1. society
2. work
3. unemployment

H. SUMMERS
LAWRENCE
Harvard University
Why
Rate
Is
the
So
Unemployment
Very
High
near
Full
Employment?
IN A WELL-KNOWN PAPER in one of the inauguralissues of the Brookings
Papers, Robert Hall posed the question "Why Is the Unemployment
1 Hall, writingin the context of the
Rate So High at Full Employment?"
3.5 percent unemploymentrate that prevailed in 1969, answered his
questionby explainingthatthe full-employmentratewas so highbecause
of the normalturnoverthat is inevitablein a dynamiceconomy where
some sectors are expandingand others are contractingand because of
the special problemsof certaindisadvantagedgroups. Hall himselfwas
pessimistic about the prospects for maintainingunemploymentconsistently below 4 percentthroughexpansionarypolicies. But he raisedthe
prospectthatsuccessfulstructuralpoliciescoulddo so. Whileaspirations
becameattunedto expectationsas unemploymentrose duringthe 1970s,
the Humphrey-HawkinsFull Employmentand BalancedGrowthAct of
1978nonetheless set an unemploymenttargetof 4 percentfor 1983.
Today,fouryeSwsinto an economicrecovery, the unemploymentrate
hovers around 7 percent. Over the past decade, it has averaged 7.6
I am gratefulto David Cutler and Louise Sheiner for extremely capable research
assistanceandusefuldiscussions.LarryKatzandJimPoterbaprovidedhelpfulcomments
on an earlierdraftof this paper.I have benefitedeven morethanusualfromthe comments
of membersof the BrookingsPanel.The NationalScience Foundationprovidedfinancial
support.A dataappendixis availablefromthe authoron request.
1. RobertE. Hall, "WhyIs the UnemploymentRate So Highat FullEmployment?,"
BPEA, 3:1970, pp. 369-402. The title of this paper is patternedafter Hall's title. The
differencesreflectincreasesin unemploymentover the pastfifteenyearsandsome doubts
aboutjust how close the Americaneconomycurrentlyis to full employment.
339
340
Brookings Papers on Economic Activity, 2:1986
percentandhas neverfallenbelow 5.8 percent.Even most forecaststhat
call for steady growthover the next five years do not foresee unemployment rates dipping back below 6 percent. It is helpful to recall that
unemploymentpeaked at 7.2 percentduringthe relativelysevere recession of 1958.Whilesome of the differencebetween recentandpast levels
of unemploymenthas resulted from cyclical developments, it is clear
thata substantialincreasein the normalor naturalrateof unemployment
has taken place. Where Kennedy-Johnsoneconomists set 4 percent as
an interimfull-employmenttarget, contemporarypolicymakerswould
regardeven the temporaryachievementof 6 percent unemploymentas
a greatsuccess.
This paper describes and explains the substantialrecent increase in
normalunemployment.The firstpartof the paperassesses the relationship between unemployment and other indicators of labor market
tightnessand describes changes in the compositionof the unemployed
population.The data reveal that the level of unemploymentconsistent
with any given level of vacancies, capacity utilization, or change in
inflationhas increased significantlyover time. It appears that little of
this movementcan be tracedto measurementdifficultiesin the Bureau
of Labor Statistics' CurrentPopulationSurvey. Rather, increases in
unemploymentare a serious problem because they are concentrated
among mature men, job losers, and the long-term unemployed. The
portrayalof risingunemploymentas the consequence of an increase in
the shareof secondaryworkersin the laborforce thatwas popularduring
the 1970sis no longeraccurate.
The second and more speculative part of the paper draws on the
dramaticvariations in state and regional economic performancethat
have taken place over the past fifteen years in an effort to get at the
causes of rising unemployment.It links observed increases in unemployment with structuralchanges in the economy that have lowered
employmentin high-wagesectors and increasedit in low-wage sectors.
The structuralchangesincludeboth macroeconomicdevelopmentsthat
have reduced the demand for the output of high-wageindustries and
labor market pressures, particularlyin unionized sectors, that have
pushed up wages in sectors where they were alreadyhigh. I conclude
that reversing the dramaticsectoral shocks of the last few years can
makean importantcontributionto reducingunemployment.
Lawrence H. Summers
341
Increasing Unemploymentin the United States
Figure 1 depicts the evolution of the total U.S. unemploymentrate
andthe unemploymentratefor marriedmen since the KoreanWar.The
fairly steady increasein both rates is interruptedonly duringthe 1960s.
While the amplitude of fluctuations in the unemployment rate has
increased, it is also clear that the normallevel of unemploymenthas
risen. A conspicuousfeatureof the datais that the ratefor marriedmen
has increasedin tandemwith the overallrate.
Table 1 presents informationon the unemploymentrate and the
employment ratio (the ratio of employed adults to the total adult
population)over the seven business cycles since 1953.Again, a secular
increasein unemploymentis evident. Withthe exception of the boom of
the 1960sand the early 1970s, each cycle has higherpeak, trough, and
averageunemploymentrates than the one that precededit. Indeed, the
unemploymentrate at the last peak, in July of 1981, was comparable
withthe ratesreachedat mostpreviouscyclical troughs.Unemployment
at the next peak is not knowable. But most forecasts do not call for
substantial declines. The most recent Congressional Budget Office
forecast,whichassumesfairlysteadygrowthuninterruptedby recession
for the next five years, calls for unemploymentto decline only to 6.0
percentby 1991.2
The secular increase in unemploymentcontrasts sharply with the
behavior of the employmentratio, which has trendedupwardsand is
today close to its historicalpeak. The rise reflects the rapidincrease in
the laborforce participationof adultwomen, from 37.6 percentin 1960
to 43.3 percentin 1970,51.3 percent in 1980, and 54.7 percent in 1985.
Since 1973, both unemploymentand employment have grown quite
rapidly.Total employmentincreased by 25 percent between 1973and
1985.
Unemploymentcan increasefor eithercyclical or structuralreasons.
Before I turn to an analysis of changes in the structureof the labor
market,it is necessary to address the possibility that rising unemploy2. Congressional Budget Office, The Economic and Budget Outlook: An Update
(GovernmentPrintingOffice,August1986).
342
Brookings Papers on Economic Activity, 2:1986
Figure 1. Aggregate and Married Male Unemployment Rates, 1955-85
Percent
10
8
6
/
\
Married males~~
'I
1960
1965
1970
1975
1980
1985
Year
Sources: U.S. Department of Labor, Bureau of Labor Statistics, Handbook of Labor Statistics (June 1986), table
25.
mentis merelya by-productof weakness in aggregatedemandin recent
years. A natural way to test that possibility is to examine how the
unemploymentrate has moved relative to other measures of cyclical
conditionsand to considerwhetherthe relationshipbetween unemployment and inflationhas changed.
Table 2 examines the trendin unemploymentrelativeto two cyclical
indicators-vacancies and capacityutilization.The resultsindicatethat
at any given level of vacancies or of capacity utilization,both overall
unemploymentand the unemploymentof middle-agedmen have in-
Lawrence H. Summers
343
Table 1. Increasing Unemployment over the Business Cycle, July 1953-July 1986a
Percent
Peak
Cycle (peak to peak)
Unemployment
rate
Jul. 1953-Jul.1957
Aug. 1957-Mar.1960
Apr. 1960-Nov. 1969
Dec. 1969-Oct. 1973
Nov. 1973-Dec. 1979
Jan. 1980-June1981
Jul. 1981-Jul.1986
2.5
4.0
5.1
3.4
4.8
6.2
7.1c
Trough
Employmentpopulation
ratiob
59.5
58.7
58.0
59.8
59.4
60.9
60.1c
Unemployment
rate
5.7
7.2
6.7
5.7
8.4
7.7
10.6
Average
Employmentpopulation
ratiob
Unemployment
rate
Employmentpopulation
ratiob
57.3
56.9
57.1
58.4
57.1
59.8
58.2
4.3
5.7
4.7
5.2
6.6
7.1
8.2
58.6
57.5
58.0
58.5
59.0
60.2
60.0
Source: U.S. Department of Labor, Bureau of Labor Statistics, Employment and Earnings, various issues.
a. Peak unemployment is the rate at the beginning of the cycle. Trough unemployment is measured at the cyclical
trough. The dating of the cycles is based on National Bureau of Economic Research chronology.
b. Employment is total civilian employment as measured from the Current Population Survey. Population is the
civilian noninstitutional population over sixteen years of age.
c. The cyclical peak has not yet been observed in this expansion.
creasedsharply.3 Overthepasttwo decadestotalunemploymentappears
to have increased by between 0.13 and 0.17 percentagepoint per year
after adjustmentfor changes in cyclical conditions. Similarincreases
from a much lower base show up in the unemploymentrate of mature
men. Ofthe roughly3.5 percentagepointincreasein totalunemployment
between its 1969 low point of 3.5 percent and the present rate, the
equations indicate that between 2.5 and 2.9 percentage points are
attributableto structuralfactors capturedby the time trend, with the
relativelysmallremainderbeing attributableto changes in the capacity
utilizationand vacancies cyclical indicators.
Estimation of Phillips curves for various periods does not yield
sufficientlyprecise estimates of the naturalrate of unemploymentto
permitdefinitivestatementsaboutits evolution. But the datado suggest
that, particularlywhen the maturemale unemploymentrate is used, the
naturalrate has increased. A calculationis instructive. Between 1965
3. See James L. Medoff, "U.S. Labor Markets: Imbalance, Wage Growth, and
Productivityin the 1970s,"BPEA,1:1983,pp. 87-120, for an earliertreatmentof changes
in therelationshipbetweenunemploymentandvacancies.As KatharineAbraham,"What
Does the Help-WantedIndex Measure?" (Brookings, 1986), emphasizes, measuring
vacanciesis not an easy problem.The index used here is the ConferenceBoard HelpWantedIndex adjusted as suggested by Abrahamfor changes in competitionin the
newspaperindustryandthe occupationalcompositionof the laborforce.
344
Brookings Papers on Economic Activity, 2:1986
Table 2. Corrected Unemployment Trends, Various Periods, 1955-85a
Percentage points
Unemployment rate
Sample
period
Control
Total
Males,
aged
35-44
1955-85
Vacancy rate
1967-85
Vacancy rate
1967-85
Capacity utilization
0.133
(0.009)
0.167
(0.030)
0.147
(0.025)
0.079
(0.011)
0.157
(0.028)
0.159
(0.020)
Source: Author's calculations based on data from BLS, Employment and Earnings, various issues.
a. Regressions of the unemployment rate on a constant, several controls, and a time trend. The controls indicated
are the contemporaneous index (vacancy rates or capacity utilization) and the first two lags of the series. Vacancy
rates are derived from the Conference Board index of help-wanted advertisements with adjustments made as suggested
in Katharine Abraham, "What Does the Help-Wanted Index Measure?" (Brookings, 1986). Standard errors are in
parentheses.
and 1974,the inflationrateas measuredusingthe GNP deflatorincreased
by 6.4 percentage points, while the unemploymentrate of men aged
thirty-fiveto forty-fouraveraged2.2 percent. Between 1980and 1985,
the inflation rate declined by 5.7 percentage points while the same
unemploymentrate averaged5.5 percent. If one assumes, in line with
the data, that a reduction of 1 percentage point in the inflation rate
requires1.5 percentagepoint years of extra unemployment,the implied
naturalrate is 3.2 percent for the earlierperiod and 4.0 percent for the
later one, an increase of one-fourth.To the extent that supply shocks
were partiallyresponsible for changes in the inflationrate over both
periods, the calculationunderstatesthe changein the naturalrate.
On balance, the evidence suggests that the current high level of
unemployment,particularlythe rate for maturemen, cannot easily be
explainedas a consequence of a cyclical decline in demand.I therefore
turnto an explorationof structuralfactors that could possibly account
for risingunemployment.
CHANGES IN LABOR FORCE COMPOSITION
One explanationfor increases in the unemploymentrate is that the
compositionof the labor force has changed so that the share of groups
with high unemploymentrates has increased. In that case, measured
Lawrence H. Summers
345
unemploymentmight increase even though the risk of unemployment
for any particularindividualwith given characteristicshadnot changed.
It would also be true that any given level of unemploymentwould
indicate less labor market slack than had once been the case and so
presumablyshouldbe a cause for less concern.
GeorgePerryput forwardan argumentof this type in consideringthe
breakdown in the Phillips curve relation during the late 1960s.4He
suggested that the increasing share of workers from groups whose
unemploymentwas typically relatively high-women and teenagershad raised the level of measuredunemploymentcorrespondingto any
given amount of labor market slack. Perry constructed an alternative
unemploymentseries by takinga fixed weighted average of the unemploymentrates of differentage-sex groups, thus controllingfor changes
in labor force composition. Since his introductionof this notion, the
construction of "Perry weighted" unemployment rates has become
standardin the estimationof Phillipscurvesandin discussionsof changes
in the naturalrate of unemployment.5
There is no reason why the logic of adjustingfor changes in labor
force composition should be applied only to changes in its age-sex
composition. Argumentssimilarto those originallymade by Perry can
be appliedto other changes in laborforce compositionas well. Some of
the changes work in the opposite direction to Perry's demographic
adjustment.More educatedworkerstend to have lower unemployment
rates than do less educated workers, and the labor force has become
more educated over the past thirty years. Likewise, mining,construction, and manufacturingtend to have higher average unemployment
rates than do services, trade, and finance, and the share of the labor
force engagedin the latterpursuitshas increased. Thus it is not clear a
priorithat changes in the compositionof the laborforce have tended to
increasemeasuredunemployment.
Table3 presentsestimatesof adjustmentsto the measuredunemployment rate for changes in labor composition by age and sex, marital
4. See George L. Perry, "ChangingLabor Marketsand Inflation,"BPEA, 3:1970,
pp. 411-41.
5. For calculationsof the naturalrate of unemploymentand potentialGNP that rely
heavilyon demographicallyadjustedunemploymentrates, see RobertJ. Gordon,"Inflation, FlexibleExchangeRates, and the NaturalRate of Unemployment,"in MartinNeil
Baily, ed., Workers, Jobs, and Inflation (Brookings, 1982), pp. 89-152.
346
Brookings Papers on Economic Activity, 2:1986
Table 3. Changes in Labor Force Composition and the Unemployment Rate, 1954-85
Percentage points
Adjustmenta
Year
Unemployment
rate
Age-sex
1954
1955
5.5
4.4
-0.3
-0.2
n.a.
0.3
n.a.
n.a.
0.3
0.2
1956
1957
1958
1959
1960
4.1
4.3
6.8
5.5
5.5
-0.2
-0.2
-0.2
-0.2
-0.2
0.3
-0.1
-0.1
0.0
-0.1
n.a.
n.a.
n.a.
0.4
n.a.
0.1
0.1
0.2
0.2
0.1
1961
1962
1963
1964
1965
6.7
5.5
5.7
5.2
4.5
-0.2
-0.2
-0.1
-0.1
0.0
0.0
-0.2
-0.1
-0.1
0.0
n.a.
0.2
n.a.
