the effect of attribute-based cultural networks on …...the effect of attribute-based cultural...
TRANSCRIPT
The Effect of Attribute-based Cultural Networks on
Selection Dynamics in Popular Music*
Noah Askin1 and Michael Mauskapf2
1INSEAD, Fontainebleau, France [email protected]
2Kellogg School of Management, Northwestern University, Chicago, IL [email protected]
ABSTRACT
In this paper we integrate insights from cultural sociology, network theory, and computer science to reexamine evaluation outcomes in popular music. Using web-based tools to construct a data set that distills songs’ musical content into a set of discrete attributes, we test whether and how these attributes affect a song’s performance on the Billboard Hot 100 charts. Our analyses suggest that cultural attributes matter, beyond the effects of artist familiarity, genre affiliation, and social influence. More specifically, we find evidence that cultural networks, or the relational patterns formed between songs’ with shared attributes, play an important role in determining songs’ popularity. Songs that sound too similar to contemporaneous songs tend to suffer, while those that are optimally differentiated are more likely to appear atop the charts. These results contribute to the growing literature on endogenous cultural effects, prompting us to reconsider some of the basic mechanisms that drive consumption behavior.
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!* For their valuable thoughts, comments, and support, we are grateful to Matt Bothner, Frederic Godart, Jenn Lena, John Levi Martin, Kevin Lewis, John Mohr, Klaus Weber, Alejandro Mantecón Guillén, The Echo Nest, and members of the University of Chicago’s Knowledge Lab and Northwestern University’s Institute on Complex systems (NICO).
!
2
INTRODUCTION
Over the last two decades, a small but growing group of scholars has worked to import and develop
new techniques to better understand patterns of cultural production and consumption (e.g., DiMaggio
1994; Mohr 1998; Weber 2005; Michel 2010; Goldberg 2011). These developments have been bolstered
by methodological advancements in computer science, which increasingly influence the work of social
scientists. Despite this bridge, however, issues of empirical measurement—how to operationalize
conceptual variables and processes, determine appropriate indicators for them, and specify discrete units
to model and test—remain a central concern for scholars interested in the study of culture (Mohr and
Ghaziani 2014).
In this paper, we integrate insights from cultural sociology, music information retrieval (MIR), and
network theory to better understand how consumers distinguish, evaluate, and select products in the
marketplace for popular music. Computer and social scientists alike have found that, while peer
recommendation plays an important role in driving consumption behavior in cultural domains (Salganik,
Dodds, and Watts 2006), a number of other factors also shape audience preferences and selection patterns.
Contingencies such as artist familiarity, payola, label size, and other producer- and song-level
characteristics all independently contribute to a song being considered a hit or a flop (Peterson 1990;
Dowd 1992; Rossman 2012). While some research has shown that cultural attributes, such as the lyrical
content of songs (Lena 2006) or the material composition of buildings (Jones et al. 2012), shape
consumer preferences and classification processes, it remains unclear how this influence asserts itself.
To better understand the role that cultural content plays in determining how products are perceived
and evaluated, we employ a unique dataset consisting of more than 25,000 songs that appeared on the
Billboard Hot 100 charts between 1958 and 2013. In addition to weekly chart position, our data includes
measures of artist, genre, and label affiliation, as well as a suite of algorithmically-determined musical
attributes that represent the sonic fingerprint of each song (e.g., tempo, valence, “danceability,” etc.).1 We
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!1 More information about these attributes and how they are constructed can be found in the Data & Methods section of this paper and in the Appendix.
!
3
leverage this data to measure how certain patterns of attribute similarity between songs influence
consumer preferences. More specifically, we argue that (1) audiences perceive and evaluate songs as
products that compete for attention and are embedded in an ecosystem of cultural production, and (2) a
song’s success is determined in part by its position within this ecosystem, or what we call “cultural
networks.” By specifying the relational systems formed among some bounded set of products, practices,
and ideas, our conceptualization of cultural networks extends related work concerning the effects of
endogenous cultural structures, which suggests that changes in tastes are driven in part by internal
processes rather than by exogenous forces (Mohr 1998; Mark 1998, 2003; Lieberson 2000; Kaufman
2004; Mohr & White 2008), Studying these networks of relations provides a new tool to map and
measure the position of cultural products in relation to one another, and informs our understanding of how
different product classes compete for consumer attention.
After a brief review of relevant prior research, we further develop our interpretation of cultural
networks and generate a series of models designed to test whether relationships based on shared attributes
between songs exist, and to what effect. After demonstrating the baseline significance of sonic attributes
in predicting songs’ success on the Billboard charts, we use cultural networks to measure and test the
effect of songs’ relative conventionality on subsequent chart performance. Our findings provide strong
evidence that cultural content matters, both independently and in the way it structures songs’ relationships
to each other. By recognizing the possibility that culture has its own sphere of influence that is partially
independent of the actors who produce and consume it, we provide a lens into how audiences evaluate
products embedded in subjective markets, and rethink some of the basic mechanisms associated with the
cultural production and consumption process.
CULTURE & THE MOVE TOWARD RELATIONALITY
Research on culture, both in the humanities and social sciences, typically emphasizes contextual
richness and interpretive complexity over large-scale comparison and hypothesis testing. While these
!
4
conditions are ideal for generating detailed descriptions of and theory about culture, meaning, and its
relationship to the social world, they have led some scholars to critique this area of research for being
theoretically rich but “methodologically impoverished” (DiMaggio, Nag, and Blei 2013). Nevertheless,
recent advancements in data collection, analysis, and visualization techniques offer scholars a host of
opportunities to advance our understanding of cultural products and processes (Bail 2014). Tools such as
lattice analysis (Mohr 1994, 1998) and topic modeling (Blei, Ng, and Jordan 2003; DiMaggio et al. 2013;
Mohr et al. 2013) empower researchers to reduce the high dimensionality of meaning to a set of discrete
principles or codes, which can then be used to map the contours of our cultural environment and
inductively thematic connections between cultural objects (DiMaggio et al. 2013; Mohr et al. 2013).
The study of social networks has also become a fruitful line of research for scholars interested in the
determinants of cultural taste and production outcomes. Networks, and the concept of relationality more
generally, are central to our understanding of how social systems operate, but their relevance to the study
of culture has typically focused on producers or consumers, rather than the content itself. For example, in
the context of Broadway musicals, Uzzi and Spiro (2005) find that when collaborations between artists
and producers display small world properties, their cultural productions are more likely to achieve critical
and commercial success. In his work on jazz, Phillips’s (2011) links cultural reproduction efforts (e.g., re-
recordings) to the disconnectedness of cities in which the music was initially recorded. And at the
intersection of culture and cognition, Goldberg (2011) has developed a network-inspired method—
relational class analysis (RCA)—to identify groups of individuals that share distinctive ways of
understanding cultural categories. RCA’s explicit emphasis on the patterns of relationships between
individuals’ attitudes, rather than the attitudes themselves, reaffirms “the relation” as an important unit of
analysis for cultural studies and social scientific inquiry more broadly.
These and other studies (e.g., Dowd and Pinheiro 2013) highlight the means through which cultural
practices have been shaped and determined: via geographic, interpersonal, and/or professional
collaboration networks. While these are sensible applications of relational analysis to the study of culture,
there are other ways in which these two areas have been integrated. Rather than simply being embedded
!
5
in a static cultural environment, social networks can be altered by a dynamic set of tastes and practices,
recasting culture and social structure as mutually constitutive (Lizardo 2006; Pachucki and Breiger 2010;
Vaisey and Lizardo 2010). This view—one that privileges culture’s role in determining social reality—is
supported by the “strong program” in cultural sociology (e.g., Alexander and Smith 2002) and related
work on the materiality of culture (DeNora 2006; Rubio 2012; Rubio and Silva 2013). Rather than
passive symbolic structures, culture of this sort is endowed with real material properties that can influence
the preferences and behaviors of actors.
Research on topics as varied as product ecology (Carroll, Kessina, and McKendrick 2010) and the
social life of ideas (Douglas and Isherwood 1996) provides evidence that systems of goods and
information acquire meaning through the relationships they form with one another. While these
relationships may be socially constructed, they are also materially constituted and defined in part by
overlapping attributes. Empirical work on popular music suggests that the ecology of cultural forms plays
an important role in determining why certain products succeed and others fail (Mark 1998), and related
research in marketing finds that cultural content (in the form of instrumentation) shapes listening
preferences (Nunes and Ordanini 2014). Nevertheless, it remains unclear how such influence is expressed.
Indeed, although existing theory explains the various ways in which cultural practices and social ties
affect each other, “we know quite a bit less about how [practices] are structured and how those structures
matter” (Mohr and White 2008: 509).
Cultural Networks: What Are They, and Why Do They Matter?
