A survival analysis of songs on digital music platform

Telematics and Informatics xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Telematics and Informatics journal homepage: www.elsevier.com/locate/tele

A survival analysis of songs on digital music platform ⁎

Hyunsuk Ima, Haeyeop Songb, , Jaemin Jungc a

Mobile Communications Business, Samsung Electronics, Gyeonggi-do, Republic of Korea Department of Media & Culture, Kunsan National University, Gunsan, Republic of Korea c Graduate School of Information and Media Management, KAIST, Seoul, Republic of Korea b

A R T IC LE I N F O

ABS TRA CT

Keywords: Survival analysis Digital music Download Streaming

This study explores and compares what factors are critical for music to succeed in download and streaming services. The weekly top 100 songs listed on the Korean music ranking charts for three years were used as the sample, and the factors affecting success on the download and streaming ranking charts were examined using the survival analysis. The results indicate that being the title track is the most critical factor for songs’ survival on the charts for both services. While songs released by major labels survived longer on the download chart, major labels are no longer superior to minor labels in streaming services. The results also indicate the impact of the superstar was positive to the survival on the streaming chart, but it was effective only in the interaction with being title track on the download. As expected, piracy showed a negative influence on the survival of songs for both download and streaming service. Theoretical and practical implications for the digital music industry were suggested.

1. Introduction The success of traditional music industry is determined by the quantity of albums sold, which is then ranked on charts. The appearance and continued presence on the charts influences the awareness, perceptions, and profits of an album (Bradlow and Fader, 2001). Therefore, having an album on the music ranking charts is an important goal for most popular music artists and their record labels (Strobl and Tucker, 2000). In is no doubt that previous studies have examined the factors affecting the success of music on ranking charts (Asai, 2008; Bhattacharjee et al., 2007; Strobl and Tucker, 2000). Several variables, including debut rank, star effect, or tie in with other media, have consistent results through all studies. Meanwhile, some variables, such as major label, have shown inconsistent results. Although previous studies found the critical determinants of music hits, those studies focused on the physical format such as CD album. In the changing music industry, while revenue from digital download and streaming consists of more than 75% of the industry revenues, the ratio of physical format sales is only limited to 21 percent. Furthermore, streaming grew from just 9% of the market in 2011 to more than half (51.4%) of total industry revenues in 2016 (RIAA, 2017). Whether the factors affecting the success of physical music sales previously are still effective in the digital music market is questionable. Thus, this study focuses on figuring out the factors affecting songs’ success in digital market. By using the established determinants such as debut rank, superstar, major label, or tie-in, and an additional variable ‘title track’ exploring the unbundling effect in digital music industry, this study attempts to identify the factors affecting a song’s survival time on the download and streaming charts. Furthermore, the effect of piracy threatening the digital music industry is also empirically tested. Since the survival time on the charts symbolizes the popular life of a song, it is the object of our analysis.

⁎

Corresponding author. E-mail addresses: [email protected] (H. Im), [email protected] (H. Song), [email protected] (J. Jung).

https://doi.org/10.1016/j.tele.2018.04.013 Received 29 August 2017; Received in revised form 22 February 2018; Accepted 23 April 2018 0736-5853/ © 2018 Published by Elsevier Ltd.

Please cite this article as: Im, H., Telematics and Informatics (2018), https://doi.org/10.1016/j.tele.2018.04.013

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

This study conducts a survival analysis of digital music market by analyzing the weekly digital top 100 hit charts in South Korea. Although a relatively small country, with a population of 47 million, South Korea has a highly developed broadband infrastructure and Koreans crave information and communication technology (ICT) services (Choi et al., 2013; Jung et al., 2017). Large-scale digital content supply and consumption are facilitated through diverse digital devices such as smartphones and tablet computers (Jung et al., 2017). In fact, Korea is a leading country for paid music streaming services. The IFPI report (IFPI, 2016a,b) shows that 41 percent of internet users in Korea subscribe to a paid music streaming service as of the end of 2016. This is the largest proportion followed by Sweden, Mexico, Brazil, and the U.S. Furthermore, the increasing popularity of Korean pop (K-pop) music has yielded the “Korean Wave” (pronounced Hallyu in South Korean), which refers to the popularity of South Korean culture throughout other Asian countries (Ryoo, 2009) and is now extending to Western countries. An example of recent K-pop popularity from a global perspective is the song Gangnam Style by Psy, which was ranked number one in YouTube views, reaching more than 2.5 billion in 2016. BTS, also known as the Bangtan Boys, received worldwide recognition by winning the Top Social Artist Award at the Billboard Music Award and became the first K-pop group to perform at the American Music Awards (Kelly, 2017; Liu, 2017). As such, South Korean music industry is noteworthy given its highly developed broadband infrastructure and music popularity. As one of the world trendsetters in digital content production and consumption, the findings from the Korean experience might be extrapolated to better understand the nature of digital music survival in other country settings. 2. Literature review 2.1. Survival analysis research Survival analysis is a branch of statistics for analyzing the expected duration of time until one or more events happen. The event can be death, occurrence of a disease, failure in mechanical systems, divorce, etc. and the time to event or survival time can be measured in days, weeks, or years, etc. It deals with time-to-event data which is complicated not only by the dynamic nature of events occurring in time but also by censoring where some events are not observed directly (Klein et al., 2016). Historically its origin to classic life table construction begun in the 1600 s and it have been applied to the various field of academia. Although the researchers who are interested in survival analysis deal with time-to-event data, the specific event in which is interested depend on their research area. For example, the lifetime of patients is of interest to medical researchers, while the failure time of manufactured items is the area of interest in engineering studies. Marketing researchers have also focused on inter-purchase time to investigate how often consumers purchase a certain product. Financial researchers predicted financial distress such as bankruptcy or business failure using survival model. Empirical research has been conducted using survival analysis methods in various fields. For example, Mazzaferri and Jhiang (1994) applied the survival analysis to find the effect of medical and surgical treatment on cancer. Chen and Lee (1993) focused on the failure of oil and gas industry, and Kauffman and Wang (2008) used survival analysis to examine the factors of the Internet business failures. Survival analysis is employed by a number of research fields to analyze the time of events. This study applied the survival analysis method to the music ranking chart and focused on the events that occur in ranking chart over time in the music industry. 2.2. Survival analysis of music industry Music is a typical experience good in that consumers recognize its value after its consumption (Nelson, 1970). These products require a personal experience, such as sampling or purchase, to evaluate quality. However, sampling or experiencing music requires considerable time and effort (Bhattacharjee et al., 2007). For example, Apple Music had over 40 million songs in 2016 (Apple, 2016), and the number of song is increasing quickly. Consumers cannot afford to listen to every song to determine what they want. Thus, they occasionally depend on predecessors’ choices to assess a song’s quality. Music is also a fashion-oriented product because it includes features of social utility that satisfy interpersonal needs (Chen et al., 2008), and the popularization of music represents a phenomenon of fashion. Thus, a consumer’s choice of music can be influenced by others. Consumers who are highly involved in fashion may perceive music with higher value when it has social utility (Chen et al., 2008). A music ranking chart is a definite indicator of music consumers’ preferences. Since popular songs are ranked on the music charts, consumers naturally depend on the charts to either reduce their risk of choice or enable them to follow the trends. Although many albums and songs are released every year, only a few of them are ranked on the charts, are lucrative, and achieve great success (Seabrook, 2003). Furthermore, having a song on the chart leads to an information cascade so that songs garner attention and increase current and future sales (Strobl and Tucker, 2000). Thus, having an album or song on the top 100 chart is the primary goal of both artists and their record labels. Previous studies considered any album on a chart as a “hit,” and examined the length of time the album remained on the chart as an object of analysis (Asai, 2008; Bhattacharjee et al., 2007; Strobl and Tucker, 2000). Those studies focused on the physical format such as an entire CD album. However, this study focuses on a song’s survival time on the chart because people now can buy one song from an album in the digital music industry. If a song continues to stay on the charts before dropping off, it is regarded as “surviving,” and otherwise “dying.” This study conducts a survival analysis and compares the factors affecting the survival time between songs on the download and streaming charts. 2

