1
EFFECTS OF SUCCESIVE SELECTION BY EXPERT SELECTORS ON EXPERIENCE GOODS
- EMPIRICAL INVESTIGATION IN THE MUSIC INDUSTRY-Meverik Vaiksaar (11153253) Supervisor: dr. J. J. Ebbers
Master thesis:
MSc BA Entrepreneurship & Management in the Creative industries
18.08.2017
2
Statement of originality
This document is written by Meverik Vaiksaar who declares to take full
responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that
no sources other than those mentioned in the text and its references have been used
in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of
completion of the work, not for the contents.
3
Abstract
In creative industries expert selectors are professional experts who earn living as one. Over time researchers have argued about expert selectors being influencers or predictors of experience goods. As influencers expert selectors influence the experience good’s performance, whereas predictors only predict what will perform well. Moreover, researchers have increasingly analyzed word of mouth’s influence next to expert selectors. Given the technological advancements and changes in creative industries this study adds a new dimension to analyzing expert selectors influence. The study introduces multiple selection periods through successive selection, whereas previous studies have analyzed expert selection only based on one selection period. The empirical setting of the study is the music industry and on-demand music streaming platform Spotify. The research analyzes the effects of Spotify expert
curated playlists on 260 songs by major and independent artists’ songs from across the world. Based on weekly successive changes in experts’ playlists the study analyzes their effect on song’s performance, word of mouth worldwide and as determinants of innovativeness and category spanning ability. The study finds evidence for successive selection and that independent artists’ gain more from Spotify experts’ playlists. Contrary to the expectation the results show that major producers are more
innovative than independent artists, yet have a higher category spanning ability as expected. Lastly, the study found that in successive selection expert selectors can effect experience goods word of mouth worldwide. Findings indicate a need for further research and exploration in the field.
4
Table of Contents
Abstract ... 3
Introduction ... 5
2. Literature review ... 8
2.1 Gatekeepers ... 8
2.2 WOM and e-WOM ... 10
2.3 E-WOM types ... 11
2.4 E-WOM and Creative industries ... 13
2.5 Selectors ... 14
2.6 Music industry ... 18
2.7 Hypotheses building ... 21
3. Research design & methodology ... 28
3.1 Research setting ... 28
3.2 Data collection ... 29
3.3 Dependent variables ... 30
3.4 Independent variables ... 31
3.5 Moderating variables ... 32
3.6 Control variables ... 32
3.7 Methods ... 33
4. Results ... 34
4.1 Descriptive statistics ... 34
4.2 Correlations ... 37
4.3 Hypotheses ... 38
5. Discussion ... 46
5.1 Hypotheses & Contributions to literature ... 46
5.3 Managerial implications ... 53
5.4 Limitations, future research ... 55
6. Conclusion ... 58
Appendix A ... 59
Appendix B ... 66
5
Introduction
Consider the following quote: “Nobody knows anything”. Today In the rapidly changing world this quote could be relevant to any industry. Yet, historically researchers have attributed it to creative industries (Caves, 2003). Namely, the experience goods creative industry produces have always been in excess and their quality susceptible to people’s subjective opinions (Hirsch, 1972, Peltoniemi 2015). Therefore, at the foundation of creative industry research lies selection system theory, which helps to explain the underlying processes of value determination and competitive processes between producers (Wijnberg 1995; 2004; Wijnberg and Gemser 2000; Priem 2007). Historically creative industries are also known for gatekeepers who have controlled the whole production value chain and as a result the competitive processes through sheer resource power (Hirsch, 1972). These include movie studios, record labels, publishing houses who have had the appropriate resources to bring experience goods to the market in the first place. However, as with many other industries the researchers have marked significant changes in gatekeepers and selectors power due to the coming of internet (Eisenmann & Bower 2000; Hirsch 2000).
Researchers illustrate how internet has brought outstanding changes to networks, distribution and information channels which in aggregate have democratized industries (Eisenmann & Bower 2000; Hirsch 2000; Soda 2004; DeFilippi et. Al 2007; Brynjolfsson et. al 2011). In essence the changes have decreased gatekeepers power from superior players to simply large companies (ie. majors), who are in competition with the growing number of small producers (ie. independents) (IFPI, 2016). Tremendous changes have also taken place in terms of communication and word of mouth effects, which in online setting (e-WOM) achieve unprecedented levels (Aral, Walker 2012). In essence, through blogs, online reviews and social media platforms large companies have lost their dominant marketing power and
6
consumers have gained power over the public eye. While studies on word of mouth existed pre-millennia, in age of internet research fields in eWOM have flourished Libai, Muller, Peres 2005; Watts and Dodds 2007; Easley and Kleinberg 2010; van Eck et. al 2011; Van der Lans and van Bruggen 2010; Kaplan and Haenlein 2011; Hinz & Skiera 2011; Aral and Walker 2012; Berger & Milkman, 2012; Nejad and Sherrell 2014; Aral, Walker 2014).
While some researchers have taken increased interest in investigating consumers influence, others have expanded work on selectors. Academics have investigated the valence and volume of reviews (Clement 2010; Berger et. al 2010), genre specificity eg. art house vs blockbuster (Gemser et. al 2007; Hennig-Thurau, Marchand & Hiller 2012) and newer research has studied selectors influence also hand in hand with e-WOM (Liu et. al 2007; Kim et. al 2013; Tsao 2014). Yet, the most notable and questioned area of research has been the extent selectors are influencers or predictors of experience goods’ sales (Burzynski & Bayer 1974; Eliahsberg &Shugan 1997; Terry 2004; Reinstein & Snyder 2005; Clement et. al 2007).
