manufacturing popularity: an ecological model of …...3 manufacturing popularity: an ecological...
TRANSCRIPT
1
Manufacturing Popularity:
An Ecological Model of Time-Based Competition*
William P. Barnett
Stanford Graduate School of Business
Mooweon Rhee
School of Business, Yonsei University
Elise Tak
Stanford Graduate School of Business
April 6, 2018
*Order of authorship is alphabetical. Thanks to the Stanford Graduate School of Business and
the Yonsei University School of Business for support. We appreciate advice from Miron
Avidan, Dror Etzion, Amir Goldberg, and Paul Ingram.
2
Manufacturing Popularity:
An Ecological Model of Time-Based Competition
Abstract
Most models of competition depict organizations or products vying for resources arrayed
over market dimensions at any given point in time. We observe that many markets, instead,
feature competition for resources arrayed over time, with organizations or products competing to
enjoy a moment of temporary advantage. We develop a model to understand such time-based
competition, and estimate the model using data on Korean popular music. The results indicate
the importance of social exposure to a product’s chances of becoming and remaining popular.
They also show that efforts by large organizations to manufacture popularity in that context have
triggered a self-defeating dynamic. These efforts improve the competitiveness of songs trying to
become popular, but also increase competition from rival songs thereby intensifying the brevity
of success.
3
Manufacturing Popularity:
An Ecological Model of Time-Based Competition
Many markets are characterized by very short product life cycles, where a product will go
from being extremely popular to dying out over a just a few months or even weeks – only to be
replaced by the next “new thing.” We see this pattern in fashion markets, of course, since they
are explicitly structured around seasonality. But many other markets share the quality of short
product life cycles (Katila and Ahuja, 2002; Abrahamson, 1991). For instance, technology-based
products such as computer storage devices and telecommunications equipment go from rapid
market penetration to obsolescence often in a matter of months (Eisenhardt and Schoonhoven,
1990). Even some inherently durable products are replaced in a rapid cadence, as in the
automobile market where new models are introduced annually. And then there is the iconic
example – cultural products such as commercial films and songs – that come and go rapidly
(Grossman, 2012; Askin and Mauskapf, 2017), as parodied in the “15 minutes of fame”
aphorism often attributed to Andy Warhol. In general, these diverse examples share a common
pattern of rapid rise and decline over time. Driven by excitement, the new-new thing takes off
dramatically until it peaks and falls, leaving the market ripe for the next new thing.
Given the prevalence of such transitory products, it is noteworthy that research on
competition rarely considers their dynamics. In the strategy field, the emphasis is on explaining
enduring competitive advantage (Porter 2008), thereby diminishing the importance of products
that come and go in fleeting moments of success. And when transitory products are considered,
often they are depicted as ephemeral - faddish in a pejorative sense notable more for “hype” than
for tangible significance, as in the Gartner “hype cycle” often referred to by strategy consultants
4
(Fenn and Raskino, 2008). Some approaches in economics and sociology allow for competitive
dynamics, but they typically envision competition as occurring at the level of markets, market
segments, or among collections of organizations such as fields, populations, categories, or
networks (Soule and King, 2008; Brynjolfsson, Hu, and Simester, 2011; Stark and Vedres,
2012). Consequently, patterns of competition in the transitory comings and goings of individual
products are overlooked by much of our extant theory (Carroll, Khessina and McKendrick,
2010). Yet judging by their prevalence, transitory products are collectively very important, even
though they are – by definition – individually fleeting.
Our aim here is to develop an ecological model of time-based competition; one that
allows us to isolate how competition shapes the short lives of transitory products. Ordinarily,
models of competition among products or organizations depict a process taking place over some
resource space. Strategies then are understood in terms of an organization’s position across that
space, and the ability of its products to compete in that position. By contrast, our model allows
organizations to contend for resources arrayed over time. It depicts each product’s time for
attracting resources as severely limited, with important implications for strategy and
organization. For starters, resolving uncertainty about a new product is problematic when time is
limited. And once a product launches, its fate hinges on whether it immediately becomes
popular. Competition in such contexts is less about position in resource space and more about
timing. Entering when a rival product is exploding to success may lead a product to end up
languishing in the shadows. Yet waiting to enter amounts to ceding the market to a competitor,
since the possible time for success is inherently limited.
We theorize about how different organizational strategies shape the success and failure of
products in such transitory markets. Our theory points to two distinct, but fundamentally
5
interrelated, mechanisms. The first, social exposure, involves efforts by organizations to draw
attention to their products so that they enjoy growth in popularity. The second important
mechanism is competition, which occurs if a product explodes to success – since this surge in
popularity often comes at the expense of others. We argue that this competitive effect is distinct
from the social exposure effect, and that the two effects together account for the frequently
observed pattern of rise and decline seen in such markets. Our ecological model parameterizes
each of these effects, and thereby allows us to investigate the effectiveness of observed
organizational strategies – in this case looking at the context of Korean popular music.
Rapid Product Cycles in Korean Popular Music
The Korean popular music industry is ideal for our study because of its transitory product
life cycles, and because of the different strategies followed by organizations. Over our study
period, 2004 through 2014, the so-called “Hallyu” (Korean wave) movement grew explosively,
with Korean-produced soap operas and music thriving in Korea but spreading as well to Japan,
China, and beyond (Seabrook, 2015). This cultural movement is perhaps most notable for
“Kpop,” shorthand for Korean popular music and dance, performed live and especially in music
videos by so-called Korean “Idol” groups. These groups typically release songs and music
videos featuring upbeat music and elaborately choreographed group dancing. In some cases,
groups release music videos that achieve considerable attention worldwide, such as “Gangnam
Style” released by Psy in 2012 (Fisher, 2012). For our study, we collected quantitative life-
history data on all Idol songs and groups, and we conducted extensive interviews and case-study
research on-site within Kpop production agencies (Identifying reference).
6
Note that “Idol” in this context refers to a genre category, rather than to a level of
popularity as the term is used in western countries. In terms of their features, Korean Idols are
similar to what are known as “boy bands” and “girl bands” in American popular culture, with
some differences. Korean Idols are usually a group, consisting of 4-6 members, although there
are rare cases of solo acts or as many as 13 in a group. Demographically, they are typically in
their teens and are predominantly made up of same-sex members. Musical sub-genre within the
Idol genre also exist. Dance music is the most popular sub-genre, although ballad, R&B and hip-
hop are also represented. Note that the sub-genres of Idol music typically bear little resemblance
to genres using these labels elsewhere in the world’s music scene. In part, this stems from the
fact that youthful and cheerful looks, even for all-male Idols, are nearly universal in Korean Idol
music.
