Digitized
by
the
Internet
Archive
in
2011
with
funding
from
Boston
Library
Consortium
Member
Libraries
B31
M415
M'Obr
P-
7
"
Massachusetts
Institute
of
Technology
Department
of
Economics
Working
Paper
Series
Are
Emily
and
Greg
More
Employable
than
Lakisha
and
Jamal?
A
Field
Experiment
on
Labor
Market
Discrimination
Marianne
Bertrand
Sendhil
Mullainathan
Working
Paper
03-22
May
27,
2003
RoomE52-251
50
Memorial
Drive
Cambridge,
MA
02142
This
paper
can
be
downloaded
without
charge from
the
Social
Science
Research Network
Paper
Collection
atMASSACHUSETTS
INSTITUTE
OF
TECHNOLOGY
SEP
1 92003
Are
Emily
and Greg
More
Employable
than Lakisha
and Jamal?
A
Field
Experiment
on Labor
Market
Discrimination
Marianne
Bertrand
and
Sendhil
Mullainathan
*May
27,2003
Abstract
We
performafieldexperimenttomeasureracialdiscriminationinthelabor market.We
respondwithfictitious resumestohelp-wanted adsin Boston and Chicagonewspapers.
To
manipulateperception of race, eachresume israndomlyassigned eithera very African American soundingname
oraveryWhite
sounding name.
The
resultsshow
significant discrimination against African-American names:White
names
receive 50 percentmore callbacksfor interviews.We
also find that raceaffects thebenefits ofabetter resume. For
White
names, ahigher qualityresume elicits30 percentmore
callbackswhereasforAfricanAmericans,itelicitsafarsmallerincrease. Applicantslivinginbetterneighborhoodsreceive
more
callbacks but, interestingly, thiseffectdoes notdifferby race.
The amount
of discriminationisuniformacross occupations and industries. Federal contractors and employers
who
list "Equal Opportunity Employer"in theiraddiscriminateasmuch
asotheremployers.We
find littleevidence that ourresultsaredrivenbyemployersinferringsomethingotherthanrace,suchas socialclass, fromthenames. These
resultssuggestthat racialdiscriminationisstill aprominent featureofthe labor market.
•University ofChicagoGraduateSchool of Business,
NBER
andCEPR;
MIT
andNBER.
Addresses: 1101E.58thStreet,R0
229D,Chicago, IL 60637; 50MemorialDrive,E52-380a, Cambridge,
MA
02142. E-mail: marianne.bertrand@gsb.uchicago.edu;mullain@mit.edu. DavidAbrams,VictoriaBede,Simone Berkowitz,HongChung,AlmudenaFernandez,Mary Anne
Guedigu-ian,ChristineJaw, RichaMaheswari, Beverley Martis, Alison Tisza, Grant Whitehorn, andChristine Yeeprovided excellent
1
Introduction
Every
measure
ofeconomic
success reveals significant racialinequality in theUS
labormarket.Compared
toWhites, African
Americans
are twice as likely to beunemployed and
earn nearly25 percent lesswhen
they are
employed
(CouncilofEconomic
Advisers, 1998).This
inequality hassparked a debateon whether
employers discriminate
by
race.When
facedwith observablysimilarAfricanAmerican
and White
applicants,do
theyfavor theWhite
one?Some
argueyes, citingeitheremployer
prejudice oremployer
perceptionthatrace signals lower productivity. Others argue that discriminationis a relicofthe past, eliminated
by
some
combination of
employer
enlightenment, affirmative actionprograms
and
the profit-maximization motive.In fact,
many
in this latercamp
even feel that stringent enforcement of affirmative actionprograms
hasproduced
an environment
of reverse discrimination.They
would
argue thatfacedwith identicalcandidates,employers
might
favorthe AfricanAmerican
one.1Data
limitations
make
itdifficultto empiricallytesttheseviews. Sinceresearchers possess far less data
on
workersthan
employers do,White and
AfricanAmerican
workers that
appear
similarto researchersmay
look verydifferent toemployers.So
any
racialdifference inlabor
market outcomes
couldjust as easilybe
attributed to thesedifferencesunobserved by
researchersastodiscrimination.
We
conduct
a fieldexperiment
to circumvent this difficulty.We
send resumes in response tohelp-wanted
ads inChicago
and Boston newspapers and measure
thenumber
ofcallbackseachresume
receivesforinterviews.
We
experimentally manipulate perceptionofraceviathename
on
theresume.We
randomly
assign very
White sounding
names
(such asEmily Walsh
orGreg
Baker) to half the resumesand
veryAfrican
American sounding
names
(such asLakisha Washington
orJamal
Jones) tothe otherhalf.Because
we
are also interested inhow
credentials affect discrimination,we
experimentally vary the quality oftheresumes used inresponse to a givenad. Higherquality applicantshave
on
average alittlemore
labormarket
experience
and
fewerholesin theiremployment
history;they arealsomore
likelytohave
an email address,have
completed
some
certificationdegree,possess foreignlanguage
skillsorhave beenawarded
some
honors.2In practice,
we
typicallysendfourresumes
in responsetoeachad:two
higherqualityand two
lowerqualityones.
We
randomly
assign toone
ofthe higherand
oneofthe lowerqualityresumesan
AfricanAmerican
sounding
name.
In total,we
respond
toover 1300employment
ads in the sales, administrative support,clerical
and customer
servicesjob categoriesand
send nearly5000
resumes.The
adswe
respond to coveralarge
spectrum
ofjob quality,from
cashierwork
at retailestablishmentsand
clericalwork
in amailroom
tooffice
and
salesmanagement
positions.We
find large racial differences in callback rates.3 Applicants withWhite
names
need to send aboutJ
This
camp
often explains thepoorperformanceofAfricanAmericansintermsofsupplyfactors. IfAfricanAmericanslackmanybasicskillsenteringthe labormarket,then theywill perform worse, even with parityor favoritismin hiring.
Increating the higher quality resumes,wedeliberatelymadesmallchangesincredentialsso as tominimizethechanceof over-qualification.
3For ease of exposition,we
10
resumes
to get one callbackwhereas
applicants with AfricanAmerican
names
need to sendaround
15resumes
to getonecallback. This 50 percentgap
in callback rates isstatisticallyverysignificant.Based on
ourestimates, a
White
name
yieldsasmany
more
callbacks asan
additional eight years of experience. Sinceapplicants'
names
arerandomly
assigned, thisgap
can onlybe
attributed tothename
manipulation.Race
alsoaffects thereward
to having a betterresume.Whites
with higher qualityresumes
receive 30percent
more
callbacksthan
Whites
with lowerqualityresumes, astatisticallysignificant difference.On
theotherhand, having a higherquality
resume
hasamuch
smallereffect forAfricanAmericans. Inotherwords,the gap
between
White and
African-Americans widens
withresume
quality.While
onemay
have expected thatimproved
credentialsmay
alleviate employers' fear that AfricanAmerican
applicants are deficient insome
unobservableskills, this is notthe case inour
data.4 Discrimination thereforeappears to bite twice,making
itharder not onlyforAfricanAmericans
to find ajobbut
alsotoimprove
their employability.The
experiment
also reveals several other aspects of discrimination. First, sincewe
randomly
assignapplicants' postal addresses to the resumes,
we
can
study the effect ofneighborhood
of residenceon
theprobability of callback.