Researchers have argued about expert selectors being influencers or predictors for years and as a result academics have increasingly added more variables into their research to find conclusive evidence for either role. These have included star power (Gazley, 2010) director power (Moul, 2007) award nominations, ex ante, ex post determinants (Brewer, 2009), user reviews (Gazley et. al 2010; Tsao 2014) and increasingly word-of-mouth (Liu et. al 2007; Kim, Park 2013; Tsao 2014). However, even with larger models what has been common throughout the years is that the studies have focused on only one time period in the analysis. In the movie industry the effect of critics’ has been tested on reviews that were published before the release (Eliashberg & Shugan 1997; Basuroy et. al 2003; Reinstein & Snyder 2005; Gemser et. al 2007) and similarly in the book industry where reviews have appeared on the opening week or earlier (Berger et. al 2010; Clement et. al 2010). With the coming of new technology and close to real
7
time data monitoring this study suggests that on demand streaming can introduce a new dimension into selectors research, which can add to the discussion of expert selectors as influencers or predictors.
In the recent years on-demand streaming has revolutionized many industries and at the helm of progress has been the creative industries. Many changes have happened in the motion picture and television fields but even more so in the music industry where on-demand music streaming has become the industry’s main driver of growth worldwide (IFPI, 2016). Consumers have wholeheartedly accepted the algorithmic recommendations but also curated playlists by the services own expert selectors. Not only do these playlists re-introduce expert selectors dynamics that were prevalent in the early days of music industry with payola (Mol & Wijnberg 2007; Cookson, 2015), they also allow to study expert selectors influence in more depth than ever. The playlists are more advanced than radio music curation with thousands of playlists worldwide fit for specific genres, moods and local tastes. Moreover, the playlists are curated on custom time periods, which can be daily, weekly, bi-weekly, monthly or even on longer periods (Morris and Powers, 2015). Thereby, this research aims to implement such nuanced time periods into selectors literature and in the study introduces the term ‘selection period’ and the concept of ‘successive selection’. Selection period is defined as time period when expert selectors evaluate goods or actors based on their own value systems and then make their evaluations public. As a new concept in selection system theory, successive selection is defined as a nuanced form of selection where after the experience goods release to the market at least more than one selection period occurs.
Considering the qualities of the music streaming platforms, I propose that research in this setting can greatly contribute to the discussion of expert selectors being as influencers or predictors after experience goods release to the market. Especially whether more selection periods through expert selectors playlist curation can influence the performance and the word of mouth around experience goods worldwide, which is yet to have been investigated. Therefore, this paper will aim to answer the following
8
research question: To what extent does successive selection by expert selectors affect the performance
and word of mouth of experience goods after introduction?
In sum, this study will make several contributions to theory and practice. For researchers, the study will add a new dimension in form of successive selection and contribute to expert selectors as influencers’ literature and selection system theory in general. Moreover, the study will be the first empirical study to analyze expert selectors influence on the fast growing on-demand streaming platforms. This will be also very useful for practitioners who can learn if good relationships with expert selectors on new streaming platforms can lead to more revenue and word of mouth around songs.
2. Literature review
In the following, the currently existing literature and research is reviewed. First, the literature on gatekeepers is presented, followed secondly by word of mouth from consumers and companies perspective. Third and the last section will be devoted to selectors in the creative industry where the hypotheses will also be presented.
2.1 Gatekeepers
Gatekeepers stand as a central concept in the creative industries (Hirsch 1972; Towse 2003, Caves 2003; Tschmuck 2006). On the large scale gatekeepers are defined as organizations, institutions with a full control in the value chain to either or both produce and distribute experience goods on the market (Hirsch, 1972). A great distinction can be made between the gatekeepers based on these two functions. In the production process creative organizations first source the talent and thereafter begin the grooming of talent or immediately proceed to production. As doing such gatekeepers offer resources, functions and expertise that is difficult for individuals to come by on own means (Peltoniemi 2015, Anand & Peterson 2000). On the marketing side gatekeepers provide distribution channels (eg. TV networks, book publishing
9
deals, news outlets) where the goods can actually be bought or experienced. What is more, the largest and most powerful gatekeepers operations usually include both functions of production and distribution (Hirsch, 1972). Large parent companies such as Disney own movie studios but also TV channels ABC and Disney channel (Giroux and Pollock, 2010).
Nowadays the functions of gatekeepers still exist but their dominant power has been compromised (Hirsch, 2000) and thus primarily remain as large companies (ie. majors) with wide operational capabilities. As for many industries the disruptive force has been the internet. Gatekeepers’ power has primarily decreased through evolutionary changes in networks, distribution and information channels (Eisenmann & Bower 2000; Hirsch 2000; Soda 2004; DeFilippi et. Al 2007; Brynjolfsson et. al 2011). First and foremost the advent of internet has allowed networks including business networks inside the creative industries to expand and democratize (Hirsch, 2000). More producers can access and gain expertise through information online which results in a wider range of competitors and potential business partners. As a result the number of small producers (eg. independent filmmakers, bands, publishers) has skyrocketed. Secondly, the number of distribution channels have grown tremendously, which has divided the audience’s attention that a gatekeeper owned distribution channel would solely control (Brynjolfsson et. al 2011). Consumers have a myriad of options from different cable TV channels, internet shows and YouTube channels which drive innovation and create completely new formats unlike ever before. Gatekeepers have adopted and created some of these channels themselves but in totality their ability to exclusively control consumers’ attention span has significantly decreased (Seabrook, 2015). Many new content outlets such as Netflix and Spotify have emerged who have tremendous reach yet also editorial power, which makes them essentially the new age selectors. Selectors will be discussed in great depth later on yet most importantly they highlight how the gatekeepers’ have had to change their general marketing efforts, which require now more segmentation and resource investments. Thirdly, the
10
increased means of communication have amplified consumers’ word of mouth effects and have enabled regular consumers to exert greater social influence (King, Racherla & Bush 2014).