Figures 1-6 illustrate the development of Idol music over our study period, and clearly
illustrate the rapid rise-and-fall characteristic of transitory products. 13,570 Idol songs were
released over this period, growing steadily in number over time as shown in Figure 1. Figure 2
shows that each week a small fraction of these songs would make the top 100 weekly rankings,
which includes Idol songs as well as songs from other popular music styles. Once popular, these
hits attracted huge fanfare, but before long would fall off the top 100 list as shown in Figure 3.
The net result was a steady increase in the number of popular Idol songs over the period, with the
genre occupying typically 10% to 50% of the top 100 as shown in Figure 4. The rapid rise-and-
fall of each song, as a function of duration, is shown in Figures 5 and 6. The steeper the slopes
of these plots, the higher is the hazard rate of a song making, and then falling off, the top 100 list,
respectively. Note that the rate of making the top 100 declined dramatically just days after a
song was released, such that after only 11 days a song that had not yet become popular would
7
essentially never become popular. Once popular, these songs fell off the top 100 list at a steady
rate, with all songs off the list within a year.
-----------------------------------
Insert Figures 1-6 about here
-----------------------------------
Perhaps the most notable implication of these patterns is the tremendous uncertainty
faced by the production agencies that are trying to create popular products in this context. One
typically sees large numbers of unsuccessful experiments in highly uncertain markets. Precisely
this pattern emerged in the K-Pop context, where only a small fraction of Idol songs became hits.
Of the 13,570 Idol songs released over our study period, only 2,316 (17%) became hits, as
defined by whether or not the song makes it to the top 100 weekly charts. Such uncertainty is a
formidable strategic and organizational challenge for music producers. In some cultural markets,
uncoordinated action in various, socially distinct contexts leads to the occasional discovery of
new, popular products in a diffuse process of experimentation (Salganick, Dodds, and Watts,
2006). Crossley (2015), for instance, documents the rise of the Punk movement in the UK from
localized social networks, just as Lena (2012) identifies performers and genre in the US starting
in the avant-garde. Yet Lena also shows that much innovation in music has been organized by
firms, originating from within the structure of the major labels.
In the case of Kpop, uncertainty in the production process is managed by organizations,
starting with talent training and group formation. Seabrook (2015) describes the system of
producing Kpop music and videos as a “machine,” with the major production agencies
controlling talent acquisition and training, songwriting, dance choreography, production,
8
distribution, and marketing. Trainees are recruited into a production agency during their early
teens, trained by the agency in dancing, singing, acting, and foreign languages for several years.
This process is highly selective, with only about 1 out of 10 trainees making it to their debut
(Rich, 2015). After training, the agencies form Idols by grouping trainees according to talents
and characteristics thought to be important to popularity at that point in time. For instance, SM
Entertainment, the largest and most successful production agency in Korea during our study
period, created the mega-hit group “Girl’s Generation” by combining 9 girls, each with an area
of specialty – English, Japanese, singing, dancing and rap (identifying reference).
The management of uncertainty in Kpop also includes the systematic creation and
production of songs and dances. Kpop music, lyrics, and dances typically are created by the
production agency, rather than by the musicians themselves. During our site visits to production
agencies, we sat in on planning meetings where real-time analysis of trends were used to
determine video and audio content. Seabrook (2015: 157) describes, “if Idols are successful,
they are often expected to churn out a full album every eighteen months or so, and a five-song
mini album each year. The charts change rapidly, and because youth and novelty are at such
premium, established groups usually don’t last long: five years is the average shelf life of an
Idol.” In fact, the saying “7-year jinx” is widely used by industry insiders to note that most Idol
groups are disbanded before or by their 7th year. These processes illustrate the extent of control
that the agency has over the Idols’ songs and their careers – and underscore that the Idol groups
are manufactured for the purpose of commercial success.
Yet these large, professionalized, multi-group production agencies are only part of the
industry. Our data include songs produced by 325 different agencies over the study period, 184
of which were small, single-band agencies that lacked elaborate organizational machinery. The
9
other 141 agencies are the multi-group agencies (69 of which grow from single to multi-group
over the study period). The single-group agencies are not only smaller than the multi-group
agencies, but by definition they have not developed routines for identifying and developing new
talent – the so-called “artists and repertoire” function that is well-developed in multi-group
agencies. Some multi-group agencies have created elaborate, global systems for talent
acquisition. Training programs are routinized within the multi-group agencies, developing
prospects and placing them into groups in order to fill out a portfolio of dancing and singing
abilities (identifying reference). Multi-group agencies also employ staffs of professional writers
to create content for their pipeline of oncoming Idol groups. Furthermore, marketing and
distribution systems are extensive in the multi-group agencies, targeting particular country and
demographic markets in order to fit features of new idol groups to the psychographic profiles of
fans (identifying reference). In these many ways, the contrast of multi-group and single-group
agencies allows us to determine how highly rationalized sourcing, production, marketing, and
distribution affects the popularity of songs. How has the manufacturing of popularity shaped
time-based competition in this industry? We address this question using an ecological model of
product demography.
Competing in Time
We model the competition of products in time as a “product ecology,” where the fates of
products depend on their features and on their competition with other products (Carroll et al.,
2010). Typically, modeling success and failure among products has involved mapping selection
processes, as firms adapt through the creation and failure of products (Sorenson, 2000). These
selection processes are particularly difficult for organizations to manage under time-based
10
competition. Generally, we know that the appearance of a new product confronts producers and
audiences with a categorization challenge (Hsu and Hannan, 2005; Hannan, Pólos, and Carroll,
2007; Pontikes, 2012). In most markets, over time as products diffuse, the process of adoption
requires categorization, and therefore leads to the emergence of classification systems (Etzion,
2014). But under the time limitations seen in popularity markets, producers must determine a
product’s features and release the product well before the trend is collectively understood.
Unable to resolve uncertainty, producers proliferate products – many of which will fail –
essentially “exploring” collectively for what will constitute the next big thing (March, 1991;
Sorenson, 2000).
For products competing over popularity, however, the birth of a product is only the first
step. Because most of the resources in popularity markets go to a handful of products that catch
on, the event of interest is the point when a product becomes popular. The initial creation of a
product is significant, but only because at that point a product becomes “at risk” of becoming
popular. Similarly, the “death” of a product in popularity markets is the point at which it ceases
to be popular – even if the product remains “alive” but obscure. In this way, the product ecology
of popularity markets involves a growth and decline process, where the two key events are the
discrete moments when products surge into, or fall out of, the spotlight.
