Sociology memo

I’d Like to Thank the Academy, Team Spillovers, and Network Centrality

Gabriel Rossman,a Nicole Esparza,b and Phillip Bonacicha

Abstract

This article uses Academy Award nominations for acting to explore how artistic achievement is situated within a collaborative context. Assessment of individual effort is particularly difficult in film because quality is not transparent, but the project-based nature of the field allows us to observe individuals in multiple collaborative contexts. We address these issues with analyses of the top-10 credited roles from films released in theaters between 1936 and 2005. Controlling for an actor’s personal history and the basic traits of a film, we explore two predictions. First, we find that status, as measured by asymmetric centrality in the net- work of screen credits, is an efficient measure of star power and mediates the relationship between experience and formal artistic consecration. Second, we find that actors are most likely to be consecrated when working with elite collaborators. We conclude by arguing that selection into privileged work teams provides cumulative advantage.

Keywords

social networks, team spillovers, culture, film

Cultural fields are characterized by a hierar-

chy of quasi-moral value, in which a set of

mechanisms consecrate certain individual

works, authors, genres, and even entire artis-

tic fields as worthy (Baumann 2001, 2007;

Becker 1978; Bourdieu 1993; DiMaggio

1981; English 2005; Motti 1994). Awards,

prizes, and honors for the ‘‘best artwork’’

or the ‘‘best artist’’ are among these conse-

crating institutions (Anand and Watson

2004; Cowen 2000; English 2005). A person

(or work closely associated with a person) is

consecrated by an award and that person gets

her name engraved on a trophy, gives an

acceptance speech, and thereafter finds her

career transformed (Lincoln 2007). Nobel-

winning scientists find that they are subse-

quently given better access to resources, their

work is evaluated more favorably, and they

are considered the real force in any collabo-

ration (Merton 1968; Zuckerman 1996).

Likewise, journalists treat ‘‘medals and tro-

phies as a legitimate measure—perhaps the

only legitimate measure—of a person’s cul-

tural worth’’ (English 2005:22).

This emphasis on individual winners af-

firms the romantic ideology of art as heartfelt

expression of the artist rather than the

aUniversity of California–Los Angeles bUniversity of Southern California

Corresponding Author: Gabriel Rossman, Department of Sociology, 264

Haines Hall, UCLA, Los Angeles, CA 90095-1551,

USA

E-mail: rossman@soc.ucla.edu

American Sociological Review 75(1) 31–51 � American Sociological Association 2010 DOI: 10.1177/0003122409359164 http://asr.sagepub.com

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

collective achievement of the artistic team or

broader art world. While especially pro-

nounced in lifetime achievement awards, all

awards following the best-of-a-profession

formula imply ‘‘overemphasis on individual

achievement and individual authorship’’

(English 2005:86). This mismatch between

the essentially collaborative nature of most

arts and the essentially individualistic nature

of most awards provides us analytic leverage

to see how individual achievements are as-

sessed when critical observers have access

only to the collaborative efforts within which

these achievements are embedded. We do not

see individual talent face to face, but through

the glass darkly of social context and team

effort. In this article, we attempt to explain

individual artistic consecration using two as-

pects of the context of cultural production:

social status and team spillovers.

THE ACADEMY AWARDS

The Academy Awards, or ‘‘Oscars,’’ are

among the most prominent awards in the

entertainment field. Not only are they com-

parable in prominence and media attention

(if not dignity) to the Nobel Prizes, but they

are the most prominent cultural prize by

a wide margin (English 2005; Levy 2003).

The various traditions and fetishes intro-

duced by the Oscars—public nominations,

sealed envelopes, and spectacular televised

ceremonies with the nominees in atten-

dance—have been so widely imitated by other

entertainment awards (and even high culture

awards, like the Man Booker Prize) that the

Oscars can fairly be called the prototype for

the institutionalization of the entertainment

award as spectacle (English 2005). The

Oscars have concrete implications for the

box office; films that win best picture earn

an estimated extra $12.7 million (Nelson et

al. 2001). Likewise, winning an Oscar can

dramatically increase the fees a previously

obscure Hollywood worker can demand

(Gumbel et al. 1998). ‘‘The very nomination

for an Oscar can have a pervasive impact on

artists’ careers, expanding their visibility and

exposure’’ (Levy 2003:90). The ceremony it-

self has been a lucrative undertaking since

NBC (and later ABC) began paying for the

broadcast rights in 1952, and this source of

revenue is a major motivating force behind

the proliferation of awards and awards broad-

casts since 1970 (English 2005). The Oscars

are now the basis for an entire ‘‘awards sea-

son’’ industry of entertainment, gossip, and

fashion journalism, as well as secondary

awards whose primary interest is the extent

to which they project the real winners—a

sort of cinema Advent to the Hollywood

Christmas that is Oscars night.

Generally, awards make claims concerning

a field’s legitimacy and boundaries (Anand

and Watson 2004), but there are other means

to institutionally consecrate a field. The orga-

nization of cultural production is an important

factor in the institutionalization of prestige, as

elevated fields tend to be only loosely coupled

to the market and proximately focused on aes-

thetic interests of the artistic community itself

(Becker 1978; Bourdieu 1993; DiMaggio

1981). Likewise, studies show that even

within such consummately commercial mass

entertainment fields as film and rock music,

critics create artistic value by drawing distinc-

tions between more and less artistic works and

workers (Baumann 2001, 2007; Motti 1994).

While the sociological literature on cultural

consecration is dominated by such mecha-

nisms as nonprofit organization and critics,

awards represent a clear and understudied

mechanism by which fields and works are

consecrated, often through the collective

action of key figures within the field itself

(Anand and Watson 2004; English 2005).

The Academy of Motion Picture Arts and

Sciences (AMPAS) was founded in 1927 as

a professional honorary organization of 36

members (it has since expanded to approxi-

mately 6,000 motion picture professionals).

Although AMPAS bestowed awards from

its inception, this was only one of its several

functions during its first decade. The early

film industry had a low reputation, stemming

32 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

from anti-immigrant sentiment about pro-

ducers and customers, and this bad reputation

was only exacerbated by a series of sex scan-

dals (Baumann 2007; Levy 2003). By the

1920s, the industry faced censorship and

other regulatory threats. To reassert the in-

dustry’s legitimacy, Louis B. Mayer,

cofounder of Metro-Goldwyn-Mayer, led

other moguls in creating two institutions:

the Hays Office (a semi-voluntary censorship

board and the precursor to the Motion Picture

Association of America) and AMPAS (a

combination cultural organization, trade

group, and labor arbitration system led by

Hollywood’s business and creative elite).

By the mid-1930s, independent unions like

the Screen Actors Guild had supplanted

AMPAS’s labor functions. Frank Capra

saved the Academy from obsolescence by re-

organizing it in 1939 to focus exclusively on

cultural and artistic concerns to the exclusion

of political and economic issues. (The trade

group functions abandoned by AMPAS

would later be adopted by the Motion

Picture Association of America, which now

aggressively pursues Hollywood’s political

interests on issues such as copyright.)

Academy members of each relevant

branch nominate individuals for the Oscars,

such that writers nominate writers, actors

nominate actors, and so on among eligible

films (Academy of Motion Picture Arts and

Sciences 2009; Levy 2003). An independent

accounting firm tallies completed nomin-

ation ballots and selects the top-five selec-

tions per award as the official nominees.

A second and final round of balloting is

given to the entire Academy. All Academy

members, regardless of branch, decide the

winners. Compared with Hollywood in gen-

eral, the Academy is very small. As of 2002,

the acting branch had 1,315 members,

approximately 2 percent of the membership

of the Screen Actors Guild (Levy 2003).

Because the Academy membership is re-

cruited from prior nominees and other artists

sponsored by the members, AMPAS is a true

academy, in the sense of an elite body where

the incumbent membership recruits new

members.

The Oscars are the most concrete and

deliberate form of the more general phenome-

non of stardom in cultural fields. As mass

communication technology allows the infinite

reproduction of cultural works, there is no

upper limit to the popularity of cultural works

and workers. This creates a massive level of

inequality known as the ‘‘superstar effect’’

(Rosen 1981). Although the original formula-

tion explains stardom as an exponential func-

tion of talent, later empirical work suggests

that celebrity is as much stochasticity as mer-

itocracy (Hamlen 1991). Current superstar

theories focus on cumulative advantage mech-

anisms, such as network externalities (Adler

1985) and information cascades (Salganik,

Dodds, and Watts 2006), whereby random ad-

vantages become self-perpetuating. Because

awards like the Oscars have beneficial career

effects (Lincoln 2007), they are almost cer-

tainly a cumulative advantage mechanism

for the creation of stardom in Hollywood.

By implication, awards generate inequality

as they separate stars from lesser actors.

In this article, we propose that the collabo-

rative context of artistic labors provides an

additional mechanism for explaining individ-

ual success, as measured by consecration

with an Oscar nomination. The project-based

nature of film implies the proximate influence

of teams and the formation of larger networks

(Faulkner and Anderson 1987; Watts and

Strogatz 1998; Zuckerman 2004). One’s past

transactions with peers may reveal and shape

the general perception of one’s place in the

pecking order. That is, does outranking

Robert DeNiro in the credits of a past film

imply that you are the sort of person whom

all voters will take seriously? Second, films

are created through the input of many artists,

including directors, screenwriters, and costars,

which allows us to test whether the efforts of

top workers spill over onto their team mem-

bers. In other words, does having Robert

DeNiro as a costar make one more likely to

be nominated for an Oscar?

Rossman et al. 33

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

REPUTATION AND STATUS

Status is important in many fields because it

provides a ready rubric for actors to quickly

assess potential trading partners. 1

Economists

use the concept of ‘‘reputation’’ to describe

a track record of honorable behavior and

high-quality output that trading partners

use to estimate the possibility of fraud or

incompetence (Axelrod 1980; Wilson 1985).

