Orchestrating Impartiality - The Impact of Blind Auditions on Female Musicians

Orchestrating Impartiality: The Impact of "Blind" Auditions
on Female Musicians
By CLAUDIA GOLDIN AND CECILIA ROUSE*
A change in the audition procedures of symphony orchestras—adoption of "blind"
auditions with a "screen" to conceal the candidate's identity from the jury—
provides a test for sex-hiased hiring. Using data from actual auditions, in an
individual fixed-effects framework, we find that the screen increases the probability
a woman will be advanced and hired. Although some of our estimates have large
standard errors and there is one persistent effect in the opposite direction, the
weight of the evidence suggests that the blind audition procedure fostered impartiality
in hiring and increased the proportion women in symphony orchestras.
(JEL n'
Sex-biased hiring has been alleged for many
occupations but is extremely difficult to prove.
The empirical literature on discrimination, de-
* Goldin; Depanment of Economics, Harvard University,
Cambridge, MA 02183; Rouse: Woodrow Wilson School,
Princeton University, Princeton, NJ 08544. Rouse acknowledges
The National Academy of Education, the NAE Spencer
Postdoctoral Fellowship Program, and the Mellon Eoundation
for financial support. We are indebted to the staff members of
the orchestras thai gave us access to their audition records and
who provided other assistance, and to the musicians who
responded to our questionnaire. We are particularly grateful to
Joanne Berry, Brigit Carr, Ruth DeSamo. Stefanie Dyson, Josh
Feidman, Barbara Haws, Oren Howard, Cindy Hubbard, Carol
Jacobs, Lynn Larsen, Bennett McClellan, Stephen Molina, Bill
Moyer, Jeffrey Neville, Stephen Novak, Deborah Oberschalp,
Stacey Pelinka, Carl Schiebler, Alison Scott-Williams, Robert
Sirineck, Harold Steiman, and Brenda Nelson Strauss. We also
thank Gretchen Jackson of the University of Michigan School
of Music. Rashid Alvi. Brigit Chen. Eric Hilfers, Serena Mayeri,
LaShawn Richburg, Melissa Schettini, Thomas Tucker,
Linda Tuch, and Lavelle (Yvette) Winfield served as our
extremely able research assistants. David Howel! of the Princeton
University Department of East Asian Studies and Jin Heum
Park kindly helped to determine the gender of Japanese and
Korean names. We thank them all. We are grateful to our
colleagues David Card, Anne Case, David Cutler, Angus
Deaton, Hank Fart>er, Lany Kalz, Alan Krueger, David Lee,
and Aaron Yelowitz for helpful conversations, and to seminar
participants al the School of Industrial and Labor Relations at
Cornell University, University of Illinois at Champaign-
Urbana, Princeton University, Utiiversity of Toronto, Harvard
University, and Vanderbilt University. We also thank two
anonymous referees for comments that have made this a better
paper. Any remaining errors are ours. Unfortunately the data
used in this article are conlidential and may not be made
available to other researchers.
riving from the seminal contributions of Gary
Becker (1971) and Kenneth Arrow (1973), has
focused mainly on disparities in earnings between
groups (e.g., males and females), given
differences in observable productivity-altering
characteristics. With the exception of various
audit studies (e.g., Genevieve Kenney and
Douglas A. Wissoker, 1994; David Neumark et
al., 1996) and others, few researchers have been
able to address directly the issue of bias in
hiring practices.' A change in the way symphony
orchestras recruit musicians provides an
unusual way to test for sex-biased hiring.
Until recently, the great symphony orchestras
in the United States consisted of tnembers
who were largely handpicked by the music
director. Although virtually all had auditioned
for the position, most of the contenders would
have been the (male) students of a select
' An extensive literature exists on occupational segregation
by sex and the possible reasons for the large differences
in occupations between men and women today and in the
past. The debate is ongoing. On the one hand are those who
believe that discrimination, either individual or societal in
nature, is the driving force, and on the other hand are those
who claim the evidence shows women and men sort among
occupations on the basis of different tastes for work characteristics.
In the former category see Paula England (1982)
and England et al. (1988); in the latter group see Solomon
W. Polachek (1979) and Randall K. Filer (1989). It should
be noted that many other studies (e.g., Ian Ayres and Joel
Waldfogel, 1994) have also attempted to measure discrimination
in atypical ways.
715
716 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
group of teachers. In an attempt to overcome
this seeming bias in the hiring of tnusicians,
most major U.S. orchestras changed their audition
policies in the 197O's and l980's makitig
them more open and routinized. Openings
became widely advertised in the union papers,
and many positions attracted more than 100
applicants where fewer than 20 would have
been considered before. Audition committees
were restructured to consist of members of
the orchestra, not just the conductor and section
principal. The audition procedure became
democratized at a time when many other institutions
in America did as well.
But democratization did not guarantee impartiality,
because favorites could still be identified
by sight and through resumes. Another set of
procedures was adopted to ensure, or at least give
the impression of, impartiality. These procedures
involve hiding the identity of the player from the
jury. Although they take several forms, we use the
terms "blind" and "screen" to describe the group."
The question we pose is whether the hiring process
became more impartial through the use of
bhnd auditions, Because we are able to identify
sex, but no other chiiracteri sties for a large sample,
we focus on the impact of the screen on the
employment of women."^
Screens were not adopted by all orchestras at
once. Among the major orchestras, one still
does not have any blind round to their audition
procedure (Cleveland) and one adopted the
screen in 1952 for the preliminary round (Boston
Symphony Orchestra), decades before the
others. Most other orchestras shifted to blind
preliminaries from the early 197O's to the late
198O's. The variation in screen adoption at various
rounds in the audition process allows us to
assess its use as a treatment.''
The change in audition procedures with
the adoption of the screen allows us to test
whether bias exists in its absence. In both our
' For an article about the blind audition process see The
Economht 0996).
' The screen may also have opened opprortunities for
individuals from less-welNknown orchestras, those trained
outside mainstream institutions, and those from minority
groups,
"" The blind audition procedures bear some resemblance
to "double-blind" refereeing in academic journals. See Rebecca
Blank (1991) for an assessment of the treatment effect
of such refereeing in the American Economic Review.
Study and studies using audits, the issue is
whether sex (or race or ethnicity), apart from
objective criteria (e.g., the sound of a musical
perfonnance, the content of a resume), is considered
in the hiring process. Why sex might
make a difference is another matter.
Our data come from two sources; rosters and
audition records. Rosters are simply lists of
orchestra personnel, together with instrument
and position (e.g., principal), found in orchestra
programs. The audition records are the actual
accounts of the hiring process kept by the personnel
manager of the orchestra. Both are described
in more detail below.
The audition records we have collected form an
uncommon data set. Our sample includes who
was advanced and hired from an initial group of
contestants and also what happened to approximately
two-thirds of the individuals in our data set
who competed in other auditions in the sample.
There are, to be certain, various data sets containing
information on applicant pools and hiring
practices (see, e.g., Harry Holzer and David Neumark,
1996). But our data set is unique because it
has the complete applicant pool for each of the
auditions and links individuals across auditions.
Most important for our study is that audition procedures
differed across orchestras in known ways
and that the majority of the orchestras in our
sample changed audition procedure during the period
of study."''
We find, using our audition sample in an individual
fixed-effects framework, that the screen
increases the probability a woman will be advanced
out of a preliminary round when there is
no semifinal round. The screen also greatly enhances
the likelihood a female contestant will be
the winner in a final round. Using both the roster
and auditions samples, and rea.sonable assumptions,
the switch to blind auditions can explain
about one-third of the increase in the proportion
female among new hires (whereas another onethird
is the result of the increased pool of female
candidates). Estimates based on the roster sample
indicate that blind auditions may accotint for 25
percent of the increase in the percentage of orchestra
musicians who are female.
^ This statement is true for the roster sample. There are
only a few orchestras that changed audition procedures
during the years of our audition data.
VOL 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIAUTY 717
I. Sex Composition of Orchestras
Symphony orchestras consist of about 100
musicians and, although the number has varied
between 90 to 105, it is rarely lower or
higher. The positions, moreover, are nearly
identical between orchestras and over time.
As opposed to firms, symphony orchestras do
not vary much in size and have virtually identical
numbers and types of jobs. Thus we can
easily look at the proportion women in an
orchestra without being concerned about
changes in the composition of occupations
and the number of workers. An increase in the
number of women from, say, 1 to 10, cannot
arise because the number of harpists
(a female-dominated instrument), has greatly
expanded. It must be because the proportion
female within many groups has increased.
Among the five highest-ranked orchestras
in the nation (known as the "Big Five")—the
Boston Symphony Orchestra (BSO), the Chicago
Symphony Orchestra, the Cleveland
Symphony Orchestra, the New York Philharmonic
(NYPhil), and the Philadelphia Orchestra—
none contained more than 12
percent women until about 1980.^ As can be
seen in Figure lA, each of the five lines
(giving the proportion female) greatly increases
after some point. For the NYPhil, the
line steeply ascends in the early 1970"s. For
the BSO, the turning point appears to be a bit
earlier. The percentage female in the NYPhil
is currently 35 percent, the highest among all
11 orchestras in our sample after being the
lowest (generally at zero) for decades. Thus
the increase of women in the nation's finest
orchestras has been extraordinary. The increase
is even more remarkable because, as
we discuss below, turnover in these orchestras
is exceedingly low. The proportion of new
players who were women must have been,
and indeed was, exceedingly high.
Similar trends can be discerned for four
other orchestras—the Los Angeles Symphony
Orchestra (LA), the San Francisco Philharmonic
(SF), the Detroit Symphony Orchestra,
and the Pittsburgh Symphony Orchestra
^The data referred to, and used in Figures I to 3, are
from orchestral rosters, described in more detail below.
(PSO)—given in Figure lB.^ The upward
trend in the proportion female is also obvious
in Figure IB, although initial levels are higher
than in Figure lA. There is somewhat more
choppiness to the graph, particularly during
the 194O's. Although we have tried to eliminate
all substitute, temporary, and guest musicians,
especially during World War II and
the Korean War, this was not always possible.
The only way to increase the proportion
women is to hire more female musicians and
turnover during most periods was low. The
number of new hires is graphed in Figure
2 for five orchestras. Because "new hires" is a
volatile construct, we use a centered five-year
moving average. In most years after the late
l950's, the top-ranked orchestras in the group
(Chicago and NYPhil) hired about four musicians
a year, whereas the other three hired
about six. Prior to 1960 the numbers are extremely
high for LA and the PSO, because, it
has been claimed, their music directors exercised
their power to terminate, at will, the
employment of musicians. Also of interest is
that the number of new hires trends down,
even excluding years prior to 1960. The important
points to take from Figure 2 are that
the number of new hires was small after 1960
and that it declined over time.
The proportion female among the new hires
must have been sizable to increase the proportion
female in the orchestras. Figure
3 shows the trend in the share of women
among new hires for four of the "Big Five"
(Figure 3A) and four other orchestras (Figure
3B).^ In both groups the female .share of new
hires rose over time, at a somewhat steeper
rate for the more prestigious orchestras. Since
the early l980's the share female among new
hires has been about 35 percent for the BSO
and Chicago, and about 50 percent for the
NYPhil, whereas before 1970 less than 10
percent of new hires were women.