We
find that living ina wealthier (ormore
educated ormore White) neighborhood
increases callback rates. But,interestingly, African
Americans
are not helpedmore than Whites by
livingina "better" neighborhood. Second, the
amount
ofdiscriminationwe measure by
industrydoes notappear
correlated to Census-based
measures
ofthe racialgap by
industry.The same
is true for theamount
ofdiscrimination
we
measure
in differentoccupations. Infact,we
findthatdiscriminationlevelsarestatisticallyindistinguishable across all the occupation
and
industrycategoriescoveredin theexperiment.We
also findthat federal contractors,
who
are thought tobe
more
severely constrainedby
affirmative action laws,do
not discriminate less;neither
do
largeremployersoremployerswho
explicitly statethatthey
arean
"EqualOpportunity Employer"
intheir ads. In Chicago,we
findthat employers locatedinmore
AfricanAmerican
neighborhoods
are slightlyless likelyto discriminate.The
rest of thepaper
is organized as follows. Section 2compares
this experiment to priorwork
on
discrimination,
and most
notably tothelabormarket
audit studies.We
describethe experimental designin Section 3
and
present the results in Section 4.1. In Section 5,we
discuss possible interpretations ofour results, focusing especially
on
two
issues. First,we
examine whether
the race-specificnames
we
havechosen
might
alsoproxy
for social classabove
and beyond
the raceofthe applicant.Using
birthcertificatesdata on mother's education for the different
names
usedin our sample,we
find littlerelationshipbetween
social
background
and
thename
specific callback rates.5 Second,we
discusshow
our resultsmap
backeffectsareabout theracialsoundingnessofnames.
We
brieflydiscussthepotentialconfounds betweenname
andracebelow andmoreextensivelyinSection5.1.4These
resultscontrastwith the view, mostly basedon non-experimentalevidence, that AfricanAmericansreceivehigher returns toskills. Forexample,estimating earnings regressionsonseveral decadesofCensusdata, Heckmanetal. (2001) show
that AfricanAmericansexperience higher returns to a high school degreethanWhites.
5
We
alsoarguethatasocial classinterpretationwouldfindithardtoexplainall ofourresults,such as
why
livinga betterto the different
models
of discriminationproposed
in the economics literature. In doingso,we
focuson
two
important results: thelowerreturns tocredentials forAfricanAmericans
and
therelativehomogeneity
of discrimination across occupations
and
industries.We
conclude that existingmodels
do
apoor
job ofexplainingthe fullsetoffindings. Section6 concludes.
2
Prior
Research
on
Discrimination
With
conventional laborforceand
householdsurveys,it isdifficulttomeasure
racialdiscriminationoranalyzeitsmechanics.6
With
surveydata, researchersusuallymeasure
discriminationby comparing
the labormarket
performance
ofWhites and
AfricanAmericans
who
reportasimilarsetofskills.But
suchcomparisons
canbe
quite misleading.
Standard
laborforcesurveysdo
not containall thecharacteristicsthatemployers observewhen
hiring,promoting
or settingwages.So one
can neverbe
sure that the AfricanAmerican and White
workers being
compared
aretruly similarfrom
the employer'sperspective.As
a consequence,any measured
differencesin
outcomes
couldbe
attributed totheseunobserved
(tothe econometrician)factorsand
not todiscrimination.
This difficulty with conventional data has led
some
authors to use pseudo-experiments.' Goldinand
Rouse
(2000), for example,examine
the effect ofblindauditioningon
the hiringprocessof orchestras.By
looking atthe treatment offemale candidates before
and
aftertheintroductionof blind auditions, theytryto
measure
theamount
ofsex discrimination.When
such pseudo-experimentscan be
found, the resultingstudy
can be
veryinformative,but
findingsuch experiments hasbeen
extremely difficult.A
different set of studies,known
as audit studies,attempt
to placecomparable
minorityand White
subjects into actual social
and economic
settingsand measure
how
eachgroup
fares in these settings.8Labor market
audit studiessendcomparable
minority (AfricanAmerican
orHispanic)and
White
auditorsin forinterviews
and measure whether
one ismore
likely toget thejob than the other.9While
the resultsvary
somewhat
across studies, minority auditors tend to performworse
on
average: they are less likelytoget called
back
for a second interview and,conditionalon
getting calledback, less likelytoget hired.These
audit studiesprovidesome
ofthe cleanestnon-laboratory evidenceoflabormarket
discrimination.But
they alsohave
weaknesses,most
ofwhich
havebeen
highlighted inHeckman
and
Siegelman (1992)and
Heckman
(1998). First,these studies requirethatboth
members
ofthe auditorpair are identical in alldimensions that
might
affectproductivityinemployers'eyes, exceptfor race.To
accomplish this,researchers6See Altonji
and Blank 1999)fora detailedreview of the existingliteratureonracialdiscriminationinthelabormarket.
7Darity
andMason(1998) describeaninterestingnon-experimentalstudy. Prior totheCivilRightsActof1964,employment
ads would explicitlystateracial biases, providing adirect measureofdiscrimination. Ofcourse, as Arrow (1998) mentions,
discrimination wasatthattime "afacttoo evidentfordetection."
8Fix and Turner
(1998) provide a survey of
many
suchauditstudies. 9Earlier hiringaudit studies include
Newman
(1978) and Mclntyre et al (1980). Three more recentstudies are Crosset al(1990), Turneretal (1991), Jamesand DelCastillo (1991). Altonjiand Blank (1999),Heckman and Siegelman(1992) andtypically
match
auditors onseveral characteristics (height, weight, age, dialect, dressingstyle, hairdo)and
train
them
forseveral daystocoordinate interviewing styles. Yet, criticsnote that this is unlikelyto erasethe
numerous
differencesthat existbetween
theauditors ina
pair.Another weakness
oftheaudit studiesisthattheyarenotdouble-blind. Auditorsknow
thepurpose
ofthestudy.
As
Turner
et al (1990) note:"The
firstday
oftraining alsoincludedan
introductiontoemployment
discrimination, equal
employment
opportunity,and
a reviewofprojectdesignand
methodology." Thismay
generate consciousorsubconsciousmotives
among
auditors togenerate dataconsistent or inconsistentwithracialdiscrimination.