Sociologists use the term ‘‘status’’ to denote

a similar concept; trading partners use status

to evaluate quality, but the origins of status

are embedded in an actor’s set of associations.

Status may be related to more direct measures

of quality, but it also contains an irreducible

element of charisma. Under certain conditions,

underlying talent will set the seed for how

others conceive of an actor, but, over time,

measurement error and self-perpetuating

dynamics of association and deference may

partially decouple status from talent (Gould

2002; Salganik et al. 2006). Status can thus

be an important intervening variable between

talent and reward (Podolny 2001). In

Hollywood, actors routinely use rubrics of rep-

utation in their assessments of the quality of

trading partners (Bielby and Bielby 1994,

1999; Faulkner and Anderson 1987).

Likewise, we expect status to color individu-

als’ evaluations of the quality of performances.

Status is often measured by constructing

a directed graph of deferential interactions,

although it is ambiguous whether such inter-

actions directly create status or merely reflect

it. While status is often measured as a social

network variable, its explanatory power for

outcomes does not follow the logic of social

capital flowing directly through a network.

Podolny (2001) uses the contrasting meta-

phors of networks as pipes, through which

social capital flows, and prisms, through

which status is refracted. Power is not cre-

ated through a continuing relationship

between individuals, but from the memory

of an interaction in others’ conceptions and

from the underlying power dynamics that

led one to defer to another. At a minimum,

directed graph centrality is an indicator of

status, itself a latent variable, and at most it

actively creates status as an example for

others. Whatever it may be, however, it is

not a set of persistent relations through which

resources and information flow.

In Hollywood, status is displayed by

position in the billing block in the same

way that investment banks show status

through placement in ‘‘tombstone’’ debt of-

ferings (Podolny and Phillips 1996). Of

course, billing rank primarily represents

prominence within a film, which itself re-

flects a casting director’s assessment of

star power. Actors’ billing also includes an

element of pure power and reputation, as

when Judi Dench had only eight minutes

of screen time in Shakespeare in Love

(1998) but nonetheless was one of only

five names on the film’s poster and won

an Oscar for her performance as Queen

Elizabeth I. For all film professions except

writers (whose credits are controlled by

their union), credits are negotiated between

the professional and the studio and include

not just the order of listing but such details

as typeface, opening versus end credits,

whether the credit appears on the screen

by itself or as a ‘‘shared card credit,’’ and

even whether an actor’s picture must appear

on all advertising (Resnik and Trost 1996).

Actors do not judge this deference in abso-

lute terms, but according to the ‘‘most

favored nation’’ principle that no other pro-

fessional should receive greater honor. For

instance, a powerful Hollywood profes-

sional may not be offended at a poster

with just the name and release date of

a film, but he would take offense if the post-

er included the name of another professional

without giving him equal prominence. In

effect, getting higher credit listing than

another demonstrates one’s higher status.

Because rank order in credits represents

a casting director’s estimate of an actor’s

star power and the bargaining power the

actor is able to exert in negotiating rank, it

should be a good measure of status (or star

34 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

power) and this should be an important pre-

dictor of Oscar nominations. This leads to

our first hypothesis:

Hypothesis 1: High-status actors are more likely

to be consecrated with an Oscar nomination.

TEAM SPILLOVERS

Stinchcombe (1963:806) distinguishes

between industries in which ‘‘individual tal-

ent is clearly a complementary factor of pro-

duction [whereas] in others it is more nearly

additive.’’ 2

He hypothesizes that wage

inequality will be greater in the former than

in the latter industries. For example, he notes

that if Alec Guinness is three times as tal-

ented as the typical actor, this will have

a much greater effect on a film’s quality

than will the efforts of a comparably prodi-

gious house painter on a home’s appearance.

This implies that Guinness’s compensation

will be much higher above the median for ac-

tors than will a talented house painter’s com-

pensation be above the median for painters.

As Stinchcombe predicts, the variance in sal-

aries for academics at research institutions is

greater than that at teaching institutions

(Abrahamson 1973). Likewise, industries

characterized by complementary productivity

exhibit greater productivity when using

incentive structures sensitive to this charac-

teristic (Petersen 1992). Ironically, just as

economics has developed an active literature

on parallel models, sociology has forgotten

Stinchcombe’s model.

In economics, a literature growing out of

the theory of the firm addresses the comple-

mentarity of production inputs such as phys-

ical capital, human capital, and raw

(unskilled) labor (Griliches 1969). Alchian

and Demsetz (1972) were the first to apply

the problem to teams of labor. They used

the example of two men lifting an object

too heavy for either to lift alone to conjecture

that the productivity of a team may be greater

than the sum of its parts. At the same time,

labor economists (echoing earlier arguments

from sociology [Stinchcombe and Harris

1969]) began to argue that professional man-

agers may have multiplicative productivity

with other inputs (Rosen 1982). Since then,

spillover models influenced by this articula-

tion have been used in development econom-

ics (e.g., Kremer 1993), labor economics

(e.g., Battu, Belfield, and Sloane 2004),

sports economics (e.g., Idson and Kahane

2000), and cultural economics (e.g., Caves

2000).

For our purposes, the key implication is

that ‘‘when complementarity exists between

labor inputs, individual productivity may be

poorly measured by treating the individual

worker separately from the character of the

organization, or team, within which he

works’’ (Idson and Kahane 2000:345). In

practice, this should mean that workers will

do their best work when they are in the com-

pany of skilled peers. 3

Several studies exploit

this property to use compensation as a proxy

for productivity in measuring peer effects

(e.g., Battu et al. 2004; Cardoso 2000;

Idson and Kahane 2000). The best studies

tend to be of professional sports, where

fine-grained measures of compensation and

productivity are available and there are

good a priori reasons to expect the relevance

or irrelevance of peer spillovers. For

instance, teammate skill has a strong effect

on points per shot in basketball (Kendall

2003). This can be safely interpreted as

a spillover because the effect is especially

strong for position dyads with frequent inter-

action and is absent in basketball free throws

and baseball batting, two situations where

teammates can provide no tangible assis-

tance. Likewise, good teammates help Tour

de France bicyclists reach better ranks

(Torgler 2007) and German soccer players

score more goals (Torgler and Schmidt

2007).

With regard to films, the theoretical litera-

ture occasionally speculates that complemen-

tarity will be important, but the idea has yet

to face an empirical test. Was Stinchcombe

(1963) right that Alec Guinness dramatically

Rossman et al. 35

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

improves any film in which he appears? Can

we go further and say that, for instance, Star

Wars (1977) was the highlight of Mark

Hamill’s career because he shared most of

his scenes with the old pro? Consider the actor

Robert Forster, who has had a long but mostly

obscure career as a character actor appearing

primarily on television and in extremely low

budget horror and crime films. Yet in 1998

he was nominated for the best supporting actor

Oscar for the role of bail bondsman Max

Cherry in Jackie Brown (1997). The film

was written and directed by Quentin

Tarantino who had previously been nominated

for best director and best original screenplay

(winning the latter) for Pulp Fiction (1994).

Forster’s costars included prior nominees

Samuel L. Jackson and Robert DeNiro.

Forster’s career immediately regressed to the

mean after Jackie Brown, demonstrating how

much his nomination for that film benefited

from Tarantino, Jackson, and DeNiro.

Extreme cases like this, where a relatively

unknown character actor gets nominated after

working with an elite team, are rare.

However, even actors who are themselves

major stars may benefit from working with

strong teams. For instance, Leonardo

DiCaprio is an A-list actor by any reckoning,

but his first nomination came from collabora-

tion with an Academy-nominated director

(Lasse Hallström) and his second from collab-

oration with Hollywood’s top director (Martin

Scorsese) and an Academy-nominated writer

(John Logan). This leads to our second

hypothesis:

Hypothesis 2: Actors who collaborate with

elite peers are more likely to have their

own contributions consecrated with an

Oscar nomination.

DATA

We used data from the Academy of Motion

Picture Arts and Science (AMPAS) and the

Internet Movie Database (IMDB) to con-

struct a dataset of the top-10 credited actors

in Academy Award eligible films from

1936 to 2005. This dataset describes

147,908 performances by 37,183 actors in

16,392 films. The Academy deems films eli-

gible for an Oscar if they are more than 40

minutes long, meet minimal technical stan-

dards for quality, and were advertised and

screened for paid admission for at least one

week in a Los Angeles County commercial

motion picture theater. We obtained nomina-

tion ballots listing the eligible films for all

years from the AMPAS Special Collection

Archives located at the Margaret Herrick

Library, Fairbanks Center for Motion

Picture Study in Los Angeles. Films appear-

ing in the IMDB but not on AMPAS ballots

were disproportionately made in recent dec-

ades, a period that saw a typical volume of

theatrically released films but exponential

growth in more obscure films (most of which

are only screened at festivals, if at all). We

leave out these noneligible films to reflect

the Academy nomination process, but the

findings are robust to this decision.

Furthermore, we include only the top-10

credited roles in each film. Although it is not

unheard of for actors ranked lower than 10 in

the credits to be nominated, it is very rare.

Limiting the dataset to the top-10 credited

roles holds constant a reasonable level of

prominence and mitigates possible confound-

ing spurious effects of variables correlated

with prominence in a film. Truncating the

data in this way also greatly reduces the

sparseness of the outcome while still including

99.2 percent of the nominees for the lead cate-

gories and 95.5 percent of the nominees for the

supporting categories. We experimented with

including lower ranked performances and the

results are similar.