Baseline models
In this spirit, our first model takes as a given that organizations create products, and then
depicts the rates at which these products experience the event of becoming popular, or their
“popularity rates”:
11
rpi = rpi* exp[γpt],
where rpi is the popularity rate of product i varying as a function of t, the time since the product
was released, γp is the duration effect to be estimated, and rpi* is the product-specific baseline
popularity rate. We can depict the dynamics that characterize time-sensitive popularity markets
in terms of two features of this model. First, we note that popularity markets are characterized
by considerable uncertainty over the features that will catch on in the next “new thing.”
Uncertainty over which products will succeed is especially great in popularity markets because
of the continuously changing social construction of the criteria that determine popularity – what
constitutes being “in.” Exemplifying this uncertainty, Figure 2 reveals very low baseline
popularity rates (rpi*) in the case of Kpop songs.
A second feature of this model is that it allows for duration dependence in the popularity
rate. This feature is particularly useful for describing popularity markets, because products in
such markets have a limited shelf-life during which they can become popular. In our model, the
limited time-at-risk for becoming popular implies γp<0, the pattern revealed for Kpop songs in
Figure 5. More generally, γp<0 indicates that, in contrast to organizations, “newness” per se
among fashion products is an advantage. However, γp<0 may also be the result of unobserved
heterogeneity, where winning products win soon and then the rest are left to wither. In either
case, a low baseline popularity rate (rpi*) that then declines with duration (γp<0) describes
popularity markets: where organizations introduce many products, but few catch on during their
brief window of opportunity.
Our second model describes the decline process of products that have become popular.
Because decline typically occurs suddenly in popularity markets, our model depicts the rates at
12
which popular products experience the event of falling into obscurity, which we refer to as their
“extinction rates”:
rei = rei* exp[γeτ],
where rei is the duration-dependent rate of extinction of product i, τ is the time since the product
became popular, γe is the duration effect to be estimated, and rei* is the product-specific baseline
extinction rate. A defining characteristic of popularity markets is that popularity is fleeting,
which implies a high baseline extinction rate rei* compared to products in markets where
advantage is more enduring. Clearly Figure 3 shows such a pattern for the case of Kpop songs,
with virtually all songs falling off the charts rapidly. Note also that duration dependence for the
extinction rate of the songs appears to be constant (see Figure 6), meaning that songs that do
endure on the charts remain in jeopardy of falling off the charts.
Organizational Strategies and Social Exposure
We assume that some organizations in popularity markets strategize in an attempt to
shape this baseline process. If some organizations can do this successfully, they will have
products with higher popularity rates and lower extinction rates. In the case of the Kpop market,
the multi-group production companies attempt to do just this. These firms manage multiple
groups, and in so doing systematically work to increase the popularity of their songs. In this
regard, one might inquire about differences in product quality, under the idea that some
organizations may have capabilities enabling them to produce higher quality products. If quality
is important to popularity, then the products created by these organizations are likely to enjoy
higher popularity rates and lower extinction rates. Here is where the high uncertainty of
13
popularity markets becomes important. High uncertainty over what constitutes quality at any
point in time is difficult for organizations to resolve in popularity markets. Only after
organizations see a trend do they have information about the features that defined quality in the
prior period. Such ex-post information collection will be of little use as a guide to forward-
looking product strategies.
For popularity products, and especially for music, uncertainty over quality raises a larger
question about whether quality can be understood as anything but socially constructed. Looking
at management fads, Strang and Macy (2001) demonstrate that boom-and-bust adoption patterns
can arise purely through imitation even without quality differences. And in a striking
experimental manipulation, Salganik, Dodds, and Watts (2006) demonstrate precisely such
faddish imitation among music listeners, where quality is controlled experimentally and powerful
social influence effects are revealed (cf. Berger and Le Mens, 2009). These authors conclude
that in the case of music, social influence increases both the inequality and the unpredictability of
success, with actual quality related to success but with a considerable amount of noise.
Consequently, the rates of success of popularity products – especially cultural products like
songs – differ especially because of differences in social visibility.
Organizations can attempt to increase the social exposure of their products through a
variety of tactics. Large, complex organizations routinely build capabilities in marketing meant
to increase the visibility of their products to audiences. For instance, Rossman (2015) found in
his analysis of hit songs in the US that the marketing and promotion activities of the major labels
are crucial to the success of songs. In Kpop, the large production companies conduct elaborate
announcements and launches of songs and groups, complete with advertising, live touring
appearances, outreach to fans and fan clubs, and merchandizing – all culminating in a song’s
14
release. It is important to note that unlike American agencies, which outsources various stages
of the song production process to external firms that specialize in the area (e.g., recording,
mastering), Korean agencies do everything from talent discovery, training, recording, etc.,
spanning the entire value chain. Because of this vertical integration, agencies have complete
control over how to strategize the song’s launch.
These activities increase the exposure of audiences to new songs, and create the
perception of widespread social acceptance of the song – both key to triggering the social
influence processes at work in the adoption of popularity products. Furthermore, the major Kpop
agencies explicitly attempt to construct systems of meanings around songs, groups, and the
Korean Wave (see Goldberg, 2011 on this idea more generally). If marketing effectively
increases the social exposure of products in popularity markets, then large organizations with
these capabilities will enjoy higher popularity rates and lower extinction rates among their
products.
Organizations can also increase the social exposure of their products by positioning them
in visible parts of the market. For example, product positioning in Kpop hinges on whether a
song is designated as the title track of an album. Although Idol songs are no longer typically
sold in CD form, but rather are watched through the internet or streamed online, they are
nonetheless packaged as “albums” featuring a title song. For instance, the hit “Gangnam Style”
referred to above was a title song. These well-positioned songs receive considerably more social
exposure. More generally, if product positioning increases social exposure in popularity
markets, then it will result in higher popularity rates and lower extinction rates for well-
positioned products.
15
A third way for organizations to increase the social exposure of their products is to
cultivate brand reputation. Branding is well known to be an effective way to increase product
recognition, and this holds for the music business generally and for Kpop where “brands” are the
identities of performing groups. When an established group releases a song, audiences will seek
out that release because it is from that group. This enhanced social exposure for “hit” groups is
likely to affect the perceived quality of their songs and the identity implications to the audience
of being associated with the song. Established, “star” groups will have an existing following of
fans who will be exposed to new songs immediately, during the key initial days after the song’s
release. This audience also will serve as a social referent for others considering listening to and
watching the new release.