As a result word of mouth and selectors have grown tremendously important in rivaling gatekeepers. Both fields have been researched heavily yet especially word of mouth since the beginning of millennia and diffusion of the internet. In the following I will cover the literature on word of mouth from consumers and companies perspective, followed by selectors literature review.
2.2 WOM and e-WOM
Word of mouth (WOM) has become an immensely popular term in this millennium. Essentially, word-of-mouth is the sharing of information in an informal way. Formality refers often to the source that the information comes from (Kaplan and Haenlein, 2010). A formal source of information would be a media outlet or institution who has recognized credibility. Informal source is a person or institution who is perceived as unbiased. Although highly popularized in the recent millennia the concept has been around for a long time. Researchers demonstrated the power of WOM already in the 1950´s when was found that in a wide variety of decision making situations individuals can be more influenced by their peers than by media (Katz and Lazarsfeld 1955). Moreover, in influencing consumer attitudes WOM was up to seven times more effective than traditional print advertising, four times more effective than personal selling and twice as effective as radio advertising (Katz and Lazarsfeld, 1955).
E-WOM essentially shares all the qualities of traditional word-of-mouth yet in an online environment where flow of information is vastly more rapid. Online on a larger scale word of mouth takes the form of information exchange between individuals and groups in general (Aral and Walker, 2014). In the most generalized definition Mangold and Faulds (2009) define online word of mouth as “online exchange of information and opinions about product, service or event”. In a business setting Hennig-Thuraru (2003) define e-WOM as the following: “any positive or negative statement made by potential,
11
actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet”.
Effectively, researchers have demonstrated with the proliferation of e-WOM companies are not anymore in full control over their campaigns or goods (Fournier et. al 2009; Gruner et. al 2014; Manchanada et. al, 2015). The speed and reach of communication has simply risen to unprecedented levels, where as traditional word of mouth could be contained in certain regions (Kaplan, Haenlein 2011). As such e-WOM has quickly become one of the hottest buzzwords in both marketing research and practice. In the following sections the literature review will cover the various e-WOM formats and eWOM research in creative industries.
2.3 E-WOM types
The researchers analyzing the types of e-WOM have investigated the various platforms and formats that people use online. These include forums, blogs, social media platforms and posts and product reviews (Lowrey, 2006; Vermeulen, 2009; Lee & Youn 2009; Dewan 2012; King, Racherla & Bush 2014.) All of the listed differ in their rates of diffusion and depth yet out of all the most centralized and largest enabler of e-WOM are social media platforms. Researchers have found that sites such as Facebook act as online extensions of people and to a great extent can replace the real world communication (Back et. al 2010; VanMeter et. al 2015). As a result such sites have quickly amassed huge numbers of users worldwide reaching to hundreds of millions. With huge user numbers social media platforms are also active enablers of virality. Virality refers to the exponential spreading of a message online through sharing, which at high levels resembles the spreading of a virus (Kaplan, Haenlein 2011). In fact, since the term was introduced a large new line of research field viral marketing has emerged, which nowadays is one of the most popular fields in both practice and research. Fundamentally, viral marketing is the marketers attempt to stimulate and capitalize on the word of mouth dynamics as the customers spread and share content related to a
12
company or brand (Kaplan and Haenlein 2010; Van der Lans, van Bruggen 2010). For their studies researchers have largely utilized social media platforms and explored the influence of networks on spreading of e-WOM (Easley and Kleinberg 2010; Hinz & Skiera 2011; Aral and Walker 2012; Aral and Walker 2014), targeting specific individuals (Gladwell 2001; Goldsmith and Foxall 2003; Watts and Dodds 2007; Kiecker and Cowles 2008; van Eck et. al 2011; Nejad and Sherrell 2014), which type of content goes viral (Berger & Milkman, 2012) and whether viral features can be designed on products (Aral, Walker 2014).
Next important type of e-WOM are users’ online reviews, where the value of reviews works very much hand in hand with network effects driven by word of mouth. Reviews often act as aggregates of consumers’ tastes or opinions about goods, brands or services which are available to other potential consumers. Studies have shown that consumers trust in others runs deep as after referrals from friends online reviews are often the second most trusted source of information (Lee & Youn 2009; Nielsen, 2013). In more detail;, researchers have found that the volume of reviews can affect consumers’ willingness-to-pay (Wu & Wu 2016) yet can also vary depending on the type of good (Cui et. al 2012) and industry eg. motion picture, books, hotels (Chevalier & Mayzlin 2006; Vermeulen, 2009; Dellarocas, Zhang and Awad 2007).
Blogs are the third most important platform that encapsulate WOM in form of a web log that can be in audio or video format. In essence blogs started out as platforms for self-expression on sites such as Tumblr, Blogspot and through rich form of e-WOM quickly started to challenge journalism as they introduced a new avenue for freedom of speech (Lowrey, 2006). Moreover, researchers have found that as such blogs can represent people’s interests and commercially effect products rate of diffusion. Dewan (2012) and Dhar, Chang (2009) aggregated blog reviews and music blogs index Hypemachine in a study and revealed that blog posts not only reflect the popularity of current music but can also increase the
13
consumption of long tail music. Hence, blogs can act as a rich e-WOM format reflecting the general popularity but also be a point of discovery to the readers or viewers of blogs.
2.4 E-WOM and Creative industries
In the creative industries word of mouth has been known to having a big effect on experience goods success long before the internet. De Vany and Walls (1996) have notably claimed that WOM is the most important variable for the long-term success of movies. While research has been conducted inside various fields in the creative industries including books (Gruhl et. al 2005; Chevalier & Mayzlin 2006; Brynjolfsson 2007) and music (Dhar & Chang 2007; Bhattacharjee 2007; Dewan 2012), by far the most extensive research has been conducted in the motion picture industry (Faber & O’Guinn 1984; Austin, 1989; Wyatt & Badger 1990; Shrum 1991; Ravid 1999; Neelamegham & Chintagunta, 1999; Godes and Mayzlin 2004; Duan et. al 2005; Liu et. al 2006; Dellarocas et al. 2007; Moul 2007; Duan et al. 2008; McKenzie 2009; Tsao 2014).