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

Table 1 Factors affecting survival on music charts. Authors

Period of study

Variables

Significance on survival time

Data

Strobl and Tucker (2000)

1991–1993

Soundtrack Greatest hit album Pre-Xmas Re-entered Top Bottom

Positive Positive Negative Negative Positive Negative

U.K. 50 popular music album chart

Bhattacharjee et al. (2007)

1995–1997 2000–2002

Debut rank Albums released Superstar Minor label Solo male Group Holiday

Negative n.s. Positive Negative Negative Negative Positive

Billboard top 100 album chart

Asai (2008)

1990 2004

Genre Star Tie-in Company

Negative Positive Positive n.s.

Japan Oricon top 100 singles and albums chart

2.3. Factors affecting survival of music Researchers have verified a variety of factors that affect the survival time and success of music on music ranking charts ( Asai, 2008; Bhattacharjee et al., 2007; Strobl and Tucker, 2000), as shown in Table 1. Several variables, including debut rank, star effect, or tie in with other media, have consistent results through all studies. For example, a top ranked song on the chart (Strobl & Tucker, 2000) or higher debut ranked song (Bhattacharjee et al., 2007) has a positive effect on survival time. A superstar effect is also beneficial for the length of survival. Meanwhile, the impact of being represented by a major company is inconsistent. Bhattacharjee et al. (2007) argued that since the major labels exert substantial control over recording, distribution, and promotion, the albums of major labels survive longer. However, Asai (2008) stated that albums of small-scale companies are not at a disadvantage since the share of major record companies is relatively low in Japan. While previous studies have verified the factors affecting the survival time of an album, this study focuses on the survival time of a song. In the digital music industry, consumers no longer need to buy an entire album. They can purchase only one song which they want as a result of the unbundling. Since an album generally consists of one major song ‘title track’ and the rest, this study includes a ‘title track’ variable to examine the effect of unbundling on music industry. By using previously studied variables and an additional variable, this study explores the factors affecting the survival of a song in the Korean digital music industry. In addition, the effect of piracy on digital music industry is also examined. These improve our understanding of the digital music market, and the difference in factors affecting songs’ success on the download and streaming chart in Korea. The explanatory variables are explained in detail as follows. 2.3.1. Initial debut rank When people make decisions, they would abandon personal information and follow other peoples’ actions. For example, once a song becomes popular and ranked on a chart, more people tend to listen to the song. This phenomenon is called the “bandwagon effect” by Leibenstein (1950). The bandwagon effect is characterized by the probability of an individual’s adoption increasing with respect to the proportion of consumers who have already done so. As more people come to believe in something, others “hop on the bandwagon” regardless of the underlying evidence. The bandwagon effect explains why there are fashion trends. Music is a typical fashion-oriented product since consumers’ tastes and preferences rapidly change and are easily influenced by other consumers. Music ranking stimulates herd behavior in consumers, and thus generates the bandwagon effect. Once songs are listed on the ranking charts based on the behavior of early adopters, the remaining consumers also become interested (Bhattacharjee et al., 2007; Strobl and Tucker, 2000). In particular, people are less likely to listen to or download a song that is ranked lower due to the extra effort required to scroll through the chart (Chen, 2009; Yoo and Kim, 2012). Due to this visibility, higher ranked songs receive more attention and survive longer. 2.3.2. Superstar Relatively fewer people earn enormous amounts of money and dominate the fields in which they engage, known as “the phenomenon of superstars” (Rosen, 1981). Adler (1985) found that when consumption requires knowledge, the phenomenon of superstars can minimize the time and effort required to search for products. Music is a typical experience product where the value is revealed only after its consumption (Nelson, 1970). Initially, music consumers do not have accurate information about the quality of a song. Since the search for information is costly, consumers naturally depend on a product’s reputation. Previous studies examined the correlation between album success and superstardom, and determined that the phenomenon exists in the music industry (Chung and Cox, 1994; Hamlen, 1991; MacDonald, 1988; Towse, 1992). Furthermore, Asai (2008) and Bhattacharjee et al. (2007) included 3