We include these organizational strategies for increasing social exposure in our models of
the popularity and extinction rates:
rpi = rpi* exp[γpt] exp[βp'Ei]
rei = rei* exp[γeτ] exp[βe'Ei]
where Ei refers to observable differences between products in the factors that increase social
exposure: organizational marketing capabilities, product positioning, and brand reputation. βp
and βe reveal the effects of these social exposure tactics on the product popularity and extinction
rates, respectively. Estimating these effects allows us to investigate our arguments about the
importance of social exposure to success in popularity markets:
Social Exposure Hypothesis: βp > 0, βe < 0.
16
Organizational Strategies and Attention-based Competition
We also expect the rates of popularity and extinction to be shaped by competition among
songs. Studies of product demography in technology markets reveal important effects of
competition affecting product entry and exit rates (Bayus, 1998; Bayus and Putsis, 1999; Astebro
and Michela, 2005; de Figueiredo and Kyle, 2006). In a time-based ecology, the possibility of
competitive effects is especially interesting because increases in a given product’s popularity
draws attention away from other products. In this way, vying for “15 minutes of fame” is an
attention-based competition. The same organizational tactics that increase social exposure, if
effective, should also increase the competitive intensity of products in this time-based product
ecology. The intensity of this competition is included in our models:
rpi = rpi* exp[γpt] exp[βp'Ei] exp[αpNpj]
rei = rei* exp[γeτ] exp[βe'Ei] exp[αeNpj]
where Npj is the number of rival, popular products j faced by product i at a given point in time.
(“Rival” products are defined as the products of other, rival organizations – not including
products of the focal organization that markets product i.) The competition coefficients αp and αe
reveal the effect of rival, popular products on the popularity and extinction rates. By sub-setting
different groups of rival products, we can test to see whether the products offered by
organizations with rationalized production and marketing capabilities generate stronger
competition, driving down popularity rates and driving up extinction rates:
Attention-based Competition Hypothesis: αp < 0, αe > 0.
Our models can also investigate whether multi-product organizations “cannibalize” by
generating competition felt by their own products. Studies of dynamic markets often find
17
evidence that firms intentionally release products to stay ahead, even when this action causes
competition with their own products (Greenstein and Wade, 1998). Similarly, Iizuka (2007)
finds “planned obsolescence” in book publishing with the rise of the used book market, an effect
similar to cannibalization. In other cases, however, firms have been found to avoid
cannibalization, as in the theater business (Chisholm and Norman, 2006). In our models, we
allow for the effects of cannibalization in order to see whether the multi-group agencies in Kpop
generate competition for their own songs.
Data and Model Estimation
To estimate these models, we collected two forms of data describing Korean popular
music. The first is a weekly ranking of k-pop songs over 20 years (January 1, 1996 to January 1,
2016) from Melon.com, the largest music database and streaming service in Korea. This dataset
includes the following variables: Idol group’s full name, group id, song name, song genre, date
of the weekly ranking, weekly rank, album name, album release date, album genre and producer
agency.
Until 2004, the weekly charts ranked the top 50 most popular songs. The rankings were
calculated by 70% album sales and 30% broadcast performance. However, starting in late 2004,
the weekly charts expanded to include the top 100 most popular songs, and the methodology for
calculating the weekly rankings also changed. After 2004, rankings are calculated by frequency
of online music streaming (40%) and the number of downloads (60%). The shift from top 50 to
top 100 reflects the growth of k-pop as an industry, in terms of the number of entrants in this
space and, relatedly, increased competition. Moreover, the importance of online streaming and
downloads in the new ranking methodology mirrors the decline of broadcast and physical album
18
sales. In these data, there are 2,193 unique groups that are observed at least once. Among them,
15% (n=339) are Idols. However, in terms of the number of times they appear in the rankings,
Idols make up 27% of rankings. Due to our interest in the Idol segment, and for consistency in
our empirical demarcation of a “hit”, we restrict the observation period from November 29, 2004
to December 31, 2014, the period after the transition to Top 100.
The second form of data is a comprehensive life history of all Idols that existed between
January 1, 1996 and January 1, 2016, which coincides with the coverage of our weekly ranking
data. To avoid survivor bias, we gathered an exhaustive list of all Idols that were ever active by
consulting two databases (Melon.com, and maniadb.com). To make the observation period
consistent with the weekly ranking dataset, we only include groups that actively released a song
at any time from November 29, 2004 to December 31, 2014. During that period, there were 578
Idol groups that released at least one song, of which 302 has produced at least one hit.
Combining these two data sets allowed us to obtain estimates of the popularity and
extinction models. Each song’s life history was broken into weekly segments so that all
independent variables could be updated over time. Actual ranking and de-listing events were
known to the day, and so for those events segments were started and stopped so that the exact
day of each event was correctly recorded. To estimate the popularity models, a subset of the data
was created including all songs from the time of their release until they rank (or until the end of
the study period). These data are summarized in Table 1a. Duration for the popularity models
was then defined as the time from the release of a song. For the extinction models, the data were
subset to include all ranked songs starting on the day they first broke into the top 100. In this
dataset, duration was defined as the time since entering into the top-100 ranking. See Table 1b
19
for a summary of these data. For both analyses, all independent variables were recorded as of
the beginning of a given segment.
The models were specified as piecewise exponential functions. The piecewise
exponential model allows for duration dependence between pieces without making restrictive
assumptions about the functional form of duration dependence. This quality is especially
advantageous in this context, since there is no existing body of work looking into the functional
form of duration dependence in the popularity rate and extinction rate for songs. Estimates of the
models were obtained using Stata’s implementation, with robust standard errors clustered by Idol
group.
------------------------------
Insert Tables 1a and 1b about here
------------------------------
Variables
The dependent variables in the models are the transitions into and out of popularity.
Songs enter the risk set for experiencing this transition on the day they are released by their
production agency. We operationalized the transition to popularity as occurring on the day when
song i first ranks in the top 100 weekly charts. For song extinction, the risk set is defined as all
Idol songs that are on the top 100 weekly charts on a given day. Extinction is then
operationalized as occurring on the day when song i falls off the weekly charts.
For the independent variables, competition was modelled by including various forms of
Npj, the number of ranked, rival songs, lagged by one week so that they measure the number of
20
rival ranked songs at the start of a given one-week time segment. To distinguish different
competitive effects coming from, and felt by, single-group agencies and multi-group agencies,
these counts were further subdivided according to this organizational distinction. In making this
distinction, we define as “single group” a production agency that up until that point has managed
only one Idol group, while “multi-group” agencies have managed more than one Idol group as of
a given point in time. Note also that the various specifications of Npj include only “rival” songs
(designated by j) – meaning songs by groups other than the group that has released focal song i.