Majority of the research conducted has analyzed eWOM from two angles: frequency or volume and valence. In their studies Wyatt, Badger (1990) and Ravid (1999) found that total number of reviews irrespective of the valence affects the revenue of movies positively. This lead some researchers to adapt volume of reviews to the frequency of reviews. The latter helped to analyze the effects of eWOM at specific times before and after the release of a movie. However, the results were no different to the previous studies as Liu et. al (2006) and Duan et al. (2008) still found only frequency of reviews to positively affect the box office revenue with no significant evidence for valence. Liu’s research highlighted that on
Yahoo! Movies website the buzz among consumers for movies peaked at one week after the movie’s
release. Despite the eWOM being negative or positive the study found that movies with most total buzz over the weeks had a higher total box office. A standout research was conducted by Dellarocas et al.
14
(2007) who found both frequency and valence to correlate with eventual box office sales numbers. The study analyzed movies released in USA in the year of 2002, which was similar to other studies that had collected data on one or two year periods. Only valence was found significant in Godes and Mayzlin (2004) study, which analyzed TV viewership ratings with consumers reviews. Hence, taken together it’s clear that the results among academics are inconsistent with one another. Yet, what can be concluded with certainty is the importance of e-WOM frequency, most studies have found to contribute to higher sales.
2.5 Selectors
Hereby the literature on selectors is reviewed. The review starts with an explanation of the selection system theory, the practical studies that have been conducted and is then followed by a deeper analysis of the music industry.
Creative industry is notorious for the oversupply of goods and artists than the market can equally consume (Peltoniemi, 2015). Not only is the competition fierce the industry has an accepted notion that ‘nobody knows anything’ due to the quality and value of experience goods being subjective. Over the years selection system theory has emerged, which helps to explain the underlying processes of value determination in the creative industry (Wijnberg 1995; 2004;Wijnberg and Gemser 2000; Priem 2007). On one end the selection system distinguishes the essential characteristics of actors and goods involved in the competitive processes. On the other the literature introduces three types of selection systems: market, peer and expert selection who ultimately make these decisions. In effect, all these three types of selectors use ‘selection’. Selection is defined as evaluations, which determine the value of actors or goods in the creative industry (Wijnberg, Gemser 2000; Wijnberg 2011). In creative industries expert selection usually occurs at time periods, which are important to the experience goods’ first release to the market (Berge et. al, 2010).
15
In market selection consumers themselves are able to ascertain the value of goods by information available. For instance, market selection applies to bakery goods to which every consumers can have a personal taste. In peer dominated market consumers look for other producers to indicate the worth and quality of goods. As an example, the visual arts industries were dominated by peer selection through guilds and academies that consisted of fellow painters evaluating each other’s work up until the rise of Impressionism in 17th century France (Wijnberg, Gemser 2000). After the movement the visual arts
industry adopted to the third selection system, which involves experts. In expert selection consumers search for professional’s expertise through expert selectors who are neither producers nor consumers. Essentially, expert selectors are professional experts in the creative industry who earn living acting as one (Wijnberg, Gemser 2000). Expert selection can occur through critics and their published reviews of films, theatre plays, musical works, books but also through tastemaker curation for instance in the music industry. Moreover, in the experts lead system the effect of time is more pronounced than in other types of selection systems. As a standard in creative industries, the expert reviews and evaluations are published before or during the first week after a good has been released to the market (Eliahsberg & Shugan 1997; Terry 2004; Reinstein & Snyder 2005). From the three selection systems expert selection is the most widely researched and will be the focal interest of this paper as well.
Selection system theory is often accompanied by classification system theory, which aims to explain selectors value determination process. Classification systems operate based on classification constructs categories, which allow for comparison between members within a category and to communicate a certain quality and value (Wijnberg, 2011). Each category is determined by certain characteristics and criterias that can be universally known or by only selectors themselves. Researchers also note that categories are in flux and constantly changing as the actors involved evolve (Khaire & Wadhwani 2010; Wijnberg, 2011). Fundamentally however, what is most important is how recognized the categories are among consumers. The three selection systems market, peer and expert indicate the
16
possible range of categories to be used by both selectors and consumers but ultimately their value is determined by the usefulness to consumers (Espeland & Stevens, 1998). Over time researchers have found selectors to consider innovativeness and category spanning ability as the two most important determinants for placing goods into categories. (Wijnberg 2000; Zuckerman & Kim, 2003; Anand & Jones 2008; Wijnberg, 2011). Innovativeness refers to the differentiating quality and overall value of a good (WIjnberg, 2000) while category spanning to the actor or good’s ability to be classified in more than one category at the same time (Zuckerman & Kim, 2003; Hsu & Hannan, 2005).
Throughout the years researchers have found expert selectors to have various effects on consumers’ consumption. Inside the books industry Clement et. al (2007) found that critics on a TV show dedicated to reviewing books have minimal effect in influencing people’s consumption of books. They found some evidence for the sentiment of reviews where stronger reviews slightly affected the sales but in general concluded the critics not to have a distinct effect in their research. In an in depth research Berger et.al (2010) conducted three studies analyzing the effect of book reviews on New York Times. Their study was unique as it counted for the already existing public awareness of goods (eg. author popularity) and consequential effect of reviews on sales. Berger et. al (2010) found that in all cases positive reviews increased the sales of books but with negative reviews the sales increased only for lesser known products. Hence, they concluded critics to have a positive effect on sales in general and can only harm the producers when a well-known author receives a negative review.