Our reading of prior research suggests that there is a gap in the way we conceptualize culture’s role
in shaping consumer preferences. We posit that cultural networks—defined here as the attribute-based
relational structures formed between some system of products, practices, or ideas—represent an important
and understudied phenomenon that can enhance our understanding of these processes. This concept
highlights two fundamental insights that we believe have not been adequately addressed by previous
research: (1) cultural preferences, tastes, and effects are contingent in part on material content; and (2)
!
6
these preferences are defined in relation to one another. Throughout the rest of the paper, we argue that
the study of cultural networks can enhance our understanding of how ecosystems of production emerge
and evolve by identifying new mechanisms that explain the ecological dynamics present in various
product spaces (Mark 1998; 2003).
The idea of a cultural network is not new (e.g., Mohr 1998; Gondal 2011), but the definition
developed in this paper is distinctive in several important ways. First, it implies a dynamic structural
approach that privileges dyadic and system-wide relationships constituted by cultural content, rather than
producers or consumers. Critically, we argue that actors’ evaluations of cultural products are shaped not
only by characteristics of the production and consumption environment, or by social influence pressures
(cf., Salganik, Dodds, and Watts 2006), but also by a product’s position within a broader ecosystem of
cultural production—its cultural network. The intuition behind this argument is relatively straightforward:
while the choices cultural producers and consumers make are shaped by their individual preferences,
relationships, and various external factors, they are also influenced by the cultural networks within which
they and their works are embedded. Put another way, producers’ and consumers’ exposure to systems of
products plays a critical role in shaping their preferences, which serve as the medium through which the
effects of cultural networks are inscribed and conferred. Although the cognition and behavior of human
actors ultimately constitute production and consumption (Weeks and Galunic 2003; Goldberg 2011),
product attributes play a significant role in shaping these processes and their associated outcomes.
Like their social counterparts, cultural networks consist of structural signatures that generate
opportunities differentially, rendering certain products more likely to succeed depending on their relative
position within the broader cultural ecosystem. For example, structural holes theory argues that the
presence of unfilled gaps between actors in a social system creates opportunities for brokerage, which in
turn leads to new and better performance outcomes for the actors who bridge those gaps (Burt 1992;
Zaheer and Soda 2009). Recent research has found that “cultural holes” operate in much the same way.
Lizardo (2014) shows how omnivorous consumers can exploit previously unconnected cultural practices
to shape consumer taste, while Vilhena and colleagues (2014) argue that communicative efficiency in
!
7
academic discourse is a function of scholars’ ability to bridge cultural holes and communicate across
fields. The more distinct and indecipherable a field’s scientific jargon, the less likely scientists are to
leverage insights (e.g., cite previously published papers) from outside their home discipline. Thus, fields
that are distinctive but not unnecessarily opaque (i.e., too far removed to be approachable) are more likely
to engage scholars from other fields.
One of the basic mechanisms that can help to explain how selection occurs in cultural networks is
product differentiation. The relative differentiation of products, practices, and even organizations plays an
important role in determining how these entities perform within or across niches (Hannan and Freeman
1977; Mark 2003). When competing for attention and other resources, products must differentiate
themselves from their peers so as to avoid crowding and confusion, but not so much as to make
themselves unrecognizable given some category or class (Kaufman 2004). In the context of fashion and
naming conventions, Lieberson (2000) invokes subtle differentiation as an endogenous explanation for
the evolution of cultural tastes and trends. Related research on optimal distinctiveness in social identities
(Brewer 1991), organizational category spanning (Zuckerman 1999; Hsu 2006), and differentiation in
consumer products (Lancaster 1975) suggests a common trend throughout the social sciences: namely,
that the road to success requires some degree of both conventionality and novelty (Uzzi et al. 2013).
Although we choose not to state any formal hypotheses a priori regarding the relationship between a
song’s relative differentiation within its cultural network and its performance on the charts, we predict
that those songs that strike a balance between “being recognizable” and “being different” will tend to
outperform their peers.
Differentiation across cultural networks can shed light on how consumers behave across a number of
different contexts, but we believe music represents an ideal setting in which to test these dynamics, due in
part to its reliance on an internally consistent grammar (e.g., discrete combinations of pitch, harmony, and
rhythm) and its inherently subjective quality. Salganik and colleagues (2006) simulated an artificial
musical marketplace to show that consumer choice is driven both by social influence and a song’s artistic
quality, but measuring quality objectively requires a comprehensive understanding of music’s form and
!
8
attributes. Due to the specialized skills needed to identify, categorize, and evaluate such attributes
reliably, work in this domain is limited. The research that has been conducted employs musicological
techniques to construct systems of comparable musical codes that may be more or less present in a
particular musical work (Cerulo 1988; La Rue 2001a, 2001b; Nunes and Ordanini 2014). Yet even if
social scientists learned these techniques, or collaborated more often with musicologists, it would be
extremely difficult to apply and automate such complex codes at scale.
Fortunately, these difficulties have been partially attenuated by the rise of digital data sources,
machine learning, and new computational methods in the study of music. Developed first by computer
scientists and then adopted by mainstream social science, these technologies have begun to filter into the
toolkits of cultural sociologists (Bail 2014). Most relevant for our purposes are the advances made in the
fields of music information retrieval (MIR) and machine learning, which have made possible new
research prospects that were previously considered impractical. Using new web-based data that includes
discrete representations of musical content in the form of sonic attributes, we investigate how a song’s
success is determined in part by its relative position within a cultural network.
DATA & METHODOLOGY
Our primary data come from the weekly Billboard Hot 100 charts, which we have reconstructed
from their inception on August 4, 1958 through May 11, 2013. These charts contain information on and
rankings of the most popular songs in public circulation, and thus serve as an appropriate proxy for
consumer preference and evaluation outcomes in the field of popular music. While the eponymous
magazine originally published the charts, the data we use comes from an online repository of Billboard
charts known as “The Whitburn Project.” Joel Whitburn collected and published anthologies of the charts
(Whitburn 1986, 1991) and, beginning in 1998, a dedicated fan base started to collect, digitize, and add to
the information contained in those guides. This augmented existing chart data, adding metadata and
additional details about the songs and albums on the various Billboard charts. Our dataset includes 25,762
!
9
observations for which we were able to obtain complete data, constituting approximately 97.5% of the
songs that appeared on the charts during this time period.
Although the algorithm used to create the charts has changed over the years—starting with a
combination of radio airplay and a survey of record stores across the country, and gradually evolving to
include actual unit sales (see Anand and Peterson [2000] for implications of this change), digital sales,
and even streams (Billboard.com 2007, 2012)—they remain the industry standard. As such, they have
been used extensively in social science research on popular music (Alexander 1996; Scott 1999; Anand
and Peterson 2000; Dowd 2004; Lena 2006; Lena and Pachucki 2013; Mol et al. 2013), and are widely
defended for their reliability as indicators of popular taste (e.g., Eastman and Pettijohn II 2014).
Genre. While exhaustive, the Whitburn data require augmentation in order to capture more fully the
multifaceted social and compositional elements of songs and artists. Although genres are often-changing
and potentially contentious (Lena and Pachucki 2013), they provide a symbolic classification system by
which producers, consumers, and critics of music structure their listening patterns and tastes (Bourdieu
1983, 1984; Frith 1996; Hesmondhalgh 1999; Lena 2012), and therefore impact the way that music is
perceived and its attributes evaluated. Further, genres play a significant role in defining and shaping
artists’ identities (e.g., Peterson 1997; Phillips and Kim 2008), which in turn help determine which
listeners seek out and are exposed to new music. Audiences consequently reinforce the artist identities
and genre structure (Negus 1992; Frith 1996), setting expectations for both producers and their products.
We collected genre data from allmusic.com, an encyclopedic music site containing extensive artist,
album, and track data. Although this and other user-generated music websites typically contain multiple
genre and style designations for each artist (and often each album or even each track), we created dummy
variables for the primary genres affiliated with each artist in our analysis. There were nineteen
possibilities, including Pop, Rock, Country, Electronica, Jazz, and Rap. We chose to use artist- rather than
track-level genre designations for several reasons: (1) comprehensiveness and comparability across our
dataset; (2) the stability and face validity associated with artist- rather than song-level genre designations
!
10
(based on a review of a random sample of songs in our dataset); and (3) the belief that genres influence
consumer preference at the level of the artist and his or her core identity, rather than through the
classification of specific songs (e.g., listeners’ exposure to songs is largely rooted in their familiarity with
and preference for particular artists located with particular genres). We also conducted analyses using
track-level genre designations, and similar results were obtained.2
Echo Nest Audio Attributes. Although genre represents an important means of symbolic classification in
music, our interest in cultural content necessitated the collection of detailed information concerning the
acoustic attributes of each song in our dataset. For these data, we turned to The Echo Nest, an online
music intelligence provider that offers access to much of their data via a suite of application programming
interfaces (APIs). This organization represents the current gold standard in MIR, having recently been
purchased by Spotify to run its analytics and recommendation engine. Using web crawling and audio
encoding technology, the Echo Nest has collected—and continuously updates—information on over 30
million songs and nearly 3 million artists. This data contains objective and derived qualities of audio, text
analyses based on artist mentions in articles and blog posts, and qualitative information about artists.
