You are on page 1of 31

This PDF is a selection from a published volume from the National

Bureau of Economic Research

Volume Title: Investment in Human Beings

Volume Author/Editor: Universities-National Bureau Committee for


Economic Research

Volume Publisher: The Journal of Political Economy Vol. LXX, No.


5, Part 2 (University of Chicago Press)

Volume ISBN: 0-87014-306-9

Volume URL: http://www.nber.org/books/univ62-3

Conference Date:

Publication Date: October 1962

Chapter Title: On-the-Job Training: Costs, Returns, and Some


Implications

Chapter Author(s): Jacob Mincer

Chapter URL: http://www.nber.org/chapters/c13572

Chapter pages in book: (p. 50 - 79)


ON-THE-JOB TRAINING: COSTS, RETURNS, AND
SOME IMPLICATIONS'
JACOB MINCER
Columbia Universityand National Bureau of Economic Research

INTRODUCTION formalprocessesof learningfromexperi-


IN TIHE contextoftheeconomist'scon- ence. Indeed, historically, skills have
been acquired mainly by experienceon
cern with education as a process of
investmentin manpower,it is impor- the job. The vast schoolingsystemand
tant to be remindedthat formalschool the delayed entryinto the labor forceare
instructionis neitheran exclusive nor a distinctlymodernphenomena.
sufficientmethod of trainingthe labor As historysuggests,it is usefulto view
force. Graduation from some level of the two broad classes oftrainingnot only
schoolingdoes not signifythe completion as a sequence of stages but also as alter-
ofa trainingprocess.It is usually the end natives or substitutes.In many cases, the
of a moregeneraland preparatorystage, same degree of occupational skill can be
and the beginningof a more specialized achieved by "shortening"formalschool-
and oftenprolongedprocess of acquisi- ing and "lengthening"on-the-jobtrain-
tionofoccupationalskill,afterentryinto ing or by the reverse.The degreeof sub-
the labor force.This second stage, train- stitutabilitybetween the two will, of
ing on the job, ranges fromformallyor- course, vary among jobs and over time
ganizedactivitiessuch as apprenticeships with changes in technology.
and other trainingprograms2to the in- When trainingis viewedas a processof
capital formationin people, threemajor
I This workwas stimulatedand made possibleby
empirical questions may be raised for
Gary Becker's fundamentaltheoreticalanalysis of
investmentin human capital. H. G. Lewis contrib- economic analysis. (1) How large is the
uted very thoughtfuland usefulcommentson the allocation of resources to the training
firstversionofthepaper. I am also indebtedforhelp- process?(2) What is the rate ofreturnon
fulcommentsto T. W. Schultz,G. H. Moore, G. P.
Shultz, Z. Griliches,and H. Gilman. Dave O'Neill this formof investment?(3) How useful
providedhighlycompetentresearchassistance. Fi- is knowledgeabout such investmentsin
nancialsupportby the CarnegieCorporationofNew explaining particular features of labor-
York is gratefullyacknowledged.
2 A good sample of a growingliteratureon the
forcebehavior?
subject includesP. H. Douglas, AmericanAppren- Recently flourishingresearchin these
ticeshipand IndustrialEducation (New York: Co- areas provides some tentative answers.3
lumbia UniversityPress, 1921); United States De- T. W. Schultz estimatedthe amount and
partmentof Labor, Bureau of Apprenticeshipand
Training,Apprenticeships Past and Present(Wash- growthof resourcesdevoted by the econ-
ington, 1955); ApprenticeTraining (Washington, omy to formaleducation. G. S. Becker
1956); and EmployeeTrainingin New JerseyIndus-
try(Washington,1960); National Manpower Coun- I G. S. Becker, "Investmentin People" (unpub-
cil, A Policyfor Skilled Manpower(New York: Co- lished manuscript,National Bureau of Economic
lumbia UniversityPress, 1954) and Improvingthe Research, 1961), and his "Underinvestmentin Col-
WorkSkills oftheNation(New York: ColumbiaUni- lege Education?" AmericanEconomicReview,Papers
versityPress, 1955); H. F. Clark, and H. S. Sloan, and Proceedings,May, 1960; T. W. Schultz,"Capi-
Classroomsin theFactories(Rutherford,N.J.: Fair- tal Formation in Education," Journal of Political
leigh DickinsonCollege, 1958); 0. N. Serbein,Edu- Economy,December, 1960, and his "Investmentin
cationalActivitiesof Business (Washington:Ameri- Human Capital," American Economic Review,
can Council on Education, 1961). March, 1961.
50

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 51

estimatedthe rate of returnto training quisition of skill or in improvementof


at higherlevels of education. In his Na- workerproductivity.The concept,there-
tional Bureau of Economic Research fore,includes schoolingand trainingob-
study, now in progress,Becker outlines tained on the job. The latter,under this
the capital-theoreticalapproach to in- definition,is a much broader concept
vestmentin people and shows it to be a than what is conveyed by the common
tool of great analytical power and of usage of the word "on-the-jobtraining."
extensiveempiricalrelevance. It includes formaland informaltraining
My firsttask in this paper is to esti- programsin a job situation, as well as
mate the amount of investmentin on- what is called "learning from experi-
the-job training. The estimates are in- ence.)"
direct, and the concept of on-the-job The method of estimatingthe volume
trainingratherbroad, but I am hopeful of investment in on-the-job training,
that resultsare at least suggestiveof the which is describedin this section,treats
ordersof magnitude involved. The esti- "learningfromexperience"as an invest-
mates and a discussion of their limita- ment in the same sense as are the more
tions are given in the firstsection of the obvious forms of on-the-job training,
paper. In the second sectionI attemptto such as, say, apprenticeshipprograms.
estimaterates of returnon some particu- Put in simple terms,an individual takes
lar formsof on-the-jobtraining,such as a job with an initiallylowerpay than he
apprenticeshipsand medical specializa- could otherwise get because he knows
tion. The resultsare then comparedwith that he will benefitfromthe experience
the rates of returnon investmentwhich gained in the job taken.4 In this sense,
includesboth components:formaleduca- the opportunityto learn fromexperience
tion and on-the-job training.In conse- involvesan investmentcost whichis cap-
quence, some tentative inferencesare tured in the estimationmethod.
formulatedabout the separate compo- While data are much more scarce and
nents. In the finalsection of the paper I the arithmeticis more arduous, calcula-
considersome preliminaryempiricalim- tionofon-the-jobtrainingcosts is guided
plications of my results. In particular, by the same theoreticalprinciples5as the
differentialsin on-the-job training are calculation of schooling costs. Costs of
related to income and employmentdif- schooling consist of direct outlays (pri-
ferentialsamong population subgroups, vate tuitionand public support), and of
classifiedby levels of education, occupa- indirect, "invisible" opportunitycosts,
tion, sex, and race. The observed be- 4This propositionis sometimesquestionedon the
havior patterns seem largely consistent basis of casual observation.Greaterlearningfrom
with the investmenthypothesis under- experienceis characteristicof workerswith greater
motivationand ability, and their earningsat the
lyingthis study, though it was not pos- early stages of the career may in some cases be as
sible in this preliminaryempirical ex- highor higherthan thoseofotherworkers.But such
plorationto controlforall other impor- findingthat people withgreaterabilityhave higher
productivitythan othersat any given stage of ex-
tant factorsat play. periencedoes not negatethe existenceofinvestment
in on-the-jobtraining,thoughit may bias the esti-
I. ESTIMATES OF COSTS OF ON-THE-JOB mation of its magnitude.
TRAINING I The conceptual and mathematicalframework
are developedand statedin Becker's"Investmentin
For the purposeofthispaper, the term Human Capital: A Theoretical Analysis," in this
"training" denotes investment in ac- Supplement.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
52 JACOB MINCER

such as foregone earnings of students gone earnings-the other part being the
resulting from the necessary reduc- differencebetween the actual marginal
tion of their labor-forceactivities while product of the trainee and the larger
at school. Once the direct outlays are amount he could produce if he did not
known,it is possible to inferthe costs of engage in training.Adding firmoutlays
an incrementof schoolingfromcompara- in this case would constitute double
tive data on earningsof two sets of indi- counting.
viduals: students,and people similar to It is likely,however,that some frac-
themwithrespectto previouseducation- tion of firmoutlays is not charged cur-
al attainment,age, sex, ability, except rentlyto the workersbut recouped by
that they are "economically active" in the firmat a later date.7The part of firm
the labor forceand do not engagein addi- outlays whichis not matched by current
tional schooling.In empiricalworkthese reductionsin wages of traineesshould be
conditionsare approximated as well as added to foregoneearnings of workers.
data permit. Unfortunately,it is impossible to esti-
According to the available calcula- mate how large a fractionof firmoutlays
tions,6foregoneearningsconstituteover are costs borne by the firm.Worse yet,
halfof total costs of schoolingand about data on costs of training(whetherborne
75 per cent of the costs borne by stu- by firmsor workers)are not only scarce
dents. Foregoneearningsbulk even more but, in principle,highlyunreliable.Such
in the costs borneby traineeson the job. items as loss of production by experi-
Indeed, nowadays it is difficultto think enced workerswho are helpingthe train-
of any important direct payments by ees or wear and tear of equipmentdo not
trainees,though in the past it was not show up in any entryas direct costs of
uncommonfor apprenticesto pay their training. Rather, they are likely to be
masters for the training.This does not hidden in the wage and depreciation
mean,however,that no directoutlaysare costs. Even if all costs of trainingwere
incurredin the trainingofworkerson the borne by firms,so that they would also
job. Firms do spend sizable sums to pay all the foregoneearningsof workers,
financeapprenticeshipsand other train- onlya fractionof costswould be revealed
ing programs: equipment must be pur- by accounting data. I conclude that an
chased and instructorspaid. These sums attempt to gauge costs of on-the-job
presumablyappear in accounts of firms training in the economy by accounting
as costs of trainingworkers,thoughsuch data of firms,even if they were made
data are rarelyavailable. available, would lead to severe under-
Should all or a part offirmoutlaysbe estimates.
added to the sum offoregoneearningsof On the other hand, working with
workersto arriveat a total figureof costs earnings data of workers to estimate
of on-the-jobtraining,indirect and di- 7Under competitiveconditions,all of the firm's
rect?The answer is no, if all of the firm costs willbe chargedto the workerifthe trainingin-
outlays are currently charged to the creases his futureproductivityin otherfirmsjust as
much as in the firmin which he is training.Some
worker in the form of a reduction in fractionofcosts will not be chargedto the workerif
wages. In this case the worker buys the trainingcontainselementsofspecificity, that is,
trainingservicesfromthe firm.The cost ifit increasesthe worker'sfutureproductivityin the
firmmorethan in otherfirms.For a fullexposition
of the purchaseis simplypart ofhis fore- see Becker, "Investment in Human Capital . . .
6 See referencesin n. 3. op. cit.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 53

their foregoneearningsalso leads to an does not satisfythese requirements.