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

the superstar factor in their survival model and found a positive effect on survival time. 2.3.3. Title track The process of “unbundling” is prevalent and influential in digital content industry. Traditional publishers sell access to some books by a pay-per-page-view or individual chapter sales. Moreover, many traditional print media companies, such as The New York Times and The Economist, have also unbundled content and sell each article for a small fee. The music industry has already undergone the same changes. Elberse (2010) found that digital music consumers are buying more music than before, but revenues from the individual songs are not enough to offset the resulting drop in album purchases. This result implies that consumption is focused on the most popular or the title song, rather than on the whole album. In the past, music was released on an album that contained the most commercialized song, referred to as the title track, and a number of other songs. Consumers had to purchase an entire album even if they liked only one song. However, people now buy or stream their one or two favorite songs instead of buying the whole album. Moreover, since promotion focuses on the title track, a title track is more exposed and will have a longer survival time than the remaining songs. 2.3.4. Major label Only a few music companies dominate more than half of the music distribution industry. Three major record labels (Universal Music Group, Warner Music Group, and Sony Entertainment) account for almost 70% of the world’s music distribution market (Music and Copyright, 2014). The labels’ strong financial base and wide network of major companies helps artists record, distribute, and promote their albums, which leads to easier access to a larger customer base (Strobl and Tucker, 2000). On the other hand, minor labels, called indie companies, are small and nimble. Minor labels lack the resources to attract leading artists to their labels and their limited distribution networks restrict their activities in the market. Therefore, small labels tend to target to niche markets (Spellman, 2006). Previous studies empirically showed that music albums produced by major labels have different influences on songs’ survival times in the U.S. and Japan (Asai, 2008; Bhattacharjee et al., 2007). Thus, this study examines the major label variable to determine the capital power of large-scale record companies. 2.3.5. Tie-in with other media Songs are often incorporated into an original soundtrack (OST) or as background music for movies or TV dramas. Some songs also receive more attention from popular audition shows. This “tie-in” with other media organizations can increase media exposure, resulting in increasing attraction to the song. Thus, the tie-in system has been a strategy to promote sales since the 1980s (Asai, 2008). This concept is also called synergy, a marketing strategy in which motion picture and recording industries simultaneously promote a single product (Denisoff and Plasketes, 1990). Strobl and Tucker (2000) showed that a soundtrack survives longer on the chart due to the popularity of the film and greater audience exposure. Asai (2008) also empirically verified that a tie-in strategy using other media can be an effective marketing strategy for promoting singles. Since many Korean songs have been released with a tie-into other media, this study also examines the effect of tie-in as an independent variable. 2.3.6. Piracy There have been conflicting evidences regarding whether piracy affects the sales of music. For example, Oberholzer-Gee and Strumpf (2007) showed that the effect of illegal file-sharing on record sales has no statistical significance. Meanwhile, Gopal, Bhattacharjee and Sanders (2006) found that online music sharing increases the propensity to buy lesser known artists’ albums. On the other hand, Liebowitz (2006) revealed that file-sharing has a close relationship with the decline of record sales. Bhattacharjee et al. (2007) also found that file sharing has a negative impact on low-ranked albums. Compared to more popular songs, there are less information and cues on less popular songs. Thus, there is a possibility for less popular songs to be damaged from sharing technology. In the streaming era, the situation is a bit different. There is no limit for experiencing and sampling music. The cost to experience music and gain knowledge on artist is almost zero. Weijters and Goedertier (2016) alluded that file sharing does not affect the music sales in streaming, while download services can be damaged. Even, Aguiar and Waldfogel (2017) found that the sales of a streaming service, Spotify, displace music piracy. This study also empirically investigates the impact of piracy on the download and streaming services from the perspective of the survival time on the chart. 2.3.7. Male group and female group An idol is an image or material object that represents a deity, or any person or thing regarding admiration, adoration, or devotion, and the term is now extensively used to include anyone who is high-spirited or worthy of respect. An idol could be an athlete, a politician, a star, a writer, a singer, or even a figure in a cartoon or comic (Chiou et al., 2005). Idolatry is an important factor that affects songs’ survival times. Wang et al. (2009) showed the positive effect of idolatry on the purchase of music CDs, and Ouellet (2007) described how consumers’ response to performers can influence their purchase of music. Improving the relationship between consumers and the artist increases consumer loyalty and a likelihood that consumers maintain an honest relationship with the artist (Chiou et al., 2005; Ouellet, 2007). The effect of idolatry can be explained by the social identity theory that consumers’ identification with a group is related to the intention to purchase sponsors’ products (Cornwell and Coote, 2005) or licensed merchandise (Kwon and Armstrong, 2002). In other words, consumers who idolize an artist are likely to make an emotional connection or attachment through sponsored property. Korea has a unique culture that idolizes groups of male and female artists. Fans are likely to download or stream their idol’s songs to make an emotional connection and attachment. Thus, the male group and female group variables are considered to examine the 4

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

effect of idolatry. 2.3.8. Control variables Increased competition is more likely when there are more songs released each year. Thus, the total number of songs registered in each year was controlled. In addition, the release of new music albums and overall album sales can vary depending on time. For example, the success of music albums may be influenced if they are released during the Christmas holiday period (Bhattacharjee et al., 2007; Montgomery et al., 2000; Strobl and Tucker, 2000). Thus, the timing of release is also controlled. 3. Methods 3.1. Model A number of models enable researchers to estimate the survival probability and hazard rate. Depending on whether the survival time data can be parameterized, these models are classified into three types: parametric, semiparametric, and nonparametric models, and each has advantages and disadvantages. Parametric models specify how the hazard varies over time. They assume the underlying hazard using several parametric distributions, such as log-normal and Weibull. Although a more sophisticated analysis is possible with a parametric model, if the wrong distribution is used, the conclusion may be biased. In contrast, nonparametric and semiparametric models provide flexibility in modeling data. A nonparametric model is a way to estimate the probability of the survival time without assuming its shape. Since these models require no assumption about the distribution of survival time, they allow the data to describe itself. However, the nonparametric analyses can only compare the survival functions of a limited number of groups, and examining the effect of one independent variable while controlling for other variables is difficult. In addition, they cannot handle quantitative variables. Fortunately, a semiparametric model is an intermediate technique that can handle the influence of the explanatory variables, even though a baseline hazard function is assumed to be model-free. Among these models, this study uses the nonparametric Kaplan-Meier method and the semiparametric Cox proportional hazard model. 3.1.1. Kaplan-Meier method The data is analyzed using the Kaplan-Meier method to verify what service leads to a longer survival time on the chart (Kaplan and Meier, 1958). The Kaplan-Meier estimate measures the fraction of subjects living for a certain amount of time after a treatment. The method includes censored data such that subjects who are lost at any point in time can be used in analysis. By comparing two Kaplan-Meier plots for the download chart and streaming chart, we can determine the difference in survival times. 3.1.2. Cox proportional hazard model Although the difference between the survival times can be verified, nonparametric models do not address the influence of each explanatory variable. Thus, this study also uses the Cox proportional hazard model, one of semi-parametric models. Proportional hazards models can assess the impact of variables on the likelihood of an event occurring. The proportional hazard model devised by Cox (1972) is presented as:

hi (t;x ) = h 0 (t )·exp(γ ′ xi ) This model contains a parametric function, γ’, and a nonparametric function, h 0 (t ) . The nonparametric function is a baseline hazard function and is a function of the time, not the individual characteristics, and the unspecified effect of time on the hazard rate. The parametric function consists of an exponential function of covariates and their coefficients. The explanatory variables x i are used to predict an individual’s hazard. For each covariate x i in the Cox model, we report the exponential form of the coefficient beta, known as the hazards ratio. A hazard ratio greater than 1.0 for variable x i indicates that it is positively associated with the probability of default, whereas a ratio less than 1.0 indicates that x i is negatively associated with the probability of default. 3.2. Data Survival analysis is different from the other methods due to its distinctive properties of skewness and censoring. Survival times are always non-negative and positively skewed. To estimate the shape of the distribution, exponential or Weibull parametric models are used. Fig. 1 illustrates the survival histories of three songs from March 2011 to March 2012. Song A entered the chart in March 2011 and dropped out (died) in July 2011. Song B entered the chart in July 2011 and died in November 2011. These observations are referred to as “uncensored” since they died within the period of the study. Finally, song C entered the chart in October 2011 and was alive until the end of the observation period. The song actually dropped out in May 2012, but this was unobserved and thus cannot be considered in the analysis. This observation is referred to as “censored.” Since survival analysis does not allow for center-censored data, the center-censored data were eliminated. For example, songs that dropped off the chart for several weeks and then reappeared were excluded from our data. The number of songs that dropped off was 190 out of 4716 for download services, and 142 out of 2914 for the streaming services. Thus, 4526 songs for download services and 2772 songs for streaming services were used in the analysis. The difference between the numbers of songs indicates that some songs are more lucrative in streaming services than download services. The data were collected and combined from five sources: 1) The Gaon music chart (http://www.gaonchart.co.kr), which is a 5

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

Fig. 1. Censoring.

music chart that ranks the popularity of songs or albums in South Korea, 2) Melon (http://www.melon.com), the largest digital music retailer in South Korea, 3) The Korea Music Copyright Association (http://www.komca.or.kr), 4) The Korea stock market, and 5) The Korea Copyright Protection Center (http://www.cleancopyright.or.kr). The Gaon music chart is the leading music chart that aggregates all online and offline data provided by record labels, distributors, and web-based music providers. The chart has public confidence in South Korea and was established to be a national chart analogous to the Billboard chart of the U.S. and the Oricon chart of Japan. The chart is compiled by the Korea Music Content Industry Association and supported by South Korea’s Ministry of Culture, Sports, and Tourism. The chart is published weekly and includes the popularity and downloading or streaming count of each song or album. A web crawler was developed to retrieve all public information for this study. The weekly rankings of the songs on the Gaon Music top 100 chart were collected from February 2011 to February 2014. Melon.com is the biggest music distributor in the Korean music market and accounts for 55% of the market share by sales. The characteristics of the songs (e.g., male/female group, title) were gathered from Melon.com. Major labels and minor labels are classified depending on whether a company is listed on Korea stock market. Company names listed on the stock market are as follows: SM, YG entertainment, JYP Ent., Loen entertainment, CJ E&M, FNC. They are classified as major labels, coupled with the three worldwide major companies, Universal Music Group, Warner Music Group, and Sony Entertainment. The Korea Music Copyright Association maintains a database of all songs that are released. The total number of registered songs during each time segment was

Table 2 Operational definition of variables. Variable

Definition

Source

Length of Survival Initial Debut Rank

Number of weeks a song appears on the top 100 charts, The rank at which a song debuts on the top 100 chart. Numerically lower ranked songs are more popular. (e.g. Rank 1 is the most popular) A binary variable representing the reputation of the artist. If a given song’s artist has previously appeared on the charts within the top 10th for the past three years, prior to the current song’s debut, then the variable is set to 1, otherwise 0. A binary variable that denotes if a song is an album’s title track. A binary variable that is set to 1 if the production label is one of the companies listed on the stock market. A value of 0 represents independent and small music labels. A binary variable representing if a song is produced and distributed by a broadcasting station (MBC, KBS, SBS, CJ E&M). A binary variable that denotes it the song’s artist is a male group. A binary variable that denotes if a song’s artist is a female group. A binary variable to control for the holiday effect (“Christmas effect”). If the album debuted in December, then the variable is set to 1, otherwise 0. Number of songs released during each time segment of the study period.

Gaon music chart

Superstar

Title track Major Label Tie-in with Other Media Male Group Female Group Holiday Number of Songs Released Number of Piracy Users Number of Pirated Content

Total number of piracy users caught by the copyright protection center during the song’s debut month Total number of pirated content caught by the copyright center during the song’s debut month

6

Melon.com Korea stock market Gaon music chart Melon.com Melon.com Korea Music Copyright Association Korea Copyright Protection Center

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

Table 3 Descriptive statistics of download and streaming charts.

Number of songs Average length of survival (week) Initial debut rank Superstar (%) Title track (%) Major label (%) Tie-in with other media (%) Male group (%) Female group (%) Holiday (%) Number of songs released Number of piracy users Number of pirated content

Download

Streaming

4526 3.1 45.8 35.1 64.6 23.8 25.3 27.8 11.5 8.5 30570.5 3862.8 263349.0

2772 5.2 58.4 37.0 76.2 24.3 26.0 26.6 13.3 7.6 30471.2 3848.6 264374.9

used to control the competition. Finally, The Korea Copyright Protection Center reports the monthly statistics for the total amount of pirated content and the number of discovered by the government. The center monitors illegally distributed content in real time. Table 2 summarizes the operational definitions and sources of the variables used in this study and Table 3 shows the descriptive statistics of our dataset. 4. Results 4.1. Length of survival Fig. 2 shows the survival function, or the probability of survival, by the services. The dotted line for the streaming service lies above the full line for the download service. These results show that the survival probability on the ranking charts is higher for streaming services. By the fifth week after appearing on the chart, almost 40% of the songs survived on the streaming chart and only 20% remained on download chart. The log rank test was conducted to verify the statistical difference between two distributions. The χ2 value is 436 and, thus, the distribution is significantly different (p < .001). Based on the Kaplan-Meier method, these results show that streaming services have longer survival periods. 4.2. Factors affecting survival To examine factors affecting survival time, the Cox proportional hazard model was conducted where the effect of the covariates can be interpreted by relative hazard ratios. The following Tables contain the results of our hazard model regarding the effect of each factor on the survival of songs. Survival estimation results in model 1 and model 2 show that a lower initial debut rank (the larger number of rank; e.g. 99, 100) increases the hazard ratio for both the download and streaming charts (hazard ratio > 1.0). This indicates that a lower debut rank (numerically larger number) reduces the song’s survival time on both the download chart and the streaming chart.