We used the Echo Nest API to collect data on 97.5% of the songs that appeared on the charts
between 1958 and 2013, including sonic features (such as tempo, mode, and key), as well as some of the
company’s own subjective creations like “valence,” “danceability,” “acousticness” (see the Appendix for
a detailed description of all attributes used in our analysis). Although we were not granted access to the
algorithms that generated these features, our conversations with leading MIR researchers support our
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!2 One of the weaknesses in our data is that these genre codes were applied in the early twenty-first century, rather than the year in which each song was originally released. While genre attributions are admittedly dynamic (Lena and Peterson 2008), we believe it is reasonable to assume that historical attributions are for the most part consistent with our data. This also helps to explain why we use primary genre attributions at the artist- rather than track-level: though genres appear and disappear over the course of our data, and those that persist have evolved, such changes have their provenance predominantly at the sub-genre level (e.g., “Hard Rock” and “Roots Rock” versus “Rock” (see McLeod 2001; Mol et al. 2013). Employing genre assignments at the artist level, particularly at the broadest level of genre classification, means that in all likelihood, misattributions are unlikely or should be relegated to fringe cases. Examining the changes in the content and meanings of genres over time represents an area that we intend to explore in the future.
!
11
belief that these measures provide the most systematic attempt at capturing songs’ material and sensory
composition. Nearly twenty years of research and advancements in MIR techniques have produced both
high- and low-level audio features that provide an increasingly robust representation of how listeners’
perceive music (Friberg et al. 2014).
While some of the Echo Nest’s data has been used to conduct research in computer science and
machine learning (e.g., Shalit, Weinshall, and Chechik 2013), it has not yet been used by social scientists
(see Serrà et al. 2012 for one notable exception). Given its relative novelty, then, it is necessary to address
its potential concerns. First, there are the substantial limitations associated with distilling complex cultural
products into a handful of discrete attributes meant to represent entire songs. Individual songs can vary
substantially in terms of their tempo, key, and general dynamics, and the data we use provides a single
number—often between 0 and 1—that captures the Echo Nest’s computers’ summation of each attribute
for a given song. These measures are inherently reductionist. Second, as the music analyzing process is an
ongoing and dynamic one, the algorithms can change over time, meaning that the values for a given
song’s attributes can similarly change. Each of these concerns is legitimate. However, because the nature
of our primary analyses is relational, the impact of such changes is substantially reduced. The same
algorithmic adjustments are applied to all of the songs, meaning that the relationships between the songs,
our main interest, should not change in any meaningful way. Moreover, having now tracked this data for
multiple years, we can say that the changes to the attributes in our data are negligible. Though far from
perfect, the Echo Nest attribute data represents the closest approximation of what people actually hear
available.
Together with the genre categories described above and a dummy variable reflecting whether a song
was released on a major or independent label, we employ ten of the Echo Nest’s primary sonic attributes
for our initial set of models. Table 1 provides the descriptive statistics for these and other variables in our
analyses.
!
12
[Insert Table 1 Here]
Song Conventionality (Primary Independent Variable). A common means of examining the effects of
networks on performance outcomes is to look at crowding (e.g., Bothner, Kang, and Stuart 2007) and
differentiation. This strategy has been particularly effective in the ecology literature (Podolny, Stuart, and
Hannan 1996; Negro, Hannan, and Rao 2010), where the presence of too many organizations can
overcrowd a consumer or product space (niches). In our analysis, we highlight the effects of crowding and
differential positioning within cultural networks by constructing a measure and testing the effects of song
conventionality. We argue that songs perceived to be sonically similar to one another will compete for a
particular audience segment’s attention. Specifically, we predict that an inverted U-shaped relationship:
songs that are too similar to or too dissimilar from their peers will tend to perform worse on the charts,
while those that are optimally differentiated will tend to perform better.
To test this prediction, we constructed a measure of song conventionality. Following research that
uses cosine similarity to estimate the distance between two distinct entities (Evans 2010; Aral and Alstyne
2011), we created a related measure by taking the ten audio attributes provided by The Echo Nest,
normalizing them each to a 0-1 scale, and then collapsing them into a single vector for each song, Vi. We
took each week’s chart (t) separately and calculated the cosine distance between each song’s vector of
attributes, using the following equation:
!"#$!"! = cos!(!!!, !!")
The resulting matrix At has dimensions matching the number of songs on each week’s charts, with cell
Aijt representing the similarity between song i and song j. As every song is perfectly similar to itself (i.e.,
has a cosine similarity of 1), we removed A’s diagonal from all calculations. We posited that simply
taking the average of each song’s row in the matrix (i.e., creating a raw conventionality score) would
open up the possibility that two dissimilar songs that “looked” similar in terms of their attributes would
!
13
actually sound very different, and end up biasing our analyses.3 Continuing our earlier line of reasoning
around artist genres, we know that consumers tend to be split into segments defined by the type of music
that they listen to, and these genres each have their own distinct traditions and histories (cf. Lena 2012).
Although omnivorous consumer behavior is on the rise (e.g., Lizardo 2014), we believe that the
perception of the distance between two songs will increase if those songs are positioned within two
different genres. To account for both the fact that listeners' perceptions of a song's attributes are likely to
be influenced by genre affiliation and the potential for the acoustic attribute values to make two distinct
songs appear more similar than they are, we weighted each song pair's raw cosine similarity by the
average similarity of those songs' primary genres. To do this, we calculated yearly attribute averages for
each genre, and then used the same cosine similarity equation to measure the average attribute overlap of
each genre pair. The resulting weights were then applied to the raw similarity measures for each song
pair. For example: if one rock song and one folk song had a raw cosine similarity of 0.75, and the average
cosine similarity between rock and folk for the year in question is 0.8, then that genre-weighted song
conventionality for those two songs would be 0.75 * 0.8 = 0.6. We calculated the average of each row in
the matrix, creating the variable genre-weighted song conventionalityt, a genre-weighted average of each
song’s distance from all other songs with which it shares a chart.
Dependent Variables. The weekly Billboard charts provide us with a real-world performance outcome
that represents the general popularity of a song, and can be compared and tracked over time. Unlike
movie box-office results or television show ratings, music sales are often tightly guarded by the content
owners, leaving songs’ diffusion across radio stations (Rossman 2012) or their chart position as the most
reliable and readily available performance outcome. In their examination of fads in baby naming, Berger
and Le Mens (2009) use both peak popularity and longevity as key variables in the measurement of
cultural diffusion processes; here we use them as our dependent variables. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!3 We have run our main set of models with this raw conventionality measure (results available by request) and found that while there is a main effect indicating that increased conventionality is harmful, there is no point of optimal differentiation (i.e., no quadratic effect), after which songs are more likely to fair well on the charts.
!
14
To measure songs’ popularity, we reverse coded peak chart position. Positive coefficients on the
independent variables indicate a positive relationship with chart position. The second performance
measure we use is the number of weeks on the chart, a measure of sustained popularity. Though these two
outcome variables are related to one another in our data set (i.e., songs that reach a higher peak chart
position are likely to remain on the charts longer, R ≈ .74), we believe it is important to look at both peak
performance and its longevity, even if the longest-lived songs in our data only made it to 76 weeks.4
For our relational analyses, we use a slightly different measure of chart success to account for the
week-to-week dynamics between songs that appear in the same cultural network. To determine the effect
of a song’s position in its network at time t on its chart performance at time t +1, we subtracted each
song’s (reverse-coded) position on the charts during the previous week (t) from its current position (t+1)
to measure its ascent up (and descent down) the charts. We consider the change in chart position critical
for our relational models because it (1) better captures the fluid nature of the charts; (2) does not overly
penalize the relatively short “shelf life” of song popularity; and (3) accounts for the fact that songs
appearing near the bottom of the charts have greater opportunity for improvement compared to those at
the top. While we could have included lagged chart position in our models to control for past
performance, the resulting standard errors when running fixed-effects analyses generate concerns (Nickell
1981), and findings are usually more robust when using change scores as a dependent variable (Morgan
and Winship 2007).
ESTIMATION & RESULTS
Although our primary interest concerns the effect of a song’s position within its cultural network on
chart performance, we thought it helpful to first demonstrate the significance of the relationship between
cultural content and consumer preference. Our initial suite of models tests the direct relationship between
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!4 Two additional features of the Billboard Hot 100 charts specifically designed to keep them current warrant mentioning. First, Billboard removes what it calls “recurrent” songs. A song is removed from consideration if it has been on the chart for 20 weeks and falls below position 50. Second, back catalog sales generally do not get counted for the Hot 100’s purposes, though their sales are likely sufficient to merit inclusion.