underestimate,to the extent that some Fortunately,an alternativeprocedure
trainingcosts are borne by firms.The based on Becker's theoreticalanalysis of
calculationreportedbelow is an estimate investment in people9 permits utiliza-
of foregoneearnings of workers,using tion of the comprehensiveincome data
Census incomedata ratherthan firmac- available in the United States Censuses.
countingdata. At least, in termsofpopu- The procedureconsists of a comparison
lation coverage,this is a completecalcu- of two average income streamsof work-
lation of what probably is the more im- ers differingby levels of schooling,such
portant component of on-the-jobtrain- as male college graduates and high-
ing costs. The alternativeprocedure,of school graduates.
using firmdata, is practically ruled out Taking thiscomparisonas an example,
because of the meagersupply of informa- the procedureinvolvesyear-by-yearesti-
tion, aside fromthe serious conceptual mation of training costs which a high-
inadequacies. However, some attempt is school graduate must incur in order to
made to supplement the estimates ob- acquire a college education and the ad-
tained fromworkers'income data with 8 One interestingexceptionis the information
ob-
fragmentary estimatesof firmcosts. tainedfroman analysisofa sampleofmorethanfour
A directcomputationofforegoneearn- hundred heads of households fromthe Consumer
Union Panel, taken in 1959. The respondentswere
ings of workers engaged in on-the-job college-educatedmales who started on their first
trainingwould be possible if data were full-timejob approximatelytwelveyears beforethe
available on their earnings during and surveydate. The correlationbetweeninitialearnings
of these individualswiththeircurrentearningswas
afterthe period of training,and on earn- used to testtheexistenceofinvestmentin on-the-job
ings of a comparisongroup of workers trainingby the predictedeffectson age-earnings
who have the same amount of formal profiles:
Consider Ye, the earningsof any individual at
schoolingand are otherwisesimilarto the time t, as consistingof four additive components:
trainees, but do not receive any on- Yt, average earnings of the group; at, an ability
the-job training.Presumably,the latter component of the individual; ct, the investment
component(a cost if negative,returnif positive);
would have a flatterage-earningsprofile and ut, a randomcomponent.
than the former.That is, traineeswould
initiallyreceivelowerearningsthan those
Yt = Yt+ at+ +
ct -ut

not training,the differencerepresenting For simplicityassume that the componentsare not


correlatedwithone another,and u is not correlated
costs of training.At a later age, earnings over time.Since fYeis the same forall individualsin
of traineeswould rise above earningsof the group,the covariance between earningsin the
the untrained,the difference constituting firstand the twelfthyear is:
a returnon the investment.Unfortunate- Cov ( Ybi Y12) = Cov ( a + c + ul, a12
ly, it is impossible to classify work-
ers empirically into such comparison + C12+u12) = Cov ( a,, a] 2 ) -Cov ( C1, C12)

groups.8 Given the group, say, of all The correlationwas foundto be veryclose to zero.
male collegegraduates,thereis no readi- Since the covariance of the ability factoris surely
positive (and roughlyequal to the variance of the
ly available statistic which would pro- ability componentof earnings),the second covari-
vide informationon differential amounts ance must be negativeand equally sizable. That is,
of on-the-job trainingreceived by sub- the larger (more negative) the initially foregone
earnings(cl), the larger (more positive) the return
groups,and no incomedata are provided twelveyears later (c12).
by such subclassifications.Even the frag- Becker,"Investmentin Human Capital .
mentaryinformationon apprenticeships op. cit.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
54 JACOB MINCER

ditional amount of trainingon the job


which is, on the average, characteristic
of college graduates. Such estimatesare
obtained on the assumptionthat the rate 1
ai
of returnis the same on each year's in- 1t 1(1/1
- ? r) S
vestment whether at school or on the
job.'0 In any given year j after high- a is a correctionfactorforfinitelife,'2it is
school graduation, those who go on to, the lengthof the workinglife.
or have graduated from,college would Figures in Table 1 were computed in
have earnings(Yj) whichequal the earn- this fashion and cumulated over the
ings of high-schoolgraduates (X1) plus workinglife. They constituteestimates
the income earned on differential invest- of training costs: these are schooling
ment in trainingmade since graduation costs before entry into the labor force
fromhigh school, providedno furtherin- and opportunitycosts ofon-the-jobtrain-
vestment in trainingwas incurredbythem ing afterward.The cumulationof annual
duringtheyear j. Costs of (incremental) costs over the workinglifestops at about
trainingin year j are, therefore,meas- fifteento twentyyears after entryinto
ured by the differencebetween Yi and the labor force,since the computedtrain-
Xj augmented by the (foregone)return ing costs decline with age after labor-
on the previous (incremental)costs. force entryand become negligible,fluc-
The procedure and the basic data tuating around zero, around age forty
utilized in it are shown in detail in the (see cols. [4], [5], and [6] in Appendix
Appendix.The firststep in the procedure Tables A5-A7). The decline of training
is to compute the rate of return(r) on with age is consistentwith a priori ex-
the investmentin trainingby which the pectations about investment behavior:
two groupsdiffer.This is done by equat- youngerpeople have a greaterincentive
ing the sum of discounted earningsdif- to invest in themselvesthan older ones,
ferencesto zero, after direct schooling 11Aftera year of additionaltraining,the income

outlays are netted out of earnings. alternativesofthe traineeare betterthan thoseindi-


cated by the age profileX>, whichassumes no addi-
Once the rate of returnis obtained, tional training.
the comparisonof net earningsstreams 12The correctionfactora is nota sufficient
correc-
Yj and Xj permits the followingstep- tion forthe effectivelengthof the workinglife.Use
by-stepcalculation of trainingcosts: let ofthisfactoralone assumesthatall ofa givencohort
surviveto a givenage and have a 100 per centlabor-
j = 1 denote the firstyear of additional forceparticipationrate (afterschooling)to thisage.
training.Then trainingcosts in year 1 A completecorrectionshouldtake intoaccountmor-
are C, = X- YI, the observed income talityratesand the fractionofa cohortwhichis out
of the labor forceat each age. Adjustmentsformor-
In year 2 the costs are C2 =
differential. talityand forlabor-forceparticipationwere not in-
(X2 + rai C1) - Y2, the observedincome corporated in the estimatingprocedure. Neither
have any significant effectson age-incomeprofilesof
differential,augmented by the (fore-
males beforethe age of fifty.The effectson income
gone) return on previous costs.'1 Pro- differentialsare small. Accordingto Becker's work
ceeding sequentially, training costs in themortalityadjustmentresultsin a smallreduction
any year j are of the rate of return,if the same mortalitytable is
used forall educationgroups.The correctionfactor
10This assumptionis later questioned.However, was used in the initial set of calculations,but dis-
the fragmentary evidence in Sec. II below suggests carded in the finalrevision,as it turnedout to be
that the assumptionof equal rates is not unreason- negligible.Leaving out all these "survival" factors
able, whenratesare computedon the sum ofprivate resultsin a small overstatementof costs, as is dis-
and public costs of training. cussed later in the text.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 55

because they can collect the returnsfor schooling enter the labor force at age
a longertime.'3 fourteenand have no foregoneearnings
The age-earningsprofileswhich are while at school; high-schoolgraduates
the basic data used in derivingestimates enterthe labor forceat age eighteen,and
of trainingcosts are presentedin Appen- their foregone earnings during high-
dix Tables At-A4. These are beforetax schoolattendanceare obtainableby com-
incomes of United States males (wage parison with incomes of elementary-
and salary in 1939, income in 1949 and school graduates of the same age; college
TABLE 1
LIFETIME INVESTMENT IN TRAINING PER CAPITA AT SCHOOL AND ON-THE JOB,
UNITED STATES MALES, 1939, 1949, 1958, BY LEVEL OF SCHOOLING
(In Thousands)

CURRENT DOLLARS 1954 DOLLARS*

Marginal Cost Total Cost Marginal Cost Total Cost


EDUCATIONAL LEVEL

School Ob Sum School SOutbe- Sum School OnJtb Sum School OnJtIe- Sum
Job job job job
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)

1939:
College........ 4.9 3.5 8.4 7.7 7.9 15.6 9.4 6.7 16.2 14.7 15.2 29.9
High school.... 2.0 2.4 4.4 2.8 4.4 7.2 3.9 4.6 8.5 5.2 8.5 13.7
Elementary
school....... 8 2.0 2.8 8 2.0 2.8 1.3 3.9 5.2 1.3 3.9 5.2
1949:
College........ 10.2 15.7 25.9 15.9 24.3 40.2 11.5 17.7 29.3 18.0 27.4 45.4
Highlschool. 4.1 4.7 8.8 5. 7 8.6 14.2 4.6 5.3 9.9 6.4 9.7 16.0
Elementary
school....... 1.6 3.9 5.5 1.6 3.9 5.5 1.8 4.4 6.2 1.8 4.4 6.2
1958:
College........ 16.4 22.5 38.9 26.0 30. 7 56. 7 15.3 21.2 36.5 24.1 28.8 52.9
High school.... 7.1 2.9 10.0 9.5 8. 2 17.7 6.6 2. 7 9.3 8.8 7.6 16.4
Elementary
school....... 2.4 5.3 7.7 2.4 5.3 7.7 2.2 4.9 7.1 2.2 4 9 7.1

Source: Appendix Tables A1-A7.


* Deflatedby theBureauofLabor Statistics'ConsumerPriceIndex.

in 1958), classifiedby age and education, studentsgraduate at ages twenty-twoto


and adjusted to approximatethe relevant twenty-three, and estimates of relevant
concepts. The adjustments involve net- income differentialsare constructedin a
ting out direct school costs and correc- similarway.
tions for part-timeemploymentof stu- For each date and education group,
dents during the period of school at- year-by-yearestimatesof marginalcosts
tendance. For these purposes, and in of trainingwere calculated by equation
order to separate school and on-the-job (1). An illustrativecalculation is shown
trainingcosts,the assumptionwas made in Appendix Table A4. Detailed annual
thatpeople withnone up to eightyearsof figuresare shown in Tables A5-A7, col-
13 Becker,"Investmentin Human Capital .... umns (t), (2), and (3). The annual esti-
op. cit. mates of marginal costs are then cumu-

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
56 JACOB MINCER

ilted horizontallyin columns (4), (5), gregate costs correspondingto the total
and (6) of Tables A5-A7, to obtain an- cost classificationsin Table 1, columns
nual total costs of schoolingand of on- (3), (4), and (5).
the-job training.Summingthe figuresin In contrastto Table 1, Table 2 repre-
each column yields, separately, lifetime sents actual opportunity costs in the
total costs of schoolingand of on-the-job economy,notexpectationsofindividuals.
training typical of groups with given The relativesizes of the two components
levels ofschoolingperperson.The results of trainingcosts, formaland on the job,
are presentedin Table 1. are also different in the two tables. This
In reading this table it is important is because the aggregative estimates in
to distinguishbetween the "marginal" Table 2 depend on the age distributionof
and "total" figures.The costs of attend- workerswith given levels of education-
ing high school,shown as marginalcosts al attainment.Secular trendsin popula-
of high-schooleducation,do not measure tion size and in educational attainments
the total costs of schoolingofthe individ- affectthe relevantage distributionsin a
ual up to and includinghigh school. For way which makes the aggregative on-
this purpose the costs of high-schoolat- the-job trainingcosts somewhat smaller
tendance must be added to the costs of in relationto school costs than is true on
elementary-school attendance. Similarly, the per capita basis.
the costs of on-the-jobtrainingof a high- Before proceeding to discussion and
school graduate as obtained by equation interpretationof the findingsone must
(1) are additional costs over and above raise questions about their validity and
the costs of on-the-jobtrainingincurred reliability.A numberof possible sources
by elementary-schoolgraduates. These of bias are easily identified.First, the
marginal costs (col. [2] in Table 1) are estimates of per capita training costs
firstdifferences of the total costs of on- (Table 1) are based on cross-sectionin-
the-job training for graduates of any come profiles.They, therefore,may ap-
particular level of schooling, shown in proximate expectations of an average
column (5) of Table 1. male of a given educational level, pro-
The estimates of on-the-job training vided the differencesbetween his earn-
costs in Table 1 are per capita mag- ings and earningsof males at the next
nitudes approximating the sum of re- lower educational level will change year
sourcesthe average male of a given edu- afteryear in the future,preciselythe way
cational level may be expected to invest they do change in the cross-sectional
in trainingon the job duringhis working comparisonfromone cohortto the next,
life. Estimates of the aggregate invest- one year older. If secular trends are ex-
ment by male workersin the economy pected to tilt both income streams up-
duringa given year are shown in Table ward by the same percentage,the returns
2. They are obtained by multiplyingthe (income differentials at a later stage of
year-by-yearcosts of training,as shown life)are likelyto increasesomewhat,with
in Tables A5-A7 (cols. [4],[5],and [6]), by 14 To obtain estimates of investmentsby all
the numberof workers'4(student enrol- workers,those with "some elementaryschooling,"
nent duringthe period of schooling) in "some highschool," and "some college" have to be
the correspondingage and educational included in the calculation. It was assumed that
theirinvestmentcosts are halfwaybetweeninvest-
group (cols. [7], [8], and [9]). The cross- mentcosts of graduatesat neighboringeducational
productsare then summed to obtain ag- levels. See notes to AppendixTables A5-A7.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-lOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 57

income differentialsat an early stage served cross-sectional age-income pro-


largely unaffected.On this assumption, filesare biased downwardat olderages in
the procedureinvolvesa small underesti- all educational groupsexcept the lowest.
mate of the rate of returnsince differen- The failure to tilt the income streams
tials later in life are heavily discounted. upward leads, as before, to an under-
In turn,this implies an understatement statementof costs, mainly at the upper
of costs, to the extent that costs are, in levels of education.
part, a positive functionof the discount For anotherreason, costs were under-
rate (eq. [11). estimatedalso at the lowerlevels of edu-
TABLE 2
AGGREGATE ANNUAL INVESTMENT IN TRAINING AT SCHOOL AND ON-THE-JOB,
UNITED STATES MALES, 1939, 1949, 1958, BY LEVEL OF SCHOOLING
(In $ Billions)