Out of all the industries most extensive research on selectors has been conducted in the motion picture industry. This can be very much due to the nature of the industry, which has embraced movie critics since the early days. At large the researchers have attempted to examine whether critics act as influencers or predictors of box office ie. sales (Eliahsberg & Shugan 1997; Terry 2004; Reinstein & Snyder 2005). In the influencers role scholars state that critics can influence the revenue the experience good will make (Eliahsberg, Shugan 1997). While as predictors, critics reviews’ can predict if the movie is going to
17
be received well but wont have any actual effect on revenue (Terry, 2004). Meaning that as influencers’ expert selectors can affect the sales of experience goods, while as predictors they can only predict what will become a success. The researchers have come to conflicting conclusions, while some attribute predictive power over the course of a longer period and others an influential effect over a longer period. Common ground between the studies is they have agreed the critics’ effect manifests on a longer period, which has been defined from four weeks onward. As a result academics have considered more variables in addition to critics, which can affect movie’s performance. Basuroy et. al (2003) were the first to question singular correlation between movie reviews and eventual performance. Their research added variables such as star power while arguing that the singular correlation alone can be spurious when there is a strong demand for a movie. In the following years researchers also added director power (Moul, 2007) award nominations, ex ante and ex post determinants (Brewer, 2009), user reviews (Gazley et. al 2010; Tsao 2014). Furthermore, since the power of e-WOM has grown tremendously most recent selectors research studies selectors and eWOM level of influence hand in hand (Gazley, Clark, Sinha 2010; Kim et. al 2013; Tsao 2014).
Another dynamic that has been explored by researchers is the type of producer. In creative industries the producers are mostly differentiated as major (ie. large corporation organizations) and independent (ie. small) producers. These two types of producers are more prevalent in some branches of creative industries than others yet undoubtedly the number of total independent producers has grown in the recent times (Hirsch, 2000). As a result researchers have investigated the operational nature of the two producer types. In the movie industry studies have analyzed the associational differences with particular movie genres and content (Bordwell & Thompson, 2001; Zuckerman & Kim, 2003), differences in scale (Zuckerman & Kim, 2003; Eliashberg et. al 2006), type of film distributor (Zuckerman & Kim, 2003) and type of exhibition in movie theaters (Eliashberg et. al 2006). In the music industry differences have been found for sales and charting (Anand & Peterson, 2000), generation of word of mouth (Dhar, Chang
18
2007), number of awards received (Anand & Watson 2004) and innovativeness (Peterson & Berger 1975; Lopes 1992; Dowd 2004).
In essence both major and independent producers are faced with the same selection system dynamics. Yet as the researchers have found, there are particular differences especially in both producers’ means of marketing. Hence the information about experience goods from major and independent producers can reach the selectors and also the consumers differently. This dynamic has been investigated by Gemser, Oostrum and Leenders (2006) whose research in the Dutch movie industry concluded that expert selectors have more impact on the independent movies than mainstream movies. Consumers that are interested in seeing the mainstream movies made by large studios where found to seek out different information such as star power and directors. Hence, there still remains a lot to be investigated in the selection system differences for major and independent producers. For future research Gemser, Oostrum and Leenders (2006) suggested music industry as a possible option to study such selectors effects in more depth. This concurs with the setting of this paper which will use music industry and on demand streaming as a case study to research these selectors effects and continue discussion whether expert selectors are influencers or predictors of experience goods. Next, after an overview of the music industry is given, the paper will continue with the hypotheses building section.
2.6 Music industry
The early days of commercial music go back to the late 1950’s, which brought the commercialization of vinyl and diffusion of local radio. However, this era is also known for payola and highly influential selectors in the field of radio. Payola is essentially defined as an illegal bribe to have one’s owned records played on demand on radio (Caves, 2000). Nowadays radio remains influential yet in the 50’s and 60’s it successfully outperformed television as the most powerful all-round medium for music. In their paper Mol & Wijnberg (2007) give a fantastic overview of the selectors dynamics in the era.
19
They explain how selectors grew in power as more and more local radio stations started appearing across America. The radio stations and subsequently the radio programmers became payola targets for the smaller independent record labels of the time. The payola practice was highly escalated by major labels whose large resources allowed to push the trend further and finally came to an end in the 60’s when payola was declared illegal by US Supreme Court (Mol & Wijnberg, 2007).
Nowadays very similar selectors dynamics can be found in the music industry’s new frontier on-demand music streaming. Streaming deviates from the music industry's traditional business model and allows customers to access a comprehensive library through a subscription or ad based tier rather than purchase individual music products (e.g., CDs or downloads) (Wlömert and Papies 2016). Consumers are thus not direct owners of the goods and can access catalogs with millions of tracks. Byproduct of such large supply are personalized listening experiences through algorithmic recommendations and playlists, which are one of the most innate features of on-demand streaming (Morris and Powers, 2015). The latter offers opportunities for user created and platform curated playlists, whose curators can be considered expert selectors. As a term music curation refers to specifically chosen set of songs based on specific criteria (Morris, Powers 2015).
From the producers standpoint streaming platforms are not resource exclusive as all producers can distribute music as experience goods on the service for a marginal cost through a distributor (Anderson, 2011). First platforms were founded post 2005 (Sinha & Mandel, 2008) and have been increasingly embraced by both independent and major labels since. This is even more evident in the growing independent labels market share, which in the last 10 years has grown from 18% to 31% (Leeds 2005; Music Business Worldwide 2017). Such growth can heavily be attributed to streaming, which has grown exponentially in the last few years and in 2015 became the recording industry’s main driver of growth (IFPI, 2016). With such numbers streaming has had a big role in decreasing the power of majors and made the market more accessible for smaller producers.
