-d) - e-rl
Enn-dn = = - penses of training are taken into account.
an, 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).
persion in the amount of training implies tive performance and hence a decline in
that aggregative skewness is greater than earnings, particularly in jobs where
it would be in its absence. In particular, physical effort or motor skill is involved.
even if it were true that abilities are dis-Thus, in general, the "life-cycle" of
tributed in a way which, ceteris paribus, earnings exhibits an inverted U-shaped
implies a symmetric distribution of earn- pattern of growth and decline typical of
ings, positive skewness would appear in many other growth curves.
that distribution as soon as choice of We have already seen that differences
training was admitted into the model. in formal training result in compensatory
Thus Pigou's paradox would persist even differences in levels of earnings as be-
in the absence of the institutional fac- tween different "occupations," the latter
tors that he invoked to explain it. defined in terms of length of formal train-
ing. This compensatory principle must,
EXTENSION OF THE MODEL
of course, also remain valid when life-
Primarily for mathematical conven- paths of earnings are sloped. An im-
ience, I have expressed differences in portant new question arises: What
training in terms of definite time periods specific assumption is to be made about
spent on formal schooling. However, the differences between these slopes in the
process of learning a trade or profession different occupations, since these in turn
does not end with the completion of imply differences in the dispersion of
school. Experience on the job is often the earnings within the occupational groups?
most essential part of the learning Casual observation suggests that pat-
process. terns of age-changes in productive per-
Just as formal training can be meas- formance differ among occupations as
ured by the length of time spent at well as among individuals. The explora-
school, the other part of the training tion of such differences is a well-estab-
process-experience-can be introduced lished subject of study in developmental
into the theoretical model in terms of the psychology. A survey of broad, rather
amount of time spent on the job. When tentative findings in this field indicates
this is done, "intra-occupational" pat- that (a) growth in productive perform-
terns of income variation, previously ance is more pronounced and prolonged
abstracted from, must emerge. By defi- in jobs of higher levels of skill and com-
nition, the amount of formal training is plexity; (b) growth is less pronounced
the same for each member of an occupa- and decline sets in earlier in manual
tion. However, the productive efficiency work than in other pursuits; and (c) the
or quality of performance on the job is a more capable and the more educated
function of formal training plus experi- individuals tend to grow faster and
ence, both measured in time units; hence longer than others in the performance of
it is a function of age. We are thus forced the same task.'8
to relax another assumption, previously These findings suggest that experience
adopted for convenience, namely, that influences productivity more strongly in
earnings are of the same size in each jobs that normally require more training.
period of an individual's earning life. The steeper growth of performance
Clearly, as more skill and experience throughout the span of working life in
are acquired with passage of time, earn- occupations
with higher levels of train-
ings rise. In later years aging often
brings about a deterioration of produc- 18 See Mincer, op. cit., chap. ii,
Note 1.
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
largely due to merging several symmet- Skewed Income Curve," Journal of the American
rical distributions which differ pri- Statistical Association (March, 1955), pp. 55-71.
22For 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.
TABLE 2
ACTUAL AND THEORETICAL COEFFICIENTS OF DISPERSION AND SKEWNESS
ALL "FULL TiME"
WORKERS WORKERS
(1) (2) (3)
Coefficient of dispersion (Theoretical* f(a) 0.43 0.43 0.43
Q9 - Q5 J (b) 0.48 0.48 0.48
Q50 lActualt 1.68 1.45 1.36
Coefficient of skewness fTheoretical* f(a) 0.11 0.11 0.11
(b) 0.22 0.22 0.22
(Q9 - Q50)- (Q50-Q5)
Q95-Q5 lActualt 0.24 0.39 0.39
* The measures of dispersion and skewness are based on differences between the Sth and 95th percen-
tiles in the distribution of training. These measures are more sensitive and of greater interest in our context
than the standard ones based on quartiles. Differences between the 5th and 95th percentiles in the educa-
tional distribution of the male U.S. population amounted to 14-16 years of schooling, according to the
1950 Census (Vol. II, Part I: U.S. Summary, Table 115, p. 236). A more appropriate estimate is that of
11 years obtained from occupational data (see n. 12 above) as the difference between the length of working
life of professional and technical groups and the least-trained laborers, which is interpreted as the number
of years of income postponement. These occupational groups are approximately coextensive with the lowest
and highest deciles of the occupational distribution; hence the 5th and 95th percentiles used in my measures
correspond to their median positions. Since the highest schooling classification in the census data is 16 years
or more, the lowest for our purposes is 5-7 years, and the middle figure is 12. Income figures by age for these
three training levels were shown in Fig. 2. The discount rate was conservatively put at 4 per cent and the
length of working life of unskilled workers at 51 years (cf. n. 12).
t Actual measures are computed from the distribution in Table 1. Unfortunately, that distribution
includes males who worked part-time (though during the full period). This factor tends to impart a down-
ward bias to "actual" skewness. A rough correction for this bias is achieved by eliminating all earnings
below $1,500 (col. 2) and below $2,000 (col. 3), even though this may result in an upward bias.
0
- 12
z3000
0
z
5-7
2000 /
15 25 35 45 55 65
AGE
Source: U.S. Censusof Population (1950), Ser. P-E, No. 5-B: Education,Tables 12, 13
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, CurrentPopulationReports,ConsumerIncome, Ser. P-60, No. 16, Table 3, p. 13
70
0
560
4) /
40 -
a. 30-
20
-9
10
C
0 10 20 30 40 50 60 70 30 90 100
Percentage of income recipients
Source: U.S. Census, CurrentPopulationReports,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
34Ibid., 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-period36workers 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-
1
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, CurrentPopulationReports,ConsumerIncome, Ser. P-60, No. 16, Table 5, p. 16
FIG. 5.-Wage and salary incomes of employed males, non-farm, United States, 1953 (14 years of age
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 Personsworking50 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*
INCOMEINEQUALITYIN DETAILEDOCCUPATIONS AND
OCCUPATIONAL STATUS,U.S., 1949
(Male Full-Year Wage and Salary Workers)t
Professional Clerical Laborers
and and and
Group Managerial Sales Craftsmen Operatives Service
Median quintile share 38.2 35.6 30.2 29.6 28.7
Midrange of quintile
shares........... 35.8-44.2 26.3-42.8 29.1-32.7 28.0-31.4 28.0-30.9
* Source: Miller, op. cit., Table C-5, pp. 193-96.
f Weeks worked, as defined in the 1950 Population Census includes all weeks in 1949 during which any work
was performed. Persons who did any amount of work during each of 50 weeks in 1949 were counted as full-year
workers. The variation in man-hours due to the sizable proportion of part-time workers in many occupations
obscures the relation we are studying. No direct information is available on the extent of part-time work in the
detailed occupations. Frequency distributions of earnings provided in Miller, op. cit., Table C-2, pp. 179-81, were
used to exclude 12 occupations in which more than 10 per cent of workers earned less than $1,000, working 50 weeks
or more in 1949. Such low earnings in groups of male, non-farm, full-period workers can only be a reflection of part-
time work or of income received to a large extent in a non-monetary form. The excluded occupations are: clergy-
men; musicians and music teachers; messengers; newsboys; attendants; private household workers; charmen;
janitors and porters; service workers (n.e.c.); fishermen and oystermen; lumbermen, raftsmen, and woodchoppers;
laborers in wood production; and laborers in wholesale and retail trade.
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.
42 These contained over 90 per cent of all male
Thus earnings of full-year employed
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,
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).