View
216
Download
0
Category
Preview:
Citation preview
8/17/2019 Diversity Policies 2006
1/30
Best Practices or Best Guesses? Assessing the Efficacy of Corporate Affirmative Action andDiversity PoliciesAuthor(s): Alexandra Kalev, Erin Kelly, Frank DobbinSource: American Sociological Review, Vol. 71, No. 4 (Aug., 2006), pp. 589-617Published by: American Sociological AssociationStable URL: http://www.jstor.org/stable/30039011
Accessed: 16/02/2009 17:08
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=asa.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.
American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to
American Sociological Review.
http://www.jstor.org
http://www.jstor.org/stable/30039011?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=asahttp://www.jstor.org/action/showPublisher?publisherCode=asahttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/30039011?origin=JSTOR-pdf
8/17/2019 Diversity Policies 2006
2/30
Best Practices or Best Guesses?
Assessing the Efficacyof Corporate
Affirmative Action and Diversity Policies
Alexandra Kalev
University of California, Berkeley
Frank Dobbin
Harvard University
Erin Kelly
University of Minnesota
Employers have experimented with three broad approaches to promoting diversity. Some
programs are designed to establish organizational responsibilityfor diversity, others to
moderate managerial bias through training and feedback, and still others to reduce the
social isolation of women and minority workers. These approaches find support in
academic theories of how organizations achieve goals, how stereotyping shapes hiring
and promotion, and how networks influence careers. This is thefirst systematic analysis
of their efficacy. The analyses rely on federal data describing the workforces of 708
private sector establishments rom 1971 to 2002, coupled with survey data on their
employmentpractices. Efforts to moderate managerial bias through diversity training
and diversity evaluations are least effective at increasing the share of white women,
black women, and black men in management. Efforts to attack social isolation through
mentoring and networking show modest effects. Efforts to establish responsibilityfor
diversity lead to the broadest increases in managerial diversity. Moreover, organizations
that establish responsibility see better effectsfrom diversity training and evaluations,
networking, and mentoring. Employers subject to federal affirmative action edicts, who
typically assign responsibilityfor compliance to a manager, also see stronger effects
from some programs. This work lays thefoundation for an institutional theory of the
remediation of workplace inequality.
Lists of best practices in diversity man-
agement have proliferated recently.
Everyone seems to have a list, from the Equal
Employment Opportunity Commission (1998)
Direct correspondence o AlexandraKalev, RWJ
Scholars Program, University of California, 140
Warren Hall, MC7360, Berkeley, CA 94720
(akalev@berkeley.edu).The authors thank Ronald
Edwards and Bliss Cartwright of the Equal
EmploymentOpportunityCommission for sharing
their data and expertise;Nicole Esparzaand Leslie
Hinkson or help with datacollection;KevinDobbin,
JohnDonohue, LaurenEdelman,JoshuaGuetzkow,
Heather Haveman, Jerry A. Jacobs, Seema
to the Presidential Glass Ceiling Commission
(1995), the women's business advocacy group
Catalyst (1998), and the Society for Human
Resources Management (2004). These lists are
Jayachandran, awrence Katz, JordanMatsudaira,
John Meyer, TrondPeterson, Daniel Schrage, Paul
Segal, Robin Stryker, Donald Tomaskovic-Devey,
Bruce Western,ChrisWinship,and four anonymous
reviewersfor suggestions; and Randi Ellingboe for
technical and editorial assistance. Supported by
National Science Foundation grant 0336642 and
Russell Sage Foundationgrant 87-02-03 and par-
tiallysupported y the RobertWoodJohnsonScholars
in Health Policy ResearchProgram.
AMERICANOCIOLOGICALEVIEW,006, VOL. 1 (August:589-617)
8/17/2019 Diversity Policies 2006
3/30
590 AMERICANSOCIOLOGICALREVIEW
loosely based on academic heories hatpoint to
causes of workplace inequality ranging from
unwitting bias (Lemm and Banaji 1999) to
dependence on networks for hiring and pro-
motion (Reskin and McBrier 2000). Whereas
there has been a great deal of researchon the
sourcesof inequality, herehas been little on the
efficacy of differentprograms or countering t.
At best, best practices are best guesses. We
know a lot about the disease of workplace
inequality,but not much aboutthe cure.
We examine the effects of seven common
diversity programs-affirmative action plans,
diversity committees and taskforces, diversity
managers,diversity training,diversity evalua-
tions for managers,networkingprograms,and
mentoringprograms-on the representation f
white men, white women, black women, and
black men in the managementranksof private
sector firms. Each of these programsmay well
increase diversity.To date, therehas been little
evidence one way or the other.This is surpris-
ing given the popularity and cost of the pro-
grams. Ourcontribution s to bringto bearrich
new data, o theoretically istinguish hree ypes
of diversityprograms,and to show that organi-
zational structures llocatingresponsibility or
changemaybe moreeffective hanprograms ar-
geting either managerialbias or the social iso-
lation of disadvantaged roups.
Previous empirical studies of antidiscrimi-
nation and diversityprogramshave been limit-
ed by data constraints. Economists first
comparedemployers who are subject to affir-
mative action requirementswith those who are
not (Ashenfelterand Heckman1976; Heckman
and Wolpin 1976; Leonard 1984). They lacked
data on employer programs. Sociologists and
economistsstudyingemployerprograms xam-
ine data at one or two points in time (but see
Baron,Mittman, ndNewman 1991), analyzing
the effects of some programswithout account-
ing for others.These studies ndicate hat some
programsmay be effective,buttheir indingsare
inconsistent (Baron et al. 1991; Edelman and
Petterson 1999; Holzer and Neumark 2000;
Konrad and Linnehan 1995; Leonard 1990;
Naff and Kellough 2003). Gender and racial
segregationhas declined remarkably ince the
1970s, when employers first adopted antidis-
crimination programs (Jacobs 1989a; King
1992;Tomaskovic-Devey t al. 2006), but there
is no hardevidence thatthese programsplayed
a role.
We obtained he federal establishment-level
data that economists have used (i.e., the annu-
al EEO-1 reports hat privatesector establish-
ments submit to the Equal Employment
OpportunityCommission [EEOC]). We then
surveyed sampleof theseestablishments n the
history of their personnel and diversity pro-
gramsso thatwe could analyzeprogram ffects
on diversity.
A strengthof the EEO-1 reports s that they
detail annual employment by race, ethnicity,
and gender n all mediumand largeprivate ec-
tor workplaces.A limitation s that they cover
only nine broad ob categories, collapsing into
management ll jobs above that of first-line
supervisor Baronand Bielby 1985; Smith and
Welch 1984). We know frompreviousresearch
thatwomen andAfricanAmericansare crowd-
ed in the lowest ranksof management.Even as
women moved into management n the 1970s
and 1980s, womenmanagers ontinued o trail
their male counterparts n both earnings and
authority Jacobs 1992). Thus our analyses
indicatewhich diversityprogramshelp women
and African Americans move at least into the
bottom ranksof managementand, mportantly,
which do not. They cannot tell us whetherany
of these practiceshelp women andminorities o
move into the executive ranks.
We find a clearpattern n the data.Structures
establishing responsibility (affirmative action
plans, diversitycommittees,and diversitystaff
positions) are followed by significant increas-
es in managerialdiversity.Programs hattarget
managerial tereotyping hrougheducationand
feedback(diversity rainingand diversityeval-
uations) are not followed by increases n diver-
sity. Programs that address social isolation
among women andminorities networkingand
mentoringprograms)are followed by modest
changes. The effects of these initiatives vary
across groups, with white women benefiting
most, followed by black women. Black men
benefit least. We also find that responsibility
structuresmake training,performanceevalua-
tions, networking, and mentoring programs
more effective. Federal affirmative action
requirements,which typically lead to assign-
ment of responsibility or compliance,also cat-
alyze certainprograms.
8/17/2019 Diversity Policies 2006
4/30
CORPORATEAFFIRMATIVEACTION AND DIVERSITYPOLICIES 591
These findings support an institutional the-
ory of inequality remediation that builds on key
precepts of organizational sociology. As Weber
(1978 [1968]) argues, executives must appoint
specialists and give them authority to achieve
specialized goals. Thus, remedies targeting indi-
vidual bias or network isolation may be less
effective than remedies that establish responsi-
ble parties. As neo-institutionalists (Meyer and
Rowan 1977) note, new programs decoupled
from everyday practice often have no impact.
Therefore, appointing a manager or committee
with responsibility for change is likely to be
more effective than annual diversity training,
periodic diversity evaluations, or decentralized
networking and mentoring programs. As struc-
tural theorists of organizational inequality claim
(Baron 1984), there is more to segregation than
rogue managers exercising bias. Thus, appoint-
ing special staff members and committees to
rethink hiring and promotion structures may be
more effective than training managers not to ask
their secretaries to make coffee, and not to
exclude minorities from football pools.