Although the first Oscars were given to

films released in 1927, we begin the analysis

with films released in 1936 because this is

when the acting awards reached their current

form. In its first two years, the Academy

lacked ‘‘official’’ nominations and only had

lists of actors ‘‘discussed’’ for the award

(Academy of Motion Picture Arts and

36 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

Sciences 2009). Furthermore, from 1927 to

1935 the Academy had only two categories

for acting, ‘‘Best Actor’’ and ‘‘Best

Actress.’’ Additionally, in five cases during

this period, actors’ nominations were not tied

to a particular film but to several films they

performed in that year. To bracket these early

irregularities, we begin our analysis in 1936

with the maturation of the nomination system

for the acting awards. However, including ear-

lier years in the dataset gives essentially the

same results. Table 1 lists the variables used

in the analysis. All variables are time-varying.

Academy nomination. The dependent

variable is the log-odds of an actor being

nominated for an Academy Award. We count

all AMPAS nominations from 1936 to 2005

in English language, feature-length films for

the categories ‘‘Best Actor in a Leading

Role,’’ ‘‘Best Actress in a Leading Role,’’

‘‘Best Actor in a Supporting Role,’’ and

‘‘Best Actress in a Supporting Role.’’ Each

nomination is attached to one film perfor-

mance. Although no one has ever been nom-

inated in the same category for two different

films in a given year, it is theoretically possi-

ble, and there have been years in which an

actor was nominated in the lead category

for one film and the supporting category for

another, as with Jamie Foxx in Ray (2004)

and Collateral (2004). Many films produce

multiple nominations, for example, Geena

Davis and Susan Sarandon were both nomi-

nated for lead actress for their performances

in Thelma and Louise (1991).

Human capital. At the most basic level,

actors should be nominated for Oscars based

on their acting ability or skill. We therefore

incorporate two human capital measures into

the model. First, we include a variable for

past acting nominations received to date.

Second, we include a past acting measure for

the number of films an actor has been in to

date. This variable is important in the literature

because experienced actors are offered more

jobs (Zuckerman et al. 2003). Likewise, if ac-

tresses do not impress the Academy within

their first five movies, they have a slim chance

of ever being nominated in the future (Levy

2003). We use a linear spline with knots at

the 5th and 20th films to allow for actors

(and especially actresses) peaking at some

point and then declining. We determined these

cutpoints by looking at the odds of nomination

at each number of past films and noting the

points at which the slope changes. One can

accrue skill and experience even from obscure

performances, so we use all IMDB film credits

to measure experience.

Status (credit centrality). Our first

hypothesis is about status, as measured by

centrality in a graph of deferential interac-

tions (Benjamin and Podolny 1999; Podolny

1993). Conceptually, status for Hollywood

actors should be equivalent to ‘‘star power.’’

Centrality efficiently summarizes an actor’s

status because it shows not just how many

peers have deferred to the actor, but how

high status these deferents are themselves.

This makes status recursively transitive.

Alternative measures of network power,

such as betweenness, are inappropriate for

measuring status because deference is more

important than brokerage or information

flow. 4

Our measure of an actor’s status is

alpha-centrality, which is similar to eigen-

vector centrality but is more appropriate for

directed graphs (Bonacich and Lloyd 2001).

We calculate the metric on a directed credit

network where one is defined as choosing an

actor when that actor outranks the other in

the credits of a film. 5

In other words, when

you accept a position in the credits of

a film, you are choosing everyone listed above

you as your superior, and everyone listed

below you chooses you as their superior.

The network consists of all such ties within

a rolling five-year window. We drop ties older

than five years from the network, but as long

as an actor continues to make films she re-

mains in the network, with her position

defined by her recent collaborations. 6

Centrality is distinct from an actor’s aver-

age rank in the credits because the key fact is

not how many others but who the actor

Rossman et al. 37

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

Table 1. Variables Used in the Analysis

Variable Description Mean SD Range

Oscar Nomination Performance was

nominated for an Oscar

.009 .094 [0, 1]

Film Id number of film in

which the performance

occurs, level-two cluster

variable

n/a n/a [1, 16,392]

Baseline Variables

Year Year of release for the

film

1967 21 [1936, 2005]

Films per Year Number of Oscar eligible

films released in the year

(in 100s)

2.633 .934 [1.21, 4.3]

Genre: Drama IMDB genre codes include

‘‘drama’’

.508 .500 [0, 1]

Genre: Comedy IMDB genre codes include

‘‘comedy’’

.318 .466 [0, 1]

Genre: Biography IMDB genre codes include

‘‘biography’’

.027 .164 [0, 1]

Major Distributor Film’s first-run U.S.

theatrical release was

by a major studio

.690 .463 [0, 1]

Cast Size Number of credited actors

in the film

24.621 18.026 [1, 382]

Release Date Date (in year) of film’s

general or Los Angeles

premiere

191.261 103.488 [1, 366]

Female Performance was

by an actress

.301 .459 [0, 1]

Human Capital

Past Films 0 to 5 Linear spline 0 to 5, number

of films the actor was in

prior to year

3.647 1.960 [0, 5]

Past Films 6 to 20 Linear spline 6 to 20, number

of films the actor was in

prior to year

6.374 6.601 [0, 15]

Past Films .20 Linear spline 21 and up,

number of films the actor

was in prior to year

8.443 21.058 [0, 248]

Past Nomination Actor is a prior Oscar

nominee

.084 .277 [0, 1]

Status (Credit Centrality) Actor’s asymmetrical

centrality in the network

of credits over the prior

five-year period

81.743 14.246 [0, 100]

Spillovers

Costars with Past

Nominations

Costars have prior Oscar

nominations for acting

.703 .993 [0, 8]

Director with Past

Nominations

Director has prior Oscar

nomination for directing

.117 .322 [0, 1]

Writers with Past

Nominations

Writers have prior Oscar

nominations for writing

.231 .514 [0, 4]

38 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

outranks. For instance, in the film The

Godfather, Part II (1974), the first three

names in the credits are, in order, Al

Pacino, Robert Duvall, and Diane Keaton.

We code this as Keaton choosing Duvall

and Pacino, Duvall choosing Pacino, and

Pacino himself choosing no one. Bruno

Kirby is further down the credits, so he

chooses all three of the leads. Being chosen

by Duvall does far more for Pacino’s centrality

than does his being chosen by Kirby, because

in the scope of the entire IMDB network,

Duvall is frequently chosen by other actors,

whereas Kirby is only seldomly chosen.

We validated this metric by comparing the

centrality of all 328 film actors who appeared

on the cover of Entertainment Weekly from

1997 to 2005 with the general population of

film actors active in this period. These prom-

inent actors had much higher centrality than

average, with their group mean at 1.7 stan-

dard deviations above the general mean.

The only major difference between the

Entertainment Weekly and centrality meas-

ures of stardom is that centrality favors

men, whereas women were equally likely to

appear on the magazine cover. However, all

of our models control for gender, so this is

not a concern (for further exploration of

potential gender differences in the effects of

centrality, see the Online Supplement at

http://asr.sagepub.com/supplemental).

Peer spillovers. Although the first-level

unit of analysis is the actor, the second level

is the film, which ties actors, writers, and di-

rectors to a collaboration. We measure each

film as a unique collaboration and attach sev-

eral traits to each of its participants. Our sec-

ond hypothesis centers on credentials of the

cast and crew involved in the collaboration.

We test this hypothesis through specifying

the number of costars, writers, and directors

with prior nominations. We use past nomina-

tions only to avoid endogeneity. Although

many actors also write and direct films, we

include past award history only for the occu-

pation the individual performs in the current

collaboration. For instance, Ben Affleck has

never been nominated for an acting Oscar,

but he won an Oscar in 1998 for original

screenplay. We thus treat Affleck as a top

writer but not as a top actor. Actors appear-

ing in films written by Affleck after 1998

may receive a spillover for the vicarious

human capital of using Affleck’s writing. 7

However, an actor who costars in a film

with Affleck (not written by him) will not

receive that spillover.

Of the 37,183 actors in the dataset, 723

wrote or directed at least one of the Oscar eli-

gible films in which they also starred. This

raises the question of whether an actor who

is also a prior directing or writing nominee

benefits from a ‘‘team spillover’’ by acting

under his own direction or using his own

screenplay, much as self-collaboration

between writers, directors, and producers de-

fines the organization of Hollywood (Baker

and Faulkner 1991). In fact, self-collabora-

tion proves to be largely irrelevant in prac-

tice; only three actors have ever been

nominated for acting in films that they also

wrote or directed after having previously

been nominated for writing or directing:

Laurence Olivier for Richard III (1955),

Warren Beatty for Heaven Can Wait (1978)

and Reds (1981), and Clint Eastwood for

Million Dollar Baby (2004). The analysis is

completely robust to dropping these three ac-

tors. In other cases, such as Orson Welles’s

quadruple nomination for Citizen Kane

(1941) (best picture, director, original screen-

play, and lead actor), the nominations are

simultaneous, so the actor is not considered

to benefit from working with a prior-nomi-

nated writer and director.

Controls. Because there are a fixed num-

ber of nominations available, the volume of

competition, by necessity, is relevant to an

actor’s chances of nomination. We therefore

control for the number of eligible films in

a given award year. We also include a binary

variable for actresses because there are fewer

roles for women than for men in Hollywood

(for every performance in our data by

a woman there are two by men) but the

Rossman et al. 39

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

same number of nominations and awards (for

a further exploration of gender differences, see

the Online Supplement).