0.0
0.0
0.1
0.1
0.1
0.0
0.0
1966
1967
1968
1969
1970
3.8
3.8
3.6
3.5
4.9
0.1
0.0
0.1
0.1
0.2
-0.1
-0.1
-0.1
0.0
0.0
0.0
-0.1
-0.1
-0.1
-0.1
-0.1
-0.1
0.0
0.0
-0.1
1971
1972
1973
1974
1975
5.9
5.6
4.9
5.6
8.5
0.3
0.4
0.5
0.5
0.7
0.0
0.2
0.2
0.3
0.4
- 0.3
- 0.2
-0.2
-0.3
-0.8
-0.1
0.0
0.0
-0.1
-0.3
1976
1977
1978
1979
1980
7.7
7.1
6.1
5.8
7.1
0.7
0.7
0.7
0.6
0.6
0.5
0.6
0.6
0.5
0.5
-0.7
-0.8
-0.7
-0.8
- 1.1
-0.2
-0.1
-0.1
-0.1
-0.2
1981
1982
1983
1984
1985
7.6
9.7
9.6
7.5
7.2
0.6
0.6
0.4
0.3
0.3
0.7
0.7
0.7
0.6
0.6
- 1.4
- 1.8
- 2.3
- 2.1
-2.1
-0.1
-0.4
-0.3
-0.2
-0.1
Marital
status
Schooling
Industry
Source: Author's calculations. Actual unemployment rate and data before 1984 used in calculations of adjustments
are from BLS, Handbook of Labor Statistics, Bulletin 2217 (June 1985). Unemployment rates by demographic group
are listed in table 27, pp. 69-73; rates by marital status are in table 50, pp. 115-18; rates by education are in table
62, p. 169; rates by industry are in table 30, p. 77. Weights for demographic civilian labor force status are from table
4, pp. 14-17; weights for marital status are from table 6, pp. 22-23; weights for educational attainment are from
table 61, p. 164; weights for industrial composition are from table 30, p. 76. Statistics for 1984 and 1985 are from
BLS, Employment and Earnings, vol. 33 (January 1986).
n.a. Not available.
a. The adjustment for each year is calculated by creating an adjusted unemployment rate using 1965 labor force
shares as weights. (See equation 1 in text.) The adjustment for changing labor force composition is then the difference
between the actual and the adjusted unemployment rate.
Lawrence H. Summers
347
status, schooling,andprimaryindustry.The adjustmentfor each yearis
calculated by first creatingan adjustedunemploymentrate as a fixed
weightedaverageof group-specificunemploymentratesusing 1965labor
force sharesas weights. Thatis,
(1)
A URt =
E si65URit.
The adjustedunemploymentrate, A UR, is the unemploymentrate that
wouldprevailin a given yearif each laborforce grouphadits 1965share,
Si65, of the laborforce. The adjustmentfor changinglaborforce composition is then the differencebetween the actual and the adjustedunemploymentrate.
While it would be ideal to estimate the adjustmentusing a single
decompositionof the laborforce into subgroups,it is not possible to do
so with the availabledata. The table thereforepresents four separate
adjustments.The age-sex adjustmentis based on a decompositionof the
labor force into the fourteen categories used by Perry.6The marital
status adjustmentdivides the laborforce into six categories-men and
women who are single, marriedwith spouse present, and widowed,
separated,or divorced.The schoolingadjustmentis based on a division
of the laborforce into six categoriesrangingfromworkerswithless than
five yearsof schoolingto those completingfouror moreyearsof college.7
Finally, the industry adjustmentis based on a decomposition of the
experiencedlabor force into categories correspondingto the one-digit
standardindustrialclassification(SIC)code.8
The changingage-sex compositionof the laborforce can accountfor
relatively little of the increase in unemploymentin recent years. The
adjustment(relativeto 1965)peaked at 0.7 percentagepoint in the mid1970sand has declined since then to only 0.3 point in 1985.The decline
in the adjustmentreflects two developments: the decline in the labor
force share of teenagersand the decliningrelative unemploymentrate
of women. The laborforce shareof teenagershas fallenfrom7.9 percent
6. Thecategoriesaremenandwomenagedsixteento nineteen,twentyto twenty-four,
twenty-fiveto thirty-four,thirty-fiveto forty-four,forty-fiveto fifty-four,fifty-fiveto sixtyfour,andsixty-fiveandover.
7. The schoolingadjustmentis calculatedusing datafor Marchof each year because
questionson educationalattainmentare askedonly in the MarchCPS.
8. Theindustryadjustmentis not strictlyparallelto the otherssince it is an adjustment
to the unemploymentrate of experiencedworkers.This noncomparabilityis inevitable
giventhatnew entrantsto the laborforce cannotmeaningfullybe assignedan industry.
348
Brookings Papers on Economic Activity, 2:1986
in 1965to 6.8 percentin 1985,a decline that is assumedhere to have no
effect on the youth unemploymentrate. If, as Michael Wachter has
argued, crowdingeffects cause increases in the youth unemployment
rate, the decline in the adjustmentin recent years wouldbe significantly
greater.9
The dramaticchange in the compositionof the laborforce in recent
years has been the increase in female labor force participation. If
unemploymentrates for men and women had maintainedthe pattern
they exhibited in the 1960s, the measuredunemploymentrate would
have increasedsubstantially.But, as I discuss later, the gapbetween the
unemploymentrates of men and women has narrowedin recent years,
so the effect is not very large.
Moreimportantthanthe age-sex adjustmentis the adjustmentfor the
changingmaritalstatusof the laborforce. It rose to 0.6 percentagepoint
in the mid-1970sbut, unlikethe age-sex adjustment,has not turneddown
since. The majormaritalstatuschangeis the dropin the fractionof men
in the laborforce who are married.In 1965, 18percentof the male labor
force had never been married,comparedwith 27 percentin 1985.Given
that unemploymentrates were three times as high for single as for
marriedmen in 1985, the effect of reductionsin the share of the labor
10
force thatis marriedis quite substantial.
Quantitatively,the most importantadjustmentfor changes in the
compositionof the laborforce involves education.Assumingno changes
in group-specificunemploymentrates, recent increases in education
should have reduced the unemploymentrate by 2.1 percentagepoints
between 1965and 1985. That adjustmentdwarfs the demographicand
11 Over the past twenty years the shareof the
maritalstatus corrections.
laborforce that received some college educationnearly doubled, from
22 to 40 percent,whilethe sharewith less thanan eighth-gradeeducation
fell from23 percentto only 7 percent.
9. See Michael L. Wachter, "IntermediateSwings in Labor-ForceParticipation,"
BPEA,2:1977,pp. 545-74. Wachterforecast,on the basisof demographicconsiderations,
a significantdeclinein the naturalrateof unemploymentin the early 1980s.
10. The adjustmentis not independentof the previousadjustmentfor changesin the
age-sex compositionof the laborforce. It also reflectsincreasesin the shareof womenin
the laborforce andto some extent reflectschangesin the age structureof the laborforce.
The rise in the proportionof singlemen notedin the text is particularlystrikingin lightof
the declinein the shareof teenagersandyoungmen in the laborforce.
11. RobertM. Solow madethis pointin "Macro-policyandFull Employment,"in Eli
Ginzberg,ed., Jobsfor Americans(PrenticeHall, 1976),pp. 46-48.
Lawrence H. Summers
349
It is arguablethat the educationaladjustmentmade here is inappropriate because the differentialsin unemploymentbetween different
educationalgroupsreflectnot the effects of educationbut ratherdifferences in the innate skills of more and less educatedworkers. Undoubtedly, the adjustmentcalculatedhere is an overestimateof the trueeffect
of increasededucationalattainmenton unemploymentfor this reason,
but the overestimatemay not be large. The premiseof policies directed
at discouragingteenagersfrom droppingout of high school is that more
schooling means less unemployment. And the fact that the relative
unemploymentrate of college graduateshas droppedas their number
has swelled casts doubt on the importanceof sortingeffects of the type
noted above.
Finally,the industryadjustmentshownin the finalcolumnof the table
indicates that changes in the industrialmix, particularlythe decline in
the share of employmentin the volatile manufacturingsector, has also
workedto reduceunemployment.In some recent years, when manufacturinghas been weak because of adversecyclical conditions,the adjustment has been quantitativelysignificant,reaching0.4 percentagepoint
in 1982.
While it is inappropriateto sum the various adjustmentsin table 3
because they are not independent,it seems clearthatmix effects cannot
account for the recent increase in unemployment.The mix effects that
should have led to decreases in unemployment,increases in education
and reductionsin manufacturingemploymentare quantitativelymuch
more importantthan the demographicmix effects that are emphasized
in most discussions of rising unemployment.Takinginto account the
changingcompositionof the laborforce does not reduce and may even
increasethe size of the rise in unemploymentthat must be explained.
WHOSE
UNEMPLOYMENT
HAS
INCREASED?
Sincechangesin laborforce compositioncannotaccountfor increases
in employment,it is naturalto ask how the increasein unemploymentin
recentyearshas been distributedacross the population.Table4 presents
unemploymentrates for various subgroupsof the populationin 1965,
1974,1978,and 1985.The years 1965,1974,and 1978arecontrastedwith
1985because each is a yearof moderatelylow but not cyclically minimal
unemployment.The broadconclusionsthatemergein the discussionare
not sensitive to the choice of years.
350
Brookings Papers on Economic Activity, 2:1986
Table 4. Unemployment Rates for Population Subgroups, 1965, 1974, 1978, and 1985
Percent
Category
1965
1974
1978
1985
4.5
5.6
6.1
7.2
14.1
15.6
15.8
19.5
6.4
2.9
2.5
2.5
3.3
3.5
4.0
8.8
4.0
2.6
2.4
2.6
3.3
4.9
9.2
4.4
2.8
2.7
2.8
4.2
5.3
11.4
6.6
4.9
4.6
4.3
3.1
7.0
Females, 16-19
Females, 20-24
Females, 25-34
15.7
7.3
5.5
16.6
9.5
6.2
17.1
10.1
6.7
17.6
10.7
7.4
Females, 35-44
4.6
4.6
5.0
5.5
Females, 45-54
Females, 55-64
Females, 65 and over
All females
3.2
2.8
2.9
5.5
3.7
3.2
3.6
6.7
4.0
3.2
3.8
7.2
4.8
4.3
3.3
7.4
10.1
2.4
11.8
2.7
11.7
2.8
12.7
4.3
7.2
8.2
6.2
10.5
6.6
10.9
9.2
10.7
Married women
4.5
5.3
5.5
5.6
Divorced, separated,
or widowed women
5.4
6.3
6.9
8.3
7.1
5.6
4.8
6.2
7.7
8.5
11.3
13.0
7.4
9.6
12.4
15.9
4.1
4.8
6.2
8.0
3.3
4.2
4.6
5.1
1.4
2.0
2.5
2.6
Total
Age-sex
Males, 16-19
Males, 20-24
Males, 25-34
Males, 35-44
Males, 45-54
Males, 55-64
Males, 65 and over
All males
Marital status
Single men
Marriedmen
Divorced, separated,
or widowed men
Single women
Educationa
Less than five years
Five to eight years
One to three years of
high school
Four years of high
school
One to three years
of college
Four or more years
of college
Sources: BLS, Handbook of Labor Statistics (June 1985). Unemployment rates by age-sex are from table 27, pp.
69-73; by marital status, from table 50, pp. 115-18; and by education, from table 62, p. 169. Data for 1985 are taken
from Employment and Earnings, vol. 33 (January 1986).
a. Education statistics for 1985 were obtained by telephone from the Bureau of Labor Statistics.
Lawrence H. Summers
351
The most dramaticrelativeincreasesin unemploymenthave occurred
amongprime-agedmales. Whileaggregateunemploymentincreasedby
18percent, from6.1 percentto 7.2 percent, between 1978and 1985,the
unemploymentrate for men aged thirty-fiveto forty-fourincreased by
75 percent, from 2.8 percent to 4.9 percent. The increase occurred
despite a rise in that cohort's laborforce nonparticipationrate from 4.3
to 5.0 percent.A similarbut less pronouncedincreasein unemployment
is observed for men in the other under-sixty-five age groups. It is
noteworthy that even going as far back as 1965, the conclusion that
unemploymentamongmaturemen has risendisproportionatelyremains
valid.
Unemploymentrates for women have risen relatively little, despite
hugeincreasesin laborforce participationrates. For women agedthirtyfive to forty-four, the unemploymentrate increased only 10 percent,
from 5.0 to 5.5 percent, during 1978-85. The relative increases in
unemploymentwere somewhat smallerfor youngerwomen and somewhatgreaterfor olderwomen. Thereis a substantialdifferencebetween
the experienceof youngmen andthatof youngwomen. Whileincreases
in the unemploymentof women aged sixteen to nineteen and twenty to
twenty-four have been relatively small since 1974, there have been
significantincreases in the unemploymentrate of young men, particularlythose aged sixteen to nineteen.
The total unemploymentrate is a weightedaverageof the unemployment rates of differentdemographicgroupswith weights dependingon
theirsharesof the laborforce. A simpleway of combiningthe effects of
changingdemographiccompositionand changinggroup-specificunemploymentratesis to ask whatcontributiondifferentdemographicgroups
make to total unemploymentin differentyears. The contributionof a
givengroupto total unemploymentis the productof its laborforce share
and its unemploymentrate. Performingthis calculation reveals two
significantdevelopments. First, the amountof unemploymentattributable to teenagers has declined in recent years. Teenagers contributed
1.2 percentagepoints to the 4.5 percentunemploymentrate in 1965, 1.5
points to the 5.6 percent unemploymentrate in 1974, 1.5 points to the
6.1 percent unemploymentrate in 1978, but only 1.3 points to the 7.2
percentunemploymentrate in 1985. Second, the bulk of the increase in
unemploymentin recent years is attributableto men aged twenty and
above, whose contributionto total unemploymentincreased from 1.9
percentagepoints in 1965to 2.3 points in 1978and 3.3 points in 1985.
352
Brookings Papers on Economic Activity, 2:1986
Dataon unemploymentratesfor differentmaritalstatusgroupsreveal
thatunemploymenthas increasedmost dramaticallyamongmarriedand
formerlymarriedmen. The rate for these groupsincreasedby about 50
percent between 1978 and 1985. For single men and women and for
marriedwomen the datareveal only very minorincreases in unemployment since 1974. These patterns cast doubt on the arguments that
increasesin measuredunemploymentare primarilythe resultof a rise in
the fraction of the populationon the marginbetween workingand not
working.'2Surely maturemen, especially those who are married,are
the groupfor whom it is least plausiblethat social changes have made
marginallaborforce attachmentattractive.
Finally, the breakdownof unemploymentrates by educationin table
4 reveals that the extent of the increase in unemploymentover the past
decade declines steadily with increasededucation. The unemployment
rateof highschool drop-outsincreasedby morethanone-fourthbetween
1978and 1985,comparedwith an increase of only 4 percentfor college
graduatesand 11percentfor workerswith some college education.The
unemploymentrate for those with only one to five years of schooling
rose by almost 50 percent. The level of unemploymentis not, however,
monotonically related to education either in the 1970s or at present.
People receivingno high school traininghave significantlylower unemployment rates than do high school drop-outs.'3That pattern at least
raises a questionaboutargumentsthatunemploymentis due to a lack of
skills on the partof workers.
WHAT
TYPES
OF UNEMPLOYMENT
HAVE
INCREASED?
The discussion so far suggests that the increase in measuredunemployment is potentially a serious social problem. Furtherevidence to
that effect can be gleaned from data on changes in the composition of
unemploymentby reason and duration.As table 5 suggests, most of the
increase in unemploymentover the last decade is concentratedamong
12. For the clearest and most persuasive statementof this view see Robert Hall's
commenton Medoff, "U.S. LaborMarkets," BPEA, 1:1983,pp. 121-23. I considerthe
argumentin moredetaillater.
13. Age effects may be at workhere. It is likelythathighschool drop-outsin the labor
forceareon averagemuchyoungerthanpeoplereceivingless thaneightyearsof schooling.
The issue cannotbe investigatedusingpublishedtabulations.