So defined, Npj can include rival songs from other groups that are managed by the same
production agency. Competitive effects within an agency would be “cannibalization,” and so to
test for such effects some models will distinguish between within- versus between-agency
competition.
The social exposure effects were modelled by including three different measures. First,
organizational marketing and distribution capabilities were measured by a binary variable
indicating whether a song was produced by a multi-group agency. Product positioning was
measured using a binary variable denoting whether a song is an album title, since these songs
receive greater social exposure through music videos, marketing around the song, and televised
performances. A third social exposure term was the number of previous hits by the group
performing a given song. This continuous variable measured the number of top-100 ranked
songs that the song’s group had prior to time t.
Several control variables are included in all models. To account for differences in
success rates between Kpop sub-genres, each song’s sub-genre is included as a fixed effect:
ballad, dance, drama, and R&B/soul. We also include calendar year in order to capture secular
trends in Kpop’s popularity over historical time.
21
Results
Tables 2a and 2b show the estimates of the popularity models. Tables 3a and 3b show the
estimates of the extinction models. All models include estimates of the effects of social exposure
specified as a function of whether a song was marketed by a multi-group agency, was a title
song, or was a song from a group with previous hits. All models also include the duration
effects, and then the different specifications in each table allow us to identify the sources and
impact of attention-based competition among songs. The final model in each table (Model 5 in
Table 2a/2b and Model 10 in Table 3a/3b) is the most complete, including the separate effects of
competition among and between songs by single-group agencies and multi-group agencies.
Starting with the duration effects in Table 2b, all models show dramatic negative duration
dependence in the popularity rate, mirroring the pattern observed in the nonparametric estimate
of Figure 5. After just 3 days, the likelihood of a song becoming popular falls precipitously –
and this is true across all models. And among all songs that rank, 79% do so within the first 7
days after release. By contrast, duration dependence in the extinction rate is relatively constant,
as shown in the duration estimates in Table 3b and consistent with the nonparametric estimate
shown in Figure 6.
--------------------------------------
Insert Tables 2a, 2b, 3a, and 3b about here
--------------------------------------
22
Next consider the social exposure variables – the effects of the number of prior hit songs
by a group, being a title track, or being produced by a multi-group agency. The effects of title
track and number of prior hits are positive and significant across all models in Table 2a and
negative and significant across all models in Table 3a. This pattern indicates higher rates of
popularity and lower rates of extinction for songs with greater social exposure due to market
position (title track) and reputation (prior hits by group). Yet the magnitudes of these effects
vary, so we review each in turn.
The number prior hits by a song’s group, at the average of that variable, make a song
34% times more likely to become popular according to the estimate in Model 5, the most
complete specification. As for the extinction rate, Model 10 estimates that songs are about 10%
less likely to fall off the charts due to the number of prior hits by the song’s group (on average).
Reputation at the level of the group strongly drives song popularity. The considerable hype
surrounding these Idol groups has powerfully affected the social exposure of their songs,
resulting in higher popularity rates and lower extinction rates.
Turning to the effects of market position - being an album title song - this social visibility
makes songs more likely to become popular and less likely to go extinct across all specifications
in Tables 2a and 3a. Model 5 shows a song to be over 12 times more likely to become popular as
a result of this exposure, an effect strong enough to offset much of the decline in popularity that
comes with duration. As for extinction, Model 10 estimates that being the title track makes
songs half as likely to fall of the charts. Popularity and extinction are strongly affected by
whether a song is a title track, our measure of market position.
We now turn to the organizational basis of social exposure, the effects of being produced
and marketed by a multi-group agency. According to the estimates in Models 1-3, this variable
23
has a positive effect on the popularity rate, and the effect is quite strong. The songs of multi-
group agencies are nearly twice as likely to become popular than are single-group agency songs.
However, this effect disappears in Models 4 and 5, which fully specify competition among
songs. In particular, both Models 4 and 5 reveal strong competition by multi-agency songs,
which drive down the popularity rate of single-agency songs. So the popularity advantage of
multi-agency songs in Models 1-3 operates through a competitive mechanism, where a lower
popularity rate among single-agency songs results from the competition they experienced from
multi-agency songs. Looking at the extinction models in Table 3a, the multi-group agency term
is never significant, but multi-agency songs generate competition in the extinction process.
Taken together, these results indicate that organizational effects on popularity and extinction in
these data hinge on competition.
To understand the estimates of competition in these models, an interpretative explanation
is in order. To allow for competition, the models include counts of the number of Idol group
songs in the rankings at any given time. The rankings, by definition, always contain 100 songs,
including Idol songs (studied here) as well as other songs not in this study. We assume that the
consideration set for deciding where to direct one’s attention, for Kpop fans, consists of Idol
songs. Consequently, the more other Idol songs are in the rankings, the more attention-based
competition felt by Idol songs trying to get into the rankings. In this light, a negative effect of
the number of ranked, rival songs on the popularity rate of an unranked song is evidence of
competition. Intuitively, such an effect means that the existence of those rival songs in the
ranking is making it less likely that the focal song will make the ranking. Similarly, in the
extinction models, a positive effect of the number of ranked, rival songs on the rate of a given
song leaving the ranking is also evidence of competition. In this case, the intuition is that the
24
existence of the rival songs in the rankings is making it more likely that the focal song will fall
off the charts.
Looking across Tables 2a and 3a, a striking pattern emerges from the various
specifications of competition. In both sets of models, statistically significant competition effects
are found only to be generated by the songs produced by multi-group agencies. Specifically,
Models 2 and 7 show competition among songs, such that more Idol songs in the top 100
decreases the chances that a new song will become popular, and increases the chances that a
popular song will be driven off the charts. Models 3 and 8 then reveal that these competitive
effects are coming entirely from multi-agency songs. In fact, the existence of ranked songs
marketed by single-group agencies make it more likely that other songs will become popular (see
Models 3-5), implying that when single-agency songs were in the rankings the pickings were
easy for songs seeking to become popular.
In the popularity models, competition from multi-group agency songs is felt only by
single-agency songs – and this competition is quite strong (see models 4 and 5). For instance,
the existence of 35 multi-group agency songs in the top 100 (the mean of that variable) decreases
the popularity rate for unranked songs by 60%.1 This powerful effect, furthermore, mediates the
advantage held by multi-group agency songs in terms of the popularity rate. After controlling for
competition from the multi-group agency songs felt by the single-group agency songs, the main
effect of multi- vs. single-group agencies falls to non-significance.