Even though the fraction of new hires who
are female rises at somewhat different times
' Our roster sample also includes the Metropolitan Opera
Orchestra and the St. Louis Symphony.
^ A centered five-year moving average is also used for
this variable.
^ In virtually all cases the share of women among new
hires has decreased in the 1990" s.
718 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
A 0.4
BSO
Philadelphia
Chicago
NYPhil
Cleveland
1940 1950 1960 1970 1980 1990
FIGURE 1. PROPORTION FEMALE IN NINE ORCHESTRAS, 1940 TO 1990"S
A: THE "BtG FIVE"; B: FOUR OTHERS
Source: Roster sample. See text.
across the orchestras, there is a discernible
increase for the group as a whole in the late
197O's to early 198O's, a time when the labor
force participation of women increased generally
and when their participation in various
professions greatly expanded. The question,
therefore, is whether the screen mattered in a
direct manner or whether the increase was
the result of a host of other factors, including
the appearance of impartiality or an increased
pool of female contestants coming out of
music schools. Because the majority of new
hires are in their late twenties and early
thirties, the question is whether the most selective
music schools were producing considerably
more female students in the early
1970"s. We currently have information by
instrument for only the Juilliard School of
Music. With the exception of the brass section,
the data, given in Figure 4, do not reveal
VOL. 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIALITY 719
22
20
18
16
(A
% 14
12 0)
t 10
^ ^
NYPhil
—•— Chicago
LA
^ ^ SF
^ ^ PSO
1950 1960 1970 1980 1990
FIGURE 2. NUMBER OF NEW HIRES IN FIVF, ORCHESTRAS, 1950 TO 199O'S
Source: Roster sample. See text.
Notes: A five-year centered moving average is used. New hires are musicians who were not with ihe orchestra the previous
year, who remain for at least one additional year, and who were not substitute musicians in the current year.
any sharp breaks in the fraction of all graduates
who are female.'" Thus, it is not immediately
obvious that an expansion in the
supply of qualified female musicians explains
the marked increase in female symphony
orchestra members; it could, therefore, be because
of changes in the hiring procedures of
orchestras.
But why would changes in audition procedures
alter the sex mix of those hired? Many of the most
renowned conductors have, at one time or another,
asserted that female musicians are not the equal of
male musicians. Claims abound in the world of
music that "women have smaller techniques than
men," "are more temperamental and more likely
to demand special attention or treatment," and that
"the more women [in an orchestra], the poorer the
'"We also have data on the .sex composition of the
graduates of the University of Michigan School of Music
and Indiana University, but not by instrument. In the Michigan
data, both for those receiving the Bachelor of Music
(BM) degree and for those receiving the Masier of Music
(MM) degree, there is no change in ihe percentage female
from 1972 to 1996. The Indiana University data, for both
BM and MM degrees and excluding voice, piano, guitar,
and early instruments, show an increase in the fraction
female from 1975 to 1996. The ratio of females to males
was 0.9 in 1975 but 1.2 in 1996.
sound."'' Zubin Mehta, conductor of the Los Angeles
Symphony from 1964 to 1978 and of the
New York Philharmonic from 1978 to 1990, is
credited with saying, "I just don't think women
should be in an orchestra."'^ Many European orchestras
had, and some continue to have, stated
poUcies not to hire women.'^ The Vienna Philharmonic
has only recently admitted its first female
member (a harpist). Female musicians, it can be
convincingly argued, have historically faced considerable
discrimination.'* Thus a blind hiring
procedure, such as a screen that conceals the identity
of the musician auditioning, could eliminate
"Seltzer (1989). p. 215.
'^ Seltzer (1989), p. 215. According to Seltzer, the fact
that new hires at the NYPbil were about 45 percent female
during Mebta's tenure as conductor suggests that Mehta's
views may have changed.
'•' In comparison with the Llnited Kingdom and the two
Germanys. tbe United States in 1990 had the highest percentage
female among its regional symphony orchestras and was a
close second lo the United Kingdom in the major orchestra
category (Juna J. Allmendinger et al., 1996).
'•* In addition, an African-American cellLst (Earl Madison)
brought a civil suit against the NYPbil in 1968 alleging
tbat their audition procedures were discriminatory becau.se
they did not use a screen. Tbe orche.stra was found not guilty
of discriminating in hiring permanent musicians, hut it was
found to discriminate in hiring substitutes.
720 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
0.6
m 0.5
B
I
04
1950
0-6
« 0.5
0.4
0.3
01
to
® 0.2
(0E
to
"- 0.1
0.0
1950
NY Phil
BSO
Chicago
Cleveland
1960 1970 1980 1990
^ v _ LA
SF
—0— Detroit
—•— PSO
v
— — — ' ——' — • 1 • 1
1960 1970 1980 1990
FIGURE 3. FEMALE SHARE OF NEW HIRES IN EIGHT ORCHESTRAS. 1950 TO 1990'S
A: FOUR OF THE "BIG FFVE"; B: FOUR OTHERS
Source: Roster sample. See text.
Notes: A five-year centered moving average is used. New hires are musicians who were not with the orchestra the previous
year, who remain for at least one additional year, and who were not substitute musicians in Ibe current year.
the possibility of discrimination and increase the
number of women in orchestras.
II. Orchestral Auditions
To understand the impact of the democratization
of the audition procedure and the
screen, we must first explain how orchestra
auditions are now conducted. After determining
that an audition must be held to fill an
opening, the orchestra advertises that it will
hold an audition. Each audition attracts musicians
from across the country and, often,
VOL. 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIALHY 721
1940 1950 1960 1970 1980 1990 2000
FIGURE 4. PROPORTION FEMALE OF JUILLIARD GRADUATES, TOTAL AND BY SECTION: 1947 TO 1995
Source: Juilliard Music School files.
from around the world.'^ Musicians interested
in auditioning are required to submit a
resume and often a tape of compulsory music
(recorded according to specific guidelines) to
be judged by members of the orchestra. In
some orchestras this prescreening is dispositive;
in others the musician has the right to
audition live in the preliminary round, even if
the audition committee rejects the candidate
on the basis of the tape.'^ All candidates are
given, in advance, most of the music they are
expected to perform at the live audition.
Live auditions today generally consist of
three rounds: preliminary, semifinal, and final.
But there is considerable variation. Although all
orchestras now have a preliminary round, some
have two final rounds and in many there was no
semifinal round until the 198O's. The preliminary
is generally considered a screening round
to eliminate unqualified candidates. As a result.
the committee is free to advance as many, or as
few, as they wish. Candidates advanced from
the semifinal round are generally considered
"acceptable for hire" by the audition committee
(which does not include the music director,
a.k.a. conductor, until the finals). Again, this
means that the committee can advance as many
as it wishes. The final round generally results in
a hire, but sometimes does not.'^
In blind auditions (or audition rounds) a
screen is used to hide the identity of the player
from the committee.'^ The screens we have
seen are either large pieces of heavy (but soundporous)
cloth, sometimes suspended from the
ceiling of the symphony hall, or look like large
room dividers. Some orchestras also roll out a
carpet leading to center stage to muffle footsteps
that could betray the sex of the candidate.'^
Each candidate for a blind audition is given a
number, and the jury rates the candidate's
'^ Orchestral auditions, particularly for the nation's most
prestigious orchestras, are national if nol international, in
scope. Many contestants, the vast majority of whom receive
no travel reimbursement, travel long distances to audition.
The auditions span the fewest number of days possible to
minimize hotel charges.
"'The tape, in this case, provides information to the
candidate of his or her likelihood of success, sparing the
musician a potentially large travel expense.
"There is one exception to this general rule. In rare
cases when ihe committee cannot decide between two or
three candidates, each is invited to play with the orchestra
before the final deci.sion is made.
"* It may also serve to hide the identity of ihe committee
from the player, although thai is not ils main function. We
use the terms "blind" and "screen" interchangeably.
'^ Or, if a carpel is not placed on the stage, the personnel
manager may ask a woman to take off her shoes and he
provides ihe compensating footstep.s.
722 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
performance next to the number on a sheet of
paper. Only the personnel manager knows the
mapping from number to name and from name
to other personal information.^" The names of
the candidates are not revealed to the juries until
after the last blind round.
Almost all preliminary rounds are now blind.
The semifinal round, added as the number of
applicants grew, may be blind. Finals are rarely
blind and almost always involve the attendance
and input ofthe music director.^' Although the
music director still wields considerable power,
the self-governance that swept orchestras in the
197O's has served to contain the conductor's
authoritarianism. The music director can ignore
the audition committee's advice, but does so at
greater peril. Once an applicant is chosen to be
a member of an orchestra, lifetime tenure is
awarded after a brief probationary period. The
basis for termination is limited and rarely used.
The positions we are analyzing are choice jobs
in the musical world. In 1995 the minimum
starting base salary for musicians at the BSO
was $1,400 per week (for a 52-week year), not
including recording contracts, soloist fees, overtime
and extra service payments, bonuses, and
per diem payments for tours and Tanglewood.^"
Are blind auditions truly blind, or can a
trained, accomplished musician identify contestants
solely from differences in playing style,
just as academics can often identify authors of
double-blind papers they get to referee? Unlike
douhie-biind refereeing, for which one sees an
^^ The personnel manager is generally a musician who
played with the orchestra for some time and knows the
players and the conductor well. The duties involve managing
the day-to-day work of the orchestra, getting substitute
musicians, making travel plans, and arranging the hiring of
new musicians.
^' It is almost always the case that if an orchestra in. say,
the spring of 1986 holds a hlind preliminary round for a
position, it will have all its candidates audition blind in that
round and in all other preliminary rounds during that season,
should there he any. That is. there is generally no discretion
on the part ot" the jury (and certainty not on the part of the
contestant) in temi.s of the audition prcK-edure, particularly
once an audition i.s underway.
" Most of the orchestra contracts in the group we have
examined have similar base salatics. Union contracts list
only the minimum or base starting salary and minimum
increments for seniority. We do not know how many musicians
have individually negotiated rates above the stated
minimum amounts.
entire paper with its distinctive writing style,
methodology, sources, and citations, the candidates
play only predetermined and brief excerpts
from the orchestral repertoire. Each
candidate typically has just 5 to 10 minutes to
play for the audition committee, particularly in
the early rounds. There is little or no room for
individuality to be expressed and not much time
for it to be detected.^ Even when an individual
musician is known in advance to be auditioning,
jury members often cannot identify that individual.
Only the rare, well-known candidate, with
an unusually distinctive musical style could
conceivably be correctly identified.
The many musicians and personnel managers
with whom we have spoken uniformly deny that
identification is possible for the vast majority of
contestants. They also observe that, although it
is tempting to guess the identity of a contestant,
particularly in the later rounds, audition committee
members, more often than not, find they
are wrong. To base a hiring decision on speculation
would not be in the best interests of the
orchestra. Further, although an individual committee
member may believe that he or she
knows the identity of a player, it would be rare
for the entire committee to be secure in such
knowledge. Thus, even if one committee member's
vote is swayed by such a belief, the committee's
vote must correspond to the consensus
view of the player's musical ability for it to
determine the outcome. Thus, auditions held
with a screen, apart from very few exceptions,
are truly blind.