Lawrence H. Summers
353
Table 5. Unemployment by Reason and Duration, Various Years, 1965-85
Percent except where otherwise indicated
Reason for unemployment
Year
Unemployment
rate
Job losers
Job leavers
1967a
1974
1978
1985
3.8
5.6
6.1
7.2
1.6
2.4
2.5
3.6
0.5
0.8
0.8
0.8
Reentrants
1.2
1.6
1.8
2.0
Duration of unemployment
New
entrants
0.5
0.7
0.9
0.9
Share of
Year
Unemployment
rate
0-5
weeks
6-14
weeks
15-26
weeks
27 or
more
weeks
Mean
duration
(weeks)b
long-term
unemploymentc
1965
1974
1978
1985
4.5
5.6
6.1
7.2
2.2
2.8
2.8
3.0
1.3
1.7
1.9
2.2
0.5
0.6
0.8
0.9
0.5
0.4
0.6
1.1
11.8
9.8
11.9
15.6
42.5
45.2
46.0
54.Od
Sources: BLS, Handbook of Labor Statistics (June 1985). Unemployment by reason for unemployment is in table
32, pp. 80-81. Unemployment by duration of unemployment is in table 31, pp. 78-79. Statistics for 1985 are from
Employment and Earnings, vol. 33 (January 1986), table 12, p. 166, and table 14, p. 167.
a. Data on reason for unemployment do not begin until 1967.
b. Mean duration of interrupted spells.
c. Fraction of the year's unemployment due to persons with more than twenty-seven weeks of unemployment as
derived from the Work Experience Survey.
d. Because 1985 data are unavailable, data from 1984 are used.
job losers. The unemploymentrate attributableto job loss rose from 1.6
percentin 1967to 2.4 percentin 1974,2.5percentin 1978,and3.6 percent
in 1985.Unemploymentattributableto job leavers has not increasedat
all since 1974, while unemploymentamong new entrantsto the work
force has increased modestly. Noticeable increases in unemployment
have also taken place among workers reenteringthe work force. For
reasonsspelledout in detailin my earlierpaperwith KimClark,I believe
thata substantialpartof the reentrantcategoryis composed of workers
whohaverecentlylostjobs. 14If even a portionof the increasein reentrant
unemploymentis addedto thejob losers category, it appearsclear that
14. Kim B. Clarkand LawrenceH. Summers,"LaborMarketDynamicsand Unemployment:A Reconsideration,"BPEA, 1:1979,pp. 13-60. We show thatmanyreentrants
haverelativelyrecentworkexperienceandreportdurationsof unemploymentvery close
to the totaltime since they last worked.The traditionalpictureof housewivesreentering
the laborforce after their childrenhave grown up is grossly inconsistentwith the facts
regardingreentrantunemployment.
354
Brookings Papers on Economic Activity, 2:1986
the bulkof the increasein unemploymentin recent years is the result of
job loss.
The data also suggest that a large part of the observed increase in
unemploymentis due to increases in the durationof unemployment.Of
the 1.1 percentage point increase in the unemploymentrate between
1978and 1985, 0.5 point, or almost half, is attributableto increases in
the numberof people reportingthemselves as out of workfor more than
twenty-seven weeks. The incidence of such long-termunemployment
has more than doubled since 1965. Only a relatively small part of the
observedincreasein unemploymentis due to an increasein the number
of peoplereportingthemselvesas unemployedforfewer thanfive weeks.
Data on unemploymentdurationare difficultto interpretbecause of
the high incidence of reportingerrors. It appears that almost threequartersof the unemployedpopulationreport their durationof unemployment inconsistently from month to month.15 There is also the
complication, emphasized by many authors, that almost half of all
unemploymentspells end in withdrawalfromthe laborforce ratherthan
in employment.'6Nonetheless, the availableinformationsuggests that
unemploymentis increasinglyconcentrated among a relatively small
groupthatis unemployedfor long stretchesof time.
An easy way to see this point is to note that doubling the mean
duration of incomplete spells of unemployment (shown in table 5)
provides an estimate of the mean durationof the completed spell for
those currentlyunemployed.17As ClarkandI arguedin ourearlierpaper,
this concept is far superiorto the morecommonlystudiedmeanduration
of a completed spell for those enteringunemploymentin assessing the
dynamicsof unemployment.The expected total durationof unemploy15. The consistency of individuals'reportedunemploymentdurationfrom monthto
monthis examinedin JamesM. PoterbaandLawrenceH. Summers,"ResponseVariation
in the CPS: Caveatsfor the UnemploymentAnalyst," MonthlyLaborReview, vol. 107
(March1984),pp. 37-43.
16. This findingis probablya consequenceof measurementerrorin the CPS survey.
See JamesM. PoterbaandLawrenceH. Summers,"ReportingErrorsandLaborMarket
Dynamics,"Econometrica(forthcoming).
17. For discussions of alternativeconcepts of the durationof unemployment,see
StephenW. Salant, "SearchTheoryand DurationData:A Theoryof Sorts," Quarterly
Journalof Economics, vol. 91 (February1977),pp. 39-57; GeorgeA. Akerlofand Brian
G. M. Main, "UnemploymentSpells and UnemploymentExperience,"AmericanEconomic Review, vol. 70 (December 1980),pp. 885-93; and Clarkand Summers,"Labor
MarketDynamics."
Lawrence H. Summers
355
ment for the unemployed is now thirty-one weeks, compared with
twenty-fourweeks in 1978,twentyweeks in 1974,andtwenty-fourweeks
in 1965.Takingaccountin some way of the shorteningof reportedspells
of unemploymentthatcan be attributedto laborforce withdrawalwould
furtherincrease the estimateddurationof joblessness for the currently
unemployedpopulation.
Additionalevidence on the concentrationof unemploymentamong
the long-termunemployedis providedby the retrospective Work Experience Survey conducted annuallyin Marchas a supplementto the
CPS. The March survey, in which respondents are asked about the
extent of their unemploymentand employmentexperience in the preceding year, makes it possible to calculate the fraction of total unemployment attributableto people experiencingdifferentamounts of unemploymentin the precedingyear. In ourearlierpaper,ClarkandI used
this data to suggest that a largefractionof unemploymentin 1969, 1974,
and 1975was attributableto the relatively small subgroupof the population that experiencedmore than six months of unemploymentin the
preceding year.18 Replicating our calculations for subsequent years
suggests that the importanceof long-termunemploymenthas increased
significantly.Whilepeople out of workfor twenty-sevenor moreweeks
accounted for 45.2 percent of all unemploymentreportedin the 1974
WorkExperienceSurvey, they accountedfor 46.0 percentof unemploymentin 1978and54.0 percentof unemploymentin 1984,the most recent
yearfor which dataare available.
Increases in normalunemploymentover the past twenty years represent a serious problem. The view that the current high level of
unemploymentis primarilythe resultof the increasedunemploymentof
secondaryworkersis simply false. In fact, the increases in unemployment have been relatively greatest for mature men with dependents.
And they have resultedprimarilyfromjobloss andincreasesin duration
of unemployment.
CPS UNEMPLOYMENT
AND
OTHER
LABOR
MARKET
INDICATORS
A numberof recent analyses have called attentionto the fact that the
observed increase in the official unemploymentrate has not coincided
18. ClarkandSummers,"LaborMarketDynamics,"table4, pp. 36-37.
356
Brookings Papers on Economic Activity, 2:1986
with substantialincreasesin otherlabormarketindicators.'9 It could be
that some flaw in the CPS measureof unemploymentaccounts for the
observed increase, though such an argumentis difficult to evaluate.
Unemploymentas reflectedin the CPS is more a state of mind than an
objective reality. The substantialimportanceof rotationgroupbias and
the sensitivityof the measuredunemploymentrateto even smallchanges
in the phrasingor the orderof the questionsaskedsuggeststhe subjective
nature of measuredunemployment.20This means that it is difficultto
examinewhetheror not the CPS is correctlymeasuringunemployment.
In an importantsense, unemploymentis what the CPS says it is.
Nonetheless, it is usefulto contrastmovementsin CPSunemployment
rateswith movementsin othervariablesthatare likely to reflectchanges
in labor market conditions. Table 6 presents estimates of the CPS
unemploymentrate, the insuredunemploymentrate, the unemployment
rateas inferredfromthe annualretrospectiveWorkExperienceSurvey,
and the discouragedworkerrate.2"A majormystery is the sharprecent
decline in the ratio of insured unemploymentto total unemployment.
The insuredunemploymentrate-the numberof recipientsof unemployment benefitsdividedby the numberof jobs covered by unemployment
insurance-was about 15 percent lower in 1985than it was in 1978and
20 percentlowerin 1985thanit was in 1974.It was only one-thirdgreater
thanit was in 1969.As GaryBurtlessexplains,one wouldexpect insured
unemploymentto be below actual unemploymentsince many of the
unemployedare ineligiblefor benefits.22Burtless also suggests that the
increasingshareof the populationcovered by unemploymentinsurance
can account for some of the pre-1980 trend decrease in the ratio of
19. See, for example,MartinNeil Baily, "LaborMarketPerformance,Competition,
and Inflation," in Baily, ed., Workers, Jobs, and Inflation, pp. 15-44; Hall's comment on
Medoff, "U.S. Labor Markets";Gary Burtless, "Why is Insured UnemploymentSo
Low?," BPEA, 1:1983,pp. 225-49; andGeorgeA. AkerlofandJanetL. Yellen, "Unemployment throughthe Filter of Memory," QuarterlyJournal of Economics, vol. 100
(August1985),pp. 747-73.
20. For a discussionof these points stressingthe ambiguityinherentin the distinction
between being unemployedand not being in the labor force, see Clarkand Summers,
"LaborMarketDynamics."
21. It wouldbe desirableto examinethe quitandlay-offratesin conjunctionwithother
labormarketindicators.Unfortunately,publicationof these turnoverdatawas discontinued after 1981.As Baily, "LaborMarkets,"argued,theirbehaviorup until 1981does not
mirrorthatof the officialunemploymentrate.
22. Burtless,"InsuredUnemployment."
Lawrence H. Summers
357
Table 6. Unemployment and Alternative Labor Market Indicators, 1960-85
Percent
Year
Unemployment
rate
Work
Experience
Survey
unemployment
ratea
Insured
unemployment
rate
Discouraged
worker rate
1960
1961
1962
1963
1964
5.5
6.7
5.5
5.7
5.2
6.0
6.6
6.2
5.7
5.2
4.8
5.6
4.4
4.3
3.8
n.a.
n.a.
n.a.
n.a.
n.a.
1965
1966
1967
1968
1969
4.5
3.8
3.8
3.6
3.5
3.9
3.2
3.1
2.9
3.0
3.0
2.3
2.5
2.2
2.1
n.a.
n.a.
n.a.
0.9
0.7
1970
1971
1972
1973
1974
4.9
5.9
5.6
4.9
5.6
4.7
5.6
5.1
4.2
5.3
3.4
4.1
3.5
2.7
3.5
0.8
0.9
0.9
0.8
0.8
1975
1976
1977
1978
1979
8.5
7.7
7.1
6.1
5.8
8.0
7.3
6.4
5.3
5.0
6.0
4.6
3.9
3.3
2.9
1.2
1.0
1.0
0.8
0.7
1980
1981
1982
1983
1984
7.1
7.6
9.7
9.6
7.5
6.8
7.4
9.4
8.4
6.9
3.1
3.5
4.6
3.8
2.8
0.9
1.0
1.4
1.5
1.1
1985
7.2
n.a.
2.8
1.0
Source: Author's calculations and BLS, Handbook of Labor Statistics (June 1985). Data for 1984 and 1985 are
from Employment and Earnings, vol. 33 (January 1986).
n.a. Not available.
a. Calculated from published tabulations as the ratio of total weeks of unemployment to labor force time.
insuredunemploymentto total unemployment.But thereis no apparent
explanationfor the divergence of these two measures in recent years.
The mysteryis deepenedby the observation,noted above, that most of
the recent increase in unemploymenthas been due to increases in the
job losercategory,the categorymosteligiblefor unemploymentbenefits.
Burtless considers a numberof possible explanationsfor the recent
358
Brookings Papers on Economic Activity, 2:1986
low level of the insuredunemploymentrate withoutfindingany that are
wholly persuasive. It appears that many people who, based on their
answers to the CPS questionnaire,appearto be eligible for unemployment insuranceare not collectingit, possibly because benefitsbegan to
be taxed in 1980 or, more plausibly, because administrativechanges
have increasedthe logisticaldifficultiesassociated with collecting benefits. It is conceivable that receipt of benefitscarriesmore stigmain the
Reaganera than it once did. Perhapsthe most plausibleexplanation,in
view of the increasingaveragedurationof unemployment,is that many
of the unemployedhave exhausted their unemploymentinsuranceeligibilityduringeithertheircurrentunemploymentspellor a previousone.
Althoughit is not clearwhatthe low insuredunemploymentrate means,
at a minimumit exonerates unemploymentinsuranceas a cause of the
high level of unemployment.If a smaller share of the labor force is
collecting benefits than used to be the case, unemploymentinsurance
can hardlybe blamedfor increasingunemployment.
The second columnof table6 follows GeorgeAkerlofandJanetYellen
inreportingthe WorkExperienceSurveyunemploymentrate,calculated
as the ratioof reportedunemploymentfor the precedingyearto reported
laborforce participation,definedas the sumof time spentemployedand
unemployed.23As they note, using data for the 1960-81 period, there
has been a tendencyfor the retrospectiveunemploymentrate to decline
relativeto the officialrate over time. Between 1974and 1984,the Work
Experience Survey unemploymentrate increased by 1.6 points; the
officialrate, 1.9 points.
Akerlofand Yellen estimate that the CPS unemploymentrate corresponding to any given Work Experience Survey unemploymentrate
rose by about0.8 percentper year through1981,a relationshipthat has
held up over the threeadditionalyearsfor whichdatahave since become
available.24The CPS rate has thus risen by about 12 percent, or 1
23. My calculationdiffersslightlyfromthat of Akerlofand Yellen, "Unemployment
throughthe Filterof Memory,"becauseI use reportedlaborforce participationfromthe
WorkExperienceSurveyas the denominatorin calculatingthe WorkExperienceSurvey
unemploymentrate ratherthan using labor force data from the CPS as they did. My
procedurereducessomewhatthe differentialbetweenthe two series.
24. The precise estimatedependson what adjustmentis madefor the changesin the
CPS institutedin 1967afterthe GordonCommissionreport.See President'sCommission
to AppraiseEmploymentand UnemploymentStatistics, Measuring Employment and
Unemployment (GPO,1962).
Lawrence H. Summers
359
percentagepoint,relativeto the retrospectivelyreportedunemployment
rate over the past fifteen years. Akerlofand Yellen find, however, that
there has been essentially no trend increase in the ratio of CPS unemployment to retrospectiveunemploymentfor either prime-agemen or
prime-agewomen over this period. They also report that about onefourth of the movement in official unemployment relative to Work
Experience Survey unemploymentcan be explainedby changes in the
composition of the labor force, particularlythe influx of women, for
whomthe ratioof retrospectiveunemploymentto officialunemployment
is particularlylow.
Citing a variety of psychological studies suggesting that the more
painful an experience the better people recall it, Akerlof and Yellen
attributethe risingdifferentialbetween the two ratesto a decrease in the
discomfortassociatedwith unemployment.They buttresstheirclaimby
noting that the ratio of retrospective unemploymentto official unemployment is highest for mature men and that it rises in recessions. A
naturalinterpretationof the Work Experience Survey informationis
thatunemploymenthas become a less painfuland salientexperiencefor
youngworkers.It mightbe moreaccurateto say thatthe unemployment
of young workers has become less salient for their parents, since one
memberof a household, typicallyan adult, providesinformationon the
labor marketstatus of all household membersin the WorkExperience
Survey, as in the CPS. The reductionin the salience of unemployment
is not surprisinggiven the sharpincreasein the shareof young people in
school. It seems reasonable to conclude from the Work Experience
Surveydatathata 7 percentunemploymentratetodayis associatedwith
less distress thanwas once the case. But the data shed little light on the
observed increase in unemployment, most of which has come from
adults.