Fig. 2. Kaplan-Meier analysis. 7

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

Table 4 Survival estimation results. Haz. Ratio (exp(β)) Model 1 (Download) Initial debut rank Superstar Title track Major label Tie-in Male group Female group Control (Songs released, Holiday) Obs. R-square

Model 2 (Streaming) ***

1.03 1.05 0.59 0.90 1.15 0.98 0.86 Included

*** ** ***

**

4526 0.46

1.02 0.85 0.69 0.95 1.38 0.95 0.75 Included

*** *** ***

***

***

2772 0.32

‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05.

Being the title song of an album has the highest explanatory power among all variables in predicting survival on the ranking charts. When a song is a title track of an album, the hazard ratio greatly declines, indicating that title tracks survive longer on the charts. Specifically, title tracks survive longer with download services than streaming services (the hazard ratio of “title” is 0.59 for the download chart and 0.69 for the streaming chart). Songs distributed by major companies have an approximately 10% lower risk rate of dropping off the download chart than minor companies. However, being a major company has no effect on the survival time of songs on the streaming chart. This result indicates that regardless of firm size, minor label companies have the potential to succeed on the streaming charts. Among the variables regarding idolatry, only the female group variable is statistically significant. The average survival time of a female group on the download chart is 4.3 (week), and that of non-female groups is 2.9 (week). Furthermore, the average survival time of a female group on the streaming chart is 7.0 (week), and that of non-female groups is 5.0 (week). In summary, songs of female groups survive longer than others. The two of our results are opposite to previous studies. To our knowledge, the superstar effect is likely to decrease the hazard ratio for both the download and streaming charts. However, the superstar effect has no influence on the download chart, although it decreases the probability of dropping off the streaming chart. This superstar effect on the download chart is inconsistent with previous results where superstars’ albums remain survive longer (Asai, 2008; Bhattacharjee et al., 2007). The coefficient of the tie-in variable indicates that songs inserted into other media have a larger risk of dropping off the charts, which is opposite to Asai (2008). Our result rather shows that a tie-in with other media has a negative effect on a song’s success. Although additional media may attract attention initially, it fails to retain the attention (Table 4). 4.2.1. Superstar effect To examine the effect of superstar, we conduct an additional test. When ‘superstar’ variable is used in download chart, model is not statistically significant. To investigate the superstar effect more thoroughly, we added interaction term between the superstar variable to the alternative model (model 3). The likelihood ratio test between model 1 and model 3 show significant results (χ2(2) = 20.59, p < 0.001). This indicates that the model is valid. When the interaction with the title track variable is not considered, the superstars’ songs would not seem to influence survival. However, the coefficient of the interaction term indicates that the superstars’ title tracks reduce the hazard rate. For download services, the interaction between superstar and title track is statistically significant (hazard ratio 0.74, p < 0.001). Meanwhile, the model 4 represents that the model added interaction term is not valid (χ2(2) = 0.40, p = n.s.). It is inferred that when superstars release their albums in download chart, consumers listen all the songs, which appears to be on the chart. However, except for the title track of each album or most popular song, the remaining tracks of the album drop out of the chart shortly. Among 1588 superstars’ song in download chart, almost half of songs are title track (49.6%). The average length of survival is 2.14 (week) in non-title track and 5.25 (week) in title track. In streaming chart, 1025 superstars’ songs are released and approximately 70% songs are title track. The average length of survival is 4.37 (week) in non-title track and 7.42 (week) in title track (Table 5). 4.2.2. Piracy To assess the effect of piracy, the number of piracy users and the number of pirated content are added to the model. Both variables make the same results. Either the number of piracy users or the number of pirated content has a negative influence on the survival of songs for both download services and streaming services (Table 6). If there are a lot of illegal download on a song’s debut month, the song does not survive for a long time. Namely, the increased amount of illegal download harms legal downloading and streaming. Although previous assertions that services can displace piracy (Aguiar and Waldfogel, 2017; Weijters and Goedertier, 2016), the effect of piracy is not differ between download and streaming services. 8

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

Table 5 Survival estimation results with interaction term. Haz. Ratio (exp(β)) Model 3 (Download with interaction) Initial debut rank Superstar Title track Major label Tie-in Male group Female group Superstar × Title Control (Songs released, Holiday) Obs. R-square Likelihood ratio

Model 4 (Streaming with interaction) ***

1.03 1.25 0.66 0.89 1.17 0.98 0.87 0.74 Included 4526 0.46 20.59

***

1.02 0.88 0.71 0.95 1.39 0.95 0.75 0.94 Included 4526 0.46 0.40

*** *** ** ***

** ***

***

***

***

***

n.s.

‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05. Table 6 Survival estimation results with piracy. Haz. Ratio (exp(β)) Model 5 (download with piracy user)

Initial debut rank Superstar Title track Major label Tie-in with other media Male group Female group Log (number of piracy users) log (number of pirated content) Superstar × Title Control (Songs released, Holiday) Obs. R-square Likelihood ratio

1.03 1.24 0.66 0.89 1.16 0.97 0.86 1.06

***

0.74 Included 4526 0.46 30.353

***

*** *** ** ***

**

Model 6 (download with piracy content) ***

1.03 1.24 0.66 0.89 1.16 0.97 0.86

*** *** ** ***

**

**

***

Model 7 (streaming with piracy user)

1.02 0.87 0.71 0.95 1.38 0.95 0.75 1.05

*** + ***

***

**

***

***

0.95 Included 2772 0.32 4.989

1.02 0.87 0.71 0.95 1.38 0.95 0.75

***

1.07 0.95 Included 2772 0.32 5.76

*

+ ***

***

***

*

**

1.07 0.74 Included 4526 0.46 28.59

Model 8 (streaming with piracy content)

n.s.

n.s.

‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘+’ 0.1.