The audition procedures of the 11 orchestras
in the roster sample are summarized in
Table \.^* Although audition procedures are
now part of union contracts, that was not the
case in the more distant past and the procedures
were not apparently recorded in any
surviving documents. We gathered information
on these procedures from various
sources, including union contracts, interviews
with personnel managers, archival documents
on auditions, and a mail survey we conducted
of orchestral musicians concerning the proce-
" Also, there is generally not a standing audition committee
that might hecome familiar with the musicians who
audition frequently.
-•* We identify the orchestras by letter, rather than by
name, to preserve confidentiality of the audition sample.
VOL. 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIAUTY 723
TABLE 1—ORCHESTRA AuDmoN PROCEDtiRE SUMMARY TABLE
Orchestra
A
B
C
D
EF
G
HIJ
K
Preliminaries
Blind since 1973
Blind since al leasl 1967
Blind since at least 1979
(definitely afler 1972)
Blind since 1986
Use of screen varies until 1981
Blind since at leasl 1972
Blind since 1986
Blind since 1970
Blind since 1979
Blind since 1952
Not blind
Semifinals
Blind (varies) since
1973
Use of screen varies
Not blind: 1991-preseni
Blind: 1984-1987
Blind since 1986; varies
until 1993
Use of screen varies
Blind since at least
1972
Use of screen varies
Not blind
Blind since 1979
Blind since 1952
Not hlind
Finals
Nol blind
Blind 1967-1969; since
winter 1994
Not blind
1st pan blind since 1993;
2nd part not bbnd
Not blind
Blind since at least 1972
Not blind
Not blind
Blind since fall 1983
Nol blind
Not blind
Notes: The 11 orchestras (A dirough K) are ibose in the roster sample described in the text. A subset of eight form the audition
sample (also described in the text). All orchestras in tbe sample are major big-city U.S. symphony orchestras and include the
"Big Five."
Sources: Orchestra union contracls (from orchestra personnel managers and libraries), personal conversations with orchesUa
personnel managers, and our mail survey of current orchestra members who were hired during the probable period of screen
adoption.
dures employed during Ihe audition that won
them their current position.
An obvious question to ask is whether the
adoption of the screen is endogenous. Of particular
concern is that more meritocratic orchestras
adopted blind auditions earlier, producing
the spurious result that the screen increased the
likelihood that women were hired.""^ We estimate
a probit model of screen adoption by year,
conditional on an orchestra's not previously
having adopted the screen {an orchestra exits
the exercise once it adopts the screen). Two
time-varying covariates are included to assess
commonly held notions about screen adoption:
the proportion female (lagged) in the orchestra,
and a measure of tenure (lagged) of then-current
orchestra members. Tenure is included because
personnel managers maintain the screen was
advocated more by younger players.
As the proportion female in an orchestra increases,
so does the likelihood of screen adoption
in the preliminary round, as can be seen in
^^ Note, however, it is unlikely that the orchestras tbat
sought to bire more women chose to adopt the screen earlier
sinee the besl way to increase the number of women in the
orchestra is to have not-blind auditions (so that one could be
sure to bire more women).
columns (1) and (2) in Table 2, although the
effects are very small and far from statistically
significant."*' We estimate a similar effect when
we assess the role of female presence on the
adoption of blind finals [see column (3)]. The
impact of current tenure, measured by the proportion
with less than six years with the orchestra,
is—contrary to general belief^negative
and the results do not change controlling for
whether the orchestra is one of the "Big Five.""^
In all. it appears that orchestra sex composition
had little influence on screen adoption, although
the stability of the personnel may have increased
its likelihood.^^
~^ An increase in tbe proportion female from 0 to 0.35,
the largest for any of the orcbestras (see Figure I), would
enhance the likelihood of adopting tbe screen in the preliminary
round by a mere 0.0021 percentage points,
^^ Our measure of tenure begins at ihe first date for
whieh we have rosters, but not earlier than 1947. Tenure
then cumulates for each member until the individual exits
the orchestra. Because tenure will increase for all orchestras
with time, we use the proportion of all members with fewer
than six years of tenure.
^^ A change in conductor could also have led to a change
in the audition policy, but we lind no supporting evidence.
For example, current players contend that Charles Munch
had complete authority in hiring at the BSO before 1952.
The BSO adopted the screen in 1952, but Munch was
724 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
TABLE 2—ESTIMATED PROBIT MODELS
FOR THE USE OF A SCREEN
(Proportion female),_,
(Proportion of orchestra
personnel with <6
years tenure),. ,
"Big Five" orchestra
pseudo R^
Number of observations
Preliminaries blind
(1)
2.744
(3.265)
10.006]
-26.46
(7.314)
1-0.058]
0.178
294
(2)
3.120
(3.271)
[0.004]
-28.13
(8.459)
[-0.039]
0.367
(0.452)
[0.001]
0.193
294
Finals
blind
(3)
0.490
(i.l63)
[0.011]
-9.467
(2.787)
[-0.207]
0.050
434
Notes: Tbe dependent variable is 1 if the orchestra adopts a
screen, 0 otherwise. Huber standard errors (with orchestra
random effects) are in parentheses. All specifications include
a constant. Changes in probabilities are in brackets.
"Proportion female" refers to the entire orchestra. "Tenure"
refers lo years of employment in the current orchestra. "Big
Five" includes Boston, Chicago. Cleveland, New York Philharmonic,
and Philadelphia. The data begin in 1947 and an
orchestra exits the exercise once it adopts the screen. The
unit of observation is an orchestra-year.
Source: Eleven-orchestra rosier .sample. See text.
m. The Role of Blind Auditions on the
Audition and Hiring Process
A. Data and Methods
Audition Records.—^We use the actual audition
records of eight major symphony orchestras
obtained from orchestra personnel managers and
the orchestra archives. The records are highly confidential
and occasionally contain remarks (including
those of the conductor) about musicians
currently with the orchestra. To preserve the full
confidentiality of the records, we have not revealed
the names of the orchestras in our sample.
Although availability differs, taken together
we obtained information on auditions dating
from the late l950's through 1995. Typically,
the records are lists of the names of individuals
conductor from 1949 to 1962. Our inability to explain the
timing of screen adoption may result from our lack of
intimate knowledge of the musical world, although it is also
difficult to explain blind refereeing policy among economies
journals (see the list in Blank, 1991).
who attended the auditions, with notation near
the names of those advanced to the next round.
For the preliminary round, this would indicate
advancement to either the semifinal or final
round. Another list would contain the names of
the semifinalists or finalists with an indication
of who won the audition."^"^ From these records,
we recorded the instrument and position (e.g,,
section, principal, substitute) for which the audition
was held. We also know whether the
individual had an "automatic" placement in a
semifinal or final round. Automatic placement
occurs when a musician is already known to be
above some quality cutoff and is invited to
compete in a semifinal or final round.''^ We also
recorded whether the individual was advanced
to the next round of the current audition.
We rely on the first name of the musicians to
detennine sex. For most names establishing sex
was straightforward."*' Sexing the Japanese and
Korean names was equally straightforward, at
least for our Japanese and Korean consultants.
For more difficult cases, we checked the names
in three haby books (Connie Lockhard Ellefson,
1990; Alfred J. Kolatch, 1990; Bruce Lansky,
1995). If the name was listed as male- or
female-only, we considered the sex known. The
gender-neutral names (e.g., Chris, Leslie, and
Pat) and some Chinese names (for which sex is
indeterminate in the absence of Chinese characters)
remained ambiguous. Using these procedures,
we were able to determine the sex of
96 percent of our audition sample.^^ We later
assess the impact that sex misclassification may
have on our results.
In constructing our analysis sample, we exclude
incomplete auditions, those in which there
were no women (or only women) competing,
rounds from which no one was advanced, and
the second final round, if one exists, for which
In rare cases, we have additional information on tbe
finalists, such as resumes,
•™ The person will be known to be above a quality eutoff
either because the individual is a current member of a
comparable orchestra or because the person was a semifinalist
or finalist in a previous audition.
^' For 13 percent of the contestants, sex was confirmed
by personnel managers, resumes, or audition summary
sheets.
•'' Most of the remainder were sexed using census data
by assigning to them the dominant sex of individuals with
their first name.
VOL 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIAUTY 725
TABLK 3—DESCRIPTIVE STATISTICS ABOLIT Ai^DmoNS. BY YEAR AND ROUND OF AUDITION
Year
Number of
auditions
Proportion
female
Number of Number of Proportion
musicians auditions female
Number of Number of Proportion
musicians auditions female
Completely blind auditions Not completely blind auditions
All
Pre-1970
1970-1979
1980-1989
1990+
Round
Preliminaries.
without
semifinals
Preliminaries,
with
semifinals
Semifinals
Finals
254
10
69
102
73
170
137
114
167
0.367
(0.013)
0.187
(0.042)
0.329
(0.026)
0.394
(0.019)
0.390
(0.027)
0.357
(0.015)
0.3%
(0.019)
0.415
(0.019)
0.430
(0.016)
43.4
(3.13)
42.5
(4.29)
44.6
(4.64)
34.3
(1.S7)
45.5
(2.54)
12.3
(0.649)
4.93
(0.448)
60
33
27
Blind
rounds
125
134
89
28
0.393
(0.029)
0.375
(0.034)
0.415
(0.049)
0.367
(0.017)
0.395
(0.019)
0.404
(0.022)
0.472
(0.040)
38.1
(1-74)
16J
( 2 ^
31.4
(2.10)
39.6
(2.73)
50.6
(4.52)
24.7
(2.33)
49.3
(17.0)
10.4
(1.21)
7.12
(0.310)
194
10
69
69
46
Not-blind
rounds
45
3
25
130
0.359
(0.015)
0.187
(0.042)
0.329
(0.026)
0.403
(0.022)
0.375
(0.033)
0.327
(0.029)
0.425
(0.205)
0.455
(0.043)
0.422
(0.017)
Notes: The unit of ob.servation for the top portion is the audition, whereas it is the round for the bottom portion (e.g..
proportion female in the top portion of the table is averaged across the auditions). Standaid eirors are in parentheses.
Source: Eight-orchestra audition sample. See text.
the candidates played with the orchestra.^^ In
addition, we generally consider each round of
the audition separately. These sample restrictions
exclude 294 rounds (199 contained no
women) and 1.539 individuals. Our final analysis
sample has 7,065 individuals and 588 audition
rounds (from 309 separate auditions)
resulting in 14,121 person-rounds and an average
of 2.0 rounds per musician."*•*
As can be seen In the bottom portion of
Table 3, 259. or 84 percent, of our 307 preliminary
rounds were blind. 78 percent ofthe
114 semifinals were blind, but just 17 percent
of the 167 final rounds were hlind. Most of
our audition sample is for the period after
1970. The blind preliminaries contained 40
• Although the results are unaffected, harp auditions are
excluded because it has typically been a female-dominated
instrument.
^ See Table Al for descriptive statistics.
candidates on average, whereas those without
the screen had 26. Women were about 37
percent of all preliminary candidates but 43
percent of finalists, and the difference holds
for both the blind and not-blind auditions.
