ArticlePDF Available

Abstract and Figures

Do the cultural works artists produce after receiving major awards change in character? As awards lessen the constraints artists typically face, we argue that award winners receive more opportunities, gain more autonomy, and are more likely to pursue unique creative paths. Empirically, we analyze the consequences of winning a major Grammy award, a high-profile (often status-shifting) honor in the popular music industry. Using a neural learning approach, we examine the subsequent artistic differentiation of albums of award winners from albums of other artists. We analyze whether the music styles and sonic content of post-Grammy albums of winners change, and whether they become more or less similar to the combined corpus of albums of other artists. In panel regression estimates, we find that after winning a Grammy, artists tend to release albums that stand out more stylistically from other artists. Surprisingly, artists who were nominated but did not win a Grammy became more similar to other artists than they were before the nomination. The findings suggest symbolic awards can regularly induce change and affect the heterogeneity of cultural products.
Content may be subject to copyright.
What’s Next? Artists’ Music After Grammy Awards*
Giacomo Negro
Emory University
giacomo.negro@emory.edu
Balázs Kovács
Yale University
balazs.kovacs@yale.edu
Glenn R. Carroll
Stanford University
gcarroll@stanford.edu
Forthcoming in American Sociological Review
*For helpful comments on the paper, we thank Noah Askin, Bertis Downs, Tim Dowd, Amir Goldberg,
Michael Hannan, Greta Hsu, Damon Phillips, Elizabeth Pontikes, Brian Reschke, Amanda Sharkey,
participants at the 2020 Nagymaros Conference, the 2020 EGOS Conference, and the 2017 Authenticity
Workshop at Northwestern University. Ron Harris (Emory) and Mason Jiang (Stanford) provided expert
research assistance.
2
What’s Next? Artists’ Music After Grammy Awards
ABSTRACT
Do the cultural works artists produce after receiving major awards change in character? As
awards lessen the constraints artists typically face, we argue that award winners receive more
opportunities, gain more autonomy and are more likely to pursue unique creative paths.
Empirically, we analyze the consequences of winning a major Grammy, a high–profile (often
status-shifting) honor in the popular music industry. Using a neural learning approach, we
examine the subsequent artistic differentiation of albums of award winners from those albums of
other artists. We analyze whether the music styles and sonic content of post–Grammy albums of
winners change, and whether they become more or less similar to the combined corpus of albums
of other artists. In panel regression estimates, we find that after winning a Grammy, an artist tends
to release albums that are more likely to stand out stylistically from other artists. Surprisingly,
artists who were nominated but did not win a Grammy became more similar to other artists than
they were before the nomination. The findings suggest that symbolic awards may regularly induce
change and affect the heterogeneity of cultural products.
Keywords:
Awards, cultural production, status, differentiation, music, Grammys, careers
1
Awards bestow honor on the achievements of individuals, groups, or organizations. A
common feature in many domains, public awards
1
carry special significance in fields of cultural
production where many regard them as acts of consecration that separate the great from the
merely good (Bourdieu 1991). Accordingly, sociologists have examined extensively the
individual, relational, and social structural factors that set the stage for consecration in these
fields. These factors include: demographic characteristics; distinct forms of valorization by
critics, peers, and consumers; institutionalization of genres; and resource mobilization from core
and peripheral positions (Allen and Lincoln 2004; Schmutz 2005; Kremp 2010; Lena and
Pachuki 2013; Cattani, Ferriani and Allison 2014; Dowd et al. 2021; Bledsoe 2021; Schmutz and
van Venrooij 2021).
Little, if any, systematic research has examined the behavioral consequences to artists of
winning a major award. What do award winners do after consecration? Do winners embark on a
journey of differentiation and innovation, or do they become stalwarts of current styles? The
question is potentially of great interest because any change in the behavior of award winners
(presumably among the most visible artists) can affect the entire field, directly or indirectly.
Award winners garner respect, set trends, and become emulated by many other artists. Learning
the post–award fates of winners also promises to offer insight into the often–fraught relationships
between artists and their more commercially–oriented production partners and recording
companies. Winning an award gives an artist leverage to reduce potentially the long-lamented
creative constraints allegedly imposed by these partners. We seek to discover whether the lifting
of these constraints can lead to greater artistic innovation and novelty.
The general sociological question raised by awards concerns how factors internal to the
system of cultural production shape and enable cultural change over time. This internalist
2
approach – which Kaufman (2004) labels ‘ecological’ – focuses on explaining cultural
differentiation and innovation based on the relations of cultural producers to one another and
their social environment. We suggest that awards enhance the winning cultural producers’ status,
increase their leverage with commercial partners, and result in greater differentiation from others
in their field. Differentiation is achieved by means of chronologically ordered positions in the
field, akin to Bourdieu’s (1985) prises de position.
Award recipients experience significant social and economic benefits. The status
enhancement associated with a major public award grants greater visibility to the recipient
(Merton 1968; Kovács and Sharkey 2014; Reschke, Azoulay and Stuart 2018), as well as more
professional opportunity (Goode 1978). For instance, film actors who receive an Oscar award
tend to get increasingly better roles and higher pay (Faulkner and Anderson 1987). We argue that
such social and economic benefits bring into line artists’ interests, incentives, and motivation,
allowing them to differentiate more after an award. At the same time, the tradeoff between art
and commerce, and that between interests of artists and their production partners, can still
impose limitations on how awards increase post–award cultural differentiation. It is reasonable to
think that while winning an award reduces the material constraints an artist faces in subsequent
production, they do not entirely disappear.
In examining these questions, we study differentiation as contained in cultural products,
described as “the discrete and apprehensible human creations – songs, paintings, newspaper
articles, meals, sermons, laws, poems, scientific papers, garments – associated with fields of
cultural production(DiMaggio 2011: 288). For popular music, album recordings lay for decades
at the center of the production system as the field’s ‘extended textual units’ (Toynbee 2000). Just
as each cultural producer exists in relation to other producers in the field, cultural products exist
3
by virtue of their interdependence (Bourdieu 1983). Cultural products can be represented by sets
of features that position them vis–à–vis one another in a space of genre categories, a system of
aesthetic conventions in which cultural fields are embedded (Becker 1982).
We analyze five decades of albums made by thousands of professional music artists. We
seek to understand whether and how artists changed their music in response to receiving a
Grammy. Watson and Anand (2006: 55) describe how one recipient puts it, saying that a
Grammy confers significant prestige, “because it’s the only major award voted for by your peers
– you can walk down the street with the thrill of knowing that you are considered worthy by
other creative people in the business” (Watson and Anand 2006: 55). Receiving a Grammy in a
ceremony broadcasted world–wide ensures public visibility, commercial success, and career
longevity (Anand and Watson 2004).
We focus on the award’s effect on the subsequent artistic differentiation of the artist’s
albums relative to other artists. We measure differentiation in terms of distance of stylistic and
(technical) sonic content from the products of other artists. The Grammy selection process
identifies publicly all award nominees as well as winners. Prior to the award, the winners and
nominees appear to be generally of similar qualities, as we show below. The difference between
winning a Grammy and being only nominated can be regarded as a quasi–experimental
treatment: winners and nominees all experience social status enhancement and greater visibility.
By comparing the albums made by the two groups to other artists, and then comparing them to a
matched sample of non–nominated artists, we can see how artists generally respond to the public
honor and success emanating from the award.
The findings show that subsequent albums of Grammy winners embody features that are
more ‘distant’ from the combined work corpus of other artists. These distancing effects of
4
winners occur primarily with stylistic rather than sonic content. The finding suggests that
differentiation in cultural fields depends on artists constructing their positions in the field
through aesthetic choices collectively coded and recognized by audiences (Lena 2012; Godart
2018). When considering the post–nomination albums of all Grammy contenders, we find that
non–winning nominees become less stylistic distant from other similar, but unnominated artists
in their subsequent albums. This perhaps surprising result matters because for any award, there
are more nominees who did not win than who did. By implication, the award system apparently
exerts a chilling impact on artistic differentiation in a cultural field even though the intentions of
award sponsors are often the reverse.
Broadly conceived, this study aims to contribute to research on change in cultural
production. We consider the dynamics between producers and their associates in organizing
cultural production after consecration. In doing so, we refocus the analysis of cultural production
in artistic fields to cultural products, overcoming criticisms of Bourdieu for neglecting products
or treating them as epiphenomena (de La Fuente 2007; Prior 2011; Beljean, Chong and Lamont
2015). In the depiction presented here, cultural products are central to how artistic producers
construct their careers and their identities.
THEORY
Differentiation in Cultural Production
Differentiation drives the internal dynamics of cultural change. Bourdieu’s (1983) field
theory connects differentiation and cultural production, as cultural producers engage in a
‘struggle’ to accumulate recognition as symbolic capital. Cultural recognition is highly prized
because it can be converted into valuable economic resources and into power in the field. To
achieve recognition, producers may break with antecedents and distinguish themselves from
5
others through ‘position–takings’ – works, services, acts, arguments, and products. Every
position–taking is defined in relation to the space of other possible position–takings, and receives
its distinctive value from these relationships (Bourdieu 1983: 313).
In this account, competition for recognition varies across production systems, defined as
either restricted or largescale. In restricted fields, such as the fine arts, science, and poetry,
cultural recognition is accorded by peers, and production is grounded in a logic of ‘art for art’s
sake’ where producers act as creators relatively autonomous from commercial considerations. In
large–scale fields, such as mainstream film and broadcasting, production is organized to resonate
with consumer audiences and non–producers. Here cultural producers depend more on consumer
tastes as well as the organizations that control the means of production and distribution to
consumers, such as media companies or advertisers. Lower artistic autonomy in large–scale
fields is expected to encourage homogeneity and repetition rather than differentiation of
products.
The separation between the two types of production systems described in Bourdieu’s
original framework is far from sharp, however. Large–scale producers face unpredictable
consumer tastes and look at their restricted–scale counterparts to produce novelty and change.
Bourdieu (1985: 35) himself noted that commercial culture defines itself in relation to legitimate
culture and renews “its techniques and themes” by borrowing from high art aimed at other
producers. In addition, more open markets for cultural goods have weakened institutionalized
cultural authority (DiMaggio 1991). Among consumer audiences, omnivorous tastes (Peterson
and Kern 1996) and appeal for atypical products (Goldberg, Hannan and Kovács 2016) imply
that differentiation is positively valued beyond fields of restricted cultural production. For
6
instance, songs that sound too much like previous and contemporaneous music find less success
in the mass market (Askin and Mauskapf 2017).
The co–existence of restricted and large–scale production in the same field revises the
ideal–typical contrast between cultural fields (Beljean et al. 2015; Schmutz and van Venroij
2021). Specifically, it provides scope for producers to invest in diverse kinds of works through
specific stylistic and genre conventions. The co–presence of restricted and large–scale
production also brings the tension between artistic and commercial values into each field. In
doing so, the borrowing of ideas and practices by large-scale producers from their restricted
counterparts reinforces the cultural significance of artistic values over commercial values.
Schmutz (2016) analyzed critical discourse of popular music between 1975 and 2005 in the U.S.
and the Netherlands. He found that elite newspapers increased their coverage of commercially
successful artists; the majority of the coverage also adopted an aesthetic perspective (although
less so in the U.S.). These findings show the cultural pre–eminence of aesthetical legitimacy for
all producers including commercial producers, rather than reflecting the domination of
commercial constraints over artistic values (see also, Baumann 2001).
In sum, cultural producers can seek differentiation in restricted as well as large–scale
cultural fields. And, importantly, greater legitimacy tends to be attributed to differentiation
achieved through artistic autonomy vis–à–vis commercial success.
Post-Award Producer Differentiation
Awards honor outstanding achievements and enhance prestige. They attract public
attention and shape positively the impressions of others; awards also signify other, more
difficult–to–observe qualities of the producer. Status–enhancing awards incentivize future effort,
which may be channeled through novel behaviors (Frey and Gallus 2017; Malmendier and Tate
7
2009). We refer to the actions taken by producers to position themselves in a cultural field in this
direction as artistic differentiation strategies.
Producers seek to expand their symbolic capital as a way to preserve their authority in a
field. Awards can encourage more artistic differentiation because of the individual advantages
accrued. The benefits of status–enhancement from a major award grant greater access to
resources and opportunities. For example, academics receiving a prestigious award are
subsequently more likely to get grants, to receive teaching releases, to attract better students, and
to collaborate with productive co–authors (Chan et al. 2014).
As awards confer a more advantageous position in the field, a producer gains the ability
to pursue personal vocations, and aspirations. The resulting strategies of action do not need to be
fully conscious. A producer’s own capacities to think, feel, and respond to a situation are shaped
by their position in the field. Bourdieu’s (1980; 1985) notion of habitus describes a practical
sense learned through experience and internalization of the structures of one’s own social space.
Accordingly, producers develop their own ideas, expressions, and styles as regulated by patterns
attached to the positions they occupy. The existence of ‘pure’ works of art and ‘disinterested’
producers depend on the field’s acceptance of disinterestedness of the producer as a strategy to
accumulate recognition (Bourdieu 1983; Toynbee 2000).
2
In other words, producers previously
celebrated are expected to pursue unique artistic aims to maintain their advantage.
Lamont (2009) argues that engaging in cultural production offers valuable subjective
experiences beyond simply hoarding capital and imposing one’s position in the field. For
instance, pleasure and curiosity are alternative types of motivations for academics doing
scientific work, and these motivations may be used in artistic work following receipt of an
award. In psychology, award recognition for valuable work is thought to increase effort and to
8
foster the desire for more complex, creative work. Amabile (1993) demonstrates that even a
simple award (such as a plaque on the wall) motivates a recognized individual to engage more
deeply in activities considered intrinsically interesting. Awards also give people the feeling that
what they do is worth pursuing and not necessarily because of the position they occupy
(Eisenberger 1992). These analyses suggest a more ‘positive’ view, holding that strategies of
artistic differentiation can be explained by more than pure structural interest.
3
Cultural production is a collective, rather than individual effort. Cultural products
represent the joint work of the producer and the people and organizations that mediate exchanges
with the audience (Becker 1982). Cooperation is often not simple––different parties have distinct
aims and interests. Cultural production thus implies interdependence between multiple
participants, and a complex balance of power. In music, for example, stories abound about how
agents or recording companies dominate aspects of artistic decision–making and how this power
leads to less artistic or less original recordings (Toynbee 2000). For instance, Singular (1997:
112) describes famed record producer David Geffen:
“Throughout his years in the record business, Geffen was vulnerable to the charge that he
was more concerned with money than with music…critics said…He knew nothing about
making records…He was a mediator between talent and commerce, who’d seen a great
opportunity and seized it…he has no sense for what has meaning.”
Record labels, film studios, and publishers typically prioritize economic profits imposing
a pursuit of larger markets, leading to what Bourdieu (1980: 287) calls “devaluation entailed in a
mass appeal”. Awards provide leverage to artists to counterbalance the interests of these other
parties involved in the production system. They grant artists some power to negotiate less
utilitarian work, and more exploration with partners who control production resources. Winning
an award potentially lessens the constraints that an artist usually must operate under. In this way,
awards facilitate the process of artistic differentiation.
9
To summarize, we see artists who win major awards to be consecrated as eminent in their
field, and we expect them to seek and afford to pursue greater artistic differentiation in their
subsequent work. This leads to the first hypothesis:
Hypothesis 1: Artists differentiate their cultural products more after winning a major
award.
The hypothesis stands in contrast to alternative arguments in which winning can lead to
less differentiation because of the potential gains of replicating products that led to consecration
and success, or because awards free producers from the pressure of distinguishing themselves.
Our analysis instead highlights the role of contextual factors such as competition for recognition,
audience expectations for novelty, and individual interests, feelings, motivation, and incentives
that artists experience in engaging in cultural production. These factors suggest a connection
between (peer) recognition and producers pursuing more differentiated positions in the field. By
most scenarios, the hypothesis also implies that major awards provide enhanced resources to
winners for subsequent cultural production, and economic success for two reasons. First, the
enhanced resources provided to the artist show that the balance in the relationship with the
commercial partners has shifted, with the artist gaining resources through strengthened leverage.
Second, additional resources are likely necessary to support the artist’s creative discovery and
exploration, even after taking into account the different amounts of resources that various types
of products can require (differnces we control for statistically). While this empirical expectation
has been partially tested before in popular music (Watson and Anand 2006), we examine it again
with the data we have collected for this study.
Constraints on Differentiation
10
The relative distribution of opportunities for distinction shapes the impetus for
differentiation in cultural production (Kaufman 2004). We consider two factors that influence
this distribution and thus represent varying constraints to creative production. The first is the
balance between differentiation and other artistic strategies. The second factor we consider is an
organizational tradeoff. Each factor suggests an additional hypothesis.
With respect to strategic balance, it is perhaps obvious that commercial success can
weaken a producer’s disposition towards expanding symbolic recognition, consecration, and
prestige. Two types of arguments bear relevance. One holds that commercial imperatives
compromise and coopt artistic activity. Critical theorists Horkheimer and Adorno [2002(1947)]
claim that there is little possibility of avoiding the power of economic capital---artists who
experience success are challenged to find artistic autonomy and thus produce more homogenous
works. The other argument is less radical, and perhaps more realistic: it suggests that both
creative and commercial criteria may be under simultaneous consideration in cultural production.
Commercial success creates individual tradeoffs for artistic differentiation because there is
simply more at stake for each participant. The cultural producer who finds success with the
consumer audience develops an ambivalent attitude towards the market. As Bourdieu (1983:
330331) puts it, these artists become “torn between the internal demands of the field of
production which regard commercial successes as suspect […], and the expectations of their vast
audience, which are to some degree transfigured into a populist mission.”
Prior success also shapes personal perceptions as well as the appreciation and exercise of
artistic autonomy. For example, music artists assimilate the commercial logic of making music
when they work with producers who make commercial music (Toynbee 2000). After awards,
more successful producers will use their specific competencies to tackle the tension between
11
pursuing a wide variety of artistic aims while simultaneously taking on the expectations of the
market. From these arguments, we formulate a second hypothesis:
Hypothesis 2: Post-award artistic differentiation of cultural products is lower for
winning artists with greater prior commercial success.