Thefinalcolumnof the tablepresentsthe "discouragedworker"rate,
estimated as the numberof discouragedworkers divided by the total
laborforce. Discouragedworkersare definedas those who cite inability
to findwork as their sole reason for not searching.Many analysts have
argued that they should properly be counted as unemployed. The
discouragedworker rate has moved in parallelwith the official unemploymentrate over the past fifteen years. If, as some have argued, an
increasingpercentage of unemploymentreflects marginallabor force
attachment,one might have expected to see a decline in the ratio of
360
Brookings Papers on Economic Activity, 2:1986
discouragedworkersto unemployedpersons. The observed increase in
discouragementover the past decade suggests that increases in unemploymentdo in fact reflectincreasesin the difficultyofjob finding.
Different labor marketindicatorscapturedifferentaspects of labor
marketperformance.It does not appearthat other labormarketindicators provide a basis for concludingthat the observed increase in CPS
unemploymentreflects measurementerror. However, they do suggest
thatthe natureof unemploymentmay have changedover the past fifteen
years.
THE SEARCH ACTIVITIES
OF THE UNEMPLOYED
Oversimplifyingslightly, people are counted as unemployedif they
reportbeing availablefor work in the CurrentPopulationSurvey week
and report having looked for work in the preceding four weeks. In
practice,the firstquestionregardingavailabilityfor workis the principal
determinantof unemploymentstatus. All survey respondentsare asked
their primaryactivity. Five answers are possible for the unemployed:
with a job (to which they expect to return),looking for work, keeping
house, in school, andother. The last categoryincludesbut is not limited
to retiredworkers. If increasingunemploymentreflects an increase in
the numberof people marginallyattachedto the laborforce, the number
reportingtheirprimaryactivityas lookingforworkshouldhave declined.
The intensityof theirsearch shouldalso have declined.
While data on the primaryactivity of the unemployedare not published, I was able to construct a time series on primaryactivity using
raw data fromthe CPS for May of each year from 1973through1984.A
conspicuousfeatureof the datareportedin table7 is thatonly a minority
of the unemployed report themselves as having a job or report their
primary activity as looking for work.25The fraction reporting their
primaryactivity as looking for work or as having a job to which they
expect to returnvaries cyclically but shows no trend during 1973-84.
The datareveal significantdeclines in the proportionof the unemployed
reportingtheir primaryactivity as keeping house or being in school, a
findingthat is supportedby the observationthat the averagenumberof
searchmethodsused by unemployedpersons has graduallyincreased.
25. See Hall's commenton Medoff, "U.S. LaborMarkets";and Hall, "The Nature
andMeasurementof Unemployment,"WorkingPaper252(NationalBureauof Economic
Research,July 1978).
Lawrence H. Summers
361
Table 7. Major Activity of the Unemployed, 1973-84
Percent except where otherwise indicated
Year
With a
job
Looking
for work
Keeping
house
In
school
Other
activitiesa
Average
number
of search
methods
used
1973
1974
6.8
6.5
34.3
35.7
27.3
25.5
18.0
17.7
13.5
14.7
1.52
1.54
1975
1976
1977
1978
1979
12.1
6.1
4.8
5.4
4.7
38.3
38.4
36.7
34.0
33.2
23.2
23.4
25.2
26.8
25.8
13.2
15.4
16.3
15.9
16.6
13.2
16.7
16.9
17.9
19.7
1.58
1.58
1.57
1.53
1.54
1980
1981
1982
1983
1984
9.4
6.3
6.8
5.1
5.0
34.2
34.5
36.4
38.6
37.0
21.6
21.9
20.4
20.1
21.3
13.0
13.9
12.8
12.4
14.2
21.8
23.4
23.6
23.9
22.5
1.58
1.59
1.63
1.64
1.63
Sources:Averagenumberof searchmethodsused is fromBLS, Handbookof LaborStatistics(June1985),table
35, pp. 85-88, and fromEmploymentand Earnings,vol. 33 (January1986).Majoractivityof the unemployedwas
computedby the authorusingdatafromthe CurrentPopulationSurveyfor Mayof each year. Figuresare rounded.
a. Includesthose unemployedwho are listedas unableto work.
The mysteryin table7 is the dramaticincreasein the numberof people
listing "other" as theirprimaryactivity. While "other" includes retirement, it is implausiblethat the large increase in the category could be
accounted for by increasingretirement.26As shown in table 8, which
presents an age-sex breakdownon the primaryactivity of the unemployed for 1974 and 1984, increases in the "other" category are not
confined to older workers for whom the retirement explanation is
plausible.The fractionof men aged twenty-fiveto thirty-fourreporting
"other" as theirprimaryactivity rose from21.3 percentto 35.3 percent
between 1974and 1984.Perhapsmoreimportant,the tablereveals large
demographicdifferencesin the natureof the changesin primaryactivity.
There appear to be quite substantialdeclines in the number of men
lookingfor workexcept for the cohortagedtwenty to twenty-four,while
the numberof women looking for work has increased slightly. More
detailed tabulationssuggest that the declines in the fraction of those
26. A phone call to the Bureauof Labor Statistics did not succeed in eliciting any
aboutthenatureof theanswerscategorizedas "other"beyondtheobservation
information
thatit includedpersonslabelingthemselvesas retired.
* Cqt
06 6
r?e'1o0ON
)N
(
4'~~O?NONON-
.}
00
ONx'I
f)
OC0
1- Ic ,:^t 0
00
b
ON
-
t
O
o~~~C;
r-NO
00
C.
O
(O
W) W) OC OC (O C)
0
0
O
C
Ic
1-
en O
OCC
O
t
O
i
m
o.
t
O
(0003OC
OC
W) 'I
cq
O
en 'I
X
'
It
ent
Cq~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
(V
-:t o
00
CA "C
O O
r- ,:t O0 OO
t2e
0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.
0~~~~~~~~~~~~~~~~~~
0
0
Cu~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~C
0~~~~~~~~~~~~~~~~~~~~~~~~.
N "C
ON
Y~~~t
t
9
<t_^
O'"
C)
"C
c l00 q"C
ONen~
,
C.
O
C)
?? mNN
00
o
06
"
.0
-
?
t
O
O0
o
lClrON
?O
0
ClClr
0
0_
t
CNO
H
Lawrence H. Summers
363
whose primaryactivity is lookingfor work are concentratedamongjob
losers and leavers.
Given the ambiguitiesassociatedwith the "other" category, it is not
clear how to interpretthese figures. They may well be related to the
greaterincreasein unemploymentamongmenthanamongwomen. They
may also have somethingto do with broadersocial trendsregardingthe
division of family responsibilitiesbetween men and women. Another
possibilityis that single men, who have increasedas a shareof the male
laborforce in recent years, feel less pressureto look for work thantheir
marriedcounterparts.
All of the informationpresentedso far on the increase in unemployment suggests that it is a serious problem. Increases in normalunemployment reflect neither measurementproblems nor changes in the
demographiccompositionof the laborforce. Rather,unemploymenthas
increasedin those segmentsof the populationwhere it is most seriousamongmarriedmen,job losers, and the long-termunemployed.
Regional Differences in Unemployment
So far my object has been more to accountfor the observedincrease
in unemploymentthan to explain it. Inevitably, aggregatetime series
dataare not richenoughto distinguishalternativeexplanationsfor rising
unemployment.In seeking explanations, I turn to informationon the
differentlabormarketexperiencesof differentpartsof the country.27
Databy statereflectwidelynotedpatternsin recentregionaleconomic
growth. Duringthe past fifteen years, for example, New Englandhas
performedextraordinarilywell, while the North Central States have
fared poorly. California's economy has done well, while Alabama,
Mississippi,and Louisianahave sufferedsignificantincreases in unemployment.
The datareveal significantvolatilityin the patternof state unemploy27. Studiesexploringaspects of the geographicdistributionof unemploymentinclude
RobertE. Hall, "Turnoverin the Labor Force," BPEA, 3:1972, pp. 709-56; Medoff,
"U.S. Labor Markets"; and Stephen T. Marston, "Two Views of the Geographic
Distributionof Unemployment,"QuarterlyJournal of Economics, vol. 100 (February
1985),pp.57-79, amongmanyothers.Theviewof geographicdifferencesinunemployment
putforwardhereparallels,in some respects, thatof HallandMarston.
364
Brookings Papers on Economic Activity, 2:1986
mentrates. The correlationbetween 1970and 1985state unemployment
rates was 0.54. Somewhat surprisingly,the correlationbetween unemployment rates in the mid-1970sand 1985was significantlylower. For
example, the correlationbetween unemploymentrates in 1976and 1985
was only 0.03, and the correlationbetween unemploymentrates in 1978
and 1985was 0.33. That volatility over the past fifteen years indicates
thatregionalinformationhas the potentialto illuminatethe causes of the
observedincreasein normalunemployment.28
EMPLOYMENT
OPPORTUNITIES
AND UNEMPLOYMENT
One explanationfor regional unemploymentdifferentialsis differences in industrialcomposition.For example,the problemsof the North
Centralareaare often attributedto its heavy relianceon manufacturing,
while the strengthof the New England economy is explained by the
growthin its "high-tech"industries.To the extent that regionaldifferences in unemploymentreflect only differences in industrialcomposition, however, they can explainonly a little of the observed increase in
aggregateunemployment.
In order to explore the importanceof such composition effects in
explainingdifferencesin state unemploymentrates, I used direct information from the CPS to compute adjusted state unemploymentrates
thatcontrolfor differencesin demographic,educational,industrial,and
occupationalcompositionamongstates. I used datafrom the May CPS
for selected years to estimate an equation relating an individual's
employmentstatusto his age, sex, andmaritalcharacteristics,two-digit
industry,one-digitoccupation,educationalattainment,and his state of
residence.29Using the coefficientson the state dummies,I constructed
adjustedunemploymentratesandthen normalizedthem so thataverage
adjustedunemploymentwouldequalaverageunemploymentas officially
reportedfor the entireyear.
28. U.S. Departmentof Labor, Bureau of Labor Statistics, GeographicProfile of
Employment and Unemployment, 1985 (September 1986) and earlier issues.
29. The variablesincludedin the equationused to constructadjustedstate unemployment rates were age, sex, age-sex, maritalstatus-sex, education, education squared,
education-sex,educationsquared-sex,race,centercity status,one-digitoccupation,twodigitindustry,and statedummies.
Lawrence H. Summers
365
Table 9 presents both actual and adjustedunemploymentrates for
each state for 1984, the most recent year for which it was possible to
computeadjustedunemploymentrates. The strikingfeatureof the data
is the similaritybetween the actual and adjustedunemploymentrates.
The correlationof the two variablesis 0.84. Whilethe adjustmentsgo in
the expected direction,reducingunemploymentin the Rust Belt states,
for example, by recognizingthe poor performanceof manufacturingin
recent years, they are not large. Before adjustmentfor industry and
occupation, the difference between the unemploymentrate in Massachusetts and that in Ohio was 4.6 percentagepoints; after adjustment,
the difference fell to 3.4 points. Only a relatively small fraction of
differences in unemploymentrates among states can be explained by
differencesin the characteristicsof workersor jobs. This conclusion is
robust. It holds for other years, for changes as well as levels of
unemployment,andfor employmentas well.
This findingsuggests that much of the difference in unemployment
rates reflects differences across states in the performanceof given
industriesratherthandifferencesin the industrialcompositionof states.
Or it could reflect differencesin labor marketconditionsthat influence
the willingness to supply labor. Table 10 presents estimates of the
relationshipbetween employmentgrowth,its components,andchanges
in unemploymentover various sampleperiods. The first column of the
tablepresentsevidence on the relationshipbetweenoverallemployment
growth and unemployment.While the relationshiphas the expected
negative sign, it is surprisinglyweak. For example, during 1970-85, a
hypotheticalstate that experienced a 30 percent, or 2 percent a year,
growthin employmentwould have enjoyed only a 0.5 percentagepoint
declineinunemployment.Overshorterperiods,therelationshipbetween
employmentgrowthand changes in unemploymentis somewhattighter
but still not strong.Between 1981and 1985,a state that experiencedan
extra 2 percent a year of employment growth would have seen its
unemploymentrate decline by only 0.7 percent.30
30. Employmentgrowthwill fail to lead to equalpercentagereductionsin unemployment if it is associated either with populationgrowth or with increases in labor force
participation.Resultsnotreportedhereindicatethatemploymentgrowthratesarestrongly
associatedwithpopulationgrowthacross statesandonly weaklyassociatedwithchanges
in laborforce participation.
Brookings Papers on Economic Activity, 2:1986
366
Table 9. Actual and Adjusted State Unemployment Rates, 1984
Percent
State and region
Actual rate
Adjusted ratea
New England
Maine
New Hampshire
Vermont
Massachusetts
Rhode Island
Connecticut
6.1
4.3
5.2
4.8
5.3
4.6
6.6
6.2
6.8
5.6
6.9
5.8
7.2
6.2
9.1
6.4
6.7
8.2
9.4
8.6
9.1
11.2
7.3
9.0
7.7
9.2
10.0
8.0
6.3
7.0
7.2
5.1
4.3
4.4
5.2
5.2
7.9
8.0
6.5
5.1
4.6
6.6
6.2
5.4
9.0
5.0
15.0
6.7
7.1
6.0
6.3
6.4
6.3
8.2
6.0
15.6
5.5
5.1
5.2
6.2
9.3
8.6
11.1
10.8
8.8
8.1
7.8
7.7
Mid-Atlantic
New York
New Jersey
Pennsylvania
East North Central
Ohio
Indiana
Illinois
Michigan
Wisconsin
West North Central
Minnesota
Iowa
Missouri
North Dakota
South Dakota
Nebraska
Kansas
South Atlantic
Delaware
Maryland
Districtof Columbia
Virginia
West Virginia
North Carolina
South Carolina
Georgia
Florida
East South Central
Kentucky
Tennessee
Alabama
Mississippi
Lawrence H. Summers
367
Table 9. (Continued)
Percent
State and region
Actual rate
Adjusted ratea
8.9
10.0
7.0
8.3
9.6
6.9
5.9
4.9
7.4
7.2
6.3
5.6
7.5
5.0
6.5
7.8
8.5
7.2
8.9
6.7
8.3
5.3
7.8
7.2
9.5
9.4
7.8
10.0
5.6
8.4
9.6
6.8
9.3
4.3
West South Central
Arkansas
Louisiana
Oklahoma
Texas
Mountain
Montana
Idaho
Wyoming
Colorado
New Mexico
Arizona
Utah
Nevada
Pacific
Washington
Oregon
California
Alaska
Hawaii
Sources: Actual unemploymentrates are from U.S. Bureauof the Census, StatisticalAbstractof the United
States, 1986(GPO, 1986),p. 409. Adjustedunemploymentrates were calculatedby the authorusingthe May 1984
CPS.
a. Adjustedunemploymentrates are computedrelativeto Washington,D.C., and are then scaled so that the
averageadjustedunemploymentrateequalsthe nationalaverageunemploymentrate.
The empiricalfindingthat changes in state unemploymentrates are
only weakly relatedto total employmentgrowthis vividlyillustratedby
Massachusetts, which in 1985 had the nation's lowest unemployment
rate. While the "MassachusettsMiracle" has been widely discussed,
the datain table 11reveal thatemploymentgrowthin Massachusettshas
actuallybeen below nationalemploymentgrowthover the past decade
despite the state's 5.5 percentagepoint reductionin its unemployment
rate.3"In an arithmeticsense, the apparentsuccess of the Massachusetts
economy is less the resultof job creationthanof circumstancesthat led
to relativelyslow laborforce growth.
Whilethe relationshipbetweentotalemploymentgrowthandchanges
in unemploymentis weak, the second and third columns of table 10
31. Fora comprehensivediscussionof the Massachusettsexperience,see HelenLadd
andRonaldFerguson,"Massachusetts'EconomicDevelopment:A CaseStudy,"Working
Paper(HarvardUniversity,KennedySchool of Government,1986).
368
Brookings Papers on Economic Activity, 2:1986
Table 10. Employment Growth and Changes in Unemployment,
Various Periods, 1970-85a
Employment change decomposed
Interval
Percentage
change in
total
employment
Change in
high-wage
employment as
a percentage
of total
employment
Change in
non-high-wage
employment as
a percentage
of total
employment
1970-85
-0.017
(0.008)
...