The percentage female among all candidates
increased over time, from 33 percent in the
1970 to 1979 period to 39 percent in the
post-1990 years (see upper portion).
Roster Data.—Our second source of information
comes from the final results of the audition
process, the orchestra personnel rosters.
We collected these data from the personnel page
of concert programs, one each year for eleven
major symphony orchestras. These records are
in the public domain and thus we have used the
orchestra names in the graphs containing those
data alone. As opposed to the auditionees, we
were able to confirm the sex of the players
with the orchestra personnel managers and
726 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
archivists. We considered a musician to be new
to the orchestra in question if he or she had not
previously been a regular member of that orchestra
(i.e., we did not count returning members
as new). We excluded, when possible,
temporary and substitute musicians, as well as
harpists and pianists. Our final sample for 1970
to 1996 has 1,128 new orchestra members (see
Table A2).
Econometric Framework.—We take advantage
of the variation that exists across orcbestras,
time, and audition round to identify the
effect of the screens on the likelihood that a
female is advanced from one round to the next
and ultimately hired. The probability that individual
i is advanced (or hired) from an audition
at orchestra j , in year /, from round r, is a
function of the individual's sex (f), whether a
screen is used (B), and other individual (X) and
orchestral (Z) factors, that is:
(1)
The screen, it will be recalled from Table 1, varies
across orchestra, time, and audition round.
Orchestras adopted the screen in different years.
Some used the screen in the preliminary round
only, whereas others used the screen for the
entire audition process. We use this variation to
estimate a differences-in-differences strategy.
In linear form, we write
(2) P,^,, ^ a + /3F,- + yBj,, + S(F,- X B,,,)
The coefficient on Bj,^, y, identified from the
men who audition with a screen, controls for
whether all individuals are more or less likely to
be advanced from a blind than from a not-blind
audition. Thus the parameter of interest is that
on the interaction between F, and Bj^,, S, which
measures the change in the probability that a
woman will be advanced if a screen is used,
relative to her auditioning without a screen (after
accounting for other blind audition effects).
We aiso test whether the use of the screen
eliminates sex differences in the likelihood an
individual is advanced from one round to the
next. Because no restrictions exist on the number
of individuals advanced from the preliminary
and semifinal rounds, there is no zero-sum
game between men and women for these
rounds.
B. The Effect of the Screen on the Likelihood
of Being Advanced
Tabulations and Regression Results With and
Without Individual Fixed Effects.—The raw
data in Tables 4 and 5 can reveal the impact on
women of changes in the audition process and
provide an important introduction to the data.
We demonstrate that in the absence of a variable
for orchestral "ability," women fare less well in
blind auditions than otherwise. But if the orchestral
"ability" of the candidate is held fixed,
the screen provides an unambiguous and substantial
benefit for women in almost all audition
rounds.
Table 4 gives the success rate by sex, round
of audition, and over time. We define "relative
female success" as the proportion of women
advanced (or hired) minus the proportion of
men advanced (or hired). The relative success
of female candidates appears worse for blind
than for not-blind auditions and this finding also
holds for each round of the audition process.
One interpretation of this result is that the adoption
of the screen lowered the average quality of
female auditionees in the blind auditions. Only
if we can hold quality constant can we identify
tbe true impact of the screen.
Because we have tbe names of the candidates,
we are able to link their success in one
audition to that in another. (In our sample, 24
percent of the individuals competed in more
than one audition.) In Table 5 we report audition
success statisfics, by round and overall, for
musicians who appear more than once in our
sample and for whom at least one audition (or
round) was blind and one was not blind. The
evidence tells a very different story from that in
Table 4, and taken together they suggest tbat
blind auditions expanded the pool of female
applicants to include more who were less qualified.
When we limit the sample to those who
auditioned both with and without a screen, the
success rate for women competing in blind auditions
is almost always higher than in those
that were not blind.
VOL 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIAUTY 727
TABLE A—AVERAGE SUCCESS AT AUDITIONS BV SEX, YEAR, AND ROUND OF AUDITION
Year
All
Pre-1970
1970-1979
1980-1989
1990+
Round
Preliminaries, without semifinals
Preliminaries, with semifinals
Semifinals
Finals
All audition.s
-0.001
(0.008)
0.053
(0115)
0.001
(0.021)
-0.006
(0.009)
-0.003
(0.010)
All rounds
-O032
(0.019)
-0.048
(0.016)
-0.030
(0.038)
0.009
(0.036)
Relative female success
Completely blind
auditions
-0.022
(0.012)
-0.039
(0.016)
-0.001
(0.017)
Blind rounds
-0.048
(0.021)
-0.052
(0.016)
-0.059
(O044)
-0.028
(0.102)
Not completely blind
auditions
0.006
(0.010)
0.053
(0.115)
0.001
(0.021)
0.010
(0.009)
-0.003
(0.013)
Noi-blind rounds
0.012
(0.040)
0.116
(0.228)
0.071
(O080)
0.016
(0.038)
Notes: For the top part of the table "success" is a "hire." whereas for the bottom portion "success" is advancement from one
stage of an audition to the next. The unit of observation for the top ponion is the audition, whereas it is the round for the
bottom ponion (e.g.. relative female success in the top ponion of the table is averaged across ihe auditions). Standard errors
are in parentheses. "Relative female success" is the proportion of women advanced (or hired) minus the proponion of men
advanced (or hired). By hired, we mean those who were advanced from the final round oul of the entire audition.
Source: Eight-orchestra audition sample. See lext.
Take the preliminary round with no semifinals,
for example, in Table 5. In the blind auditions 28.6
percent of the women are advanced, as are 20.2
percent of the men. But in the not-blind column,
just 19.3 percent of the women are advanced,
although 22.5 percent ofthe men are. Even though
a woman has a small advantage over a man when
the screen is used (by 8.4 percentage points), her
success rate, relative to that of a man, is increased
by 11.6 percentage points above that in the notblind
regime. Note that because these are the same
women. Table 5 suggests that a woman enhances
her own success rate by 9.3 percentage points by
entering a blind preliminary round. Not only do
these differences suggest that women are helped
by the screen, the differences are large relative to
the average rate of success.^^
Women's success is also enhanced by the
'^ Because of the inlrequency of position availability, it
is unlikely there was much gaming by women (e.g.. trying
out only lor blind auditions), although the change in the
screen in the finals and for the overall audition
(termed "hired" in the table). For the finals, a
woman's success rate is increased by 14.8 percentage
points moving to blind auditions
(23.5 - 8.7) and is enhanced by a hefty 28.1
percentage points above that of men. All success
rates are very low for auditions as a whole,
hut the female success rate is 1.6 times higher
(increasing from 0.017 to 0.027) for blind than
for not-blind auditions. The only anomalous
result in the table concems the semifinals, to
which we return later. We now show that these
results stand up to the controls we can add.
including the year of the audition and the
instrument.^''
general environment of auditions could have altered the
pool of contestants.
'^ We do not discuss the regression analog to Table
4. that is, the analysis without individual fixed effects,
because we have firmly established thai individual fixed
effect.^ matter. Table A3 shows the results of regressions
728 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
TABLE 5—AVERAGE SUCX:ESS AT AUDITIONS BY SEX AND STAGE OF AUDITION FOR THE SUBSET
OF MUSICIANS WHO AUDITIONED BOTH BLIND AND NOT BLIND
Women
Men
Women
Men
Women
Men
Women
Men
Women
Men
Proportion
advanced
0.286
(0.043)
0.202
(0.026)
0200
(0.092)
0.083
(0.083)
0.385
(0.061)
0.368
(0.059)
0.235
(0.106)
0.000
(0.000)
0.027
(0.008)
0.026
(0.005)
Blind
Number of
person-rounds
Preliminaries without
112
247
Proportion
advanced
semifinals
0193
(0.041)
0.225
(0.031)
Preliminaries with semifinals
20
12
Semifinals
65
68
Finals
17
12
"Hired"
445
816
0133
(0.091)
0.000
(0.000)
0.568
(0.075)
0.295
(0.069)
O087
(0.060)
0.133
(0.091)
0.017
(0.005)
0.027
(0.005)
Not blind
Number of
person-round.s
93
187
15
8
44
44
23
15
599
1102
Notes: The unit of observation is a person-round. Standard errors are in parentheses. For the round in question, only musicians
who auditioned more than once and who auditioned at least once behind a screen and at least once withoui a screen are
included. "Hired" means those who were advanced from the final round out of the entire audition. Blind in the "hired"
category means for all rounds. Not blind in the "hired" category means that at least one round was not blind. This difference
in the definition of what consfitutes a "blind" round or audition is one reason why the number of observations in the first four
panels is less (han the number of observations in the "hired" panel. The number of observations also differ because we exclude
auditions or rounds in which no individual is advanced or in which there are only women or no women. Finally, unlike in
subsequeni tables, we exclude a (ew candidates for whom we could not detennine or impute their sex. Note that the binding
constraint for the preliminaries is the not-blind category, for which we have only one orchestra. The binding constraint in the
"hired" category are the blind auditions, for which we have (at most) three orchestras. Musicians can appear more than once
in either the blind or not-blind categories.
Source: Eight-orchestra audition sample. See text.
The results given in Table 6 are the regression
analogs to the raw tabulations in Table
5.-*^ Because the effect of the blind procedure
comparable to those in Table 6 but without individual fixed
effects.
^^ In the (total) subsample of individuals auditioning
both with and without a screen, all eight orchestras in our
audition sample are represented, and seven of the orchestra.s
changed audition policy during our sample time frame. The
sample sizes in Table 6 are considerably larger than those in
could differ by the various rounds in the audition
process, we divide audition rounds into
the three main rounds (preliminary, semifinal,
and final) and also separate the preliminaries
into those that were followed by a semifinal
Table 5. The reason is that the regressions in Table 6 Include
ail individuals whether or not they auditioned more than
once, whereas Table 5 includes only those who auditioned
at least twice, bhnd and not blind.
VOL 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIAUTY 729
TABLE 6~LiNEAR PROBABILITY EsirMATES OF THE LIKELIHOOD OF BEING ADVANCED: WITH INDIVIDUAL FIXED EFFECTS
Blind
Female x Blind
Number of auditions attended
Years since last audition
Automatic placement
"Big Five" orchestra
Total number of auditioners in
round (-100)
Proportion female at the audition
round
Principal
Substitute
p-value of //(,: Blind + (Female
X Blind) = 0
Year fixed effects?
RNumber
of observations
Preliminaries
Without
semifinals
(!)