5. Conclusion and discussion In an environment that is transitioning from download to streaming, this study used music ranking charts as success indicators and compared factors affecting song survival on both download and streaming charts. This study analyzed the weekly top 100 hit charts from 2011 to 2014 in the Korean music market. Based on a literature review, the established variables were debut rank, superstar, major label, tie-in, and male/female group, and the additional variable “title” was included. Furthermore, the effect of piracy was also empirically tested. A lower debut rank (the larger number of rank; e.g. rank 99, 100) had a significant negative effect on the song’s survival time on the both charts. In other words, once a song is ranked on the chart, it guarantees longer success on the chart. Consumers have difficulty in choosing what to listen because digital music services provide a lot of songs. Therefore, they rely more on the ranking information, and the ranking chart has become more important in digital music services. A title song can be either the best or the most popular song on an album, and the promotion and marketing capabilities are mainly focused on the title track. Thus, the title tracks usually have more exposure than non-title songs. As Elberse (2010) stated, even though consumers now buy more digital music, the revenues from the unbundled purchases do not offset the decrease in whole album sales. The results of the survival analysis indicated that being the title track is the most important variable to most significantly reduce the probability of dropping off both the download and streaming charts. The descriptive statistics of samples demonstrated that almost 25% of songs were tied with other media. The results show that being tie-in songs increase the chance to drop off the chart shortly. Tie-in with other media apparently helps to put songs on the chart owing to the higher chance of exposure. Although several songs had a chance to be exposed by movies or TV dramas, they generally did not survive longer on the chart. 9

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

Regarding the effect of piracy, this study found the similar results with previous studies. First, the number of legal download was obviously damaged by illegal download. The result is not different with the study of Bhattacharjee et al. (2007) that the piracy reduces albums’ survival time. However, the assertions that streaming services can displace the piracy (Aguiar and Waldfogel, 2017; Weijters and Goedertier, 2016) were not supported. This study produced a consistent result that the piracy declines the time surviving on the chart. From the perspective of songs’ survival, the higher number of illegal download harms songs’ success on the chart. There were two different results affecting songs' survival between download and streaming chart (i.e. superstar effect and major label effect). Previous studies have examined the correlation between music success and superstardom. Many studies concluded that superstar effect exists in the music industry. Furthermore, Asai (2008) and Bhattacharjee et al. (2007) included the superstar factor in their survival model and found a positive effect on survival time. In this study, superstar effects, which was not significant to song’s survival time for download services, became positive when the model includes the interaction term between superstar and title track. Although it did not be investigated in this study, we can think several reasons about why the difference happens between them. The one of possible reason is unbundling and accessibility. Music is a typical experience good where the value is revealed only after its consumption (Nelson, 1970). Superstars’ album was safe choice for consumers. However, after the album unit was unbundled, consumers can buy a single song with pay per song. Since they could buy a single song, consumers no more depend on superstars’ name value. This result implies that consumption is focused on the most popular or the title song, rather than on the whole album despite superstar released the album in download chart. However, streaming services give unlimited access to many songs at a low cost. Consumers tend to dip their toe in non-title superstar songs. Nonetheless, consumption is focused on the most popular or the title song like download chart. In this data, approximately 49.7% of superstars’ songs are title track in download chart. Meanwhile, approximately 69.4% of superstars’ songs are title track in streaming chart. This result addresses that non-title track is difficult to take a place on the ranking chart. Since unlimited access creates information overload for consumers, they depend on chart to mitigate their overload. The average survival time of songs on the charts was much longer for streaming services than download services. As a result, the number of songs that have been ranked on the charts for the sample period was 4526 for download services and 2772 for streaming services. Since there are fewer songs ranked on the streaming chart and the chart enables songs to garner more attention and increase current and future sales, revenues are more likely to be concentrated by fewer songs in the streaming service. Therefore, we might assume that superstar effect is dominate in streaming services due to unlimited accessibility despite of unbundling. Another distinctive result between download and streaming chart is major-label effect. Major labels generally have stronger capital than minor labels and can, accordingly, more easily promote and distribute music. Therefore, albums of major labels survive longer than those of minor labels (Bhattacharjee et al., 2007). However, although this held true for the download chart, the opposite result was found for the streaming chart. Once a song is ranked on the streaming chart it can survive longer regardless of the size of the label. Thus, streaming services provide minor labels the opportunity to succeed, while they decrease the influence of major companies. Two distinctive result between download and streaming chart represent the changing environment of music sales strategies. Therefore, this study suggests several implications for stakeholder in music industry. 5.1. Implications Streaming services are obviously different from download services. The way of consuming content has changed from ownership to access, and consumers have different values or preferences for ownership and access. Thus, this study figured out that factors affecting songs’ success on download and streaming services are different. Thus, it is important to understand the difference between the services, and respond to the change. This study has several practical implications. First, Korea is a precursor of music streaming services. According to the IFPI report (2016a), global music streaming accounts for 43 percent of digital revenues, and has grown more than fourfold over the five years. Premium subscription services have expanded dramatically in recent years, and about 68 million people now pay for a music subscription. In contrast, Korea digital music market has been led by streaming services very early. IFPI represents streaming in Korea accounts for approximately 91% of total digital revenues in 2014 (IFPI, 2015). By analyzing Korea music market data, it would be possible to predict how streaming services affect the success of songs. Second, music producers should focus on their main products for marketing and promotion. As Elberse (2008) stated, gaining profit from the tail is difficult in the digital music industry. Although online platforms improve the availability of additional products, they may also create information overload due to excessive options. Consequently, most consumers rely on external information, such as ranking charts or media exposure. The results of this study show that success online cannot be guaranteed until business capacity is concentrated on a certain product that gains attention from the audience. A song which is a title track and receives greater support can survive longer. Based on the finding, it is possible to establish a music release strategy. Instead of releasing a whole album at a time, release of a single album including one title track can be a good way to draw consumers’ attention. Several artists such as Big Bang have adopted this strategy for their new album. It is necessary to invest in and focus on the one thing. Third, in the streaming environment, minor labels can have the same opportunity to succeed as major labels. In other words, minor labels’ songs can survive as long as major labels’ songs even with less capital and marketing power. The cost to search for and acquire songs has significantly diminished through streaming services. Consumers can easily discover a wide variety of songs that they have not consumed previously. Thus, it is crucial to increase the exposure of the songs yet to be discovered, and minor labels can promote and market songs through low-cost social network services (SNS). Fourth, superstars can have a different strategy when utilizing download services. When superstars release their albums, all their songs are often listed on the chart. Although most songs drop out of the chart shortly, a title track remains on the chart for a long time. 10