Because the Academy seems to prefer cer-

tain genres, we include a set of control dum-

mies for drama, biography, and comedy

(Levy 2003). The IMBD genre codes are not

mutually exclusive and most films have mul-

tiple genres; for instance, many dramas are

really hybrids of drama and other genres. 8

We include a binary variable indicating that

a major firm controlled the film’s first-run

U.S. distribution rights. These firms include

all variations and subsidiaries of Viacom/

Paramount, Metro-Goldwyn-Mayer/Loew’s,

Fox, Warner Bros., Radio-Keith-Orpheum,

Sony/Columbia, Disney, United Artists,

MCA/Universal, and Orion (Compaine and

Gomery 2000; Vogel 2001). The variable de-

scribes which company advertised a film and

distributed it to theaters. In recent decades,

this company may not have originally pro-

duced the film but only purchased or licensed

it. We measure release date as day of the year

for the Los Angeles or general release of the

film. 9

The Academy may be more likely to

notice films with larger budgets because

they have more resources. Ideally, we would

include budget as a control. Unfortunately,

studios are loath to release budget figures,

so the IMDB has enormous levels of missing

data for this variable. The IMDB has budget

data for about 10 percent of Oscar-eligible

films, and a cursory spot-check shows that

many expensive films have missing data. It

would thus be inappropriate to impute values

of zero. We experimented with including the

budget variable by first using the Bureau of

Labor Statistics Consumer Price Index defla-

tor to make figures comparable across years

and then breaking the variable into a dummy

set of quartiles with missing data as the

omitted category. Including this dummy set

in our models does not appreciably change

the other effects. As suggested by the fact

that the most expensive films in the data

are historical epics like Cleopatra (1963),

large casts are expensive. Because the

IMDB data on credited roles are much

higher quality than its budget data, we use

cast size as a proxy for budget in our models.

The major distributor variable also helps

proxy for budget, and the restriction to eligi-

ble films screens out most truly obscure and

tiny films. Results including the budget

dummy set are available from the first author

on request.

ANALYSIS AND RESULTS

We attempt to predict which performances

will be nominated for an Academy award

for acting. In all models, the log-odds of an

Oscar nomination are the outcome, and the

unit of analysis is the performance, with the

universe constrained to the top-10 credited

roles in Oscar-eligible films. To the extent

that one is interested in an actor being nom-

inated for an Oscar in a time period (rather

than a film), one can conceive of this article

as estimating nomination given the prior con-

dition that an actor appears in a top-10

credited role of a nonobscure film.

Zuckerman and colleagues (2003) model an

outcome approximating this precondition.

Table 2 presents a correlation-variance-

covariance matrix for the variables used in

the models. As previously mentioned, the da-

taset includes 147,908 performances by

37,183 actors in 16,392 films. From 1936

to 2005, there were a total of 1,326 Oscar

nominations and 279 wins; less than 1 per-

cent of the performances in the dataset

were nominated and it is even rarer to win.

Of the 1,326 nominations, 766 actors had at

least one nomination. Meryl Streep had the

most nominations (13) and Katherine

Hepburn had the most wins (four).

An issue complicating our modeling is that

performances are not independent but exhibit

cross-classified clustering; our 147,908 per-

formances contain multiple observations of

our 37,183 actors and 16,392 films. Ideally,

we would model both types of autocorrela-

tion. Unfortunately, this is not feasible

40 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

T a b le

2 .

C o rr

e la

ti o n

-V a ri

a n

c e -C

o v a ri

a n

c e

M a tr

ix

1 2

3 4

5 6

7 8

9 1 0

1 1

1 2

1 3

1 4

1 5

1 6

1 7

1 .

O s c a r

N o m

in a ti

o n

a .0

1 .0

0 .0

0 .0

0 .0

0 .0

0 .0

6 .5

7 .0

0 .0

1 .0

2 2

.0 1

.0 0

.0 5

.0 1

.0 0

.0 0

2 .

F il

m s

p e r

Y e a r

2 .0

3 .8

7 2

.0 4

2 .0

5 2

.0 1

2 .0

2 2

6 .8

1 2

3 .1

7 2

.0 1

.3 1

1 .4

1 4 .2

6 2

.0 2

2 .9

7 2

.1 6

2 .0

2 2

.0 1

3 .

G e n

re :

D ra

m a

.0 7

2 .0

9 .2

5 2

.0 8

.0 1

.0 1

.7 8

2 .2

7 .0

1 2

.0 6

2 .2

5 2

1 .0

0 .0

1 2

.3 4

.0 8

.0 2

.0 2

4 .

G e n

re :

C o m

e d

y 2

.0 2

2 .1

2 2

.3 2

.2 2

2 .0

1 .0

2 1 .4

7 .2

1 .0

1 .0

1 2

.0 5

2 .4

2 .0

0 .5

5 .0

1 .0

0 .0

0

5 .

G e n

re :

B io

g ra

p h

y .0

5 2

.0 3

.1 2

2 .0

8 .0

3 .0

0 .3

5 .6

1 .0

0 .0

0 2

.0 1

2 .0

6 .0

0 .0

4 .0

1 .0

0 .0

0

6 .

M a jo

r D

is tr

ib u

to r

.0 4

2 .0

5 .0

4 .0

8 .0

3 .2

1 1 .2

3 .9

6 .0

0 .0

6 .0

8 2

.4 1

.0 1

1 .1

1 .0

8 .0

2 .0

3

7 .

C a s t

S iz

e .0

3 2

.4 0

.0 9

.1 8

.1 2

.1 5

3 2 4 .9

4 1 0 1 .6

5 .1

2 2

2 .0

4 2

1 6 .2

4 2

6 3 .5

7 .2

6 3 0 .0

9 2 .8

4 .4

0 .3

5

8 .

R e le

a s e

D a te

.0 6

2 .0

3 .0

4 .0

0 .0

4 .0

2 .0

5 1 0 7 0 9

.2 8

2 4 .8

5 2

9 .4

3 2

4 1 .1

8 .8

9 2

1 .9

4 7 .3

6 2 .3

9 2 .3

4

9 .

F e m

a le

.0 4

2 .0

3 .0

3 .0

5 .0

0 .0

0 .0

1 .0

1 .2

1 2

.1 1

2 .5

3 2

1 .5

9 .0

1 2

.3 9

.0 1

.0 0

.0 0

1 0 .

P a s t

F il

m s

0 to

5 .0

3 .1

7 2

.0 6

.0 1

2 .0

1 .0

7 2

.0 6

2 .0

2 2

.1 2

3 .8

4 8 .6

3 1 1 .4

3 .1

1 1 7 .1

0 .1

6 .0

1 .0

5

1 1 .

P a s t

F il

m s

6 to

2 0

.0 3

.2 3

2 .0

8 2

.0 2

2 .0

1 .0

3 2

.1 4

2 .0

1 2

.1 7

.6 7

4 3 .5

7 7 2 .8

3 .4

8 4 7 .2

1 .3

5 .0

3 .1

6

1 2 .

P a s t

F il

m s

. 2 0

2 .0

1 .2

2 2

.1 0

2 .0

4 2

.0 2

2 .0

4 2

.1 7

2 .0

2 2

.1 6

.2 8

.5 2

4 4 3 .4

5 .7

9 7 4 .6

0 2

.4 2

2 .0

7 .0

0

1 3 .

P a s t

N o m

in a ti

o n

s .1

3 2

.0 7

.0 6

.0 1

.0 3

.0 7

.0 5

.0 3

.0 5

.2 0

.2 6

.1 3

.0 8

.7 4

.0 4

.0 1

.0 2

1 4 .

S ta

tu s

(C re

d it

C e n

tr a li

ty )

.0 4

.2 2

2 .0

5 .0

8 .0

2 .1

7 .1

2 .0

0 2

.0 6

.6 1

.5 0

.2 5

.1 9

2 0 2 .9

6 1 .4

5 .2

3 .5

8

1 5 .

C o s ta

rs .0

6 2

.1 7

.1 5

.0 3

.0 7

.1 8

.1 6

.0 7

.0 2

.0 8

.0 5

2 .0

2 .1

6 .1

0 .9

9 .0

9 .1

3

1 6 .

D ir

e c to

r .0

9 2

.0 5

.1 3

2 .0

2 .0

9 .1

4 .0

7 .0

7 .0

1 .0

2 .0

2 2

.0 1

.1 1

.0 5

.2 7

.1 0

.0 4

1 7 .

W ri

te rs

.0 5

2 .0

2 .0

9 .0

2 .0

6 .1

4 .0

4 .0

4 .0

1 .0

5 .0

5 .0

0 .1

1 .0

8 .2

6 .2

7 .2

6

N o te

: T

h e

d ia

g o n

a l

o f

th is

ta b le

s h

o w

s v a ri

a n

c e .

T h

e n

u m

b e rs

a b o v e

th e

d ia

g o n

a l

a re

c o v a ri

a n

c e s

a n

d th

o s e

b e lo

w th

e d

ia g o n

a l

a re

c o rr

e la

ti o n

s .

a B

e c a u

s e

‘‘ O

s c a r

n o m

in a ti

o n

’’ is

a b in

a ry

v a ri

a b le

w it

h a

fr e q u

e n

c y

o f

.0 0 9 ,

it n

e c e s s a ri

ly h

a s

a lo

w c o rr

e la

ti o n

a n

d c o v a ri

a n

c e

w it

h o th

e r

v a ri

a b le

s ,

e v e n

if th

o s e

v a ri

a b le

s a re

in fa

c t

h ig

h ly

a s s o c ia

te d

w it

h n

o m

in a ti

o n

.

41

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

because observed combinations of the classes

are extremely sparse; any given actor appears

in less than 1 percent of all films and any

given film has less than 1 percent of all actors

in its cast. Such a data structure makes any

analysis attempting to fully account for all

auto-correlation underidentified. We are

therefore forced to settle for approximation

and triangulation.

We experimented extensively with vari-

ous specifications. First, we alternated

between random effects by film or by actor.