Now we consider the organizational tradeoff. Cultural producers rarely are solitary
creators, particularly in cases where production is embedded in complex and bureaucratic
organizational processes. Managers of large firms seek efficiency of operations and design
cultural products to appeal to a broad audience hampering differentiation and innovation
(Godart, Seong, and Phillips 2020). They are less likely to invest in products deemed unlikely to
offer large sales, and place demands on artists to maximize the market potential of their products.
Musicians and other artists routinely complain about these pressures and the constraints they
place on artistic work (Toynbee 2000).
van Venrooij and Schmutz (2018) argue that smaller record labels follow a ‘professional
logic’ that imposes fewer market–based limitations and facilitates the production of less
conventional music. Larger labels follow instead a ‘commercial logic’ that shapes the sound of
artists in the mold of accepted genres and styles. The intent is to secure, among other things,
acceptance of the music by retailers, streaming services, etc. Following this logic results in
external constraints on artistic autonomy for artists and affords more power to the managers and
marketers who uphold established aesthetic conventions. For example, music artist George
Michael described his contentious relationship with major label Sony Records:
“[S]ince the Sony Corporation bought my contract along with everything and everyone
else at CBS Records, I have seen the great American music company I joined as a
teenager become a small part of the production line for a giant electronics corporation,
who quite frankly, have no understanding of the creative process…Musicians do not
come in regimented shapes and sizes, but are individuals who change and evolve together
12
with their audiences. Sony obviously views this as a great inconvenience” (Rule
1992:56),
Receipt of an award increases an artist’s visibility and signals their qualities to audiences.
Managers in large firms can see awards as providing opportunities to make the most of such
visibility and favor works that reduce market risk and increase profits. By contrast, managers in
smaller firms face smaller audiences and smaller risks and may be more willing to invest in
works which can bring mainly symbolic profits and the corresponding intellectual authority, at
least in the short term. From these arguments, we derive a third hypothesis:
Hypothesis 3: Post-award artistic differentiation of cultural products is lower for
winning artists making these products with largescale organizations.
CONTEXT
Popular Music and the Grammys
Popular music combines features of large–scale and restricted cultural production (van
Venrooij 2011; Schmutz, 2016; Schmutz and van Venrooij 2021). Music creation can bring
together artists and audiences in arenas that are not fully ‘commodified,’ such as jazz players out
of regular session work (Lena 2012). However, the production and distribution of recorded
music can be capital intensive activities with high fixed costs. Artists and record labels make
money from selling a large number of records to the widest possible audiences and other revenue
streams associated with making and distributing music (e.g., concerts, streaming). Audience
tastes are difficult to predict and success is uncertain (Caves 2000). The uncertainty stimulates a
constant demand for new products and rewards strategies that spread the risk of market failure
across different offerings. Toynbee (2000: xxi) sees music-making as the intersection of: (1)
artists’ dispositions to play, write, record, and perform in a particular way; (2) the pattern of
positions taken by all artists in the field; and (3) the works through which the artists
13
communicate the musical practices, textual forms, and codes. Music artists are “designers and
assemblers who take pieces of what is already heard and recombine them” (Toynbee 2000: xiv).
Some critics consider the tensions between art and commerce a cliched argument. But
Negus (1995: 325) argues that for participants in music production “these ideas are part of the
way in which they make sense of what is happening to them.” Frith (1981:61) noted that
economic success for artists is important because “failing to sell records and reach audiences
through the medium of the market means failing as a musician.” However, commercial pressure
for sales can conflict with artistic ambitions and artists who become very successful risk
commodifying themselves. Toynbee (2000: 32) argues that the industry has embraced the cult of
the author: “When audiences demand that music makers are creators, the music business must
guarantee minimum conditions of independence for them.
Music production involves the mobilization of various resources owned and controlled by
distinct participants. These participants include agents, managers, promoters, producers, and
record labels, each of whom contributes unique resources (Roy and Dowd 2010). Cooperation
between them is not simple because different parties often have different aims and interests. For
example, record labels produce, distribute, and promote an artist’s music. Large labels control
abundant resources but the supply of artists competing for them is even more abundant. The
distribution tips the balance strongly in favor of the labels in their dealings with the artists, which
results in less autonomy for artists (Toynbee 2000).
The collective work of making music takes place within a system of aesthetic
conventions (Becker 1982). In music, such conventions include genres and styles (Negus 1995).
Rather than being based only on formal musical properties or sonic qualities alone, genres and
styles are influenced by social relations, identities, characteristics of producers and audiences,
14
and production technologies. These influences include ways of playing an instrument, the use of
voice, forms of expression and details of presentation (van Venrooij and Schmutz 2015). Artists
attempt to find their place in the field by positioning their works within this system of
conventions.
In popular music, multiple audiences provide cultural legitimacy (Schmutz and van
Venrooij 2021). The Grammy awards represent the most significant form of recognition from
peers in music. The Grammys were established in 1959 by the National Academy of Recording
Arts and Sciences (NARAS). The intent was to honor musical accomplishments (“high artistic
achievement”) of entertainers in the industry. The NARAS credo reads that “sales and mass
popularity are the yardsticks of the music industry…they are not the yardsticks of this academy”
(Schipper 1992, in Watson and Anand (2006: 43)). Until 2020, those firms registered with the
NARAS or individual members (artists and other professionals in the industry) could submit
recordings for consideration for a Grammy. A screening committee of more than 150 experts
examined each submission and determined whether it was entered in an appropriate category.
4
Voting members chose entries for award nominations, and the five entries that received the most
votes in each category comprised the final list of nominees. A subsequent round of balloting
among members determined the winners in each category. In voting, NARAS members were
instructed not to be influenced by sales, personal friendships, or other extraneous factors.
The Grammy promoters proclaim “disinterest in commerce while enabling commercial
exploitation that comes from improved artistic reputation” (Watson and Anand 2006: 54–55).
Participation in tournament rituals such as the Grammys is a “privilege endowed upon influential
social actors in an organizational field and an instrument of status contests among them” (Anand
and Watson 2004: 60). When we interviewed the manager of a Grammy-winning major rock
15
band, he said: “Anything that can build awareness is a good thing for the artist. […] Winning is
better than to be nominated but the awareness increases in both cases.” He added: “Winning [a
Grammy] is an experience that can change the artist […]. The artist’s response is also sensitive
to how you make people notice; it means a lot to the creator.” Grammys provide the opportunity
to make and sell more records influencing decisions about the type of music that gets produced,
distributed, and consumed.
DATA
The statistical analysis we report combines data from multiple sources about the music
albums of all nominees, winners, and other artists not nominated for Grammy awards who were
active in the market during the analysis period (1959-2018) and appeared in the data sources.
The first source is the list of Grammy awards compiled by the NARAS (grammy.com).
We collected data for the nominated artists and winners of the four major general–field awards.
Artists refer to the main performer – an individual such as Carole King or a group such as
America as credited on the recording.
5
Each year the Grammys system considers work
accomplished in the previous year. Our data cover the awards from 1959, the first year in which
the Grammy ceremony was held, to 2018. The NARAS considers four major awards as ‘general
field’ and does not restrict entry by genre: Album of the Year for the performer and the
production team of a full album; Record of the Year for the performer and the production team of
a single song; Song of the Year for the writer or composer of a single song; and Best New Artist.
Our analysis focuses on these four major and highly visible awards that have significant impact
on artists’ careers (English 2005).
The second data source is AllMusic (allmusic.com), an online music guide considered the
most extensive music database on popular music (Mount 2013). We collected data on the music
16
releases of all artists from the database. AllMusic offers descriptive and editorial content to
consumers, although it also provides data to online and traditional music stores. The content
includes basic information on artists and their released albums such as names, titles, year of
release, and production credits indicating the production team involved in the recording of an
album. The entries are curated collectively by a group of popular music historians, critics, and
passionate collectors. (We attempted to contact the editors at AllMusic to discuss their practices
of reviewing and music categorization but received no response to our inquiries.)
Our analyses focus on music albums. During the study period, albums were the most
important format of music recording, containing multiple tracks of similar quality and coherent
themes. We excluded other types of recordings: singles/EPs, compilations, videos, and re–
releases. AllMusic categorizes albums using genre and stylistic tags to describe the aesthetic
characteristics of the music.
Musical genres describe broad aesthetic categories, such as Blues, Country, Jazz,
Pop/Rock, and Rap. Styles include more specific categories ranging from “Experimental,” to
“New Wave” and “Punk.” Musicologist Allan Moore notes that genres characterize “what an art
work is set out to do” and refer to context of musical gestures, while styles characterize “how it
[the art work] is actualized” and the “manner of articulation” of musical gestures (Moore 2001:
441). Moore describes the relationship between musical genres and styles as a loose hierarchy of
styles within genres. In the data, many styles are used only for one genre, for example “British
Rock” only for Pop/Rock albums. Other styles are used for multiple genres, for example
“Fusion” for Jazz albums like Miles Davis’s Bitches Brew, but also Pop/Rock albums like Joe
Satriani’s Surfing with the Alien.
17
Table 1 lists the primary genres of the albums (N=21) in the AllMusic data and their
relative frequency. The number of genre categories in the data is small and stable with only Rap
as a new genre introduced during the study period; roughly 85 percent of the albums are
associated with a single genre, and the remainder have multiple genres (less than 0.5 percent of
these have five or more genres). The list of styles (N=832) is too long to include in the table, and
we report those styles that have at least 5,000 albums in the data. On average, each album is
associated with 2.5 styles.
[Insert Table 1 about here]
The third data source is Echo Nest/Spotify, an online music intelligence provider that
offers data access via Application Programming Interfaces (APIs).
6
We used this source to code
some sonic features of the music. Askin and Mauskapf (2017) used this source to measure sonic
attributes of songs in the Billboard charts. We followed the methodology described in their Table
1 and coded information on the same attributes (acousticness, danceability, energy,
instrumentalness, key, liveness, mode, speechiness, tempo, and valence) for the songs in albums
in the AllMusic database assigning a quantitative value for each feature.
Sonic fingerprint data summarize some technical attributes of a sound signal, what music
theorist Dannenberg (2010) calls ‘texture’. By contrast, styles denote the general impression or
intention provided by the music. Dannenberg (2010: 49) notes that “style, especially in popular
music, includes an important sociological component, so we should not expect style to be purely
a matter of how something sounds.” Styles describe more the activity of making sound including
behaviors, artistic practices, and social identities; styles reflect attributions established through
perceptions and social interaction between producers and audiences (DeNora 1997). More
generally, Godart (2018: 106) defines styles as “durable and recognizable patterns of aesthetic
18
choices.” Styles unite works of individual artists of a given place and era; they are observable
and codified in a social and historical context (Lena and Peterson 2008; Lena 2012).
The sonic attributes from Echo Nest/Spotify and the styles from AllMusic represent
distinct aspects of the music and we do not interpret them as measuring the same information
‘objectively’ vs. ‘subjectively.’ Echo Nest CEO Jim Lucchese described the sonic attributes as
‘machine listening’ of what people hear, and social and cultural attributes such as styles as
‘cultural analytics’ that provide editorial insight (Ransbotham 2015). Both types of features are
viewed as significant to represent music and taste profiles. Indeed, the correlation between the
two variables is positive but low (0.12). Research on music information retrieval finds that
context–based data tends to explain more accurately than sonic features how music classification
systems are organized (Wang, Li, and Ogihara 2010; Oramas et al. 2017).
The fourth data source is the industry publication Billboard. It publishes weekly charts
showing the popularity of recorded music in the U.S. We collected information from the
Billboard 200 chart, which has tracked the 200 most popular albums and extended plays
according to sales since 1967. (The chart was first published as a top–10 list in 1956 and became
a top–200 in May 1967.) Higher positions in the chart indicate larger sales and greater market
success for artists. The chart has social and cultural value, and the media often discuss the
albums and artists that debut or reach the top of the list.
AllMusic served as the core organizing data base for the analyses. In these data, we found
albums for all but one of the Grammy artists, and 99 percent of the albums in the Billboard
charts. We also found a total of 125,340 albums whose songs had sonic features information on
Echo Nest/Spotify. These 125,340 albums are the main dataset used for the analyses and to
19
generate the matched sample described next. Appendix A provides further details on the analysis
files.
Measurement
Outcome Variables. The main outcome variable measures the artistic differentiation of
an artist’s post–award music. Artistic differentiation is represented as the distance of an artist’s
albums from albums in the same genre(s) made by other artists over the previous three years.
We constructed three variables. The first measures distance based on stylistic as well as sonic
content. The second variable uses only stylistic content, and the third only sonic content. For
stylistic content, we used the style labels in AllMusic. For sonic content, we used the sonic
features provided by Spotify. The style–only and the sonic–only measures are special cases of
the first variable, incorporating only the relevant subsets of input information. All three measures
are based on a neural learning model estimated at the album level.
The neural learning network uses two kinds of information: (1) the set of styles
associated with an album and the genre(s) into which the album is classified; and (2) the sonic
fingerprint data. We constructed the outcome variables in two major steps. In the first step, we
used both the AllMusic style descriptors and the Echo Nest/Spotify sonic data to determine the
coordinates of the position of each album in the genre space. The classification layer outputs a
21–dimensional vector for each album’s location in the space of genres. These values are the
predicted probabilities that the album is classified in a given genre. As illustration, the location
of album A may be represented with a 1–by–21 vector such as A=(0.20,0.14,0,0,0, 0.23,
0.14,0.2,0,0, …, 0.04, 0.01). Interpreting this vector implies that the predicted probability that
album A is Pop/Rock is 0.20, that it is Jazz is 0.14, that it is Electronic is 0, etc. Similarly, album
B may be represented with a vector such as B=(0.32,0.12,0.02,0.05,0, 0.06, 0.04,0.02,0,0, …,
20
0.02, 0.08). These values are predicted probabilities and sum to 1. We opted to use a neural
learning algorithm to locate albums in the genre space because it is well–suited to combine input
information of different types (text information such as styles, continuous variables such as
tempo, and categorical variables such as key). We have no clear a priori expectation of the
functional form to use to combine the various types of stylistic and sonic information. A neural
learning network is flexible about handling and exploring different ways in which the various
styles and sonic features can be combined.
7
[Insert Figure 1 about here]
In the second step of variable construction, we used as input the coordinates of the
albums in the genre space to calculate the pairwise distances of all albums from each other. The
dimensions in a classification layer are orthogonal, and we calculated a simple Euclidean
distance between the vectors. For example, the distance between albums A and B is calculated as
sqrt[(0.2-0.32)^2+(0.14-0.12)^2+(0-0.02)^2+…+(0.01-0.08)^2]. Formally,
!"#$%
&
'()
*
+
,- &
'!.)!
*
"
"#
!$#
. With such pairwise album distance measures in hand, we then proceeded to
calculate the average distance between albums. For example, if a genre has albums A, B, and C,
and a new album D is released, then to measure the newest album’s average distance from the
previous albums, we calculated dist(D,<A,B,C>)=(dist(A,D)+dist(B,D)+dist(C,D))/3. The range
of each of the distance variables is between 0 and 1.4, with an average of 0.2. Figure 1 shows the
structure of the neural learning approach we followed. Appendix B provides more details.
[Insert Figure 1 about here]
To illustrate how the measure and its components contribute to represent an artist’s
music, consider the artist Jody Watley. Ms. Watley won the Best New Artist Grammy in 1987
based on her eponymous album. AllMusic classified it in the R&B genre with styles
21
“Contemporary R&B” and “Dance–Pop” (artistic distance = 0.330). Later, Ms. Watley arguably
broadened her creative boundaries beyond Dance. Her next albums included Intimacy and
Affection, labeled in genres R&B and Rock/Pop with styles “Dance–Pop” and “Urban.These
albums carved out a more unique voice for the artist, one focused on more introspective themes
and smooth sounds (distance = 0.490 and 1.004, respectively). Recently, Ms. Watley moved to a
personal blend of electronic club music in the “House” style in the R&B genre with Midnight
Lounge (distance = 1.088). Finally, she returned again to more conventional R&B with The
Makeover (distance = 0.494).
We also looked for effects assumed by the relationship between awards and
differentiation as measured by album sales charts, and resources in terms of production credits
for post–award albums. We analyzed the peak position of each album in the Billboard 200 album
chart. The chart is based on sales of albums in the United States both at the retail, and later,
digital level. We calculated the highest position reached by each album on the chart as a measure
of audience evaluation, and reverse–coded it for more intuitive interpretation (so that higher
chart positions indicate greater success). The Billboard chart ranks the 200 best–selling albums
in the U.S., and albums may not reach the lowest position. We coded the albums that did not
rank on the chart with a position of 201. We also used a log–transformation of the original peak
position variable to smooth differences between positions at the bottom of the chart. To account
for truncation of the chart position variable, we estimated the number of weeks the album spent
on the chart as an alternative measure of success.
We examined the amount of resources involved in the musical production of post–award
albums. For this analysis, we measured the number of distinct production credits associated with
each album of an artist. Production credits (from AllMusic) list everyone who makes a
22
significant contribution to the creation of a music album. Credits include creative as well as
technical inputs including producers, mixers, engineers, backup musicians, and songwriters. The
number of entries in the production credits reflects the level of resources used to make the music.
Consistent with previous research, we expect that following receipt of an award an artist
will be given more resources to make their music, and that more resources will be reflected in
more production credits. The albums in the data have an average of ten credits; the average
declined slightly over time (the correlation between number of credits and calendar year is -
0.07). Variation between genres suggests some differences in production systems: Electronic
albums have the lowest average with six credits, and Holiday and Stage & Screen the highest
with 32. In the middle range are Classical and Folk with 15 credits, and Pop/Rock with nine. The
models include dummies for each genre of the album to control for differences in production
resources across these categories.
Covariates. The main explanatory variable codes the awarding of a major Grammy. The
Grammy awards studied cover the four general–field categories: Album of the Year, Record of
the Year, Song of the Year, and Best New Artist. The annual Grammy ceremony for the
recordings of any given year (e.g., 2018) is typically held in February of the following year (e.g.,
2019). The analysis is conducted on albums released by the artist after the award, and we
calculated a running lagged count of Grammy wins. When we expand the analysis to include
non–nominated artists, we also created a similar count of Grammy nominations.