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

Despite the availability of one popular song such as a title track, the other songs are sold together when an album is released. A superstar’s fandom may download the whole album because of the desire for owning the product. Eventually, unbundled products are sold to a superstar’s fandom as if they were bundled. Especially, non-title songs can survive longer as a superstar is more popular. Thus, a cross selling strategy which sells a title track and other songs together can be employed by superstars in the download service. The negative effect of piracy on digital music market is examined through the survival analysis. The prior studies argued that streaming services can reduce the piracy behavior (Aguiar and Waldfogel, 2017; Weijters and Goedertier, 2016). The result of this study is similar with what those studies claimed. When there is much piracy, songs do not survive longer on the both download and streaming charts. It indicates that the piracy is a substitute for both downloading and streaming. On the basis of the advantage of streaming services such as unlimited access to many songs at a low cost, it would be possible for music industry to widen the paid users and reduce the piracy. Finally, another suggestion for producers is to provide a song at low or no cost initially. While this may decrease profits in the short-term, a price adjustment can be an effective way to expose a product to consumers easily. If the song succeeds and is ranked on the chart using this price strategy, it can bring more revenue in the long-term. This strategy is even more effective in streaming services since the survival time is much longer than in download services. Furthermore, hoarding behavior, though unethical, can occur during which fans and brokers coordinate their efforts to push a certain song up the chart (Won, 2015). 5.2. Limitations This study is significant in that it does not rely on user recall to assess levels of media usage. The self-reported usage method often yields suspect results in the complex user environment in this age of ubiquitous media consumption (Webster, 2005). Despite numerous studies regarding consumer perceptions about ownership and access, there is little analysis using industry data, especially focusing on the digital music industry over time. This study verified how access-based consumption can affect the music market by using industry data. This facilitates researchers and marketers in comprehending the music market in more detail. Despite some contributions, it is not without limitations. Songs often gain popularity through SNS diffusion, such as Psy’s Gangnam Style. However, since the titles of songs use everyday language, it is difficult to gather social data. As a result, this study did not consider the effect of social media as a promoting tool. Future studies can consider the effect of SNS to reinforce our research findings. Furthermore, this study used proxies to measure production cost, advertising cost, and piracy since exact cost and illegal download per each song were not available. This study used the variable “major label” as a proxy measure of the production and advertising cost. If the exact costs and number of piracy are considered in the further analysis, the optimized costs for production and marketing would be measured. Besides, instead of the number of piracy per each song, the total number of piracy users and pirated content for each month were adopted to examine the influence of piracy on downloading and streaming. Although this study tried to make robust result by using two different measures: the number of piracy users and pirated content, the exact effect would be discovered with the piracy number per each song. The characteristics of content itself could not be considered in the study. For example, genre, lyric, or mood of music can be one of crucial factors to determine the success of songs. Text analysis or music chord analysis can be used to examine the quality of content itself in the future research. Finally, these results have generalizability issues. South Korea has a highly developed broadband infrastructure, which is beneficial to consume digital content and the music industry is growing very fast in Korea and is expanding their influence on other countries. Thus, exploring music survival in the Korean music industry as one of the examples might contribute to better understand the nature of digital music industry. However, song survival on the ranking charts could vary based on a country’s culture or technical infrastructure. Future research is necessary to investigate various contexts of digital music consumption in other regions and countries. References Adler, M., 1985. Stardom and talent. Am. Econ. Rev. 75 (1), 208–212. http://dx.doi.org/10.2307/1812714. Aguiar, L., Waldfogel, J., 2017. Streaming reaches flood stage: does Spotify stimulate or depress music sales? Int. J. Indus. Org. Online first. http://dx.doi.org/10. 1016/j.ijindorg.2017.06.004. Apple, 2016. The all-new Apple Music. Retrieved from https://www.youtube.com/watch?v=CQY3KUR3VzM. Asai, S., 2008. Factors affecting hits in japanese popular music. J. Media Econ. 21 (2), 97–113. http://dx.doi.org/10.1080/08997760802069895. Bhattacharjee, S., Gopal, R.D., Lertwachara, K., Marsden, J.R., Telang, R., 2007. The effect of digital sharing technologies on music markets: a survival analysis of albums on ranking charts. Manage. Sci. 53 (9), 1359–1374. http://dx.doi.org/10.1287/mnsc.1070.0699. Bradlow, E.T., Fader, P.S., 2001. A Bayesian lifetime model for the “Hot 100” Billboard songs. J. Am. Stat. Assoc. 96 (454), 368–381. Chen, Y.C., 2009. Essays on Mobile Advertising and Commerce. Science Technology and Management, Harvard University, Boston, MA (Ph.D. dissertation). Chen, K.C., Lee, C.W.J., 1993. Financial ratios and corporate endurance: a case of the oil and gas industry. Contemp. Account. Res 9 (2), 667–694. Chen, Y.C., Shang, R.A., Lin, A.K., 2008. The intention to download music files in a P2P environment: consumption value, fashion, and ethical decision perspectives. Electron. Commer. Res. Appl. 7 (4), 411–422. Chiou, J.S., Huang, C., Lee, H., 2005. The antecedents of music piracy attitudes and intentions. J. Bus.Ethics. 57 (2), 161–174. Choi, J., Jung, J., Lee, S., 2013. What causes South Korean users to switch from a local to a global social network site? The cultural, social, economic, and motivational factors of Facebook’s globalization. Comp. Hum. Behav. 29 (6), 2665–2673. http://dx.doi.org/10.1016/j.chb.2013.07.006. Chung, K.H., Cox, R.A.K., 1994. A stochastic model of superstardom: an application of the yule distribution. Rev. Econ. Stat. 76 (4), 771–775. http://dx.doi.org/10. 2307/2109778. Cornwell, T.B., Coote, L.V., 2005. Corporate sponsorship of a cause: the role of identification in purchase intent. J. Bus. Res. 58 (3), 268–276. Cox, D.R., 1972. Regression models and life-tables. J. R. Stat. Soc. A. Stat. 34 (2), 187–220.