Second, we alternated between specifying

random effects by one class and bootstrap-

ping on the other. The results are largely

robust with effects maintaining direction

and significance across specifications,

although parameters associated with a class

decrease in significance slightly when ran-

dom effects are specified for that class.

Because all results are basically consistent

regardless of how and by what class we

model auto-correlation, we focus on pre-

senting results for models with random ef-

fects for film. We settled on this

specification for two reasons. First, very lit-

tle auto-correlation is associated with actors

but a moderate amount is associated with

films. Second, because our key hypothesis

treats human capital, a set of actor-level

traits, as a theoretical baseline, and peer

spillovers, a set of film-level traits, as the

object of our most novel hypothesis, this

specification gives a conservative bias to

our hypothesis testing. Were a perfect esti-

mation possible, it would probably leave

the effects of film-level parameters (i.e.,

peer spillovers) unchanged and slightly

reduce the effects of actor-level parameters

(i.e., human capital). Alternate specifica-

tions are available from the first author on

request.

We begin our analysis in Table 3 with

a nested set of models using these sets of

variables: controls, human capital, status,

and spillovers. All of these models include

film random effects. Model 1 presents re-

sults from the baseline model. The effects

of each of these control variables are strong

and in the predicted direction. An actor is

most likely to be nominated in a year with

few films per year (i.e., fewer actors are

competing for nominations). Similarly,

female actors have a better chance of being

nominated for any given role than do males,

because there are fewer roles for women and

thus less competition for a fixed number of

nominations. Actors are also most likely to

be nominated for dramatic or biographical

genre films that were released late in the

year. These baseline effects maintain their

strength and direction in all subsequent

models. Performances in films distributed

by major studios are also much more likely

to be nominated. Unlike the female and dra-

matic genre effects, however, the benefit of

a major studio release is somewhat attenu-

ated as more variables are introduced in suc-

cessive models. This implies that much, but

not all, of the reason why major studio films

accrue disproportionately more acting nom-

inations is because they recruit top actors

and put them in teams with other elite

workers.

Model 2 adds human capital variables to

the baseline. The acting experience variable

is specified as a linear spline so coefficients

should be read as marginal but cumulative.

For example, to calculate the effects of expe-

rience for an actor who has worked on seven

films, multiply the ‘‘past films 0 to 5’’ coef-

ficient by five and multiply the ‘‘past films 6

to 20’’ coefficient by two and take the sum of

these two multiples. As expected from previ-

ous research (Levy 2003; Zuckerman et al.

2003), the odds of Oscar nomination increase

with an actor’s experience in the film indus-

try up to a point, then level off and eventually

decline slightly. Although these results are

based on a spline, if we specify experience

as a quadratic or a dummy set we obtain sim-

ilar results. A history of Academy nomina-

tions strongly predicts future Academy

nominations, an illustration of Merton’s

(1968) Matthew Effect. Ideally, we would

be able to perfectly model stable aspects of

42 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

human capital with fixed-effects, but because

of the nature of the dependent variable this

would severely constrain the scope of the

analysis and the interpretation. Nonetheless,

using fixed-effects as a robustness check,

we arrive at similar results.

Table 3. Logistic Regression Models of Academy Award Nominations, 1936 to 2005

Models

Variable 1 2 3 4

Baseline

Films per Year (100s) 2.287*** 2.272*** 2.400*** 2.390***

(.056) (.058) (.060) (.062)

Genre: Drama 2.127*** 2.062*** 2.109*** 1.936***

(.125) (.128) (.131) (.134)

Genre: Comedy .004 2.022 2.046 2.030

(.110) (.113) (.115) (.117)

Genre: Biography 1.311*** 1.277*** 1.311*** 1.144***

(.172) (.175) (.179) (.180)

Major Distributor 1.125*** .952*** .847*** .544***

(.117) (.120) (.123) (.126)

Cast Size .004 .003 2.001 2.002

(.002) (.002) (.002) (.002)

Release Date .007*** .006*** .007*** .006***

(.000) (.000) (.000) (.000)

Female .849*** .837*** .873*** .864***

(.062) (.068) (.069) (.069)

Human Capital

Past Films 0 to 5 .144*** .023 .025

(.026) (.029) (.029)

Past Films 6 to 20 .016* .005 .005

(.008) (.008) (.008)

Past Films .20 2.007** 2.008** 2.008**

(.002) (.002) (.002)

Past Nominations 1.904*** 1.774*** 1.739***

(.078) (.079) (.081)

Status (Credit Centrality) .049*** .047***

(.005) (.005)

Spillovers

Costars with Past Nominations .163***

(.041)

Director with Past Nominations 1.224***

(.110)

Writers with Past Nominations .359***

(.075)

Intercept 210.134*** 210.999*** 214.177*** 214.083***

(.286) (.308) (.490) (.497)

rfilm .538 .542 .556 .550

Log Likelihood 26148.926 25670.363 25619.257 25502.304

Note: N performances 5 147,908; N films 5 16,392. Standard errors are in parentheses. Please see the Online Supplement for additional specifications that break out nominations by gender and lead/ supporting category, treat the blacklist as a natural experiment, and interact key variables with period. Alternate specifications that use more complex genre specifications, include MPAA ratings, and treat best picture nomination (a measure of film quality) as a mediating variable are available from the first author on request. *p \ .05; ** p \ .01; *** p \ .001 (two-sided z-tests).

Rossman et al. 43

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

Model 3 adds a measure of status, the

asymmetrical centrality to the credit network.

Credit centrality strongly predicts nomina-

tion, supporting Hypothesis 1. A one stan-

dard deviation increase in credit centrality

is roughly equivalent to having a major

film distributor, and a two standard deviation

increase is equivalent to having a prior nom-

ination. ‘‘Past films 0 to 5’’ now drops out

and remains out in all subsequent models

(alternate specifications of age also drop

out against status). We do not interpret this

to mean that experience is a spurious effect

of centrality, but that centrality mediates

the effect of experience. If two actors start

out playing small roles but one remains

a character actor while the other moves up

to playing leading roles (with respectable

supporting costars), they will have the same

amount of experience but the persistent char-

acter actor will have much lower centrality

than the actor who now has leading roles.

Because few actors debut in starring roles,

it is meaningful to think of experience as pro-

viding the opportunity for increasing status

(as measured by centrality), or to view it

more substantively as the development of

star power. Unfortunately, the IMDB lacks

detailed data on other dimensions of actor

quality, such as dramatic training, diction,

or physical attractiveness. If such aspects

of actor quality were available, we specu-

late that their effects would also be largely

mediated by the intervening variable of sta-

tus or star power. Being a prior Oscar nom-

inee still has a large effect when credit

centrality enters the model, but this is

consistent with our interpretation because

having a prior Oscar nomination is concep-

tually very similar to credit centrality. In

both cases, an actor’s peers are acknowl-

edging his star power as someone who de-

serves the deference of outranking other

actors in the credits or being honored with

an Oscar nomination. 10

Model 4 adds spillover effects from col-

laborating with prior-nominated individu-

als. 11

Elite costars, writers, and directors all

significantly affect an actor’s likelihood of

being nominated, net of the focal actor’s

human capital and status. This supports

Hypothesis 2, which predicts that top peers

will make one more likely to be nominated.

The model’s results show that it is good to

work with an elite team but do not reveal

the mechanisms behind this pattern.

Writers, directors, and costars likely vary in

how they can contribute to an actor’s conse-

cration. The most basic distinction should be

between spillovers offered by costars and

those from behind-the-camera talent.

Costars may offer chemistry through the

interaction of the performance. Directors

and actors may have a relationship similar

to that of orchestra conductors and their

musicians, in which musicians expect the

conductor to not only have technical compe-

tence at music but a clear ambition and

ability to communicate it (Faulkner 1973).

Writers provide the story within which an

actor’s performance is situated and the dia-

logue that constitutes it—the more plausible

these are, the greater the room for the actor’s

performance. Finally, elite collaborators,

regardless of occupation, will attract atten-

tion and legitimacy for the film. In the

Online Supplement, we use the blacklist as

a natural experiment to explore the distinc-

tion between spillovers of prestige and of

talent. That analysis is inconclusive, but it

suggests that even secret collaborations can

have substantial spillovers and thus (at least

for writers) spillovers are mostly of talent

rather than prestige. Note, though, that writ-

ers lack any direct equivalent to either actor

marquee value or director auteurship.

The greatest spillover coefficient is for

directors, but this is in part because a film

can have only one Oscar-nominated direc-

tor but several Oscar-nominated writers.

Furthermore, the Academy has just one di-

recting award but two categories for writing

(and between 1940 and 1956, there were

three categories). Any given directing nom-

ination may therefore signify greater talent

and prestige than any given writing

44 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

nomination. Indeed, taking into account the

range of the variables, the effects of direc-

tors and writers are comparable, although di-

rectors remain more important. Costars have

less of an effect than do writers and direc-

tors. This distinction appears to be because

costars provide spillovers and competition,

and these partially cancel each other out.

This interpretation is supported in the

Online Supplement, where we treat lead

and supporting nominations as alternative

outcomes and show that prior-nominee co-

stars are very beneficial for supporting nom-

inations but a wash for lead nominations.