20
As a field of research streaming is as new to researchers as to the practitioners. The initial research has greatly focused on the overall economics of streaming as the music industry has attempted to recover from the severe losses ever since the continuous decline in 2000 (Aguiar and Waldfogel 2015; Wlömert and Papies 2016). The researchers have primarily investigated the effects of streaming’s cannibalization on downloads, CD’s and the feasibility of subscription vs ad based tiers. However, as recent industry figures have proved the feasibility of streaming as a business model more academics are interested in the usage aspects of streaming from a consumers perspective (Morris and Powers 2015; Lee et. al 2016; Datta, Knox and Bronnenberg 2016). Concurrently are also the practitioners who are paying increasing attention to leveraging revenues from streaming to maintain the growth.
Main practice to leveraging revenues has been currently playlists, which are exactly the reason to pose parallels to the early days of music industry. More specifically the industry has noted the assurgence of payola through playlists (Cookson, 2015; Peoples 2015). In streaming, playlists serve as a personalized listening tool but also as a point of discovery. The most popular are playlists curated by the expert selectors of the platforms themselves, which the platforms usually measure through the number of ‘followers’ or ‘fans’ (Morris and Powers, 2015). When in the 50’s radio programmers chose a specific selection of songs to go on air essentially the same is done in streaming without the live show aspect. Selectors curate playlists on a custom period, which can be weekly, bi-weekly monthly or even on a longer period. What is more, the chosen songs can be from previous periods, newly released and be fit for playlists with specific genres, moods or occasions (Morris and Powers, 2015). As stated above academically the researchers are only beginning to examine the usage aspects of streaming. With this paper I’d like to conduct my study in this rising field of research and explore the influence of expert selectors further. I will pose my arguments in the next hypotheses building section.
21
2.7 Hypotheses building
The main intention of this paper is to analyze expert selectors level of influence on experience goods in successive selection, which is yet to have been investigated. As stated earlier, selection is defined as evaluations, which determine the value of actors or goods in the creative industry (Wijnberg, Gemser 2000; Wijnberg 2011). Expert selectors are professional experts in the creative industry who earn living acting as one (Wijnberg, Gemser 2000). In creative industries expert selection usually occurs at time periods, which are important to the experience goods’ first release to the market. For instance, in the movie industry researchers have analyzed the effect of critics’ reviews on sales, which were published before the release (Eliashberg & Shugan 1997; Basuroy et. al 2003; Reinstein & Snyder 2005; Gemser et. al 2007). Similarly in the book industry, researchers have analyzed reviews that appeared in the opening week or earlier (Clement et. al 2007; Berger et. al 2010). After collecting reviews on either of the periods researchers have then analyzed the effects of these reviews on future sales. Consequently, the results have over the years divided scholars between expert selectors as influencers (Ravid 1999; Terry 2004; Reinstein & Snyder 2005) or predictors (Eliahsberg, Shugan 1997; Basuroy 2003; Hennig-Turau 2012). Which is why the researchers have then added more variables next to expert selectors in order to find more conclusive evidence for either role. Today, researchers haven’t come to definitive conclusions on either role (Brewer, 2009; Gazley, 2010; Kim, Park 2013; Tsao 2014) while next to expert selectors evidence has been found for star power (Gazley, 2010), director power (Moul, 2007), user reviews (Brewer, 2009) and increasing influence of word of mouth (Liu et. al 2006; Kim, Park 2013; Tsao 2014), which all can significantly impact experience goods sales.
The expansion of researchers’ models is understandable as more exogenous variables can potentially affect experience goods’ sales. Through larger models it becomes clearer how much influence expert selectors really have when their influence is not studied in isolation. However, in all the years of
22
research no study has investigated something fundamental as whether selectors influence or lack thereof can be dependent on different time periods or when selection occurs on multiple time periods. Currently all conducted time series studies (Eliashberg & Shugan 1997; Basuroy et. al 2003; Reinstein & Snyder 2005; Berger et. al 2010) have analyzed expert selectors influence on experience goods sales based on singular expert evaluations, which have appeared on only one time period. These have been either before or during the first week of experience good’s release to the market. However, in reality expert evaluations and curation can take place at more than just one time period and throughout experience good’s life cycle after being released to the market (Morris and Powers, 2015). Firstly, in order to analyze expert selection on multiple time periods this paper proposes the term ‘selection period’. Selection period is defined as time period when expert selectors evaluate goods or actors based on their own value systems and then make their evaluations public. Secondly, as a new concept in selection system the paper proposes successive selection, which utilizes selection periods. Successive selection is defined as a nuanced form of selection where after the experience goods release to the market at least more than one selection period occurs.
Given the technological advancements and on-demand streaming, music industry has the potential to become the new frontier for selectors and successive selection research. Effectively, the playlists employed by the streaming platforms experts allow to study users’ consumption close to real time and study expert selectors influence at almost any point of time after a song has been released. Expert selectors on these platforms curate playlists on more frequent periods than other selectors in the past, which can be under daily, weekly, bi-weekly or monthly curation (Morris and Powers, 2015). Such approach by the playlist experts’ results in multiple selection periods after a song has been released on the streaming platforms. This allows to analyze expert selection in more depth than ever and get a more nuanced understanding if expert selectors are influencers or predictors. As a historical reference music industry has the case of payola, which first brought into light the expert selectors influence in the
23
recording industry through radio curation (Mol & Wijnberg 2007). Similar phenomenon has been now documented on music streaming platforms (Cookson, 2015). Since payola’s period in music industry was short lived and illegal to say the least, no empirical study was able to investigate the expert selectors influence. Therefore, empirical research on the legal on-demand music streaming platforms can prove to be a good alternative. Based on the discussion above and presented literature I present the first hypothesis.