!
15
a song’s audio attributes and its position on the charts. To do this, we ran pooled, cross-sectional OLS
regressions models for each of our two outcome variables on all of the songs in our data set. Importantly,
we included a set of dummy variables for the number of songs an artist had previously placed on the
charts as a means of capturing artist popularity and visibility. We thought this necessary, in light of the
fact that artists receive differential levels of institutional support (e.g., marketing or PR), which is related
to the number of hits they have previously had. Moreover, these dummies also help to capture “superstar”
effects (Krueger 2005; Elberse and Oberholzer-Gee 2006), which may aid certain artists due to their
previous songs’ success.5 We also included dummies for each decade to examine differences in these
relations over time.
Results
Table 2 presents estimates for three models demonstrating the relationship between genre assignments,
audio attributes, and chart performance. Model 1 includes eighteen genre categories and the major label
affiliation dummy (“pop,” the nineteenth genre, serves as the reference category, as does indie label
affiliation), as well as controls for song length, artists’ previous chart success, and decade. Although these
indicators account for only a small portion of the variance in peak position reached on the chart (R2 =
.044), some expected patterns emerge. First, we find that most niche genres simply are not as successful
as major genre categories: songs from the blues, country, electronica, jazz, Latin, and vocal genres tend to
peak at lower chart positions than those songs in the broad and diverse pop genre. Songs from major
labels tend to generally do better than their indie peers, and there is a curvilinear effect of song length..
Most interestingly, the dummies representing artists’ previous chart success provide evidence of a
“sophomore slump,” a term used to describe musicians’ failure to produce a second song or album as
popular as their first. These results suggest that an artist’s second and third “hit” songs do not perform as
well as their first. Another way to interpret this is result is that “one hit wonders” receive a performance
boost, above and beyond what the inherent quality of a song would predict. Moreover, the positive
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!5 We also ran these models with fixed effects for artists and found similar results.
!
16
coefficient for songs released by artists with more than 10 previously charting songs reveals evidence of a
potential superstar or Matthew Effect (Merton 1968).
[Insert Table 2 Here]
Next, we add songs’ sonic attributes to test the effect of cultural content on peak chart position.
Coefficients for tempo, mode (major key), danceability, and liveness are all positive, while those for
energy (intensity/noise), valence (emotional scale), and acousticness are negative. These results provide
preliminary evidence that cultural attributes are tied to song success above and beyond the social factors
that we include in our model. Finally, we use this model to estimate a song’s duration on the charts.
Slightly different patterns emerge. For example, some niche genres appear to benefit from their faithful
audience base. Although genres like country, electronica, and rap were less mainstream than rock or pop
throughout most of the Hot 100’s history (Peterson 1997; Eastman and Pettijohn II 2014), they benefited
from fervent fan bases that are clearly delineated from other genres on the chart. This combination may
explain why songs from these genres tend to last longer on the charts, despite not enjoying the peak
positions reached by pop and rock. Results also support the earlier finding that more upbeat songs—
highly “danceable” songs, in particular—enjoy more sustained success on the charts. The same is true for
songs that sound more “live” (as opposed to those that sound more computer generated, or overly
produced). Conversely, higher levels of energy, “acousticness,” and emotional valence all contribute to
shorter shelf lives for songs. We also ran logistic regression models predicting a song’s entry into the Top
40, Top 10, and Number 1 spot. All of these supported the results reported in Table 2, and are thus not
reported here (results available upon request).
This first set of models confirms that certain attributes and genre assignments do, in fact, predict a
song’s position and duration on the charts. However, while these results establish some face validity for
our attribute measures, they do not allow us to tell a coherent story about how particular attribute
configurations cohere to affect evaluation outcomes. Without interaction effects or clustering analyses,
which are difficult to interpret across a ten dimensional attribute space, it is unclear how to interpret these
!
17
findings beyond their correlational implications. We address these limitations by employing our song
conventionality measure to test how songs’ differentiation across a cultural network affects changes in
chart position. Before presenting regression results, we first wanted to determine whether there is a
general associative relationship between chart position and song conventionality. To do this, we
calculated the relative conventionality of songs that reached Number 1 compared to other segments of the
chart (e.g., Top 10, Top 40, and the rest of the Hot 100). The resulting trend (standardized across
observations) appears in Figure 1, and suggests that the most successful songs in our dataset are also the
least conventional: songs that appear toward the top of the charts tend to be more novel than other songs
on the charts (differences are statistically significant from one another, p < .001).
[Insert Figure 1 Here]
Given the relatively high levels of conventionality across our dataset (M = 0.796), and the limited
variance therein (SD = 0.0593), we argue that, while adhering to existing song-writing prescriptions is
likely to help a song achieve some degree of popularity, those songs that truly separate themselves from
the pack tend to exhibit some degree of differentiation or novelty on one or more attributes. This is
especially true for songs that reach the top of the charts. For example, Lionel Richie’s 1984 hit “Hello”
reached the top of the charts on May 5, 1984, and featured a conventionality score of .663 the week
before it peaked—nearly three standard deviations less conventional than other songs on the chart that
week. According to The Echo Nest, much of the song’s novelty can be attributed to its low “danceability”
(2.8 SD below the mean for that week) and above average “liveness” (3.0 SD above the mean). In 1992,
Whitney Houston’s version of power ballad “I Will Always Love You” reached the top of the charts with
a conventionality score of .576, some 3.5 standard deviations lower than the mean for the songs with
which it was competing. The song’s relatively low energy score (2.8 SD below the mean) and high
“acousticness” (3.4 SD above the mean) contributed to its distinctiveness and, we argue, its success. The
!
18
associative relationship between attributes and chart position for these songs the week before they reached
the top of the charts is represented graphically in Figures 2 and 3. We provide a table detailing these and
other examples of Number 1 songs that exhibit distinction across attributes in our dataset in the Appendix.
While these figures do not prove a causal relationship between conventionality and chart position, they
offer some descriptive support for our differentiation prediction.
[Insert Figures 2 & 3 Here]
We now turn to the results presented in Table 3, which examine the impact of song conventionality
on changes in chart position. The three models we estimated include linear and quadratic control variables
for the number of weeks a song has already been on the charts. We do this in light of the fact that the
average “life span” of a song is just over 11 weeks, and songs generally follow a parabolic trajectory
through the charts. Most songs initially improve in chart position, but begin to decline after about 6
weeks. It is relatively rare—though not unheard of—for songs to move up substantially after that point in
time. Additionally, all of these models include song-level fixed effects, which allow us to control for the
effect of time-invariant factors of each song, including the artist, his or her label, the marketing budget,
and the song’s individual audio attributes. Having already established the relevance of these attributes, we
remove their direct influence on songs’ movement up or down the charts, and focus instead on the effect
of cultural networks via conventionality. All independent variables and controls are lagged one week
(time t).
[Insert Table 3 Here]
Model 4 employs the linear genre-weighted song conventionality variable to predict a song’s weekly
change in position on the charts. The coefficient here is strongly negative, indicating that in a given week,
!
19
songs that are more similar to their peers are likely to suffer a performance discount in subsequent weeks.
After accounting for the natural decay that songs experience after appearing on the chart (see the negative
coefficient for Weeks on charts), a 1 SD increase in conventionality results in a song falling an additional
½ a position each week, which is not insignificant given the average debut position of songs on the charts
(82). This story changes, however, when a quadratic term for song conventionality is added. Although the
linear term in model 5 is only marginally significant, the sign of its coefficient is now positive, while the
squared term is significant and negative. This finding suggests that the least conventional songs in our
sample appear to benefit from attribute crowding, rather than differentiation, although this effect is
limited to those songs that are most sonically distinctive—approximately 2% of our sample, or songs that
are greater than 2.5 standard deviations below the mean for genre-weighted song conventionality.
This result suggests that songs from typically unrepresented genres, or those in mainstream genres
that are particularly unique, benefit from the appearance of more similar sounding songs on the charts,
which serve as a kind of touchstone for listeners to compare and evaluate similar cultural products that are
otherwise distinctive. Nevertheless, for the vast majority of songs that appear in our dataset, increased
levels of conventionality predict a drop in chart position. Taken together, these findings offer strong
support for the importance of a focal song’s conventionality relative to its peers in determining how it is
perceived and evaluated by consumers. A song’s success is thus in part a function of its position with a
cultural network, which is dynamic and changes week to week.
In our final model, we add fixed effects for each week’s chart to control for temporal differences in
the dispersion of conventionality.6 This represents our most conservative estimate of the effects of song
conventionality on changes in chart position, and further reinforces the role of optimal differentiation on
selection dynamics. All else equal, a single standard deviation increase in a song’s conventionality score
from the mean (e.g., 0.796 to 0.855) is associated with a small drop in chart position. Conversely, a 1 SD
increase from the minimum conventionality score in our dataset (0.226) results in a nearly two-position !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!6 With over 2,500 chart-weeks, we do not show the coefficients for these dummy variables. However, they are predominantly significant, supporting their inclusion. They also reveal temporal differences, as the charts from 1964 to 1970 have negative coefficients, while those from 1978 forward have positive coefficients.