...
1975-85
-0.035
(0.026)
-0.073
(0.027)
-0.083
(0.026)
- 0.087
(0.026)
-0.082
(0.029)
-0.151
(0.047)
-0.167
(0.045)
-0.168
(0.044)
-0.145
(0.045)
-0.144
(0.051)
0.037
(0.035)
0.000
(0.038)
-0.002
(0.043)
- 0.025
(0.046)
-0.017
(0.053)
-0.057
(0.032)
-0.083
(0.038)
-0.150
(0.054)
-0.126
(0.062)
-0.227
(0.062)
-0.139
(0.057)
-0.147
(0.064)
-0.326
(0.074)
-0.260
(0.096)
-0.182
(0.134)
0.004
(0.047)
-0.036
(0.049)
-0.049
(0.059)
-0.076
(0.066)
- 0.232
(0.062)
1976-85
1977-85
1978-85
1979-85
1980-85
1981-85
1982-85
1983-85
1984-85
Source:Author'scalculationsusingdatafromBLS, Employment and Earnings, variousissues.
a. Dependentvariableis the changein unemployment,regressedon a constantand the percentagechange in
employmentof nonagricultural
wage and salaryworkers(firstcolumn)and alternativelyas the percentagechange
decomposedinto high-wageand non-high-wageemployment(last two columns).High-wageemploymentis defined
as employmentin manufacturing,
and publicutilities.Standarderrorsare
mining,construction,and transportation
in parentheses.
indicate that the relationshipbetween growth in employmentin highwage industries-manufacturing,construction,mining,and public utilities-and unemploymentis significantlystrongerthan the relationship
between overall employmentgrowth or growth in low-wage industries
and unemployment.32
The estimates in the last two columns come from
32. See Alan B. Kruegerand Lawrence H. Summers, "EfficiencyWages and the
Wage Structure,"WorkingPaper 1952(NationalBureauof Economic Research, June
1986),for an examinationof the level of wages in differentindustriescontrollingfor the
Lawrence H. Summers
369
Table 11. Employment Growth and Unemployment in Massachusetts and the
United States, Various Periods, 1976-85
Percent unless otherwise indicated
Change in
ratio of
high-wage
employment
to total
employmenta
Change in
unemployment
(percentage
points)
Change in
employment
Massachusetts
1976-80
1980-85
1976-85
- 3.9
- 1.6
-5.5
9.7
8.4
18.9
7.2
0.3
7.5
United States
1976-80
1980-85
1976-85
-0.6
0.1
-0.5
11.9
7.9
20.7
3.2
-0.5
2.7
Period
Source:Datafor UnitedStatesare fromBLS, Handbookof LaborStatistics(June1985),and, for 1984and 1985,
from Employment and Earnings, vol. 33 (January 1986). Massachusetts data are from Statistical Abstract of the
United States, 1986 and earlier issues.
a. High-wageemploymentis definedas employmentin manufacturing,
mining,construction,and transportation
and publicutilities.
separateregressions in which the decomposed employmentchange is
substitutedfor the total. Creatingor avoidingthe loss of "good" highwage jobs appearsto be more potent in reducingunemploymentthan
creatinglow-wagejobs. Furtherestimates not reportedhere that allow
for a nonlinearrelationshipbetween changes in high-wageemployment
and unemployment suggest that the loss of high-wagejobs has an
especially large impact on unemployment. During 1979-85, a fairly
representativeperiod, every one hundredhigh-wagejobs that were lost
raisedunemploymentby twenty-fiveworkers. In contrast, the creation
of high-wagejobs had only a minorimpacton unemployment.The data
in table 10 also reveal that over periodslonger than a single year, there
is essentially no relationshipbetween growthin low-wage employment
andunemployment.
As table 11indicates,whileMassachusettsdidnot experienceunusual
growthin totalemployment,its high-wageemploymentgrowthexceeded
differentcharacteristicsof theirworkers.We estimatewage premiumsof 12 percentfor
manufacturing,12 percent for construction,25 percent for mining,and 25 percent for
public utilities. Very similarresults are obtained using data on workers who change
industries.
370
Brookings Papers on Economic Activity, 2:1986
thatof the rest of the country-though not by enoughto accountfor the
extraordinaryperformanceof its economy.
These findingson the relationshipbetween changes in employment
andunemploymentareinstructive.They suggestthatin analyzingrecent
changesin unemploymentin the United States, it is not enoughto focus
on the determinationof the total level of employment.33It is also
necessary to examine the composition of employmentgrowth and to
considerthe incentives individualsmay have to remainunemployed.
WHAT IS INVOLUNTARY
UNEMPLOYMENT?
As countless analysts have pointed out, the notion of invol-dntary
unemploymentinvolves importantlogical difficulties. The argument
usually goes somethinglike this: virtuallyeveryone counted as unemployed could findsome type ofjob at some wage; even if not, the option
of self-employmentis surelyopen; in the sense thatthereis some option
open to all the unemployed, there is a voluntary component to all
unemployment.34Carefulcritics of the concept of involuntaryunemployment are quick to stress that labelingunemploymentas voluntary
does not make it benignor socially inconsequential.But they do stress
that a proper analysis of its causes requires recognizingits voluntary
element.
The standardresponse to this line of argumentis usually to conjure
up images of the GreatDepression, to highlightthe personaland social
costs of unemployment,and then to take refuge in some notion that
unemploymentis involuntary only when "reasonable" jobs are not
available.Withoutsome specificationof whatis meantby a "reasonable"
job, the concept of involuntaryunemploymentis vague, but, at the same
time, it does seem to capture an importantaspect of what many see
happeningover the course of cyclical fluctuations.
Perhapsthe most coherent set of attempts to justif the concept of
involuntaryunemploymentrelies on some notion of segmented labor
33. Increasingunemploymentin Europehas been associated with a cessation of job
creationdistinguishingit sharplyfromthe UnitedStates.
34. For perhapsthe best-knownrecent attack on the concept of involuntaryunemployment, see Robert E. Lucas, Jr., "UnemploymentPolicy," American Economic
Review, vol. 68 (May 1978, Papers and Proceedings, 1977), pp. 353-57.
Lawrence H. Summers
371
markets.35Where employed workers of a given ability do not receive
equal compensation,a meaningfuldefinitionof involuntaryunemploymentis possible. A workermay be definedas involuntarilyunemployed
if he is unableto get ajob at a wage that other workersof his abilityare
receiving, even if he could get an alternativelower-wagejob. If labor
marketsaresegmentedso thattherearedifferencesinemployedworkers'
compensationunrelatedto differences in their ability, it is possible to
observeunemploymentthathas bothvoluntaryandinvoluntaryaspects.
It is voluntary in the sense that unemployed workers decline some
opportunitiesto work. But it is involuntaryin the sense thatothers with
the sameabilityas the unemployedareworkingat wages the unemployed
would be willing to accept. Segmented labor markets raise another
possibilityas well. Some of the unemployedmay preferlow-wagejobs
to being unemployed, but choose to remain unemployed in order to
queuefor high-wagejobs.
Any explanationof involuntaryunemploymentthat relies on labor
marketsegmentationmust account for the differencesin the wages of
equallyskilledworkersin differentindustries.More specifically,it must
explain why high-wageemployers who face an excess supply of labor
do not reducewages. A convincingsegmented-marketinterpretationof
unemploymentshouldalso be ableto explainwhy workerswouldchoose
to remainunemployedin order to wait for high-wagejobs, ratherthan
wait while workingat lower-payingpositions. I take up these issues in
turn.
There are three broadclasses of explanationsfor the failureof highwage employersto reduce theirwages in the face of an excess supplyof
labor.The most obvious is that there are institutionalimpedimentsthat
make it impossible. Unions are one such impediment;regulationsare
35. For argumentsalong the lines sketchedhere, see JohnR. Harrisand MichaelP.
Todaro,"Migration,UnemploymentandDevelopment:A Two SectorAnalysis,"American EconomicReview, vol. 60 (March1970),pp. 126-42; RobertE. Hall, "The Rigidity
of Wages and the Persistenceof Unemployment,"BPEA, 2:1975, pp. 301-35; Ian M.
McDonaldand Robert M. Solow, "Wages and Employmentin a Segmented Labor
Market," Quarterly Journal of Economics, vol. 100 (November 1985), pp. 1115-41; and
JeremyI. Bulow and LawrenceH. Summers,"A Theory of Dual Labor Marketswith
Applicationto IndustrialPolicy,Discrimination,andKeynesianUnemployment,"Journal
of LaborEconomics,vol. 4 (July1986),pp. 376-414.
372
Brookings Papers on Economic Activity, 2:1986
another.36But even in the absence of these institutions, there are
substantialdifferentialsin the wages differenttypes of employerspay to
similarworkers.37
A secondclass of explanationsforwage differentials-labeled efficiencywage theories-holds thatfirmsfindit profitableover
some rangeto increasetheirwages even in the face of an excess supply
of labor. By paying higherwages, firmsenhance productivitythrough
improvedwork-forcemorale, reduced turnoverand hiringcosts, and
increasedworkereffort.38A thirdclass of explanations-insider-outsider
theories-involves the notionof rentsharingbetweenworkersandfirms.
Because hiringand trainingnew workersis costly, incumbentworkers
have leverage and so are able to capturea share of the rents that firms
earn. Firmsare thereforeunableto reduce wages even in the face of an
excess supply of labor. An importantpiece of evidence to this effect is
that high-wageindustriesand high-wagefirms tend to pay all types of
workershighwages.39
All threeclasses of explanationsforthefailureof high-wageemployers
to reduce wages when laborsupplyis excessive supportthe plausibility
of segmentedlabormarketsandthusexplainthe existence of involuntary
unemploymentas I have definedit. But on the argumentsdeveloped so
far,unemploymentexists only becausethe unemployedpreferremaining
unemployedto acceptingworkin low-wageindustries.Giventhe general
empiricalfindingthat labor supply is relatively inelastic, it is unlikely
36. For the importanceof regulation,see Hall, "The Rigidityof Wages"; for the
effects of unions,see McDonaldand Solow, "WagesandEmployment."
37. This point has been recognizedby institutionallaboreconomistsfor many years.
A recent review of the evidence on wage differentialsmay be foundin Alan B. Krueger
andLawrenceH. Summers,"Reflectionson theInter-industry
WageStructure,"Working
Paper1968(NationalBureauof EconomicResearch,June 1986).Similarconclusionsare
reached in WilliamDickens and Lawrence Katz, "Industryand OccupationalWage
PatternsandTheoriesof WageDetermination"(HarvardUniversity,1986).
38. Foranexcellentsummaryof variousefficiencywagetheoriesanda strongargument
for their relevance to macroeconomics,see Joseph E. Stiglitz, "Theories of Wage
Rigidity,"inJamesL. Butkiewicz,KennethJ. Koford,andJeffreyB. Miller,eds., Keynes'
Economic Legacy: Contemporary Economic Theories (Praeger, 1986), pp. 153-206. For
a surveyof some of the relevantempiricalliterature,see LawrenceF. Katz, "Efficiency
WageTheories:A PartialEvaluation,"in StanleyFischer, ed., NBERMacroeconomics
Annual 1986 (MIT, 1986), pp. 235-76.
39. This findingis reportedby WilliamDickens and LawrenceKatz, "Interindustry
WageDifferencesandIndustryCharacteristics"(HarvardUniversity,1986);andby Erica
Lynn Groshen, "Sources of Within IndustryWage Dispersion: Do Wages Vary By
Employer?"(Ph.D. dissertation,HarvardUniversity,1986).
Lawrence H. Summers
373
that a large numberof workerswill be willing to work at high- but not
low-wagejobs. A more compellingexplanationof involuntaryunemploymentwould explain why workers choose to forgo low-wage work
in orderto seek high-wagework.
In theirpaperon unemploymentin less developed countries, Harris
and Todaro offered a very plausible answer to this question. They
explainedthat the high-wagejobs were in the city while the low-wage
jobs were in ruralareas. It was thus impossibleto queuefor a high-wage
job while holdinga low-wagejob. The marketequilibratedwhen unemployment in the city was sufficientlyhigh and the chance of getting a
high-wagejob in the city sufficientlysmall that workers would opt for
the certaintyof a low-wagejob in the country.
Suchanexplanationis notplausiblefordevelopedeconomies.Perhaps
the most plausible explanationfor what has been called "transitional
unemployment"is that workers who have lost high-wagejobs find it
difficult to accept their fate and so prefer remainingunemployed to
acknowledgingthe permanenceof theirloss by takinga low-wagejob.40
In a society where status is highlyboundup with one's occupation,it is
to be expected that workerswho lose attractivehigh-wagejobs will be
reluctantto accept lesserjobs.
Also, there are fixed costs for workersas well as firmsin enteringan
employmentrelationship,so that workerswho expect to returnto highwagejobs in a relatively short time may find it difficultor undesirable,
or both, to take ajob at a low-wagefirm.Somethingof this sort mustlie
behind firms' reluctanceto hire "overqualifiedworkers." The unemployed may also feel thatacceptinga low-wagejob suggeststo potential
high-wageemployers that they are not qualifiedfor betterjobs and so
reduces theirchance of gettingthem. Finally, in some circumstancesit
maybe moreefficientto searchwhile remainingunemployedthanwhile
working.
Whilethe idea of transitionalunemploymentcan easily be criticized
by pointing to the costs to workers of remainingunemployed, the
empiricalobservationthat total employmentgrowthin a given state has
only a very limitedimpacton unemploymentdoes suggestthat a theory
40. I borrow the term "transitionalunemployment"from McDonaldand Solow,
"SegmentedLaborMarket."I use it to referto the unemploymentof workerstransiting
in bothdirectionsbetweenthe high-andlow-wagesectorsof the economy.
374
Brookings Papers on Economic Activity, 2:1986
of transitionalunemploymentis preferableto simple theories based on
wage rigidities. If involuntary unemployment were caused by rigid
wages, one would expect it to be sharplyreducedby movementsin the
demandfor labor, and thus employment,in a given state. The fact that
it is not makes it worthwhileat least to explore labor supply aspects of
the determinationof unemployment.The empiricalfindingthat reductions in high-wageemploymentincrease unemploymentand that this
impactis not easily offset by growthin low-wage employmentsuggests
thatthe transitionalunemploymentsufferedby personslosinghigh-wage
jobs is significant.
UNIONS
AND UNEMPLOYMENT
To understandthe causes of the high and rising normal level of
unemployment,it may be desirableto focus on the factors influencing
the extent of "transitionalunemployment." Without embracingany
theoryof the cause of transitionalunemployment,the precedingdiscussion suggests that its extent is likely to be determinedby the size of the
wage differentialsbetween high- and low-wagejobs, the availabilityof
high-wagejobs,andthe costs, pecuniaryandnonpecuniary,of remaining
unemployed. Althoughall these factors are difficultto quantify,it is a
reasonable conjecturethat in areas where the level of unionizationis
high, ceteris paribus,there shouldbe more transitionalunemployment.
This is especially the case when, as in recent years, the economy is
subjected to large intersectoral shocks. High and rising union wage
premiumsare likely to cause job losses in the unionized sector of the
economy andalso to makethose who lose high-wagejobsmorereluctant
to accept alternative lower-wage employment. The empirical work
presented in table 12 examines the conjecture that unions increase
unemployment.