H1a: The larger the number of expert selectors playlists in successive selection, the larger the performance of experience goods
When acknowledging the type of producer only Gemser, Oostrum, Leenders (2006) have investigated whether expert selectors support influences major and independent producers differently. This field has received very little attention as studies investigating the two producer types have mostly focused on differences in scale (Zuckerman & Kim, 2003; Eliashberg et. al 2006), type of film distributor (Zuckerman & Kim, 2003) exhibition (Eliashberg et. al 2006), sales and charting (Anand & Peterson, 2000), generation of word of mouth (Dhar, Chang 2007) number of awards received (Anand & Watson 2004) and innovativeness (Peterson & Berger 1975; Lopes 1992; Dowd 2004).
The unique study by Gemser, Oostrum, Leenders (2006) found that expert selectors reviews influence the performance of only independent producer’s experience goods as consumers sought out different information for majors’ experience goods. However, their research was conducted very locally in The Netherlands and in motion picture industry, which leaves plenty of room to build upon and expand the research to more industries. It is interesting to investigate if independent producers in the music industry gain more from playlist experts’ support as major producers are known to have larger resources and artist fanbases (Dowd, 2004). Therefore, this study will analyze similar dynamics as Gemser et. al (2006) yet in the case of successive selection in the music industry. I present the next hypothesis.
24
H1b: The effect of expert selectors playlists on performance in successive selection is larger for independent owned than majors owned experience goods
2.7.1 Innovativeness and Category spanning
Not only does on-demand music streaming enable to research successive selection by expert selectors, it also enables to analyze how expert selectors categorize experience goods through classificatory processes. The latter can be done through expert curated playlists, where songs are determined to be fit based on certain categories and criteria.
Previous research has identified innovativeness and category spanning as the two most important determinants for categories by which the selectors indicate experience goods value to the consumers (Wijnberg 2000; Hsu 2006; Hsu 2009; Wijnberg 2011; Vergene & Wry 2014). Importance of innovativeness has been highlighted by Wijnberg (2000) who noted the change in selection systems during the assurgence of Impressionism in France. The movement influenced which types of art modern museums display nowadays, whereas before Impressionism museums exhibited the works of only dead artists. Wijnberg states the reason behind the change were art critics and museums own expert selectors who started to judge the painters of the period on their level of innovativeness. In empirical studies, innovativeness in the music industry has been tested by Peterson & Berger’s (1975) seminal study. Their research explored the industry’s different stages of market concentration from 1950’s to early 1970’s in relation to innovativeness. In the study innovativeness was measured through number of new artists and total genres on the charts, while market concentration indicated the degree of competition between record labels in the market. The study set out to disprove Schumpeter (1950) innovation theory that oligopolistic market structure with few major players leads to more innovation. Peterson and Berger found strong evidence for their hypotheses. Throughout the period when major labels controlled the market under high market concentration, a smaller number of new artists and hit songs were found in the charts. In the periods
25
where competition increased, which the authors note came primarily from new independent labels, more new artists and genres such as rock & roll were found in the charts.
Since then, more studies have built upon Peterson & Berger and expanded on the analyzed time period (Rothenbuhler & Dimmick 1982; Dimaggio & Stenberg 1985; Burnett 1992; Lopes 1992; Huygens et. al 2001; Dowd 2004). Even more so, the researchers have coined a now popular term ‘open systems’, which refers to the decentralized production major labels adopted during the 1970’s (Denisoff 1986; Garofalo 1987; Sanjek 1988; Lopes 1992). Effectively majors labels started to diversify the number of labels they owned and acquired new small independent labels for talent. Yet even with open systems researchers have not found major labels to be more innovative under high market concentration. Studies have found that for majors decentralized production helps mitigate the negative effects of high market concentration but ultimately their innovativeness depends on the diversity and innovativeness of their open systems (Lopes 1992; Huygens et. al 2001; Dowd 2004). Therefore, studies agree that majors can be more innovative than independents but that depends on their open system of production.
When viewing Peterson, Berger (1975) and future studies from selection system theory perspective it is evident that no expert selectors were involved in the determination of innovativeness since the studies used market level data. The authors analyzed innovativeness on the market level through open systems and new artists on the charts without considering the subjective quality of the experience goods. The latter however is critically important in the creative industries. Therefore, this study will build upon Peterson & Berger (1975) work and apply expert selectors perspective to analyzing innovativeness between major and independent producers. Moreover, the study will do this similarly to first hypothesis under successive selection. Based on the presented literature and majority of research supporting independent producers as more innovative, I present the following hypothesis.
26
H2a: Independent owned experience goods are determined as more innovative by expert selectors in successive selection
In their study, Peterson & Berger (1975) also made a link to the second main category determinant ie. category spanning ability. They stated that in late 50’s major labels started to homogenize more traditional genres such as jazz and blues which resulted in their respective pop formats under high market concentration. Such trend continued in later periods as more new genres emerged. Similar findings were reported by Lopes (1992) who described the standardization and homogenization of traditional genres by majors after the 70’s. Hereby, it is worthwhile to revisit the term category spanning ability, which refers to the actor or good’s ability to be classified in more than one category at the same time (Zuckerman & Kim, 2003; Hsu & Hannan, 2005). When linking this term to Peterson, Berger (1975) study it becomes evident how majors started to essentially create category spanning experience goods, which would enable the songs to be classified under multiple categories. As was the case with innovativeness, future studies since Peterson & Berger have remained studying diversity of production at market level (Black & Greer 1987; Lopes 1992; Alexander 1994) without considering the subjective quality of songs. Therefore, this study will analyze category spanning from expert selectors perspective and based on the arguments presents the following hypothesis.