To help get a sense of how the effects

work together, we compared the effects in

Model 4 at the median to those at the 95th

percentile for several of the variables. 12

Holding everything else constant, such

a boost in credit centrality increases the

log-odds by .55. Changing centrality, acting

experience, and an actor’s prior Oscar nomi-

nations increases the log-odds by 1.38. In

other words, a median performance in

a median film has about a .58 percent chance

of being nominated, whereas an exceptional

actor appearing in the same film has about

a 1.00 to 2.26 percent chance of being nom-

inated. For an actor with a prior nomination,

high centrality, and exactly five prior films,

the predicted probability is about 5.34

percent if the actor appears in the median

film. If we contrast a median actor in

a median film with a similar actor in a film

at the 95th percentile for peer spillovers

(with the film being otherwise at the

median), the predicted probability shifts

from about .58 to 4.42 percent. Finally, plac-

ing an exceptional actor in a film with excep-

tional spillovers increases the log-odds by

4.00, for a predicted probability of about

24.16 percent. Spillovers thus matter about

as much as the observed aspects of individual

status and accomplishment. However,

because good actors with weak spillovers

and weak actors with good spillovers both

have weak chances of being nominated, it

is more appropriate to say that an actor needs

both exceptional personal status and excep-

tional spillovers to have an appreciable

chance at an Oscar nomination. Another

way to read this is that exceptional actors

are most likely to be nominated when they

work with exceptional peers, a theme we

explore in Model 5 of Table 4.

Model 5 presents fixed-effects models by

actor as a robustness check against the error

modeling strategy used in Models 1 through

4 of specifying random effects for film. 13

Fixed-effects models allow each actor to

have a different intercept, fully accounting

for any stable characteristic of an actor and

thus excluding the possibility of omitted vari-

able bias for any stable trait of the actor.

Fixed-effects models, however, can only esti-

mate off cases with variance in the outcome;

they cannot simply ask the question ad-

dressed in Models 1 through 4, ‘‘What as-

pects of films and actors get a performance

nominated for an Oscar?’’ Instead, they

must ask the narrower question, ‘‘Among

those actors who at some point are nominated

for an Oscar, what were the traits of the films

for which they were nominated?’’

Model 5 is similar to Model 4, but to

avoid collinearity and perfect prediction

problems with the fixed-effects it is limited

to actors with at least one nomination.

Likewise, to avoid regression to the mean is-

sues, we drop actors after their first nomina-

tions and the model omits the number of

prior films spline. 14

The film-level effects

in Model 5 are consistent with those in

Model 4. Hypothesis 1 continues to be sup-

ported, despite the fact that the fixed-effects

capture all stable aspects of the actor—in-

cluding those that we lack measures for. If

centrality were entirely reducible to fixed or

fairly stable traits like physical attractive-

ness, it would have no effect net of the

fixed-effect. That a large centrality effect

persists is consistent with our interpretation

of individual talent as providing the opportu-

nity for the development of star power,

which in turn directly affects the likelihood

of consecration.

Rossman et al. 45

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

We argue that between status and actor

fixed-effects, we are capturing all of the

fixed and much of the time-varying aspects

of actor star power. Therefore, the parsimoni-

ous explanation for the remaining effects for

film-level variables is that they are real and

not reducible to selectivity effects, such as

the sorting of good actors to work with

good teams. In particular, as Hypothesis 2

predicts, it remains very beneficial to work

with elite costars, directors, and writers.

The consistency of the film-level results,

whether we model substantive actor traits

(and film random effects) or actor fixed-ef-

fects, gives us strong confidence that the re-

sults are robust to specification. Despite

very different specifications in both cases,

we see that actors are most likely to be nom-

inated for Oscars when they have high status

and are working with an elite team.

CONCLUSIONS

We find that personal- and contextual-level

factors shape an actor’s chances of being cul-

turally consecrated. At the level of the actor,

we find strong effects for status, as measured

by centrality in the asymmetric network of

screen credits. This is not to say that mem-

bers of the Academy’s acting branch are

actually tallying who outranks whom in

screen credits, let alone calculating eigenvec-

tors, but rather that credit centrality is an effi-

cient summary of status or star power.

Furthermore, because centrality renders

experience insignificant and remains strong

with actor fixed-effects, we interpret status

as mediating the effects of more tangible tal-

ents and accomplishments. This is consistent

with the cumulative advantage literature

arguing that status is endogenous and par-

tially autonomous from objective quality

(Gould 2002; Salganik et al. 2006).

At the contextual level, we find that the

prestige and merits of a film actor’s collabo-

rators, particularly the writer and director,

greatly increase her own chances for recogni-

tion. Collaborations are ubiquitous in many

fields; this is well-noted in the networks liter-

ature because most large public datasets use

collaborations to measure ties. With few ex-

ceptions (e.g., Uzzi and Spiro 2005), how-

ever, this is often treated as a nuisance

rather than a source of intrinsic interest. In

this article, we show that one’s immediate

collaborative context is an important deter-

minant of individual accomplishment. This

finding ties into a growing economic

Table 4. Logistic Regression Model of Context of Academy Award Nominations among Actors Nominated at Least Once, 1936 to 2005, with Actor Fixed-Effects

Variable Model 5

Baseline

Films per Year (100s) 2.536***

(.095)

Genre: Drama 1.377***

(.154)

Genre: Comedy 2.077

(.135)

Genre: Biography .691***

(.188)

Major Distributor .146

(.145)

Cast Size .004

(.003)

Release Date .004***

(.000)

Status (Credit Centrality) .098***

(.012)

Spillovers

Costars .134**

(.044)

Director .735***

(.115)

Writers .162*

(.079)

Log Likelihood 2975.864

Note: N 5 6,655 performances by 624 actors. Standard errors are in parentheses. Because of the perfect prediction problem, we include only actors who are nominated at least once. For the same reason, we omit the variables female, past nomination, and the ‘‘past films’’ spline. A specification using film fixed-effects is available from the first author on request. *p \ .05; ** p \ .01; *** p \ .001 (two-sided z- tests).

46 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

literature and a dormant sociological litera-

ture on intrateam spillovers (e.g., Saint-Paul

2001; Stinchcombe 1963).

Parsing out exactly why it is beneficial to

work with good peers is difficult. Most of the

literature assumes that it is an effect of skill

complementarities. For complex collabora-

tions where it is difficult to measure individ-

ual effort, parsing out each worker’s

marginal contribution of value becomes an

intractable problem. Therefore, each work-

er’s productivity is in part determined by

the aggregate production; if top workers

bring up this aggregate then all team mem-

bers benefit (Alchian and Demsetz 1972).

An alternative interpretation would view

spillovers as concerning prestige and reputa-

tion rather than skill and talent. That is, an

actor’s immediate collaboration may give

legitimacy to a performance, in much the

same way that an actor’s long-term pattern

of association reflects or constitutes the ac-

tor’s status. We can return to our familiar

case of Robert Forster as an example.

According to the conventional skill spillover

interpretation, Forster benefitted from work-

ing with Tarantino because it meant he

finally got to work with a decent script. By

contrast, according to the prestige spillover

interpretation, Academy members figured

that if Tarantino chose to work with Forster

then Forster must be a diamond in the rough,

and Tarantino’s fame meant that, unlike most

of Forster’s films, Academy voters actually

saw Jackie Brown (1997). Unfortunately, it

is almost impossible to cleanly identify spill-

overs of skill versus those of prestige, so the

distinction must remain speculative until

a future study disentangles them. Our analy-

sis in the Online Supplement suggests that

spillovers are mostly of skill, but the num-

bers involved are too small to be conclusive.

While the micro-mechanisms of spillovers

are not entirely understood, such a lacuna

should not obscure the basic finding that

spillovers matter.

Because collaborations with high-quality

peers help determine an individual’s success,

the process by which collaborations form is

an important intervening mechanism for indi-

vidual life-chances. From World War II to

the mid-1970s, it may have been adequate

to note that there was a persistent wage pre-

mium for ‘‘core’’ workers at large, capital-

intensive employers over ‘‘periphery’’ work-

ers at smaller firms (Beck, Horan, and

Tolbert 1978; Katz et al. 1989). In the post-

industrial economy, however, large firms,

while still important, are increasingly less

so as work is increasingly flexible and orga-

nized around teams rather than firms

(Hollister 2004; Lazear and Shaw 2007).

Under such decentralized conditions, a mech-

anism for inequality is not so much for whom

but with whom a worker will labor.

Much of the theoretical and empirical lit-

erature in economics emphasizes matching of

workers by skill (Kremer 1993; Saint-Paul

2001), with matching in marriage markets

being a common analogy (Becker 1973).

Most of these models predict that high-qual-

ity labor inputs will not only sort together but

also attract the best inputs of other kinds,

such as cheap access to finance capital. In

such circumstances, homogenously excellent

teams will see success greater than the sum

of their individual excellence. Likewise, ho-

mogenously mediocre teams will see failure

greater than the sum of their individual medi-

ocrity. As predicted, corporate demography

compatible with sorting workers into firms

by productivity highly predicts inequality

(Sørensen and Sorenson 2007).

While the idea of team sorting is fairly

unfamiliar in the recent sociology literature,

it is closely analogous to ideas in the residen-

tial segregation literature. Neighborhood ef-

fects emerge directly or indirectly from

neighbors themselves (Massey and Denton

1993). When one lives among the affluent,

one has access to their social capital, the

labor markets and consumer amenities cre-

ated by their entrepreneurship and consump-

tion, state benevolence ensured by their

political power and tax base, rising property

values due to the desire of others to enjoy

Rossman et al. 47

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

such pleasant company, and various other

positive externalities. By contrast, living

among the impoverished implies a vacuum

of such desirable features and exposure to

crime. Just as sociology has long recognized

that our neighbors expose us to externalities,

we ought to extend this understanding to our

co-workers. Likewise, as sociology recog-

nizes that sorting into neighborhoods is an

important intervening mechanism for life-

chances, so should we understand sorting

into work teams.