The argument that organizations should struc-
ture responsibility for reducing inequality may
seem commonsensical, but today's popular
diversity programs often focus on changing
individuals. In the academy generally and in
management studies particularly, methodolog-
ical individualism now holds sway. Theorists
prescribe solutions that change incentives for,
and beliefs of, individuals with the idea that
most problems of management are problems
of motivation rather than structure. Thus the
most popular program that is not federally man-
dated is diversity training, designed to attack
bias. Managerial bias is also the target of diver-
sity evaluations that offer feedback to man-
agers. Networking and mentoring programs
may appear to operate at the collective level, but
they are designed to fix a lack of specific
human and social capital in individual workers.
Next, we describe the three categories of
diversity practices, link them to theories of
inequality, and summarize the (scant) evidence
about the effects of workplace antidiscrimina-
tion programs. Then we review the research on
the effects of the Civil Rights Act and presi-
dential affirmative action edicts on employ-
ment-hitherto the main body of research on the
effectiveness of antidiscrimination measures.
After a discussion of data and methods, we
present the results from analyses of white men,
white women, black women, and black men in
management.
THREE APPROACHES TO INCREASING
MANAGERIAL DIVERSITY
Scholars often presume that practices designed
to attack known causes of inequality actually
will reduce it, as Reskin (2003) argues, making
a leap of faith between causes and remedies.
Thus, for example, although we know from
experimental psychology that unconscious bias
is endemic, and likely contributes to workplace
inequality, we can only hope that the prevailing
treatments-diversity training and diversity
evaluations-diminish inequality. Under-
standing the cause of malaria and understand-
ing its treatment are two different things.
Whether a prescription for inequality is effec-
tive is an inherently empirical question. Current
prescriptions are not based in evidence.
Our goal is to take a first step toward devel-
oping an empirically based theory of remedia-
tion for organizational inequality. We sketch
three mechanisms for remediating workplace
inequality rooted in different social science lit-
eratures and discuss the popular human
resources (HR) measures thought to put these
theories to work. One mechanism, based in
arguments from Max Weber and organization-
al institutionalists, is the creation of special-
ized positions as the way to achieve new goals.
Another mechanism, based in theories of stereo-
typing and bias, involves training and feedback
as the way to eliminate managerial bias and its
offspring, inequality. A third mechanism, based
in theories of social networks, involves pro-
grams that target the isolation of women and
minorities as a way to improve their career
prospects.
ORGANIZATIONALHANGE: TRUCTURESF
RESPONSIBILITY
We begin with a canonical insight from orga-
nizational theory. Organizational sociologists
and psychologists find that workers ignore
newly announced organizational goals and con-
tinue to pursue old goals with old routines. The
decoupling of formal goals and daily practice
may occur because individuals face information
overload, and thus stick to the familiar, or
8/17/2019 Diversity Policies 2006
5/30
592 AMERICANSOCIOLOGICALREVIEW
because the old ways of doing things have been
imbued with meaning and value over time
(Orton and Weick 1990; Selznick 1949).
Institutionalistsargue that decoupling is com-
mon in programs responsive to regulatory
demands, uch as civil rightsprograms Dobbin
et al. 1988; Edelmanand Petterson1999; Scott
2001; Sutton and Dobbin 1996). Thus, for
instance,academicdepartments aveabandoned
the old-boysystem of hiring n favorof open ob
advertisement,but department hairs still ask
their pals for leads. Some arguethat managers
may simply not perceive t as in their nterest o
promote gender and racial integrationof jobs
(Jacobs 1989b).Decoupling s particularlyike-
ly when there is no office or expertto monitor
progress, as Max Weber(1978 [1968]) hinted
when he argued hat executives should appoint
specialists to pursue specialized goals.
If Weberand the institutionalists re correct,
where diversity efforts are everyone'srespon-
sibilitybut no one'sprimary esponsibility,hey
aremore ikelyto be decoupled. n organizations
that do not assign responsibility for diversity
goals to a specific office, person, or group,
these goals may fall by the waysideas line man-
agers juggle competing demands o meet pro-
duction quotas, financial targets, and the like
(Edelman 1990; Meyer and Rowan 1977).
Scholars (Reskin 2003; Sturm2001) and con-
sultants (Winterle 1992) alike advise ongoing
coordination and monitoring of diversity
progress by dedicated staff members or task
forces. Three common approaches an be used
to establish responsibilityfor diversity,as dis-
cussed in the following sections.
RESPONSIBILITYAND AFFIRMATIVEACTION
PLANS. ssign responsibility or setting goals,
devising means, and evaluatingprogress;this
was Weber'sadvice to bureaucrats. he agency
LyndonJohnson et up in 1965 to monitoraffir-
mativeactionamong ederal ontractors ncour-
aged this approach. In 1971, the Office of
Federal Contract Compliance (OFCC, which
later gained a P for programs o become
OFCCP)orderedcontractors o write affirma-
tive action plans in which they annuallyevalu-
ate their own workforces,specify goals for the
fair representationof women and minorities
based on labor market analyses, and sketch
timetables for achievement of these goals
(Shaeffer 1973:66).
The order also specifies that firms should
assign responsibility o a staff member: He or
she must have the authority, esources,support
of and access to top management o ensurethe
effective implementation of the affirmative
action program (U.S. Department of Labor
2005). By collecting and reviewing ocal infor-
mation annually, he affirmativeaction officer
can track underutilization of women and
minorities and keep managers nformedabout
their departments' progress (Linnehan and
Konrad 1999:410; Reskin 2003:13) or initiate
constructivedialogue about making further
progress(Sturm2001).
The few studies hatexamineeffects of affir-
mative action plans are inconclusive. Baron et
al. (1991), studying annual data from 89
California state agencies between 1975 and
1981, found that, all else being equal, agencies
with affirmativeaction programsmade signif-
icantly slower progress in gender desegrega-
tion of jobs. Yet those agencies were more
integratedoriginally,so it may be that preex-
isting affirmativeaction programshad left lit-
tle roomfor improvementsee also Edelman nd
Petterson 1999:126; Leonard 1990:65). In a
study of 3,091 federal contractorswith affir-
mative action plans JonathanLeonard 1985b)
shows that the goals employers set for hiring
white women, blackwomen, andblackmen did
have positive effects, although the goals were
wildly optimistic. Goals apparentlydo not act
as quotas because virtually no employer ever
achieves its writtengoals.
Federal ontractors rerequired o writeaffir-
mative action plans, but contractor tatus does
not correspondperfectlywith the presence of a
plan. Many contractors ail to write plans or to
update hem (Bureauof NationalAffairs 1986;
Leonard 1990:55). Up to one fourth of firms
with affirmative ctionplansarenot contractors.
They createplans to bid for contractsor to set
diversity goals (Bureau of National Affairs
1986; Reskin 1998). In our sample,7 percentof
contractorsnever had a plan, and 20 percentof
firms thathad neverhad a contractwroteplans.
OVERSIGHTVIA STAFFPOSITIONS ND DEPART-
MENTS. ollowing the classic bureaucratic dic-
tum (Weber 1978 [1968]), some organizations
appoint ull-time taffmembersor createdepart-
mentsto monitordiversity nsteadof leaving he
task to line managersor assigning it to staffers
8/17/2019 Diversity Policies 2006
6/30
CORPORATEAFFIRMATIVEACTION AND DIVERSITYPOLICIES 593
with otherresponsibilities.As a newly appoint-
ed diversity manager in a high tech company
explained o us in 2001: As he organization as
started o grow, hey realized hey neededsome-
one in there to really pay attention o affirma-
tive action and compliance and ... efforts on
diversity.... So the position was createdat the
beginning of this year.
Big militarycontractorswere the first to cre-
ate special positions, in the wake of Kennedy's
initialaffirmative ctionorder n 1961. Edelman
and Petterson 1999) show that equal opportu-
nity departmentsdo not increase gender and
racial diversity on their own, but that they do
expand diversityrecruitmentprograms,which
in turn mprovediversity.We includea measure
for recruitmentprograms o isolate the effects
of diversitystaff positions.
OVERSIGHTAND ADVOCACYVIA COMMITTEES.
From the late 1980s, experts have advised
employers o appointdiversitycommitteesand
task forces comprising people from different
departments, professional backgrounds, and
managerial levels. Committees typically are
charged with overseeing diversity initiatives,
brainstorming o identify remedies, and moni-
toring progress.The diversitytask force at the
accounting and consulting giant Deloitte &
Touche, for instance, createda series of ongo-
ing groupsresponsible or analyzing he gender
gap, recommending emedialsteps, and estab-
lishingsystems ormonitoring esultsandensur-
ing accountability Sturm2001:492).
These three strategies share a focus on
responsibility.An organizationwith any one of
these has assignedresponsibility or progress o
a person or group-an affirmativeaction offi-
cer, a diversity manager or department,or a
committee or task force. That person or group
monitorsprogressregularly.Affirmativeaction
officers also write explicit annual goals for
progress,as do some staffers and committees.