As
psychologistsknow
verywell,thesedemand
effectscanbe
strong. Itisverydifficultto insurethat auditors willnot
want
todo
"agood
job." Sincetheyknow
thegoal oftheexperiment, theycan alter their behavior in front ofemployers to express (indirectly) their
own
views.Even
a small beliefby
auditors that employerstreat minorities differently can resultin apparent discrimination. This effect isfurther magnified
by
the fact that auditors are not in fact seekingjobsand
are thereforemore
free to lettheirbeliefs affect the interview process.
Finally, audit studies are extremely expensive,
making
it difficult to generate largeenough
samples tounderstand
thenuances
and
possiblemitigatingfactorsof discrimination. Also, thesebudgetary
constraintsworsen
theproblem
ofmismatched
auditorpairs. Costconsiderations forcetheuseofafewpairs of auditors,meaning
thatany
one
mismatched
paircan easily drivethe results. Infact,these studies generallytendtofind significant differences in
outcomes
across pairs.Our
study
circumventstheseproblems. First,becausewe
only relyon resumes
and
notpeople,we
canbe
suretogenerate comparabilityacrossrace. Infact,since raceis
randomly
assignedtoeach resume, thesame
resume
willsometimes
be associated withan
AfricanAmerican
name
and sometimes
with aWhite
name.
This guarantees that
any
differenceswe
findaredue
solelyto theracemanipulation. Second,theuseofpaperresumes
insulates usfrom
demand
effects.While
the research assistantsknow
thepurpose
ofthestudy,ourprotocolallowslittle
room
forconsciousorsubconsciousdeviationsfrom
thesetprocedures. Moreover,we
canobjectively
measure whether
the randomization occurred asexpected.This
kindof objectivemeasurement
is impossiblein the caseofthe previous auditstudies. Finally, because of relatively low marginalcost,
we
can send out alarge
number
ofresumes. Besides giving usmore
precise estimates, this largersample
sizealso allowsusto
examine
themechanics
ofdiscriminationfrom
many
more
angles.101
A
similar "correspondence"techniquehasbeen usedina fewU.K.studies. See Joweiland Precott-Clarke (1970),Brown
and Gay (1985)andHubbockand Carter (1980). Theseearlierstudies have verylimitedsamplesizeandfocus exclusivelyon
documenting gapincallbacks betweenthe minorityand non-minoritygroups.
Some
of these studiesfailtofullymatchskills3
Experimental Design
3.1
Creating
a
Bank
of
Resumes
The
firststep oftheexperimental designisto generate templatesfortheresumesto besent.The
challengeisto
produce
a setofrealisticand
representativeresumes
without usingresumes
that belongto actualjobseekers.
To
achieve this goal,we
start with resumes of actualjob searchers but alterthem
sufficiently tocreate distinct resumes.
The
alterations maintain the structureand
realismofthe initialresumes
withoutcompromising
theirowners.We
begin withresumes postedon two
jobsearchwebsitesasthebasisforour artificialresumes.11While
the
resumes
postedon
these websitesmay
not be completely representative ofthe average jobseeker,theyprovide apracticalapproximation.12
We
restrictourselves topeople seekingemployment
inour experimentalcities (Boston
and
Chicago).We
alsorestrictourselves to fouroccupationalcategories: sales, administrativesupport,clericalservices
and customer
services. Finally,we
furtherrestrictourselves toresumes
postedmore
than
sixmonths
prior tothestart ofthe experiment.We
purge the selectedresumes
ofthe person'sname
and
contactinformation.During
this process,we
classify theresumes
within each occupational category intotwo
groups: highand
low quality. Injudgingresume
quality,we
usecriteriasuch as labormarket
experience, careerprofile,existence ofgaps in
employment and
skills listed.Such
a classification is admittedly subjective but it ismade
independently ofany race assignmenton
theresumes
(which occurs laterintheexperimentaldesign).To
further reinforcethe qualitygap between
thetwo
setsofresumes,we
add
toeach high qualityresume
asubset ofthefollowingfeatures:
summer
or while-at-schoolemployment
experience,volunteeringexperience,extra
computer
skills, certification degrees, foreign languageskills,honors
orsome
militaryexperience. Thisresume
quality manipulation needs tobe
somewhat
subtle to avoidmaking
a higher qualityjob applicantover-qualified for a given job.
We
try toavoid thisproblem by
making
sure that thefeatures listedabove
are not all
added
at oncetoa given resume. Thisleaves us with a high qualityand
alow quality pool ofresumes.13.
To
minimizesimilarityto actualjobseekers,we
useresumes
from Boston
jobseekers toform
templatesforthe resumes to
be
sentout inChicago
and
used theChicago
resumes
toform
templatesfortheresumes
to
be
sent out in Boston.To
implement
this migration,we
alter thenames
ofthe schoolsand
previousemployers
on
theresumes.More
specifically,foreachBoston
resume,we
usetheChicago resumes
toreplacea
Boston
schoolby
aChicago
school.14We
also usetheChicago
resumes
toreplace aBoston employer
by a
Thesitesarewww.careerbuilder.
com
andwww.americasjobbank.com.12lnpractice,wefound
largevariation inskilllevelsamongpeople postingtheir resumes onthesesites.
In Section4.2and Table3,weprovide a detailedsummaryofresumecharacteristicsby qualitytype. 14
We
Chicago employer
inthesame
industry.We
useasimilarprocedure to migrateChicago
resumestoBoston.15This produces distinct but realistic looking resumes, similarin their education
and
career profiles to thissub-populationofjob searchers.16
3.2
Identities
of
Fictitious
Applicants
The
next step is to generate identities forthe fictitiousjob applicants: names, telephonenumbers,
postaladdresses
and
(possibly) e-mail addresses.The
choice ofnames
is crucialto our experiment.17To
decideon which names
areuniquely AfricanAmerican and which
areuniquelyWhite,
we
usename
frequencydata
calculated from birth certificates ofall babies
born
in Massachusettsbetween
1974and
1979.We
tabulatethese
data by
race to determinewhich
names
are distinctivelyWhite
and which
are distinctively AfricanAmerican.
Distinctivenames
are those that have the highest ratio of frequency in one racialgroup
tofrequencyinthe otherracial group.
As
a checkof distinctiveness,we
conducted
a surveyinvariouspublicareasinChicago.Each
respondentwas
given aresume
with aname
and
asked to assess features ofthe person,one
ofwhich
being race. Ingeneral, the
names
led respondents to readily attribute the expected race for the person but therewere
afew exceptions
and
thesenames
were
disregarded.18The
final list offirstnames used
for thisstudyarereportedinAppendix
Table 1.The
table alsoreportsthe frequency ofthese
names
in the Massachusetts birth certificates data.19The
AfricanAmerican
firstnames
usedinthe experimentareremarkably
common
inthe population. This suggests thatby
usingthesenames
asan
indicator ofrace,we
are actually coveringalargesegment
ofthe AfricanAmerican
population.20Applicantsineach race/sex/ city/resumequalitycellare allocatedthe
same
phone number.