In investigatingthe relationshipbetweenunionizationandunemployment, the criticalempiricalproblemis eliminatingotherfactorsthatmay
be correlatedwith both. Most obviously, the high-wagesector of the
economy tends to be more highly unionized than other parts of the
economyandin recentyears has sufferedhighunemployment.I address
this issue in several ways. First, I estimate the relationshipbetween
unionizationandbothactualunemploymentratesandthe unemployment
rates adjusted for changes in the composition of the labor force, as
375
Lawrence H. Summers
Table 12. Unionization and State Unemployment Rates, 1970-85a
Year
Intercept
1970
3.43
(0.493)
4.97
(0.681)
5.55
(1.06)
3.23
(0.784)
5.18
(1.35)
0.044
(0.019)
0.059
(0.023)
0.053
(0.028)
0.053
(0.032)
0.048
(0.041)
5.45
(0.750)
4.59
(1.62)
3.46
(2.01)
6.31
(1.20)
3.60
(2.60)
0.085
(0.036)
0.081
(0.038)
0.117
(0.045)
0.040
(0.061)
0.145
(0.085)
1.67
(0.666)
0.10
(0.805)
-1.57
(1.22)
3.05
(1.11)
-0.77
(1.73)
0.039
(0.026)
-0.009
(0.027)
0.062
(0.033)
-0.019
(0.045)
0.063
(0.053)
1970
1970
1970
1970
1985
1985
1985
1985
1985
1970-85
1970-85
1970-85
1970-85
1970-85
Union
Share in
highwage
industries
Region
dummies
Instrumental
variables
R2
...
No
No
0.096
-0.063
(0.022)
-0.027
(0.036)
...
No
No
0.231
Yes
No
0.540
No
Yes
0.093
...
Yes
Yes
0.561
...
No
No
0.309
No
No
0.099
Yes
No
0.554
...
No
Yes
0.073
...
Yes
Yes
0.549
...
No
No
0.044
No
No
0.323
Yes
No
0.609
...
No
Yes
- 0.054
...
Yes
Yes
0.585
0.035
(0.053)
0.042
(0.057)
0.106
(0.027)
0.034
(1.47)
Source: Author's calculationsbased on data from the following sources. Unemploymentand manufacturing
employmentdatafor 1970are fromU.S. Bureauof the Census,Censusof Population,1970,vol. 1, Characteristics
of the Population,section 1, U.S. Summary(GPO, 1972),pp. 1-469. Unionizationrates for 1970are from BLS,
Directory of National Unions and Employee Associations 1979, BLS Bulletin 2079 (September 1980), p. 109.
andmanufacturing
Unemployment
employmentfor 1985arefromEmploymentand Earnings,vol. 33 (January1986).
Unionizationratesusedin 1985regressionsare 1982ratesobtainedfromLeo TroyandNeil Sheflin,UnionSourcebook
(WestOrange,New Jersey, IndustrialRelationsDataand InformationServices, 1985),table7.2, p. 7-4, as no more
recentunionizationratescouldbe obtained.
a. Dependentvariableis the unemploymentrateregressedon a constantandthe level of unionizationand, where
indicated,the percentof workersin high-wageindustries(manufacturing,
construction,mining,and transportation
and public utilities), region dummies,and a dummyinstrumentalvariablefor a state with a right-to-worklaw.
Regionaldivisionscorrespondto the nine U.S. Censusdivisions.The regressionexcludes the Districtof Columbia
since no independentunionizationfiguresfor D.C. were publishedin 1970.Standarderrorsare in parentheses.
376
Brookings Papers on Economic Activity, 2:1986
described earlier. Because the results are broadly similarfor the two
concepts and because data on officialunemploymentare availableover
alongertimespan,only resultsusingofficialunemploymentarereported.
Second, in severalof the specificationsreportedin table 12, I controlfor
the shareof the high-wagesector (or alternativelythe shareof manufacturing)in total employmentand for regions in estimatingthe effect of
unionizationon unemployment.Third, I treat unionizationas endogenous and use the presence or absence of a right-to-worklaw as an
instrument.Nineteen states have such laws, almost all of which were
put in place before 1960,so it is plausibleto take the presence of a rightto-work law as exogenous. A numberof investigatorshave found that
right-to-worklaws have a significanteffect on unioncoverage.41
The results in table 12 supportseveral conclusions. First, there is a
clear and substantively significant impact of unionization on state
unemploymentrates. The estimates for 1985, controllingfor both the
region and the share of employmentin high-wageindustries, suggest
that an increase of 10 percentagepoints in a state's unionizationrate
increasesits unemploymentrateby 1.2percentagepoints. Because there
aresubstantialregionaldifferencesin the degreeof unionization(in 1982,
Texas, with 12.5 percent of its work force unionized, was the fortieth
most unionized state, while Pennsylvania, with 27.0 percent, ranked
tenth), those differences can account for quantitativelyimportantregional variations in the extent of unemployment. Second, there is
suggestive evidence that the impact of unionizationon unemployment
has increased over time. The estimated equation for the change in
unemploymentbetween 1970and 1985,holdingconstantthe high-wage
shareof employmentand region, suggests that a state with a 20 percent
unionizationrate, approximatelythe sample average, experienced an
increase in unemploymentof 1.2 percentagepoints relative to a hypothetical state that had no unions. In this sense, a significantpart of the
observed increase in normal unemploymentin recent years may be
attributedto the effects of unions.42
41. See, for example, Henry S. Farber, "Right-to-WorkLaws and the Extent of
Unionization," Journal of Labor Economics, vol. 2 (July 1984), pp. 319-52.
42. Theincreaseis partiallyoffsetby thedeclineinunioncoverageof about4percentage
pointsbetween 1970and 1982.The coefficientestimatesfor 1985implythatthe declinein
union membershipreduced the unemploymentrate by about 0.5 percentagepoint. Of
course,to the extent thatunioncoveragedeclinedbecause unionmemberswere laid off,
the declinein unionmembershipmay, over a long transitionperiod,actuallyhave further
increasedunemployment.
Lawrence H. Summers
377
In their widely read book WhatDo Unions Do?, RichardFreeman
andJamesMedoffestimatethathighlyunionizedstates have on average
an unemploymentrate that is 1 percentage point higher than that of
"low" union states.43That findingparallelsthe one reportedhere. But
they reportbeing unable to find any relationshipbetween unionization
and the fraction of the populationemployed in a state, and they infer
from that that unions may draw workers into the labor force but that
they do not reduceemployment.My own findingsdiffer.Table 13reports
estimatesof the impactof unionizationon the employmentratio. While
the impactis estimatedless precisely thanthe impactof unionizationon
the stateunemploymentratesin table 12,the resultsstronglycorroborate
the conclusions reached using unemploymentdata. In fact, in most
specifications, the impact of unionizationon the employment ratio is
greaterthan its impact on the unemploymentrate. For example, when
the region and the share of employment in high-wage industries are
controlledfor, the data suggestthatin 1985an increaseof 10percentage
points in the fractionof the workforce thatwas unionizedwould reduce
the employmentratioby almost4 percent. Likewise, in most but not all
specifications, the impact of unionization on the employment ratio
increased by more between 1970 and 1985 than did the impact of
unionizationon unemployment.
In results not reportedhere, I have explored the robustnessof these
conclusionsin a numberof ways. First, I have estimatedthe relationship
between unionization,unemployment,and the employmentratio using
datafor every year between 1970and 1985.The dataconfirmthe upward
trend in the impact of unionization. In fact, the trend appears more
dramaticwhen results for the early 1970sare comparedwith those for
the early 1980s. Second, I have estimated the effects of unions on
unemploymentratesseparatelyfor maleandfemaleworkers.Theresults
indicatethatunionshave a somewhatgreaterimpacton maleunemployment. Third, I have reestimatedthe equationsafter combiningsmaller
states into largerunits as was done by the CPS in the early 1970s. The
reestimatehas little impacton the results. Furthercorroborationfor the
conclusions reached here comes from the work of other investigators
usingdataon metropolitanareasratherthanstates. Forexample,Edward
Montgomery,using data on forty-fourstandardmetropolitanstatistical
43. RichardFreemanand JamesMedoff, WhatDo UnionsDo? (Basic Books, 1984),
pp. 120-21.
378
Brookings Papers on Economic Activity, 2:1986
Table 13. Unionization and State Employment Ratios, 1970-85a
Year
Intercept
1970
4.021
(0.027)
4.024
(0.040)
4.170
(0.064)
3.998
(0.043)
4.170
(0.088)
-0.024
(0.107)
0.081
(0.133)
-0.329
(0.169)
0.075
(0.177)
-0.379
(0.267)
4.123
(0.029)
4.192
(0.061)
4.175
(0.083)
4.098
(0.046)
4.192
(0.106)
-0.193
(0.139)
-0.193
(0.143)
-0.363
(0.188)
-0.063
(0.232)
-0.387
(0.349)
0.114
(0.021)
0.171
(0.029)
0.055
(0.044)
0.100
(0.033)
0.000
(0.062)
-0.186
(0.082)
-0.116
(0.098)
0.021
(0.118)
-0.128
(0.137)
0.083
(0.188)
1970
1970
1970
1970
1985
1985
1985
1985
1985
1970-85
1970-85
1970-85
1970-85
1970-85
Union
Share in
highwage
industries
Region
dummies
Instrumental
variables
R2
...
No
No
0.001
-0.139
(0.132)
0.025
(0.213)
...
No
No
0.032
Yes
No
0.393
No
Yes
...
Yes
Yes
0.315
...
No
No
0.038
-0.258
(0.200)
0.042
(0.237)
...
No
No
0.071
Yes
No
0.451
No
Yes
0.021
...
Yes
Yes
0.447
...
No
No
0.096
-0.252
(0.097)
-0.145
(0.147)
...
No
No
0.252
Yes
No
0.577
No
Yes
0.087
...
Yes
Yes
0.487
-0.017
Source:Author'scalculationsusingdatafromthe followingsources.Manufacturing
employmentdatafor 1970are
from Census of Population, 1970, U.S. Summary, pp. 1-469. Laborforce participationdata are from the same
volume,pp. 1-350. Unionizationratesfor 1970are fromDirectory of National Unions and Employee Associations,
1979, p. 109. Employmentdata for 1985were obtainedfrom Employment and Earnings, vol. 33 (January1986).
Populationby stateis the averageof the 1984populationover age fourteenandthe 1984populationover age eighteen
multipliedby the growthrateof total populationbetween1984and 1985.Unionizationratesused in 1985regressions
are 1982ratesobtainedfromTroy and Sheflin,Union Sourcebook, since no morerecentunionizationratescouldbe
obtained.
a. Dependentvariableis the log of the employmentratioregressedon a constant,the level of unionization,and,
whereindicated,the percentof workersin high-wageindustries(manufacturing,
construction,mining,and transportation and publicutilities),regionaldummies,and a dummyinstrumentalvariablefor a state with a right-to-work
law. Regionaldivisions correspondto the nine U.S. Census divisions. The regressionexcludes the District of
Columbiasince no independentunionizationfigures for D.C. were publishedin 1970. Standarderrors are in
parentheses.
Lawrence H. Summers
379
Table 14. Annual Percentage Changes in the Employment Cost Index,
by Union Status, 1973-85
Employment cost index
Cumulative
difference
Year
Union
Nonunion
1973a
1974a
1975a
5.7
7.5
8.6
5.5
8.0
6.0
-0.3
2.3
1976
1977
1978
1979
1980
8.1
7.6
8.0
9.0
10.9
6.8
6.6
7.6
8.5
8.0
3.6
4.6
5.0
5.5
6.6
1981
1982
1983
1984
1985
9.6
6.5
4.6
3.4
3.6
8.0
6.1
5.2
4.5
5.1
8.2
8.6
8.0
6.9
5.4
0.2
Source:RichardB. Freeman,"In Searchof UnionWageConcessionsin StandardDataSets," IndustrialRelations,
vol. 25 (Spring1986),table4, p. 139.
a. Estimatedfromchangesin majorcontractsettlements.
areas in 1983, finds a statistically significant,though not substantially
large, negativeimpactof unionizationon employment."4
The conclusionthat the impactof unionson unemploymenthas been
increasing is not surprisinggiven that the spread in wages between
unionizedand nonunionizedworkers has also increased, at least until
recently.Table 14presentssome informationon changesin the employment cost index for unionizedand nonunionizedworkersduring197385 and shows that unionwage premiumsincreasedduringthe 1970sand
have declined somewhat, but not enough to reverse their previous
increase, duringthe 1980s.Analyses using survey data on the wages of
individualsalso suggestthatunionwage premiumsrose duringthe 1970s
but find less evidence of a decline in the 1980s than is suggested by
newspaperheadlinesand the employmentcost index.45
44. Edward Montgomery, "The Impact of Regional Difference in Unionism on
Employment," Economic Review of the Federal Reserve Bank of Cleveland, 1:1986,
pp.2-11.
45. See RichardB. Freeman, "In Search of Union Wage Concessions in Standard
DataSets," IndustrialRelations,vol. 25 (Spring1986),pp. 131-45; and Peter Linneman
and MichaelWachter,"Rising Union Premiumsand the DecliningBoundariesAmong
Noncompeting Groups," American Economic Review, vol. 76 (May 1986, Papers and
Proceedings, 1985), pp. 103-08.
380
Brookings Papers on Economic Activity, 2:1986
The coincidence of rising union wage premiumsand an increasing
impact of unionizationon state unemploymentrates, along with the
widely observed decline in employment growth in unionized firms,
makes it plausiblethat unionpower has accountedfor a significantpart
of the increase in normalunemploymentin recent years. The fact that
the loss of unionizedjobs resultedin increased unemploymentdespite
the rapidcreationof jobs in the low-wage service sector provides some
supportfor the "transitional"theoryof unemploymentadvocatedhere.
FURTHER
EVIDENCE
ON TRANSITIONAL
UNEMPLOYMENT
The discussion so far suggests that increasingunion wage premiums
duringthe 1970scontributedto the risingrate of normalunemployment
by causing an increase in transitionalunemployment.Transitionalunemploymentis a likely concomitantof any increasein the importanceof
noncompetitivewage differentials,whethercaused by unions or by the
efficiencywage andrent sharingconsiderationsdiscussed above.
Table 15, which is drawnfrom the work of Linda Bell and Richard
Freeman, presents several different estimates of the extent of wage
dispersion in the economy during 1970-85. Each of the measures
indicates that wage dispersionhas increased.46Rising wage dispersion
does not necessarilyindicatean increase in the importanceof noncompetitive wage differentials. It could occur because increases in the
demandfor labor in high-wageindustriesmoved firms along upwardslopingshort-runlaborsupplycurves. However, using severaldifferent
datasources, Bell andFreemanfindthatthe correlationacrossindustries
between employmentgrowth and wage growth over the decade of the
1970s was negative, which suggests that shocks in the wage-setting
process thatmovedfirmsalongtheirlabordemandcurvespredominated.
Withoutinvokingthe considerationsleadingto labormarketsegmentation, noted above, it is difficultto accountfor these shocks.
It is likely that efficiency wage or rent-sharingconsiderationsled to
increases in noncompetitivewage differentialsduringthe 1970s. This
inferenceis supportedby evidence that the gap between the wages paid
by smallfirmsand those paid by large firmsincreasedduringthe 1970s
46. LindaAnnBellandRichardB.Freeman,"DoesaFlexibleIndustryWageStructure
Increase Employment?:The U.S. Experience," in LindaAnn Bell, "Essays on Labor
MarketEfficiency"(Ph.D. dissertation,HarvardUniversity,May 1986).
Lawrence H. Summers
381
Table 15. Dispersionin Wagesand Compensation,1970-85
Standarddeviationof log
Year
Average
hourly
earnings in
manufacturing
National
income and
product
accounts
compensation
Census of
manufacturers
wages
1970
1971
1972
1973
1974
0.215
0.226
0.237
0.240
0.241
0.255
0.266
0.278
0.280
0.285
0.221
0.222
0.237
0.242
0.240
1975
1976
1977
1978
1979
0.253
0.257
0.258
0.267
0.270
0.303
0.311
0.316
0.319
0.324
0.247
0.252
0.260
0.269
0.279
1980
1981
1982
1983
1984
0.270
0.277
0.282
0.286
0.291
0.335
0.339
0.349
n.a.
n.a.