Table 2 presents descriptive data about the Grammy–winning and Grammy–nominated
artists. Between 1959 and 2018, a total of 1,036 distinct artists received nominations. The
majority of these artists (over 60 percent) obtained only one nomination in their career. The artist
with the highest number of Grammy nominations in the four major general field categories is
23
Frank Sinatra, who received 22 nominations and won four awards. 278 artists received at least
one Grammy. Of them, only 28 percent won more than one award. The artists who won the most
Grammys – seven each – in the major general field categories are Adele and Paul Simon.
A second covariate measures prior commercial success. We calculated the variable as the
lagged average peak position in the Billboard 200 chart for each artist. The variable is reverse–
coded to simplify interpretation. A positive coefficient implies a positive association with
success in the market. We included this variable and interaction terms with the Grammy
variables to test the second hypothesis that artistic differentiation increases less with an artist’s
prior commercial success.
[Insert Table 2 about here]
The third covariate considers the influence of type of record label on artist behavior. It
relies on a distinction between major labels with global promotion and distribution networks, and
independent labels with relatively limited reach (Dowd 2004; Lena and Pachucki 2013). Major
record labels jointly account for about twothirds of the market share of the U.S. music market.
The list of major labels currently includes: Universal Music Group, Sony Music Entertainment,
Warner Music Group, and their associated subsidiaries. These companies show distinct
production and distribution strategies (Dowd 2004). We coded the major label variable as a
dummy equal to one if an album was released by a major label or an associated subsidiary, and
zero otherwise.
8
We included this variable and interaction terms with the Grammy variables to
test the third hypothesis that artistic differentiation will increase less if an artist’s albums are
released by larger companies.
9
We included a set of control variables in the analysis. First, we included calendar time to
account for temporal trends that correlate with macro factors that can influence cultural
24
production such as technology, and industry structure (Peterson and Anand 2004). Music
production is partitioned in genres, and genres vary in types and social trajectories (Lena and
Peterson 2008). To account for such heterogeneity, we included dummies for each of the 21
primary music genres in which each album is classified by AllMusic. Genres also change over
time, and we included as controls the interaction terms between each music genre and calendar
time. The Billboard 200 chart started in 1967, so our estimations with this data cover the period
1967–2018.
Our analyses use fixed effects estimators, and to avoid perfect collinearity of individual
variation for the calendar time and experience variables, experience was log–transformed. Table
3 contains the summary statistics and correlations for the variables in the main analysis.
[Insert Table 3 about here]
Matching for Comparisons
Our interest resides in the possible post–award differentiation in the recordings of artists
winning a major Grammy award. Our analysis draws on two comparisons. To establish internal
validity of the award effect, one comparison focuses on winners of a Grammy and other artists
who were nominated but did not win the prize. To establish external validity, another comparison
includes the combined sample of Grammy nominees and winners, and a sample of non–
nominated artists that we matched on observable characteristics.
We leveraged some features of the Grammy award process as a quasi–exogenous status
shift. In the ideal empirical analysis, we would compare a Grammy winner’s post–award albums
to those of the same artist had they not been nominated or won the award. This counterfactual
cannot be observed, so we considered a plausible empirical proxy for the hypothetical outcome
without the status increase. A starting point would be to compare average post–award recordings
25
of winners to the average among all non–winning artists. This approach would provide a valid
estimate of the treatment effect if assignment to the treatment group were random. However,
this assumption likely does not hold here.
An alternative approach is to compare winners to another group of potential recipients
who are similar along most important dimensions including merit (Malmendier and Tate 2009).
Nominees shortlisted for the award offer an appropriate matched sample for the winners (Kovács
and Sharkey 2014). Winning artists are often not necessarily better than the other nominated
artists. Ginsburgh and Weyers (2014) show this lack of differentiating quality in the Queen
Elisabeth piano contest in Belgium, in which the random order of appearance of musicians
influences the probability of becoming a finalist in the competition.
The data supports this assumption. In Appendix C, we compare the means of the outcome
variables and covariates before their first nomination or win (Table C.1). In all the t–tests, the
group of Grammy winners does not differ statistically from the group of nominees prior to the
award. It is noteworthy that artistic differentiation from other artists before the award does not
differ between winners and nominees. This addresses an alternative heuristic suggesting that: (1)
winners are rewarded because of higher (or lower) pre–award artistic differentiation than
nominees; and (2) nominees show greater conformity post–award because they follow the pre–
award artistic differentiation of winners. More specifically, Table C.2 also establishes that
albums that won a Grammy for Best Album are similar to those that were nominated but did not
win across a host of variables. Tables C.3–C.6 show similarity in artistic differentiation between
the two groups of albums for each of the other awards and conjointly. Note this does not mean
that the award selection system is unbiased. Rather, the data suggest that bias can occur earlier in
the nomination process than in the process selecting the winner.
26
We sought to establish a general comparison of award winners beyond other nominated
artists. The differences between nominated and non–nominated artists can be difficult to assess,
thereby confounding the counterfactual comparison between the groups of artists in standard
regressions. For instance, on average nominated artists at the time of first nomination show
higher artistic differentiation than non–nominees (0.392 vs. 0.324; p~0). To ensure a more
plausible comparison with Grammy–nominated artists, we use fixed effects estimators to control
for omitted variable bias due to unobserved heterogeneity. In addition, we constructed a matched
sample of non–nominated artists using “coarsened exact matching” (CEM), a nonparametric
method that reduces data covariate imbalance and increases the comparability of the units in the
sample (Iacus, King, and Porro 2012).
In the CEM procedure, units receiving a ‘treatment’ are matched to a ‘control’ group.
Treatment here means an artist received a Grammy nomination, and each artist–year dyad
represents the unit to match. The CEM procedure matches units in the two groups within the cut–
points for every covariate, and ensures that matched units have similar values. To improve the
comparison, we also included the lagged dependent variable as a matching covariate. CEM
reduces systematic differences in the composition of the groups. The procedure calculates the
imbalance statistic L1, a distance measure based on the difference between the multidimensional
histogram of all pretreatment covariates in the treated group and that in the control group. A
good matching solution would produce a reduction in its value overall and for each variable. Our
data show a substantial reduction in imbalance, not only in the means, but also in the marginal
and joint distributions of the data. Some imbalance remained for the experience variable. One
approach to deal with this common situation is to add the variable with imbalance as an
additional control to the statistical model. Accordingly, the regressions included a variable of the
27
experience of the artist in music, calculated as number of years since the artist’s first album
release.
Appendix D (Tables D.1–D.3) provides details about the covariates included in the CEM
matching and the matched dataset. The final number of observations in the main regressions
(45,012 albums for 36,808 artists) is the result of the dataset pruned from matching for the period
jointly covered by the multiple data sources (1967–2018). Table 4 summarizes the key details of
the analyses of the study.
[Insert Table 4 about here]
FINDINGS
Artistic Differentiation. As a test of the first hypothesis, Table 5 shows estimates of the
effects of Grammy awards on the artistic differentiation of an artist’s albums. We start with the
distance measure that uses both stylistic and sonic content. In Model 5.1, we restrict the analysis
to Grammy–nominated artists and included albums for these artists until they were nominated or
won a second award, if they did. This approach allows us to isolate the effect of the Grammy
count covariate from confounders by estimating the effect of a first–time win relative to a first–
time nomination. (As described earlier, the tests in Appendix C show that winning and non–
winning Grammy–nominated artists and albums are similar along the dimensions that the data
could measure.) Model 5.1 also excluded non–nominated artists because while the matching
procedure reduced the data imbalance between Grammy–nominated and non–nominated artists,
it also assumes that all the important matching characteristics have been measured. If some
characteristics related to artistic merit are not observable among the non–Grammy artists, then
the matching cannot compensate for these confounding factors. Model 5.1 includes 2,570
observations for 713 artists. With such a limited data panel structure, a fixed–effects model can
28
yield less reliable estimates, and the smaller number of observations (less than 2 percent of the
sample) suggested the use of ordinary least squares (OLS) regression. These estimates use robust
standard errors to mitigate specification error. In this model, we estimated a positive coefficient
of a Grammy win
&/ + 010223!4.56789 + 0100:*
, suggesting that winners increase their
differentiation in their subsequent albums after receiving the award. The finding is consistent
with the first hypothesis in the sample of Grammy–nominated artists.
[Insert Table 5 about here]
All additional regressions reported here included fixed effects for artists and for genres as
well as for interactions between genres and time, and the other control variables to isolate the
effects of winning and nomination from stable differences (e.g., demographic characteristics)
between artists. The estimations expanded the analysis sample to the full set of matched albums
of Grammy–nominated artists and the sample of unnominated artists.
In Model 5.2 we find that when an artist wins a Grammy, their subsequent albums show
greater artistic differentiation (in styles and sonic features) from the albums of other artists
&/ +
010;<3!4.56789 +010=*
. So, we find further support for the first hypothesis in the full
analysis sample.
Model 5.3 replicates the previous specification and also includes a covariate measuring
the running count of prior Grammy nominations that did not result in an award. This
specification intends to isolate the effect of winning from that of being nominated and also to test
whether albums of award contenders become more differentiated after being nominated, which
we would expect to a lesser extent. We continue to find that winning a Grammy is associated
with subsequently more differentiated albums
&/ + 010<23!4.56789>0*
. Surprisingly,
however, receiving a nomination but not winning the award shows subsequent lower
29
differentiation
&/ + .010;;3!4.56789>0*
. These findings support the first hypothesis for
winners but not for non-winning nominees in the full matched data sample.
To illustrate the negative differentiation effect for nominees, we point to Grammy
nominee Charlie Byrd. Mr. Byrd was nominated for Record of the Year (Desafinado) and Album
of the Year (Jazz Samba) in 1963. His blend of Jazz and International (especially Brazilian)
genres later became popular with the term “bossa-nova,” also one of the styles used by AllMusic
for Jazz Samba in addition to “Brazilian Jazz,” “Brazilian Traditions,” “Samba,” and “World
Fusion”. This album was praised for its artistic merit, and its artistic distance value is 1.273. In
the years after Jazz Samba, bossa-nova became part of the mainstream in North American music.
In the next twelve years, Mr. Byrd released a number of albums in the same genres and styles or
with minor changes including Bossa Nova Pelos Passaros (distance = 0.336), Traveling Man
(0.391), Byrdland (0.396), Hollywood Byrd (0.380). These albums have the same genres and
styles except for “Bop” instead of “Samba”. Another album of his, Great Guitars (0.289)
substituted “Guitar Jazz” and “Bop” styles for “Samba” and “World Fusion”. Accolades for
these albums suggest that Mr. Byrd continued to combine classical and bossa nova guitar with
few exceptions to his usual playing style.
These estimated equations include a host of controls. We see a negative and significant
coefficient for the experience variable measured as log–years since first album. In additional
analyses not shown in detail here, we examined the stability of these effects, whether the effects
of awards are subject to a ‘decay’ over time. We do not find evidence supporting this idea. When
we re-estimated Model 5.3 adding interaction terms between the Grammy variables and the
number of years since an award win or nomination, we found that differentiation significantly
30
increases with years since winning
&/ + 0100;3!4.56789 + 010;*
and significantly decreases
with years since having been nominated
&/ + .0100:3!4.56789 + 010?*
.
10
In the next two model specifications, we test the effect of Grammy awards on artistic
differentiation based separately on stylistic content and sonic content. Model 5.4’s specification
parallels that of Model 5.3 but the estimation uses only the sonic information from Spotify. Here,
neither a Grammy win nor a Grammy nomination shows a statistically significant association
with sonic distance. (We return to these findings later.)
In Model 5.5, we measure artistic differentiation using only styles from AllMusic. In this
model, the same pattern of Model 5.3 appears and is statistically significant: (1) artistic
differentiation increases after winning a Grammy; and (2) non-winning Grammy nominations
show the opposite effect, lowered differentiation. Comparing Model 5.3 and Model 5.5, shows
greater explanatory power of the model measuring distance with styles only than of the one using
stylistic and sonic content (
@"+01:A!5$1010B*
. Accordingly, we report additional analyses
using artistic differentiation based only on styles.
[Insert Table 6 about here]
In Table 6, we test the second and third hypotheses. In Models 6.1 and 6.2, we included
the variable past commercial success and its interactions with the Grammy variables in the main
model specification (4.5). The argument is that awards lead artists along different creative paths
but past success creates tension between economic and artistic interests. We find a negative
significant coefficient for commercial success, and the main effects of the Grammy variables are
unchanged (6.1). The interaction of past success with Grammy wins shows a significant negative
effect
&/ + .0100:3!4.56789 + 0100A*
, and the interaction with nominations is not
statistically significant &
/ + 01000;3!4.56789 + 01:;
* (6.2). Past success in the market
31
appears to weaken the artistic differentiation of award winners but leaves unaffected the
nominees. These estimates support the second hypothesis for Grammy–winning artists.
In Figure 2, we plot the adjusted predictions to compare the effects on stylistic distance of
winning a Grammy vs. not winning with past success ranging from the bottom of the chart to the
top. The linear prediction of stylistic distance decreases from .26 to .13 for non–Grammy
winners (dashed gray line), a 50 percent reduction. Stylistic distance decreases more, roughly 74
percent from .34 to .09, for artists who won one Grammy (black solid line). The graph also
shows lower stylistic distance overall when Grammy winners rank around the middle of the
chart, and declining further with more success.
[Insert Figure 2 about here]
In Models 6.3 and 6.4, we included the variable major record label and its interactions
with the Grammy variables. The estimates support the third hypothesis. The interaction between
major label and Grammy wins is negative and significant
&/ + .010<03!4.56789 + 0100:*
,
while the interaction with nominations is not significant
&/ + 0100=3!4.56789 + 01<=*
.
Albums released by award winners with major labels show less artistic differentiation.
Figure 3 shows the adjusted predictions of winning a Grammy and releasing an album
with a major label graphically. For artists who do not win a Grammy, the linear prediction of
artistic differentiation is similar for independent and major labels (0.258 vs. 0.253). Artistic
differentiation increases 29 percent to 0.333 for Grammy winners who work with independent
labels and 9.8 percent to 0.279 for Grammy winners who work with major labels. Model 6.5
includes both sets of interactions and shows the same pattern of the previous two specifications.
[Insert Figure 3 about here]
32
Billboard Rankings and Production Credits. We next analyzed the implicit empirical
expectations about the effects of awards on audience appeal and production resources. The data
in these analyses include Grammy winners, nominees and the matched group of non–nominated
artists. Table 7 presents the analysis of the relationship between Grammy awards and success in
the consumer market (cf., Peacock and Hu 2013). Model 7.1 contains artist fixed–effects
regressions for position of each album in the year of the Grammy award in the Billboard 200
album chart. (Peak position is reverse–coded.) Instead of the lagged running counts of Grammy
wins and nominations, we used a dichotomous variable equal to one for albums containing
Grammy–nominated or –winning music to better isolate the award effect (the awards are given
for music released in the previous year), and zero otherwise.
[Insert Table 7 about here]
The estimates in Model 7.1 imply that after a Grammy nomination, an artist’s album
lands roughly 22 positions higher in their peak performance in the Billboard chart (net of
controls). Winning the award implies a gain of 27 positions to the chart, although an F–test
indicates that the extra five positions are not a statistically significant advantage
&C&:(D:<B* +
010;D3!4.56789 +01<B*
. These estimates suggest that recognition from awards results in
economic benefits. In Model 7.2, we find the same pattern by replacing the reverse–coded peak
position with the negative of the log–transformation of the original peak position variable to
reduce differences between positions at the bottom of the chart. Model 7.3 estimates the log of
the number of weeks the album spent on the chart.
11
Grammy nominations and wins increase the
time on the chart.
12
We assume that this enhanced market power affords the artists involved
greater leverage in their relationships with commercially oriented partners such as record labels.
33
Table 7 also examines an implied effect of enhanced artist leverage, an expected positive
association between Grammy awards and the resources used in production of subsequent albums.
In Model 7.4, the outcome variable is the number of production credits and the covariates
measure again the running count of Grammy wins and nominations. The regression specification
includes artist fixed effects. We find that a Grammy nomination leads to an increase of almost
four additional production credits to an artist’s subsequent albums
&/ + =1A?D3!4.56789>0*
.
Winning a Grammy award results in approximately two additional credits in each subsequent
album, but the coefficient is not statistically significant. The average number of credits per album
is ten, and receiving an award increases the level of future production resources by about half, an
exceptional change in production resources.
13
These estimates demonstrate that awards provide
more resources for those nominated, an outcome consistent with the assumption that artists
earning awards gain leverage in their relationship with recording companies.
14
Further robustness checks. Finally, we examined the robustness of the Grammy awards
effects to a series of confounds and measurement artifacts. We determined that the effects of
awards measure distance in styles rather than: (1) making music simply categorized as more
‘mainstream’ or Pop/Rock (Regev 2015); (2) shifting the comparison set of genres to include a
wider set of artists and styles; (3) spanning of multiple genres; or (4) spanning of multiple styles.
We also found that stylistic distance represents a more general form of differentiation rather than
specific differentiation from certain subgroups of artists, or from an artist’s prior own music.
Specifically, we found that the main findings hold across various comparison groups: when
winners and nominees are compared to all other artists, to winners only, to nominees or winners,
or to non-Grammy winners or non-nominees, and when stylistic distance from one’s music is
controlled for. In unreported estimates, we also used stylistic distance from the artist’s prior own
34
albums as the outcome and did not find significant effects of Grammy wins or nominations,
suggesting that stylistic distance is primarily a differentiation from others. Finally, we also do not
observe bias in the relationship between awards and expert ratings (in AllMusic reviews) which
would suggest that stylistic distance simply reflects changes in perceptions of winners rather
than, as we argue, differences in the music as perceived by audiences. The findings are also not
sensitive to excluding any one of the four awards included, or excluding albums with fewer
styles or excluding albums in specific genres such as Classical where the music can recombine
works composed earlier in time. The details of these robustness tests are reported in Appendix E.
We also explore in Appendix F whether the genre and style measures show obvious signs of
endogeneity bias.
DISCUSSION
In the Introduction, we suggested that a promising (but perhaps uncanny) pathway to
studying change in a field of cultural products involves looking at how artists respond to status–
shifting awards. Winning an award gives an artist recognition and potential market power,
thereby offering leverage to offset the usual boundaries imposed by commercially-oriented
recording companies. Once these constraints are lessened, do award–winning artists become
more unique or less unique relative to other artists? The answer carries broad significance
because an award system that exerts systematic effects on the work of the more visible and
highly regarded artists will likely induce change on other artists and the entire field.