1 From Models 4 and 5, the effect of multi-group agency songs on the popularity rate of single-group agency songs
is -.0259. This estimate indicates a multiplier effect on the popularity rate of exp[-0259 x 35] at the observed mean
of 35 multi-group agency songs, which equals .40. A multiplier of .40 indicates a reduction of the rate to .40 of
what it would have been otherwise, or a 60% reduction.
25
Competition in the extinction models also is driven by the multi-group agency songs.
Model 7 shows that existing songs in the hit rankings drive up each other’s rate of falling off the
ranking. But Model 8 reveals this effect to be generated entirely by multi-group agency songs.
Models 9 and 10 then show that this competition is felt by all songs, but it is twice as strong
against the single group agency songs, more than doubling their extinction rate. Model 10 shows
that the weaker competitive effect of multi-group agency songs on other multi-group agency
songs holds when we create a separate term to isolate cannibalization among an agency’s own
songs. The coefficient of that term is large, suggesting perhaps that there is cannibalization or at
least intentional sequencing of marketing campaigns, but the effect is not statistically significant.
Discussion
Markets featuring time-based competition, with their distinctive pattern of rapid rise and
fall among individual products, are commonplace. Yet models of competition have rarely
attempted to explain such markets. In fact, most models of competition depict organizations
vying for resources at a given point in time over some resource space. In markets characterized
by time-based competition, by contrast, products vie for resources arrayed over time. Each
product has a narrow window of time during which it has the potential to rapidly grow popular
and thereby attract resources. Most products never make that transition, and those that do only
enjoy their success for a brief window of time until they are replaced by the next new thing. In
this light, how do organizations shape competition over popularity in such markets?
To address this question, we developed an ecological model of time-based competition.
Using data from Korean popular music, we were able to estimate the model and the results were
26
revealing. First, our decision to isolate the rates of song popularity and song extinction paid off.
Song popularity turned out to be strongly duration dependent, with nearly 80% of ranked songs
reaching popularity within the first 7 days from there release – and most of these within 3 days.
After just 11 days, any song not yet on the charts was unlikely to ever succeed. Mitigating
against these odds were differences among organizations that made some songs more likely to
become popular. Various sources of social exposure worked to the advantage of some songs.
And yet even with these effects the odds of any particular song succeeding remained quite low.
This challenge has been met by sophisticated production and marketing efforts among
some of the organizations competing in this context. The efforts of the large, multi-group
production agencies were able to increase the popularity of their songs by competing away rival
songs marketed by less sophisticated agencies, but this advantage also generated a self-defeating
mechanism. The successful songs produced by multi-group agencies generated considerable
competition among these agencies, dramatically shortening the period over which their
successful songs enjoyed popularity. These organizations have succeeded in manufacturing
popularity, but in so doing they have intensified the brevity of their successes.
Identifying these distinct effects required that we estimate an ecological model of time-
based competition. The advantage of our model is that it distinguishes patterns at the level of
individual songs from patterns due to competitive dynamics among songs. Yet such ecological
models are descriptive, and as such may potentially suffer from endogeneity. It is worth noting,
however, that endogeneity will typically work against finding evidence of competitive effects.
For example, looking at the effects of single-group agencies, songs marketed by these
organizations do not appear to generate competition in our estimates. In fact, in the popularity
models these songs increase the popularity of multi-group agency songs. Interpreted causally,
27
this finding would indicate that the market is made ripe for a song trying to become a hit when
the existing hits are from single-group agencies. But it is also possible that this effect is
somehow endogenous, reflecting a pattern where conditions favorable to becoming a hit helps all
Idol songs. Generalizing this concern, such endogeneity is likely across all the density effects,
and probably works to weaken evidence of competition where we find it.
Our results imply that research into competition among organizations would benefit from
explicitly modelling time-based competition more broadly. We estimated our models in a
market known for faddish popularity, where time-based competition is particularly acute. But
many other markets are also characterized by time-based competition: technology products, other
cultural products, and many manufacturing industries such as fast-fashion or markets
characterized by seasonality all feature time-based competition. Even competition among
organizations in some non-market contexts may feature time-based competition, such as
ideological competition among movements or candidates during waves of political fervor.
Similarly, markets that have already been analyzed by more traditional models of crowding at a
given moment in time might be revisited. Moving to the product level, and allowing for time-
based competition, research may find that competing over momentary popularity is far more
common than our current body of knowledge would suggest.
28
References
Abrahamson, E., 1991. Managerial Fads and Fashions: The Diffusion and Rejection of
Innovations. Academy of Management Review, 16(3), pp.586-612.
Askin, N. and Mauskapf, M., 2017. What Makes Popular Culture Popular? Product Features and
Optimal Differentiation in Music. American Sociological Review, 82(5), pp.910-944.
Åstebro, T. and Michela, J.L., 2005. Predictors of the Survival of Innovations. Journal of
Product Innovation Management, 22(4), pp.322-335.
Bayus, Barry L. 1998. “An Analysis of Product Lifetimes in a Technologically Dynamic
Industry.” Management Science, 44: 763-775.
Bayus, Barry L. and William P. Putsis, Jr. 1999. “Product proliferation: An empirical analysis of
product line determinants and market outcomes.” Marketing Science, 18: 137-153.
Berger, Jonah and Gael Le Mens. 2009. “How Adoption Speed Affects the Abandonment of
Cultural Tastes.” Proceedings of the National Academy of Sciences, 106:8146-8150.
Brynjolfsson, E., Hu, Y. and Simester, D., 2011. Goodbye Pareto Principle, Hello Long Tail: The
Effect of Search Costs on the Concentration of Product Sales. Management
Science, 57(8), pp.1373-1386.
Carroll, Glenn R., Olga M. Khessina, and David G. McKendrick. 2010. “The Social Lives of
Products: Analyzing Product Demography for Management Theory and Practice.”
Academy of Management Annals, 4: 157-203.
Chisholm, Darlene C. and George Norman. 2006. “When to Exit a Product: Evidence from the
U.S. Motion-Picture Exhibition Market.” American Economic Review, 96: 57-61.
29
Crossley, Nick. 2015. Networks of Sound, Style, and Subversion: The Punk and Post-punk
Worlds of Manchester, London, Liverpool, and Sheffield, 1975-80. Manchester:
Manchester University Press.
de Figueiredo, John M. and Margaret K. Kyle. 2006. “Surviving the Gales of Creative
Destruction: The Determinants of Product Turnover.” Strategic Management Journal, 27:
241-264.