0.282
n.a.
n.a.
n.a.
n.a.
1985
0.293
n.a.
n.a.
Source: Linda Ann Bell and Richard B. Freeman, "Does a Flexible Industry Wage Structure Increase Employment?:
The U.S. Experience," in Linda Ann Bell, "Essays on Labor Market Efficiency," (Ph.D. dissertation, Harvard
University, May 1986), table 1, p. 51.
n.a. Not available.
even after adjustmentfor unionization.47Given the empiricalevidence
presentedabove regardingthe impactof unionizationon employment,
it seems plausiblethatthe generalincreasein labormarketsegmentation
over the past fifteen years has tended to raise the normal rate of
unemployment,thoughthe propositionis difficultto test.
Finally, the transitionalunemploymentexplanationfor risingunemployment is also consistent with the informationon the nature of the
increasein unemploymentpresentedin the firstpart of this paper. It is
most plausiblyjob losers who would wait to regain high-wagejobs.
Investing in waiting for a high-wagejob makes much more sense for
mature marriedmen, who as a group have a very low employment
turnoverrate, thanfor otherdemographicgroupsthathave muchhigher
47. Nicole Gerris,"TheChangingSize WageEffect" (Undergraduate
thesis, Harvard
University,1983).
382
Brookings Papers on Economic Activity, 2:1986
turnoverrates. Personslosing high-wagejobs are most likely to experience protractedspells of unemployment.Sectoralshocks leadingto the
loss of high-wagejobs would also lead to plant shutdowns, reducing
capacity and thereby raisingcapacity utilization. In addition, sectoral
shocks that hit at high-wage industries could easily account for the
change in the vacancies-unemploymentrelationship,if job losers were
reluctantto accept low-wageemploymentin expandingsectors.
Conclusions
The analysis in this paper suggests that the rise in normalunemployment over the past twenty years represents a serious problem. The
additionalunemploymentis concentratedamong maturemarriedmen
who have lost jobs and are likely to be out of work for periods of six
months or more. Increased unemployment cannot be convincingly
dismissed as the consequence of marginallabor force attachmentor
measurementproblemsin the CPS. Nor is it simplythe resultof cyclical
weakness in the economy. Persistently high unemploymenthas coincided with relativelyhighvacancy and capacityutilizationrates.
These conclusions have importantimplicationsfor economic policy.
First, they suggest that while high unemploymentis a serious problem,
expansionaryaggregatedemandpoliciesareunlikelyto be ableto reduce
it to the levels of the 1950s and 1960s without creating excessive
inflationarypressures, unless they reverse the structuralchanges that
have taken place in recent years. Increasedunion wage premiumsand
wage dispersionmore generallymean that in equilibriummore people
will lose high-wagejobs and choose to remainunemployedlonger than
was previouslythe case. The lattereffect is magnifiedby the increasing
tendency for the unemployed to be in families with other working
members. Between 1977 and 1985, the share of unemployedmarried
maleswho hadanotherfamilymemberworkingfull timeincreasedfrom
37.4 percentto 43.6 percent.
Second, while expansionary policies are not likely to reduce the
equilibriumunemploymentrate, stablefiscal and monetarypolicies can
probably make a significantcontribution.Since workers losing highwagejobs are the ones most likely to choose transitionalunemployment
Lawrence H. Summers
383
over takinga low-wagejob, policies that temporarilycontractthe highwage sectorof theeconomyarelikelyto createstructuralunemployment.
Theremay be importantasymmetriesbetween the effects of expansionary and contractionarypolicies. Policies that hurtthe high-wagesector
may create much more transitionalunemploymentthan policies that
promote it can alleviate if new high-wagejobs are taken by workers
otherthanthose previouslylaid off.
The recent fiscal-monetarymix and the associated squeeze on the
high-wagemanufacturingsector are instructive.Whenthe manufacturing sector is squeezed, unemploymentincreases sharplyas those who
lose jobs wait to get them back. The eventual abnormalincrease in
manufacturingoutputthatwill be necessaryto service the tradedebt the
United States is now incurringis unlikelyto reduce unemploymentby
as muchas the contractionincreasedit.
Comments
and Discussion
Katharine G. Abraham: Lawrence Summersarguestwo mainpoints.
First, the bulk of the increase in "normal" U.S. unemploymentsince
the mid-1960scannot be attributedsimply to shifts in the demographic
compositionof the workforce, buthas been concentratedamongmature
men who lose theirjobs and experience extended spells of unemployment. Second, this increasein maturemale unemploymenthas resulted
in large partfrom high and growingnoncompetitivewage differentials,
whichhave contributedto employmentdeclinesin the high-wagesectors
in the face of demandshocks and led those who lose high-wagejobs to
hold out longerin hopes of gettingtheiroldjobs back.
Let me start with Summer's "stylized facts" regardingthe decomposition of the observed increase in the U.S. unemploymentrate. It is
true that no more than a smallpartof the observed increase in the U.S.
unemploymentrate between 1965and 1985can be attributedsimply to
shifts in the demographiccompositionof the labor force towards more
unemployment-prone
workers.However, Summers'ssuggestionthat the
increasinglevel of educationalattainmentamonglaborforceparticipants
has actually worked to decrease the unemploymentrate is open to
question. Not having more than an eighth-gradeeducation or having a
college degreebothobviously meansomethingvery differentfor today's
new entrants than for the cohorts that are currentlyretiring.Relative
educationalattainmentwithin an age cohort probablydoes affect employmentprospects;relativeeducationalattainmentacross age cohorts
is unlikelyto matterin the same way.
Having rejected the idea that demographicshifts in the composition
of the labor force can explain the growth in unemployment,Summers
considershow the increasein unemploymenthas been distributedacross
384
Lawrence H. Summers
385
the laborforce. He emphasizesthe fact that, over the 1965to 1985period
taken as a whole, there have been larger cumulative increases in
unemploymentamong maturemen than among other groups. Whathe
does not emphasizeis that the patternof unemploymentgrowthlooked
very different prior to the late 1970s than it has since then. This is
important,because it means that no single explanationfor the upward
driftin U.S. unemployment,includingSummers'sexplanation,is likely
to applyto the entire 1965to 1985period.
Summers's tables 4 and 5 show that increases in unemployment
between 1978and 1985were relativelylargerfor men aged thirty-fiveto
forty-fourand for marriedmen thanfor other groupsand that the share
of unemploymentattributableto job losers and to those experiencing
long spells of unemploymentrose. But between 1965 and 1974, the
largest increases in unemploymentoccurredamongyoung men, young
women, andolderwomen;menagedthirty-fiveto forty-fourexperienced
virtuallyno changein theirunemploymentrate, and the unemployment
rates for older men actuallyfell. Between 1974and 1978,the increasein
unemploymentwas spreadrelativelyevenly across age-sex groups.The
statistics on unemploymentby marital status group show a similar
pattern:only duringthe 1978to 1985perioddo marriedmenanddivorced,
separated,and widowed men fare relatively worse than other groups.
Data on unemploymentby reason-job losers, job leavers, and so onare not available before 1967, but the data on the distribution of
unemploymentshowthatlong-termunemploymentbecamedramatically
moreimportantonly beginningin 1978.
Because using any particularyears for such an assessment may be
misleading,I have examinedsome of these trendsusingannualdata. My
table 1 reports annual unemploymentrates since 1954 for men aged
thirty-fiveto forty-fourandfor marriedmen. These unemploymentrates
were strikinglystableup untilthe last few years. Regressionsof the two
maturemale unemploymentrates on a constant, a time trend, and the
overallunemploymentrateare summarizedin my table 2. These models
bearon the questionof whethermaturemales'unemploymentrateshave
in fact risen morerapidlythanthe unemploymentrates of othergroups,
as Summershas asserted. Looking at the model for the entire 1954to
1985period, the answer to that question would appear to be a simple
"no." However, fittingseparatesubperiodmodelsindicatesthatmature
men's unemploymentactually trended downwardsrelative to overall
Brookings Papers on Economic Activity, 2:1986
386
Table 1. Aggregate and Mature Male Unemployment Rates, 1954-85
Unemployment rate
Year
Males
aged
35-44
Married
malesa
Aggregate
1954
1955
4.1
3.1
n.a.
2.6
5.5
4.4
1956
1957
1958
1959
1960
2;6
2.8
5.1
3.7
3.8
2.3
2.8
5.1
3.6
3.7
4.1
4.3
6.8
5.5
5.5
1961
1962
1963
1964
1965
4.6
3.6
3.5
2.9
2.6
4.6
3.6
3.4
2.8
2.4
6.7
5.5
5.7
5.2
4.5
1966
1967
1968
1969
1970
2.0
1.7
1.6
1.5
2.4
1.9
1.8
1.6
1.5
2.6
3.8
3.8
3.6
3.5
4.9
1971
1972
1973
1974
1975
3.1
2.7
2.0
2.6
4.9
3.2
2.8
2.3
2.7
5.1
5.9
5.6
4.9
5.6
8.5
1976
1977
1978
1979
1980
4.1
3.5
2.8
2.9
4.1
4.2
3.6
2.8
2.8
4.2
7.7
7.0
6.0
5.8
7.1
1981
1982
1983
1984
1985
4.5
6.9
7.1
5.2
4.9
4.3
6.5
6.5
4.6
4.3
7.6
9.7
9.6
7.5
7.2
Source: U.S. Department of Labor, Bureau of Labor Statistics, Handbook of Labor Statistics, Bulletin 2217 (June
1985), and BLS, Employment and Earnings, vol. 33 (January 1986).
n.a. Not available.
a. Data for married males start in 1955.
unemploymentthroughthe late 1970s,thoughit may have risen relative
to overallunemploymentsince then. Significantly,the coefficientsin the
earliersubperiodmodel are quite robustto changes in both the starting
yearandthe endingyear used to estimatethe model.
Lawrence H. Summers
387
Table 2. Unemployment Trends among Mature Males, Various Periods, 1954-85a
Males aged 35-44, u1
Period
Married males,
Ulb
Trend
Civilian
unemployment, u
Trend
Civilian
unemployment, u
-0.02
(5.6)
0.04
(8.2)
-0.02
(15.6)
1.62
(16.9)
1.48
(27.2)
1.46
(29.6)
-0.01
(7.6)
0.01
(1.7)
-0.02
(15.1)
1.56
(25.9)
1.53
(16.3)
1.48
(38.2)
-0.02
(5.2)
-0.02
(4.0)
-0.02
(3.4)
-0.02
(2.3)
0.00
(0.2)
1.45
(18.7)
1.44
(16.0)
1.46
(14.1)
1.41
(13.6)
1.50
(15.1)
- 0.02
(8.1)
-0.02
(6.4)
-0.02
(7.1)
- 0.02
(5.9)
-0.01
(3.0)
1.43
(36.6)
1.43
(31.4)
1.47
(32.6)
1.48
(29.5)
1.47
(27.4)
-0.02
(4.0)
-0.02
(2.8)
-0.01
1.44
(16.0)
1.43
(13.4)
1.44
-0.02
(6.4)
-0.01
(3.4)
-0.01
1.43
(31.4)
1.42
(20.1)
1.42
(2.2)
(12.5)
(3.1)
(19.8)
-0.01
(1.9)
1.50
(12.0)
-0.01
(2.7)
1.46
(18.8)
Long periods
1954-85b
1979-85
1954-79b
Vary start year
1963-79
1964-79
1965-79
1966-79
1972-79
Vary end year
1964-79
1964-80
1964-81
1964-82
Sources:Author'scalculationsusingdatafromtable 1.
a. Equationestimatedis Inul = A + b, t + b2 Inu,wherethe dependentvariableis the naturallogarithmof the
unemployment
ratefor malesaged 35-44 and for marriedmales;t is a time trendand Inuis the naturallogarithmof
the civilianunemploymentrate;t-statisticsare in parentheses.
b. Datafor marriedmalesstartin 1955.
In sum, the data show very clearly that only since the late 1970shas
increasedunemploymentamongmaturemalejoblosers grownmarkedly
enoughto accountfor a disproportionateshareof the overall growthin
unemployment.Summers'stransitionalunemploymentstory,then,should
be considereda story aboutthe past six or seven years, not a storyabout
the past twenty years.
Thetransitionalunemploymentstoryhastwo parts.Summersbelieves
both that noncompetitive wage differentialshave contributedto the
effects of adverse shocks on layoffs and plant closings, particularlyin
388
Brookings Papers on Economic Activity, 2:1986
the union sector, and thatwideningwage differentialshave led recently
displaced workers to choose to remain unemployedlonger than they
wouldotherwisehave done. Summersis less explicitthanhe mighthave
been concerningthe respective roles played by demandfactors per se
versus noncompetitivewage differentialsin swellingthe flows of laid-off
workers into unemploymentin recent years. However, it is plausible
thatwage rigidityin the high-wagesectors has contributedto job loss in
those sectors. I findit less plausiblethat reducedwillingnesson the part
of today's displacedworkersto accept alternateemploymenthas been
a significantfactorin raisingthe unemploymentrate.
Onereasonformy skepticismregardingthis secondpartof Summers's
transitionalunemploymentstory is thatI findit hardto believe thatmost
unemployed mature men, particularlyunemployed marriedmen, are
reallyin a positionto wait very long for goodjobs. WhileI don't want to
maketoo muchof anecdotalevidence, GeneralMotors'experiencewith
its GuaranteedIncome Stream (GIS) programcomes to mind here.
Under that program,high-senioritylaid-off workers are entitled to a
guaranteed income until they reach retirement age. The condition
attachedto the guaranteeis that if offered a job at anotherGM plant,
even ajob halfwayacross the country, the laid-offworkermust accept
it. Thejobs are not necessarilyattractive:the workerwho moves starts
over at the bottomof the plantseniorityladder,whichmaymeanworking
the night shift, performingan onerous task, and being vulnerable to
temporarylayoff. According to Al Warren,GM's Vice-Presidentfor
IndustrialRelations, marriedmen almost always accept offered jobs
ratherthanlose theirGIS eligibility,whereas single men are morelikely
to turnjobsdown. His interpretation:marriedmen simplycannotafford
to be withoutajob, even if it is a worsejob thanthey held before.
Statistics on growth in the union-nonunionwage differentialand in
cross-industry wage differentials are the main evidence underlying
Summers's view that the unemployed now have reason to hold out
longerin hopes of gettingtheiroldjobs back. But the data on the union
compensationpremiumin table 14 imply that while the union premium
did grow substantiallyfrom the early 1970sthrough1982,by 1985it had
fallenbackto its 1978level. Similarly,the Bell andFreemandatain table
15 show that cross-industrywage dispersiongrew rapidlybefore 1978,
but less rapidly between 1978 and 1985. Thus, the data do not offer
strongsupportfor the view that the relativeattractivenessof unionjobs
Lawrence H. Summers
389
or high-wage-industry
jobs increasedsubstantiallyover the 1978to 1985
periodduringwhich maturemale unemploymenthas risen.
It should also be noted that neither growth in the union-nonunion
wage differentialnor growthin the dispersionof averagewages across
industriesnecessarilyimpliesanincreasein the dispersionof wage offers
availableto unemployedworkers, as would be necessary for standard
searchmodels to predictan increase in the durationof unemployment.
Any conclusion regardingthe dispersion of wages among individuals
requiresknowing somethingboth about the relative wages in different
sorts ofjobs and aboutthe relativesharesof employmentin thosejobs.
All else the same, if the shareof employmentin high-wagejobs shrinks
and the share of employmentin average-wagejobs rises, overall wage
dispersionwill tendto fall. Overthe sameperiodsthatincreasesin unionnonunionandcross-industrywage differentialshave occurred,the union
shareof totalemploymentfell, andthe high-wageindustrieswherewage
growth was most rapid experienced below-average or even negative
employmentgrowth. Withoutmore information,it is impossibleto say
what the net effect on overall wage dispersion was. But some added
informationis available.James Medoff has looked at the dispersionof
hourlywage rates using CurrentPopulationSurvey data for the period
1973 to 1984. He finds no increase in the dispersion of hourly wages
amongindividuals.' Thus, even if it can be assumedthatincreasesin the
dispersionof wages paid to employed workerstranslateinto increased
dispersion of the wage offers made to the unemployed, the evidence
does not establishthat any such increase has occurred.