In this study, we investigated these issues in the field of popular music and the Grammy
awards. Consistent with our motivation, we found that the Grammys show significant effects on
artists’ subsequent creative strategies. In the main analyses, artists’ recorded music shifted after
Grammy nomination. For winners, music albums released after the award show greater stylistic
35
distance from the albums of other artists. By contrast, albums released after an artist is
nominated but does not win become stylistically closer to the albums of other artists. We also
found that market success increases for artists after they receive a nomination for a major
Grammy award, resulting in higher Billboard chart positions for their albums. Recognition by
the Grammys also results in enhanced resources for their future albums. Both effects imply that
artists gain leverage over recording companies through Grammy recognition.
Consider again now the chilling effect of non-winning award nominations on artistic
differentiation. How to explain it? Social psychologist Fritz Heider (1958:141) suggests that
award effects for winners might not extend to those shortlisted for the prize, pointing to negative
affective reactions of “near success” such as “exasperation, heightened frustration.” Indeed,
‘silver medalists’ often express disappointment over almost winning (Medvec, Madey, and
Gilovich 1995). And, employees who do not win a corporate award can feel they have no chance
of succeeding and become demotivated (Frey and Gallus 2017).
Other research also suggests a conformity response from award contenders. First, awards
provide not only recognition but also information about the performance and status hierarchy.
They signal to non–winners that their past strategies did not earn them the award (Neckermann
and Yang 2017). Non–winners can focus subsequently on actions relevant for winning the
award. For example, when the criteria for winning are not easily observable, non–winners can
follow how the award treated products in the past, and imitate more the features of the winners
(Rossman and Schilke 2014). Second, contenders receive information that their actions
represented a normative deviation. In response, they may update their beliefs and adapt their
choices. Even with unpredictable outcomes, contenders perceive that that their previous actions
deviated from the norm and will increase conformity (Hoogveld and Zubanov 2017).
36
Given typical award intentions of enhancing creativity and quality, the result that non–
winning nominations decrease subsequent artistic differentiation prompts the question whether
the music world be better off not publishing the list of nominees, as with awards such as the
Nobel? Such a change in the system would no longer incentivize conventional behavior by
shortlisted artists. On the other hand, publicizing a shortlist promotes album sales even if an
artist does not win.
It seems worthwhile to address the empirical finding that the sonic-based differentiation
of artists’ albums is not affected by a Grammy win or nomination. One possible reason for this
non-finding is that a pure sonic-based characterization, as captured by the Echonest/Spotify data,
is simply not rich enough to capture albums’ location in the artistic space. While sonic features
are important, styles are more general: styles incorporate information not only about the sonic
features but also aspects such as lyrical content, choice of instrumentation, playing approach,
aesthetic and political ideals (Toynbee 2000; Lena and Peterson 2008; Lena 2012). Because the
style data contains more information, it has higher predictive power---in our sample, we find
82% accuracy for genre classification based on styles, but only 73% accuracy based on sonic
features. This pattern is well-documented in research on music information retrieval and
classification. This research finds that context–based data tends to explain more accurately than
sonic features how music classification systems are organized (Wang, Li, and Ogihara 2010;
Oramas et al. 2017). In supervised learning tasks, distinct context–based approaches have higher
predictive power on average than sonic–content approaches (Turnbull et al. 2009). Audio
features also show more limited power in predicting preferences for songs or stream counts
(Nijkamp 2018). The findings are consistent with this research.
37
This study contributes to sociological research on cultural production, particularly what
Kaufman (2004) labeled the internalist, or endogenous, approach (see also, Abbott 2001). This
approach builds on Bourdieu’s field theory, but focuses on mechanisms of cultural dynamics
alternative to social structure and group interests, or economic factors. The theory also follows
field theory’s idea that differentiation results from the competitive process of position-taking by
producers in their field. We extend this account in four ways.
First, we examine the consequences, rather than the determinants of cultural consecration.
Consecration can be viewed an end-point of cultural production, while we argue that it also feeds
back on cultural production by triggering strategies of artistic differentiation among producers.
Second, the content of differentiation is modulated by the distribution of the award system. The
path of separation between consecrated and non-consecrated artists continues post-award for
winners. While the desire for novelty can direct audience attention toward new producers,
established producers innovate and pursue less conventional paths (Kremp 2010). The findings
show that the separation between winners and other artists applies to nominees who are also
consecrated yet revert to the conventional. Third, as a trigger for differentiation awards do not
exclude the role of structural interests (an award indeed enhances the winner’s status position);
rather, awards interact with structural interests as well as motivation, incentives, and affect to
shape strategies of cultural production (Beljean et al. 2015). Finally, the study builds on the idea
of correspondence between field positions of cultural producers and products: it develops a
methodology to examine an iterative process of differentiation of products that leads to long-
term differences among producers (Prior 2011).
The findings suggest to us two interesting ‘paradoxes’ for the analysis of cultural
production. First, prior market success limits differentiation of artists post-award. While
38
audiences expect novelty from producers, the support they provide to those same producers curbs
the same process that satisfies their needs or wants. Indirectly, the success that awards engender
inhibits subsequent differentiation. Recognition and success may become a curse of sorts for
cultural production. Future research might explore how artists themselves perceive the spoils of
the awards—how aware they are, how intentionally they plot their trajectories in the cultural
field, as well as how the position-takings of other artists who emulate them evolve.
Second, consecration leads to subsequent artistic work that may fulfill individual aims
but does not necessarily appeal to consumers or critics. Askin and Mauskapf (2017) showed that
atypicality of popular music songs increases consumer success up to a point. In Appendix G, we
also show that inclusion in lists of Best Albums compiled by influential music critics is affected
by Grammy nominations but appears unaffected by artistic differentiation. This pattern suggests
that award winners follow consecration by innovating in ways that usually do not meet new
acclaim. Yet moving away from the work that was consecrated may reinforce the winners’ place
among the greats. After the award, increased differentiation can reduce direct comparisons to an
artist’s previous work (and perhaps this is related to artists sometimes resisting being associated
with their earlier labels). In addition, other artists who follow in the winners’ steps usually
experience lower chances of recognition because certain aesthetic strategies have already been
rewarded. And in this context, nominees – who differentiate less than winners – may fare
relatively well: they do not receive full consecration for their work but receive positive audience
response because they do not differentiate as much. All these concurring effects suggest that the
overall impact of awards on the creative careers of winners, near-winners and other artists
deserves comprehensive examination.
39
Research on social status in markets shows multiple advantages of recognition for
winners, from greater public attention to outsized credit, increased productivity, and resources
that cumulate over time (Zuckerman 1967; Sauder, Lynn, and Podolny 2012; Kovács and
Sharkey 2014). In science, an individual’s rise in social status after a scientific prize can elevate
general interest in their domain, but it can also capture the attention that local audiences had
allocated to their domain neighbors. For example, neighboring scientific articles attract less
attention when authors of papers near them receive a prestigious prize (Reschke et al. 2018; Jin,
Ma, and Uzzi 2021). This is especially the case if the star is highly differentiated from others.
The “crowding out” effects documented in science can apply to cultural production such
as popular music; in fact, it may be an important factor involved in enduring phenomena such as
canonization. Canonization declares that some works or artists are of highest importance. It may
be that post-award differentiation makes the consecrated work more salient and contributes to
establish its value over time, as it will be separated from the prior work of other artists as well as
the winner’s. In fact, a canon lays claim to permanence but is not independent of time, place, and
context (Dowd et al. 2021). Future work on cultural production might examine canonization as a
social process involving diverse actors, products, institutions, and discourses that collaborate and
compete. This process is also clearly shaped by inequalities of access to power and discourse,
ideology, class, or gender.
The limitations of the present analysis should be recognized. First, while we focused on
the four major Grammy awards, many other awards obviously exist, Grammys and non–
Grammys. First, it would be useful to explore whether the findings generalize to settings in
which awards come with substantial financial gain. For instance, the Nomura Art Award and the
Nobel Prize come with million–dollar cash awards. We know that post–award motivation
40
associated with monetary prizes declines (Gubler, Larkin, and Pierce 2014). It may be that
awards result broadly in greater conformity including for winners. Conversely, it would be
relevant to examine symbolic awards in more restricted production fields where the economic
benefits of the award are more limited and post–award strategies towards more conventional
work may be less expected.
One example is the Pulitzer Prize for music, a highly prestigious award usually reserved
for critically acclaimed artists. We collected data on all Pulitzer Prize winners and shortlisted
artists, merged the data with the AllMusic dataset and replicated the analysis of our main model
(5.5). We found again that winners become more differentiated post–award, but the nomination
did not have any significant effect on subsequent albums for shortlisted artists. The estimates are
presented in Appendix H. While this is only initial evidence of this idea, it appears promising to
examine the general differentiation effects of awards as well as the ways in which variation in
type of award and field structure shapes creative work.
A second, and related limitation is that we studied only one cultural field. Even with
music’s vast impact, it comprises just one aspect of cultural life. Awards play a central role in
fields for books, movies, theater, dance, paintings, sculpture, architecture, politics and even
science. Do the patterns uncovered here occur in other cultural fields?
Finally, we could only observe subsequent producer strategies applied to products; we do
not see the cognitive and psychological processes that affect artists. While changes in artists’
stylistic positioning may plausibly be attributed to changes in artistic choices, the content of an
album can also be influenced by extraneous factors such as contractual obligations. Future
research on the music industry could delve deeper into how such factors influence musicians’
artistic choices, perhaps by studying closely specific cases.
41
REFERENCES
Abbott, Andrew. 2001. Chaos of Disciplines. Chicago: University of Chicago Press.
Allen, Michael P., and Anne E. Lincoln 2004. “Critical Discourse and the Cultural Consecration
of American Films.” Social Forces 82(3):871–894.
Amabile, Theresa. M. 1993. “Motivational Synergy: Toward New Conceptualizations of
Intrinsic and Extrinsic Motivation in the Workplace.” Human Resource Management Review
3(3):185–201.
Anand, Narasimhan, and Mary R. Watson. 2004. “Tournament Rituals in the Evolution of Fields:
The Case of the Grammy Awards.” Academy of Management Journal 47(1):59–80.
Askin, Noah, and Michael Mauskapf, 2017. “What Makes Popular Culture Popular? Product
Features and Optimal Differentiation in Music.” American Sociological Review 82(5):910–
944.
Baumann, Shyon. 2001. “Intellectualization and Art World Development: Film in the United
States.” American Sociological Review 66(3):404–426.
Becker, Howard S. 1982. Art Worlds. Berkeley, CA: University of California Press.
Beljean, Stefan, Phillipa Chong, and Michèle Lamont. 2015. “A Post-Bourdieusian Sociology of
Valuation and Evaluation for the Field of Cultural Production.” Pp. 38–48 In Routledge
International Handbook of the Sociology of Art and Culture, edited by L. Hanquinet, and M.
Savage. New York: Routledge.
Benner, Mary J., and Joel Waldfogel. 2016. “The Song Remains the Same? Technological
Change and Positioning in the Recorded Music Industry.” Strategy Science 1(3):129–147.
Bledsoe, Ashlee. 2021. “Walk the (Gendered and Racialized) Line: Retrospective Consecration
and the Rock and Roll Hall of Fame.” American Behavioral Scientist 65(1):59–82.
42
Borjas, George J., and Kirk B. Doran. 2015. “Cognitive Mobility: Labor Market Responses to
Supply Shocks in the Space of Ideas.” Journal of Labor Economics 33(S1):S109–S145.
Bourdieu, Pierre. 1980. “The Production of Belief: Contribution to An Economy of Symbolic
Goods.” Media, Culture and Society 2: 261–293.
Bourdieu, Pierre. 1983. “The Field of Cultural Production, Or: The Economic World Reversed.”
Poetics 12(4–5):311–356.
Bourdieu, Pierre. 1985. “The Market for Symbolic Goods.” Poetics 14(1–2):13–44.
Bourdieu, Pierre. 1991. Language and Symbolic Power. Cambridge, MA: Harvard University
Press.
Bourdieu, Pierre. 1998. Practical Reason. On the Theory of Action. Stanford: Stanford, CA:
University Press.
Bradler, Christiane, Robert Dur, Susanne Neckermann, and Arjan Non. 2016. “Employee
Recognition and Performance: A Field Experiment.” Management Science 62(11):3085–
3099.
Cattani, Gino, Simone Ferriani, and Paul D. Allison. 2014. “Insiders, Outsiders, and the Struggle
for Consecration in Cultural Fields: A Core–Periphery Perspective.” American Sociological
Review 79(2):258–291.
Caves, Richard E. 2000. Creative Industries: Contracts Between Art and Commerce. Cambridge,
MA: Harvard University Press.
Chan, Ho Fai, Frey, Bruno S., Gallus, Jana, and Benno Torgler. 2014. “Academic Honors and
Performance.” Labour Economics 31(December):188–204.
43
Dannenberg, Roger B. 2010. “Style in Music” Pp. 45–58 in The Structure of Style: Algorithmic
Approaches to Understanding Manner and Meaning, edited by S. Argamon, K. Burns, and S.
Dubnov Berlin: Springer-Verlag.
DeNora, Tia. 1997. Beethoven and the Construction of Genius. Musical Politics in Vienna,
1792–1803. Berkeley, CA: University of California Press.
de La Fuente, Eduardo. 2007. “The ‘New Sociology of Art’: Putting Art Back into Social
Science Approaches to the Arts.” Cultural Sociology 1(3):409–425.
DiMaggio, Paul J. 1991. “Social Structure, Institutions, and Cultural Goods: The Case of the
U.S.” Pp. 133–135 In Social Theory for a Changing Society, edited by P. Bourdieu, and J. S.
Coleman. Boulder, CO: Westview Press.
DiMaggio, Paul J. 2011. “Cultural Networks.” Pp. 286–310 in The Sage Handbook of Social
Network Analysis, edited by J. Scott, and P. J. Carrington. London: Sage.
Dowd, Timothy J. 2004. “Concentration and Diversity Revisited: Production Logics and the US
Mainstream Recording Market, 1940–1990.” Social Forces 82(4): 1411–1455.
Dowd, Timoty J., Trent Ryan, Vaughn Schmutz, Dionne Parris, Ashlee Bledsoe, and Dan
Semenza. (2021). “Retrospective Consecration Beyond the Mainstream.” American
Behavioral Scientist 65(1):116–139.
Eisenberger, Robert. 1992. “Learned Industriousness.” Psychological Review 99(2): 248–267.
English, James F. 2005. The Economy of Prestige: Prizes, Awards and the Circulation of
Cultural Value. Cambridge, MA: Harvard University Press.
Faulkner, Robert R., and Andy B. Anderson. 1987. “Short–Term Projects and Emergent Careers:
Evidence from Hollywood.” American Journal of Sociology 92(4):879–909.
44
Frey, Bruno S., and Susanne Neckermann. 2008. “Awards: A View from Psychological
Economics.” Zeitschrift für Psychologie/Journal of Psychology 216(4):198–208.
Frey, Bruno S., and Jana Gallus. 2017. Honours versus Money. The Economics of Awards.
Oxford: Oxford University Press.
Frith, Simon. 1981. Sound Effects: Youth, Leisure, and the Politics of Rock. New York: Pantheon
Books.
Ginsburgh, Victor, and Sheila Weyers. 2014. “Nominees, Winners, and Losers.” Journal of
Cultural Economics 38(4):291–313.
Godart, Frédéric. 2018. “Why is Style Not in Fashion? Using the Concept of “Style” to
Understand the Creative Industries.” Research in the Sociology of Organizations 55:103–
128.
Godart, Frédéric, Seong, Sorah, and Damon Phillips. 2020. “The Sociology of Creativity:
Elements, Structures, and Audiences.” Annual Review of Sociology 46:489–510.
Goldberg, Amir, Michael T. Hannan, and Balázs Kovács. 2016. “What Does It Mean to Span
Cultural Boundaries? Variety and Atypicality in Cultural Consumption.” American
Sociological Review 81(2):215–241.
Goode, William T. 1978. The Celebration of Heroes: Prestige as a Social Control System.
Berkeley, CA: University of California Press.
Gubler, Timothy, Larkin, Ian, and Lamar Pierce. 2016. “Motivational Spillovers from Awards:
Crowding Out in a Multitasking Environment.” Organization Science 27(2):286–303.
Heider, Fritz. 1958. The Psychology of Interpersonal Relations. New York: John Wiley & Sons.
45
Hoogveld, Nicky, and Nick Zubanov. 2017. “The Power of (No) Recognition: Experimental
Evidence from the University Classroom.” Journal of Behavioral and Experimental
Economics 67(April):75–84.
Horkheimer, Max, and Theodor W. Adorno. and Max. 2002(1947). Dialectic of Enlightenment.
Stanford, CA: Stanford University Press.
Iacus, Stefano M., King, Gary, and Giuseppe Porro. 2012. “Causal Inference without Balance
Checking: Coarsened Exact Matching.” Political Analysis 20(1):1–24.
Jin, Ching, Yifang Ma, and Brian Uzzi. 2021. “Scientific Prizes and the Extraordinary Growth of
Scientific Topics.” Nature Communications 12: 5619.
Kaufman, Jason. 2004. “Endogenous Explanations in the Sociology of Culture.” Annual Review
of Sociology 30:335–357.
Kovács, Balázs, and Amanda J. Sharkey. 2014. “The Paradox of Publicity: How Awards Can
Negatively Affect the Evaluation of Quality.” Administrative Science Quarterly 59(1):1–33.
Kremp, Pierre–Antoine. 2010. “Innovation and Selection: Symphony Orchestras and the
Construction of the Musical Canon in the United States (1879–1959).” Social Forces
88(3):1051–1082.
Lamont, Michèle. 2009. How Professors Think. Inside the Curious World of Academic
Judgement. Cambridge, MA: Harvard University Press.
Lena, Jennifer C. 2012. Banding Together. How Communities Create Genres in Popular Music.
Princeton, NJ: Princeton University Press.