Eisenhardt, K.M. and Schoonhoven, C.B., 1990. Organizational Growth: Linking Founding
Team, Strategy, Environment, and Growth among US Semiconductor Ventures, 1978-
1988. Administrative Science Quarterly, pp.504-529.
Etzion, Dror. 2014. “Diffusion as Classification.” Organization Science, 25: 420-437.
Fisher, M., 2012. Gangnam Style, dissected: The Subversive Message within South Korea’s
Music Video Sensation. The Atlantic, 23.
Fenn, J. and Raskino, M., 2008. Mastering the Hype Cycle: How to Choose the Right Innovation
at the Right Time. Harvard Business Press.
Goldberg, Amir. 2011. “Mapping Shared Understandings Using Relational Class Analysis: The
Case of the Cultural Omnivore Reexamined.” American Journal of Sociology, 116: 1397-
1436.
Greenstein, Shane M. and James B. Wade. 1998. “The Product Life Cycle in the Commercial
Mainframe Computer Market, 1968-1982.” RAND Journal of Economics, 29: 772-789.
Hannan, Michael T., László Pólos, and Glenn R. Carroll. 2007. Logics of Organization Theory:
Audiences, Codes, and Ecologies. Princeton: Princeton University Press.
Hsu, Greta and Michael T. Hannan. 2005. “Identities, Genres, and Organizational Forms.”
Organization Science, 16: 474-490.
30
Iizuka, Toshiaki. 2007. “An Empirical Analysis of Planned Obsolescence.” Journal of
Economics and Management Strategy, 16: 191-226.
Katila, R. and Ahuja, G., 2002. Something Old, Something New: A Longitudinal Study of
Search Behavior and New Product Introduction. Academy of Management Journal, 45(6),
pp.1183-1194.
Lena, Jennifer C. 2012. Banding Together: How Communities Create Genres in Popular Music.
Princeton: Princeton University Press.
March, James G. 1991. “Exploration and Exploitation in Organizational Learning.” Organization
Science, 2: 71-87.
Pontikes, Elizabeth G. 2012. “Two Sides of the Same Coin: How Ambiguous Classification
Affects Multiple Audiences’ Evaluations.” Administrative Science Quarterly, 57: 81-118.
Pontikes, Elizabeth G. and William P. Barnett. 2016. “The Non-consensus Entrepreneur:
Organizational Responses to Vital Events.” Administrative Science Quarterly,
forthcoming.
Porter, M.E., 2008. Competitive strategy: Techniques for Analyzing Industries and Competitors.
Simon and Schuster.
Rich, N. 2015. Hit Charade: Meet the bald Norwegians and Other Unknowns Who Actually
Create the Songs That Top the Charts. The Atlantic. October 2015.
Rossman, Gabriel. 2015. Climbing the Charts: What Radio Airplay Tells Us About the Diffusion
of Innovation. Princeton: Princeton University Press.
Salganik, Matthew J., Peter Sheridan Dodds, and Duncan J. Watts. 2006. “An Experimental
Study of Inequality and Unpredictability in an Artificial Cultural Market.” Science, 311:
854-856.
31
Seabrook, John. 2015. The Song Machine: Inside the Hit Factory. New York: W.W. Norton.
Sorenson, Olav. 2000. “Letting the Market Work for You: An evolutionary Perspective on
Product Strategy.” Strategic Management Journal, 21: 577:592.
Soule, S.A. and King, B.G., 2008. Competition and Resource Partitioning in Three Social
Movement Industries. American Journal of Sociology, 113(6), pp.1568-1610.
Stark, D. and Vedres, B., 2012. Political holes in the economy: The Business Network of
Partisan Firms in Hungary. American Sociological Review, 77(5), pp.700-722.
Strang, David and Michael W. Macy. 2001. “In Search of Excellence: Fads, Success Stories, and
Adaptive Emulation.” American Journal of Sociology, 107: 147-182.
32
Table 1a. Description of the Song Popularity Data
Independent variables Mean Std Dev Min Max
All Songs
# top rival Idol songs 35.25 9.57 8 58
# top rival Idol songs, single-agencies 0.81 1.84 0 17
# top rival Idol songs, multi-agencies 34.42 9.89 4 55
# top rival Idol songs, same multi-agency 0.01 0.16 0 13
Song produced by multi-group agency 0.78 0.41 0 1
Song is an album title 0.17 0.37 0 1
# previous hits by song’s group 12.29 14.78 0 80
Multi-group Agency Songs
# top rival Idol songs 35.68 9.45 8 58
# top rival Idol songs, single-agencies 0.72 1.64 0 17
# top rival Idol songs, multi-agencies 34.94 9.68 4 55
# top rival Idol songs, same multi-agency 0.01 0.18 0 13
Song is an album title 0.15 0.36 0 1
# previous hits by song’s group 13.41 14.68 0 80
Single-group Agency Songs
# top rival Idol songs 33.68 9.82 8 58
# top rival Idol songs, single-agencies 1.11 2.42 0 17
# top rival Idol songs, multi-agencies 32.56 10.40 4 55
Song is an album title 0.21 0.41 0 1
# previous hits by song’s group 8.26 14.45 0 70
33
Table 1b. Description of the Song Extinction Data
Independent variables Mean Std Dev Min Max
All Songs
# top rival Idol songs 35.78 9.52 8 58
# top rival Idol songs, single-agencies 0.98 2.18 0 17
# top rival Idol songs, multi-agencies 34.67 9.93 4 55
# top rival Idol songs, same multi-agency 0.03 0.36 0 10
Song produced by multi-group agency 0.87 0.34 0 1
Song is an album title 0.74 0.44 0 1
# previous hits by song’s group 16.86 16.03 0 80
Age of song when first in top 100 6 2.92 1 27
Multi-group Agency Songs
# top rival Idol songs 36.65 9.11 8 58
# top rival Idol songs, single-agencies 0.81 1.88 0 17
# top rival Idol songs, multi-agencies 35.68 9.40 4 55
# top rival Idol songs, same multi-agency 0.03 0.37 0 10
Song is an album title 0.77 0.42 0 1
# previous hits by song’s group 16.64 16.25 0 80
Age of song when first in top 100 5.77 2.62 1 25
Single-group Agency Songs
# top rival Idol songs 29.98 10.11 8 58
# top rival Idol songs, single-agencies 2.12 3.37 0 17
# top rival Idol songs, multi-agencies 27.77 10.57 4 55
Song is an album title 0.57 0.50 0 1
# previous hits by song’s group 18.30 14.44 0 70
Age of song when first in top 100 7.51 4.13 3 27
34
Table 2a. Piecewise Exponential Hazard Models of Song Popularity:
Estimated Entry Rate of Songs into the Top 100 by Korean “Idol” Groups, 2004-2014†
Independent variables Model 1 Model 2 Model 3 Model 4 Model 5
Song/Group/Org Characteristics
Song produced by multi-group agency .6374* .6643* .7204* .1045 .1073
(.1407) (.1406) (.1428) (.4796) (.4796)
Song is an album title 2.464* 2.476* 2.495* 2.500* 2.501*
(.1365) (.1344) (.1347) (.1349) (.1349)
# previous hits by song’s group .0247* .0248* .0244* .0239* .0239*
(.0033) (.0034) (.0033) (.0036) (.0036)
Competitive effects on all songs
# rival songs (ranked in the top100) -.0110*
(.0046)
# rival songs by single-group agencies .0816*
(.0212)
# rival songs by multi-group agencies -.0094*
(.0042)
Competitive effects
on single-group agency songs
# rival songs by single-group agencies .0764* .0764*
(.0315) (.0314)
# rival songs by multi-group agencies -.0259* -.0259*
(.0126) (.0126)
Competitive effects
on multi-group agency songs
# rival songs by single-group agencies .0756* .0757*
(.0284) (.0284)
# rival songs by multi-group agencies -.0061
(.0049)
# rival songs by other multi agencies -.0062
(.0048)
# rival songs by same multi agency .1690
(.0968)
Log-likelihood -5644.09 -5632.42 -5591.36 -5585.14 -5583.61
Degrees of freedom 13 14 15 17 18 †All models include fixed effects for sub-genre: ballad, dance, drama, R&B/soul. Data include 2,315 ranking
events among 13,585 songs released by 579 groups. Duration effects reported in Table 2b.