!
20
improvement on the subsequent chart. The benefits associated with conventionality end around 0.7,
encompassing approximately6% of the songs that appear in the Hot 100. Above this point, sounding too
much like your peers harms your subsequent performance on the charts. Viewed another way, songs that
are novel are more likely to ascend the charts until they become too distinct, after which point they too
begin to suffer.
DISCUSSION & CONCLUSIONS
To summarize, our results findings provide compelling evidence that the material attributes attached
to cultural products matter, both independently and in the way they structure the ecosystem of cultural
production and consumption. Controlling for many of the social factors that contribute to a cultural
product’s success, we find that consumer assessments of popular music are shaped in part by the content
of the songs themselves, perhaps suggesting that listeners are more discerning than we sometimes give
them credit for (cf. Salganik, Dodds, and Watts 2006). Our empirical proxy for the effect of cultural
network configurations—song conventionality, a concept that can easily be transferred to other domains
of cultural analyses—significantly affects how songs are perceived and evaluated by audiences.
Specifically, while most popular songs suffer a performance discount for being too similar to their peers,
we find that this effect is attenuated and even reversed for the most novel songs in our dataset. These
results suggest that songs’ conventionality relative to their peers plays a significant role in determining
how they perform on the Hot 100 charts, and support the argument that those songs perceived to be
optimally distinct are more likely to achieve superstar success.
We believe that the ideas presented in this paper make several methodological, theoretical, and
empirical contributions. First, we import methods traditionally associated with computer science and big
data analysis to enhance our understanding of large-scale cultural dynamics. While these tools necessarily
simplify the intrinsic high-dimensionality of culture, they also empower us to learn new things that might
otherwise remain unknown. Although many new cultural measurement tools originate from advances in
!
21
computer science and other disciplines, social scientists must critically develop and apply them
appropriately and thoughtfully (Bail 2014). Other scholars have mapped meaning structures (Carley 1994;
Mohr 1994, 1998), charted diffusion patterns (Rossman 2012; Long and So 2013), and introduced the link
between cultural content and consumption behavior (Lena 2004, 2006; Jones et al. 2012), but there has
been no systematic attempt to theorize and measure how attributes influence the emergence and diffusion
of new production and consumption patterns. In this paper, we have introduced a rich dataset capable of
exploring these dynamics,, generating new insights into the world of popular music and cultural
consumption more broadly.
Second, we develop and test the effects of cultural networks. This concept serves both as a tool to
map ecosystems of cultural products and as a means to understand consumption behavior. We argue that
the system of relations between product attributes is theoretically and analytically distinct from—though
integrally connected to and mediated through—networks of cultural producers and consumers. In so
doing, we raise the possibility that cultural content asserts its own partially autonomous influence over
evaluation outcomes through the crowding and differentiation of products (here, songs) within some
ecosystem of cultural production. This conceptualization of cultural attributes is dynamic, and pushes
network scholars to theorize new ways in which mapping techniques might be used to describe different
kinds of relationships. Although existing research on networks focuses largely on interpersonal ties,
substantive relationships exist between all sorts of actors, objects, and ideas. Continuing to redefine
traditional conceptions of “nodes” and “edges” might help scholars rethink how products, practices, and
ideas assert influence or agency, thereby addressing a critical issue in social theory more broadly (Berger
and Luckmann 1966).
Finally, our findings speak to the inherent difficulty in perfecting or even practicing so called “hit
song science” (Dhanaraj and Logan 2005; Pachet and Roy 2008). It is certainly true that producers and
artists have more tools and data at their disposal than ever before, providing them with incredibly detailed
information about the constitutive elements of popular songs that might in turn help them to craft their
own hits (Thompson 2014). Nevertheless, while writing catchy songs may become easier with the
!
22
emergence of these tools, our results suggest that the artists who try to reverse engineer hit songs may be
neglecting two important points. First, songs that are too similar sounding to other songs are going to
have a difficult time reaching the top of the charts. Second, the characteristics of songs that are released
contemporaneously will have a significant impact on a focal song’s performance. Because a song’s
reception is partially contingent on how much it differentiates itself from its peers, and producers cannot
control what songs are released concurrently with their own, we find that the crafting of a hit song may be
more art than science.
While we believe these findings provide important insights into the dynamics of a multi-billion
dollar industry, we also recognize several important limitations of our study. Although the data we use to
measure sonic attributes is relatively comprehensive and sophisticated, it represents a significant
distillation of a song’s musical complexity. Reducing such a high dimensional object into ten fixed
attributes inevitably simplifies its cultural fingerprint and alters its relationships with other like-objects.
As MIR tools improve, so too will our ability to map the connections between songs. Our data also does
not allow us to account for listeners’ interpretations of attributes or lyric similarity between songs.
Moreover, the bounded nature of our sample—which includes only those songs that achieve enough
success to appear on the charts in the first place—limits the generalizability of our conclusions. In the
future we expect to use additional data to conduct comparative analyses that match these songs with those
that never made it to the Billboard Hot 100, encouraging us to better define the scope conditions of our
findings in order to learn more about the effects of cultural networks on performance outcomes. We also
hope to conduct more dynamic analyses to better understand the nature and implications of specific
cultural structures that appear in our dataset. Carving the chart into distinct segments, estimating our
effects for different time periods and genres, and mapping the social life of individual songs via their
chart trajectory should provide additional insight into the dynamic nature of cultural production and
consumption processes, which differ across time and place.
Although we provide robust evidence for how musical attributes affect songs’ performance on the
charts, our explanation of evaluation outcomes is limited to characteristics of the production environment.
!
23
The analyses presented in this paper do not account for the external consumption environment, making it
difficult to identify the mechanisms by which audiences shape the composition of the charts. It is also
unclear how contingent our findings are on the domain of cultural production being studied (e.g., popular
music), and if the findings are limited further by our focus on such a small subset of popular music.
Collecting additional data and employing other methods (e.g., instrumental variables, cluster analyses,
binary random classifiers) will allow us to better account for the complex and likely interdependent nature
of these dynamics. Nevertheless, we believe that the concepts and findings developed in this paper can
inform the study of many empirical domains, especially those in which relatively subjective qualities
determine how audiences differentiate, evaluate, and select products for consumption.
!
24
REFERENCES
Alexander, J. C. and P. Smith. 2002. The Strong Program in Cultural Theory: Elements of a Structural Theory. Pp. 135-150 in Handbook of Sociological Theory, edited by J. Turner. New York: Kluwer Academic/Plenum Publishers.
Alexander, P.J. 1996. Entropy and Popular Culture: Product Diversity in the Popular Music Recording Industry. American Sociological Review 61(1):171–74.
Anand, N., and R.A. Peterson. 2000. When Market Information Constitutes Fields: Sensemaking of Markets in the Commercial Music Industry. Organization Science 11(3): 270–284.
Appadurai, A. 1986. Introduction: Commodities and The Politics of Value. In The Social Life of Things: Commodities in Cultural Perspective, edited by A. Appadurai. Cambridge: Cambridge University Press.
Aral, S. and M. Van Alstyne. 2011. The Diversity-Bandwidth Trade-Off. American Journal of Sociology 117(1): 90–171.
Bail, C.A. 2014. The Cultural Environment: Measuring Culture with Big Data. Theory and Society 43: 465–482.
Berger, J. and G. Le Mens. 2009. How Adoption Speed Affects the Abandonment of Cultural Tastes.
Proceedings of the National Academy of Sciences 106(20): 8146–8150.
Berger, P. and T. Luckmann. 1966. The Social Construction of Reality: A Treatise in the Sociology of Knowledge. Garden City, NY: Doubleday & Company, Inc.
Billboard.com. 2007. “Billboard Hot 100 To Include Digital Streams Billboard.” Billboard.com. 2012. “Hot 100 Impacted by New On-Demand Songs Chart Billboard.” Bothner, M. S., R. Haynes, W. Lee, and E.B. Smith. 2010. When Do Matthew Effects Occur? The
Journal of Mathematical Sociology 34(2):80–114. Bothner, M.S., J.-H. Kang, and T.E. Stuart. 2007. Competitive Crowding and Risk Taking in a
Tournament: Evidence from NASCAR Racing. Administrative Science Quarterly 52(2): 208–247.
Bothner, M.S., Y.-K. Kim, and E.B. Smith. 2011. How Does Status Affect Performance? Status as an Asset vs. Status as a Liability in the PGA and NASCAR. Organization Science 23(2):416–33.