1939 1949 1958

EDUCATIONAL LEVEL

School Job Total School Job Total School Job Total

CurrentDollars

College. .......... .l1 1.0 2.1 3.8 4.3 8.1 8.7 8.7 17.4
Highschool...... . 1.8 1.4 3.2 3.4 3.8 7.2 8.4 3.8 12.2
Elementary.... .9 .6 1.5 2.1 .9 3.0 4.5 1.0 5.5

All levels....... 3.8 3.0 6.8 9.3 9.0 18.3 21.6 13.5 35.1

1954 Dollars

College........... 2.1 1.9 4.0 4.3 4.7 9.0 8.1 8.1 16.2
Highschool..... .. 3.5 2.7 6.2 3.8 4.2 8.0 7.8 3.5 11.3
Elementary....... .9 1.1 2.8 2.4 1.0 3.4 4.2 .9 5.1

All levels.....7.7.3 5.7 13.0 10.5 9.9 20.4 20.1 12.5 32.6

Source: Appendix Tables A1-A7.

Another bias is introduced by using cation. I compared the earningsstream


the cross-sectionalpatterns as approxi- of elementary-school graduateswiththat
mations for the true earnings streams. of persons with one to four years of
This is themisreportingofyearsofschool- schoolingratherthan with persons with
ing by Census respondents.Accordingto zero schooling.The groupwithno school-
Denison, the olderthe groupin an educa- ing is small, and its compositionso dif-
tion class, the larger the fraction of ferentfromthat ofthe othergroups (it is
persons reportinga level of education heavily weighted with farm workers,
higherthan the one they reportedat the single persons,and non-whites)that its
previous Census."5This means that ob- age-earningsprofilecould not serve as a
bench mark. To the extentthat persons
15E. F. Denison, "A Note on Education, Eco-
nomic Growth,and Gaps in Information,"in this with zero to four years of schooling
Supplement. undergosome on-the-jobtraining,which

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
58 JACOB MINCER

is undoubtedly true, the costs of such The extentto which earningsof more
training have been omitted from my trained persons exceed earnings of less
estimates. trained persons is, therefore,an over-
An opposite bias is impartedby omis- estimateof the returnon training.Part
sion ofthe survivalfactors,as mentioned of the observed return is a return to
previously (n. 12). Lack of adjustment these "ability" factors.But, forthe same
for mortality,for example, means that reasons, the observed data are likely to
earnings differentialsat later ages are underestimatethe costs incurred:ifmore
overstated. Costs are thereforeoveresti- capable high-schoolstudents enter col-
mated,because the rate of returnis over- lege,theirforegoneearningsare probably
estimated, though by a small amount. underestimatedby the observed earn-
The 1949 and 1958 income figuresin- ings of the less capable high-schoolgrad-
clude property income in addition to uates who did not go on to college. It is
labor income,and thistoo tends to widen difficultto say, a priori,how large such
differentials between profilesnoticeably biases may be. But, if a correctionfor
at later ages. This is because ofa positive the "ability" factor involves a decrease
correlationof propertyincome with age in returnand a simultaneousincrease in
and witheducation. The resultis a slight cost via income differentials, it is clear
overestimateof costs by an overestimate that the relative decline in the rate of
of the rate of return. returnmust be larger than the relative
A more serious question is posed by increase in costs.16 Accordingto Becker
the assumption that differencesin in- an adjustmentforclass standingof high-
come streams of the groups compared schoolgraduatesbringsthe rate ofreturn
are attributable to differencesin train- down by about 15 per cent. If costs are
ing. Such an assumptiondisregardsother underestimated,this figuremeasuresthe
factors which may affect shapes and maximumamount of bias, when the di-
levels of age profiles.Biases will arise if mensionof ability which is measured by
these other factorsare not independent class standing is taken into account.
of the classificatorycriteria:forexample, Other factors may account for more.
the higherthe years of schoolingand the Once again, the bias need not be in one
higherthe age, the lower the fractionof direction.To the extentthat the restric-
males who are non-white.Farmers and tive factors under discussion affectre-
farmlaborersare disproportionately dis- turns (earnings differentialsafter the
tributed in the low years of schooling trainingperiod) withoutaffectingincome
and low age classes. Restrictionof esti- differentialsduring the trainingperiod,
mates to non-farmwhites (as in 1939) the rate of return and costs are over-
avoids the distortions,but such data estimated. This is because costs, as we
were not available forall the periods. It computed them, are in part a positive
is clear, however, that, even in data functionof the rate of return.
whichare quite homogeneousby Census Possibly the largest source of down-
criteria, certain selective or restrictive ward bias in the estimationof costs was
factorsare not neutralwithrespectto the already mentioned:the omissionof costs
educational classification: people who 16 The rate of returnis a ratio of returnsto costs,

undertake more training are likely to r = k/c. If only c were increased,with k left the
have higherintelligencequotients,higher same, the relative (per cent) decrease in r would
equal the relative increase in c. But, since k is de-
parental income and education, more creased,therelativedecreasein r is strongerthanthe
motivationand information. relativeincreasein c.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 59

of training which are borne by firms. the declinein aggregateon-the-jobtrain-


These costs do not show up in theincome ing for the elementary-schoolclass was
data at all. As a simpleexample,take the not a resultof a declinein costs per head
case of a firmwhich pays half the costs but a decline in the number of heads.
of training,the otherhalf being paid by Similarly,the increasein on-the-jobcosts
the worker.Later on, the firmcaptures in the aggregateforthe college class also
half of the returns.Rates of returnare consisted mainly in an increase in the
not affected,and foregoneearnings of numberof heads ratherthan in training
workersare cut in half. costs per head, particularlyin the second
It is not possible to arriveat an over- decade.
all notion of the directionof bias with- One featureof the findingsin Table 1
out knowingmoreabout the magnitudes is worthyof closer attention: on-the-job
of each possible error. But, if there is trainingis a larger quantity the higher
some reason to believe that totals are the level of education. This is not a tru-
underestimated,thereare reasons to be- ism as in the case of schooling,wherethe
lieve that the distortionis weaker when marginalquantitiesof schoolingare posi-
it comes to relative sizes of subtotals in tive by definition.There is nothingin the
the classificationsof Tables 1 and 2. If calculation of on-the-job training costs
ability factors bias costs in the com- that would make the marginalquantities
parison of college and high school, they necessarilypositive. In other words,the
have similar effectsin the high school positiveassociationbetweenschooltrain-
and elementaryschool. ing and on-the-jobtrainingis not defini-
The striking finding in Table 1 is tional; it is an empiricalinferencefrom
that the opportunitycosts of on-the-job the observedincomedata. More training
trainingper male are almost withoutex- seems to involve more of both formsof
ception somewhat higher than costs of training,though not in any fixed pro-
a comparable increment of schooling. portion.This is reasonable: school educa-
But while per capita amounts of formal tion is a prerequisite,a basis on whichto
schooling (as measured by costs in con- build the further,more specialized train-
stant dollars) grew between 1939 and ing.
1958 at all levels, the corresponding Some independent evidence on this
quantitiesof on-the-jobtrainingper cap- positive association is provided by re-
ita grewmainlyat the highereducational cent Department of Labor estimates of
levels. amounts of school and on-the-jobtrain-
On an aggregativebasis (Table 2) on- ing,bothmeasuredin school-gradeequiv-
the-jobtrainingcostswerea littlesmaller alents, requiredforthe acquisition of oc-
than schoolingcosts in 1939 and grewat cupational skill in four thousand de-
a slower rate than the former.Formal tailed occupations."7From the fourthou-
education expendituresgrew rapidly at sand occupations listed in the publica-
all levels duringthe 1939-58 period. On-
tion, a sample of 158 occupations was
the-job trainingexpendituresgrew just
selected on the basis of comparability
as fastas schoolingat the highesteduca-
withthe 1950 Census occupationalbreak-
tional level, increased before 1949 and
decreased afterwardat the high-school 17 United States DepartmentofLabor, Bureau of

level, and continuouslydeclined at the EmploymentSecurity,United States Employment


Service,Estimatesof WorkerTrait Requirements fcr
elementary-school level. The per capita 4,000 Jobs as Definedin the Dictionaryof Occupta
figures(Table 1) indicate,however,that tional Titles(Washington,1956).

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
60 JACOB MINCER

down. The two measures of school and number of college-equivalent years of


on-the-job training requirementsgiven on-the-jobtrainingimbedded in the la-
in rank form,were correlatedwith coef- bor force(includingfemales)."8 The esti-
ficient+.86. mate was 1.66 and 1.72 for 1939 and
The positive association between 1949,respectively.But these are average
schooling and on-the-job traininghelps quantitiesforthe whole age distribution,
in understandingtrends.It suggeststhat figuresrepresenting a stock. We are inter-
an expansion of education is likely to ested in the flowof currentinvestmentin
bring about an expansion of on-the-job on-the-jobtraining,and this is incurred
training, a development indicated in mainlyby the youngerage groups.These
Tables 1 and 2. To the extent that an groupshave highereducation levels than
expansion of education is induced by a the labor forceas a whole and are, there-
decrease in its price relativeto the price fore,likelyto investmore also in on-the-
of on-the-jobtraining,some substitution job training.In 1949 the age group 18-29
will take place, and education may grow had a median schoolingof 12 years com-
at the expense of on-the-job training. pared to a labor forcemedian of 10 years.
Such factors,among others,may under- The discrepancy between means was
lie the slowergrowthof on-the-jobtrain- even greater.Since the investmentin on-
ing than of schooling. More precisely, the-job trainingis higherat higheredu-
the data suggestslow or no growthof on- cational levels, an upward adjustment is
the-jobtrainingat the lower educational required. Using the ratio of medians to
levels and pronouncedgrowthat upper revise Eckaus' estimatesupward, rough-
educational levels. This findingsupports ly in proportion,yields 1.99 and 2.06
popular impressionsabout the changing years for 1939 and 1949 respectively.
levels of on-the-jobtraining:a shiftfrom In terms of equivalent college costs
apprenticeshipsto technicians,scientific per year, 2.06 years of training would
personnel, and executive development cost about $6,000 per memberof the la-
programs.Such shiftsmay, in the aggre- bor forcein 1950, accordingto Table 1.
gate, reflectthe upward trendin supplies The average femaleinvestsin on-the-job
of labor with high levels of educational trainingabout one-tenthas much as the
attainmentand possibly some substitu- average male,"9and the number of fe-
tion phenomenaat the lower levels. The males was slightlyover a third of the
questions about trends are very intri- total laborforcein the age group 18-29.
guing, but the data do not lend them- Hence, the implicitcost (C) on-the-job
selves to more than conjectures. trainingincurredper male in 1949 is:
Turning to bodies of data other than
$5,200 =2C?+ --C
the comprehensiveincome statistics, I
tried to exploit them, though not very C= $7,500 .
intensively,fortwo purposes: (1) to pro-
18 R. S. Eckaus, "Education and Economic
vide some empiricalchecks on the relia-
Growth," in Economics of Higher Education, ed.
bility of estimates based on foregone Selma J. Mushkin (Washington:United States De-
incomes of workers, (2) to form some partmentof Health, Education, and Welfare[forth-
guesses about firmcosts or outlays. coming]),Tables 1 and 2. College equivalenceis im-
in UnitedStates DepartmentofLabor, Bureau
1. On the basis of the BLS publication plied
of EmploymentSecurity, United States Employ-
on skill requirementsfor 4,000 occupa- mentService,op. cit.,p. 111.
tions, Eckaus estimated the average 19See Part III below.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
()N-'TJ'E-J( )B TRAININ(G: COSTS, RETURNS, AND) IMPLICATI(O)NS 61