Sociological studies examining team for-

mation in Hollywood have found extremely

strong evidence of sorting among writers,

producers, directors, cinematographers, and

actors (Faulkner and Anderson 1987;

Zuckerman 2004). Furthermore, there is evi-

dence that sorting is partially mediated by

brokerage and ascribed characteristics

(Bielby and Bielby 2002; Bielby and Bielby

1999). More generally, by studying how

advantaged collaborations form and what

benefits accrue to them, one can ultimately

explain the distribution of resources through

many fields and industries. Indeed, the sort-

ing and spillover mechanism may even be

able to explain the third-world poverty trap

(Kremer 1993). From our perspective, there

is a very good reason that Academy Award

acceptance speeches are so long—they

should be, because an actor’s collaborators

are largely responsible for his achievement.

If the rest of us had occasion for acceptance

speeches, due humility would suggest a simi-

lar practice.

Acknowledgments

The authors would like to thank Xiao Chen, Paul

DiMaggio, Charles Kirschbaum, Pierre Kremp, Bill

Mason, Christine Percheski, Laura Piersol, Martin

Ruef, Ezra Zuckerman, the workshop participants at

UCSD and the Princeton University CACPS seminar,

the California Center for Population Research, UCLA

Academic Technology Services, the Robert Wood

Johnson Scholars in Health Policy Research Program,

the ASR reviewers, and the Academy of Motion

Picture Arts and Sciences.

Notes

1. In the sense that we use the term, ‘‘status’’ means

personal esteem and is conceptually distinct from

Weber’s discussion of status groups, although they

may overlap in two senses. First, others may use an

individual’s apparent membership in a status group

to inform their estimate of that person’s status.

Second, membership in status groups may provide

social closure and structure to social networks in

such a way that has implications for individuals’ sta-

tus (Tilly 1998).

2. Whereas Stinchcombe (1963) and related models

(Saint-Paul 2001) assume that actors benefit from

association with high-quality peers, the ‘‘weakest

link’’ or ‘‘O-ring’’ theory notes that one can suffer

from association with low-quality peers in situa-

tions where risk-averseness is desirable so as to

avoid catastrophic failure (Jacobs 1981; Kremer

1993).

3. Spillovers can backfire, however, if observers are

conscious of these dynamics and therefore discount

the contributions of the least-skilled worker in

a team, as when junior scientists receive little credit

for coauthoring with laureates (Merton 1968).

Furthermore, some studies that find spillovers

involve situations for which it is implausible to

allege complementarity because work is parallel

rather than collaborative, as with fruit pickers

(Bandiera, Barankay, and Rasul 2005). This litera-

ture argues that social expectations of diligence

are the key mechanism, with workers conforming

to the locally modal level of productivity.

4. Even though it is inappropriate for measuring star

power, betweenness could have other applications

to Hollywood. For instance, it has implications for

such issues as continued ability to find work or abil-

ity to transcend typecasting.

5. We base the metric on all IMDB film credits. This

allows an actor to receive deference not only from

lower ranked costars, but from bit players as well.

Likewise, acts of deference that occur in the credits

of obscure films also inform our understanding of

star power. Calculating centrality on all IMDB

film credits is practical because few performances

in our dataset (i.e., top-10 credits in Oscar eligible

films) have centrality scores at or close to zero.

We experimented with calculating centrality using

only our dataset, and this version of the metric

has a .7 correlation with centrality based on all

IMDB film credits. The two versions also behave

similarly in regression models.

6. In addition to the five-year window, we also experi-

mented with deleting ties after 3, 7, and 10 years; in

all cases, results are similar.

7. In fact, well after we originally wrote this specula-

tion into early drafts of this article, Affleck co-

wrote his second screenplay, Gone Baby Gone

48 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

(2007), and one of the actors in this film, Amy

Ryan, was nominated for her performance.

8. With the exception of having a special category for

‘‘best documentary,’’ the Academy does not make

explicit use of genre distinctions. All of our genre

data come from the IMDB, which does not draw

distinctions such as ‘‘main genre’’ versus ‘‘second-

ary genre’’ but treats all genre codes applied to

a film as having implicitly equal weight. Our spec-

ification uses only a few of the available IMDB

genre codes and assumes that their effects are addi-

tive. We experimented with more nuanced specifi-

cations that include the full set of IMDB genre

codes and are sensitive to potentially combinatorial

effects of multiple genre codes. The findings are

robust to this more nuanced specification. We also

experimented with specifying MPAA ratings, which

had no effect. These specifications are available

from the first author on request.

9. The IMDB is missing release date information for 3

percent of Oscar-eligible films. In these cases, we

imputed values by drawing from a uniform distribu-

tion. This imputation introduces measurement error

that should raise the standard error of release date

slightly but leave the coefficient and effects on

other variables essentially the same.

10. Because status is measured through social networks

and Oscar nominations are determined by (a subset

of) peers, an alternative interpretation of centrality

might be that it measures connection to voters, not

status. We therefore experimented with using nom-

inations for the Golden Globes, which are not a peer

award, and the results are similar. Likewise, our re-

sults are robust to treating recent collaborations

with prior nominees as a proxy for network ties to

the Academy. We can thus state with a fair degree

of certainty that our findings of status and spillover

effects are not spurious effects of past collaboration

ties to nominating peers. To use Podolny’s (2001)

terminology, we are certain that our ‘‘prisms’’ are

not really ‘‘pipes.’’

11. To test for changes associated with changes to

Hollywood’s organization over time, we experi-

mented with an alternative to Model 4 that allows

key effects to vary by period. Status effect and spill-

overs from writers and directors are appreciable and

statistically significant in all periods, although spill-

overs from costars are not statistically significant in

the blockbuster era (post-1975).

12. Although we can say precisely how much a change

in values shifts the log-odds, it is more difficult to

say what the change in the predicted probability

will be, because assuming a random effect of zero

gives predicted probabilities that are an order of

magnitude lower than the empirical nomination

rate of .9 percent for the performances in the data-

set. To estimate an adjustment, we generated mod-

els with and without the random effects (which have

similar coefficients but different intercepts) and cre-

ated predictions where all traits are at the median.

The difference between the two predictions is 2.02

on the logit scale, and we found that adding this fig-

ure to the Model 4 predictions gives more reason-

able predicted probabilities. This adjustment is for

the illustrative purpose of calculating predicted

probabilities only, with the intercept and all other

figures in the regression tables remaining

unchanged. Absent the adjustment, there would be

no ordinal change in the predicted probabilities,

but all of them would be much lower (indeed

implausibly low). Furthermore, the proportion of

the binary variable ‘‘drama’’ is just barely over .5

and so the median is one. If the vignettes instead

assume a value of zero for ‘‘drama,’’ the predicted

probabilities will be much lower.

13. We also experimented with a model using fixed-ef-

fects by film and substantive independent variables

by actor. This model has a similarly constrained

interpretation but confirms that status is a strong

predictor.

14. Because we limit the analysis to first nomination, we

must omit the ‘‘actor’s prior nominations’’ variable.

Likewise, we omit the acting experience variables

because when the population is limited to only actors

who experience nomination and they exit after their

first nomination, the acting experience variables have

no substantive interpretation but behave as a count-

down to the inevitable event. This can be thought of

as a special case of a survival analysis with no censor-

ing and where age is a predictor variable.

References

Abrahamson, Mark. 1973. ‘‘Talent Complementarity

and Organizational Stratification.’’ Administrative

Science Quarterly 18:186–93.

Academy of Motion Picture Arts and Sciences. 2009.

‘‘About the Academy Awards’’ (http://www.oscars

.org/awards/academyawards/about/).

Adler, Moshe. 1985. ‘‘Stardom and Talent.’’ American

Economic Review 75:208–12.

Alchian, Armen A. and Harold Demsetz. 1972.

‘‘Production, Information Costs, and Economic

Organization.’’ American Economic Review

62:777–95.

Anand, N. and Mary R. Watson. 2004. ‘‘Tournament

Rituals in the Evolution of Fields: The Case of the

Grammy Awards.’’ Academy of Management

Journal 47:59–80.

Axelrod, Robert. 1980. ‘‘Effective Choice in the

Prisoner’s Dilemma.’’ The Journal of Conflict

Resolution 24:3–25.

Baker, Wayne E. and Robert R. Faulkner. 1991. ‘‘Role

as Resource in the Hollywood Film Industry.’’

American Journal of Sociology 97:279–309.

Rossman et al. 49

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

Bandiera, Oriana, Iwan Barankay, and Imran Rasul.

2005. ‘‘Social Preferences and the Response to

Incentives: Evidence from Personnel Data.’’

Quarterly Journal of Economics 120:917–62.

Battu, Harminder, Clive R. Belfield, and Peter J. Sloane.

2004. ‘‘Human Capital Spillovers in the Workplace:

Evidence for the Service Sector in Britain.’’

International Journal of Manpower 25:123–38.

Baumann, Shyon. 2001. ‘‘Intellectualization and Art

World Development: Film in the United States.’’

American Sociological Review 66:404–26.

———. 2007. Hollywood Highbrow: From

Entertainment to Art. Princeton, NJ: Princeton

University Press.

Beck, E. M., Patrick M. Horan, and Charles M. Tolbert

II. 1978. ‘‘Stratification in a Dual Economy: A

Sectoral Model of Earnings Determination.’’

American Sociological Review 43:704–720.

Becker, Gary S. 1973. ‘‘A Theory of Marriage: Part I.’’

Journal of Political Economy 81:813–46.

Becker, Howard S. 1978. ‘‘Arts and Crafts.’’ American

Journal of Sociology 83:862–89.

Benjamin, Beth A. and Joel M. Podolny. 1999. ‘‘Status,

Quality and Social Order in the California Wine

Industry.’’ Administrative Science Quarterly

44:563–89.

Bielby, Denise and William T. Bielby. 2002.

‘‘Hollywood Dreams, Harsh Realities: Writing for

Film and Television.’’ Contexts 1:21–27.