Bourdieu, P. 1983. The Field of Cultural Production, or: The Economic World Reversed. Poetics 12(4–5): 311–356.
_________. 1984. Introduction. In Distinction: A Social Critique of the Judgment of Taste. Pp. 1–7. Cambridge, MA: Harvard University Press.
_________. 1996. The Rules of Art: Genesis and Structure of the Literary Field. Stanford: Stanford University Press.
!
25
Bradlow, E.T., and P.S. Fader. 2001. A Bayesian Lifetime Model for the ‘Hot 100’ Billboard Songs. Journal of the American Statistical Association 96(454): 368–381.
Brewer, M.B. 1991. The social self: On being the same and different at the same time. Personality and Social Psychology Bulletin 17: 475–482.
van Berchum, M. 2014. Linked Sources: A Network Approach to the Repertory of Sixteenth-Century
Polyphony. Leonardo 47(3): 272. Burt, R. 1992. Structural holes: The social structure of competition. Cambridge, MA: Harvard University
Press. ______. 2004. Structural Holes and Good Ideas. American Journal of Sociology 110: 349-399. Carley, K.M. 1994. Extracting Culture through Textual Analysis. Poetics 22: 291-312.
Carroll, G.R., O.M. Kessina, D.G. McKendrick. 2010. The Social Lives of Products: Analyzing Product Demography for Management Theory and Practice. The Academy of Management Annals 4(1): 157–203.
Cerulo, K. 1988. Analyzing Cultural Products: A New Method of Measurement. Social Science Research 17: 317-352.
DeNora, T. 2006. Music as Agency in Beethoven’s Vienna. Pp. 103-120 in Myth, Meaning, and Performance: Toward a New Cultural Sociology of the Arts, edited by R. Eyerman and L. McCormick. Boulder, CO: Paradigm Publishers.
Dertouzos, J.N. 2008. Radio Airplay and the Record Industry: An Economic Analysis. National Association of Broadcasters, USA.
Dhanaraj, R. and B. Logan. 2005. Automatic Prediction of Hit Pop Songs. Proceedings of the International Society for Music Information Retrieval: 488-491.
DiMaggio, P. 1994. Introduction: Meaning and measurement in the sociology of culture. Poetics 2(4):
263–371. ________. 2011. Cultural Networks. In J. Scott and P.J. Carrington (eds.), The Sage Handbook of Social
Network Analysis. Pp. 286–298. London: Sage Publications. DiMaggio, P., M. Nag, and D. Blei. 2013. Exploiting Affinities between Topic Modeling and the
Sociological Perspective on Culture: Application to Newspaper Coverage of U.S. Government Arts Funding. Poetics 41: 570–606.
Douglas, M. and B.C. Isherwood. 1996. The World of Goods: Towards an Anthropology of Consumption.
New York: Routledge. Dowd, T.J. 1992. The Musical Structure and Social Context of Number One Songs, 1955 to 1988. An
Exploratory Analysis. In R. Wuthnow (ed.), Vocabularies of Public Life: Empirical Essays in Symbolic Structure, Pp. 130–157. London. Routledge.
!
26
. 2004. Concentration and Diversity Revisited: Production Logics and the U.S. Mainstream Recording Market, 1940-1990. Social Forces 82(4): 1411–1455.
Dowd, T.J. and D.L. Pinheiro. 2013. The Ties among the Notes: The Social Capital of Jazz Musicians in Three Metropolitan Areas. Work & Occupations 40(4): 431–464.
Eastman, J.T. and T.F. Pettijohn II. 2014. Gone Country: An Investigation of Billboard Country Songs of the Year across Social and Economic Conditions in the United States. Psychology of Popular Media Culture.
Elberse, A. and F. Oberholzer-Gee. 2006. Superstars and Underdogs: An Examination of the Long Tail Phenomenon in Video Sales. Division of Research, Harvard Business School.
Evans, J.A. 2010. Industry Induces Academic Science to Know Less about More.” American Journal of Sociology 116(2): 389–452.
Friberg, A., W. Schoonderwaldt, A. Hedblad, M. Fabiani, and A. Elowsson. 2014. Using Listener-based Perceptual Features as Intermediate Representations in Music Information Retrieval. Journal of the Acoustical Society of America 136(4): 1951–1963.
Frith, S. 1996. Music and Identity. In S. Hall and P. Du Gay (eds.), Questions of Cultural Identity. Pp. 108–125. London: Sage.
Fuhse, J.A. 2009. The Meaning Structure of Social Networks. Sociological Theory 27(1): 51–73.
Giles, D.E. 2007. Survival of the Hippest: Life at the Top of the Hot 100. Applied Economics 39(15): 1877–1887.
Goldberg, A. 2011. Mapping Shared Understandings Using Relational Class Analysis: The Case of the Cultural Omnivore Reexamined. American Journal of Sociology 116(5): 1397-1436.
Gondal, N. 2011. The Local and global Structure of Knowledge Production in an Emergent Research
Field: an Exponential Random Graph Analysis. Social Networks 33: 20–30.
Hannan, M.T. 2010. Partiality of Memberships in Categories and Audiences. Annual Review of Sociology 36(1): 159–181.
Hannan, Michael T. and John H. Freeman. 1977. “The Population Ecology of Organizations.” American Journal of Sociology 82(5):929–64.
Hesmondhalgh, D. 1999. Indie: The Institutional Politics and Aesthetics of a Popular Music Genre. Cultural Studies 13(1): 34–61.
Hsu, G. 2006. Jacks of all Trades and Masters of None: Audiences’ Reactions to Spanning Genres in Feature Film Production. Administrative Science Quarterly 51(3): 420–450.
Jones, C., M. Maoret, F.G. Massa, and S. Svejenova. 2012. Rebels with a Cause: Formation, Contestation, and Expansion of the De Novo Category “Modern Architecture,” 1870–1975. Organization Science 23(6): 1523–1545.
!
27
Kaufman, J. 2004. Endogenous Explanation in the Sociology of Culture. Annual Review of Sociology 30: 335–357.
Krueger, A.B. 2005. The Economics of Real Superstars: The Market for Rock Concerts in the Material World. Journal of Labor Economics 23(1):1–30.
Lancaster, K. 1975. Socially Optimal Product Differentiation. The American Economic Review 65(4): 567–585.
La Rue, J. 2001a. Significant and Coincidental Resemblance between Classical Themes. The Journal of Musicology 18(2): 268-294.
_______. 2001b. Fundamental Considerations in Style Analysis. The Journal of Musicology 18(2): 295-312.
Lena, J.C. 2006. Social Context and Musical Content of Rap Music, 1979-1995. Social Forces 85(1): 479–495.
_________. 2012. Banding Together: How Communities Create Genres in Popular Music. Princeton: Princeton University Press. Retrieved
Lena, J.C. and Peterson, R.A. 2008. Classification as Culture: Types and Trajectories of Music Genres. American Sociological Review 73(5): 697-718.
Lena, J.C. and M.C. Pachucki. 2013. The Sincerest Form of Flattery: Innovation, Repetition, and Status in
an Art Movement. Poetics 41: 236–264. Lieberson. S. 2000. A Matter of Taste: How Names, Fashions, and Culture Change. New Haven: Yale
University Press.
Lizardo, O. 2006. How cultural tastes shape personal networks. American Sociological Review 71: 778-807.
________. 2011. Cultural Correlates of Ego-Network Closure. Sociological Perspectives 54(3): 479-487. _________. 2014. Omnivorousness as the Bridging of Cultural Holes: A Measurement Strategy. Theory
and Society 43: 395–419. Long, H. and R. So. 2013. Network Science and Literary History. Leonardo 46(3): 274 –.
Mark, N. 1998. Birds of a Feather Sing Together. Social Forces 77(2): 453–485.
_______. 2003. Culture and Competition: Homophily and Distancing Explanations for Cultural Niches. American Sociological Review 68(3): 319–345.
McLeod, Kembrew. 2001. “Genres, Subgenres, Sub-Subgenres and More: Musical and Social Differentiation Within Electronic/Dance Music Communities.” Journal of Popular Music Studies 13(1):59–75.
Merton, R.K. 1968. The Matthew Effect in Science: The Reward and Communication Systems of Science Are Considered. Science 159(3810): 56.
!
28
Michel, J. B. et al. 2010. “Quantitative Analysis of Culture Using Millions of Digitized Books.” Science 331(6014):176–82.
Mohr, J.W. 1994. Soldiers, Mothers, Tramps, and Others: Discourse Roles in the 1907 New York City
Charity Directory. Poetics 22: 327–357. ________. 1998. Measuring Meaning Structures. Annual Review of Sociology 24: 345–370. Mohr, J.W., R. Wagner-Pacifici, R.L. Breiger, and P. Bogdanov. 2013. Graphing the Grammar Motives
in National Security Strategies: Cultural Interpretation, Automated Text Analysis and the Drama of Global Politics. Poetics 41: 670–700.