This compares with our estimates of is understandablein view of differences


$8.600 costs of on-the-job training of in age and in educational backgrounds
male high-schoolgraduates (Table 1, col. already acquired.
[5]), the modal group in the population. 2. Several recent surveys of training
A similar calculation for 1939 yields activities in firmshave shown that such
about $3,600 to be compared with our functionsare carried by many firms.20
estimate$4,400. Elementsofsubjectivity Of course, only formallyarranged pro-
in the BLS-derived figures make the grams are described -in such surveys.
comparison difficult,but the fact that Unfortunately, questionsabout costs are
the two sets of estimates are not very seldomraised in these surveys.Undoubt-
farapart is encouraging. edly,it would be difficultto interpretthe
Another piece of supplementaryevi- financialdata, even if they were forth-
dence is provided by data on the distri- coming.In only one of the recentstudies
TABLE 3
GI BILL EXPENDITURES, BY LEVEL AND TYPE OF TRAINING, 1945-55

No. Con GI BILL EXPENDITURES ALL MALES, AGGREGATES

LEVEL OF TRAINING VETERANS


(MILLIONS)
$ Billions Per cent $ Billions Per cent
(1) (2) (3) (4) (5)

College.......... 2.2 5.5 38.1 4.5 40.8


High school ...... 1.4 2.2 15.3 3.3 30.0
Tradeschool..... 2.1 3.3 23.1
On the job ....... 2.1 3.5 24.5 3.2 29.2

Total ......... 7.8 14.5 100.0 11.0 100.0

Source: Cols. (1) and (2), President's Commission on Veterans' Pensions, ReadjustmentBenefits,
Staff Report (No. IX, Part B [Washington: Government Printing Office,September 12, 1956]),
'll 99-2-24 30-3.

bution of federal expenditures on the were such questions asked, with these
GI Bill for 1945-55. The expenditures results:
and theirdistributionare given in Table Althoughquestions were asked concerning
3. In columns (3) and (5) we compare total expendituresfor in-companyeducation,
the percentagedistributionsof expendi- fewfirmsreplied-Perhaps thechiefreasonwas
thatoftenthebooksofthefirmwerenotkeptin
tures:costsofcollegetrainingofveterans a mannerthat would make it easy to separate
duringthe ten-yearperiod are compared educationalcosts fromother costs. Other rea-
withcosts of college of all males in 1949; sonscenteredaroundquestionsofallocationand
a similar comparisonof veterans' costs items to be consideredas costs-The data re-
is made with marginal costs of high ported are not comparable,since some of the
figuresincludesalariesand some excludethem.
school,and withtotal costs of on-the-job It is notcertainthatthefigures reportedinclude
training of high-schoolgraduates. The all in-companyprograms.In one case it was
distributions(col. [3] and col. [5]) look specificallystated that the figurereportedwas
for one program.21
reasonably comparable. The greater se-
20 Clark and Sloan, op. cit., Serbein,op. cit.,and
lectivityof veterans toward college and
the 1960 New JerseySurvey of the Bureau of Ap-
vocational training (trade schools and prenticeshipsand Training.
on the job) in comparisonto all males 21 Serbein,op. cit.,pp. 9-10.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
62 JACOB MINCER

If the scant financialreplies shown in study estimates were made of costs of


this surveyare blownup to an aggregate, labor turnoverto the firm.The concept
the resultis an estimatebelow $1 billion of replacementcost includes hiringcosts
for 1957, undoubtedly a severe under- such as advertising,recruitment,inter-
estimateof even those currentfirmout- views, and separation costs; on-the-job
lays whichare easily identifiable.Smaller trainingcosts are definedmore compre-
case studies indicate that firmexpendi- hensivelyas "the expense broughtabout
tures on formaltrainingprogramsmust by sub-standardproductionof new em-
be much larger: estimates range from ployees while learning their job assign-
$85 for an operative in training22to ments and becoming adjusted to their
over $10,000 for an executive training work environment;the dollar value of
program.23 Accordingto the recentcom- time spent by supervisorsand other em-
prehensivesurvey of New Jerseyindus- ployees who assist in breaking in new
triesmade by the Bureau of Apprentice- employees on theirjob assignment,and
ship and Training,24the proportion of costs of organized trainingprograms."26
workersparticipatingin formaltraining These trainingcosts per workerreplace-
programsin 1959 was about 5 per cent. ment were estimated at about $230. If
Of these 20 per cent enrolledin manage- hiringand separation costs are included
mentdevelopmentprograms,10 per cent the figure doubles. Multiplying these
in apprenticeships,10 per cent in tech- costs of a replacementby the total num-
nical (semiprofessional)training,12 per ber of replacementsin industryin 195827
cent in sales training,and the rest in yieldsan estimateof $7 billion.Inclusion
short programs of operative training, of hiringand separation costs raises the
orientation,safety,etc. Applyingalmost estimate to $14 billion. The assumption
any vaguely reasonable dollar figures- that all of these costs are borne by the
from$85 per operativeto a conservative firmsis, of course, highly questionable.
$2,000 per executive trainee per annum, How much is shiftedback to the trainee
and projecting to the aggregate labor in the formof a wage reduction is not
forcein recentyears, yields an estimate known.At the same time,a large part of
of$2-$3 billion.But this,ofcourse,misses the opportunity cost of workers-the
all costs incurred in informaltraining, differencebetween what they did pro-
which is the typical situation: only 16.2 duce while in training and what they
per cent of firmsin New Jerseyhad for- could produce if they did not train-is
mal trainingprograms. also missed in these figures.
One estimatewhichtakes into account All these heroic attempts to estimate
"invisible" costs of firms,includingcosts firmcosts add up to an uncomfortable
in informaltraining processes, can be range of uncertaintywhen it comes to
obtained using figuresshown in a recent answering the question: how much of
study of Californiafirmsby the Ameri- firmcosts should be added to the esti-
can Management Association.25In this mates of foregoneincomes of workers?
It is possible that billions of dollars are
22
"Training 1\Ianpower,"Fortune, July, 1951.
involved, but it is not clear how many.
23 Clark and Sloan, op. cit.,p. 3.
Besides firm costs. two more items
24
See referencescited in n. 2.
25Merchants and Manufacturers Association, 26bid.

Labor Turnover:Causes, Costs and Methodsof Con- 27 About thirtymillion,using the observedaver-

trol(New York, February,1959). age monthlyreplacementrate of 4 per cent.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 63

must be added to our estimatesin Table other can be approximatedin a residual


2 to get total costs of on-the-jobtraining fashion.What is immediatelyimportant,
in the economy: trainingcosts incurred the larger the differencebetween the
by women and trainingexpendituresin rates of returnon investmentin school-
the Armed Forces. The latter are esti- ing and in on-the-jobtraining,the less
mated at $1.6 billion28in 1959, and the accurate are the cost estimates in the
formerat $1.4 billion29 in 1958.According precedingsection,as well as the various
to Table 2, aggregate opportunitycosts recent estimates of rates of return on
of male workerswere about $13.5 billion (school) education. If the rate of return
in 1958. Additionof the two itemsbrings on schoolingexceeds the rate on on-the-
the figureup to $16.5 billion,more than job training,the estimates are on the
half of the aggregate costs of schooling low side.
(males and females) in 1956.30The ad- It is not obvious, on a priorigrounds,
dition of possibly several billion dollars whetherthe money rate of returnto on-
of costs borne by firmsnarrowsthe dif- the-job trainingis likelyto be smalleror
ferencebut may not close it. Since most largerthan the rate on formaleducation.
of the on-the-job trainingcosts are in- It could be argued that non-pecuniary,
curredby and spent on male workers,it ''consumption" elementsmay be a more
is probably correctto say that, in the importantpart of the real returnto for-
male half of the world,on-the-jobtrain- mal education then to on-the-jobtrain-
ing-measured in dollar costs-is as im- ing. If so, and ifthiswere the only differ-
portantas formalschooling. ence,the moneyrate of returnon school-
ing would appear smaller than the rate
II. ESTIMATES OF RATES OF RETURN
to on-the-jobtraining.Larger public sub-
An estimate of rates of returnto on- sidies to formal education would also
the-job training is both desirable and have this effect,if returnsare computed
difficultto obtain. The rate of return on total costs (privateand public). These
computedby equating the presentvalues argumentsare based on an assumption
of net earningsof two education groups of equality of the real (pecuniary and
should not be interpretedas a rate of re- non-pecuniary)private rate of returnin
turn on schooling costs. The computed both trainingsectors.
rate is some average of rates of returnto One could argue, however,that larger
schoolingand to on-the-jobtraining.The impedimentsto a flowof investmentinto
hybrid rate depends on the weights formaleducationmake forhigherratesof
(costs) of the two trainingcomponents return to schooling than to on-the-job
and on the rates on each component.8'If training.Income constraintsare less se-
the rate on one componentis known,the vere in the latter case as costs are more
spread out over time. Perhaps more im-
28
Includes militaryschools and graining pro-
gramsbut excludesbasic trainingand depreciation portantis that this investmentis under-
of equipment (estimated by R. C. Blitz in "The taken at a later age and in the contextof
Nations Educational Outlay," in Mlushkin(ed.), a concrete,existingworksituation:there
Economicsof HigherEducationt
29 Based on 1949 estimates for female college
is much less uncertaintyabout future
graduates (see Part III, below).
31It also depends on timing.The chronologically
30Accordingto Schultz,the total cost of school- earlier component receives greater weight (see
ingwas $28.7 billionin 1956 ("Investmentin Human Becker, "Investment in Human Capital. . . ," op.
Capital," op. cit.). cit.).

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
64 JACOB MINCER

prospects,about one's own abilities and an average wage Wa, afterwhich he be-
motivations, etc. These circumstances comes a journeyman receivingan aver-
tend to produce a lower real rate of re- age wage Wmn.A suitable alternativeoc-
turnto on-the-jobtrainingand may well cupation,32where almost no trainingis
reduce the moneyrate on it to a lower involved, is the operative,and his aver-
level than the money rate on formal age wage is W0. The annual wage differ-
education. ential d = Wa - Wo is negative during
There are no comprehensivedata com- the training period and positive after-
parable to the Census classificationsby ward, k = W - Wo assumed constant
formal education level from which to for the rest of the workinglife. Under
compute rates to on-the-job training. these assumptions, and disregardinga
The rates shown in Table 4 were esti- negligiblecorrectionforthe finitenessof
mated fora few selected skillsforwhich working life, the rate of return (r) is
easily obtained from:33
TABLE 4
RATES oF RETURN ON APPRENTICESHIP (1+ r)n= 1 + dk (2)
TRAINING, SELECTED TRADES, 1949
wheren is the numberof years of train-
AssUNFPTIONSABOUT ALTERNATIVE ing, or lengthof the apprenticeship.
INCOME STREAMS Estimates in column (1) ofTable 4 are
based on comparisonsof earningsof ap-
TRADES Operatives
Assuming a prentices, journeymen, and operatives
Operatives 10 Per Cent
in Same
with
with Return on in the same industries.While operatives
Industries Highest
Schooling
Additional and correspondingcraftsmen had the
Schooling
(1) (2) (3) same median schooling,the apprentices
had two to threemore years of schooling
Metal....... 16.4 10.4 9.5
Printing.... 16.0 12.6 9.0 than the othertwo groupsin 1949. Thus
Building ..... 18.3 11.3 9.7 k, the differencebetween earnings of
journeymenand operatives,is computed
Source: Table A8.
correctly,holding formal schooling the
same. But foregoneearningsof appren-
tolerablygood data are available. These
tices are underestimated:having more
referto apprenticeshiptraining in the
schooling than the operativeswithwhom
several industriesin whichthey are con-
they are compared, the apprenticescould
centrated. All estimates are for 1949.
earn more in alternative jobs. With re-
The rates of returnon apprenticeship
turns correct and costs underestimated,
trainingwerecomputedin threedifferent
figuresin column (1) are too high.
ways providing a range of estimates,
32 This occupation is more appropriateas an al-
fromthe highestvalues in column (1) to
ternative,in termsof educationalbackground,than
the lowestin column(3) ofTable 4. How- laborers.Clericalworkis an alternative,but it prob-
ever, the lowest values (col. [3]) are ably containsmoreon-the-jobtrainingthan opera-
conceptuallythe soundest.The computa- tive jobs, which involve at most a few monthsof
training.
tionsinvolveequating to zero the present
33 Calculated from
value of differentialsbetween earnings
of workerswho served an apprenticeship
and earnings of their assumed alterna- El(1+ r)i k~ -J7 r)1 +
tive occupational groups. During the The assumptionof infinitelife creates a negligible
periodof trainingthe apprenticereceives error.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 65