*p<.05, robust standard errors (clustering on groups) in parentheses.
35
Table 2b. Duration effects from the Song Popularity Models
Independent variables Model 1 Model 2 Model 3 Model 4 Model 5
Duration 0-3 days from release -8.176* -7.848* -8.074* -7.546* -7.550*
(.4017) (.4350) (.4276) (.6077) (.6079)
Duration 3-5 days from release -6.536* -6.207* -6.430* -5.903* -5.908*
(.3862) (.4142) (.4080) (.5908) (.5910)
Duration 5-7 days from release -6.141* -5.812* -6.029* -5.502* -5.507*
(.3778) (.4029) (.4011) (.6013) (.6017)
Duration 7-14 days from release -7.798* -7.469* -7.685* -7.156* -7.162*
(.3732) (.4057) (.4035) (.5992) (.5994)
Duration 14-21 days from release -9.594* -9.263* -9.470* -8.941* -8.946*
(.5152) (.5416) (.5292) (.6937) (.6939)
Duration above 21 days from release -11.40* -11.07* -11.27* -10.74* -10.75* (.7346) (.7489) (.7515) (.9016) (.9017)
*p<.05, robust standard errors (clustering on groups) in parentheses.
36
Table 3a. Piecewise Exponential Hazard Models of Song Extinction:
Estimated Exit Rate of Songs from the Top 100 by Korean “Idol” Groups, 2004-2014†
Independent variables Model 6 Model 7 Model 8 Model 9 Model 10
Song/Group/Org Characteristics
Song produced by multi-group agency -.1227 -.1925 -.2103 .1705 .1686
(.1338) (.1293) (.1342) (.3905) (.3888)
Song is an album title -.6521* -.6631* -.6781* -.6790* -.6764*
(.0750) (.0726) (.0691) (.0681) (.0685)
# previous hits by song’s group -.0059* -.0067* -.0067* -.0064* -.0063*
(.0029) (.0029) (.0029) (.0030) (.0030)
Competitive effects on all songs
# rival songs (ranked in the top100) .0130*
(.0035)
# rival songs by single-group agencies -.0050
(.0157)
# rival songs by multi-group agencies .0126*
(.0034)
Competitive effects
on single-group agency songs
# rival songs by single-group agencies -.0119 -.0117
(.0175) (.0175)
# rival songs by multi-group agencies .0232* .0231*
(.0096) (.0095)
Competitive effects
on multi-group agency songs
# rival songs by single-group agencies .0005 .0008
(.0214) (.0214)
# rival songs by multi-group agencies .0105*
(.0037)
# rival songs by other multi agencies .0105*
(.0037)
# rival songs by same multi agency .0477
(.0463)
Log-likelihood -3004.02 -2987.99 -2987.12 -2984.17 -2983.86
Degrees of freedom 13 14 15 17 18 †All models include fixed effects for sub-genre: ballad, dance, drama, R&B/soul. Data include 2,290 song exit
events among 2,321 songs at risk over 21,389 song-week segments. Duration effects reported in Table 3b.
*p<.05, robust standard errors (clustering on groups) in parentheses.
37
Table 3b. Duration effects from the Song Extinction Models
Independent variables Model 6 Model 7 Model 8 Model 9 Model 10
Duration 0-14 days in top 100 -4.010* -4.370* -4.304* -4.604* -4.609*
(.2334) (.2516) (.2472) (.4216) (.4206)
Duration 14-30 days in top 100 -3.572* -3.938* -3.874* -4.172* -4.178*
(.2397) (.2630) (.2561) (.4339) (.4329)
Duration 30-90 days in top 100 -3.927* -4.278* -4.216* -4.509* -4.514*
(.2430) (.2642) (.2584) (.4257) (.4247)
Duration 90-180 days in top 100 -3.524* -3.857* -3.796* -4.086* -4.091*
(.2594) (.2774) (.2710) (.4366) (.4354)
Duration 180-270 days in top 100 -3.199* -3.511* -3.438* -3.731* -3.735*
(.2682) (.2823) (.2790) (.4321) (.4306)
Duration above 270 days in top 100 -3.772* -4.110* -4.055* -4.353* -4.357* (.3971) (.4216) (.4159) (.5541) (.5539)
*p<.05, robust standard errors (clustering on groups) in parentheses.
38
Figure 1. Number of Idol Song Releases Each Week
39
Figure 2. Number of Idol Song Entries into Top 100
40
Figure 3. Number of Idol Song Exits From Top 100
41
Figure 4. Songs by Kpop Idols as a proportion of the Korean Top 100
42
Figure 5. Kaplan-Meier Estimate of Log Survivor Function of Ranking in Top 100
43
Figure 6. Kaplan-Meier Estimate of Log Survivor Function for De-Listing from Top 100