BEHAVIORALHANGE: EDUCING IAS
THROUGH DUCATIONNDFEEDBACK
Social psychologists trace inequality to bias
amongmanagers.Stereotyping s a natural og-
nitive mechanism. It is inevitable given our
innocent endency o makeassociationsbetween
categories and concepts (Gorman 2005;
Heilman 1995; Lemm and Banaji 1999). The
implicit associations we make between race,
gender,ethnicity,and social roles can have the
effect of reproducing existing patterns of
inequality (Jost, Banaji, and Nosek 2004).
Managers may unwittingly select women for
jobs traditionally dominated by women and
men for obs dominatedby men, with the effect
of preserving between-group differences.
Moreover in-group preference is widespread
(TajfelandTurner1979) and may likewise con-
taminate managerial judgment (Baron and
Pfeffer 1994; Reskin 2000). Rosabeth Moss
Kanter 1977) sketches he earlyresearchon in-
group preference to support her theory of
homosocial reproduction-white men promot-
ing their clones. Kanterargues that managers
preferto hire their own for reasons of commu-
nication and trust.
Twocorporatenitiatives re hought o count-
er stereotyping and in-group preference.
Diversitytraining s thought o make managers
awareof how bias affectstheiractionsandthose
of subordinates. Diversity evaluations are
thought to provide managers with feedback
showing the effects of their decisions on diver-
sity.
EDUCATIONVIA DIVERSITYTRAINING.Social
psychological researchshows that giving peo-
ple informationabout out-groupmembersand
about tereotypingmay reducebias (Fiske 1998;
Nelson, Acker, and Melvin 1996). Diversity
trainingprovidesmanagerswith such informa-
tion. It can be traced to the equal opportunity
sensitivity rainingprograms hata handfulof
major corporations put together in the mid-
1970s in response to the first equal opportuni-
ty consent decrees and court orders (Shaeffer
1973). By the late 1980s, quite a few corporate
trainersandpsychologistshad developed rain-
ing modules designedto familiarizeemployees
with antidiscriminationaw, to suggest behav-
ioral changes that could address bias, and to
increase culturalawarenessand cross-cultural
communication Bendick, Egan, and Lofhjelm
1998).
Employersusually offer trainingeither o all
managersor to all employees. We look at the
effects of training offered at least to all man-
agers.Some studiesof diversity raining uggest
that it may activate rather than reduce bias
(Kidder et al. 2004; Rynes and Rosen 1995;
Sidanius,Devereux,andPratto2001). Research
8/17/2019 Diversity Policies 2006
7/30
594 AMERICANSOCIOLOGICALREVIEW
on diversity training programs has seldom
explored heireffectson workforce omposition,
but one study of federal agencies (Naff and
Kellough 2003) did show that a broaddiversi-
ty program had a negative effect on the pro-
motion of minorities (Krawiec2003:514).
FEEDBACKVIA PERFORMANCE VALUATIONS.
Feedback s thought o reducebias by directing
managerial attention and motivation (Reskin
2003:325). Laboratory xperimentsshow that
when subjects know that their decisions will
be reviewedby experimenters, hey show lower
levels of bias in assigning jobs (Salancik and
Pfeffer 1978; Tetlock 1985). Evaluatingman-
agers on their diversity performance creates
oversight and provides feedback. As early as
1973, the HarvardBusiness Review noted that
as one criterionof a line manager'sperform-
ance appraisal, ome companieshave included
his success in effectively implementingequal
opportunity programs (Fretz and Hayman
1973:137). By the mid-1980s, a study of nine
exemplary firms found that managers n each
firm received regular equal opportunityper-
formanceevaluations Vernon-Gerstenfeldnd
Burke 1985:59-60). To ourknowledge,no stud-
ies assess the effects of diversityevaluations.
TREATINGSOCIALISOLATION:NETWORKING
AND MENTORING
MarkGranovetter1974) brought nsightsabout
social networks,pioneeredby both sociologists
and psychologists, to the study of how people
find jobs. Students of inequality have since
speculated that differential network contacts
and differentialresources accruingfrom these
contacts may explain part of the continuing
inequality between whites and blacks, and
betweenmen andwomen (Blair-Loy 001; Burt
1998; Ibarra 1992, 1995; McGuire 2000;
Petersen, Saporta, and Seidelm 1998). White
men are more likely than others to find good
jobs through network ties because their net-
works are composed of other white men who
dominate the upper tiers of firms (Burt 1998;
Reskin and McBrier 2000, but see Fernandez
and Fernandez-Mateo 2006; Mouw 2003).
Social networksalso encourage rust, support,
and nformal oaching BaronandPfeffer 1994;
Castilla 2005; Kanter 1977). Networking and
mentoringprogramsdesigned specifically for
women and minorities are thought to provide
useful contactsand nformationThomas2001).
Both types of programswere pioneered in the
1970s and then revived in the 1990s as part of
diversitymanagement fforts Wernick 994:25;
Winterle 1992:21).
NETWORKING ROGRAMS.
Diversitynetwork-
ing programs or women andminoritiesvary in
structure. ome takethe form of regularbrown-
bag lunchmeetings,whereasothers nclude av-
ish national conferences (Crow 2003). These
programsmay be initiatedby employees or by
HR managers.They provide a place for mem-
bers to meet and share informationand career
advice. Some networks also advocate policy
changes,such as those involving amilypolicies
and domestic-partner benefits (Briscoe and
Safford2005). Althoughnetworkingmay occur
without any organizational mpetus, we exam-
ine formal networkingprograms hat employ-
ers support hrough elease ime forparticipants,
meeting space, funding,newsletters,and email
lists.
MENTORINGROGRAMS.n 1978, the Harvard
Business Review published an article titled
EveryoneWho Makes It Has a Mentor hat
made mentors a must-have for aspiring man-
agement trainees (Lunding, Clements, and
Perkins1979;see also Roche 1979). Proponents
of formal mentoringprogramsarguethat they
can level the playing field, giving women and
minorities he kinds of relationships hat white
men get through the old-boy network.
Mentoringprogramsmatch aspiringmanagers
with senior mentors,with the two meeting for
career counseling and informal advice.
Empirical tudies,such as BurkeandMcKeen's
(1997) survey of universitygraduates, uggest
a relationship between mentoring and career
success among women, but do not rule out the
possibility that ambitious women seek men-
tors. One study of randommentor assignment
within a single firm found that, in general,
mentees have improved social networks and
tacticalknowledge,whichmayhelp theircareers
(Moore2001). Othershavefound hatcross-race
mentoring relationships often fail (Thomas
2001), and that same-sex mentoring does not
have a positive effect on job placement n aca-
8/17/2019 Diversity Policies 2006
8/30
CORPORATEAFFIRMATIVEACTION AND DIVERSITYPOLICIES 595
demic departments of economics (Neumark
and Gardecki 1996).
ADVERSE FFECTS FDIVERSITYRACTICES
Some argue that affirmativeaction and diver-
sity programs can backfire (Bond and Pyle
1988; Linnehanand Konrad1999). First,exec-
utives may believe that women and minorities
benefit from reverse discriminationand thus
may not deserve their positions (Heilman,
Block, and Stathatos 997;but see Taylor1995).
Second,because of the elusive natureof cogni-
tive bias, conscious attemptsat thoughtregu-
lation -such as diversity rainingand diversity
evaluations- may even backfire, leading to
exaggerated stereotypingunder conditions of
diminished capacity, or when self-regulation
efforts are relaxed (Nelson et al. 1996:31).
Indeed,management onsultants ndresearchers
find mixed reactionsto diversitymanagement
among white males, who report that they are
tiredof being made to feel guilty in every dis-
cussion of diversity... of being cast as oppres-
sors (Hemphill and Haines 1997). Third,
coworkers and executives may have negative
reactions when they perceive minorities as
attempting o obtain power by individual and
collectivemeans Ragins 1995:106),andexec-
utives may fear that networking will lead to
union organizing(Bendick et al. 1998; Carter
2003; Friedman and Craig 2004; Miller
1994:443; Society for Human Resources
Management2004). Finally,some studies find
thatraciallydiversework groups communicate
less effectivelyandare ess coherent Baughand
Graen 1997; Townsendand Scott 2001; Vallas
2003; Williams and O'Reilly 1998). Taken
together, this research suggests that diversity
programs may inhibit management diversity,
particularly or blacks.
THE CIVILRIGHTS ACT, AFFIRMATIVE
ACTION EDICTS,AND DIVERSITY
PRACTICES
Although there is little researchon the effects
of corporate iversityprograms, he Civil Rights
Act and presidentialaffirmativeaction orders
havebeen shownto increasediversity.The Civil
RightsAct covers virtuallyall employers,mak-
ing research n its effectsdifficult Donohueand
Heckman 1991). The effects of presidential
affirmativeaction orders can be examined by
comparing ederal contractors ubjectto these
orders with noncontractors.Six studies using
EEOC data for periods of 4 to 6 years between
1966 and 1980 show that black employment
grew more quickly among contractors
(Ashenfelter ndHeckman1976;Goldsteinand
Smith 1976; Heckman and Payner 1989;
HeckmanandWolpin1976).Affirmativeaction
hadnegligibleeffects on whitewomen Leonard
1989:65). Contractor ffects on blacks, espe-
cially black women, declined from the early
1980s (Leonard 1990:58), coincident with the
Reaganadministration'solicy of deregulation.