In column(2) thisdefectis correctedto titioners.The computation utilizes age-


a large extent. In the calculation, k is income profilesof independentmedical
the same as before,but d was computed specialists,startingwith an initialperiod
froma comparisonof wages of appren- of residency,with the income profileof
tices with wages of operatives whose independentgeneral practitioners,start-
schoolinglevels are closerto levels of ap- ing with the firstyear in practice. Esti-
prentices,regardlessof industryattach- mates of income in money and kind of
ment. As Table A8 shows, however,me- residentswere obtained fromAmerican
dian schoolingof these operativesis still Medical Association sources;"7 earnings
about a year less than of apprentices,so from1950 Census sources."8The calcula-
rates may still be overestimated. tion on before-taxincomes showed a re-
In column(3) the same k is used again, turn of 12.7 per cent. A rough adjust-
but the opportunitycost is computedby ment fortaxes broughtthe rate down to
adding to d (as computedin col. [1]) a re- 11.3 per cent. It is difficultto judge
turnon additional years of (high-school) whetherthis is highor low in comparison
education84by which apprenticesexceed with apprenticeships.89
the operativeswith whom they are com- Table 5 compares estimated rates of
pared in column(1). This bringsthe rates returnon apprenticeshipsand on train-
down to the levels shown in column (3). ing at the college level.
The estimates probably sufferfrom Generalizingboldly, a comparison of
several biases. Operativeshave some on- columns (1) and (2) suggeststhat money
the-job training, but so do craftsmen rates of return(beforetax) on totalcosts
after completion of apprenticeships.If (public and private) are similarforschool
the additional trainingof the latter ex- and on-the-jobtraining.Figures in col-
ceeds that of the former,the rates of
37 Journal of theAmericanMedical Association,
returnon apprenticeshipsare overesti- September 22, 1956, pp. 277 ff.,and October 10,
mated. On the other hand, abstraction 1959, pp. 665 ff.
fromsecular rates of growth,as in the 38
"Income of Physicians," Survey of Current
general case,35 may have the opposite Business,July,1951.
effect.It is also possible that union re- 39The 1950 rate of returnto medical specializa-
strictions on entry to apprenticeships tion may have been above equilibrium.The propor-
tion of specialistsamong physicianswas less than
resultedin higherreturnsin the several half in 1950 and increasedto about two-thirdsby
fieldsselected in Table 4 than in other 1960 (accordingto Medical Economics,1961). If this
kinds of-on-the-jobtraining.36 was a supplyshiftin responseto a highlevel of de-
mand, the rate of returnon specializationshould be
For a comparisonwith another high less today than in 1950. Data frommedical sources
level of skill, I computedrates of return (Physicians Earnings and Expenses, published by
on medical specialization,comparingin- Medical Economics,1961) indicate that in 1959 the
money income differential between specialists and
comes of residentsand specialists (after generalpractitioners is no largerthan it was in 1949,
residency)with incomesof generalprac- despitethe factthat the average incomesof special-
ists rose over 60 per cent duringthe period, resi-
34A 10 per cent rate was used. Higher rates costs
dencieslengthenedsomewhat,and opportunity
would lowerthe figuresin col. (3) even more. clearlyincreased.If the data are reliable,it would
36 See Part I, above. seem that ratesofreturntodayare a fewpercentage
36 However,accordingto a recentstudyby H. G. pointslowerthan in 1949. Incidentally,estimatesof
Lewis, the impactof unionismon wage differentials rates of returnon specializationin medicinehave
was verysmall in the 1945-50 period ("Union Ef- little bearing on the question of alleged monopoly
fectson Relative Wages," in Aspectsof Labor Eco- returnsin medicine.Whateverthe barriersto entry
nomics [National Bureau of Economic Research into medicine,once a medical degreewas obtained,
Conference,1960 (New York, 1960)]). institutionalobstaclesto specializationare weak.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
66 JACOB MINCER

umn 1 are weightedaverages of returns But this is not at all obvious. More de-
on the two sectors; similarityof average tailed data and intensive research are
and componentmeans that rates on each needed.
component are alike. It does appear,
III. ON-THE-JOB TRAINING AS A FACTOR IN
however,that private rates of returnare
INCOME AND EMPLOYMENT BEHAVIOR
lower for the selected instances of on-
the-job trainingthan for total training In the firstsection of this paper, the
at collegelevels. If the selectedinstances economictheory ofinvestmentin people
can be generalized,the rate of returnon was used to bringthe very elusive proc-
collegeeducation per se is somewhatun- ess ofon-the-jobtrainingunderthe meas-
derestimatedby the figuresin column uring rod of money. In this section the
(1). Apparently,the greater ease of in- theory will be used to produce addi-
vestingin on-the-jobtrainingoutweighs tional measurementsand to explain, in
part, certain well-knownbut not well-
TABLE 5 understoodpatterns of income and em-
RETURNSTO "EDUCATION" AND TO ON-THE- ploymentin population subgroups. The
JOB TRAINING, 1950 empiricalanalyses sketchedbelow are no
morethan preliminary,but perhaps they
PER CENT
are sufficiently indicative.
A calculation of (marginal) on-the-job
College On-the-Job trainingcosts per capita for female col-
Level* Trainingt
(1) (2) lege graduatesin 1949 providedtwo esti-
mates: (a) $830, (b) $2,160. The com-
Total costs............... 11 9.0-12.7
Private costs beforetax.... 14 parable figure for males was $15,700
Privatecosts aftertax .1.3.. 8. 5-11 .3 (Table 1, col. [2]). The calculation is the
* Source: G. S. Becker, "Underinvestment in College Edu-
same as the one underlyingTable 1. It
cation?" op. cit. is based on a comparisonof net earnings
t Range based on column 2 and 3 of Table 4, and on return
to medical specialization. of college and high-school graduates,
given in Table A9. Estimate (a) is based
the possibly greater consumption ele- on earningsdata adjusted for(multiplied
ments in college education. Anotherin- by) labor-forcerates of women in the
triguingimplicationis that the apparent, various age groups (Table A9, cols. [3]
but not clearly documented, stability and [4]); estimate (b) is based on the
over time in the rates of returnto train- unadjusted earnings (Table A9, cols. [1l
ing (both in school and on the job) may and [2]). The adjustment for participa-
conceal a decline in the rate of returnto tion rates assumes that the return on
formaleducation,given that investment investmentin trainingof women (at col-
in education seems to have grownfaster lege and on the job) is obtainable only
than in on-the-jobtraining,at least at in the labor market.If it is believed that
the lower levels. this investmentin trainingresults also
These conclusionsare hazardous. The in the same amount of productivityin-
rates are not adjusted forabilityfactors. crease in the "home industry,"earnings
If thereis a greaterselectivity(based on should not be adjusted by labor-force
ability) for admission into college, dif- rates. This certainlycannot be assumed
ferencesbetween adjusted rates in the of investmentson the job, but may be
two sectors may disappear, or reverse. true of schooling.The estimate (b) based

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 67

on unadjusted earnings is, of course, Some direct evidence on scant female


larger. Both assumptions are extreme, participation in on-the-job training is
and, in principle, provide limits for a provided in a recent internationalsur-
correctestimate.40 vey.4' In all countriessurveyed,appren-
While formaleducation costs are not ticeshipsare shorterforwomen than for
much smallerforfemalesthan formales, men. They are half the length of male
investmentsin on-the-job training are apprenticeshipsin the United States in
very small, about one-tenth (taking a bookbindingand in the garmentindus-
middlefigurebetweenthe two estimates) try,where women concentrate.In other
of the amounts invested by males. The industries,numbers of women appren-
figuresmay not be highly reliable, but tices are negligible,perhaps because of
theirsmallnessis quite reasonable,in the physicalrequirementsbut not because of
light of investmenttheory: the average any legal obstacles. It is interestingto
female expects to spend less than half findthat, in contrastto other countries,
her workinglife in the labor force. In applicationsforapprenticeshipsby wom-
particular,she has a high probabilityof en were quite numerous in the early
droppingout of the work forcefor pro- postwar years in Germanyand Austria.
longedperiodsofchild-rearing soon after, By 1949 in these countries,the number
and possiblyduring,the trainingperiod. of skilled women in trades previously
It is clear that returnson prolongedon- considered male was quite pronounced
the-job trainingwould be small. Hence and increasing. Because of the war-
pecuniaryincentivesto invest in on-the- caused imbalance in the sex ratio in the
job trainingleading to higher levels of young age groups,unfavorablemarriage
skill are weak. And even when a girl prospects of young females clearly in-
plans on a career,that is, expects to be creased worker and employer expecta-
permanentlyattached to the labor force, tions of their more permanent attach-
the opportunityforinvestingin on-the- ment to the labor force. Larger invest-
job trainingis likely to be limited. So ment in on-the-jobtrainingbecame eco-
long as thereare some elementsof speci- nomical to both parties. Aside frompa-
ficityin any trainingprogramsor promo- triotism,such motivations may play a
tional schemesof the firm,the employer role in the increased labor-forcerates
will prefermen to women trainees,even and job-trainingof women duringwars
if the latter professoccupational ambi- in all countries.And the willingnessof
tions.This also impliesthat to the extent employersto trainwomenas well as men
that women do obtain specifictraining is enhanced by governmentalsubsidies
they bear a larger fractionof the total of the trainingfunction.
costs of such training than men and, Returningto our estimates: the small
therefore,that the differencebetween amounts of investment in on-the-job
on-the-jobtrainingcosts (includingthose training by females were derived from
borne by employees) for women and female age-income profiles. This pro-
those formen is even largerthan is sug- cedure is, of course, equivalent to a
gestedby our estimate. hypothesiswhich emphasizesthe lack of
40Empiricalevidence on labor-forcebehavior of on-the-jobtrainingas the factorrespon-
marriedwomenis moreconsistentwiththe firstthan sible forboth the observedflatnessof fe-
with the second assumption(see my "Labor Force
Participationof Married Women,"in AspectsofLa- 41 "The Apprenticeshipsof Women and Girls,"
borEconomics,op. cit.). InternationalLabor Review,October,1955.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
68 JACOB MINCER

males' age-incomeprofilesand the small cation ofwomenis morestronglyfocused


differentialbetweenobservedincomesof on the "consumption" sphere, and re-
women of different levels of formaledu- turns are in larger part non-pecuniary
cation. than for males. Hence the apparently
A recentdetailed study of income dif- smallermoney rate of return.
ferentials between males and females In Table 6 a 1949 comparison of
showsthat wage rates approach equality trainingcosts of Negro and white males
when the detailed job specificationis indicates much smaller investmentsin
identical for both sexes.42The rougher on-the-jobtrainingby Negroes, though
the occupationalclassification,the bigger the investmentsare not negligible.The
the wage differentialsat the higherskill investmentin on-the-jobtrainingis also
levels. Lack of on-the-job training fits smaller in relationto investmentin for-
these phenomenaquite well. mal schooling,suggestinga lesser access
TABLE 6
COSTS PER NON-WHITE MALE OF SCHOOL AND ON-THE-JOB
TRAINING, 1949
(In $ Thousands)

TOTAL COSTS
MARGINAL TOTAL COSTS OF ALL UNITED
COSTS STATES MALES
EO(UCATIONAL LEVEL

On the On the On the


School School J bSchool
Joh Joh Job

College.... .. 8.05 3.98 13.20 7.87 15.9 24.3


High school...... 3.92 0.46 5.15 3.89 5.7 8.6
Elementaryschool. . 1. 23 3.43 1.23 3.43 1.6 3.9

Source: Table A10 and Table 1.