-0.017
(0.039)
0.125
(0.068)
0.053
No
0.748
5,395
(2)
0.003
(0.046)
0.111
(0.067)
-0.020
(0.014)
-0.005
(0.007)
-0.154
(0.035)
-0.003
(0.081)
0.118
(0.139)
-0.079
(0.037)
0.165
(0.081)
0.063
Yes
0.775
5,395
With semifinals
(3)
0.109
(0.172)
0.013
(0.215)
0.342
No
0.687
6.239
(4)
0.224
(0.242)
-0.025
(0.251)
0.010
(0.010)
-0.006
(0.005)
-0.059
(0.024)
0.014
(0.031)
0.312
(0.134)
-0.078
(0.019)
0.123
(0.093)
0.285
Yes
0.697
6,239
Semifinals
(5)
0.026
(0.089)
-0.179
(0.126)
0.089
No
0.774
1.360
(6)
0.102
(0.096)
-0.235
(0.133)
0.015
(0.030)
-0.005
(0.013)
-0.096
(0.064)
0.006
(0.081)
-0.371
(0.521)
0.104
(0.218)
-0.082
(0.066)
0.167
(0.183)
0.170
Yes
0.794
1,360
Finals
(7)
-0.154
(0.150)
0.308
(0.196)
0.222
No
0.811
1,127
(8)
-0.060
(0.149)
0.331
(0.18!)
0.126
(0.028)
0.016
(0.015)
-0.069
(0.073)
-0.059
(0.084)
-0.262
(0.756)
0.067
(0.159)
-0.185
(0.076)
0.079
(0.217)
0.042
Yes
0.878
1.127
Notes: The unit of observation is a person-round. The dependent variable is 1 if the individual Is advanced to the next round
and 0 if not. Standard errors are in parentheses. All specifications include individual fixed effects, an interaction for the sex
being missing and a blind audition round, a dummy indicating if years since last audition is missing, and (in columns (3)-(8)]
whether an automatic placement is missing.
Source: Eight-orchestra audition sample. See text.
round and those that were not. In the evennumbered
columns we include year and instrument
fixed effects, as well as individual
and audition covariates. The individual correlates
are whether the musician had an automatic
placement in a semifinal or final round,
years since the last audition in the sample,
and the number of previous auditions in
which we observe the musician to have competed.
We also control for the total number of
musicians in the round, the proportion female
among contestants, and whether the audition
is for a principal or substitute position.
Because 42 percent of the individuals in our
sample competed in more than one round in our
data set (24 percent of the musicians competed
in more than one audition) and 6 percent competed
both with and without a screen for a
particular type of round (e.g., semifinal), we are
able to use an individual fixed-effects strategy
to control for contestant "ability" that does not
change with time. In all columns of Table 6 we
include individual fixed effects, in which case
the identification is from individuals who auditioned
both with and without a screen.^** The
'^ There are 639 person rounds comprised of individuals
who auditioned at a preliminary round that was not followed
by a semifinal round [columns (I) and (2) of Table 6], hoth
with and without a screen; on average the.se individuals
competed in 2.7 such preliminary rounds. There are 55
person-rounds comprised of individuals who auditioned at a
preliminary round that was followed by a semitinal round
[columns (3) and (4)], both with and without a screen; on
average these individuals competed in 2.4 such preliminary
rounds. There are 223 person-rounds comprised of individuals
who auditioned at a semifinal [columns (5) and (6)],
730 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
effect ofthe screen here, therefore, is identified
from differing audition procedures both within
and across orchestras.^^"^ Note that we include a
dummy variable for whether the orchestra is
among the "Big Five," to control for the quality
of the orchestra.
The coefficient of interest is the interaction
between "Female" and "Blind." A positive
coefficient would show that screened auditions
enhance a woman's likelihood of advancement.
Because screened auditions are
more likely to take place in later years than
auditions without screens, the interaction between
"Female" and "Blind" might simply
reflect the fact that female musicians get better
over time. Note, however, that for this
effect to bias the coefficient, female musicians
would have to improve faster with time
than male musicians. Nevertheless, we have
also included (in the individual covariates)
the number of previous auditions the musician
attended in our sample, the number of
years since the last audition in the sample,
and whether the candidate was an automatic
placement. The coefficient on "Blind" reveals
whether blind auditions change the likelihood
that all contestants are advanced.
As in the raw tabulations of Table 5, we
find that the screen has a positive effect on the
likelihood that a woman is advanced from the
preliminary round (when there is no semifiboth
with and without a screen; on average these individuals
participated in 2.8 semifinal rounds. Finally, there are 67
person-rounds comprised of individuals who auditioned at a
final round [columns (7) and (8)]. both with and without a
screen; on average these individuals participated in 2.4 final
rounds. It should he noted that the number of person-rounds
off of which we are identified in Table 6 can aiso be found
in Table 5. with one exception. There are 223 person-rounds
comprised of individuals who auditioned at the semifinal,
both with and without a screen, in Table 6 and only 221 in
Table 5 because there are two individuals we could not sex.
We include these individuals in the regressions in Table
6 and add a dummy variable indicating that the sex is
missing.
^^ An analysis of variance (ANOVA) across the entire
sample, that is pooling ail rounds, indicates that 19 percent
of the variation in the use of the screen is across orchestras.
Looking by audition round reveals that 73 percent of the
variation in preliminaries, 53 percent of the variation in
.semifinals, and 71 percent of the variation in finals is across
orchestras. By contrast, in Table 7 (which includes a subset
of the orchestras, see table notes), just 1 percent of the
variation in the use of the screen is across orchestra.s.
nal) and from the finals.*" The effects, moreover,
are statistically significant in both cases.
The effect in the semifinal round, however,
remains strongly negative."' In addition, the
magnitudes of the effects in Table 6 are similar
to those implied by the raw tabulations
(Table 5). For preliminaries that are not preceded
by a semifinal, the blind audition increases
the likelihood that a woman will be
selected by about 11 percentage points. For
female musicians who made it to the final
round, the individual fixed-effects regression
result indicates that the screen increases the
likelihood of their winning by about 33 percentage
points.42
Assessing Potential Biases.—A concem with
the preceding fixed-effects analysis is that, as
noted earlier, female musicians who are improving
over time are those who switch from
not-blind to blind auditions and that the growth
rate of their "ability" is faster than that of men.
We attempted to address this potential bias by
including several individual time-varying co-
""' An exception occurs when preliminaries are followed
by semifinals. There are, however, only three preliminary
rounds that are not blind when there is also a semifinal
round (see Table 3). Thus the coefficients in columns (3)
and (4) of Table 6 are identified using very few separate
audition rounds. We also note that when we estimate fixedeffects
logit models we obtain results similar to those in
columns (I) and (2) in Table 6 (and in Table 7). Because of
the small samples with the identifying requirements of the
fixed-effects logit. standard errors for the estimates in columns
(3)-(8) of Table 6 could not be computed. Further, for
the results without individual fixed effects. logits and linear
probability models give qualitatively similar results.
"" This result on the semifinal.s is robust across time,
instrument, position, and orchestra. One interpretation is
that it represents a form of affirmative action by the audition
committees. Committee.s may hesitate to advance women
from the preliminary rotmd if they are not confident of the
candidate's ability. On the other hand, semifinals are typically
held the same day as are preliminaries and give the
audition committee a second chance to hear a candidate
before the finals. Thus, audition committees may actively
advance women to the final round only when they are
reasonably confident that the female candidate is above
some threshold level of quality. If juries actively seek to
increase the presence of women in the final round, they can
do so only when there is no screen.
"- As noted earlier, an obvious explanafion for the importance
of the individual fixed effects in the estimation is
that the screen altered the pool of female applicants; however,
we have been unable to show this empirically.
VOL 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIAUTY 731
variates (in the even-numbered columns of Table
6). The inclusion of these individual
covariates had little effect on the estimated effect
of the screen.
A related concem is that those individuals
who get hired at their first audition, and therefore
do not contribute to the identification ofthe
effect in the presence of individual fixed effects,
are more able musicians than those who audition
multiple times. (Alternatively, some individuals
who audition and are not hired may get
discouraged and not audition again and are
therefore worse than those who audition multiple
times.) Although this is a potential source of
bias, it is important to remember that only a
very small number of musicians win an audition
in any given year, since there are just a handful
of auditions (for a given instrument) among the
major orchestras. Furthermore, many of the
contestants in our sample did audition at least
twice.
In addition, there are three pieces of empirical
evidence that suggest this potential
source of bias is not a major problem in our
data. First, we control for the number of previous
auditions in the even columns of Table
6, and this control does not change the results
significantly. Second, there is no significant
difference in the proportion female among
those who auditioned both with and without a
screen and those who auditioned only once
(or who auditioned under only one policy
regime). Finally, the coefficient estimates
generated when the sample is restricted to
those who auditioned at least three times are
not perceptibly different from those generated
from the full sample or from the sample of
individuals who auditioned both with and
without a screen. (These results are presented
in Table A4.)
A third potential bias is that, because the
effect ofthe screen is partially identified from
differing audition procedures across orchestras,
the results in Table 6 may indicate that
orchestras that use screens are less discriminatory
against women than those that do not.
Specifically, because we include individual
fixed effects, a bias would arise if women
who are improving faster than average are
more likely to audition for orchestras that use
screens and are more likely to be advanced
because these orchestras are intrinsically less
discriminatory. Our sample contains only one
orchestra per audition round that changed
policy. As a result, we cannot separate the
estimation by audition round and include orchestra
fixed effects. We can. however, poo!
the audition rounds for the three orchestras
that changed audition policy during our sample
frame and include both individual and
orchestra fixed effects.•*' These results are
presented in Table 7.
In column (1) of Table 7 we include individual
fixed effects, in which case the identification is
from individuals who auditioned both with and
without a screen. We add orchestra fixed effects in
column (2) such that the identification now is from
individuals who auditioned for a particular orchestra
both before and after the orchestra began using
a screen."*^ Finally, in column (3) we exclude
individual but keep orchestra fixed effects to illustrate
the importance of individual fixed effects.
Again, the coefficient on "Blind" shows whether
all musicians are more likely to be advanced when
the audition is blind. The interaction between
whether the individual is female and whether the
audition is blind indicates whether women receive
an extra boost relative to men when the screen is
used.
The coefficient of interest is positive in
columns (1) and (2) but negative in column
(3), similar to the difference between the
tabulations in Tables 4 and 5. In addition,
the estimated effect of the blind auditions on
the success of women is similar to that in
Table 6. The point is that individual fixedeffects
estimation matters; orchestra fixed effects,
however, do not matter. In all cases,
blind auditions increase the probability of advancement
for both men and women. More
•'^ We do not include the type of audition round since we
have only one orche.stra that changed procedures for the
preliminaries, one that changed for the semifinals, and one
thai changed for the finals (and for which there were musicians
who auditioned for that orchestra and audition round
with and without a screen). We have also estimated these
regressions separately for each of these three orchestras.
Although ihe point estimates are not statistically significant.
the magnitudes are quite similar to those presented in Table
6 for the corresponding round of the audition.
"" In this subsampie. there are 1,776 person-rounds comprised
of individuals who auditioned for a particular orchestra,
both behind and without a screen; on average these 552
individuals competed in 3.2 audition rounds.
732 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
TABLE 7—LINEAR PROBABILITY ESTIMATES OF THE LIKELIHOOD OF BEING ADVANCED: WITH
INDIVIDUAL AND ORCHESTRA FIXED EFFECTS
Blind
Female x Blind
Female
p-value of H^:
Blind + (Female X Blind)
Individual fixed effects?
Orchestra fixed effects?
Year fixed effects?
Other covariales?