These studies do not look at whether federal
contractors increased black employment by
adoptingantidiscrimination ractices.The two
exceptions are a study by Leonard (1985b)
showing that employers who set high recruit-
ment goals see more change and a study by
Holzer and Neumark (2000) showing that
employers subject to affirmative action law
expandrecruitment ffortsandhiremore appli-
cants from disadvantaged roups.We examine
the effect of affirmative action orders and
explore he possibility hatbeing subject o such
orders by being a federal ontractor)enders he
seven diversityprogramsmore effective.
In summary,we expect the differentsorts of
diversityprograms o vary n efficacy.If assign-
ing organizational esponsibility s more effec-
tive than targeting he behaviorof individuals,
then affirmative action plans, diversity com-
mittees, and full-timediversity taff will be fol-
lowed by broader increases in diversity than
will eitherdiversity rainingand diversityeval-
uations,or networking ndmentoring rograms.
By the same logic, the latter ourprogramsmay
be more effective when implemented n organ-
izations with responsibilitystructures.Finally,
we examinewhether ffirmative ctionoversight
rendersprogramsmore effective.
ALTERNATIVE OURCES OF CHANGE
IN THE MANAGERIALWORKFORCE
We include n the analysesother actors hought
to affect management diversity. We cannot
include factors hatdo not vary with time, such
as industry r location,becauseourfixed-effects
models account for such stable traits.
8/17/2019 Diversity Policies 2006
9/30
596 AMERICANSOCIOLOGICALREVIEW
LEGALENVIRONMENT
Legal enforcement, through OFCCP compli-
ance reviews, lawsuits, and EEOC charges,
should increase employers'hiring and promo-
tion of women and minorities (Baron et al.
1991:1386; Donohue and Siegelman 1991;
Kalev and Dobbin forthcoming;Leonard1984;
Skaggs 2001).
ORGANIZATIONALTRUCTURES
Organizational ize andthe availability f man-
agerial obs createnew opportunities Baronet
al. 1991), but also more competition. Konrad
and Linnehan (1995) and Leonard (1990:52)
find that ncreaseddemand or managers avors
white women, but not African Americans.
Unionization tends to preserve segregationby
favoringold timers hrough eniorityprovisions
(Blau and Beller 1992; Milkman 1985; but see
Kelly 2003; Leonard 1985a). Formalization f
personnel systems can reduce favoritism
(Dobbinet al. 1993;ReskinandMcBrier2000),
although t also can create separatecareer ra-
jectories or different roups BaldiandMcBrier
1997; Baron and Bielby 1985; Elvira and
Zatzick 2002). Legal counsel may sensitize
employers o diversity n promotiondecisions,
and recruitment ystems targetingwomen and
minoritiescan increasediversity Edelmanand
Petterson 1999; Holzer and Neumark 2000).
Finally,work/family olicies mayremoveobsta-
cles to the promotion of women (Williams
2000).
TOPMANAGEMENTOMPOSITION
The diversityof the top management eam may
affect managerial hires through homosocial
reproductionor social closure (Kanter 1977;
Tomaskovic-Devey1993).
LABORMARKET NDECONOMICNVIRONMENT
Firms can more easily increase managerial
diversitywhen internaland external aborpools
are diverse (Cohen, Broschak, and Haveman
1998; Shenhavand Haberfeld 1992). Demand
for workers romunderrepresentedroupsmay
be higher in industrieswith more federal con-
tractors. n hardeconomic imes,blackmen, and
to a lesser extent women, are more vulnerable
than white men to being laid off (Elvira and
Zatzick 2002; Kletzer 1998). Finally,growing
industries can offer more attractive obs, and
both women and minorities have historically
been relegated o less attractive ectors (Reskin
and Roos 1990:298).
DATA AND METHODS
We conducted a fixed-effects analysis of lon-
gitudinaldataon the workforcecompositionof
708 establishments o assess changes in mana-
gerialcompositionafter he adoptionof each of
seven diversity practices. The data cover the
period1971-2002. Fixed-effectmodels account,
implicitly, or organizations'unobservedchar-
acteristics that do not vary over time and that
may affect diversity.
EEOCDATA
The Civil Rights Act of 1964, as amended,
requiresprivateemployerswith more than 100
employees and government contractors with
more than 50 employees and contracts worth
$50,000 to file annual EEO-1 reports. These
reportsdetail the race, ethnicity,and gender of
employees in nine broad occupational cate-
gories. There are no better data on workforce
composition for a methodological iscussionon
using EEO-1reports, ee Robinsonet al. 2005).
We obtainedthe data from the EEOC through
an IntergovernmentalersonnelAct (IPA)agree-
ment.
Some argue hatemployersreclassifiedjobs
in the 1970s, moving women and minorities
into managementcategories to improve their
federal reports (Smith and Welch 1984).
Leonard 1990:53) notes that purereclassifi-
cation would cause black losses in the lower
occupations[in the EEO data],which is gener-
ally not observed. Jacobs (1992:298) shows a
declining gender earningsgap consistent with
realprogress,noting hat thepredominantrend
has been toward eal, f slow progress nto man-
agementon the partof women. n our sample,
few firms show sudden ncreases or women or
blacks in management,but we checked results
for robustnessby eliminating hese cases, and
the results did not change. We also eliminated
establishment-year pells from before 1990, as
discussed later,and the findings held up.
8/17/2019 Diversity Policies 2006
10/30
CORPORATEAFFIRMATIVEACTION AND DIVERSITYPOLICIES 597
ORGANIZATIONALURVEY ATA
We drew a random sample of establishments
fromthe EEO-1 database or our organization-
al survey. For that sample, we constructed a
dataset comprising all EEO-1 reports for the
years 1971-2002, interpolating or the missing
years of 1974, 1976, and 1977. Establishments
enter the datasetwhen they begin filing EEO-
1 reports.To ensure hatwe wouldbe ableto fol-
low establishmentsover time, we chose half of
the sample from establishments hat had been
in the dataset ince 1980 andhalf from hosethat
had been in the dataset since 1992. We also
stratifiedby size, selecting 35 percentof estab-
lishments with fewer than 500 employees in
1999, and by industryto represent he manu-
facturing, ervice,and rade ectors.We sampled
from food, chemicals, computer equipment,
transportationquipment,wholesale rade, etail
trade, nsurance,business services, and health
services. Corporatediversitycan be influenced
by acquisitions, pin-offs,andplantclosings, so
we sampledestablishments, electing no more
than one per parent irm.
We conducted a longitudinal survey of
employment practices at each establishment
covering he years 1971-2002, in collaboration
with the PrincetonSurveyResearchCenter.We
drewon the experiencesof otherswho had con-
ducted organizational urveys of employment
practices (particularlyKalleberg et al. 1996;
Kelly 2000; Osterman 1994, 2000). We com-
pleted 833 interviews,for a responserate of 67
percent, which compares favorably with the
rates of those other organizational urveys. In
preparation,we conducted41 in-person nter-
views with HR managers rom randomly am-
pled organizationsn fourdifferent egions,and
20 pilot phone interviews. Data from those
interviews are not included in the analyses
reported n this discussion.
We began by writing to the HR director at
each establishment.We asked for permission o
conduct an interview and for the name of the
person who could best answerquestions about
the establishment's istoryof HR practices.The
typical ntervieweewas an HR managerwith 11
yearsof tenure.We scheduledphone interviews
at the convenience of the interviewees, and
explained in advance the nature of the infor-
mation needed. We asked whether the estab-
lishmenthad everused each personnelprogram,
when it was adopted,and whetherand when it
hadbeen discontinued. rogram iscontinuation
was rare.When a respondentcould not answer
a question, we sent a copy of that question by
email or fax, askedthat she consult recordsand
colleagues, and called back to fill in the blanks.
During our in-personpilot interviews, respon-
dents routinelypulled out manualswith copies
of policies and lists of adoption and revision
dates. Nonetheless, because responses about
events long past may be inaccurate,we repli-
cated the analyses using only establishment-
year spells for 1990 to 2002, as discussed later.
We matched survey data for each establish-
ment with annual EEO-1 records, creating a
datasetwith annual establishment-year pells.
After excluding 10 cases that had EEO-1 data
available or fewer than 5 years, 13 cases with
excessive numbersof missing values for EEO-
1 or survey data,and 102 cases thatwere miss-
ing the adoption date for at least one key
program,our final dataset included 708 cases
and 16,265 establishment-year cells, with a
medianof 25 years of dataper establishment,
minimum of 5 years, and a maximum of 32
years. We collected dataon national,state, and
industry mployment rom he Bureauof Labor
Statistics.