Lena, Jennifer C., and Mark Pachucki. 2013. “The Sincerest Form of Flattery: Innovation,
Repetition, and Status in an Art Movement.” Poetics 41(3):236–264.
46
Lena, Jennifer C., and Richard A. Peterson. 2008. “Classification as Culture: Types and
Trajectories of Music Genres.” American Sociological Review 73(5):697–718.
Malmendier, Ulrike, and Geoffrey Tate. 2009. “Superstar CEOs.” Quarterly Journal of
Economics 124(4):1593–1638.
Medvec, Victoria H., Madey, Scott F., and Thomas Gilovich. 1995. “When Less is More:
Counterfactual Thinking and Satisfaction among Olympic Medalists.” Journal of Personality
and Social Psychology 69(4):603–610.
Merton, Robert K. 1968. “The Matthew Effect in Science”. Science. 159(3810):56–63.
Moore, Allan F. 2001. “Categorical Conventions in Music Discourse: Style and Genre.” Music
and Letters 82(3):432–442.
Mount, Andre. 2013. Multimedia Review. AllMusic. Journal of the Society for American Music
7(3):359–361.
Neckermann, Susanne, and Xiaolan Yang. 2017. “Understanding the (Unexpected)
Consequences of Unexpected Recognition.” Journal of Economic Behavior & Organization
135:131–142.
Negus, Keith. 1995. Where the Mystical Meets the Market: Creativity and Commerce in the
Production of Popular Music. The Sociological Review 43(2):316–341.
Nijkamp, Rutger. 2018. Prediction of Product Success: Explaining Song Popularity by Audio
Features from Spotify Data. Bachelor Thesis. University of Twente.
Oramas, Sergio, Oriol Nieto, Francesco Barbieri, and Xavier Serra. 2017. “Multi-Label Music
Genre Classification from Audio, Text, and Images Using Deep Features.” Suzhou: 18th
International Society for Music Information Retrieval Conference.
47
Peacock, David E. and Gongzhu Hu. 2013. “Analyzing Grammy, Emmy and Academy Awards
Data Using Regression and Maximum Information Coefficient. 2013 IIAI International
Conference on Advanced Applied Informatics:74–79.
Peterson, Richard A., and Narasimhan Anand. 2004. “The Production of Culture Perspective.”
Annual Review of Sociology 30:311–334.
Peterson, Richard A., and Roger M. Kern. 1996. “Changing Highbrow Taste: From Snob to
Omnivore.” American Sociological Review 61(5):900–907.
Phillips, Damon J., and Ezra W. Zuckerman. 2001. “Middle–Status Conformity: Theoretical
Restatement and Empirical Demonstration in Two Markets.” American Journal of Sociology
107(2):379–429.
Prior, Nick. 2011. “Critique and Renewal in the Sociology of Music: Bourdieu and Beyond.”
Cultural Sociology 5(1):121–138.
Ransbotham, Jim. 2015. “Analytics in E Major.” MIT Sloan Management Review 56(3):
http://mitsmr.com/1BOgUXw.
Regev, Motti. 2015. “Cultural Globalization Pop-Rock and Musical.” in The Routledge Reader
on the Sociology of Music, pp. 201-210. Edited by John Shepherd and Kyle Devine.
Routledge: New York and London.
Reschke, Brian P., Pierre Azoulay, and Toby E. Stuart. 2018. “Status Spillovers: The Effect of
Status-conferring Prizes on the Allocation of Attention.” Administrative Science Quarterly
63(4):819–847.
Rossman, Gabriel, and Oliver Schilke. 2014. “Close, But No Cigar: The Bimodal Rewards to
Prize–Seeking.” American Sociological Review 79(1):86–108.
48
Roy, William G. and Dowd, Timothy J. 2010. “What Is Sociological About Music?” Annual
Review of Sociology 36:183–203.
Rule, Sheila. 1992. The Pop Life. New York Times, 25 November: 56.
Sauder, Michael, Freda Lynn, and Joel M. Podolny. 2012. “ Status: Insights from Organizational
Sociology.” Annual Review of Sociology 38:267–283.
Schipper, Henry. 1992. Broken Record: The Inside Story of the Grammy Awards. Secaucus, NJ:
Birch Lane Press.
Schmutz, Vaughn. 2005. “Retrospective Cultural Consecration in Popular Music.” American
Behavioral Scientist 48(11):1510–1523.
Schmutz, Vaughn. 2016. “Commercialization and Consecration: Media Attention to Popular
Music in the US and the Netherlands, 1975–2005.” Poetics 59(1): 82–95.
Schmutz, Vaughn, and Alex van Venrooij. 2021. “Harmonizing Forms of Legitimacy in the
Consecration of Popular Music.” American Behavioral Scientist 65(1):83–98.
Singular, Stephen. 1997. The Rise and Rise of David Geffen. Secaucus, NJ: Birch Lane Press.
Toynbee, Jason. 2000. Making Popular Music: Musicians, Creativity and Institutions. London:
Arnold.
Turnbull, Douglas, Luke Barrington, Mehrdad Yazdani, and Gert Lanckriet. 2009. “Combining
Audio Content and Social Context for Semantic Music Discovery.” Proceedings of the 32nd
International ACM SIGIR Conference on Research and Development in Information
Retrieval.
van Venrooij, Alex. 2011. “Classifying Popular Music in the United States and the Netherlands.”
American Behavioral Scientist 55(5):609–623.
49
van Venrooij, Alex, and Vaughn Schmutz. 2015. “Classification in Popular Music.” Pp. 799–804
In International Encyclopedia of the Social and Behavioral Sciences 2nd Ed. Vol. 3, edited by
J. D. Wright. Oxford: Elsevier.
van Venrooij, Alex, and Vaughn Schmutz. 2018. “Categorical Ambiguity in Cultural Fields: The
Effects of Genre Fuzziness in Popular Music.” Poetics 66:1–18.
Watson, Mary R., and Narasimhan Anand. 2006. “Award Ceremony as an Arbiter of Commerce
and Canon in the Popular Music Industry.” Popular Music 25(1):41–56.
Wang, Dingding, Tao Li, and Mitsunori Ogihara. 2010. “Are Tags Better than Audio Features?
The Effect of Joint Use of Tags and Audio Content Features for Artistic Style Clustering.”
11th International Society for Music Information Retrieval Conference, pp. 57–62.
Zuckerman, Harriet. 1967. “Nobel Laureates in Science: Patterns of Productivity, Collaboration,
and Authorship.” American Sociological Review 32(3):391–403.
50
ENDNOTES
1
While no standard definition exists, awards typically involve: (1) programs of public
recognition to winners; (2) the formal bestowal of something tangible, such as money or a
certificate or trophy; and (3) scarcity or competition, such that not everyone can win (Frey and
Neckermann 2008).
2
For Bourdieu (1998:77), interest is “to be there”, to be invested in the field, “to recognize the
game and to recognize its stakes”.
3
In economics, Borjas and Doran (2015) find that awards incentivize unconventional work. They
see this behavior as arising from a tradeoff between labor and leisure. Awards increase ‘wealth’
through the social value of the prize, which in turn stimulates more consumption of leisure over
labor. Borjas and Doran (2015) suggest that awards shift producer attention (‘cognitive mobility’
in their words) toward less conventional paths. For example, mathematicians who had received
the Fields Medal produced less of the pure mathematics that the medal was awarded for, and
frequently branched out into different areas from those they had pursued before, increasing their
consumption of more “enjoyable” work topics.
4
This particular selection and voting system was installed in 1995. This temporal change does
not alter the pattern of findings that we report. In the Discussion section, we address whether the
revision of the nomination process in 2021 could affect the findings of the study.
5
Our analyses use unique identification numbers as assigned by AllMusic. Our regressions
control for fixed artist characteristics but do not link cases when an artist is part of multiple
performing acts (e.g., Paul Simon, and Simon & Garfunkel).
6
In 2014 Echo Nest was acquired by Spotify and its data integrated, so we used Spotify’s Web
API (https://developer.spotify.com/documentation/web-api/).
51
7
In supplemental analyses, we used a multinomial logit regression approach to calculate the
distance variables. When we used these logit–based variables in our analyses, we found a pattern
of results similar to what we report (see Table B.1. in Appendix B).
8
Coding the label types reliably over time can be elusive. We coded the variable to the best of
our knowledge, based on research that analyzed strategic change in the music industry during
time periods in our data (Benner and Waldfogel 2016), and music industry webpages (e.g.,
https://en.wikipedia.org/wiki/Lists_of_record_labels).
9
Additional analyses indicate that the effect of the major label variable is not a spurious
association between artists changing label or changing label type from the previous albums.
10
We considered several alternative accounts for our findings. One account is regression to the
mean (Zuckerman 1967; Malmendier and Tate 2009). This account would suggest less
conventional work for winners after an award acknowledging their exceptional achievements.
Yet we find more--not less--differentiation in post–award creative work of winners. And we find
systematic asymmetries in creative strategies between winners and nominees who also were
selected for their exceptional achievements. Another account would suggest that reward systems
increase conformity of all candidates. Winners follow the same path because they are inclined to
reciprocate the honor bestowed on them. Contenders perceive that their past behavior did not
conform to the apparent norm in the group, and will conform more in the future (Bradler et al.
2016). Alternatively, one could expect both winners and nominees to conform less but for
distinct reasons, the winners because they achieved recognition, and the nominees because they
need to do something different to try again to win. Instead, we find differences between winners
and nominees. One final account is middle–status conformity (Phillips and Zuckerman 2001).
This argument would suggest less conventional behavior for those at the top (and at the bottom)
52
of the status ranking and more conventional behavior for those in the middle, which is consistent
with our findings. Here the analysis of audience effects shows that award nominees and winners
are separated by minor differences in recognition and in gains if at all, not whole categories.
Also, in Phillips and Zuckerman (2001) the risk of illegitimacy is a major reason why those with
middle status would be expected to conform. Here it is unlikely that award nominees can become
illegitimate by engaging in more creative, non–conventional behavior. Rather, in creative
industries producers are praised for their non–conformity.
11
One advantage of this variable relative to is that zero weeks is not an attributed value.
12
In unreported analyses we examined the relationship between Grammy awards and critical
response. We used data on the year–end top–40 critics list published by Village Voice from 1971
to 2018 based on a polling of hundreds of popular music critics (Schmutz and van Venrooij
2021). This list was published at the end of each year, typically after the announcement of the
Grammy nominations but before the winner’s selection. This timeline makes the causal path
connecting ratings and awards more unclear to model and the reported statistical associations
need be interpreted with caution. We find positive yet moderate correlation. Correlation is
slightly higher between albums first receiving Grammy nominations and then ending on the
Village Voice list (0.21) than for albums first on the Village voice list that win a Grammy (0.16)
suggesting that Grammys predict critics appeal more than the reverse. The same correlations
decrease as the same artists accumulate awards and nominations (0.06 and 0.02 respectively)
suggesting also that critics regularly add more artists to the list who are not yet consecrated.
These data suggest benefits from the Grammys that translate in popular demand and to some
extent critical appeal, and positive correlations between distinct forms of recognition (Schmutz
and van Venrooij 2021).
53
13
In unreported regressions, we explored whether Grammy awards are associated with more
complex productions, as measured by the number of distinct production roles in the credits (e.g.,
arranger, engineer, art director, etc.). We find that the number of roles – six on average –
increases by 20 percent with nominations
!" # $%&'() *+ , -./01234
, but not with wins.
14
Artistic differentiation does not appear to be the simple product of greater amount of resources
used in music production. While the resource effects for production credits do not differ
significantly between nominees and winners, the effects for artistic differentiation go in opposite
direction for the two groups.
54
Figure 1. Neural Learning Approach Used to Measure Stylistic and Sonic Distance
loudness
tempo
instrumentalness
...
P(Pop-Rock)
P(Jazz)
P(Electronic)
P(Blues)
...
acapella
acid
afro-beat
...
waltz
worldbeat
urban cowboy
P(Rap)
55
Figure 2. Average Marginal Effects of Winning a Grammy Award and Past Market
Success on Stylistic Distance
Figure 3. Average Marginal Effects of Winning a Grammy Award and Major Record
Label on Stylistic Distance
0 .1 .2 .3 .4
Linear Prediction
Bottom of the Chart Top of the Chart
Past Market Success on Billboard Chart
No Grammy Wins
1 Grammy Win
Adjusted Predictions
.26 .28 .3 .32 .34
Linear Prediction
Independent Label Major Label
No Grammy Wins
1 Grammy Win
Adjusted Predictions
56
Table 1. Albums in ALLMUSIC Archive, by Primary Genre and Primary Style
Primary
Genre
Percent of
total albums
Style*
Percent of
total albums
Avant–Garde
0.69
Adult Pop/Rock
1.06
Blues
1.34
Alternative Pop/Rock
4.36
Childrens
1.07
Brazilian Traditions
0.83
Classical
13.79
Chamber Music
5.67
Comedy/Spoken
0.59
Choral
3.04
Country
2.56
Christmas
0.70
Easy Listening
1.05
Club/Dance
5.10
Electronic
9.04
Concerto
3.28
Folk
2.41
Contemporary Pop/Rock
0.90
Holiday
0.25
Gospel
2.80
International
8.11
Hard Bop
0.91
Jazz
7.89
Heavy Metal
1.57
Latin
4.34
Indie Rock
6.10
New Age
1.60
Japanese Traditions
1.37
Pop/Rock
33.49
Keyboard
3.27
R&B
2.22
Latin Pop
0.89
Rap
4.64
Opera
2.45
Reggae
0.81
Orchestral
2.25
Religious
2.34
Post–Bop
0.97
Stage & Screen
0.82
Symphony
2.25
Vocal
0.95
Vocal Music
2.40
Total
100
Total
52.17
* Includes styles with at least 5,000 albums. Total number of styles is 832.
57
Table 2. Distribution of GrammyWinning and GrammyNominated Artists
Number of
Wins*
Number of
Artists
Number of
Nominations
Number of
Artists
0
738
1
612
1
220
2
192
2
44
3
86
3
16
4
58
4
10
5
28
5
4
6
16
6
1
7
13
7
3
8
4
1,036
9
9
10
4
11
5
12
4
13
0
14
2
15
1
16
1
17
0
18
0
19
0
20
0
21
0
22
1
1,036
* Reports number of wins among nominated artists (and includes wins = 0 for artists who are nominated
and do not win)
58
Table 3. Summary Statistics and Correlations of Variables Used in the Regression Analyses (N=45,012)
Variable
Mean
S.D.
1
2
3
4
5
6
7
8
9
10
1
Stylistic and sonic distance
0.256
0.161
2
Stylistic distance
0.297
0.2
0.501
3
Sonic distance
0.304
0.149
0.643
0.123
4
Peak position in Billboard 200
9.21
35.336
-0.035
0.141
-0.079
5
Number of production credits
12.642
20.86
0.029
0.293
-0.060
0.419
6
Grammy win
0.016
0.189
0.034
0.089
0.007
0.130
0.157
7
Grammy nomination
0.081
0.542
0.046
0.140
0.002
0.232
0.267
0.713
8
Past commercial success
6.559
20.737
-0.020
0.139
-0.072
0.591
0.400
0.239
0.372
9
Experience
1.023
1.377
-0.071
0.306
-0.062
0.212
0.403
0.111
0.206
0.289
10
Major record label
0.096
0.295
-0.021
0.150
-0.054
0.388
0.331
0.138
0.246
0.378
0.355
11
Time
2003
8.495
-0.061
-0.215
-0.012
-0.136
-0.070
-0.095
-0.160
-0.280
-0.085
-0.277
59
Table 4. Summary of the Regression Analyses in the Study
Outcome
Main covariate
Estimation method
Data
Analysis
Artistic differentiation
from other artists
(Hypothesis 1)
Lagged running
count of Grammy
wins
OLS with robust standard
errors across artists
Grammy-
nominated artists
from start of
recording career
to second
nomination
Table 5
(Model 5.1)
Artistic differentiation
from other artists
(Hypothesis 1)
Lagged running
counts of Grammy
wins and
nominations
Fixed-effects estimators
with robust standard errors
across artists
Matched sample
of Grammy-
nominated and
non-nominated
artists
Table 5
(Model 5.2-
5.3-5.4-5.5)
Artistic differentiation
from other artists
(Hypotheses 2 and 3)
Interaction terms
between lagged
running count of
Grammy wins,
lagged running count
of Grammy
nominations, and
prior commercial
success and major
record label
Fixed-effects estimators
with robust standard errors
across artists
Matched sample
of Grammy-
nominated and
non-nominated
artists
Table 6
Peak position in
Billboard (Assumption
for Hypothesis 1)
Dummy for albums
containing music for
which the artist won
or was nominated for
a Grammy award
Fixed-effects estimators
with robust standard errors
across artists
Matched sample
of Grammy-
nominated and
non-nominated
artists
Table 7
Number of production
credits (Assumption for
Hypothesis 1)
Lagged running
count of Grammy
wins and
nominations
Fixed-effects estimators
with robust standard errors
across artists
Matched sample
of Grammy-
nominated and
non-nominated
artists
Table 7
60
Table 5. Regression Estimates of Artistic Differentiation from Other Artists
Model 5.1
Model 5.2
Model 5.3
Model 5.4
Model 5.5
Variable
Stylistic &
Sonic
Distance
Stylistic &
Sonic
Distance
Stylistic &
Sonic
Distance
Sonic
Distance
Stylistic
Distance
Grammy win
0.077**
(0.024)
0.025*
(0.012)
0.029*
(0.014)
-0.010
(0.011)
0.035*
(0.015)
Grammy nomination
-0.031**
(0.005)
0.002
(0.004)
-0.022***
(0.006)
Experience
-0.024***
(0.005)
-0.027***
(0.004)
-0.015***
(0.003)
-0.041***
(0.004)
Time
-0.004
(0.005)
-0.006
(0.005)
-0.006
(0.004)
-0.001
(0.005)
Primary genre
Included
Included
Included
Included
Primary genre X Time
Included
Included
Included
Included
Constant
0.434**
(0.006)
8.850
(10.850)
13.536
(9.983)
13.142
(7.470)
2.109
(10.333)
R2
0.01
0.09
0.17
0.05
0.16
Observations
2,570
45,012
45,012
45,012
45,012
Notes: Estimates are obtained with artist fixedeffects regression (Model 5.2, 5.3, 5.4, 5.5) and OLS (Model
5.1). Robust standard errors in parentheses. The data include Grammy nominees including winners (Model
5.1); Grammy nominees including winners, and a matched sample of nonGrammy nominees (Model 5.2,
5.3, 5.4, 5.5, 5.6). The matched group was selected using Coarsened Exact Matching (CEM) procedure.