RNumber
of observations
= 0
Include
fixed
(1)
0.404
(0.027)
0.044
(0.039)
0.000
Yes
No
Yes
Yes
0.615
8.159
individual
effects
(2)
0.399
(0.027)
0.041
(0.039)
0.000
Yes
Yes
Yes
Yes
0.615
8,159
Exclude individual
fixed effects
(3)
0.103
(0.018)
-0.069
(0.022)
-0.005
(0.019)
0.090
No
Yes
Yes
Yes
0.048
8.159
Note.s: The unit of observation is a person-round. The dependent variable is 1 if the person is
advanced to the next round and 0 if not. Standard errors are in parentheses. All specifications
include an interaction for the sex being missing and a blind audition: "Other covariales" include
automatic placement, years since last audition, number of auditions attended, size of the audition
round, proportion female in audition round, whether a principal or substitute position, and a
dummy indicating whether years since last audition is missing. These regressions include only the
orchestras thai changed their audition policy during our sample years and for which we observe
individuals auditioning for the audition round both before and after the policy change. These
regressions include 4,836 separate persons and are identified off of 1.776 person-rounds comprised
of individuals who auditioned both before and after the policy change for a particular orchestra.
Source: Eight-orchestra audition sample (three orchestras of which are used; see Noles). See text.
important, even though the effect is not statistically
significant, the blind procedure has a
positive effect on women's advancement.'^^
Finally, sex misclassification may also bias
our estimates because, if the misclassification
errors are uncorrelated with the equation error,
the estimated effect of the screen will be attenuated
(see. e.g.. Richard Freeman. 1984). To
address this potential problem, we use a lesssubjective
assessment ofthe probability that the
individual is male or female. A U.S. Bureau of
the Census tabulation, based on the postenu-
*^ Although the results from these three orchestras may
not generalize lo the other five, it should be noted that the
coefficient estimate in column (3) of Table 7 is similar to
that derived from a similar regression on the entire sample.
This result is nol surprising because the primary reason we
are able to include both individual and orchestra fixed
effects for Ihese three orchestras is because they have unusually
good record keeping, which allows us to observe the
results of many auditions rather than another reason that
might be correlated with how meritocratic the orchestra is.
meration survey ofthe 1990 census, gives us the
proportion female and male of the top 90 percent
of all names.*^
In Table 8 we estimate the same specifications
given by columns (2), (4), (6), and (8) of
Table 6 and column (2) of Table 7 using the
census data in two ways. First, we simply replace
our female covariate with the census probability."^^
Note that we also use a census
estimate ofthe percentage ofthe audition round
that is female (slightly changing our sample
size), and a census estimate ofthe percentage of
our sample for which the sex is indeterminate.
In addition, our interaction term is constructed
using the census probabilities. Second, we use
'^ These data can be dowrdoaded from http://www.census.
gov/ftp/pub/genealogy/names. A possible problem with the
data is that names are generational; a male name in one
generation may become female in another.
•"^ We do not impute census probabilities for the individuals
whose sex we know with certainty (see footnote 31).
VOL 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIALITY 733
TABLE 8—LINEAR PROBABILITY ESTIMATES OF THE LIKELIHOOD OF BEING ADVANCED: ADDRESSING SEX MISCLAS.SIFICATION
Blind
Female X Blind
Other covariates?
Individual fixed effects?
Year fixed effects?
R^
Number of observations
Blind
Female x Blind
Other covariates
Individual fixed effects?
Year fixed effects?
R'
Number of observations
Part A; Preliminary rounds
Without semifinals
OLS
-0.012
(0.043)
0.139
(0.066)
Yes
Yes
Yes
0.771
5,696
Part B: Semifinal and final round!
Semifinals
OLS
0.100
(0.083)
-0.242
(0.120)
Yes
Yes
Yes
0.776
1,600
IV
-0.197
(0.700)
-0.193
(0.429)
Yes
Yes
Ye.s
1.360
IV
0.057
(0.045)
0.137
(0.068)
Yes
Yes
Yes
5.395
Preliminaries
OLS
-0.174
(0.093)
0.272
(0.188)
Yes
Yes
Yes
6.546
i. and with orchestra fixed effects
OLS
-0.028
(0.125)
0.160
(0.171)
Yes
Yes
Yes
0.848
1,509
Finals
IV
-0.025
(0.141)
0.324
(0.181)
Yes
Yes
Yes
1.127
With semifinals
IV
0.290
(0.241)
-0.035
(0.251)
Yes
Yes
Yes
6.239
With orchestras fixed
effects
OLS
0.010
(0.028)
0.069
(0.035)
Yes
Yes
Yes
0.654
8,882
IV
0.061
(0.033)
0.052
(0.036)
Yes
Yes
Yes
8.159
Notes: The unit of observation is a person-round. The dependent variable is 1 if tbe individual is advanced to the next round
and 0 if not. Standard errors are in parentheses. The instruments are tbe census probability tbat tbe individual is female, a
dummy for wbether tbe person has been sexed with certainty, and proportion female calculated using the census data and an
interaction between whether the census data are missing and a screen has been used. The "'OLS" columns use these as
regressors. All specifications include an inleraclion for tbe sex being missing and a blind audition; "Otber covariates" include
automatic placement, years since last audition, number of auditions attended, whether a "Big Five" orchestra, size of tbe
audition round, proportion female at the audition round, whether a principal or substitute position, and a dummy indicating
wbether years since last audition and automatic auditioti are missing. These are the same specifications a.s in columns (2). (4).
(6), and (8) of Table 6 and column (2) of Table 7. The sample sizes change because in the even-numbered columns we simply
replace our female covariate witb the census probability and also use a census estimate of tbe percentage of Ibe audition round
that is female, which changes the sample size slightly.
Source: Eigbt-orchestra audition sample. See text.
the census probability as an instrument for our
estimate (and for the percentage of the audition
that is female, the percentage missing sex, and
the interaction between female and whether the
audition is bhnd).
The results are quite robust across these
different methods for addressing potential
measurement error. More important, the coefficients
and their standard errors are generally
similar in magnitude to those in Tables 6
and 7. With the exception of the semifinal
round, the screen appears to have increased
the likelihood that a woman would be advanced.''**
** Another potential bias is from tbe short panel, which
may affect the consistency ofthe estimates (Hsiao. 1986).
We address tbe extent of tbis sbort panel problem in two
ways. We first restrict our sample to tbose whom we observe
auditioning at least ibree times (for the same round).
Second, we restrict tbe estimation to those who auditioned
at least once in a blind round and al least once in a not-blind
round (tbose off of wbom we are identified). The results do
noi cbange markedly from those in Table 6, sbowing tbat
the short panel may not be a problem. See Table A4.
734 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
TABLE 9—LINEAR PROBABILITY ESTIMATES OF THE EFFECT OF BLIND AUDITIONS
ON THE LTKELIHOOD OF BEING HiRED WITH INDIVIDUAL FiXED EFFECTS
Completely hiind auditioti
Completely blind audition X female
Year effects?
Other covariates?
RNumber
of observations
Without
(1)
-0.024
(0.028)
0.051
(0.046)
No
No
0.855
4.108
semifinals
(2)
0.047
(0.041)
0.036
(0.048)
Yes
Yes
0.868
4,108
With
(3)
0.001
(0.009)
0.001
(0.016)
No
No
0.692
5.883
semifinals
(4)
0.006
(0.011)
-0.004
(0.016)
Yes
Yes
0.707
5,883
Al!
(5)
0.001
(0.008)
0.011
(0.013)
No
No
0.678
9,991
(6)
0.005
(0.009)
0.006
(0.013)
Yes
Yes
0.691
9,991
Noles: The unit of observation is a person-round. The dependent variable is 1 if the individual is advanced (or hired) from
the final round and 0 if not. Standard errors are in parentheses. All specitications include individual fixed effects, whether the
sex is missing, and an interaction for sex being missing and a completely blind audition. "Other covariates" are the size of
the audition, the proportion female at the audition, the number of individuals advanced (hired), whether a "Big Five"
orchestra, the number of previous auditions, and whether the individual had an automatic semifinal or final.
Source: Eight-orchestra audition sample. See text.
C. The Effect of the Screen
on the Hiring of Women
Using the Audition Sample.—Our analysis,
thus far, has concerned the rounds of the audition
process and the degree to which the screen
enhances the Hkelihood of a woman's advancing
from one round to the next. We tum now to
the effect of the screen on the actual hire and
estimate the likelihood an individual is hired out
ofthe initial audition pool.'*^ Whereas the use of
the screen for each audition round was, more or
less, an unambiguous concept, that for the entire
process is not and we must define a hlind audition.
The definition we have chosen is that a
blind audition contains all rounds that use the
screen. In using this definition, we compare
auditions that are completely blind with those
that do not use the screen at all or use it for the
early rounds only. We divide the sample into
auditions that have a semifinal round and those
that do not, because the previous analysis suggested
they might differ.
The impact of completely blind auditions on
the likelihood of a woman's being hired is given
in Table 9, for which all results include individ-
•"^ There are four auditions in which the committee could
not choose between two players and therefore asked each to
play with the orchestra. We consider btJth to be winners.
The results are not sensitive to this classification. For this
analysis we exclude audilions with no women, all women,
or no winner: these exclusions do not change the results.
ual fixed effects.-^" The impact of the screen is
positive and large in magnitude, but only when
there is no semifinal round. Women are about 5
percentage points more likely to be hired than
are men in a completely blind audition, although
the effect is not statistically significant.
The effect is nil, however, when there is a
semifinal round, perhaps as a result of the unusual
effects of the semifinal round. The impact
for all rounds [columns (5) and (6)] is about 1
percentage point, although the standard errors
are large and thus the effect is not statistically
significant. Given that the probability of winning
an audition is less than 3 percent, we
would need more data than we currently have to
estimate a statistically significant effect, and
even a 1-percentage-point increase is large, as
we later demonstrate.
" In Table 9 we are identified off of individuals who
competed in auditions that were completely blind and those
that were not completely blind (that is, any one round could
not be blind). The unil of observation is the person-round
and there are 92 fulfilling this criterion for auditions without
a semifinal [columns (I) and (2)i; on average these persons
competed in 3.6 auditions in this sample. There are 625
person-rounds fulfilling this criterion that included a semifinal
[columns (3) and (4)] and on average these persons
competed in 3.5 auditions in this sample. Finally, there are
911 person-rounds fulfilling this criterion across all audition
[columns (5) and (6)] and on average these persons competed
in 3.5 auditions in this sample. The sample off of
which we are identified is larger for ail auditions than for the
sum of ihe other two because some individuals auditioned
both with and without a semifinal round.