11

Telematics and Informatics xxx (xxxx) xxx–xxx

H. Im et al.

Denisoff, R.S., Plasketes, G., 1990. Synergy in 1980s film and music: formula for success or industry mythology? Film. Hist. 4 (3), 257–276. http://dx.doi.org/10. 2307/3815137. Elberse, A., 2008. Should you invest in the long tail? Harv. Bus. Rev. 86 (7/8), 88–96. Elberse, A., 2010. Bye-bye bundles: the unbundling of music in digital channels. J. Mark. 74 (3), 107–123. Gopal, R.D., Bhattacharjee, S., Sanders, G.L., 2006. Do artists benefit from online music sharing? J. Bus. 79 (3), 1503–1533. Hamlen Jr, W.A., 1991. Superstardom in popular music: empirical evidence. Rev. Econ. Stat. 74 (3), 729–733. IFPI, 2015. IFPI Digital Music Report 2015. Retrieved from http://www.ifpi.org/downloads/Digital-Music-Report-2015.pdf. IFPI, 2016a. IFPI Global Music Report 2016. Retrieved from http://www.ifpi.org/news/IFPI-GLOBAL-MUSIC-REPORT-2016. IFPI, 2016b. IFPI Music Consumer Insight Report 2016. Retrieved from http://www.ifpi.org/downloads/Music-Consumer-Insight-Report-2016.pdf. Jung, J., Song, H., Kim, Y., Im, H., Oh, S., 2017. Intrusion of software robots into Journalism: the public’s and journalists’ perception of news written by algorithms and human journalists. Com. Hum. Behav. 71, 291–298. http://dx.doi.org/10.1016/j.chb.2017.02.022. Kauffman, R.J., Wang, B., 2008. Tuning into the digital channel: evaluating business model characteristics for Internet firm survival. Inform. Technol. Manage. 9 (3), 215–232. Kaplan, E.L., Meier, P., 1958. Nonparametric estimation for incomplete observation. J. Am. Stat. Assoc. 53, 457–481. Kelley, C., 2017. K-Pop superstars BTS to perform for first time on an American Awards Show at 2017 AMAs. Billboard. Retrieved from https://www.billboard.com/ articles/columns/k-town/8023046/bts-american-music-awards-debut-2017. Klein, J.P., Van Houwelingen, H.C., Ibrahim, J.G., Scheike, T.H. (Eds.), 2016. Handbook of Survival Analysis. CRC Press. Kwon, H.H., Armstrong, K.L., 2002. Factors influencing impulse buying of sport team licensed merchandise. Spo. Mark. Q. 11 (3), 151–163. Leibenstein, H., 1950. Bandwagon, snob, and Veblen effects in the theory of consumers' demand. Q. J. Econ. 64 (2), 183–207. Liebowitz, S.J., 2006. File sharing: creative destruction or just plain destruction? J. Law Econ. 49 (1), 1–28. Liu, M., 2017. Bigger than Bieber? K-pop band BTS beats US stars to win Billboard Music Award. Retrieved from https://edition.cnn.com/2017/05/22/entertainment/ k-pop-bts-billboard-music-awards/index.html. MacDonald, G.M., 1988. The economics of rising stars. Am. Econ. Rev. 78 (1), 155–166. Mazzaferri, E.L., Jhiang, S.M., 1994. Long-term impact of initial surgical and medical therapy on papillary and follicular thyroid cancer. Am. J. Med. 97 (5), 418–428. Montgomery, A.L., Moe, W.W., Hall, D., 2000. Should record companies pay for radio airplay? Investigating the relationship between album sales and radio airplay. Working paper, Marketing Dept., The Wharton School, University of Pennsylvania. Music & Copyright, (2014, May 6). UMG and WMG see gains in recorded-music market share in 2013, while Sony/ATV dominates music publishing. Retrieved from https://musicandcopyright.wordpress.com/2014/05/06/umg-and-wmg-see-gains-in-recorded-music-market-share-in-2013-while-sonyatv-dominates-musicpublishing/. Nelson, P., 1970. Information and consumer behavior. J. Polit. Econ. 78 (2), 311–329. Oberholzer-Gee, F., Strumpf, K., 2007. The effect of file sharing on record sales: an empirical analysis. J. Polit. Econ. 115 (1), 1–42. Ouellet, J.F., 2007. The purchase versus illegal download of music by consumers: the influence of consumer response towards the artist and music. Can. J. Adm. Sci. 24 (2), 107–119. http://dx.doi.org/10.1002/cjas.16. RIAA, 2017. News and Notes on RIAA 2016 Latin Shipment and Revenue Statistics. Retrieved from https://www.riaa.com/reports/news-notes-2016-riaa-shipmentrevenue-statistics/. Rosen, S., 1981. The economics of superstars. Am. Econ. Rev. 71 (5), 845–858. Ryoo, W., 2009. Globalization, or the logic of cultural hybridization: the case of the Korean wave. Asian. J. Commun. 19 (2), 137–151. Seabrook, J., 2003. The money note. New Yorker. Retrieved from http://www.newyorker.com/magazine/2003/07/07/the-money-note. Spellman, P., 2006. Indie Power: A Business-Building Guide for Record Labels, Music Production Houses, and Merchant Musicians. Retrieved from. second ed. MBS Business Media, Boston. Strobl, E.A., Tucker, C., 2000. The dynamics of chart success in the UK pre-recorded popular music industry. J. Cult. Econ. 24 (2), 113–134. Towse, R., 1992. The earnings of singers: an economic analysis. In: Cultural Economics. Springer Berlin Heidelberg, pp. 209–217. Wang, C.-C., Chen, C.-T., Yang, S.-C., Farn, C.-K., 2009. Pirate or buy? The moderating effect of idolatry. J. Bus. Ethics 90 (1), 81–93. http://dx.doi.org/10.1007/ s10551-009-0027-y. Webster, J.G., 2005. Beneath the veneer of fragmentation: television audience polarization in a multichannel world. J. Commun. 55 (2), 366–382. http://dx.doi.org/ 10.1111/j.1460-2466.2005.tb02677.x. Weijters, B., Goedertier, F., 2016. Understanding today’s music acquisition mix: a latent class analysis of consumers’ combined use of music platforms. Mark. Lett. 27 (3), 603–610. Won, H.J., 2015. Can you trust the K-pop charts? Korea Herald. Retrieved from http://www.koreaherald.com/view.php?ud=20151027001137. Yoo, B., Kim, K., 2012. Does popularity decide rankings or do rankings decide popularity? An investigation of ranking mechanism design. Electron. Commer. Res. Appl. 11 (2), 180–191.

12