Because of our stratified sampling design
and the response pattern,we were concerned
thatrespondentsmight not represent he popu-
lationof establishments hatfile EEO-1reports
in the sampled industries. We constructed
weights based on the inverseprobability hatan
establishment rom each stratum industryby
size and by time in the EEO-1 dataset)would
complete the survey.We replicatedall reported
analysesusing weights,andthe resultsremained
intact.We reportunweightedresults in the fol-
lowing discussion (Winshipand Radbill 1994).
We also were concerned that employers who
refusedto participatemight systematicallydif-
fer, on factors affecting diversity, from those
who participated.We included in the models
predictedvalues from a logistic regressionesti-
mating the probabilityof response (Heckman
1979). This did not change our results.
Covariates n that model were industry,estab-
lishment status (headquarters, ubunit, stand-
alone status),size, contractor tatus,managerial
diversity, nd contactperson'sposition.The last
variablewas obtained n the initial contact,the
others from the EEO-1 data.
8/17/2019 Diversity Policies 2006
11/30
598 AMERICANSOCIOLOGICALREVIEW
DEPENDENTVARIABLES
The dependentvariablesare the log odds that
managersare white men, white women, black
women, and black men . For each group, odds
are calculated as the proportionof managers
from that group divided by the proportionnot
from that group (proportion/(l- proportion)).
Figure 1 presents he trends, n percents, n our
sample. Between 1971 and 2002, management
jobs held by white men decline from 81 to 61
percent in the average establishment.
Management obs held by white women rise
from 16 to 26 percent, whereas those held by
black women rise from 0.4 to 2 percent, and
those held by black men rise from 1 to 3.1 per-
cent. There also is a significant rise in the rep-
resentationof othergroups,notablyHispanics,
during his period,which s why the percentages
do not sum up to 100 percent.
Black women and men showed dramatic
changes in their proportions in management
relative o the baseline,quadrupling nd ripling,
respectively,but saw small changes in percent-
age points. Because the absolute changes for
blacks are relatively small we log the depend-
ent variables.We use log odds, rather han log
proportion,because the distribution s close to
normal (Fox 1997:78).1In a sensitivity analy-
sis, log proportionperformed very similarly.
The dependentvariable s measuredannually,
one year after the independent variables.
Changing he lag to 2, 3, or 4 yearsdoes not alter
the findings. Our sample is designed to inves-
tigatethe effects of diversityprograms n work-
force composition in private sector
establishments large enough to file EEO-1
reports.We do not claim to describe he nation's
managerialworkforce.Nationally epresentative
samples such as the CurrentPopulationSurvey
include the public and nonprofit sectors, in
which the gains of women and minoritieshave
1Because og-odds logit) s undefined t values
of 0 and 1, we substituted with 1/2Nj,and 1 with
1-1/2Nj,whereNj is the number f managersn
establishment (Hanushek ndJackson1977;Reskin
andMcBrier 000).Theresultswererobust o dif-
ferent ubstitutionsor0. Wechose heone hatkept
the distributionnimodal ndclosest o normal.To
ensure hat he substitutionoesnot drive he find-
ings,we include binary ariableorno groupmem-
bers n management.
been larger.Furthermore,ational iguresreflect
the change in women's representation n man-
agementassociatedwith service sector growth
(e.g. Jacobs 1992), whereas our data track a
relatively stable set of firms.
AFFIRMATIVECTION LANSANDDIVERSITY
PRACTICES
Figure2 showsthe prevalence f all sevendiver-
sity programsamong the 708 employers ana-
lyzed later. By 2002, affirmative action plans
were used in 63 percent of the workplaceswe
study, ollowedby training n 39 percent,diver-
sity committees n 19 percent,networkingpro-
grams forwomenandminorities) n 19 percent,
diversity valuations or managers n 19 percent,
diversity taff n 11 percent,andmentoringpro-
grams forwomenandminorities) n 11 percent.
The bivariate correlations and joint frequen-
cies of the seven programsare not shown here
(see Online Supplement,ASR Web site: http://
www2.asanet.org/joumals/asr/2006/toc052.html).
In the analysesreported n the following dis-
cussion,we use binaryvariables o represent he
presenceof the sevendiversity rograms. orsix
programs,we asked whether the organization
had ever had the program, when it was first
adopted, ndwhen (if ever) t was discontinued.
For the seventhpractice,diversitytraining,we
asked when it was first and last offered. If an
employerhad gone for 3 yearswithout raining,
we treated he programas defunct.We collect-
ed additional nformation boutdiversity rain-
ing because our n-person nterviewssuggested
that it varied across organizationsmore than
the other programs,but we found significant
similarities n trainingprograms. n 70 percent
of the establishmentswith training for man-
agers, trainingwas mandatory. ncluded n 80
percent of the trainingprogramswas a discus-
sion on the legal aspectsof diversity, nd98 per-
cent were conducted with live facilitators, as
opposed to being offered exclusively via the
Web or video. Although some organizations
offered trainingnot only to managers,but also
to all employees, we reporteffects of training
for managersbecause managersmade promo-
tion decisions. Training or all employees had
nearly identical effects in the models.
Because the measures are binary, coded 1
for all the years he program s in place,program
effects are estimated for the entire period of
8/17/2019 Diversity Policies 2006
12/30
CORPORATEAFFIRMATIVEACTION AND DIVERSITYPOLICIES 599
percent
of
managers
90%
80%
70%
60%
50%
40%
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002
Year
S---- White Men -0--- White Women -4- Black Men - - - - Black Women
Figure 1. Percent of Managers: White Men and Women and Black Men and Women, 1971-2002
Note: Based on EEO-1 reports 1971-2002 sampled for Princeton University Human Resources Survey 2002.
Varying N. Maximum N = 708; EEO = equal employment opportunity.
percent
of
Workplace
40%
30%
20%
1970 1972 1974 1976 1978 1980 1982
--- AA Plan
--A-- Diversity Committee
-0-- Diversity Training
- - -Mentoring for Women/Minorities
1984 1986 1988 1990 1992 1994 1996 1998 2000 2002
Year
-- FullTimeEO/AAStaff
-4- DiversityEfforts n Mgrs'Evaluations
-)-- Networking for Women/Minorities
Figure 2. Percent of Private-SectorWorkplaceswith Affirmative Action Plans and Diversity Programs, 1971-2002
Note: Based on Princeton University Human Resources Survey, 2002. Varying N. Maximum N = 708.
8/17/2019 Diversity Policies 2006
13/30
600 AMERICANSOCIOLOGICALREVIEW
the program's xistence(not merelyfor the year
after initiation).
For six of the programs, between 2 and 4
percent of the respondents who reported the
program'sadoption could not tell us the exact
year.For he seventhpractice,affirmative ction
plan, the figure was 8 percent.We eliminated
cases with missing data on any of these vari-
ables. The resultswere virtually denticalwhen
we imputedmissing datafor variablesof inter-
est and retained these cases in the analysis.
Missing adoption dates for control variables
were imputed using ordinary least squares
(OLS) regression, with industry,age of estab-
lishment, and type of establishmentas covari-
ates. Omittingcases with imputeddata did not
substantiallyalter the findings.
CONTROL ARIABLES
All measures included in the analyses vary
annually.Table 1 presentsdefinitions and data
sources for key variablesas well as means and
standarddeviations (based on all organization-
al spells). Descriptive tatistics or the entire ist
of control variables are not shown here (see
Online Supplement,ASR Web site). Because
the fixed-effects method estimates variation
within the organization, t captures hangeover
time. For example, in the models, the variable
organizational size captures the effect of a
change in size on change in managerialdiver-
sity. These models effectively ignore measures
that do not change, such as industry,but cross-
case variation n those measures s capturedby
the fixed effects.
LEGALENVIRONMENT.e include a binary
variablebased on the EEO-1 reports ndicating
whether he establishment s a federalcontrac-
tor subject to affirmative action regulation.
Legal enforcement s measuredusing threesur-
vey variables that capturethe establishment's
experience with Title VII lawsuits, EEOC
charges, and affirmative action compliance
reviews. Each is coded 1 from the year of the
firm's first enforcementexperience.More than
one thirdof establishment-year pells had pre-
viously faced a lawsuit; more than one third
had faced an EEOC charge;and nearly 15 per-
cent had faced a compliancereview (only con-
tractorsare subject to compliancereviews).
ORGANIZATIONAL STRUCTURES. Organi-
zationalsize andavailability f managerialobs
are measured using EEO-1 data on the total
numberof employees in the establishmentand
the number of managerial employees.