Bielby, William T. and Denise Bielby. 1994. ‘‘‘All Hits

Are Flukes’: Institutionalized Decision Making and

the Rhetoric of Network Prime-Time Program

Development.’’ American Journal of Sociology

99:1287–1313.

———. 1999. ‘‘Organizational Mediation of Project-

Based Labor Markets: Talent Agencies and the

Careers of Screenwriters.’’ American Sociological

Review 64:64–85.

Bonacich, Phillip and Paulette Lloyd. 2001. ‘‘Eigenvector-

Like Measures of Centrality for Asymmetric

Relations.’’ Social Networks 23:191–201.

Bourdieu, Pierre. 1993. The Field of Cultural

Production. New York: Columbia University Press.

Cardoso, Ana Rute. 2000. ‘‘Wage Differentials Across

Firms: An Application of Multilevel Modeling.’’

Journal of Applied Econometrics 15:343–54.

Caves, Richard E. 2000. Creative Industries: Contracts

Between Art and Commerce. Cambridge, MA:

Harvard University Press.

Compaine, Benjamin M. and Douglas Gomery. 2000.

Who Owns the Media? 3rd ed. Mahwah, NJ:

Lawrence Erlbaum Associates.

Cowen, Tyler. 2000. What Price Fame. Cambridge, MA:

Harvard University Press.

DiMaggio, Paul. 1981. ‘‘Cultural Entrepreneurship in

Nineteenth-Century Boston. Part 1: The Creation of

an Organizational Base for High Culture in

America.’’ Media, Culture and Society 4:33–50.

English, James F. 2005. The Economy of Prestige:

Prizes, Awards, and the Circulation of Cultural

Value. Cambridge, MA: Harvard University Press.

Faulkner, Robert R. 1973. ‘‘Orchestra Interaction: Some

Features of Communication and Authority in an

Artistic Organization.’’ The Sociological Quarterly

14:147–57.

Faulkner, Robert R. and Andy B. Anderson. 1987.

‘‘Short-Term Projects and Emergent Careers:

Evidence from Hollywood.’’ American Journal of

Sociology 92:879–909.

Gould, Roger V. 2002. ‘‘The Origins of Status

Hierarchies: A Formal Theory and Empirical

Test.’’ American Journal of Sociology 107:1143–78.

Griliches, Zvi. 1969. ‘‘Capital-Skill Complementarity.’’

Review of Economics and Statistics 51:465–68.

Gumbel, Peter, John Lippman, Lisa Bannon, and Bruce

Orwall. 1998. ‘‘What’s an Oscar Worth?’’ The Wall

Street Journal, March 20, p. W1.

Hamlen, William A. 1991. ‘‘Superstardom in Popular

Music: Empirical Evidence.’’ Review of Economics

and Statistics 73:729–33.

Hollister, Matissa N. 2004. ‘‘Does Firm Size Matter

Anymore? The New Economy and Firm Size

Wage Effects.’’ American Sociological Review

69:659–76.

Idson, Todd L. and Leo H. Kahane. 2000. ‘‘Team

Effects on Compensation: An Application to Salary

Determination in the National Hockey League.’’

Economic Inquiry 38:345–57.

Jacobs, David. 1981. ‘‘Toward a Theory of Mobility and

Behavior in Organizations: An Inquiry into the

Consequences of Some Relationships between

Individual Performance and Organizational

Success.’’ American Journal of Sociology 87:684–

707.

Katz, Lawrence F., Lawrence H. Summers, Robert E.

Hall, Charles L. Schultze, and Robert H. Topel.

1989. ‘‘Industry Rents: Evidence and

Implications.’’ Brookings Papers on Economic

Activity, Microeconomics 1989:209–90.

Kendall, Todd D. 2003. ‘‘Spillovers, Complementarities,

and Sorting in Labor Markets with an Application to

Professional Sports.’’ Southern Economic Journal

70:389–402.

Kremer, Michael. 1993. ‘‘The O-Ring Theory of

Economic Development.’’ Quarterly Journal of

Economics 108:551–75.

Lazear, Edward P. and Kathryn L. Shaw. 2007.

‘‘Personnel Economics: The Economist’s View of

Human Resources.’’ Journal of Economic

Perspectives 21:91–114.

Levy, Emanuel. 2003. All About Oscar: The History and

Politics of the Academy Awards. New York:

Continuum.

Lincoln, Anne E. 2007. ‘‘Cultural Honours and Career

Promotions: Re-conceptualizing Prizes in the Field

of Cultural Production.’’ Cultural Trends 16:3–15.

50 American Sociological Review 75(1)

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from

Massey, Douglas S. and Nancy A. Denton. 1993.

American Apartheid: Segregation and the Making

of the Underclass. Cambridge, MA: Harvard.

Merton, Robert K. 1968. ‘‘The Matthew Effect in

Science.’’ Science 159:56–63.

Motti, Regev. 1994. ‘‘Producing Artistic Value: The

Case of Rock Music.’’ The Sociological Quarterly

35:85–102.

Nelson, Randy A., Michael R. Donihue, Donald M.

Waldman, and Calbraith Wheaton. 2001. ‘‘What’s

an Oscar Worth.’’ Economic Inquiry 39:1–6.

Petersen, Trond. 1992. ‘‘Individual, Collective, and

Systems Rationality in Work Groups: Dilemmas

and Market-Type Solutions.’’ American Journal of

Sociology 98:469–510.

Podolny, Joel M. 1993. ‘‘A Status-Based Model of

Market Competition.’’ American Journal of

Sociology 98:829–72.

———. 2001. ‘‘Networks as the Pipes and Prisms of the

Market.’’ American Journal of Sociology 107:33–60.

Podolny, Joel M. and Damon J. Phillips. 1996. ‘‘The

Dynamics of Organizational Status.’’ Industrial

and Corporate Change 5:453–71.

Resnik, Gail and Scott Trost. 1996. All You Need to

Know About the Movie and TV Business. New

York: Fireside.

Rosen, Sherwin. 1981. ‘‘The Economics of Superstars.’’

American Economic Review 71:845–58.

———. 1982. ‘‘Authority, Control, and the Distribution

of Earnings.’’ Bell Journal of Economics 13:311–23.

Saint-Paul, Gilles. 2001. ‘‘On the Distribution of Income

and Worker Assignment under Intrafirm Spillovers,

with an Application to Ideas and Networks.’’

Journal of Political Economy 109:1–37.

Salganik, Matthew J., Peter Sheridan Dodds, and

Duncan J. Watts. 2006. ‘‘Experimental Study of

Inequality and Unpredictability in an Artificial

Cultural Market.’’ Science 311:854–56.

Sørensen, Jesper B. and Olav Sorenson. 2007.

‘‘Corporate Demography and Income Inequality.’’

American Sociological Review 72:766–83.

Stinchcombe, Arthur L. 1963. ‘‘Some Empirical

Consequences of the Davis-Moore Theory of

Stratification.’’ American Sociological Review

28:805–808.

Stinchcombe, Arthur L. and T. Robert Harris. 1969.

‘‘Interdependence and Inequality: A Specification

of the Davis-Moore Theory.’’ Sociometry 32:13–23.

Tilly, Charles. 1998. Durable Inequality. Berkeley, CA:

University of California Press.

Torgler, Benno. 2007. ‘‘‘La Grande Boucle’:

Determinants of Success at the Tour de France.’’

Journal of Sports Economics 8:317–31.

Torgler, Benno and Sascha L. Schmidt. 2007. ‘‘What

Shapes Player Performance in Soccer? Empirical

Findings from a Panel Analysis.’’ Applied

Economics 39:2355–69.

Uzzi, Brian and Jarrett Spiro. 2005. ‘‘Collaboration and

Creativity: The Small World Problem.’’ American

Journal of Sociology 11:447–504.

Vogel, Harold L. 2001. Entertainment Industry

Economics: A Guide for Financial Analysis, 5th

ed. Cambridge, UK: Cambridge University Press.

Watts, Duncan J. and Steven H. Strogatz. 1998.

‘‘Collective Dynamics of ‘Small-World’ Networks.’’

Nature 393:440–42.

Wilson, Robert. 1985. ‘‘Reputations in Games and

Markets.’’ Pp. 27–62 in Game-Theoretic Models of

Bargaining, edited by A. Roth. Cambridge, UK:

Cambridge University Press.

Zuckerman, Ezra W. 2004. ‘‘Do Firms and Markets

Look Different? Repeat Collaborations in the

Feature Film Industry, 1935–1995.’’ Working

Paper, Sloan School of Management, MIT.

Zuckerman, Ezra W., Tai-Young Kim, Kalinda Ukanwa,

and James von Rittmann. 2003. ‘‘Robust Identities

or Nonentities? Typecasting in the Feature-Film

Labor Market.’’ American Journal of Sociology

108:1018–74.

Zuckerman, Harriet. 1996. Scientific Elite: Nobel

Laureates in the United States. 2nd ed. New

Brunswick, NJ: Transaction.

Gabriel Rossman is an Assistant Professor of Sociology

at the University of California-Los Angeles and an

Alfred P. Sloan Industry Studies fellow. His research in-

terests include the production of culture, cultural capital,

and economic sociology. He is currently working on

several projects that track the diffusion of pop songs

between radio stations.

Nicole Esparza recently joined the faculty of the School

of Policy, Planning, and Development at the University

of Southern California after completing a Robert Wood

Johnson Scholar in Health Policy Research postdoctoral

fellowship at Harvard University. Her research interests

include organizational dynamics, urban inequality, and

economic sociology.

Phillip Bonacich, Emeritus Professor of Sociology at

University of California-Los Angeles, has long been

interested in network measures of status, power, and

centrality. His current work is on models of social

exchange in networks.

Rossman et al. 51

at PENNSYLVANIA STATE UNIV on March 5, 2016asr.sagepub.comDownloaded from