Thisguaran-teesthat
we
canpreciselytrackemployer
callbacksineachofthesecells.The
phone
lineswe
useare virtualones
with
only avoicemailbox
attached toit.A
similaroutgoingmessage
isrecordedon
each ofthevoiceboxes
buteachmessage
is recordedby
someone
oftheappropriateraceand
gender. Sincewe
allocatethe
same
phone
number
forapplicants with differentnames,
we
cannot use a personname
inthe outgoing message.15Notethat
forapplicants with schooling orworkexperience outside of theBostonorChicagoareas, weleavethe school or
employernameunchanged.
16
We
alsogenerate asetofdifferentfonts,layoutsandcoverlettersto furtherdifferentiatethe resumes. Theseareapplied at
thetimetheresumesare sentout.
17
We
chosenameover other potential manipulationsofrace, suchasaffiliationwith a minority group,becausewefeltsuch
affiliations
may
especiallyconveymorethanrace.18Forexample, MauriceandJeromearedistinctivelyAfricanAmerican namesina frequency senseyetarenot perceived as suchby
many
people.19
We
also triedtousemore WhitesoundinglastnamesforWhiteapplicantsandmoreAfricanAmerican soundinglastnames
forAfrican American applicants. The lastnames usedforWhite applicantsare: Baker, Kelly, McCarthy, Murphy, Murray,
O'Brien, Ryan, Sullivan and Walsh.
The
last names used forAfrican American applicants are: Jackson, Jones, Robinson,Washington andWilliams.
20
One
mighthowever wonder whetherthis isan atypical segmentof the African Americanpopulation. InSection 5.1,
we
While
we
do
not expect positivefeedbackfrom
anemployer
totake place via postal mail,resumes stillneedpostal addresses.
We
thereforeconstructfictitiousaddressesbasedon
realstreetsinBoston
and Chicago
using the
White
Pages.We
selectup
to3 addressesineach 5-digitzipcode
inBoston and
Chicago.Within
cities,
we
randomly
assign addressesacrossallresumes.We
alsocreate 8emailaddresses, 4forChicago
and
4 for Boston.21
These
email addresses are neutral with respect toboth
raceand
sex.Not
all applicantsaregiven an email address.
As we
explained above, the email addresses areused almost exclusivelyfor thehigherqualityresumes. This procedureleavesuswith a
bank
ofnames,
phone numbers,
addressesand
e-mailaddresses
which
we
can
assigntothetemplate
resumeswhen
respondingtotheemployment
ads.3.3
Responding
to
Ads
The
experimentwas
carriedon between
July 2001and January 2002
inBoston
and between
July 2001and
May
2002in Chicago.22Over
thatperiod,we
collected allemployment
ads in theSunday
editions ofThe
Boston
Globeand The
Chicago Tribune inthesales, administrative support,and
clericaland customer
servicessections.
We
eliminateany ad where
applicants are askedtocall orappearin person. In fact,most
ads
we
surveyed inthesejob categoriesasked forapplicants to faxin or(more
rarely) mail in theirresume.We
log thename
(when
available)and
contact information for each employer, along withany
informationon
the positionadvertisedand
specificrequirements (such aseducation,experience, orcomputer
skills).We
alsorecord
whether
ornot thead
explicitlystatesthatthe employerisan
equal opportunity employer.For eachad,
we
use thebank
ofresumes tosample
fourresumes (twohigh-qualityand two
low-quality)that fit the job description
and
requirements as closely as possible.23 Insome
cases,we
slightly altertheresumes
toimprove
thequalityofthe match, suchasby
adding
theknowledge
ofaspecificsoftwareprogram.One
ofthehighand
oneofthe lowqualityresumesselected arethendrawn
atrandom
to receiveAfricanAmerican
names, the other highand
lowresumes
receiveWhite
names.
24We
usemale and
femalenames
forsalesjobs,
whereas
we
use nearlyexclusively femalenames
foradministrativeand
clericaljobsto increasecallback rates.25
Based on
sex, race, cityand resume
quality,we
assign aresume
the appropriatephone
number.
We
also select atrandom
a postal address. Finally, e-mail addresses areadded
tomost
ofthehigh qualityresumes.26
The
final resumesareformatted, with fonts, layoutand
coverletterstylechosenatThee-mailaddresses are registeredon Yahoo.com,Angelfire.com or Hotmail.com.
22Thisperiodspans
bothtight andslacklabormarkets. Inour data,thisisapparentascall-backrates (andnumberofnew
ads) dropped precipitously after September 11th, 2001. Interestingly, however, the amount of discrimination wemeasure in
thesetwoperiodsisthesame.
23
Insomeinstances,ourresumebankdoes not havefourresumesthatareappropriatefora givenad. Insuchinstances,we
send onlytworesumes.
Thoughthesamenamesare repeatedly used inourexperiment,weguarantee that nogivenad receivesmultipleresumes
with thesamename.
25Male nameswere used
fora few administrative jobsinthefirst monthofthe experiment.
26
In thefirst month ofthe experiment, a few high quality resumes weresent withoutemail addressand afewlow quality
random.
The
resumes
arethen faxed (orin a fewcasesmailed) tothe employer.27 All inall,we
respond tomore than
1300employment
ads overthe entiresample
periodand send
close to5000
resumes.3.4
Measuring Responses
We
measure whether
a givenresume
elicits a callback or e-mailback
foran
interview. For eachphone
oremail response,
we
use the content ofthemessage
leftby
theemployer
(name
ofthe applicant,company
name,
telephonenumber
for contact) to match,the response to the correspondingresume-ad
pair.28Any
attempt by
employers to contact applicants via postal mailcannotbe
measured
in ourexperiment
sincetheaddresses are fictitious. Several
human
resourcemanagers
confirmed to us that employers rarely, ifever,contact applicantsvia postal mailtoset
up
interviews.3.5
Weaknesses
of
the
Experiment
We
have
already highlighted thestrengths ofthisexperiment
relative toprevious audit studies.We
now
discuss its weaknesses. First, our
outcome measure
is crude, even relative to the previous audit studies.Ultimately,one cares
about whether
an applicantsgets the joband about
thewage
offered conditionalon
getting thejob.