Experience is logtransformed. Dummies for primary genre and for interactions between primary genre of
each album and time trend are included but not reported. *** p < 0.001, ** p < 0.01, * p < 0.05 (two
tailed).
61
Table 6. Regression Estimates of Artistic Differentiation from Other Artists – Effects of Commercial Success and Major
Record Labels
Model 6.1
Model 6.2
Model 6.3
Model 6.4
Model 6.5
Variable
Stylistic Distance
Grammy win
0.030*
(0.014)
0.080**
(0.023)
0.035*
(0.015)
0.075***
(0.019)
0.108***
(0.025)
Grammy nomination
-0.021***
(0.006)
-0.033**
(0.009)
-0.021***
(0.006)
-0.024***
(0.007)
-0.034**
(0.019)
Past commercial success
-0.001***
(0.0001)
-0.001***
(0.0001)
-0.001***
(0.0001)
Grammy win X Past commercial success
-0.001**
(0.0002)
-0.001**
(0.0002)
Grammy nomination X Past commercial success
0.0002
(0.0001)
0.0001
(0.0001)
Major record label
-0.008
(0.005)
-0.005
(0.005)
-0.004
(0.005)
Grammy win X Major record label
-0.050**
(0.016)
-0.043**
(0.016)
Grammy nomination X Major record label
0.004
(0.006)
0.003
(0.006)
Experience
-0.037***
(0.004)
-0.037***
(0.004)
-0.040***
(0.004)
-0.040***
(0.004)
-0.037***
(0.004)
Time
-0.001
(0.005)
-0.001
(0.005)
-0.001
(0.005)
-0.001
(0.005)
-0.002
(0.005)
Primary genre
Included
Included
Included
Included
Included
Primary genre X Time
Included
Included
Included
Included
Included
Constant
3.057
(10.318)
3.214
(10.315)
2.526
(10.335)
2.905
(10.328)
3.851
(10.312)
R2
0.16
0.16
0.16
0.16
0.17
Observations
45,012
45,012
45,012
45,012
45,012
Notes: Estimates are obtained with artist fixedeffects. Robust standard errors in parentheses. The data include Grammy nominees including
winners, and a matched sample of nonGrammy nominees. The matched group was selected using Coarsened Exact Matching (CEM) procedure.
Experience is logtransformed. Dummies for primary genre and for interactions between primary genre of each album and time trend are
included but not reported. *** p < 0.001, ** p < 0.01, * p < 0.05 † p < 0.10 (twotailed).
A62
Table 7. Regression Estimates of Peak Position in Billboard 200 Chart, Number of Weeks
Spent in Billboard 200 Chart, and Number of Production Credits
Model 7.1
Model 7.2
Model 7.3
Model 7.4
Variable
Peak Position
in Billboard 200
Peak Position
in Billboard 200
Number of
Weeks in
Billboard 200
Number of
Production
Credits
Grammy win
27.341**
(8.015)
0.486***
(0.131)
0.374**
(0.142)
2.108
(1.906)
Grammy nomination
21.767***
(4.067)
0.643***
(0.066)
0.376***
(0.072)
3.648***
(0.711)
Experience
-0.159
(1.766)
-0.001
(0.029)
-0.029
(0.031)
0.934
(0.588)
Major record label
5.649***
(1.068)
0.051***
(0.017)
0.112***
(0.019)
3.643***
(0.529)
Time
0.578
(1.078)
0.001
(0.018)
0.004
(0.019)
0.202
(0.376)
Primary genre
Included
Included
Included
Included
Primary genre X Time
Included
Included
Included
Included
Constant
-1141.962
(2156.116)
-7.649
(35.190)
-8.184
(38.282)
-389.791
(753.259)
R2
0.09
0.09
0.16
0.16
Observations
45,012
45,012
45,012
50,333
Notes: Estimates are obtained with artist fixedeffects. Robust standard errors in parentheses. In Model 7.1,
Peak position is reversecoded so that a positive coefficient implies a better position in the chart. In Model
7.2, Peak position is a negative logtransformation and a positive coefficient implies a better position in the
chart. In Model 7.3, Number of weeks in Billboard 200 is logtransformation of number of weeks that the
album spent in the chart (if the album charted multiple times, the variable measures the total number of
weeks spent in the chart), and a positive coefficient implies better performance. The data include Grammy
nominees including winners, and a matched sample of nonGrammy nominees. The matched group was
selected using the Coarsened Exact Matching (CEM) procedure. Experience is logtransformed. Dummies
for primary genre and for interactions between primary genre of each album and time trend are included but
not reported. *** p < 0.001, ** p < 0.01, * p < 0.05, † p < 0.10 (twotailed).
A63
APPENDIX A
Construction of Data Files
The music album dataset draws on three sources: AllMusic; EchoNest/Spotify; and,
Billboard. This appendix gives an overview of how these sources were combined to create the
files used in the analyses.
The AllMusic data contain information on recordings from multiple content pages
organized by: artist basic information; artist discography; record basic information; record styles;
and record credits. The file containing artist basic information was used as a master file into
which the other content was merged. The merging yielded a file with 1,264,644 recordings for
474,294 individuals/groups listed as primary artists. Recordings that were not albums
(EPs/singles, and compilations or videos), or were credited to ‘Various Artists’, soundtrack
records, studio cast records, karaoke records were excluded. This process produced a dataset of
1,035,743 albums for 421,572 artists. These albums also had release year information; the
release year ranged from 1938 to 2018. About 8 percent of the records were listed under multiple
artists (for example, Lulu is a collaborative album between Lou Reed and Metallica:
https://www.allmusic.com/album/lulu-mw0002223324). We collected the data in Fall 2018.
We submitted the list of these albums to the Spotify API, an interface that allows users
and programmers access to information about tracks, albums and artists. The query matched
125,340 albums (matched on artist name and album title) on which the artistic differentiation
measures were calculated. Spotify’s API provided sonic information for albums at the track level
and because we conduct analyses at the album level, we aggregated the track-level information
to the album level by taking the average sonic feature values of the tracks.
A64
The data for the albums in the Billboard 200 chart were collected from the Billboard
magazine archive. A total of 27,462 albums appeared on the Billboard 200 chart between 1967
and 2018. Of the total, 27,199 (99 percent) were found and merged in the AllMusic album and
artist dataset. The remaining albums in the analyses that did not reach the Billboard 200 chart
were assigned a position of 201.
Finally, the data for artists nominated for (and recipients of) Grammy awards were coded
from www.grammy.com. A total of 2,169 individuals/groups were identified. Of these, 1,037
show primary roles as performing artists with an album discography and all but one – Canadian
artist Bryan Adams who asked to be removed from the database
(https://rateyourmusic.com/discussion/music/why-is-bryan-adams-no-longer-on-
allmusic_com/1/) – were found in the AllMusic data. The other individuals/groups do not have
primary roles as artists and do not have album discographies to analyze (they still appear in
AllMusic as credited personnel such as producer, engineer, etc.).
A65
APPENDIX B
Application of the Neural Learning Model
We applied a neural learning model to represent albums in the genre space. This model
combines the sonic and stylistic features of the albums. The sonic content provided by Spotify
includes both continuous (such as tempo) and binary (e.g., minor/major key) variables, and can
be inputted directly to the algorithm. Ten pieces of information define the sonic fingerprint, so
the sonic content for each album can be stored in a 1 X 10 vector.
AllMusic provides a list of styles, which we converted to a numeric format. We did this
by turning the categorical style information into a binary format: for each album we coded a 1 X
832 sized vector of 0’s and 1’s (832 is the count of unique style labels in the data used for
estimation), where one denotes the cases where the album is assigned the style label, and zero
otherwise.
1
Combining the sonic and style content, the input vector comprises a 1 X 842-sized vector
for each album. For the measure using sonic information only, we used the 1 X 10 sonic vector
as an input. For the style information only measure, we used the 1 X 832 style vector.
Next, we included a “hidden layer” of size 842. This hidden layer, as illustrated in Figure
1, allows the neural network to capture the importance of any two-step interaction effects
between the input variables.
2
Finally, because we applied a supervised learning algorithm that
1
Styles are not strictly nested in genres. Styles can be diagnostic of genres. For example, the
“Symphony” style is quite likely a “Classical” music record. But styles can also be diagnostic of
multiple genres. In the example above, the style “Fusion” leads to high prediction of the genres
“Jazz” and “Pop/Rock.The algorithm learns common style combinations as well. For example,
“Trumpet”(style) + “Bop”(style) = “Jazz”(genre) but “Trumpet(style)” + ”Drums(style)” =
“Latin”(genre).
2
We experimented with including additional hidden layers. The prediction power did not
improve significantly, and we opted to present the simpler case with one hidden layer.
A66
predicts genres, the network outputs a representation of the albums in the space of genres that
maximizes the prediction power of a softmax function on genres. In other words, the neural
learning algorithm learns to weigh the sonic and stylistic vectors such that the predicted location
of the album in the genre space will be close (i.e., minimal distance) to the observed genres
assigned to the album. The neural network was trained in batches of 64 for 3 epochs, which
maximized out-of-sample fit without leading to overfitting. The network learning algorithm was
implemented using the keras package in Python 3.7. The detailed code is available from the
authors.
The final learned neural network provides a highly accurate representation of the albums:
using data on sonics and styles, it predicts the genre of the album with 84% accuracy. Using style
data only as an input, we achieve 82% prediction accuracy, while we achieve 73% prediction
accuracy with the sonic data only. These results illustrate that both sonic and stylistic
information could be useful in predicting album genres, but styles provide significantly higher
prediction power.
Alternative Approach: Multinomial Logit
In the main set of analyses, we use the albums’ location in the genre space as predicted
by the neural learning model. This model does not rely on a pre-specified functional form of the
relationship between styles or sonic data, and genre assignment; it also allows for multi-class
categorization (it can also handle albums classified in multiple genres, such as Jazz and
Pop/Rock).
One alternative to the neural learning model is multinomial logit regression. Multinomial
logits are limited to single-class categorizations (i.e., an album is either Jazz or Pop/Rock), and
must have the functional form pre-specified in the estimation equation (e.g., linear, quadrative,
A67
interactions, etc.). Because multinomial logit models are more commonly used in social science
research, we also investigated how a multinomial logit model would work in our setting, and
whether the results are robust to the distance measures calculated based on multinomial logit
estimates.
To estimate the multinomial logit model, we used those albums that have a single label
assigned to them (about 85% of all albums in our sample). In the regressions (which we
estimated using the mlogit command in Stata), each album is represented by one observation,
where the outcome variable is the observed genre assignment (e.g., Classical, Jazz, Rap,
Pop/Rock etc.), and the covariates are the values from the stylistic and sonic vectors. This
approach implies 842 covariates in the models that utilize both stylistic and sonic data. We enter
these covariates in a linear additive way, estimating the weights of each. Because of the
categorical nature of the dependent variable, the mlogit command estimates 21 equations for
each album (one for each possible observed genre). Formally, the model estimates regressions
such as the ones below (Greene 2018):
!"#$%&'()*+ ,$-- . /0
. 1#2!345%6!7 2"345%6"7 8 7 2#$"345%6#$" 7 2#$$39:;<=>;(!7 8 7 2#%"39:;<=>;(!&0
After estimating the model with these parameters, we calculated the predicted
probabilities of genre assignment of each album in each genre which, when combined, give a 1 X
21 vector. This vector is in the same exact format as the output of the neural learning model, and
we use these genre weights in the same ways as with the neural learning model to calculate
distances between albums. The average of these distances for each album was again used as the
outcome variable for the artistic differentiation analyses.
The predicted probabilities can be calculated for the whole set of albums (not only for
single genre albums). We also note that the predicted probabilities with the multinomial logit
A68
model and the neural learning models are highly similar, their pairwise correlation is 0.93. Below
we report the estimates of the main models (5.3-5.4-5.5) that use the artistic differentiation
measures obtained with the predicted vectors from the multinomial logit models. The pattern of
results is similar to what we report in the main text.
Table B.1. Regression Estimates of Artistic Differentiation from Other Artists –
Multinomial Logit Measurement
Model 1
Model 2
Model 3
Variable
Stylistic &
Sonic
Distance
Sonic
Distance
Stylistic
Distance
Grammy win
0.025†
(0.013)
-0.013
(0.011)
0.042**
(0.014)
Grammy nomination
-0.013*
(0.005)
0.003
(0.004)
-0.015**
(0.005)
Experience
-0.026***
(0.004)
-0.011**
(0.003)
-0.032***
(0.004)
Time
0.002
(0.004)
-0.004
(0.004)
-0.001
(0.005)
Primary genre
Included
Included
Included
Primary genre X Time
Included
Included
Included
Constant
-2.069
(9.833)
8.192
(7.896)
1.440
(9.828)
R2
0.10
0.04
0.11
Observations
45,012
45,012
45,012
Notes: Estimates are obtained with artist fixed-effects regression. Robust standard
errors in parentheses. The data include Grammy nominees including winners, and
a matched sample of non-Grammy nominees. The matched group was selected
using Coarsened Exact Matching (CEM) procedure. Experience is log-
transformed. Dummies for primary genre of each album and dummies for
interactions between primary genre of each album and time trend are included but
not reported. *** p < 0.001, ** p < 0.01, * p < 0.05, p < 0.10 (two-tailed).
A69
APPENDIX C
Comparisons of Grammy Winners and Nominees
Table C.1. Pre-Award Differences between Winners and Nominees
Winners
Nominees
P-value
Variables in Analysis
Differentiation from other artists
0.413
(0.014)
0.392
(0.010)
0.22
Differentiation from other artists stylistic content
0.435
(0.014)
0.419
(0.010)
0.36
Differentiation from other artists sonic content
0.309
(0.007)
0.299
(0.004)
0.22
Number of production credits
33.618
(0.924)
31.572
(0.642)
0.07
Peak position in Billboard 200 of artist’s albums
46.881
(2.508)
41.393
(1.643)
0.08
Experience as recording artist
6.633
(0.207)
7.039
(0.149)
0.11
Works with major record label =1
0.543
(0.017)
0.515
(0.011)
0.17
Other Variables
Maximum weeks in Billboard 200
45.482
(2.293)
44.513
(1.532)
0.73
AllMusic rating of artist’s albums
3.825
(0.028)
3.774
(0.020)
0.14
Village Voice Best Album list = 1
0.010
(0.003)
0.008
(0.002)
0.53
Elapsed time between albums (years)
1.422
(0.038)
1.453
(0.035)
0.54
Repeated collaborations
0.800
(0.074)
0.677
(0.043)
0.15
Network constraint
0.111
(0.006)
0.120
(0.004)
0.20
Number of genres of artist’s albums
1.446
(0.077)
1.282
(0.039)
0.07
Number of moods of artist’s albums
8.349
(0.240)
8.622
(0.162)
0.35
White artist = 1
0.617
(0.014)
0.645
(0.009)
0.10
Female artist = 1
0.182
(0.011)
0.180
(0.007)
0.89
Notes: Standard errors in parentheses. Values of p<0.05 indicate significant differences between the winners and
nominees. None of the comparisons between the two groups differs significantly prior to the award.
A70
Table C.2. Comparison between Grammy-Nominated and Grammy-Winning Albums, Best
Album Award
Winners
Nominees
P-value
Difference
Variables in Analysis
Artistic differentiation from other artists
0.397
(0.058)
0.329
(0.026)
0.29
Artistic differentiation from other artists stylistic content
0.429
(0.052)
0.448
(0.031)
0.76
Artistic differentiation from other artists sonic content
0.266
(0.014)
0.277
(0.008)
0.49
Number of production credits
73.302
(6.330)
61.969
(2.694)
0.07
Peak position in Billboard 200 of artist’s albums
11.628
(5.475)
5.942
(1.184)
0.10
Other Variables
Maximum weeks in Billboard 200
45.482
(2.293)
44.513
(1.532)
0.73
AllMusic rating of artist’s albums
4.500
(0.028)
4.419
(0.051)
0.40
Village Voice Best Album List = 1
0.310
(0.062)
0.185
(0.026)
0.06
Number of genres
1.345
(0.080)
1.339
(0.042)
0.95
Number of styles
4.035
(0.210)
3.872
(0.129)
0.51
Number of moods
16.069
(0.820)
16.128
(0.557)
0.97
Pop/Rock album
0.517
(0.066)
0.515
(0.332)
0.98
Notes: Standard errors in parentheses. Values of p<0.05 indicate significant differences between the winners and
nominees. None of the comparisons between the two groups differs significantly prior to the award.