All in all, I am more preparedto believe that the recent increase in
mature male unemploymentreflects increases in the numbers of displaced workers than that it reflects on increased propensity of those
displacedworkersto hold out in hopes of gettingtheiroldjobs back.
MichaelL. Wachter: LawrenceSummershas done an admirablejobof
describingthe currentpattern of unemployment.Overall I agree with
his diagnoses of the problem. In my comments I want to discuss four
aspects of the Summerspaper:whetherthe economy is indeed nearfull
1. JamesL. Medoff, "The Structureof HourlyEarningsAmongU.S. PrivateSector
Employees:1973-1984"(HarvardUniversity,December1984).
390
Brookings Papers on Economic Activity, 2:1986
employment; the appropriateadjustmentsto the unemploymentrate
based on labor supply considerations;the impact of noncompetitive
wages on "wait" unemployment;and the policy implications of the
researchfindings.
The first issue is the assumptionin the title of the paper that the
economy is "near" full employmentat 7 percent. AlthoughSummers
offers some evidence that is suggestive of a nonacceleratinginflation
rate of unemployment(NAIRU) of 7 percent, the evidence is decidedly
mixed, andSummersseems unwillingto takea strongstandon the issue.
For example,he mentionsthatthe traditionalapproachof solvinga wage
inflation equation for the unemploymentrate consistent with stable
inflationdoes not lead to robustequilibriumunemploymentrates. Based
on his evidence, a reader(such as myself) who believed thatthe NAIRU
was 6 percent ratherthan 7 percent would not be convinced to change
his view of the matter.
Whetherthe economy's NAIRU is as low as 6 percent, or as high as
7 percent, is importantto Summers'sinterpretationof the evidence. If
we are nearfull employment,then it is legitimateto compare 1985with
earlierfull-employmentyears such as 1965 and 1978. In addition, the
composition of the currentpool of unemployedcan be interpretedas
representingstructuralratherthan cyclical problems, and the change
from 1965to 1985can be studiedas reflectinglong-termstructuraltrends
in the unemploymentproblem.
The strongestevidence that the economy is not at its NAIRU is the
trendin wage andprice inflationrates. Recent growthrates of wage and
price are at levels not experienced since the early 1960s and, more
important,thereis littleevidencethatinflationratesareaboutto increase.
For example, the rate of increase of average hourlyearnings(AHE) is
now below 3 percenton an annualbasis. Thatmeasureof core inflation
was 6.9 percentin 1982and then declinedannuallyto 4.9 in 1983,3.2 in
1984,3.0 in 1985,and 2.6 percent for the four quartersending with the
second quarterof 1986. The AHE currentrate of wage growth may be
artificiallydepressed because of compositionalshifts in the economy,
but even after those shifts are correctedfor, the conclusion is that the
labormarketis slack. The employmentcost index, which is less subject
to compositionalbias, is increasingat a ratecloser to 4 thanto 3 percent,
and here againthe 1985-86 trendhas been down ratherthanup.
Addingto the evidence that the economy is not at NAIRU is the fact
Lawrence H. Summers
391
that the unemploymentrate has also been declining slowly along with
wage growth rates. Near full employment one might expect that the
movementup the short-runPhillipscurve would offset any downward
shift in that relationshipdue to the lag in the adjustmentof inflation
expectations. That is, near full employmentone might anticipatethat
even slowly decliningunemploymentrateswouldresultin at least stable,
and probably increasing, core rates of inflation. But this is not yet
happening.
If the economy's NAIRU is 6 ratherthan 7 percent, then there is a
substantialamountof cyclical unemploymentleft in the economy. The
point is importantsince individualsin differentage-sex, education, and
industrycategoriesexperiencedifferentpercentagepointimprovements
when the economy makes its final approachto NAIRU. Hence Summers's conclusionswithrespectto whichgroupshave sufferedincreased
unemploymentrates since 1965mightreflectthe remainingcyclical gap
ratherthan structuralforces.
Puttingaside the NAIRU issue, I believe that Summers makes an
importantcontributioninterpretingtoday's structuralunemployment.
His interpretationis based on supply-sidecharacteristicsof the labor
force andthe relatedimportanceof the growthin "wait" unemployment.
Summersargues that earlier attempts, includingmy own, to adjust
for demographicfactors were too focused on age-sex differences in
unemploymentand ignoredother importantdifferences, such as educationalattainment.If unemploymentdiffersamonggroupswith different
levels of education, industry of employment, or marital status, the
omission of those factors results in a biased correction for age-sex
effects.
In defense of earlieradjustments,I note thatthe youthunemployment
problemof the 1970swas reasonablywell approximatedby controlling
for age and sex. In 1979,for example, one-halfof the unemployedwere
youngworkersbetweenthe ages of sixteenandtwenty-four.Distinctions
based on maritalstatus and educationalattainmentaddlittle insightand
might even introduce a bias. Youths are always disproportionately
unmarried,not divorced, and, especially in the case of teenagers,
characterizedby a low level of educationalattainment.Hence, interaction variablesbetween those characteristicsand age are important,and
theirexclusionalso poses omittedvariableproblems.In the 1970sit was
difficultto decipherwhether changes in those rates reflected short-run
392
Brookings Papers on Economic Activity, 2:1986
age-baseddemographiceffects or longer-runeffects that would prove
independentof age. Now that youth unemploymentis declining, there
is more supportfor includingthose other variablesbased on the belief
that trendsin those rates are not drivenby age factors.
Summers'sanalysisindicatesthat shifts in unemploymentby marital
status are an importantpartof the structuralstory. I believe that this is
correctbutthatthe underlyingmechanismis maritalstatusin conjunction
with the numberof wage earnersin a family. It is certainlythe case that
the numberof familieswith two wage earnershas been increasing,and
search theory implies that these families should have higher rates of
unemployment. Similarly, single-parent, female-headed families are
likely to experience high unemploymentincidence because of frequent
transitioninto andout of the laborforce. Unfortunately,the quantitative
importanceof these factorsis difficultto test given the availabledata.
Withrespectto laborsupplycorrections,Summersarguesthatearlier
age-sex adjustmentsin estimatingNAIRU shouldresultin a findingof a
lower NAIRU today. Hence if NAIRU today is 7 percent, then those
correctionscanbe presumedto be incorrect.Butthisconclusionassumes
that the economy has returnedto full employment, a point that is not
proved in the paper. My own recent work does suggest that youth
unemploymentin particularhas been decliningrelativeto adultrates. If
the currentNAIRU is 6 percent, then adjustingfor the GNP cyclical gap
results in youth unemploymentrates lower than their 1978full-employment levels. The one importantexception to this conclusion is black
males aged eighteento twenty-four.
I agree strongly with Summers's argumentsconcerningthe importance of union wage differentialsin causing unemployment.Although
Summersrefersto this type of unemploymentas transitionalunemployment, it is useful to use the term "wait" unemployment. Whereas
transitionalunemploymentcan representalmost any form of frictional
or structuralunemployment,the labor economics literatureuses the
wait unemploymentterminology to refer to that unemploymentthat
arises from the rationingof high-wagejobs in a non-market-clearing
sector. Summers'sconclusions with respect to the impact of noncompetitivewages on unemploymentsupportmy currentresearchwithPeter
Linneman.
In calculatinga cross-sectionaltime series of union wage premiums
for each year between 1973and 1984,Linnemanand I found that union
Lawrence H. Summers
393
wage premiumsincreasedstronglyduringthe late 1970sandearly 1980s,
often by as much as 50 percent over the period. We then used the
estimatedindustry-unionpremiumsas explanatoryvariablesin a secondstage cross-sectionaltime series regressionto explainchangesin unionnonunion shares of employment. The results from this second-stage
equationshow a quantitativelylarge and statisticallysignificantimpact
of industry-unionpremiumson employmentshares.These resultsacross
industries support Summers's findingswhere the data base is across
states.
Althoughthe evidence supportsthe contention that noncompetitive
wage differentialscause wait unemployment,it does not tell us whether
the problem is likely to prove long lasting or transitional.Although
Summersmay be tiltingtowardthe long-lastingnatureof the problem,a
case can be madethat it is transitional.
Whetherone waits for a job with a noncompetitivewage or quickly
accepts a competitive wage depends not only on the size of the wage
differentialbut also on the probabilityof being hiredand receiving that
wage premiumjob. In the 1970swait unemploymentwas concentrated
among young or entry-level workers who could expect some turnover
in noncompetitivehigh-wagepublic servicejobs and, to a lesser extent,
in unionjobs. There was good incentive to look for those jobs, either
because they were fundedby the governmentand relativelyinsensitive
to economicconditionsor because the unionsector's structuralproblem
hadnot yet emergedacross a broadrangeof industries.
In other words, duringthe 1970sthe probabilityof findingajob with
a noncompetitivewage was higherthan it has been in the 1980s. More
specifically, the probabilityof a successful wait was high enough to
encourage waiting on the part of the new entrant group or those
unemployedfrom nonpremiumwagejobs. Duringthe 1980s,the probabilityof landingsuch ajob has been declining.Those who have lost such
ajob mightstill be expected to wait in hope of landinga newjob. In any
cyclical bounce they might be recalled. Moreover, the process of
adjustingone's reservationwagedownwardtakestime.But new entrants
orworkersemployedelsewhereareunlikelyto be attractedto the queue.
Theirprobabilityof findingajob in the decliningunion sector is simply
too low.
My final comment concerns Summers's short section covering his
policyconclusions. The Summerspaperconcludes with the recommen-
394
Brookings Papers on Economic Activity, 2:1986
dationthatrisingratesof adultmaleunemployment,althougha problem,
should not be addressedby expansionarymonetaryand fiscal policies.
Althoughthe paperis silent on whetherrisingadultmaleunemployment
rates should be addressedby structuralpolicies, that appearsto be the
implicitmessage. The issue is relevantbecause of the currentdebate on
whether fundingfor disadvantagedyouth programsshould be shifted
towardprogramsthat assist displacedadultmales.
The problemof structuralunemployment,however, cannotbe easily
addressed. There is a need to targetfundingtowardparticulargroups.
The issue is to identify variablesthat rankpockets of structuralunemployment in terms of their need for policy assistance. There is no
particularreason to assume that groups with increasingrates of unemployment are the ones to be targeted for assistance. Such a variable
would lead toward continuing assistance for disadvantagedyouths,
depending on their family income. If this were used to rank groups,
disadvantagedyouths and single-parentfamilies would continue to be
targeted.
GeneralDiscussion
MartinBaily pointedout that while the adjustmentsSummersmakes
for post-1965changesin the age-sex compositionof the laborforce alter
the 1985unemploymentrate only slightly, even Summers'sprocedure
shows thatthe changesadded 1.0percentagepointto the unemployment
rate between 1954and 1978. The Perry procedureweighted each agesex groupby the relativeearningsof its workersand yielded an adjustmentfor the same periodin the neighborhoodof 2.0 percentagepoints.
ChristopherSims noted that the models whose results are shown in
tables 12 and 13 provide only weak evidence for union effects on
unemploymentand employment. The most reliably specified models
includeregionaldummiesandaninstrumentedunionvariable;thelargest
t-statistic on any union variable in any of these models is only 1.7.
Summers felt that controllingfor regional effects probably obscured
some of the union effects and noted that regionalcontrols increasedthe
standard errors but did not decrease the magnitudes of the union
coefficients.
Jeffrey Sachs stressed that increased unemploymentcould not be
blamed directly on union actions. In his view, a more convincing
Lawrence H. Summers
395
interpretationis thatexogenous supplyshocks have led to bothdeclining
employmentin the union sector and decliningwages in the nonunion
sector, producingthe pattern of correlationsSummersreports in the
second half of the paper. Summers agreed that rising union wage
premiumsprobablyreflectedwage rigidityin the face of adverse shocks
ratherthanpureunionpush.
StanleyFischer, notingthatin previousworkSummershadexplained
high Europeanunemploymentas a consequence of employedworkers'
keeping wages high even when others are out of work, asked why the
same interpretationwas not applicableto the United States. Summers
replied that a key differencebetween the United States and Europe is
the presence of a large nonunion sector in the United States. The
explanationsthat he and others have offered for the rigidityof union
wages do not apply to the U.S. nonunion sector. To explain U.S.
unemployment,one must explain why the U.S. nonunionsector does
not soak up any unemploymentgeneratedin the high-wageunionsector.
This is the reason for introducingthe transitionalunemploymentinterpretation.
Wayne Vroman questioned Summers's contention that remaining
unemployedto wait for a goodjob has become moreattractivein recent
years. According to Summers, people who lose high-payingjobs find
thatthe prospectivewage in theirnext best alternativejobhas declined.
Thus they wait longer to regain their formerjobs. But Vroman cited
evidence that other sources of family income-unemployment benefits
and earnings of other family members-have declined since the late
1970s,so that familyincome duringan unemployedworker'swait for a
job is lower than it used to be. The effect would be to shorten the
equilibriumdurationof unemployment.
Vromannoted that both he and Gary Burtless have documented a
declinein the availabilityof unemploymentinsurancebenefits.A similar
decline is shown in table 6 of the Summerspaper. The cutbacks, which
occurredafter 1979, have been largest in programsfor the long-term
unemployed.They have also been largestin the majorindustrialstates
thathave experiencedespecially highunemploymentin the 1980s.Thus
the cutbackin unemploymentinsurancebenefitshas been concentrated
amongthe sameworkerswhose unemploymentexperiencehasworsened
the most.
Vromanaddedthat evidence on other family income sources for the
396
Brookings Papers on Economic Activity, 2:1986
unemployed pointed in the same direction of reduced incentives and
ability to hold out for high-payingjobs. Pretransferpoverty rates for
individualsin each categoryof unemploymentdurationwere uniformly
higherin 1983than in 1976. Real wages of other family memberswere
stagnantover the period.Vromanconcludedthatthese findings,coupled
with the cutbacksin unemploymentinsurancebenefits, stronglysuggest
that the unemployed in general, and the long-term unemployed in
particular,were worse off waitingfor a job in the 1980sthan they were
in the 1970s.
Alan Blindersuggestedthat one directtest of Summers'stransitional
unemploymenthypothesiswouldbe to look at unemploymentdurations
for people groupedby the differencebetween theiractualwage and the
wage predictedfor someone with theirhumancapitalcharacteristics.If
Summers'stheoryis correct,those with higher-than-predicted
wages on
their previous jobs should have longer-than-averageunemployment
durations.
RobertGordonproposedthe followingquestionregardingthe nature
of today's unemployment:if the U.S. monetary-fiscalpolicy mix, the
tradedeficit,andthe relativedeclinein manufacturingemploymentwere
reversed, how much would the marriedmale unemploymentrate fall?
In Gordon'sview, it was hardto imaginethatit would not returnto near
its level before the trade deficit grew. In this sense, the increased
unemploymentof recent years fundamentallyreflects demand rather
than supply developments. MartinFeldstein agreed that an important
part of mature male unemployment reflected the dramatic loss of
manufacturingjobs due to growth in the U.S. trade deficit. While he
would not advocategeneralizeddemandstimulus,Feldsteindid predict
that the improved health of the manufacturingsector that could be
expected from the lower dollar should reduce the unemploymentrate
among these older men. Robert Hall also regardedthe unprecedented
shrinkagein durablegoods productionin recent years as an important
cause of currenthigh mature male unemployment.He reasoned that
those who lose "career" jobs may hold two or three jobs in quick
succession before findinganother"career"job, so that a singlejob loss
may generateseveralroundsof unemployment.