These same phenomena,however,are to on-the-job training than to formal


possiblyattributableto differential mar- education. Again, fragmentary direct
ket discriminationagainst women ap- evidence abounds on the small propor-
pearingat the moreskilledjob levels and tions of Negroes in apprenticeshipsand
increasingwith levels of skill. The cal- othertrainingprograms.
culation based on Table A5 indeed re- Conversely, the smaller amounts of
vealed a somewhatlowerrate (about two on-the-jobtrainingreceived by Negroes
percentage points) of return on total than by whites is an interpretationof
training of women than of men. The income differentials:the relative flat-
lower rate may reflect discrimination. ness of theirage-incomeprofilesand the
Anotherexplanationwhich is consistent in earningsby edu-
smaller differentials
withthe investmenthypothesis43 is that,
43 Yet anotherexplanation,suggestedby Becker
in view of the expected smaller rate of ("Underinvestment in College Education?" op. cit.)
participationin the labor market, edu- is that the personalmoneyreturnsshownabove un-
derstatethe moneyreturnswhichactuallyaccrue to
42 H. Sanborn, "Male-Female Income Differen- women as familymembers.Accordingto this argu-

tials" (unpublisheddoctoraldissertation,University ment familyincome differentials are the relevant


of Chicago, 1959). measures.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 69

cation (even when the latter are stand- Oi relatedthe severityofcyclicalchanges


ardized in termsof cost). The lesser on- (1929-33) in employment to levels of
the-job trainingrelative to school train- wages in a particularindustryand found
ing of Negroes is an elementin theiroc- an inverse correlationbetween the two.
cupational distribution. It creates an He also correlatedaverage wages by in-
even lower skill concentrationin the oc- dustrywith turnoverrates fora number
cupational distributionthan would be of industriesat a given time. 1Hereagain
predicted by the educational distribu- the (partial) correlationwas negative. Oi
tion. As in the sex comparison this re- interpretshis results as favorable evi-
sultsin a statisticalfindingthat the ratio dence forthe investmenthypothesis,on
of non-whiteto white incomes declines the assumption that wage levels (by
with increasing level of formal educa- occupation and industry)are a proxyfor
tion.44 amounts of specifictraining.
It has long been observed that at This is a bold assumption. Even if
lowerlevels of skill and education work- cross-sectionalwage differentials (by oc-
ers are affectedby a strongerincidenceof cupation and industry) representedre-
unemploymentthan those at higheroc- turnsto trainingonly,these conceptually
cupational and educational levels. The reflectreturnsto two formsof training:
reasons forthis phenomenonhave never school trainingwhich is "general," and
been clarified. on-the-jobtrainingwhich may be "gen-
In his analysisofinvestmentin people, eral" or "specific." It is not easy to see
Becker points out that, for a given de- why the total returnshould be particu-
mand situation,turnoverand unemploy- larly strongly correlated with what is
ment rates are likelyto be milderunder probably the smallest component: that
conditionsof specificon-the-jobtraining part of on-the-jobtrainingwhich is spe-
than elsewhere. Specific training is de- cific.Oi did not attemptto segregatethe
finedas an investmentwhich increases explanatory factors into "general" and
the worker's marginal product in the "specific" components of training be-
firmin which he is trained more than cause his data did not permitstandardi-
elsewhere.Accordingto this theorymar- zations by education or by age. Without
ginal products of specifically trained such standardizationsthe resultsare am-
workersexceed theirwages, but the lat- biguous. The wage rate reflectsschooling
terare higherthan in alternativeemploy- as well as on-the-jobtraining: a higher
ments.45Hence employershave more in- rate will prevail with very little on-the-
centiveto retainsuch workers,and these job trainingbut sufficiently more school
have more incentiveto remain with the training.This mightobscurethe relation
firm.The differential behavioris implicit which is tested. Conversely,the lack of
both forcross-sectionalobservationsand controlforage makes fora spuriouscor-
forcyclicalchanges. In a recentstudy,a relationbetweenthe wage rate and turn-
similar hypothesis was elaborated and over. Larger proportionsof youngerpeo-
put to an empiricaltest by Walter Oi.46 ple in an industry,or occupation, mean
both more turnover and lower wages.
44 See M. Zeman, "A Quantitative Analysis of
White-Non-white Income Differentials in the
In an attemptto get a strongertest of
United States" (unpublisheddoctoral dissertation, 46
"Labor as a Quasi-fixedFactor of Production"
Universityof Chicago, 1955). (unpublished doctoral dissertation,University of
45 Becker,"Investmentin People," op. cit. Chicago, 1961).

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
70 JACOB MINCER

the investmenthypothesisand more in- on-the-job training in the occupation.


sightinto factorsaffectingturnoverand For given values of X2, larger XI will
unemployment,I ran a multiple regres- thereforetend to reflectmore on-the-job
sion relatinga hybridunemploymentand training.Thus the sign of the partial re-
turnover variable to average full-time gressioncoefficientof XI is expected to
incomesin 1949 of males in eighty-seven be positive. Conversely, for given oc-
detailed occupations, standardizing by cupational wage levels Xi, the higherthe
educationallevel, age, and industrialdis- schoolingX2, the less on-the-jobtraining
tribution.The dependentvariable (y) is in the occupation. Unless formalschool-
the proportionof wage and salary work- ing itselfhas an effecton turnoverand
ers who worked fiftyto fifty-two weeks unemployment,the sign at X2 should be
in 1949. This variable reflectsboth dif- negative. The two additional variables
ferential turnover and unemployment used in the regression,age, X3 and in-
incidenceamong the groups,so it is well dustrialcomposition,X4, standardizefor
suited for the purpose.47 The independ- factorsother than training.Among per-
ent variables are full-timemean incomes sons less than twenty-fiveyears of age
in the occupations (X1), median years of there is more job and labor-forcemo-
schooling (X2), proportion of workers bility than among older people, even
less than twenty-five years old (X3), and when the other variables are held con-
(X)4 proportionof workersemployed in stant.X4 crudelystandardizesforeffects
durable-goodsmanufacturingand in con- of short-rundemand fluctuationsby in-
struction. dustry.
The rationale for the choice of inde- Using these variables, the following
pendent variables is as follows: accord- regressionwas obtained (all variables are
ing to the investment hypothesis, the measuredas deviationsfromtheirmeans;
turnoverplus unemploymentvariable Y standard errorsof regressioncoefficients
is a positive functionof specifictraining are in parentheses):
costs, part of which are borne by work-
y= 2.08X1+ 1.86X2-2.29X3-.74X4
ers, part by firms.Unfortunately,there
are no data or readily available proxies (1.04) (.46) (.68) (.21)
forspecificcosts. I shall assume that such R2 = .65
costs are positively related to the total
of on-the-jobtraining.This is a much All variables are statisticallysignificant.
weaker assumption than that of a posi- All signs,except that forX2, conformto
tive correlationof specifictrainingcosts expectations.In particular,the positive
with wage rates. effectof Xi is consistentwith the invest-
Consider now the average wage X1 in ment hypothesis.
an occupation. This wage will tend to be Even if formaleducation per se had
higher,the higheris the average educa- no effecton employmentstability, the
tion X2 and the greater the amount of effectsof on-the-job training (reflected
in the coefficientat X1) would explain
47 The variable is also affectedby seasonality.

The obvious cases where seasonalityis stronghad the previouslydescribedsystematicpat-


fewerthan 50 per cent of workersemployedyear- ternsof unemploymentrates of workers
round.To avoid arbitrariness,all occupations(more
classifiedby educational levels. As we
than twenty)with y < 50 per cent were excluded
fromthe analysis. have seen in Table 1, more on-the-job

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 71

trainingis receivedby workersat higher serve the purpose. As we have seen (in
educational levels. this section and Table 1) the amounts
However, in terms of the investment investedin on-the-jobtrainingdiffersub-
hypothesis, which emphasizes specific stantially among the groups compared
trainingin this context,the positive sign within the same educational levels. It
at X2 is puzzling. Could it possibly re- also appears that differencesin amounts
verse if the analysis were expanded to of on-the-job trainingincrease with in-
include such variables as urbanization, creasing educational level in both race
unionization,race, marital status? Such and sex comparisons.If on-the-jobtrain-
an expansion, if feasible, would be de- ing were a major factorin explainingdif-
sirable. I experimentedwith inclusionof ferentialsin employmentstability, the
two easily accessible variables: X5, per- investment hypothesis would predict
centageofmales olderthan fifty-five, and higherunemploymentrates for Negroes
X6, percentage of non-whitesin an oc- than forwhitesat each educational level
cupation. Neither was statistically sig- and an increasingdifferentialin rates the
nificant.Their inclusiondid not increase higher the educational level. A similar
the correlationcoefficient, nor did it af- prediction would apply to the female-
fect the coefficientof X2. The inclusion male comparison.
of the racial variable X6, however,low- Data shown in Table 7 are differences
ered the coefficient of Xi and weakened between unemploymentrates of Negro
its reliability. and white males classifiedby age and
Is stabilityof employmentaffectedby education in 1950. Negro unemployment
training,regardlessof whetherit is gen- rates are higherin almost all classifica-
eral or specific,acquired at school or on tions; the differenceis negligibleat the
the job? One could argue,to be monistic, lowest educational levels and, generally,
that educationallevels are more strongly increases with education. The differen-
correlatedwith specifictrainingthan is tials remainpositive,but decrease at the
on-the-job training. For example, the highesteducationallevel.Similarpatterns
employermay be using informationon have been observed by Harry Gilman
educational attainment as an index of for an occupational breakdown of the
capability or suitabilityfor selection to Negro and white male labor force,both
specificon-the-jobtraining.If so, the co- for cross-sectionaldifferencesand cycli-
efficientof education (at X2) "catches" cal changes.48In the occupational break-
more of the effectsof specific train- down,the differentials increasewith skill
ing than does the coefficientat Xi. How- level in the "blue-collar" groups; dif-
ever, theremay be good reasons for the ferentials remain positive but the in-
behaviorofX2 otherthan the investment crease is halted in the "white-collar"
hypothesis,and it remainsan open ques- groups. Additional factors,such as dif-
tion for some significantexploration of ferentialindustrialattachmentsof "blue-
unemploymentphenomena. collar" and "white-collar" groups are
Anotherway of discerningthe effects likely to be responsiblefor some of the
of on-the-job training on employment deviations from the theoretical predic-
stabilityis to comparepopulationgroups tions. A multivariateanalysis is clearly
with the same amount of formaleduca- 48
"Discrimination and the White-Non-white
tion but differing in on-the-jobtraining. Unemployment Differentials"(doctoraldissertation,
Comparisons by race and sex should Universityof Chicago).

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
72 JACOB MINCER

desirable. But, by and large, even the dom, differences (Table 8). The levels are
gross comparisons suggest that the in- similarand decline with increasingedu-
vestment hypothesis is relevant in ex- cation in both groups. Does this mean
plainingdifferencesin unemploymentin- that formaleducation affectsunemploy-
cidence of Negro and white labor.49 ment rates and on-the-jobtrainingdoes
Comparisonof unemploymentrates of not? This would be, prima facie, incon-
males and females, classifiedby educa- sistentwith the other findings.A multi-
tion, show only small, apparently ran- variate analysis is needed in which the
net effectof the trainingfactorcould be
49The turnoverregressionanalysis describedbe- isolated, in orderto resolvethis puzzle.50
foreis also suggestive:once the levels of education
and of on-the-jobtrainingwere taken into account, 50The prevalenceof womenin cyclicallyinsensi-
the racial factor did not seem to have any dis- tive jobs (clerical,government,teaching,and nurs-
cernibleeffectson turnoverplus unemployment. ing) is an obviouslyplausible explanation.