Table C.3. Comparison between Grammy-Nominated and Grammy-Winning Albums, Best
Song Award
Winners
Nominees
P-value
Difference
Artistic differentiation from other artists
0.513
(0.065)
0.356
(0.027)
0.12
Artistic differentiation from other artists stylistic content
0.513
(0.064)
0.448
(0.030)
0.37
Artistic differentiation from other artists sonic content
0.266
(0.011)
0.293
(0.013)
0.11
A71
Table C.4. Comparison between Grammy-Nominated and Grammy-Winning Albums, Best
Record Award
Winners
Nominees
P-value
Difference
Artistic differentiation from other artists
0.431
(0.072)
0.407
(0.032)
0.77
Artistic differentiation from other artists stylistic content
0.477
(0.076)
0.500
(0.033)
0.78
Artistic differentiation from other artists sonic content
0.283
(0.027)
0.276
(0.010)
0.80
Table C.5. Comparison between Grammy-Nominated and Grammy-Winning Albums, Best
New Artist Award
Winners
Nominees
P-value
Difference
Artistic differentiation from other artists
0.378
(0.061)
0.369
(0.025)
0.89
Artistic differentiation from other artists stylistic content
0.433
(0.067)
0.410
(0.027)
0.72
Artistic differentiation from other artists sonic content
0.312
(0.030)
0.276
(0.010)
0.25
Table C.6. Comparison between Grammy-Nominated and Grammy-Winning Albums, Best
Album Award + Best Song Award +Best Record Award + Best New Artist Award
Winners
Nominees
P-value
Difference
Artistic differentiation from other artists
0.404
(0.018)
0.385
(0.018)
0.66
Artistic differentiation from other artists stylistic content
0.457
(0.038)
0.461
(0.019)
0.93
Artistic differentiation from other artists sonic content
0.286
(0.013)
0.284
(0.007)
0.85
A72
APPENDIX D
Data Matching
We implemented “coarsened exact matching” (CEM), a nonparametric method that
reduces data covariate imbalance and increases the comparability of the units in a sample (Iacus,
King, and Porro 2012). The CEM method was applied to artist-year observations. The CEM
procedure matches units in the two groups that are within the cut-points for every covariate,
ensuring that matched units have similar values. The covariates included in the CEM matching
were: dummy variables for each primary genre used by AllMusic to classify an artist; a dummy
variable for individual (vs. group) artist; the cumulative average of the number of genres for an
artist’s albums; the cumulative average of the number of styles for an artist’s albums; the
cumulative average of the number of moods for an artist’s albums; the cumulative number of the
number albums released with a major (vs. independent) label; years of tenure in the industry; the
cumulative number of albums that reached the top-10 positions in the Billboard 200 chart; and
the lagged artistic differentiation from other albums in the same primary genre in the previous
three years. In particular, matching on the lagged differentiation variable helps to account for
possible reversion to the mean effects. Artistic differentiation is not defined on a natural scale.
To improve covariate balance, in supplemental analyses we conducted an alternative matching
procedure using ten cut-points based on each decile of the distribution of this variable for the
matching. This alternative matching approach produced estimates that are similar to those
reported here.
Implementing CEM found matches for 62 percent of the observations for the treated
group, and led us to retain roughly 32 percent of the control group data. Table D.1 provides the
matching summary.
A73
Table D.1. Matching Summary
(Number of strata = 9,275; Number of matched strata = 943)
Control
Treated
All
122,998
2,342
Matched
42,920
1,454
Unmatched
80,078
888
The CEM procedure calculates the imbalance statistic L1, a distance measure based on the
difference between the multidimensional histogram of all pretreatment covariates in the treated
group and that of the control group. While the value of the L1 statistic does not have a specific
interpretation, a good matching solution reduces its value. In our data, the multivariate L1
distance is 0.9825884 before the matching, and .7919578 after the matching. These values
suggest that the matching method did indeed reduce the data imbalance.
Tables D.2 and D.3 provide more details about data imbalance before and after the
matching. Table D.2 reports the original L1 statistic before the matching, computed for each
variable used in the CEM procedure separately. The additional columns report univariate
measures of difference between treated and control units: means, and quantiles of the
distributions of the two groups for the minimum, 25th, 50th, 75th, and maximum percentiles for
each variable.
Table D.3 provides the same information after the CEM procedure. Multiple measures
are provided because balancing only the means between the treated and control groups does not
necessarily guarantee balance in the rest of the distribution. Table D.3 reports the same
information with the results of the matching. By comparing the imbalance results to the original
imbalance, we see a substantial reduction in imbalance, not only in the means, but also in the
marginal and joint distributions of the data. The experience variable continues to show some
A74
significant imbalance even after the matching. We sought to adjust for the remaining imbalance
by including this variable in the regression models.
The CEM procedure was conducted on the subset of records that were present in the
AllMusic and Spotify data. The final number of observations in the main regressions (N=45,012,
corresponding to 36,808 distinct artists) is the result of the dataset pruned from matching for the
period jointly covered by the three data sources (1967–2018) and for which covariate
information is available.
A75
Table D.2. Imbalance in Unmatched and Matched Samples Before CEM Matching
Variable
L1
Mean
Min
25%
50%
75%
Max
Artist’s genre:
Avant-garde
0
0
0
0
0
0
0
Blues
0.0085
0.0085
0
0
0
0
0
Childrens
0
0
0
0
0
0
0
Classical
0.0127
-0.0127
0
0
0
0
0
Comedy/Spoken
0.0082
0.0082
0
0
0
0
0
Country
0.08731
0.08731
0
0
0
0
0
Easy Listening
0.01298
0.01298
0
0
0
0
0
Electronic
0.00728
0.00728
0
0
0
0
0
Folk
0.01344
0.01344
0
0
0
0
0
Holiday
0
0
0
0
0
0
0
International
0.00507
0.00507
0
0
0
0
0
Jazz
0.03671
-0.03671
0
0
0
0
0
Latin
0.00018
-0.00018
0
0
0
0
0
New Age
0
0
0
0
0
0
0
Pop/Rock
0.19544
-0.19544
0
0
-1
0
0
R&B
0.12225
0.12225
0
0
0
0
0
Rap
0.03158
-0.03158
0
0
0
0
0
Reggae
0.00029
-0.00029
0
0
0
0
0
Religious
0.00724
0.00724
0
0
0
0
0
Stage & Screen
0.00516
0.00516
0
0
0
0
0
Vocal
0.06053
0.06052
0
0
0
0
0
Number of genres of artist’s albums
0.23609
0.26045
0
0
0
1
-3
Number of styles of artist’s albums
0.54278
1.7521
0
2
3
2
-3
Number of moods of artist’s albums
0.63857
5.4138
0
4
7
8
-35
Albums in Billboard 200 chart
0.5382
0.5382
0
0
1
1
0
Albums in top-10 positions in Billboard 200 chart
0.23605
0.23605
0
0
0
0
0
Experience as recording artist
0.57562
1.6642
0
2.1972
2.1401
1.3863
-0.04652
Albums with major record label
0.38808
0.38808
0
0
1
1
0
Group artist = 1
0.02929
0.02929
0
0
0
0
0
Lagged artistic differentiation of artist’s albums
0.32516
0.06304
0.12529
0.02641
0.05828
0.09469
-0.11531
A76
Table D.3. Imbalance in Unmatched and Matched Samples After CEM Matching
Variable
L1
Mean
Min
25%
50%
75%
Max
Artist’s genre:
Avant-garde
0
0
0
0
0
0
0
Blues
0
0
0
0
0
0
0
Childrens
0
0
0
0
0
0
0
Classical
3.3e-14
-4.e-140
0
0
0
0
0
Comedy/Spoken
0
0
0
0
0
0
0
Country
1.3e-13
-1.7e-13
0
0
0
0
0
Easy Listening
9.3e-15
-9.3e-15
0
0
0
0
0
Electronic
0
0
0
0
0
0
0
Folk
2.6e-15
-2.8e-15
0
0
0
0
0
Holiday
0
0
0
0
0
0
0
International
1.1e-15
-9.3e-16
0
0
0
0
0
Jazz
1.4e-13
-6.0e-14
0
0
0
0
0
Latin
0
0
0
0
0
0
0
New Age
0
0
0
0
0
0
0
Pop/Rock
7.3e-13
-9.9e-13
0
0
0
0
0
R&B
1.3e-13
-1.16e-13
0
0
0
0
0
Rap
3.8e-14
-4.3e-14
0
0
0
0
0
Reggae
0
0
0
0
0
0
0
Religious
2.6e-15
-2.8e-15
0
0
0
0
0
Stage & Screen
6.0e-15
-6.5e-15
0
0
0
0
0
Vocal
4.6e-14
-5.3e-14
0
0
0
0
0
Number of genres of artist’s albums
1.7e-13
-1.8e-13
0
0
0
0
0
Number of styles of artist’s albums
6.8e-13
-1.8e-13
0
0
0
0
0
Number of moods of artist’s albums
0.00864
0.0151
0
0
0
0
0
Albums in Billboard 200 chart
7.3e-13
-7.3e-13
0
0
0
0
0
Albums in top-10 positions in Billboard 200 chart
1.8e-13
-2.4e-13
0
0
0
0
0
Experience as recording artist
0.05019
0.00924
0
0
0
0
0.0378
Albums with major record label
7.3e-13
-7.0e-13
0
0
0
0
0
Group artist = 1
1.0e-12
3.5e-13
0
0
0
0
0
Lagged artistic differentiation of artist’s albums
3.0e-04
6.9e-05
0.03811
0
0
0
.
A77
APPENDIX E
Robustness Tests
To examine the robustness of the findings, we addressed some potential confounds for
the artistic differentiation effects (Hypothesis 1). One concern is that artistic differentiation does
not reflect making music different from other artists but simply music viewed as more
‘mainstream.’ This concern is twofold. First, it can shift the genres used to classify an artist’s
music. In particular, the label Pop/Rock may be used more as a generic descriptor post-award
and indicate that an artist’s music has lost some of its distinctive features of another genre
(Regev 2015). Second, if albums are more likely to be classified as Pop/Rock, then the measure
of artistic differentiation shifts the comparison set from one genre to others (say R&B albums to
R&B and Rock/Pop albums), which can include a wider (and more diverse) set of artists and
styles. Model 1 and 2 in Table E.1 used the model specification of artistic differentiation based
on style descriptors (Model 5.5) to estimate, respectively, the probability that an album is
classified in the Pop/Rock genre, and the probability that an artist’s album is classified in a
different genre (primary or secondary) from the prior album of the same artist. The models used
the Grammy win and nomination variables as main covariates and additional controls. The
estimations do not show statistically significant coefficients for the Grammy variables,
suggesting that the main effects reported in Model 5.5 are not due to an artist’s music shifting
closer to the Pop/Rock mainstream or a widening of the stylistic set of genres.
A related concern is that greater artistic differentiation of an artist’s music simply reflects
the spanning of multiple genres or styles. Models 3 and 4 in Table E.1 estimated the number of
music genres and styles in which an album was categorized. The models do not show significant
coefficients for the Grammy win and nomination variables, suggesting that awards exert effects
A78
on making music that is different from others rather than ‘generalist’ music that simply combines
multiple genre and style features.
We expected that awards would affect artistic differentiation from all other artists.
However, artistic differentiation can imply a move away from one’s own work too. To separate
these effects, in Table E.2, Model 1 re-estimated the main specification (5.5) and added a
variable measuring stylistic distance from the artist’s prior own albums. A significant positive
coefficient for this variable leaves the effects of the Grammy variables unaffected. This finding
suggests that creative paths following awards remain distinct for nominees and winners relative
to other artists, and that differentiation from one’s prior work also implies a shift away from
what others are doing. In unreported estimates, we used stylistic distance from the artist’s prior
own albums as the outcome and did not find significant effects of Grammy wins or nominations,
suggesting that stylistic distance is primarily a differentiation from others.
In Table E.2, we also examined whether this effect is more specific to other artists
consecrated by peers or the consumer public, as in a process that leads to differentiation based on
status (Models 2–7). We re-estimated the main specification (Model 5.5) using a measure of
stylistic distance limited to prior Grammy winners or nominees, and to Grammy non-winners
and non-nominees. We also estimated stylistic distance from artists who entered the Billboard
charts and those who did not. The estimates generally show coefficients consistent with the main
findings. These coefficients display statistical significance for all models for Grammy
nominations and three out of six for Grammy wins, perhaps due to the small sample on which
differentiation is calculated: significance is greater for differentiation from non-Grammy winners
and non-nominees (who are lower status, but also for differentiation from artists in Billboard
A79
(who are higher status). These findings suggest the presence of a more general form of
differentiation.
Next, we examined measurement artifacts in Table E.3. We re-estimated the main model
(5.5) to see whether the findings depend on any one of the specific awards included, and
excluded the variables for each of the four awards in sequence. The patterns are similar to the
main findings (but one coefficient in one of the models does not show statistical significance).
Bias in the estimates can also result from some albums having limited information about musical
styles to calculate stylistic distance. To check robustness, we replicated Model 5.5 including only
albums with at least two styles. The point estimates are similar and suggest no different
interpretation. Next, we excluded albums of classical music in the sample because these albums
often contain music composed before. While the selection of the compositions to include in a
classical album and their performance involves artistic work, such acts differ in some ways from
the production of original work in other genres (Toynbee 2000). The estimates show a similar
pattern to the previous models.
We also examined possible bias in the evaluation of genres, particularly Pop/Rock
albums, which can shape selection in the data or likelihood of categorization. We reasoned that if
artistic differentiation is simply a measure of change in reception rather than production, the
differences between winners and nominees could be explained by changing audience
preferences. These differences in turn could be reflected in evaluation bias by critics, in this case
the AllMusic experts. An indication of this effect could be observed in differences in evaluation
by genre.
In Table E.4. we examined whether critical ratings, number of stars from one to five in
the case of AllMusic, vary by genre. Using fixed-effects regressions with artist experience and
A80
year or release as additional controls and Pop/Rock as the reference category, we do not find
significant differences in ratings between genres with the exception albums in the Vocal genre.
In unreported analyses, we also explored the effects of interactions between the Grammy win
and nomination variables and genre dummies. The patterns do not differ from Table E.4.
Finally, in other analyses (details not reported here for brevity), we explored interactions
of the Grammy effects with time periods. We did not find evidence of moderating effects on
stylistic distance of interactions associated with: (1) the changes in the Grammy selection and
voting system; (2) the agreement between music distributors and retailers to affix stickers for
Grammy nominees and winners to improve music marketing campaigns; (3) the trends in audio
formats for music recording and reproduction (vinyl, cassette, CD, digital); or (4) the levels of
market concentration in the U.S. record music industry.
A81
Table E.1. Regression Estimates of Genre Classification
Model 1
Model 2
Model 3
Model 4
Variable
Pop/Rock
Album
Different
Genre
from Prior
Albums
Number of
Genres
Number of
Styles
Grammy win
0.244
(0.679)
0.517
(0.863)
0.052
(0.31)
0.032
(0.122)
Grammy nomination
0.157
(0.175)
-0.184
(0.200)
-0.015
(0.012)
0.028
(0.047)
Experience
-0.088
(0.164)
1.014**
(0.317)
0.028**
(0.009)
-0.472***
(0.036)
Time
-0.038**
(0.015)
-0.068
(0.94)
-0.031**
(0.011)
-0.020
(0.043)
Primary genre
Included
Included
Included
Primary genre X Time
Included
Included
Included
Constant
64.452**
(22.285)
39.501
(86.444)
Log likelihood
-635.855
-450.979
R2
0.08
0.39
Observations
2,082
1,815
45,012
45,012
Notes: Models 1 and 2 are fixed-effects logit regressions; Models 3 and 4 are fixed-effects regressions.
Robust standard errors in parentheses. The data include Grammy nominees including winners, and a
matched sample of non-Grammy nominees. The matched group was selected using Coarsened Exact
Matching (CEM) procedure. Experience is log-transformed. Dummies for primary genre of each album and
dummies for interactions between primary genre of each album and time trend are included but not reported
(Model 2, 3, and 4). *** p < 0.001, ** p < 0.01, * p < 0.05 (two-tailed).
A82
Table E.2. Regression Estimates of Artistic Differentiation from Other Artists
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Model 7
Variable
Stylistic
Distance
from All
Artists
Stylistic
Distance
from Other
Grammy
Winners
Stylistic
Distance
from Non-
Grammy
Winners
Stylistic
Distance
from Other
Grammy
Nominees
Stylistic
Distance
from Non-
Grammy
Nominees
Stylistic
Distance
from Other
Artists in
Billboard
Stylistic
Distance
from Other
Artists Not
in Billboard
Grammy win
0.036*
(0.014)
0.005
(0.029)
0.035*
(0.015)
0.029
(0.020)
0.035*
(0.015)
0.034*
(0.036)
0.014
(0.014)
Grammy nomination
-0.021**
(0.006)
-0.028**
(0.011)
-0.030***
(0.006)
-0.030***
(0.008)
-0.029***
(0.006)
-0.027***
(0.014)
-0.012*
(0.006)
Stylistic distance from
prior own albums
0.138***
(0.011)
Experience
-0.044***
(0.004)
0.011
(0.008)
-0.031***
(0.005)
-0.023***
(0.006)
-0.031***
(0.005)
-0.043***
(0.005)
-0.027***
(0.005)
Time
0.001
(0.005)
0.002
(0.010)
-0.007
(0.005)
0.003
(0.007)
-0.007
(0.005)
0.024
(0.006)
0.009
(0.005)
Primary genre
Included
Included
Included
Included
Included
Included
Included
Primary genre X Time
Included
Included
Included
Included
Included
Included
Included
Constant
-0.471
(10.232)
-3.505
(20.313)
15.382
(10.908)
-5.78
(14.203)
15.533
(10.893)
-46.766**
(12.203)
17.86
(10.257)
R2
0.18
0.01
0.48
0.23
0.47
0.34
0.11
Observations
45,012
45,012
45,012
45,012
45,012
45,012
45,012
Notes: Estimates are obtained with artist fixed-effects regression. Robust standard errors in parentheses. The data include Grammy nominees including
winners, and a matched sample of non-Grammy nominees. The matched group was selected using Coarsened Exact Matching (CEM) procedure.
Experience is log-transformed. Dummies for primary genre and for interactions between primary genre of each album and time trend are included but
not reported. *** p < 0.001, ** p < 0.01, * p < 0.05, † p < 0.10 (two-tailed).
A83
Table E.3. Regression Estimates of Artistic Differentiation from Other Artists
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Variable
Stylistic Distance
Excludes
Best
Album Award
Excludes
Best
New Artist
Award
Excludes
Best
Song Award
Excludes
Best
Record Award
Excludes
Albums with
Fewer than 2
Styles
Excludes
Classical Music
Albums
Grammy win
0.041*
(0.018)
0.034*
(0.015)
0.052**
(0.019)
0.026
(0.020)
0.032*
(0.014)
0.035*
(0.014)
Grammy nomination
-0.032***
(0.008)
-0.022**
(0.006)
-0.030***
(0.008)
-0.019*
(0.008)
-0.026***
(0.005)
-0.022***
(0.006)
Experience
-0.041***
(0.004)
-0.041***
(0.004)
-0.041***
(0.004)
-0.040***
(0.004)
-0.013
(0.007)
-0.040***
(0.004)
Time
-0.001
(0.005)
-0.001
(0.005)
-0.001
(0.005)
-0.001
(0.005)
0.001
(0.006)
0.002
(0.005)
Primary genre
Included
Included
Included
Included
Included
Included
Primary genre X Time
Included
Included
Included
Included
Included
Included
Constant
2.063
(10.330)
2.079
(10.334)
2.169
(10.333)
2.191
(10.339)
1.622
(11.091)
5.153
(10.443)
R2
0.16
0.16
0.16
0.16
0.21
0.17
Observations
45,012
45,012
45,012
45,012
12,274
44,226
Notes: Estimates are obtained with artist fixed-effects regressions. Robust standard errors in parentheses. The data include Grammy nominees (including
winners) and a matched sample of non-Grammy nominees. The matched group was selected using Coarsened Exact Matching (CEM) procedure. Experience is
log-transformed. Dummies for primary genre of each album and dummies for interactions between primary genre of each album and time trend are included but
not reported. *** p < 0.001, ** p < 0.01, * p < 0.05 (two-tailed).