Our
procedure, however, simplymeasures
callbacksfor interviews.To
the extent that thesearch process has even
moderate
frictions,one would
expect that reduced interview rateswould
translateintoreduced joboffers. However,
we
arenotable to translateourresultsintogapsinhiring ratesorearnings.Another weakness
is that the resumesdo
not directly report race but instead suggest race throughpersonal names. This leadsto varioussources ofconcern. First, while the
names
arechosen tomake
racesalient,
some
employersmay
simply notnoticethenames
ornot recognize theirracial content.As
aresult,ourfindings
may
under-estimate the extent ofdiscrimination. Relatedly,becausewe
arenot assigning racebut only race-specific
name,
our results are not representative ofthe average AfricanAmerican (who
may
not
have
such a racially distinct name).29Finally,
and
this isan
issue pervasive inboth
ourstudy
and
the pair-matchingaudit studies,newspaper
adsrepresentonly
one
channelforjob search.As
iswellknown
from
theexistingliterature,social networksare a
common
means
throughwhich
people find jobsand
one
that clearly cannotbe
studied here. Thisomission
would
affect our results if AfricanAmericans
use social networksmore
or ifless discriminatingemployersrely
more on
networks.3027
Aspartofthefaxingprocess,westripal!identifiersfromtheoutgoingfaxtoguaranteethatemployerscould notseethat
allfour faxes originatefromthesamelocale.
28Veryfewemployersused email to contactanapplicant back.
29As AppendixTable
1indicates,the AfricanAmericannames
we
usearehoweverquitecommon among
AfricanAmericans,makingthislessofaconcern.
We
return tothis issue inSection5.1.4
Results
4.1
Is
There
Discrimination?
Table 1 tabulates average callback rates
by
racial soundingness ofnames. Included in bracketsunder
eachrate is the
number
ofresumes
sent in that cell.Row
1 presents ourresults for the fulldata
set.Resumes
with
White
names
have a 10.08 percent chance ofreceiving a callback. Equivalentresumes
with AfricanAmerican
names
have a 6.70percent chance ofbeingcalled back. This representsa difference in callbackrates of 3.35 percentage points, or 50 percent, that can solely
be
attributed to thename
manipulation.Column
4shows
thatthisdifferenceisstatistically significant.31Put
inotherwords,these resultsimply that aWhite
applicantshould expecton
averageone
callbackfor every 10ads sheorheapplies to; on the otherhand,
an
AfricanAmerican
applicantwould
needto applyto 15 different adsto achievethesame
result.32How
large are theseeffects?While
the costofsending additionalresumesmight
notbe
largeperse, this50
percentgap
couldbe
quite substantialwhen
compared
totherate of arrival ofnew
job openings. In ourown
study,thebiggestconstraining factorwas
thelimitednumber
ofnew
job openings each week.Another
way
tobenchmark
themeasured
return to aWhite
name
is tocompare
it to thereturns to otherresume
characteristics. For example, in Table5,
we
willshow
that, atthe averagenumber
ofyears ofexperienceinour sample,
an
extra yearofexperienceincreasesthelikelihood ofacallbackby
.4percentagepoint.Based
on
thispoint estimate,the return toaWhite
name
isequivalent to about 8 additionalyears of experience.Rows
2and
3breakdown
thefullsample
of sentresumes
intotheBoston
and Chicago
markets.About
20 percent
more
resumeswere
sent inChicago than
in Boston.The
average callback rate (across races)is lowerin
Chicago
than in Boston. Thismight
reflect differences in labormarket
conditions across citiesover theexperimental periodor
maybe
differencesin theability oftheMIT
and Chicago teams
ofresearchassistantsinselecting
resumes
thatwere
good matches
for a given helpwanted
ad.The
percentagedifferenceincallback ratesis
however
strikinglysimilaracrossboth
cities.White
applicantsare48 percentmore
likelythan African
American
applicants to receive a callback inChicago and
52 percentmore
likely in Boston.These
racial differencesarestatisticallysignificant in both cities.Finally,
rows
4 to 7 breakdown
the fullsample
into femaleand male
applicants.Row
4 displays theaverageresultsforallfemale
names
whilerows
5and
6break the femalesample
into administrative (row5)and
salesjobs (row 6);row
7displaysthe averageresultsforallmale
names.As
notedearlier,femalenames
were
usedinboth
salesand
administrativejobopeningswhereas
male
names were
used closeto exclusivelyforsalesopenings.33
Looking
acrossoccupations,we
findasignificantracialgap
in callbacksforboth males31
Thesestatistical testsassume independenceofcallbacks.
We
havehowever verifiedthat theresultsstaysignificant whenweassumethatthe callbacks are correlatedeitherattheemployerorfirst
name
level.32ThisobviouslyassumesthatAfricanAmericanapplicantscannotassessa
prioriwhichfirmsaremorelikelyto discriminate againstthem.
33Only about6percent ofallmale resumes weresent
inresponse to an administrative job opening.
(49 percent)
and
females (50 percent).Comparing
males to femalesin sales occupations,we
find a largerracial
gap
among
males (49percent versus25percent). Interestingly,females in salesjobsappear to receivemore
callbacksthan males; however, this (reverse) gendergap
isstatistically insignificantand
economicallysmaller
than
any
ofthe racial gaps discussed above.Rather
than
studying the distribution of callbacks at the applicant level, one can also tabulate thedistribution of callbacks at the
employment ad
level. InTable 2,we
compute
thefraction ofemployers thattreat
White
and
AfricanAmerican
applicantsequally,thefraction ofemployersthat favorWhite
applicantsand
thefraction ofemployersthat favorAfricanAmerican
applicants.Because
we
sendup
to fourresumes
in response to each
sampled
ad, the 3 categoriesabove
can each take 3 different forms.Equal
treatmentoccurs
when
eitherno
applicant gets calledback,one White
and
one AfricanAmerican
get calledback
ortwo Whites
and two
AfricanAmericans
get calledback.Whites
are favoredwhen
eitheronly oneWhite
getscalled back,
two Whites
and no
AfricanAmerican
get calledback
ortwo
Whites and one
AfricanAmerican
get calledback. African
Americans
arefavoredin the othercases.As
Table2 indicates, equaltreatment occursforabout
87.5percent ofthehelp-wanted
ads.As
expected,the
major
sourceofequaltreatmentcomes from
the high fractionofadsforwhich no
callbacks arerecorded(82.5 percentoftheads).
Whites
are favoredby
nearly 9percent oftheemployers,with a majorityoftheseemployers contacting exactly one
White
applicant. African Americans,on
the otherhand,
arefavoredby
only
about
3.7 percentofemployers.We
formally testwhether
thereissymmetry
in thefavoring ofWhites
over African
Americans
and
AfricanAmericans
overWhites by
employers.We
find that the differencebetween
thefractionofemployersfavoringWhites and
the fraction ofemployersfavoringAfricanAmericans
(5.11 percent) is statisticallyverysignificant (p
=
.0000).4.2
Do
African
Americans
Receive
Different
Returns
to
Resume
Quality?