TABLE 7
NEGRO-WHITE UNEMPLOYMENT DIFFERENTIALS,* BY AGE AND EDUCATION,
UNITED STATES MALES, CIVILIAN LABOR FORCE, 1950

AGE
YEARS OF

SCHOOLING TOTAL
25-29 30-34 35-44 45-54 55-64

0 .......... 0.9 .8 -1.1 .3 .6 - .2


1-4........ 0.0 .2 .3 .3 .6 .3
5-7........ .6 1.4 1.5 1.3 1.3 1.8
8.......... 4.4 3.1 3.3 2.3 2.8 3.5
9-11....... 5.8 3.5 4.4 3.3 2.5 4.7
12.......... 5.6 4.7 4.0 2.8 3.9 4.4
13-15....... 4.8 4.9 4.0 .4 3.8 3.8
16 or more... 0.0 3.0 .9 .8 1.7 1.2

* Negro minus white unemployment rate.


Source: U.S. Census of Population, 1950, Special Reports: Education, Table 9.

TABLE 8
MALE-FEMALE UNEMPLOYMENT DIFFERENTIALS,* BY AGE AND EDUCATION,
CIVILIAN LABOR FORCE, 1950

AGE
YEARS OF

SCHOOLING
25-29 30-34 35-44 45-54 55-64

0. 2.6 -1.4 .3 -.4 -1.5


1-4.2.5 2.2 1.4 .4 -1.4
5-7..5 1.2 .3 .1 - .4
8. .3 1.0 .3 -.1 - .5
9-11..8 .7 .5 .2 .3
12. -.1 .7 .1 -.3 - .6
13-15. -.6 .2 .1 -.4 - .6
16 or more -.8 .7 .1 -.2 - .6

* Female minus male unemployment rate.


Source: U.S. Census of Population, 1950, Special Reports,Education, Table 9.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
ON-THE-JOB TRAINING: COSTS, RETURNS, AND IMPLICATIONS 73

SUMMARY vestments in on-the-job training, such


The empirical exploration described as apprenticeshipsand medical speciali-
in this paper was designed to achieve zation, was not different fromthe rate of
several purposes: (1) to estimate the returnon total costs of collegeeducation,
amount of resourcesinvested in on-the- both unadjusted forabilityfactors.How-
job trainingas distinguishedfrominvest- ever, the private return,that is, the re-
ments in the formaleducational system, turn on private costs seems to be higher
(2) to estimate rates of returnon such in formal education than in on-the-job
investments,(3) to investigatethe rele- training.These findingsraise questions
vance of these investments to certain about possible downward biases in the
well-knownbut not well-understoodpat- calculated rates of returnto education.
terns of income and employment be- (3) The last section of the paper is a
havior of population groups. preliminaryanalysis of differentialin-
Since the research was exploratory come and employmentpatternsof popu-
rather than intensive, the conclusions lation groups, classified by education,
reached are very tentative. Brieflystat- occupation, sex, and race. The analyses
ed: (1) Investmentin on-the-jobtraining are incomplete, but they suggest that
is a verylarge componentof total invest- new empirical knowledge about forms
ment in education in the United States and amounts of investmentsin people
economy. Measured in terms of costs, can lead to a significantincrease in our
it is as importantas formaleducation for understandingof such major areas of
economic behavior as income distribu-
the male labor force and amounts to
tion, unemploymentincidence, and la-
more than a half of total (male and fe-
bor mobility.
male) expenditureson school education.
Empirical ventures into unexplored
Aggregate and per capita investments territoryare hazardous. The marginsof
in on-the-jobtraininghave been increas- errorare difficultto assess, and they are
ing since 1939, though at a slower rate likely to be large. At least the findings
than investmentsin formal education. should provoke further research. The
It seems,however,that on-the-jobtrain- need formore,better,and different data
ing has grown at a much fasterrate at is evident. I hope that some guides for
higherskill levels than at lower ones. futureresearchdo emergefromthis pre-
(2) The rate of returnon selected in- liminarywork.

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
APPENDIX
TABLE Al
NET AVERAGE WAGE AND SALARY INCOMES,* BY YEARS OF SCHOOLING
AND AGE, WHITE URBAN MALES, UNITED STATES, 1939
(In Dollars)

YEARS OF SCfOOLIN'G
AGE

16 or More 12 7-8 1-4

Less than 14t.... -850 -850 -850 -340


14-15+.-115 -115 281 258
16-1711..... ...-103 -103 352 315
18-19+.-452 481 443 373
20-21+.-400 755 579 431
22-24 .......... 1,028 947 750 503
25-29......... 1,661 1,244 959 648
30-34 .......... 2,395 1,606 1,179 802
35-44 .......... 3,147 2,073 1,434 916
45-54 .......... 3,483 2,286 1,570 1,018
55-64.......... 3,147 2,105 1,439 950

* All income data are before tax.


t This now shows total rather than annual costs of elementary school per student.
I Gross earnings of high-schooland of college students were assumed to be one quarter of earn-
ings of elementary-schoolgraduates and of high-school graduates, respectively.
Source: Wage and Salary Incomes: Unpublished National Bureau of Economic Research mate-
rials of G. S. Becker, based on 1940 Population Census. Direct costs per student were derived from
Tables 3, 5, and 6 in T. W. Schultz, "Capital Formation by Education," Journal of Political Econ-
omy, December, 1960, and from Biennial Survey of Education in the United States, 1939-40.

TABLE A2
NET AVERAGE INCOMES,* BY YEARS OF SCHOOLING
AND AGE, UNITED STATES MALES, 1949
(In Dollars)

YEARS OF SCHOOLING

16+ 12 8 1-3

Less than 14t. .. -1,576 -1,576 -1,576 -394


14-17+ . . -205 -205 676 670
18-191 . . -91( 1,071 1,079 720
20-2111..... . - 753 1,745 1,523 952
22-24. . 2,284 2,356 1,929 1,192
25-29. . 3,44t1 2,975 2,341 1,474
30-34 . . 4,846 3,576 2,680 1,667
35-44 ..... . 7,085 4,055 3,029 1,814
45-54.. ...... 8,116 4,689 3,247 1,990
55-64 ... .... 7,655 4,548 3,010 1,892

*'ee n e ilnTable Al. Hereincomeincludesproperty income.


t Seen, tin Table Al.
+
t Seen. in Table Al.
Source:Incomedata derivedfrom1950 Censusof Fopulation,Ser. P-E, No. SB, Education,
Tables 12 and 13 (also H. P. Miller,"Incomein Relationto Education,"AmericanEconomicRe-
view,December,1960,Table 1. DirectcostsperstudentderivedfromT. W. Schultz,op. cit.,and
BiennialSurveyofEducation,1948-50.

74

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
TABLE A3
NET AVERAGE INCOMES,* BY YEARS OF SCHOOLING
AND AGE, UNITED STATES MALES, 1958
(In Dollars)

YEARS OF SCHOOLING

A GE- _ _ _ _ _ _ _ - _ _ _ _

16+ 12 8 0-4

Less than 14t** -2,400 -2, 400 -2,400 -600


14-17t . -224 -224 1,208 1,080
18-21t.-682 2,800 1,910 1,532
22-24.3,663 3,537 2,520 1,931
25-29. 5723 4,381 3,223 2,387
30-34.7,889 5,182 3,848 2,757
35-44. 10,106 6,007 4,403 3,023
45-54. 11214 6,295 4,337 3,008
55-64......... 10,966 6,110 3,960 2,956

Source: Income data derived fromthe March, 1959, CurrentPopulation Survey,and Miller, op.
cit. Direct costs per student derived fromStatistical Abstractof the United States, 1960.
* See n. * in Table A2.
t See n. t in Table Al.
$ See n. I in Table Al.

TABLE A4
ILLUSTRATIVE CALCULATION OF ANNUAL INCREMENTAL COSTS OF
INVESTMENT IN SCHOOLING AND IN ON-THE-JOB TRAINING
MALE COLLEGE GRADUATES, 1939
(r = 11.0 Per Cent)*

NetNet
Earnings, Returns onReunoAl
Eringh Net Earnings Differentials Last Year's Previous Costs Costt at
of College in Earnings (j - 1 Age i
Age School Cost
Graduates Graduatest ([]- [2]) (r. Ci-t) r-Ck
=31+851)

(1) (2) (3) (4) (5) (6)

18i... 409 -468 877 . .


......... ........... 877
19.... 563 -437 1,000 96 96 1,096
20.... 717 -407 1,124 121 217 1,341
21.... 793 -391 1,184 148 365 1,549
22. .. 870 870 0 170 535 535
23.. 947 1,028 - 81 59 594 513
24 . 1,021 1,186 -165 56 650 485
5. . . 1,095 1,344 -249 53 703 454
26. 1,169 1,502 -333 50 753 420
27. 1,244 1,661 -417 46 799 382
28 1,316 1,807 -491 42 841 3,50
29 1,388 1,954 -566 39 880 314
30. 1,460 2,101 -641 35 915 274
31. 1,533 2,248 -715 30 945 230
32. 1,606 2,395 - 789 25 970 181
33. 1,668 2,495 -827 20 990 163
34. 1,730 2,595 -865 18 1,008 143
35. 1,792 2,695 -903 16 1,024 121
36. 1,854 2,795 -941 13 1,037 96
37. 1,916 2,895 -979 10 1,017 68
38. 1,978 2,995 -1,017 7 1,054 37
39. 2,041 3,096 -1,055 4 1,058 3

* Obtained by equating to zero the present value of col. (3) (continued to age 65).
t Age-earnings profilesfromTable Al, interpolated within age groups.
t School cost for ages 18-21; on-the-job training cost thereafter.

75

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
TABLE A5*
ESTIMATED COST OF SCHOOLING AND OF ON-THE-JOB TRAINING, BY AGE AND
LEVEL OF EDUCATION, UNITED STATES MALES, 1939

MARGINAL COSTS ($) TOTAL COSTS ($) "EMPLOYMENT" (THOUSANDS)

AGE Elementary High Coll Elementary High College Elemen- Hi-h


School School olege School School ([1I+[21+ tary College
(r= l1.0) School
(r= 20.9) (r= 12.5) (4]=[11) [11+[21) (31) School
(1) (2) (3) (4) (5) (6) (7) (8) (9)

14..... 510 0 0 510 510 510 ......... ......... .........