A84
Table E.4. Regression Estimates of Critical Ratings#
Model 1
Variable
AllMusic Rating
Experience
-0.314***
(0.038)
Time
0.020***
(0.003)
Avant-Garde
-0.213
(0.249)
Blues
0.183
(0.202)
Children’s
0.157
(0.378)
Classical
0.294
(0.275)
Comedy/Spoken
0.161
(0.467)
Country
0.014
(0.123)
Easy Listening
-0.008
(0.343)
Electronic
-0.083
(0.265)
Folk
0.210
(0.165)
Holiday
0.004
(0.160)
International
0.001
(0.202)
Jazz
0.158
(0.136)
Latin
-0.410
(0.384)
New Age
-0.425
(0.743)
R&B
0.257
(0.173)
Rap
0.163
(0.786)
Reggae
0.255
(0.988)
Religious
0.188
(0.307)
Stage & Screen
-0.090
(0.182)
Vocal
-0.429*
(0.198)
Constant
-36.162***
(5.260)
R2
0.02
Observations
12,105
Notes: Estimates are obtained with artist fixed-effects regression. Robust standard errors
in parentheses. The data include Grammy nominees including winners, and a matched
sample of non-Grammy nominees. The matched group was selected using Coarsened
Exact Matching (CEM) procedure. Experience is log-transformed. # Omitted category is
Pop/Rock. *** p < 0.001, ** p < 0.01, * p < 0.05, † p < 0.10 (two-tailed).
A85
APPENDIX F
Endogeneity Checks
We examined possible bias that might arise from endogeneity in the classification of
genres and styles in the AllMusic data. A specific concern is that albums may be categorized into
a genre not as a result of the music an artist made, but because the artist or album was nominated
or won a Grammy.
First, we conducted additional analyses and modeled the probability that a genre is
assigned to an album by AllMusic using genre predicted by the sonic features from the
EchoNest/Spotify data, the Grammy win and nomination variables. If album categorization was
influenced by Grammys, then we would expect to see statistically significant coefficients for the
win and nomination variables. The estimates included additional controls. One such control is
number of genres. About 10 percent of the albums in the data have two or more genres, which
should increase the likelihood of any one genre to be assigned to an album. To estimate this
model, we expanded the album dataset to include all possible album-genre combinations.
The estimates in Model 1 in Table F.1 show that the probability of genre assignment is
significantly associated with genre predicted by sonic features and multiple genres, but not the
Grammy win or nomination variables. In Model 2, we excluded albums with multiple genres
rather than controlling for them and observed the same pattern of findings. In these models, we
do not find evidence of main effects of Grammy nominations and wins on genre assignments.
We conducted a related test predicting an album’s genre count as a function of the concentration
of the predicted album likelihoods from sonic features, the Grammy variables, and other
controls. The goal was to test whether winning albums were seen as more boundary spanning
than they actually are. We used a Herfindahl index to measure concentration of the likelihoods.
A86
For example, if an album is predicted P(Pop/Rock)=0.7, P(Jazz)=0.3, then the Herfindahl would
be 0.49+0.09=0.58. The estimates were obtained with negative binomial regressions, and did not
show evidence of Grammys influencing assigned genre counts.
We also examined the probability a style may be assigned to an album by AllMusic as a
function of the Grammy win and nomination variables. Similar to assigning a genre, if album
categorization was influenced by Grammys, then we would expect to see statistically significant
coefficients for the win and nomination variables. To predict styles, we could use a deep learning
model that predicts styles from sonic features, but we do not have enough data — the model
would have a very low prediction performance. Instead, we estimated a model with fixed effects
for style and additional controls. We used a linear probability model because a logit model of the
kind estimated for Model 1 and 2 could not converge. In Model 3, we present estimates that
show winning a Grammy and being nominated do not have significant effects on observing a
specific musical style.
A87
Table F.1. Regression Estimates of Observed Genre Categorization
Model 1
Model 2
Model 3
Variable
Observed Genre
Observed Genre
Observed Style
Grammy win
0.016
(0.064)
0.009
(0.067)
0.0001
(0.0001)
Grammy nomination
-0.017
(0.025)
-0.018
(0.026)
-0.0001
(0.0001)
Genre predicted
7.251***
(0.013)
7.293***
(0.014)
Genre spanning album
0.470***
(0.035)
Experience
0.026*
(0.011)
0.026*
(0.011)
0.0001**
(0.00004)
Time
-0.003
(0.011)
-0.004
(0.017)
2.90e-08
(6.20e-06)
Primary genre
Included
Included
Included
Primary genre X Time
Included
Included
Included
Constant
0.003
(0.012)
Log likelihood
-187,754.16
-179,434.67
R2
0.03
Observations
2,751,021
2,720,021
137,416,891
Notes: Estimates are obtained with artist fixed-effects regression. Robust standard errors in parentheses. The
data include Grammy nominees including winners, and a matched sample of non-Grammy nominees. The
matched group was selected using Coarsened Exact Matching (CEM) procedure. Experience is log-transformed.
Dummies for primary genre of each album and dummies for interactions between primary genre of each album
and time trend are included but not reported. *** p < 0.001, ** p < 0.01, * p < 0.05, † p < 0.10 (two-tailed).
A88
Next, we describe an earlier data collection effort that allowed us to examine whether and
how frequently the data was updated. For a separate earlier project, we conducted an initial data
collection from the AllMusic archive that coded information in the database as it appeared
between May and October of 2010. When we sought the data for this project in late 2018, rather
than merging old and new data, we collected the information from the AllMusic archive anew.
This situation gave us the opportunity to compare the old and new data, and to determine
whether the old data were later reclassified. We conducted two comparisons, one specific and the
other more general. The specific comparison was for the artists and albums nominated for the
2011 Grammy Awards. In 2010, we collected the data before the 2011 Grammy nominations and
selections (nominations take place between November and December, and winners are
announced in March of the next year). All the artists shortlisted for the four major prizes
analyzed in the study were included in the data. For this comparison, we did not observe any
case of reclassification of genres and styles for winners and nominees.
We also examined the re-categorization of Grammy winners and nominees in the whole
of the data. We found only one album of a Grammy-winning artist that was recategorized since
2010, and this album is Seal’s 2008 album Soul. Note that this specific album did not contain
music that was nominated or won a Grammy – Seal won two Grammys for Best Song of the Year
and Best Record of the Year but in 1995. So it seems unlikely that the recategorization of the
2008 album Soul was due to 1995 awards. Seal’s Soul album was categorized as Pop/Rock and
R&B in 2010, and in 2018 was only Pop/Rock. In terms of styles, the album had styles “Adult
Contemporary,” “Neo Soul,” and “Soul” and now was “Adult Contemporary” and “Adult
Contemporary R&B”. “Neo Soul” is considered a substyle of “Contemporary R&B”:
https://www.allmusic.com/style/neo-soul-ma0000004426. This single case of reclassification
A89
corresponds to 0.20 percent of the albums of Grammy winners, similar to the value for the whole
sample. For nominees, we observed a reclassification of seven albums or 0.6 percent of the
albums of Grammy nominees. Four are now Holidays and used to be R&B, Pop/Rock, or
religious. As mentioned earlier, the exclusion of albums in the Holidays genre produces a pattern
of findings similar to what is reported in the main analyses.
We also compared the data more generally beside the music that received Grammy
nominations of wins. We examined recategorization of all albums in the data and found that 0.18
percent of the albums (68 of them) showed a change in genre between 2010 and 2018. Table F.2
reports regression estimates of overlap in genre categorization for the albums with year of release
and genre dummies as covariates. We did not find any significant coefficient for the release year
dummies and for brevity we do not report them. For genres, we find one significant coefficient
for Holiday albums. (Avant-Garde was used as the omitted category.) Nine albums were
recategorized as Holiday: 4 were previously Pop/Rock, 3 R&B, 1 Religious, and 1 Vocal. While
the significant result for Holiday indicates that albums in this genre can introduce some bias in
the estimates, the very small number of cases also suggests that the bias is unlikely to change the
pattern of results we report. When we re-estimated our main model (Model 5.5) excluding
Holiday albums (N=49) the effects of the Grammy win and nomination variables remain very
similar to those reported in the main analysis
!" # $%$&'( )* + ,-./0 # $%$&1
and
!" #
+$%$2$( )* + ,-./03$1
.
A90
Table F.2. Regression Estimates of Genre Overlap Between First and Second Data
Collections
Model 1
Variable
Genre Overlap
Blues
-0.062
(0.035)
Children’s
0.0004
(0.001)
Classical
-0.005
(0.003)
Comedy/Spoken
-0.0004
(0.001)
Country
0.001
(0.001)
Easy Listening
-0.041
(0.030)
Electronic
-0.029
(0.016)
Folk
-0.015
(0.011)
Holiday
-0.561***
(0.123)
International
-0.024
(0.017)
Jazz
-0.001
(0.001)
Latin
0.001
(0.001)
New Age
0.0003
(0.001)
Pop/Rock
-0.001
(0.001)
R&B
-0.007
(0.004)
Rap
-0.0002
(0.001)
Reggae
-0.001
(0.001)
Religious
-0.079
(0.044)
Stage & Screen
0.001
(0.002)
Vocal
-0.005
(0.007)
Year dummies
Included
Constant
-36.162***
(5.260)
R2
0.09
Observations
38,236
Notes: Estimates are obtained with artist fixed-effects regression. Robust standard errors in parentheses. *** p <
0.001, ** p < 0.01, * p < 0.05, † p < 0.10 (two-tailed).
A91
APPENDIX G
Analysis of Critical Response to Grammy Nominations
We collected data on the year–end top–40 critics list published by Village Voice from
1971 to 2018. Based on a polling of hundreds of popular music critics (van Venrooij and
Schmutz 2021), this list is published at the end of each year, typically after the announcement of
the Grammy nominations but before the winner’s selection. We estimated a logit regression of an
album appearing on this list of best albums with Grammy nomination in any of the general
interest categories and stylistic distance as main covariates (additional controls include artist
experience, major record label release, time trend, primary genre dummies and their interaction
with the time trend). In Table G.1, Model 1 shows a positive and significant statistical
association between Grammy nominations and being listed among the best albums of the year.
The estimates show a positive but not significant association between stylistic distance of an
album and critics listing it among the best albums of the year. These findings suggest that while
attention and legitimation with the critics’ audience are positively associated with nomination for
an award, they are not necessarily associated with artistic differentiation post-award.
A92
Table G.1. Regression Estimates of Likelihood of Being Listed Best Albums of the Year by
The Village Voice
Model 1
Variable
Listed in Pazz & Jop
Grammy nomination
1.825***
(0.242)
Stylistic distance
0.320
(0.428)
Experience
0.272***
(0.061)
Major record label
0.842***
(0.143)
Time
0.035
(0.050)
Primary genre
Included
Primary genre X Time
Included
Constant
0.434**
(0.006)
Log likelihood
-377.402
Observations
38,750
Notes: Estimates are obtained with random-effects logit regression. Robust standard errors in parentheses.
The data include Grammy nominees including winners, and a matched sample of nonGrammy nominees.
The matched group was selected using Coarsened Exact Matching (CEM) procedure. Experience is log
transformed. Dummies for primary genre and for interactions between primary genre of each album and
time trend are included but not reported. *** p < 0.001, ** p < 0.01, * p < 0.05 (twotailed).
A93
APPENDIX H
Analysis of Pulitzer Prize for Music
Named after pioneer journalist Joseph Pulitzer, the Pulitzer Prize for Music was first
given in 1943 and is awarded for a distinguished musical composition “of significant dimension
by an American that has had its first performance in the United States during the year.” While
the prize includes a monetary component of fifteen thousand dollars, its significance is primarily
symbolic. The jurors for the award include past winners, music composers, academics, critics,
and other artists. Considered perhaps the highest achievement in musical excellence, the Pulitzer
Prize was typically awarded to Classical and Avant-Garde music. The definition and entry
requirements beginning with the 1998 competition were broadened to attract a wider range of
American music, particularly Jazz. In 2018, the prize was also awarded to the first Rap artist
(Kendrick Lamar).
In Table H.1 we replicate the estimations of the main analyses of artistic differentiation
(Models 4.2–4.5) using dummies for post-Pulitzer Prize nomination and win as covariates. The
prize was awarded in many instances to compositions that were not recorded or artists who had a
limited number of album recordings. The estimates are obtained with random-effects regressions.
We find similar patterns in the regressions with the Pulitzer Prize as well, ensuring a more
general validity of our findings that awards increase artistic differentiation for winners and
decrease differentiation for nominees.
A94
Table H.1. Regression Estimates of Artistic Differentiation from Other Artists Following
Pulitzer Prize in Music
Model 1
Model 2
Model 3
Model 4
Variable
Stylistic &
Sonic
Distance
Stylistic &
Sonic
Distance
Sonic
Distance
Stylistic
Distance
Pulitzer win
0.079*
(0.038)
0.122***
(0.018)
0.006
(0.019)
0.128***
(0.021)
Pulitzer nomination
-0.121***
(0.020)
-0.119***
(0.022)
-0.056*
(0.023)
Experience
-0.028
(0.043)
-0.021
(0.043)
-0.021
(0.045)
-0.022
(0.044)
Time
-0.003
(0.002)
-0.003
(0.002)
0.023
(0.031)
-0.004*
(0.002)
Primary genre
Included
Included
Included
Included
Constant
6.876*
(3.104)
6.979*
(3.164)
-3.887
(2.335)
7.473*
(3.275)
R2
0.76
0.77
0.12
0.63
Observations
61
61
61
61
Notes: Estimates are obtained with random effects regression. Robust standard errors in parentheses. The
data include Pulitzer nominees including winners, and a matched sample of non-Pulitzer nominees. The
matched group was selected using Coarsened Exact Matching (CEM) procedure. Experience is log-
transformed. Dummies for primary genre of each album are included but not reported.
*** p < 0.001, ** p < 0.01, * p < 0.05 (two-tailed).
REFERENCES IN THE APPENDIX
Greene, William H. 2018. Econometric Analysis. 7th Ed. Upper Saddle River, NJ: Prentice Hall.
Iacus, Stefano M., King, Gary, and Giuseppe Porro. 2012. “Causal Inference without Balance
Checking: Coarsened Exact Matching.” Political Analysis 20(1):1–24.
Schmutz, Vaughn, and Alex van Venrooij. 2021. “Harmonizing Forms of Legitimacy in the
Consecration of Popular Music.” American Behavioral Scientist 65(1):83–98.
Toynbee, Jason. 2000. Making Popular Music: Musicians, Creativity and Institutions. London:
Arnold.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Fast growing scientific topics have famously been key harbingers of the new frontiers of science, yet, large-scale analyses of their genesis and impact are rare. We investigated one possible factor connected with a topic’s extraordinary growth: scientific prizes. Our longitudinal analysis of nearly all recognized prizes worldwide and over 11,000 scientific topics from 19 disciplines indicates that topics associated with a scientific prize experience extraordinary growth in productivity, impact, and new entrants. Relative to matched non-prizewinning topics, prizewinning topics produce 40% more papers and 33% more citations, retain 55% more scientists, and gain 37 and 47% more new entrants and star scientists, respectively, in the first five-to-ten years after the prize. Funding do not account for a prizewinning topic’s growth. Rather, growth is positively related to the degree to which the prize is discipline-specific, conferred for recent research, or has prize money. These findings reveal new dynamics behind scientific innovation and investment. Scientific revolutions have famously inspired scientists and innovation but large-scale analyses of scientific revolutions in modern science are rare. Here, the authors investigate one possible factor connected with a topic’s extraordinary growth—scientific prizes.
Article
Full-text available
This review integrates diverse characterizations of creativity from a sociological perspective with the goal of reinvigorating discussion of the sociology of creativity. We start by exploring relevant works of classical social theory to uncover key assumptions and principles, which are used as a theoretical basis for our proposed definition of creativity: an intentional configuration of cultural and material elements that is unexpected for a given audience. Our argument is enriched by locating creativity vis-à-vis related concepts—such as originality, knowledge, innovation, atypicality, and consecration—and across neighboring disciplines. Underlying the discussion are antecedents (structure, institutions, and context) and consequences (audiences, perception, and evaluation), which are treated separately. We end our review by speculating on ways in which sociologists can take the discussion of creativity forward. Expected final online publication date for the Annual Review of Sociology, Volume 46 is July 30, 2020. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.
Article
Cultural consecration demonstrates the values of a field, and it is influenced by several different factors. Previous research on consecration in rock music has examined the impact of critical, professional, and popular legitimacy, as well as the gender of the performers, on the likelihood of albums being retrospectively consecrated. The study builds on and extends this research by examining the impact of these different forms of legitimacy and gender, while also examining the impact of performers’ race on (a) the likelihood of being inducted into the Rock and Roll Hall of Fame and (b) the wait-time that performers face in that induction.
Article
This review integrates diverse characterizations of creativity from a sociological perspective with the goal of reinvigorating discussion of the sociology of creativity. We start by exploring relevant works of classical social theory to uncover key assumptions and principles, which are used as a theoretical basis for our proposed definition of creativity: an intentional configuration of cultural and material elements that is unexpected for a given audience. Our argument is enriched by locating creativity vis-à-vis related concepts-such as originality, knowledge, innovation, atypicality, and consecration-and across neighboring disciplines. Underlying the discussion are antecedents (struc-ture, institutions, and context) and consequences (audiences, perception, and evaluation), which are treated separately. We end our review by speculating on ways in which sociologists can take the discussion of creativity forward.