Our
results so far suggest a substantialamount
of discrimination in job recruitment. Next,we
would
liketo learn
more
about
how
employers discriminate.More
specifically,we
askhow
employers respond toimprovements
in AfricanAmerican
applicants' credentials.To
answer this question,we
examine
how
theracial
gap
in callback ratesvariesby
resume
quality.As we
mentioned
in section 3, formost
oftheemployment
adswe
respond
to,we
send four differentresumes:
two
higher qualityand
two
lower quality ones. Table 3 gives a better sense ofwhich
factorsenter into this subjective classification. It displays
means and
standard deviations of themost
relevantresume
characteristics for thefullsample (column
1) aswell asbrokendown
by
race(columns
2and
3)and
resume
quality (columns4and
5). Sinceapplicants'names
are randomized, thereisno
differenceinresume
characteristics
by
race.Columns
4and
5document
theobjective differencesbetween resumes
subjectivelyclassifiedashigh
and
low. Higherquality applicantshave
on average an extra yearoflabormarket
experience,fewer
employment
holes (wherean employment
hole is defined as a period of at least 6months
without areportedjob), are
more
likelyto haveworked
whileat schooland
to reportsome
military experience. Also,higher quality applicants are
more
likelytohavean
email address,tohavereceivedsome
honorsand
tolistsome
computer
skillsand
otherspecialskills(such asacertificationdegreeor foreignlanguageskills)on
theirresume.
Note
thatthehigherand
lower qualityresumes do
not differon average with regard toapplicants'educationlevel.34
About 70
percentofthesentresumes
report acollegedegree.35The
last 5 rows of Table 3show
summary
characteristics of the applicants' zipcode
address.Using
1990
Census
data,we compute
thefractionhigh-schooldropouts, fraction collegeeducatedormore,fractionWhites, fractionAfrican
Americans
and
log(median percapitaincome)
for eachzipcode usedin theexper-iment. Sinceaddressesare
randomized
within cities, theseneighborhood
qualitymeasures
areuncorrelatedwith
race orresume
quality.The
differencesincallback ratesbetween
highand
lowresumesarepresentedin Table4.The
firstthing tonoteisthatthe
resume
qualitymanipulation works: higherqualityresumesreceivehigher callbackrates.As
row
1 indicates,we
recordacallback rate ofmore than
11 percentforWhite
applicantswith a higherqualityresume,
compared
to 8.8 percent forWhite
applicants with lower quality resumes. This is a statisticallysignificantdifference of2.51percentagepoints,or30percent.
Most
strikingly,AfricanAmericans
experiencemuch
less ofan
increase in callback rate forsimilarimprovements
in their credentials. AfricanAmericans
with
higher qualityresumesreceiveacallback 6.99percent ofthe time,compared
to 6.41percentforAfricanAmericans
with lowerquality resumes. This is only a .58 percentagepoint, or 9 percent, increaseand
thisincreaseis notstatisticallysignificant.
Instead of relying
on
the subjective quality classification,Panel
B
directly usesresume
characteristicsto classify the resumes.
More
specifically,we
use arandom
subsample
of one-third of theresumes
toestimate a probit regression ofthe callback
dummy
on
theresume
characteristics listed in Table 3.We
further control for a sex
dummy,
a citydummy,
6 occupationdummies
and
a vector ofjob requirementsas listed in the
employment
ads.36We
then use the estimated coefficientson
theresume
characteristics torank
the remaining two-thirds oftheresumes by
predicted callback.We
classify as "high"resumes
thosethat have
above
median
predicted callback; similarly,we
classify as "low"resumes
those that havebelow
median
predicted callback.As
one
can seefrom
Panel B, qualitatively similar resultsemerge from
thisanalysis.
While
AfricanAmericans
do appear
to significantly benefitfrom higherqualityresumes
under
thisalternativeclassification,they benefit
much
lessthan
Whites.The
ratio of callback rates forhigh versuslowqualityresumes is 1.66forAfricanAmericans,
compared
to 2.81 for Whites.34This
reflects the factthat all sent resumes, whetherhigh or lowquality, arechosen tobegood matches fora given job opening.
35This
varies from about 50 percent fortheclerical andadministrativesupport positions to more than 80 percentfor the executive,managerial andsalesrepresentativespositions.
36Seesection 4.4 formore
detailsontheseoccupationcategoriesandjob requirements.
In
Table
5,we
directly report the results of race-specificprobit regressions ofthe callbackdummy
on
resume
characteristics.We
however
start incolumn
1 withresults for thefullsample
ofsent resumes.As
one
can see,many
oftheresume
characteristics have the expected effecton
the likelihood ofa callback.The
additionofan email address, honorsand
special skills allhave
a positiveand
significant effecton
thelikelihood of
a
callback.37 Also,more
experienced applicants aremore
likely to get called back: at theaverage
number
ofyears ofexperienceinoursample
(8years), each extra yearofexperienceincreasesthelikelihoodofacallback
by about
.4percentagepoint.The
most
counterintuitiveeffectscome
from computer
skills,
which appear
tonegativelypredict callback,and
employment
holes,which appear
to positively predict callback.The same
qualitativepatternsholdincolumn
2where
we
focuson
White
applicants.More
importantly,the estimated returns to
an
email address, additionalwork
experience, honorsand
special skillsappear
economically stronger for that racial group. For example, at the average
number
ofyears ofexperience inour sample, each extra yearofexperienceincreasesthelikelihoodofacall
back
by
about.7percentagepoint.Also,
working
whileatschool, whilenot astatisticallysignificant determinantof callbackin thefullsample,yields positivereturns for
White
applicants.As
might have been
expectedfrom
thetwo
previous columns,we
findthat the estimatedreturnson
theseresume
characteristicsarealleconomicallyand
statisticallyweaker
forAfricanAmerican
applicants(column
3). Infact, alltheestimatedeffectsforAfrican
Americans
arestatisticallyinsignificant,exceptforthe returntospecialskills.
Resume
characteristicsthusappear
lesspredictive of callback ratesforAfricanAmericans
than
theyarefor Whites.To
see thismore
clearly,we
predict callback rates usingeither the regressionincolumn
2 orthe regressionincolumn
3.The
standard deviation ofthe predicted callbackfrom
column
2is.064,
whereas
it is only 0.034from
column
3. Insummary,
employers simplyseem
topay
less attention ordiscount
more
thecharacteristicslistedon
theresumes
with AfricanAmerican
sounding names.Taken
atface value, these results suggest that African
Americans
may
face relatively lower individual incentivestoinvestinhigherskills.38
4.3
Applicants'
Address
An
incidental feature ofour experimental design is therandom
assignment ofaddresstothe resumes. Thisallows usto
examine whether
and
how
an
applicant's residentialaddress, all elseequal, affectsthelikelihoodofacallback. In addition,
and most
importantlyforour purpose,we
canalsoaskwhether
AfricanAmerican
applicants arehelped
more
by
residinginmore
affluentneighborhoods.37Note
that the e-mail addressdummy,becauseitisclose toperfectly correlatedwith the subjectiveresumequalityvariable,
may
inpartcapturesomeotherunmeasuredresumecharacteristicsthatmay
haveledus to categorize a givenresumeas higherquality.