14 85.... - 388 0 85 388 388 105.7 ......... .........
15 .... 98 455 0 98 455 455 105.7 ......... .........
16...... 110 545 0 110 545 545 193.2 ......... .........
17..... 125 643 0 125 643 643 193.2 ......... .........
18.... 142 254 877 142 396 877 246.1 283.1 .........
19.... 133 200 1,096 133 333 1,096 246.1 283.1 .........
20 ...... 122 139 1,341 122 261 1,341 288.7 349.6 .........
21.... 108 148 1 549 108 256 1 549 288.7 349.6 .........
22 .... 92 158 535 92 250 785 322.8 377.9 164.3
23 .... 71 170 513 71 241 754 322.8 377.9 156.7
24 .... 70 169 485 70 239 724 322.8 377.9 149.9
25 .... 69 168 454 69 237 691 368.7 331.2 131.1
26 ...... 67 167 420 67 234 654 368.7 331.2 129.1
27 .... 65 166 382 65 231 613 368.7 331.2 127.4
28 ...... 65 159 350 65 224 574 368.7 331.2 127.1
29 ...... 66 151 314 66 217 531 368.7 331.2 128.8
30 .... 67 142 274 67 209 483 375.7 256.4 119.7
31 .... 68 131 230 68 199 429 375.7 256.4 112.6
32 ...... 69 118 181 69 187 368 375.7 256.4 108.6
33 ...... 64 105 163 64 169 332 375.7 256.4 104.7
34 .... 58 90 143 58 148 291 375.7 256.4 101.2
35 .... 51 73 121 51 124 245 360.4 167.1 90.0
36 ...... 42 54 96 42 96 192 360.4 167.1 67.2
37 .... 31 37 68 31 67 135 360.4 167.1 67.2
38 ..... 18 14 37 18 32 69 360.4 167.1 67.2
39 ..... 4 ......... 3 4 4 7 360.4 167.1 67.2
.
Total cost of on-the-jobtraining 2,000 4,400 7,900

* Cols. (1), (2), (3) obtained by the method represented by eq. (1) in the text and illustrated in Table A4. Schooling costs are
above the broken lines; on-the-job costs below it. r is the internal rate of returnon the marginal costs. Columns terminate at ages
when costs become zero. Thereafter they turn negative and positive forseveral runs; but they are small, and their sum is negligible.
Cols. (4), (5), (6) are horizontally cumulated costs foreach year of training, separately forschooling (above the broken line), and
for training on the job (below the broken line). Vertical sums (rounded) of training costs in col. (4), (5), (6) are shown in the bot-
tom row. These are entered in col. (5) of text Table 1. Figures in col. (2) of text Table 1 are firstdifferencesof figuresin col. (5),
not vertical sums of col. (1, 2, 3) in Tables A5-A7.
Col. (7) includes male workerswith eight years of education, plus half the workerswith less than eight years and half the workers
with more than eight and less than twelve years of schooling.
Col. (8) includes workers who have high-school education, plus half of the "some high-school" and of "some college" groups.
Col. (9) includes workers who have college education or more, plus half of the "some college" group.
In principle, the employmentfigures(cols. [7], [8], [91) are supposed to representnumbers of workers of a given educational cate-
gory by numbers of years elapsed since completion of schooling, and not by age. Clearly, all college students do not graduate at age
twenty-two.Very few graduate at an earlier age, but large proportions do at later ages. The number of college graduates aged twenty-
two, therefore,severelyunderestimates the number of persons who are in theirfirstyear after college graduation. The bias in numbers
of workers,of course, reverses at later ages. However, since highercosts of on-the-job training decline with age, aggregate costs (Table
2) would be underestimated. This bias is roughlycorrected at the college level (col. [9]) by the use of graduation rather than employ-
ment data. No such correction was made at the lower levels. Graduation at the lower levels cannot be equated with labor-force
participation, and the problem of bias is less acute anyway: age dispersion at graduation and cost figuresare much smaller.
Source: Cols. (7), (8), (9) 1940 Census of Population, Education, Tables 75, 76, 1950 Census of Population, G-E, No. 5B, Educa-
tion, Table 9. Bureau of Labor Statistics, Special Labor Force Reports, No. 1, February, 1960, Table D; United States Department
of Health, Education, and Welfare, Earned Degrees Conferredby Higher Educalional Institutions,1948-58; Biennial Surveyof Educa-
tion, before 1948.

76

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
TABLE A6*
ESTIMATED COSTS OF SCHOOLING AND OF ON-THE-JOB TRAINING, BY AGE AND
LEVEL OF EDUCATION, UNITED STATES MALES, 1949

MARGINAL COSTS ($) TOTAL COSTS ($) "EMPLOYMENT" (THOUSANDS)

Elemen-
tary High Clee Elemen- HihElemen- Higlh
AeHgE School School College Hih College tary
(r= 22.2) (r= 11.8) School School Sho olg
(1) (2) (3) (4) (5) (6) (7) (8) (9)

14.... 1,182 0 0 1,182 1,182 1,182 ......... ......... .........


14 .... 375 777 0 375 777 777 98.3 ..... .. .........
15 .... 382 939 0 382 939 939 98.3 ......... .........
16 .... 377 1,121 0 377 1,121 1,121 184.1 ......... .........
17 .... 401 1,309 0 401 1,309 1,309 184.1 ......... .........
18 .... 316 544 1,881 316 860 1,881 233.1 425.5 .........
19 .... 263 538 2,268 263 801 2,268 233.1 425.5 .........
20 .... 231 441 2,778 231 672 2,778 244.7 415.7 .........
21 .... 202 383 3,304 202 585 3,304 244.7 415.7 .........
22 .... 157 363 1,143 157 520 1,663 285.1 443.7 342.0
23 .... 125 329 1,273 125 454 1,727 285.1 443.7 266.7
24 .... 130 315 1,329 130 445 1,774 285.1 443.7 204.7
25 .... 129 307 1,335 129 436 1,771 303.5 476.8 118.3
26 .... 123 293 1,311 123 416 1,727 303.5 476.8 114.6
27 .... 108 268 1,294 108 376 1,670 303.5 476.8 112.0
28 .... 114 264 1,267 114 378 1,640 303.5 476.8 138.6
29 .... 104 255 1,260 104 359 1,619 303.5 476.8 169.7
30 .... 102 225 1,252 102 327 1,579 329.5 442.1 169.2
31 .... 94 196 1,218 94 290 1,508 329.5 442.1 173.1
32 .... 76 148 1,150 76 224 1,374 329.5 442.1 164.2
33 .... 45 161 1,075 45 206 1,281 329.5 442.1 157.8
34 .... 30 154 1,008 30 184 1,192 329.5 442.1 150.1
35 6.... 1 167 884 16 183 1,067 379.3 387.5 139.5
36.... ......... 151 763 ......... 151 914 . ........ 387.5 125.8
37 . .... ........ . 143 599 ......... 143 742 ......... 387.5 125.8
38. ......... 149 432 ......... 149 581 ......... 387.5 125.8
39. ......... 156 228 ......... 156 384 ......... 387.5 125.8
40. ......... 129 47 ......... 129 176 ......... 267.5 115.8
41. ......... 89 17 ......... 89 106 ......... 267.5 115.8
42 . ........ 65 . . ....... 65 65 . ........ 267.5 115.8
43. ......... 17 .................. 17 17..... 267.5 115.8
Total cost of on-the-jobtraining.. 3,902 8,600 24,300

* See notesto Table AS.

77_

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
TABLE A7*
E'STIMATED COSTS OF SCHOOLING AND OF ON-THE-JOB TRAINING, By AGE AND
LEVEL OF EDUCATION, UNITED STATES MALES, 1958

MARGINAL COSTS ()TOTAL COSTS ($) "EMPLOYMcaNT" (THOUSANDS)

Elementary High Elemen- Elemen-


AGE School
Scool Schoolt
Scool College High Hg
ACE (r= 11. 5) tary School College tary o I
High College
(r =19.3) (r =15. 1) School School Sho
(1) (2) (3) (4) (5) (6) (7) (8) (9)

14.....1,800 0 0 1,800 1,800 1,800 .......... ....


14....1 296 1,266 0 296 1,266 1,266 65.8 .....
15... . ~314 1,538 0 314 1,538 1,538 65.8 .....
16... ~~303 1,917 0 303 1,917 1,917 73.8 ..........
17... ~~300 2,338 0 300 2,338 2,338 73.8 ..........
18... 297 225 3,246 297 522 3,246 191.5 361.4 .....
19.... 293 224 3,776 293 517 3,776 191.5 361.4 .....
20..... 289 223 4,368 289 512 4,368 182.4 432.3 .....
21... . 284 222 5,027 284 506 5,027 182.4 432.3 .....
22 .... 278 220 2,090 278 498 2,588 182.4 432.3 385. 7
23.... 271 217 2,01 271 488 2,489 182.4 432.3 360.0
24....1 262 214 1,902 262 476 2,378 182.4 432.3 335.3
25 .... 251 211 1,891 251 462 2,353 254.5 502.1 285.4
26... 237 208 1,880 237 445 2,325 254.5 502.1 289.0
27.... 221 204 1,660 221 425 2,085 254.5 502.1 304.4
28 .... 202 200 1,528 202 402 1,930 254.5 502.1 332. 7
29 .... 180 195 1,367 180 375 1,752 254.5 502.1 387.3
30 .... 161 189 1,197 161 350 1,547 254.5 502.1 392.2
31 ..... 153 183 1,149 153 336 1,485 254.5 502.1 359.5
32.... 144 175 1,096 144 319 1,415 254.5 502.1 264.2
33 ...... 133 165 1,037 133 298 1,335 254.5 502.1 192.2
34 .... 120 154 971 120 274 1,245 254.5 502.1 125.9
35 .... 104 141 898 104 245 1,143 323.5 501.6 117.1
36 .... 85 126 815 85 211 1,026 323.5 501.6 114.0
37.... 63 109 719 63 172 991 323.5 501.6 140.6
38 .... 37 89 616 37 126 742 323.5 501.6 171.7
39. ._ 6 67 501 6 73 574 323.5 501.6 175.6
40.. .. 43 423 ..... 43 466 . .....501.6 165.0
41 ........ 16 339 ..... 16 355 . .....501.6 165.0
42..... .......... 245 . .
.......... 246 . .....I.....165.0
43 .....144 .....I. . ..... 144 . . .... 165.0
44 ... .... ...... 27 . .......... 27 . ...........165.0
TotalCost of on-the-jobtraining. 5,300 8,200 30,700

*See notes to Table A5.

78

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions
TABLE A8
AVERAGE WAGE AND SALARY INCOME AND MEDIAN YEARS OF SCHOOLING OF APPRENT1CES,
OPERATIVES, AND JOURNEYMEN IN THREE INDUSTRY GROUPS, 1949

METAL TRADES PRINTING AND PUBLISHING CONSTRUCTION


(4 YEARS)* (5.5 YEARS) * (3.8 YEARS)*

Schooling Wage Schooling Wage Schooling Wage

Apprentices.12.2 $2,480 12.2 $2,525 11.8 $2,576


Operatives(in same industry) 9.0 3,015 10.4 3,239 8.8 2,937
Withmoreschoolingt.11.3 3,286 11.3 3,500 11.3 3,208
Assuminga 10 per cent returnon
schooling. ..
............... .
.......... 3,415 .......... 540 ..........
3, 3,340
Journeymen .................... 9.5 3,534 10.9 4,138 8.9 3,216

* Average length of apprenticeship.


t In industries where they are found.
I This returnis added to the wage figurein second row. k = row 5 minus row 2; di = row 2 minus row 1; d2 = row 3 minu
row 1; d3 = row 4 minus row 1.
Source: U.S. Census of Population, 1950: Special Reports,Occupational Characteristics,Tables 10 and 23.

TABLE A9
NET AVERAGE INCOMES OF FEMALES WITH AND TABLE A10
WITHOUT ADJUSTMENT FOR LABOR-FORCE MEAN INCOMES OF NON-WHITE MALES, BY AGE
PARTICIPATION RATES, BY LEVEL OF EDUCA- AND EDUCATION LEVEL, UNITED
TION AND AGE, 1949 STATES, 1950
(In Dollars) (In Dollars)

ADJUSTED* EDUCATION
UNADtJUSTED

AGE High High AGE No Elemen- High colege


AE Hi(hl Col ege School College School- tary High Colle
(1) (2) (3) (4) ing School School or More

18-19 .......... 970 - 786 970 - 786 18-19.570 570 809 809 .......
20-21.......... 1,468 -706 1,468 -706 20-21.......... 808 1 177 1,349 .......
22-24.......... 1,614 1,900 734 1,313 22-24.......... 997 1,520 1,783 1,555
25-29.......... 1,635 2,120 520 939 25-29..........1,109 1,747 2,137 2,121
30-34.......... 1,674 2,293 532 1,016 30-34. 1,187 1,916 2,374 2,950
35-44 .......... 1,859 2,600 662 1,277 35-44 .11. , 300 2,008 2,453 3,437
45-54.......... 2,062 2,907 767 1,608 45-54 ..........1,254 2,068 2,419 3,639
55-64.......... 1,968 2,974 559 1,448 55-64..........1,108 1,921 2,238 3,246

* Observed average incomes multiplied by labor-forcerates Source: Computed from distributions given in U.S. Census
after age twenty-two. Rates from Gertrude Bancroft, The of Population, 1950, Vol. IV, Special Reports, Education, Table
American Labor Force (New York: John Wiley & Son, 1958), 12.
Table D, p. 62.
Source: U.S. Census of Population, 1950, Special Reports,
Education, Tables 10 and 12.

79

This content downloaded from 198.71.7.231 on Tue, 07 Apr 2015 12:59:19 UTC
All use subject to JSTOR Terms and Conditions

You might also like