Mohr, J.W. and A. Ghaziani. 2014. Problems and Prospects of Measurement in the Study of Culture.
Theory and Society 43: 225–246.
Mohr, J. and H. White. 2008. How to Model an Institution. Theory and Society 37(5): 485-512.
Mol, J.M., M.M. Chiu, I.O. Paleo, and N.M. Wijnberg. 2013. Calling Out Around the World, Are You Ready for a Brand-New Beat? Genre Formation in Popular Music. Working Paper.
Morgan, S.L. and C. Winship. 2007. Counterfactuals and Causal Inference: Methods and Principles for Social Research. Cambridge, UK: Cambridge University Press.
Negro, Giacomo, Michael T. Hannan, and Hayagreeva Rao. 2010. “Categorical Contrast and Audience Appeal: Niche Width and Critical Success in Winemaking.” Industrial and Corporate Change 19(5):1397–1425.
Negus, Keith. 1992. Producing Pop: Culture and Conflict in the Popular Music Industry. London: Edward Arnold.
Nickell, S. 1981. Biases in Dynamic Models with Fixed Effects. Econometrica: Journal of the Econometric Society 1417–1426.
Nunes, J.C. and A. Ordanini. 2014. I Like the Way it Sounds: The Influence of Instrumentation on Pop Song’s Place in the Charts. Musicae Scientiae: 1–18.
Pachucki, M.A. and R.L. Breiger. 2010. Cultural Holes: Beyond Relationality in Social Networks and
Culture. Annual Review of Sociology 36: 205-224.
Pachet and Roy 2008. Hit Song Science is Not Yet a Science. Proceedings of the International Society for Music Information Retrieval: 355-360.
Padgett, J.F. and W.W. Powell. 2012. The Problem of Emergence. In The Emergence of Organizations
and Markets, ed. by J.F. Padgett and W.W. Powell. Pp. 1-29. Princeton, NJ: Princeton University Press.
Peterson, R.A. 1990. Why 1955?: Explaining the Advent of Rock Music. Popular Music 9(1): 97–116.
___________. 1997. Creating Country Music: Fabricating Authenticity. Chicago: University of Chicago Press. Retrieved (http://books.google.com/books?hl=en&lr=&id=J3zWpIOLB-MC&pgis=1).
!
29
Phillips, D.J. 2011. Jazz and the Disconnected: City Structural Disconnectedness and the Emergence of a Jazz Canon, 1897–1933. American Journal of Sociology 117(2): 420–83.
Phillips, D.J., and E.W. Zuckerman. 2001. Middle-Status Conformity: Theoretical Restatement and
Empirical Demonstration in Two Markets. American Journal of Sociology 107(2): 379–429.
Podolny, J.M., T.E. Stuart, and M.T. Hannan.1996. Networks, Knowledge, and Niches: Competition in the Worldwide Semiconductor Industry. American Journal of Sociology 102(3): 659–689.
Rossman, G. 2012. Climbing the Charts: What Radio Airplay Tells Us about the Diffusion of Innovation. Princeton, NJ: Princeton University Press.
Rubio, F.D. 2012. The Material Production of the Spiral Jetty: A Study of Culture in the Making. Cultural Sociology 6(2): 143-161.
Rubio, F.D. and E.B. Silva. 2013. Materials in the Field: Object-trajectories and Object-positions in the Field of Contemporary Art. Cultural Sociology 7(3):161-178.
Salganik, M.J., Dodds, P.S. and Watts, D.J. 2006. Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market. Science 311: 854-856.
Schudson, M. 1989. How Culture Works: Perspectives from Media Studies on the Efficacy of Symbols. Theory and Society 18: 153-180.
Scott, A. J. 1999. The US Recorded Music Industry: On the Relations between Organization, Location, and Creativity in the Cultural Economy. Environment and Planning A 31(11): 1965–1984.
Serrà, J., Á. Corral, M. Boguñá, M. Haro, and J.L. Arcos. 2012. Measuring the Evolution of Contemporary Western Popular Music. Scientific Reports 2.
Shalit, U., D. Weinshall, and G. Chechik. 2013. Modeling Musical Influence with Topic Models. In Proceedings of the 30th International Conference on Machine Learning (ICML-13). Pp. 244–252.
Swidler, A. 1986. Culture in Action: Symbols and Strategies. American Sociological Review 51: 273–286.
Thompson, D. The Shazam Effect. The Atlantic Magazine. December 2014.
Uzzi, B. and J. Spiro. 2005. Collaboration and Creativity: The Small World Problem. American Journal of Sociology 111: 447-504.
Uzzi, B., S. Mukherjee, M. Stringer, and B. Jones. 2013. Atypical Combinations and Scientific Impact.
Science 342: 468–472. Vaisey, S. and O. Lizardo. 2010. Can Cultural Worldviews Influence Network Composition? Social
Forces 88(4): 1595-1618. Vilhena, D.A., J.G. Foster, M. Rosvall, J.D. West, J. Evans, and C.T. Bergstrom. 2014. Finding Cultural
Holes: How Structure and Culture Diverge in Networks of Scholarly Communication. Sociological Science 1: 221–238.
!
30
Weber, K. 2005. A Toolkit for Analyzing Corporate Cultural Toolkits. Poetics 33: 227-252. Weeks, J. and C. Galunic. 2013. A Theory of the Cultural Evolution of the Firm: The Intra-Organizational
Ecology of Memes. Organization Studies 24(8): 1309–1352. Whitburn, J. 1986. Joel Whitburn’s Pop Memories, 1890-1954: The History of American Popular Music:
Compiled from America’s Popular Music Charts 1890-1954. Record Research. _________. 1991. Joel Whitburn’s Top Pop Singles, 1955-1990: Compiled from Billboard’s Pop Singles
Charts, 1955-1990. Record Research.
Zaheer, A. and G. Soda. 2009. Network Evolution: The Origins of Structural Holes. Administrative Science Quarterly 54(1): 1-31.
Zuckerman, E.W. 1999. The Categorical Imperative: Securities Analysts and the Illegitimacy Discount.
American Journal of Sociology 104(5): 1398–1438.
!
31
TA
BL
ES &
FIGU
RE
S
Table 1. C
orrelations and Descriptive Statistics for V
ariables in Analyses
[1][2]
[3][4]
[5][6]
[7][8]
[9][10]
[11][12]
[13][14]
[15][16]
[17][18]
[19]
[1] Change in (inverted) Position
1[2] G
enre-weighted Song C
onventionality0.01
1[3] W
eek on charts-0.44
-0.031
[4] Major label dum
my
-0.04-0.02
0.061
[5] Song tempo (norm
alized 0-1)-0.01
0.09-0.01
0.001
[6] Song energy-0.03
0.280.02
0.010.15
1[7] Song speechiness
-0.01-0.11
0.04-0.01
-0.050.03
1[8] Song acousticness
0.05-0.29
-0.08-0.03
-0.03-0.67
0.051
[9] Minor/M
ajor Mode (0 or 1)
0.010.67
-0.020.01
-0.01-0.06
-0.080.08
1[10] Song danceability
-0.030.20
0.07-0.02
0.110.11
0.20-0.27
-0.131
[11] Song valence0.00
0.27-0.04
-0.090.16
0.450.19
-0.31-0.11
0.571
[12] Song liveness0.00
-0.13-0.01
0.02-0.01
0.140.06
-0.010.00
-0.24-0.02
1[13] Song key
0.000.04
0.010.01
0.000.02
0.02-0.02
-0.160.02
0.04-0.02
1[14] Song tim
e signature-0.02
0.120.04
0.020.03
0.150.05
-0.18-0.03
0.140.10
0.000.00
1[15] Song length (seconds)
-0.10-0.11
0.200.17
-0.010.05
0.11-0.19
-0.090.13
-0.08-0.02
0.020.09
1[16] 1st charting song dum
my
0.01-0.01
0.04-0.07
0.020.02
0.040.00
-0.020.07
0.06-0.05
0.01-0.01
-0.061
[17] 2nd or 3rd charting song dumm
y0.01
0.010.01
-0.01-0.01
0.000.02
-0.01-0.01
0.030.01
-0.010.00
0.01-0.01
-0.311
[18] 4th-10th charting song dumm
y-0.01
0.00-0.01
0.050.00
0.01-0.01
-0.010.00
-0.02-0.02
0.040.00
0.010.03
-0.39-0.34
1[19] > 10th charting song dum
my
-0.010.00
-0.050.03
-0.01-0.03
-0.040.03
0.03-0.07
-0.050.02
-0.01-0.01
0.05-0.32
-0.28-0.35
1M
ean0.54
0.808.72
0.700.44
0.590.07
0.260.73
0.600.58
0.245.27
3.90215.64
0.260.21
0.300.23
Standard Deviation
10.200.06
6.900.46
0.090.22
0.090.29
0.440.15
0.250.22
3.560.47
48.710.44
0.410.46
0.42M
inimum
-790.23
10
0.160.00
0.020.00
00
00.01
01
620
00
0M
aximum
960.91
761
11.00
0.961.00
10.99
1.001
117
5701
11
1
!