Unionization is coded 1 when the establish-
ment has at least one contract. Substituting
with a measure of core job unionization does
not alter he results.FormalHR policies involve
a count of hiring, promotion, and discharge
guidelines; job descriptions; promotion lad-
ders; performanceevaluations;pay grade sys-
tem: and internal ob posting. Legal counsel is
measuredwith a binary variable for the pres-
ence of an in-house attorney.Targeted ecruit-
ment policy is a binary measure of special
diversity ecruitment fforts.Work-family up-
port counts paid maternity eave, paid paterni-
ty leave, flextimepolicies, andtop management
support or work-family programsas assessed
by our respondents.
TOP MANAGEMENTOMPOSITION.op man-
agement team diversity is measured with the
percentage of the top 10 positions held by
women and/or African Americans, based on
survey data.We asked about the percentageat
10-year intervals and interpolatedvalues for
the interveningyears.
LABORMARKET ND ECONOMIC NVIRONMENT.
The diversity of the establishment's internal
laborpool is measuredwith two variablesbased
on the EEO-1 reports: he percent of the focal
group n nonmanagerialobs and the percent n
the corejob. To determine he EEO-1 category
that held the core job, we asked respondents
about the single biggest job in the organiza-
tion. We include a variablecoded 1 when there
are no membersof the focal group in manage-
ment. Diversity of the establishment's xternal
labor pool is capturedby two sets of variables
on industry and state labor forces from the
CurrentPopulation Survey. Industryemploy-
mentvariablesare ogged.We use the industry's
percent of governmentcontractors based on
EEO-1 data)to measuredemand or underrep-
resentedworkers n affirmativeaction sectors.
Economic conditions are measured with the
yearly state unemploymentrate, and industry
size is measured as total annual industry
8/17/2019 Diversity Policies 2006
14/30
CORPORATEAFFIRMATIVEACTION AND
DIVERSITYPOLICIES 6ol
Table
1.
Selected
Variables
Used
in
Analysis
of
Managerial
Workforce
Composition
Mean
Standard
Deviation
Minimum
Maximum
Type
Data
Outcome
Variables
(percent)
Managers
who
are
white
men
70.0
23.6
0
100
Continuous
EEO-1
Managers
who
are
white
women
22.2
21.2
0
100
Continuous
EEO-1
Managers
who
are
black
women
1.4
4.2
0
66.7
Continuous
EEO-1
Managers
who
are
black
men
2.4
5.9
0
100
Continuous
EEO-1
Affirmative
Action
and
Diversity
Measures
Affirmative
action
plan
.422
.494
0
1
Binary
Survey
Full
time
EEO/diversity
staff
.045
.206
0
1
Binary
Survey
Diversity
committee
.052
.222
0
1
Binary
Survey
Diversity
training
.064
.244
0
1
Binary
Survey
Diversity
evaluations
of
managers
.102
.303
0
1
Binary
Survey
Networking
programs
.064
.244
0
1
Binary
Survey
Mentoring
programs
.033
.179
0
1
Binary
Survey
Legal
Environment
Affirmative
action
status
(government
contract)
.455
.498
0
1
Binary
EEO-1
Compliance
review
.149
.356
0
1
Binary
Survey
Discrimination
lawsuits
.341
.474
0
1
Binary
Survey
EEOC
charges
.314
.464
0
1
Binary
Survey
Organizational
Structures
Percent
managers
in
establishment
.124
.090
.002
.789
Continuous
EEO-1
Establishment
size
702
827
10
12,866
Continuous
EEO-1
Union
agreement
.254
.436
0
1
Binary
Survey
Formal
HR
policies
4.917
2.516
0
9
Count
Survey
In-house
attorney
.277
.448
0
1
Count
Survey
Special
recruitment
for
women
and
minorities
.156
.363
0
1
Binary
Survey
Work-family
accommodations
.912
.978
0
4
Count
Survey
Top
Management
Composition
(percent)
Top
managers
who
are
minorities
3.471
10.239
0
100
Continuous
Survey
Top
managers
who
are
women
16.445
23.575
0
100
Continuous
Survey
Note:
N
=
16,265.
Labor
market
and
economic
environment
variables
are
included
in
the
analyses
but
not
shown
here.
See
note
to
Table
2
or
a
etailed
list
of
variables
not
shown
here
(see
entire
list
of
control
variables
on
Online
Supplement,
ASR
Web
site:
http://www2.asan
EEO
=
equal
employment
opportunity;
HR
=
human
resources.
8/17/2019 Diversity Policies 2006
15/30
602 AMERICANSOCIOLOGICALREVIEW
employment,both fromthe CurrentPopulation
Survey.
METHODS
We use pooled cross-sectional ime-seriesmod-
els, with fixed effects for bothestablishment nd
year (Hicks 1994; Hsiao 1986). We use fixed
effects for establishmentso account or unmea-
sured, ime-invariant haracteristics hat might
affect outcome variables (for recent empirical
examples of these methods appliedto individ-
uals, see Budig and England 2001; Western
2002). This specification, achieved by sub-
tracting the values of each observation from
the establishment mean (Hsiao 1986:31),
strengthens our causal inferences about the
effects of affirmativeaction plans and diversi-
ty practices by ruling out the possibility that
organizations hat adoptedthose practiceshad
stableunobservedpreferences or diversity.To
capture environmentalchanges, such as legal
and cultural hifts, we use a binaryvariable or
each year, omitting 1971. The large numberof
parametersnvolved n estimating ixed-effects
models rendersthem less efficient than other
estimators.However, we prefer these to alter-
native models because they provide the most
stringent tests of our hypotheses. The estab-
lishment and year fixed effects also offer an
efficient means of dealing with nonconstant
varianceof the errors heteroskedasticity)tem-
ming from the cross-sectional and temporal
aspects of the pooled data.
Because our dependentvariablesare meas-
ured as parts of the same whole (the whole
being management obs), we expect their error
terms to be correlated.Ordinary east squares
would husproduceunbiasedandconsistent,but
inefficient, estimators.We use seemingly unre-
lated regression, which takes into account
covariance between the errors and produces
unbiased, efficient estimators (Felmlee and
Hargens 1988; Greene 1997; Zellner 1962).
Simultaneous stimationalso allows us to com-
parethe effect of each diversitypracticeacross
groupswith formalchi-square ests (Kalleberg
and Mastekaasa2001; Zellner 1962).
FINDINGS
The analysis shows substantial ariation n the
effectiveness of diversity programs. Some
increasemanagerialdiversityacross the board,
whereas others have meager effects, or posi-
tive effects for some groupsandnegativeeffects
for others.The most effectivepracticesare hose
thatestablish rganizationalesponsibility: ffir-
mative action plans, diversity staff, and diver-
sity task forces. Attempts to reduce social
isolationamongwomen andAfricanAmericans
through networking and mentoring programs
are ess promising.Leasteffective areprograms
for taming managerialbias througheducation
and feedback.
DIVERSITYROGRAMST WORK
In Table 2, we report models of managerial
diversity. (Selected control variables are pre-
sented;the remainingcoefficients can be seen
on the OnlineSupplement,ASRWebsite). Each
dependentvariable s the (natural) og odds of
managersbeing from a certaingroup.To trans-
form he coefficient[p romrepresenting hange
in log odds to representingpercentagechange
in odds, it should be exponentiated: exp([3)-
1]*100. Once exponentiated in this way the
coefficient representsthe average percentage
change n the odds thatmanagers refroma cer-
taingroup,associatedwith a change n the inde-
pendentvariable. n the discussionbelow we use
'odds for [group]'as a shorthand.We also pro-
vide an illustrativesummary of the results in
proportion erms.
The R2 figures for these fixed-effects mod-
els represent the percentage of the variance
explained by the predictors when the unique
effectsof each establishment reexcluded.A log
likelihood ratio test shows that the variables
reported n Table 2 significantly improve the
model fit (chi(28) = 405.66; p < .001), as com-
pared with the baseline models that have no
variables epresenting iversityprograms avail-
able on request).