38This of course assumesthatthechanges
inwageoffersassociatedwith higherskills arethesameacrossraces,or at least
not systematicallylarger forAfrican Americans.
We
performthisanalysisinTable6.We
start(columns
1,3and
5)by
discussingtheeffectofneighborhood
of residence across all applicants.
Each
of thesecolumns
reports the results ofa probit regression ofthecallback
dummy
on
a specific zipcode
characteristicand
a citydummy.
Standard
errors are correctedfor clustering of the observations at the
employment ad
level.We
find a positiveand
significant effect ofneighborhood
qualityon
thelikelihoodofacallback. ApplicantslivinginWhiter (column
1),more educated
(column
3) orhigherincome (column
5)neighborhoods
have a higherprobability of receivingacallback. Forexample, a 10 percentage pointincreasein the fraction ofcollege-educated in zipcode ofresidence increases
thelikelihoodofacallback
by
.6percentagepoint.In
columns
2, 4and
6,we
further interact each ofthe zipcode
characteristicabove
with adummy
variable for
whether
the applicant is AfricanAmerican
or not.Each
ofthe probit regressions in thesecolumns
also includesan
AfricanAmerican
dummy,
a citydummy
and an
interaction ofthe citydummy
with
the AfricanAmerican
dummy.
There
isno
evidence that AfricanAmericans
benefitany
more than
Whites from
livingin amore
White,more
educated zipcode.The
estimated interactionsbetween
fractionWhite
and
fraction college educated with the AfricanAmerican
dummy
are economically very smalland
statistically insignificant.
We
do
findan
economicallymore
meaningful effect of zipcode
median income
level
on
the racialgap
incallback; that effect, however, isstatistically insignificant.In
summary,
while the quality of theneighborhood
ofresidence is a significant factor in employers'decision to contact a job applicant, African
Americans do
notappear
to benefitmore than Whites from
livinginbetterneighborhoods.
These
findingsare interesting. Indeed, ifghettosand
bad neighborhoods
areparticularly stigmatizingfor African Americans,
one might have
expected AfricanAmericans
tobe
helpedmore
by
having a "good" address.Our
resultsdo
not lendsupporttothis hypothesis.4.4
Job
Characteristics
Inthissection,
we
turntojobcharacteristicsand
askwhether
discriminationvaries basedon
thespecificjobrequirements oroccupation listed in the
employment
ads.Column
1 ofTable7 givesadescription of thespecificrequirementsstatedinthesample
ofadswe
respond
to.
About
80 percentofthe adsstatesome
form ofrequirement.About
43 percent ofthe ads requiresome
minimum
experience, ofwhich
roughly 50 percent simply askfor"some
experience," 25 percent lessthan
two
yearsand
25 percent at least 3 years of experience.About
44 percent ofadsmention
some
computer
knowledge
requirement,which
can range from Excel orWord
tomore
esoteric software programs.Good
communication
skillsareexplicitlyrequiredinabout
12percentoftheads. Organizationskillsarementioned
7 percentofthe time. Finally, only
about
11 percent ofthe ads list an expliciteducation requirement.Of
these, 8.8 percent require a high school degree, 49 percent
some
college (such as an associate degree)and
therest at leasta 4-yearcollegedegree.39
How
do
these differentjob requirements affect the racialgap
in callback?To
answer
thisquestion,we
show
incolumn
2theresultsof variousprobit regressions.Each
entryin thiscolumn
isthemarginal effectofthe requirement
on
the racialgap
in callback. Specifically, each entrycomes from
a separate probitregression
where
we
regressacallbackdummy
on an
AfricanAmerican
dummy,
therequirementdummy
and
theinteraction oftherequirement
dummy
with the AfricanAmerican
dummy. The
reportedcoefficientisthaton
theinteraction term.The
point estimatesshow
no
consistenteconomic
patternand
are allstatisticallyinsignificant.
Measures
ofjob quality, such as experience,computer
skills or education requirementsdo
not predict the extent ofdiscrimination.
Equally
surprising,communication
or other inter-personal skillrequirements have
no
effecton
thelevelofdiscrimination. In short,discrimination appearstovarylittleby
job requirements.40
Panel
A
ofTable 8 investigateswhether
discrimination varies significantly across occupations.As
we
mention
earlier, the specificsubsectionsoftheSunday
newspapers
help-wantedsectionsthatwe
sample
forthisstudy broadly relate tosales
and
administrativepositions. Thisisreflected in theoccupationaldistri-butionreportedin
column
1.To
form
this classification,we
usethejobdescription listed intheemployment
ad
and
map
it intoone
ofsix following categories: executivesand
managerial occupations; administrativesupervisors; sales representatives; sales workers; secretaries
and
legal assistants; clerical occupations.41 Inthe 19905 percent Census, these occupations accountfor
about
46.5 percent of totalsalariedprivate sectoremployment
inChicago
and
Boston. AfricanAmericans
accountforabout
4percentofemployment
intheseoccupations in
Boston
and
10.5percent in Chicago. For comparison, AfricanAmericans
account for about4.8 percent of total salaried private sector
employment
inBoston and
11.8 percent in Chicago.We
report inthesecondtolastcolumn
ofTable
8average earningsmeasures
foreachoftheseoccupationcategories.
These measures
arecomputed from
the1990
5 percentCensus
forBoston
and
Chicago.More
specifically,
we
first estimatea micro
wage
regression oflog(annual earnings)on
8 educationdummies,
aquadratic in age, a sex
dummy
and
a citydummy.
We
thencompute
themean
residual earnings in eachoccupationcategory.
As
expected, thesampled
occupationcategories areverydifferent.The
firsttwo
(about22%
ofthe data) correspond to better jobson
average that involve,among
other things, the supervisionofother workers; these
two
categories are associated with the highest earnings. Sales representativesand
secretaries (about
50%
ofthe data)involvelesssupervisionbutarestillfairlywellpayingjobs.The
lasttwo
occupations (salesworkers
and
clerical workers) correspond tothe lowestpaying jobs in our sample.Otherrequirements sometimes mentioned include typing skills for secretaries (withspecific words perminute minimum
thresholds)and,morerarely,foreignlanguageskills.
40Otherways ofestimating these
effectsproduce a similar non-result.
Among
otherthings,weconsidered including a citydummy
orestimatingtheeffectsseparatelybycity;wealsoestimatedonesingleprobit regressionincludingallrequirementsatonce.
41
Thesecategories respectivelycorrespondtothe followingCensus 1990occupation codes: 000-042; 303-307; 253-262; 263-302;
234and313; 308-402(excluding313).