VOL 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIAUTY 735
TABLE 10—PROBIT ESTIMATES OF THE EFFECT OF BLIND AUDITIONS ON THE SEX OF NEW
MEMBERS: 1970 TO 1996
Any blind auditions
Only hlind preliminaries and/or
semifinals
Completely blind auditions
Section:
Woodwinds
Brass
Percussion
/7-value of test: only blind preliminaries
and/or semifinals = completely blind
pseudo R'
Number of observations
Any blind
auditions
(1)
0.238
(0.183)
[0.075]
-0.187
(0.114)
[-0.058]
-1,239
(0.157)
[-0.284]
-1.162
(0.305)
[-0.235]
0.106
1,128
Only blind preliminaries
and/or semifinals vs.
completely blind
auditions
(2)
0.232
(0.184)
[0.074]
0.361
(0.438)
[0.127]
-0.188
(0.114)
[-0.058]
-1.237
(0.157)
[-0.284]
-1.164
(0.305)
[-0.235]
0.756
0.106
1,128
Notes: The dependent variable is t if the individual is female and 0 if male. Standard errors
are in parentheses. All specifications include orchestra fixed effects and orchestra-specific
time trends. Changes in probabilities are in brackets; see lexi for an explanation of how they
are calculaied. New members are those who enter the orchestra for the first time. Returning
members are not considered new. The omitted section is strings.
Source: El even-orchestra roster sample. See text.
The Roster Data.—The roster data afford
us another way to evaluate the effect of the
screen on the sex composition of orchestras.
Using the rosters we know the sex of new hires
each year for 11 orchestras, and we also have
information {see Table I) on the year the screen
was adopted by each orchestra. We treat the
orchestra position as the unit of observation and
ask whether the screen affects the sex of the
individual who fills the position. We model the
likelihood that a female is hired in a particular
year as a function of whether the orchestra's
audition procedure involved a screen, again relying
on the variation over time within a particular
orchestra. Thus, in all specifications, we
include orchestra fixed effects and an orchestraspecific
time trend.
The roster data extend further back in time
than do the audition data and could conceivably
begin with the orchestra's founding, although
there is no obvious reason to include many
years when none used the screen. We report, in
Table 10, the effects ofthe screen on the hiring
of women from 1970 to 1996 using a probit
model. The screen is first defined to include any
blind auditions [column (1)]. In column (2) we
estimate separate effects for orchestras using
blind preliminary (and semifinal) rounds but not
blind finals and those with completely blind
auditions.
To interpret the probit coefficient, we first
predict a base probability, under the assumption
that each orchestra does not use a screen. We
then predict a new probability assuming the
orchestra uses a screen. The mean difference in
the probabilities is given in brackets.
THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
The coefficient on blind in column (I) is
positive, although not significant at any usual
level of confidence. The estimates in column (2)
are positive and equally large in magnitude to
those in column (1). Further, these estimates
show that the existence of any blind round
makes a difference and that a completely blind
process has a somewhat larger effect (albeit
with a large standard error),"^' According to the
point estimates in column (1) of Table 10, blind
auditions increase the likelihood a female will
be hired by 7.5 percentage points. The magnitude
of the effect must be judged relative to the
overall average and, for the period under consideration,
it was about 30 percent.^^'^ Thus blind
auditions increased the likelihood a female
would be hired by 25 percent.
Making Further Sense of the Results on Hiring.—
The audition sample results suggest that
blind auditions increase the probability of eventual
success for a female candidate by 5 percentage
points, but otily if there is no semifinal
round. The average effect for both types of
auditions is closer to 1 percentage point (with a
large standard error). The following example,
using assumed values based on the actual data,
demonstrates that an increase of about 2 percentage
points in the probability of a woman's
success out of an audition can explain the entire
change in female hires, allowing the share of
candidates who are female to increase from 0.2
to 0.3. Thus an increase of I percentage point—
our point estimate—can account for a substantial
share.
Consider two regimes: one without the screen
(not blind) and another with the screen (blind).
In the not-blind regime, assume that 20 percent
ofthe candidates are female and that in the blind
regime 30 percent are female.'^^ We know that
^' We have also attempted to interact the effect of Wind
auditions with section dumtnies. We find thai Ihe main
effect of blind auditions is almost identical to Ihai for the
string section, which is not surptising given thai the strings
comprise 65 percent of the observaiions. In addition, fewer
than 4 percent ofthe musicians hired into the percussion and
brass sections are female.
" See Table A2.
•''•' The fraction female in the not-blind regime (taking it
lobe the period before 1970) is 0.187 in our data (see Table
3). In the blind regime it was beiween 0.35 and 0.4. We
have chosen the more conservative 0.3 in the example
in the era (say, before 1970) when few orchestras
used the screen for the preliminary round
(see Table 1), 10 percent (that is, 0.0996) of
new hires were women. Also assume that 30
candidates enter each audition, independent of
audition regime, and that one musician is hired
out of each audition. Using these assumptions,
taken from the actual data, the success rate for
the typical female audition candidate in the notblind
regime will be 0.0166 and that for the
typical male will be 0.0375. If in the blind
regime, however, the percentage of new hires
who are female increases to 35 percent (its
approximate figure for the past 10 years), the
success rate for a female audition candidate
must have increased to 0.0389 (and that for a
male must have decreased to 0.0310). That is.
for consistency with the data on percent female,
the success rate for female candidates would
have had to increase by about 2.2 percentage
points, moving from the not-blind to the blind
regime. Our point estimate is that about half of
that increase—1 percentage point—was the result
of the effect of the screened audition process.
Using the example we just offered, the increase
in the probability of a woman's being
hired out of an audition accounts for 66 percent
of the total increase in the fraction female
among new hires. Half of the 66 percent comes
from the switch to blind auditions.-^"* The other
half could have resulted, for example, from a
because we want to use a number that is independent of the
switch to using the screen. That is, we would like to use a
fraction female that is solely Ihe result of increases in
female participation in general but independent of changes
in audition procedures,
''•' The proportion female among new hires is (/i • A • a).
where n = the number of audition candidates (in this
example n = 30); A = the success rate of the average
female candidate, which may be enhanced by the screen (in
this example A increases from 0.0166 to 0.0389 or by 2.2
percentage points, about half of which is due to the screen,
based on our estimates); and a = the fraction female among
candidates (a.ssumed here to increase from 20 to 30 percent
independent of A). The percentage of the total change accounted
for by ihe change in A is given by (n • a •
AA)/A(H • A • a) or on average by [(30 • 0.25 • 0.022)/
(0.35 - 0.0996) = 66 percent, (The 0.25 figure is the
average of that in the treatment period and that previously,)
Since half is accounted for by the screen, aboul 33 percent
of the increase in the proportion female among new hires
comes from the blind audition process.
VOL. 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIALTTY 737
greater acceptance of female musicians by music
directors. The remainder (34 percent) of the
increase in the fraction female among new hires
is accounted for by the increased percentage
female among audition candidates. That portion
comes primarily from the increase in the fraction
female among music school graduates.
The point estimates from the roster data also
suggest that a substantial portion ofthe increase
in female hires across the two regimes, notblind
and blind, can be explained by the change
in audition procedures. In the not-blind regime
about 10 percent of all hires are female but in
the blind regime about 35 percent are, a difference
of 25 percentage points. The estimates in
column (I) of Table 10 show that the switch to
the blind regime increases the likelihood a
woman will be hired by 7.5 percentage
points—30 percent of the total change—although
we emphasize that the coefficient is imprecisely
estimated.
One may wonder why there was disparate
treatment of female musicians before the screen
was used. A great orchestra is not simply a
collection of the finest musicians. It is, rather, a
group of great musicians who play magtiificently
as an ensemble. Substantial amounts of
specific human capital are acquired on the job
and tenure differences by sex, therefore, could
influence hiring decisions.^'' Leaves of absence
are ordinarily allowed for medical (including
maternity) and professional reasons. We find,
using the roster sample from I960 to 1996, that
the average female musician took 0.067 leaves
per year, whereas the average male musician
took 0.061, a difference that is not statistically
significant, and that their length of leave was
trivially different. Tenure differences were also
small and some specifications show that women
accumulated more years with an orchestra,
given their starting year and orchestra.^^ Tum-
^' Musicians of the Vienna Philharmonic made this argument
in a radio broadcast by the West German State
Radio in February 1996 [translation provided by William
OsborneJ. See also New York Times (1996) in which a
player for the Vienna Philharmonic argued that female
musicians would cost the orchestra considerably more because
substitutes would have to be hired if they became
pregnant.
*• The general sf)ecification is number of actual years with
an orcheslra as a function of the starting year, section dummies,
and a female dummy, for the period since 1959. The
over and leaves of absence do not appear to
differ by sex and thus should not have rationally
influenced hiring decisions.
IV. Conclusion
The audition procedures ofthe great U.S. symphony
orchestras began to change sometime in the
197O's. The changes included increasing the number
of candidates at auditions—a democratization
of the process—and using a physical screen during
the audition to conceal the candidate's identity
and ensure impartiaUty. We analyze what difference
blind auditions have meant for female
musicians.
We have collected, from orchestral management
files and archives, a sample of auditions for
eight major orchestras. These records contain the
names of all candidates and identify those advanced
to the next round, inciuding the ultimate
winner of the competition. The data provide a
unique means of testing whether discrimination
existed in the various rounds of a hiring pnxess
and even allow the linkage of individuals across
auditions. A strong presumption exists that discrimination
has limited the employment of female
musicians, especially by the great symphony orchestras.
Not only were their numbers extremely
low until the l970's, but many music directors,
ultimately in charge of hiring new musicians, publicly
disclosed their belief that female players had
lower musical talent.
The question is whether hard evidence can
support an impact of discrimination on hiring.
Our analysis of the audition and roster data
indicates that it can, although we mention various
caveats before we summarize the reasons.
Even though our sample size is large, we identify
the coefficients of interest from a much
smaller sample. Some of our coefficients of
interest, therefore, do not pass standard tests of
statistical significance and there is, in addition,
one persistent result that goes in the opposite
direction. The weight ofthe evidence, however,
is what we find most persuasive and what we
coefficient on the female dummy is -0.299 with a large
standard error (the mean of tenure is 11.7 years). With the
addition of orchestra fixed effects, the coefficient on the female
dummy is -fO,062. again with a large standard error. The
difference in tenure by sex, therefore, is extremely small.
738 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
have emphasized. The poitit estimates, moreover,
are almost all economically significant.
Using the audition data, we find that the
screen increases—by 50 percent—the probability
that a woman will be advanced from
certain preliminary rounds and increases by
severalfold the likelihood that a woman will
be selected in the final round. By the use of
the roster data, the switch to blind auditions
can explain 30 percent of the increase in the
proportion female among new hires and possibly
25 percent of the increase in the percentage
female in the orchestras from 1970 to
1996." ^ As in research in economics and other
fields on double-blind refereeing (see, e.g..
'^The point estimate for the increased likelihood a
woman would be a new hire, as a result of the adoption of
blind auditions, is 7,5 percentage points using the roster data
(see Table 10). Because the percentage female among new
hires increased from 10 to 35 percent from before 1970 to
the 199O's, our estimate implies that 30 percent of the 25
percentage-point increase can be explained by the adoption
Blank, 1991), the impact of a blind procedure
is toward impartiality and the costs to the
journal (here to the orchestra) are relatively
small. We conclude that the adoption of the
screen and blind auditions served to help female
musicians in their quest for orchestral
positions.
of the screen. How this increase affected the percentage
female in the orchestra depends on the sex composition of
the orchestra, retirement (or lumover), and the time frame.