32
Table 2. Pooled, Cross-Sectional OLS Models Predicting Billboard Hot 100 Peak Chart Position & Longevity (1 of 2)
Continued
[1] [2] [3]Peak (inverted) Peak (inverted) Weeks on charts
Blues -18.92** -22.95** 0.116(2.015) (2.091) (0.424)
Children's 5.463 7.658 -1.394(8.031) (8.617) (2.163)
Comedy -5.465+ -7.436* -0.332(2.943) (3.241) (0.605)
Country -5.423** -8.205** 2.259**(1.027) (1.065) (0.326)
Easy -4.529* -5.434* 2.743**(2.206) (2.269) (0.462)
Electronica -9.715** -11.20** 1.462**(1.660) (1.751) (0.536)
Folk -2.621 -5.991** 1.815**(2.238) (2.298) (0.451)
Gospel -7.806+ -10.20* 1.049(4.101) (4.489) (0.911)
Jazz -11.29** -12.98** 1.152**(1.844) (1.955) (0.414)
Latin -7.179* -9.153** 0.979(3.467) (3.553) (1.116)
New Age -4.937 -3.711 1.920(10.34) (10.40) (2.614)
R&B 1.579 -0.954 3.032**(1.009) (1.049) (0.312)
Rap -1.654 -4.078** 1.140**(1.126) (1.250) (0.374)
Reggae -0.481 -4.567 2.409*(3.538) (3.566) (0.973)
Rock 1.411 -0.689 2.753**(0.947) (0.986) (0.304)
Spoken Word -9.479 -8.762 -1.108(8.690) (10.10) (0.852)
Vocal -6.459** -7.616** 2.016**(1.358) (1.425) (0.350)
World -3.340 -4.224 0.883(6.566) (7.486) (1.302)
Major Label Dummy 3.078** 2.626** 0.347**(0.422) (0.432) (0.0971)
Song length (seconds) 0.183** 0.170** 0.0550**(0.0255) (0.0262) (0.00504)
Song length2 -0.000241** -0.000224** -8.74e-05**(5.28e-05) (5.41e-05) (1.07e-05)
All Music Genres - Pop is reference
genre
!
33
Table 2 continued
2nd or 3rd charting song -2.765** -4.367** -1.326**(0.563) (0.581) (0.138)
4th-10th charting song 2.974** 0.936+ -1.043**(0.535) (0.548) (0.128)
> 10th charting song 4.543** 2.569** -1.770**(0.568) (0.582) (0.135)
1960s -4.258** -3.605** -1.307**(1.153) (1.184) (0.217)
1970s -6.656** -4.868** -0.0557(1.249) (1.299) (0.245)
1980s -4.085** -3.713** 2.068**(1.336) (1.398) (0.276)
1990s -6.762** -7.265** 4.427**(1.388) (1.455) (0.307)
2000s -6.829** -7.622** 5.149**(1.386) (1.449) (0.321)
2010s -13.39** -13.76** 2.273**(1.526) (1.584) (0.399)
Song tempo (0-1) 8.231** 0.894+(2.166) (0.511)
Song energy -7.211** -2.367**(1.398) (0.330)
Song speechiness 0.422 -0.387(2.738) (0.661)
Song acousticness -2.385* -0.524*(0.978) (0.216)
Minor/Major Mode (0 or 1) 1.277** 0.250*(0.449) (0.105)
Song Danceability 11.97** 3.438**(1.844) (0.427)
Song valence -3.956** -0.820**(1.145) (0.265)
Song liveness 7.437** 1.553**(0.900) (0.208)
Song key 0.0950+ 0.0204(0.0542) (0.0126)
Song time signature 0.445 0.102(0.387) (0.0786)
Constant 31.22** 29.46** -0.402(2.917) (3.801) (0.794)
Observations 25,731 24,369 24,369R-squared 0.044 0.049 0.184
Robust standard errors in parentheses** p<0.01, * p<0.05, + p<0.1
Artist previous charting song
buckets
Decade dummies
Echo Nest Attributes
!
34
Table 3. Fixed Effects Models Predicting Billboard Hot 100 Songs’ Change in Position
!
[4] [5] [6]Δ Chart Position Δ Chart Position Δ Chart Position
Genre-weighted Song Conventionality -6.821** 41.95+ 49.90*(2.422) (22.59) (22.68)
Genre-weighted Song Conventionality2 -32.68* -35.64*(14.69) (15.20)
Weeks on charts -2.079** -2.080** -2.843**(0.0115) (0.0115) (0.0753)
Weeks on charts2 0.0378** 0.0378** 0.0210**(0.000435) (0.000434) (0.000330)
Constant 19.16** 1.160 -652.6**(1.930) (8.725) (108.7)
Fixed Effect for Chart Week? No No YesObservations 253,642 253,642 253,642R-squared 0.448 0.448 0.533Robust standard errors in parentheses** p<0.01, * p<0.05, + p<0.1
!
35
Figure 1. Differentiation A
cross the Billboard H
ot 100
!0.05%
!0.04%
!0.03%
!0.02%
!0.01% 0%
0.01%
0.02%
0.03%
Num
ber%1%Rest%of%Top%10%
Rest%of%Top%40%Rest%of%Top%100%
Genre-W
eighted Avg. Song
Conventionality
(standardized)
Peak%Chart%Posi-on%
!
36
Figure 2. Surface Plot Showing C
orrelation Betw
een A
cousticness, Energy and C
hart Performance, B
illboard H
ot 100, Novem
ber 21, 1992
Figure 3. Surface Plot Showing C
orrelation Betw
een A
cousticness, Energy and C
hart Performance, B
illboard H
ot 100, May 5, 1984
1
20
40
60
80
100 (Absolute Value of SDs)
Chart Position
(Absolute Value of SDs)
1
20
40
60
80
100 (Absolute Value of SDs)
Chart Position
(Absolute Value of SDs)
!
37
APPENDIX
Table A1. Echo Nest Audio Attributes8
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!8 This list of attributes includes all but one of the attributes provided by The Echo Nest’s suite of algorithms: loudness. This variable was cut from our final analysis at the suggestion of the company’s senior engineer, who explained that loudness is primarily determined by the mastering technology used to make a particular recording, a characteristic that is confounded through radio play and other forms of distribution.
Attribute Scale Definition
Acousticness 0-1 Represents the likelihood that the song was recorded
solely by acoustic means (as opposed to more electronic / electric means)
Danceability 0-1 Describes how suitable a track is for dancing. This
measure includes tempo, regularity of beat, and beat strength.
Duration Seconds Length of the track.
Energy 0-1 A perceptual measure of intensity throughout the track. Think fast, loud, and noisy (i.e., hard rock)
more than just dance tracks.
Key 0-11 (integers only)
The estimated, overall key of the track, from C through B.
Liveness 0-1 Detects the presence of the live audience during the recording. Heavily studio-produced tracks score low
on this measure.
Mode 0 or 1 Whether the song is in a minor (0) or major (1) key
Speechiness 0-1 Detects the presence of spoken word throughout the track.
Tempo Beats per minute (BPM) The overall average tempo of a track.
Time Signature Beats per bar / measure Estimated, overall time signature of the track.
Valence 0-1 The musical positiveness of the track
!
38
Table A
2. Sample #1 Song A
nalysis
TrackA
rtistW
eek Hit
#1
Standardized song conventionality
Week before reaching #1
Select points of distinctiveness, week
before reaching #1W
eeks at #1
Previous Songs on
Chart
Runaround Sue
Dion
16-Oct-61
-2.684929!!1 SD
above average energy!!1.5 SD
below average acousticness
24
Bridge O
ver Troubled W
aterSim
on & G
arfunkel21-Feb-70
-2.001217
!!2 SD below
average energy! 1.6 SD
below average tem
po! 2.1 SD
above average acousticness! 2.1 SD
below average danceability
610
Hello
Lionel Richie
05-May-84
-2.852536!!2.8 SD
below average danceability
!!3 SD above average liveness
25
I Will A
lways Love You
Whitney H
ouston21-N
ov-92-3.531555
!!2.8 SD below
average energy! 1 SD
below average tem
po! 3.4 SD
above average acousticness! 2.7 SD
below average danceability
1415
Just Dance
Lady GaG
a10-Jan-09
-1.43895!!1.4 SD
below average energy
! 1.2 SD above average valence
! 1.6 SD above average danceability
30
Someone Like You
Adele
10-Sep-11-1.757673
!!2.4 SD below
average energy! 1.2 SD
above average tempo
! 2.6 SD above average acousticness
! 1.8 SD below
average danceability
52