The
secondand
thirdcolumns
ofPanelA
ofTable 8 respectivelyreport the callback ratesforWhite and
African
American
applicants in eachofthese occupation categories.Column
4 reports theratioofcallbackrateswhile
column
5 reportsthe difference.The
averagecallback rate across racesseems
to decreasewiththe level ofjob quality.
Whites
receivemore
callbacksthan
AfricanAmericans
in alljobcategories.While
the racial
gap
in callback rates variessomewhat
across occupations,we
cannot reject the null hypothesisthatthe
gap
isthesame
across all categories.We
however
brieflydiscuss the point estimatesby
occupationand
how
they relate toCensus
measuresofaverage earnings.
The
smallestgap
appears tobe
for executivejob positions,where Whites
only facea
33%
higher chancethan
AfricanAmerican
of being called back.The
second lowestgap
is for clericaljobs, with
Whites
being calledback
38%
more
often. Interestingly, thesetwo
job categories areat oppositeends of the job quality spectrum, as captured
by
our earnings measure. This suggestsno
monotonous
relationship
between
jobqualityand
the extentofdiscrimination.The
highest discriminationratiohappens
for administrativesupervisors, thesecond highest job quality category in ourdata. For suchjobs,
Whites
are
64%
more
likelytoget acallback.In thelast
column
ofTable 8,we
reporttherace differences in earningsby
occupation.These
areagaincomputed
from
the 19905percentCensus
forBoston
and
Chicago.The
racialgapsinearningaredefined asthe
White-
AfricanAmerican
differencesinmean
residual log(annualearnings)by
occupation,where
residualearningsareestimatedasexplained above. Interestingly,theredoesnot
seem
tobe
amonotonous
associationbetween
theseestimatedracialgapsin earningsand
theracial gapsin callbacks. For example,executiveand
managerial positions have the lowest racial
gap
in callbacks but thesecond highest racialgap
inearnings;similarly,administrativesupervisorshave the highest racial
gap
incallbacksbut the secondlowestracialgap
inearnings.
On
the otherhand, sales representatives are associated with thesecond-highest racialgap
incallbacks
and
the highest racialgap
in earnings.These
results are interestingand
potentially suggest thatreadily available
measures
ofracialdifferences,suchasthe racialgapin earningsused above,may
notgiveareliable depiction ofdiscrimination patterns in the
economy.
However, because the occupationaldifferencesincallbackarenot significant,these resultsshould not
be
overstated.To
summarize,
we
findthatthelevel ofdiscriminationis remarkably uniformacross avariety ofoccupa-tions
and
job requirements. This occursdespitethefact thatsome
ofthejobswe
study
arehigher earnings,such asmanagerial positions orhigh
end
salesjobs.424.5
Employer
Characteristics
The
finalset ofpossibledeterminantsofdiscriminationwe
consider areemployer characteristics. Collectingsuch
employer
information is not obvious asmost
of this information is not readily available from the42Obviously,wecannot
ruleout thatdifferentpatternsmightexistwhenonemoveseven higherinthe job qualitydistribution.
employment
adswe
respondto. In fact,the only pieceofemployer informationwe
can
directlycollectfrom
the
employment ad
iswhether
ornottheemployer
explicitlystatesbeingan
"EqualOpportunity
Employer."In severalcases,the
name
oftheemployer
isnot evenmentioned
inthead
and
theonlypiece ofinformationwe
can
relyon
isthefaxnumber
which
applicationsmust be
submittedto.We
proceededasfollows. Foremployment
adsthatdo
notlistaspecificemployer,we
usethefaxnumber
to identifythe
company name
viaweb
reverse-lookupservices.Based
on
thecompany
names,
we
usethreedifferentdata sources (Onesource Business Browser,
Thomas
Registerand
Dun
and
Bradstreet Million DollarDirectory, 2001) to track
company
information such as totalemployment,
industryand
ownershipstatus.Using
thissame
setofdatasources,we
alsotrytoidentifythespecificzipcodeofthecompany
(orcompany
branch) that
resumes
aretobe
sentto. Finally,we
use theFederalProcurement
and
Data
Center websitetofindalistof
companies
thathavefederalcontracts.43The
racialdifferenceincallback ratesforthesubsample
where employer
characteristicscouldbe determined
isverysimilarinmagnitude
tothatinthe fullsample.Column
1 ofTable
9reportssummary
employer
characteristics for theavailable data.Sample
sizes foreachvariable arereportedinparentheses.
Twenty-nine
percentofallemployersexplicitlystatethat theyare"Equal
Opportunity
Employers". Eleven percent are federalcontractors and, therefore,might
facegreaterscrutiny
under
affirmativeaction laws.The
averagecompany
sizeisaround
2000employees
butthereisalotof variation across firms. Finally, 73 percent ofthe firms are privately held, 15percent are publiclytraded
and
12 percent are non-profits.The
secondcolumn
ofTable 9presentsthemarginal effectofeach ofthese characteristicson
discrimina-tion.
As
before, each entry correspondsto a separate probit regressionwhere
we
regress acallbackdummy
on an
AfricanAmerican
dummy,
theemployer
characteristicinthatrow and
theinteractionoftheemployer
characteristicwith the African
American
dummy. The
reported coefficient is thaton
the interaction term.First, neitherthe "Equal
Opportunity Employers"
nor the federal contractorsappear
to discriminateless.Infact, each ofthese employercategories is associated with
more
discrimination,eventhough
theseeffectsare noisily estimated. Second,
we
findno
effect ofemployer
sizeon
the degree of discrimination.44 Pointestimates indicatethat publicly traded firms
seem
to discriminatemore,
while not-for-profit organizationsseem
to discriminateless; however, theseeffects areagain noisily estimated.45Panel
B
ofTable
8documents
how
the racialgap
in callbacks varies acrossbroad
industry categories.Our
experiment
coversavarietyof industries.Around
8percentofthe jobsare inmanufacturing, 3 percentin transportation
and communication,
22 percentare inwholesaleand
retailtrade, 8 percent arein financeinsurance
and
real estate, 27 percent are in businessand
personal servicesand
15 percent are in health,43This
website (www.fpdc.gov)isaccurateuptoandincludingMarch21,2000.
44Similar
resultsholdwhen wemeasure employersizeusing atotal salesmeasureratherthananemploymentmeasure.
45
Ofcourse, ourmeasurementofdiscriminationbyfirmtype
may
notbe agoodindicatorofthe fractionofAfricanAmericansactually employed in these firms. For example, "Equal Opportunity Employers"
may
receive a higher fraction of AfricanAmericanresumes. Theiractual hiring
may
therefore lookdifferent from thatofnon "Equal Opportunity Employers" whenoneconsiders thefullsetofresumes theyreceive.