ORGANIZATIONALRESPONSIBILITY.Coeffi-
cients for the diversityprogramsrepresent he
change in the log odds that managersare from
a certain group that is attributable o the pres-
ence of a practice,averagedacross all years of
the program's xistence.After employers et up
affirmative ctionplans,the odds for white men
in managementdecline by 8 percent; he odds
for white womenrise by 9 percent;andthe odds
for blackmen rise by 4 percent.These numbers
represent the estimated average difference
8/17/2019 Diversity Policies 2006
16/30
CORPORATEFFIRMATIVECTIONAND
DIVERSITYOLICIES
603
Table2. FixedEffectsEstimates f
the
Log
Oddsof
White
Men andWomenandBlack
Women ndMen
in
Management,
971-2002
WhiteMen WhiteWomen BlackWomen
Black Men
Organizational esponsibility
Affirmative ctionplan -.078** .086** .005 .039*
(.017) (.017) (.014) (.015)
Diversity
ommittee
-.081** .175** .242** .114**
(.028)
(.029) (.024)
(.026)
Diversity
taff -.055
.104**
.123**
.128**
(.033) (.034) (.028) (.030)
Managerial
ias
Diversity raining
-.038
-.001 -.066** .031
(.021) (.022) (.018) (.019)
Diversity
valuations .028 .061*
-.027 -.081**
(.027)
(.028) (.023) (.025)
Social Isolation
Networking rograms -.083** .080** .012 -.096**
(.027) (.028) (.023) (.024)
Mentoring rograms
-.011 -.004 .213**
.037
(.033) (.035)
(.029) (.031)
Legal
Environment
Governmentontract .032
.006 -.039*
-.027
(.019)
(.019) (.016)
(.017)
Compliance
eview
-.083** .077** .020 .081**
(.020) (.020)
(.017) (.018)
TitleVII lawsuit
-.107**
.141** .044**
.029*
(.015)
(.016) (.013)
(.014)
EEOC
charge
-.007 .014
.019 .034*
(.016) (.017) (.014) (.015)
Organizational
tructures
Proportionmanagers
n
establishment
-.896**
.309**
-4.499**
-3.989**
(.108) (.112)
(.092)
(.099)
Establishment
ize
(log)
-.021
-.023*
-.661**
-.515**
(.012)
(.012)
(.010) (.011)
Union
agreement
-.053 -.068*
-.007
-.029
(.033)
(.034) (.028)
(.030)
Formal
ersonnel olicies
-.002
-.003 -.016**
-.015**
(.004) (.004)
(.003)
(.003)
In-house
ttorney
-.100**
.126** -.040*
.021
(.023)
(.024) (.020)
(.021)
Targetedecruitmentolicy -.071** .108** .131** .099**
(.021)
(.021)
(.018)
(.019)
Work-family
ccommodations
-.078**
.065**
.026**
.004
(.008)
(.009)
(.007)
(.008)
TopManagement
omposition
Proportion
minorities
n
top management
-.002
-.002
.007**
.012**
(.001)
(.001)
(.001)
(.001)
Proportion
omen
n
top management
-.002**
.004**
.002**
-.002*
(.001)
(.001)
(.001)
(.001)
R2
64 parameters)
.3335 .3146
.3636
.2799
Note:
Log
likelihood atio
est;
X2
28)
=
405.66;
p
<
.001. Datashownarecoefficients rom
seemingly
unrelated
regression
with standard rrors n
parentheses.
ariablesncluded
n the
analyses
butnot shownhereare 8
vari-
ablesfor
proportion
f each
group
n
non-managerial
obs
and
n
core
ob
in each
establishment;
binary
vari-
ables
for no workers roma
group
n
management;
variables
or
proportion
f each
group
n stateand
ndustry
labor
orces;
proportion
f contractorirms n
industry;
ndustry mployment;
ndstate
unemployment
ate
full
resultson Online
Supplement,
SRWebsite:
http://www2.asanet.org/journals/asr/2006/toc052.html).
nalyses
also includeestablishment
nd
year
fixed
effects.
All
independent
ariables re
aggedby
1
year,excluding
proportion
f
managerial
obs.
N
(organization-year)
16,265;
N
(organizations)
708.
EEOC
=
Equal
Employment Opportunity
Commission.
*
p
<
.05;
**
p
<
.01
(two
tailed
test).
8/17/2019 Diversity Policies 2006
17/30
604 AMERICANSOCIOLOGICALREVIEW
between having a plan and the counterfactual
condition of not having a plan for the entire
period of the plan'sexistence.These resultsare
consistent with Leonard's 1990) finding that
affirmativeactionplan goals areeffective.Note
that the coefficient for black women is not sig-
nificant here. When we introduced industry
interactions,we discovered hat n manufactur-
ing (computers, electronics, transportation),
affirmative action plans had negative effects
on black women, whereas in service (retail,
insurance, usinessservices),affirmative ction
planshadpositiveeffects (resultsavailableupon
request).Creating diversity ommittee ncreas-
es the odds for white women, across the period
of the committee's existence, by 19 percent.
The odds for black women rise 27 percent,and
the odds for black men rise 12 percent.
Employerswho appoint ull-timediversity taff
also see significant increases in the odds for
white women (11 percent), black women (13
percent), and black men (14 percent) in man-
agement.
As noted, he coefficients n Table2 represent
the averagechanges in log odds that managers
are from a certain group. The effect of each
programon the percent of women and minori-
ties in management will vary depending on
where organizationsbegin (Fox 1997:78). For
example, an 8 percent decrease in the odds of
managersbeing white men resulting romadop-
tion of affirmativeaction plan would translate
to a decline of 2.6 percent in the percent of
white men in managementf they constituted 0
percent before adoption, but it would mean a
larger decline of 4.3 percent if they made up
only 50 percent at the baseline (Petersen
1985:311).
PROGRAMSFOR REDUCINGMANAGERIAL IAS.
Programsdesigned to reduce managerialbias
through ducation diversity raining)and feed-
back (diversity evaluations)show one modest
positive effect and two negative effects across
the threedisadvantaged roups.Diversity rain-
ing is followedby a 7 percentdecline n the odds
for black women. Diversityevaluationsare fol-
lowed by a 6 percentrise in the odds for white
women, but an 8 percentdecline in the odds for
black men. These mixed effects are anticipated
in the literature. s noted, aboratory tudiesand
surveys often show adverse reactionsto train-
ing (Bendick et al. 1998; Nelson et al. 1996).
Moreover, critics argue that trainers define
diversitybroadly o include groupsnot covered
by federal civil rights law (parents,smokers),
and therebydrawattentionaway from protect-
ed groups (Edelman, Fuller, and Mara-Drita
2001; Kochanet al. 2003; Konrad ndLinnehan
1995).
PROGRAMSFOR REDUCING OCIAL SOLATION.
Networkingandmentoringprograms,designed
to countersocial isolation, show modest effects
on managerial iversity.Networking s followed
by a rise in the odds for white women and a
decline in the odds for white men and black
men. The negative coefficient for black men is
anticipated y qualitative esearch Carter 003;
Friedman nd Craig2004) showing thatwhites
can develop negative attitudes owardAfrican-
American organizing. In contrast, mentoring
programsshow a strong positive effect on the
odds for black women. These findings suggest
that having personal guidance and support at
workcan facilitatecareerdevelopment Castilla
2005) for black women, whereasnetworking s
more effective for white women.
GENDER AND RACIAL PATTERNS.Overall, it
appears that diversity programs do most for
white women and more for black women than
for black men. Black men gain significantly
less from affirmative action than do white
women (chi-sq(1) = 4.15, p < .05), and signif-
icantly less from diversity committees than do
blackwomen (chi-sq(l) = 22.47, p.< .01). Three
programs show negative effects on African
Americans,whereasno programshows a neg-
ative effect on white women. We hesitate to
overinterprethis pattern,but note thatthere is
something of a trade-offamong groups.
Table3 evaluates he magnitude f the effects
of programson the proportion f each group n
managementbased on the coefficients in Table
2. Proportionn year of adoption s the mean
proportion of each group in management,
amongadopters,n theiractualyearsof program
adoption i.e.,just before reatment). Estimated
proportionwith practice shows the predicted
mean proportionafter the practice is in place.
Thus, for example, the proportion of white
women among managers n the averageestab-
lishment adopting an affirmative action pro-
gram was 0.132, and the net effect of the
8/17/2019 Diversity Policies 2006
18/30
CORPORATE
FFIRMATIVECTION
NDDIVERSITY
OLICIES
605
Table3.
Estimated
Average
Differences
n
Managerial omposition
Due
to
Adoption
of
Affirmative
Action
and
Diversity
Practices
WhiteMen
WhiteWomen
Black
Women
BlackMen
Affirmative
ActionPlan
Proportionnyearof adoption .783 .132 .017 .024
Estimated
roportion
ith
practice
.769
.142
.017
.025
Percent
ifferencedue to
adoption
-1.8%**
7.6%**
.0% 4.2%**
Diversity
Committee
Proportion
n
year
of
adoption
.630
.230 .014
.020
Estimated
roportion
ith
practice
.611
.262
.018 .022
Percent ifference
due
to
adoption
-3.0%**
13.9%**
29.8%** 10.0%**
Diversity
Staff
Proportion
n
year
of
adoption
.724
.157
.014
.021
Estimated
roportion
ith
practice
.713 .171
.016 .024
Percent
ifferencedue to
adoption
-1.5%
8.9%**
14.3%**
14.3%**
Diversity
Training
Proportion
n
year
of
adoption
.687 .194 .017 .022
Estimated
roportion
ith
practice
.679 .194
.016
.023
Percent
ifference ue to
adoption
-1.2% .0% -5.9%**
4.5%
Diversity
Evaluations
Proportion
n
year
of
adoption
.720
.160 .017
.024
Estimated
roportion
ith
practice
.726
.168 .017
.022
Percentdifference ue to
adoption
.8%
5.0%
.0%
-8.3%**
Networking rograms
Proportion
n
year
of
adoption
.702 .193 .014
.020
Estimated
roportion
ith
practice
.684 .206 .014
.018
Percent
difference ue
to
adoption
-2.6%** 6.7%**
.0%
-10.0%**
Mentoring rograms
Proportion
n
year
of
adoption
.690
.216 .017 .021
Estimated
roportion
ith
practice
.688
.215 .021 .022
Percent ifferencedue to
adoption
-.3%
-.5%
23.5%** 4.8%
Note:
Estimates asedon coefficients
presented
n
Table
2.