We assume a 25-year time frame (from 1970 to 1995) and
two retirements (thus two hires) per year. An increase in the
percentage female among new hires from 10 percent (its
level pre-t970) lo 17.5 percent (10 + 1.5%) implies that in
25 years, 13.75 women (out of 100) will be in the orchestra,
or an increase of 3.75. The actual increase was 15 women,
meaning 25 percent of the increase can be explained by the
adoption ofthe screen. We assume in this example that the
age distribution of the 100 players in 1970 is uniform
between ages 25 and 74, that all hires occur at age 25. and
that men and women are drawn from the same age di.stribution.
APPENDIX
TABLE AI—SAMPLE DESCRIPTIVE STATISTICS, AUDITION DATA
Advanced
Blind
Female
Female X Blind
Missing female
Missing female X Blind
Years since lasl audition
Years since last audition.
missing
Automatic placement
Number of auditions
attended
"Big Five" orchestra
Total number of
auditioners
Proponion female at round
Principal
Substitute
Number of observations
(person-rounds)
Without
Mean
0.184
0.793
0,376
0,305
0,002
0,002
2,480
0,663
—
1.611
0.607
44.348
0.375
0.192
0.025
Preliminaries
semifinals
Standard
deviation
0.387
0.405
0,485
0,461
0,047
0.043
1.661
0,473
—
1.137
0.488
22.202
0,206
0.394
0.157
5,395
With semifinals
Mean
0.185
0.976
0.374
0,362
0.002
0.002
2.621
0.505
—
2.147
0.323
64,279
0.373
0.368
0,005
Standard
deviation
0.388
0.132
0.484
0.481
0.047
0.047
2.209
0.500
—
1.717
0,467
35.914
0.239
0,482
0,071
6.239
Semifinals
Mean
0,349
0.808
0,410
0.325
0.004
0.004
2,432
0.386
0.267
2.490
0.213
15.054
0.407
0.353
0.010
1,
Standard
deviation
0.477
0.394
0.492
0,469
0.066
0,061
2.393
0,487
0.443
1.886
0.409
7.187
0.211
0.478
0.101
360
Mean
0.200
0.122
0.411
0.056
0
0
2.272
0.505
0.137
2,051
0.391
8,622
0.411
0.278
0.021
1
Finals
Standard
deviation
0.400
0.328
0.492
0.230
0
0
1.895
0.500
0,345
1,513
0,488
4.445
0,213
0.448
0.141
,127
Source: Eighl-orchestra audition sample. See text.
VOL. 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIALITY 739
TABLE A2—SAMPLE DESCRIPTIVE STATISTICS, ROSTER DATA: 1970 TO 1996
Proponion female among new hires
(Proportion female), _,
Only blind preliminary auditions
All auditions blind
Section:
Strings
Woodwinds
Brabs
Percussion
Number of observations
Mean
0.293
0.179
0.572
0.104
0.642
0.158
0.165
0.035
Standard deviation
0.455
0.081
0.495
0.305
0.480
0.365
0.371
0.185
1,128
Note: Means are musician weighted, nol audition weighted.
Source: Eleven-orchestra roster sample. Sec text.
TABLE A3—LINEAR PROBABILITY ESTIMATES OF THE LIKELIHOOD OF BEING ADVANCED: BY ROUND
Female
Female X Blind
Blind audition
/)-value of //„:
Female + (Female X
Blind) = 0
Other covariates?
Instrument fixed
effects?
Year fixed effects?
Orchestra fixed effects?
R'
Number of observations
(person-rounds)
Preliminaries
Without
semifinals
(1)
0.007
(0.025)
-0.062
(0.028)
0.015
(0.022)
0.000
Yes
Yes
Yes
No
0.062
5,395
(2)
0.011
(0.025)
-0.067
(0.028)
0.040
(0.030)
0.000
Yes
Yes
Yes
Yes
0.070
5,395
With
semifinals
(3)
-0.054
(0.069)
0.005
(0.070)
0.024
(0.057)
0.000
Yes
Yes
Yes
No
0.033
6.239
(4)
-0.085
(0.069)
0.037
(0.070)
0.027
(0.062)
0.000
Yes
Yes
Yes
Yes
0.045
6,239
Semifinals
(5)
0.103
(0.061)
-0.142
(0.066)
0.053
(0.049)
0.210
Yes
Yes
Yes
No
0.074
1.360
(6)
0.099
(0.061)
-0.137
(0.067)
0.115
(0.078)
0.222
Yes
Yes
Yes
Yes
0.081
1,360
Finals
(7)
0.002
(0.028)
-0.091
(0.075)
0.058
(0.058)
0.207
Yes
Yes
Yes
No
0.064
1,127
(8)
0.0004
(0.028)
-0.078
(0.075)
0.123
(0.089)
0.271
Yes
Yes
Yes
Yes
0.068
1.127
Notes: The dependent variable is I if the individual is advanced to the next round and 0 if not. Standard errors are in
parentheses. All specifications include dummies indicaling wheiher the sex is missing, and an interaction for the sex
being missing and a blind audition. "Other covariates" include automatic round, number of auditions attended, whether
a "Big Five" orchestra, size of round, proportion female at the round, and whether a principal (including assistant and
associate principal) or substitute position; except in columns (2), (4), (6), and (8) for which "Other covariates" include
only automatic placement and numher of auditions attended. These results are comparable to those in Table 6 but
without individual hxed effects.
Source: Eight-orchestra audition sample. See text.
740 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000
TABLE A4—LINEAR PROBABILITY ESTIMATES OE THE LIKELIHOOD OF BEING ADVANCED: ASSESSING SHORT-PANEL BIAS
Blind
Female X Blind
/j-value of HQ.
Blind + (Female X Blind) = 0
Other covariates?
individual fixed effects?
Year fixed effects?
Number of observations (person-rounds)
Blind
Female X Blind
p-value of H^:
Blind + (Female X Blind) = 0
Other covariates?
Individual fixed effects?
Year fixed effects?
R'
Number of observations (person-rounds)
Part A;
Without semifinals
I"
-0.024
{0,066)
0.126
(0.095)
0,233
Yes
Yes
Yes
0.491
1.025
II"
-0,042
(0.063)
0,095
(0.071)
0.502
Yes
Yes
Yes
0,537
639
Preliminary rounds
Preliminaries
With semifinals
I^
-0,047
(0,383)
-0,035
(0.403)
0.807
Yes
Yes
Yes
0.423
1,928
II"
-0.095
(0.744)
0.041
(0.275)
0.943
Yes
Yes
Yes
0,732
55
Part B; Semifinals and finals, and wilh orchestra fixed effects
Semifinals
I-' II"
0.060 0.169
(0.133) (0.109)
-0.179 -0.284
(0.195) (0.142)
0.438 0.298
Yes Yes
Yes Yes
Yes Yes
0.438 0.593
269 223
P
0.123
(0.356)
0.157
(0,408)
0,212
Yes
Yes
Yes
0.721
127
With orchestras
Finals fixed
II" I'
-0.140 0.084
(0,449) (0,047)
0.403 0.042
(0,415) (0.051)
0,587 0,011
Yes Yes
Yes Yes
Yes Yes
0.728 0,506
67 2,321
effects
0.352
(0.056)
0.021
(0,041)
0,000
Yes
Yes
Yes
0,603
1,776
Notes: The dependent variable is I if the individual is advanced lo the next round and 0 if not. Standard errors are in
parentheses. These are the same specifications as in columns (2), (4), (6), and (8) of Table 6 and column (2) of Table 7.
Source: Eight-orchestra audition sample. See text,
" Includes those who auditioned at least three times (for the relevant round).
" Includes those who auditioned at least once in a blind audition and at least once in a not-blind audition (for the relevant
round).
REFERENCES
Allmendinger, Jutta; Hackman, J. Richard and
Lehman, Erin V. "Life and Work in Symphony
Orchestras." Music Quarterly, Summer
1996, 50(2), pp. 194-219.
Arrow, Kenneth. "The Theory of Discrimination,"
in Orley Ashenfelter and Albert Rees, eds..
Discrimination in labor markets. Princeton, NJ:
Princeton University Press, 1973, pp. 3-33.
Ayres, Ian and Waldfogel, Joel. "A Market Test
for Race Discrimination in Bail Setting."
Stanford Law Review, May 1994, 46(5), pp.
987-1047.
Becker, Gary. 77ze economics of discrimination,
2nd Ed. Chicago: University of Chicago
Press, 1971 [orig. pub. 1957],
Blank, Rebecca. "The Effects of Double-Blind
versus Single-Blind Refereeing: Experimental
Evidence from the American Economic
Review." American Economic
Review, December 1991, 81(5), pp. 1041-
67.
Economist, The. "American Orchestras: All
Ears." November 30, 1996.
Ellefson, Connie Lockhard. The melting pot book
of baby names, 2nd Ed. Cincinnati, OH: Better
Way Books, 1990.
VOL 90 NO. 4 GOLDIN AND ROUSE: ORCHESTRATING IMPARTIAUTY 741
England, Paula. "The Failure of Human Capital
Theory to Explain Occupational Sex Segregation.""
Journal of Human Resources. Summer
1982. 17(3), pp. 358-70.
England, Paula; Kilbourne, Barbara Stanek; Farkas,
George and Dou, Thomas. "Explaining
Occupational Sex Segregation and Wages:
Findings from a Model with Fixed Effects."
American Sociological Review, August 1988,
53(4), pp. 544-58.
Filer, Randall K. "Occupational Segregation,
Compensating Differentials, and Comparable
Worth," in Robert T. Michael, Heidi I.
Hartmann, and Brigit O'Farrell, eds.. Pay
equity: Empirical inquiries. Washington,
DC: National Academy Press, 1989, pp.
153-70.
Freeman, Richard. "Longitudinal Analyses of
the Effects of Trade Unions." Journal of
Labor Economics, January 1984, 2(1), pp.
1-26.
Holzer, Harry and Neumark, David. "Are Affirmative
Action Hires Less Qualified? Evidence
from Employer-Employee Data on
New Hires." National Bureau of Economic
Research (Cambridge, MA) Working Paper
No. 5603, June 1996.
Hsiao, Cheng. Analysis of panel data. New
York: Cambridge University Press, 1986.
Kenney, Genevieve and Wissoker, Douglas A.
"An Analysis of the Correlates of Discrimination
Facing Young Hispanic Job-Seekers."
American Economic Review, June 1994. 84(3),
pp, 674-83.
Kolatch, Alfred J. The Jonathan David dictionary
of first names. New York: Perigee
Books, 1990.
Lansky, Bruce. 35,000+ baby names. New
York: Meadowbrook Press, 1995.
Neumark, David (with Bank, Roy and Van Nort,
Kyle D.). "Sex Discrimination in Restaurant
Hiring: An Audit Study." Quarterly Journal of
Economics, August 1996, 7/7(3), pp. 915-41.
New York Times. "Vienna Boys' Choir Hands the
Baton to a Woman." October 16, 1996.
Polachek, Solomon W. "Occupational Segregation
among Women: Theory, Evidence and a
Prognosis," in Cynthia B. Lloyd, Emily Andrews,
and Curtis Gilroy, eds.. Women in the
labor market. New York: Columbia University
Press, 1979, pp. 137-57.
Seltzer, George. Music matters: The performer
and the American Federation of Musicians,
Metuchen, NJ: Scarecrow Press, 1989.