*
p
<
.05;
**
p
<
.01
(two
tailed
est).
program,
with control for
other
factors,
is
to
raise white women
proportion
to 0.142.
Similarly,
he
proportion
f blackwomen
among
managerswas 0.014 intheaverage irmadopt-
ing
a
diversity
committee,
and
adoptionbrings
black women to
0.018,
an increase of
almost
30%.
The
third
row,
based on the
first two
rows,
reports
he
percentage
hange
overthe baseline
resulting
from
program
adoption.
Tables
2
and 3
support
our
contention
that
programs stablishingorganizational
esponsi-
bility
are
more
broadly
effective than hose
that
address
managerial
bias
or
social isolation
among women and African Americans.
Organizations
hat
structure
esponsibility
see
consistent
positive
effects for
white
women,
black
women,
and black
men.
Coefficients
for
control variables
are con-
sistent
with
expectations,
with one
possible
exception.
The
negative
effect
of formal
per-
sonnel
policies
is not consistent with the idea
that
bureaucracy mpedes
cronyism
or
bias
in
promotion
decisions
(Reskin
and McBrier
2000), butis consistentwiththe argumenthat
formalization eads to
the
needless
inflationof
educational
prerequisites
Collins 1979),
and
with
findings
that
the
determinants f
promo-
tion
differ
systematically
or
whites andblacks
even
when formal
personnel
systems
exist
(Baldi
andMcBrier
1997).
Other oefficients
of
control variables show that
although growth
and
unionization
have
not
improveddiversity,
and
although egal
staffhad
only
limited
effects,
targeted recruitmentprograms,work/family
accommodations,
and
top
management
team
diversity
show
positive
effects
on
managerial
diversity.
Coefficients
for the
labor
marketand
economic
environment
measures,
not
shown
here,
are
in
the
expected
directionas well
(see
Online
Supplement,
ASR
Web
site).
8/17/2019 Diversity Policies 2006
19/30
606
AMERICANOCIOLOGICAL
EVIEW
DOEs ORGANIZATIONAL ESPONSIBILITY
IMPROVEPROGRAMEFFECTIVENESS?
It
is
possible
that
some
programs
work best in
combination with others
(MacDuffie
1995;
Perry-SmithandBlum2000). Ourfindingthat
organizational responsibility
structures
have
broadereffects
than
other
programs
suggests
that
perhaps
training,
evaluation,
mentoring,
and
networking
would be more successful in
combination
with
responsibility
tructures.
We
undertake everal
analyses
of
program
combi-
nations.
First,
we
explore
the
possibility
that
he
sim-
ple
number
of
programs
matters.
Perhaps
our
measures
apture
not the effects of discrete
pro-
grams
so
much as
an
orientation oward
hang-
ing workplacedemography.We introducehree
binary
variables
epresenting
he
presence
of
any
one,
two,
and three or more
programs.
Across
the
16,265
organization-year pells
of
data,
49
percent
had no
programs,
34
percent
had
one
program,
10
percent
had two
programs,
and 7
percent
had three or more
programs.
n the
top
panel
of Table
4,
we
report
the
effects of the
Table4. Fixed-EffectsEstimates f theLogOddsof WhiteMenandWomen ndBlackWomen nd Men n
Management
ith
Bundles
of
Programs,
971-2002
White White Black Black
Men
Women Women Men
Adoption
f
One or More
AA
Plans&
Diversity
Programs
Only
one
program
-.043** .056** -.009
.026
(.016) (.016) (.013) (.014)
Two
programs
-.091**
.121**
.020 .024
(.023) (.023) (.019) (.021)
Three
or
more
programs
-.158**
.232** .127**
.046
(.029) (.030) (.025) (.027)
R2
(60
parameters)
.3323 .3124 .3569
.2767
Interaction
ith
Responsibility
tructures
Responsibility
tructures
-.063** .081**
.007 .042**
(.017)
(.017) (.014)
(.015)
Diversity
raining
-.026 -.064
-.046 .026
(.036)
(.038) (.031) (.033)
x
Responsibility
tructure
-.026 .132** .044
.040
(.042)
(.043)
(.036)
(.038)
Diversity
valuations
.294**
-.042
-.065
-.077
(.057) (.059) (.049) (.052)
x
Responsibility
tructure
-.326** .136*
.057
.009
(.061) (.063) (.053)
(.057)
Networking
rograms
-.090
.163** -.026
-.172*
(.050)
(.052)
(.043) (.046)
x
Responsibility
tructure
-.003
-.088
.073
.118*
(.056)
(.058)
(.048)
(.051)
Mentoring
rograms
.140** -.101
-.042
.127*
(.066) (.068)
(.057)
(.061)
x Responsibility
tructure
-.183*
.133
.344**
-.108
(.074)
(.076)
(.063)
(.068)
R2
(66
parameters)
.3347
.3136 .3602
.2785
Note.:
Data
shownare
coefficients rom2
seemingly
unrelated
egression nalyses
with standard
rrors
n
paren-
theses.
Responsibility
tructuresncludeaffirmative
ction
plans,diversity
ommittees nd
diversity
taff.
The
analyses
ncludeestablishment
nd
year
ixed effects andall the
controlvariables
ncluded
n
the models
present-
ed
in
Table
2
(for
coefficientsof
control
variables,
ee
Online
Supplement,
SR
Web
site:
http://www2.
asanet.org/journals/asr/2006/toc052.html).
(organization-year)
16,265;
N
(organizations)=
08.
*
p
<
.05;
**
p
<
.01
(two
tailed
test).
8/17/2019 Diversity Policies 2006
20/30
CORPORATEAFFIRMATIVEACTION AND DIVERSITYPOLICIES 607
numberof programs n models parallel o those
presented n Table2 (results or the controlvari-
ables are availableon the Online Supplement,
ASRWebsite).We compared oefficients orthe
binary count variablesusing t tests. For white
women, the sheernumberof programsmatters;
one is better hanzero, two better hanone, and
three or more are better than two. For white
men, we find the opposite pattern,suggesting
that each additionalprogramreduces the odds
for white men. Forblackwomen, havingone or
two programs s not significantlydifferent rom
having none. Having three is significantly dif-
ferent. For black men, none of the count vari-
ables show an effect significantlydifferent rom
having no programs.Hence, for white women,
the more programs he better. For blacks, the
numberof programsmatters ess thanthe con-
tentof the programs. his is not surprising iven
that some practices n Table2 show no effects,
or even negative effects, on blacks.
Althougheach additional rogram, egardless
of content,does not always ranslatento greater
diversity,particular undlesof programsmight
operatewell together.To test this idea, we ran
(in models otherwise dentical o those in Table
2) all two-way interactionsbetween affirma-
tive action plan, diversitycommittee, diversity
staff, raining, valuation, etworking, ndmen-
toring.(The bivariate orrelations nd oint fre-
quenciesof the sevenprograms represented n
the Online Supplement, ASR Web site.) The
two-way interactionsamong training, evalua-
tion, networking, ndmentoringdid not indicate
that any pairs operatedbetter than individual
programs. But two-way interactions with
responsibility structures did render training,
evaluation, networking, and mentoring more
effective. For ease of presentation,we collapse
the three responsibilitystructures nto a single
variable, nteracting t with the four other pro-
gram variables. The second panel in Table 4
includesestimates rommodelswith these inter-
actions(results or the controlvariablesarepre-
sented on the Online Supplement, ASR Web
site).
Diversity training, evaluation, networking,
and mentoringprogramsare more effective in
firms with responsibility tructures.Withdiver-
sity trainingand evaluations, he responsibility
structure interaction positively affects white
women. With networking, the responsibility
structure interaction positively affects black
men, and with mentoring, t positively affects
black women. Note that he noninteracted ari-
able,responsibility tructure, ontinues o show
the expected effects for white men, white
women, and black men. The overall pattern s
strikingand suggests that hese authority truc-
turesrender he otherprogramsmore effective.
Yeteven with responsibility tructuresn place,
none of these programsshow the sort of con-
sistent pattern across outcomes that we find
for, say, diversitycommittee.
Do AFFIRMATIVECTIONORDERSMEDIATE
PROGRAMFFICACY?
In Table 2, we also examine whetheraffirma-
tive action enforcement shows direct effects.
Employers who sign a government contract,
and hereby ecome subject o affirmative ction
regulation,do not see increases in managerial
diversityas a directresult.When we interacted
contractortatuswith the period1971-1980, the
results did not supportearly researchers' ind-
ings
Recommended