H2b: Major owned experience goods are determined to span more categories by expert selectors in successive selection
27
2.7.2 Selectors and Word of mouth
The latest studies in field of selectors have analyzed the influence of expert selectors and eWOM hand in hand (Plucket et. al 2009; Gazley, Clark, Sinha 2010; Kim et. al 2013; Tsao 2014). The main question researchers have attempted to answer is whether consumers’ consumption of experience goods is more affected by expert selectors or word of mouth. Majority of studies have found that consumers value the opinions of others through e-WOM more than those of expert selectors (Gazley, Clark, Sinha 2010; Kim et. al 2013; Tsao 2014). Such findings are also supported by studies conducted in the pre-internet era (Boor, 1992; Holbrook, 1999). Some researchers however have found that the two groups can assess experience goods differently, which affects their level of influence (Ginsburgh & Weyers 1999), while others have disagreed and proposed an interaction effect (Plucker et. al 2009; Kim et. al 2013).
In their study Kim et. al (2013) decided to analyze expert selectors and eWOM influence on movies internationally as most researchers had focused on USA market in the past. Similar to majority of studies e-WOM was found to be more influential across all territories, while expert selectors had significant effect only in America. As one limitation of the study the authors noted cultural discount because only Hollywood movies were used. Cultural discount refers to the decline in value of a good from foreign markets when the good does not harmonize with the local culture (Fu and Lee, 2009). Therefore, the influence of word of mouth and expert selectors can be skewed if experience goods don’t fit the local market. Moreover, Kim et. al (2013) noted that future studies should consider other local market idiosyncrasies such as an interaction effect between expert selectors and eWOM. The last scope of this study will build up on Kim et. al (2013) and further explore expert selectors influence worldwide while analyzing experience goods from across the world. What is more, the study will investigate if an interaction effect occurs between the two groups and whether expert selectors reviews can influence word of mouth through expert selectors playlists. Based on the arguments made, I present the last hypothesis.
28
H3: The larger the number of expert selectors playlists in successive selection, the larger the experience goods’ word of mouth worldwide
3. Research design & methodology
3.1 Research setting
The empirical setting of this quantitative research is on-demand music streaming platform Spotify. The objective of the research is to monitor and consequently analyze an unbiased sample of songs that are on Spotify and whether their performance is influenced by Spotify employed playlist curators’ favorability over time, while differentiating for the song’s producer type.
As a platform Spotify is quickly evolving into the most important music service for all types of producers and artists with the highest growing user account in the industry. It is currently in 61 countries worldwide and has 100m+ active users with 60m paying subscribers (Music Business Worldwide, 2017). While there are various other on-demand streaming services such as Apple Music, Deezer, Pandora, Spotify currently provides the best opportunities to conduct academic research for the following reasons.
Firstly, Spotify has by far the highest number of active users and arguably represents on-demand music streaming much as Netflix does for on-demand video streaming. Hence, any academic research on the platform can be of high relevance to the new and evolving music industry as a whole. Secondly, Spotify has open API policies to conduct active research for any interested party, which other platforms do not provide. Thirdly, Spotify has arguably the most in depth playlist infrastructure, which allows to analyze the playlist curators’ effects on specific songs over time in a controlled academic study. The curators employ active playlist curation, which on average is on a weekly to bi-weekly basis. Considering the user-base of Spotify and playlist curation as a form of selection, Spotify playlist curators very much qualify as
29
expert selectors. Moreover, since playlist curators are more active than other expert selectors in the past this allows the study to investigate expert selectors level of influence in more depth and more frequent time periods. Lastly, Spotify features Viral Charts on their platform, which are aggregates of social media (eg. Facebook, Twitter) and song sharing, which are very useful to study word of mouth and virality. Viral Charts are available for all Spotify territories and thus allow to analyze word of mouth worldwide and in specific territories.
3.2 Data collection
This study was done in a collaboration with Spotontrack to collect data for all the samples involved in the research. Spotontrack utilizes Spotify API to retrieve real time data about playlists and Viral Charts activity for any song that is on Spotify. In order to achieve an objective selection of samples, a four week selection of original tracks from Spotify’s global New Music Friday (NMF) playlist was chosen. The data collection period for the samples ran from February 2017 to end of April 2017. The global New Music Friday playlist was chosen because the playlist features the most prominent new music from artists worldwide and is curated weekly on every Friday. Friday is also the official new music release day in the music industry which consequently coincides with other marketing and retail activities (Billboard, 2015). Furthermore, the global NMF playlist is comprised of a diverse selection of most popular music genres (eg. pop, rap, electronic, rock, Christian), which prevents any specific artist or genre bias. What is even more important, the selection includes experience goods from both major and independent producers.
In the end, the average weekly sample selection from the global NMF playlist amounted to 70. Across all four weeks the study identified and excluded 29 songs, which were not original songs and either remixes or other alternative versions of already released music. After excluding the non-original songs the total sample size resulted at 260. During the study playlist and performance data was collected in 5 successive periods for all songs through Spotontrack. First period being on day of the release, followed by
30
1 week, 2 weeks, 3 weeks and 4 weeks after the release date. Such method was applied to samples from all four global NMF’s and effectively registered the addition and removal of any Spotify curated playlist during the 4 week period. Effectively, the method enabled to exclude any performance in previous periods and register only the most recent weeks’ playlists. Only exception in the study was Viral Charts data, which due to complications were not collected weekly but in aggregate solely after the 4 week period.
3.3 Dependent variables
The first dependent variable of the study was song’s performance. The song’s performance was
operationalized through number of plays and measured in count. As indicated in previous section, the data on performance was collected in weekly intervals. As a limitation of the study, it is very important to point out that the number of plays are total number of plays each song has generated during the weekly period on Spotify. The study does not identify direct sources of plays through the various types of playlists (eg. Spotify curated, fan created, 3rd party, algorithmic or charts), which can in effect all contribute. This
limitation will be further addressed in the discussion section.
The second dependent variable of the study was worldwide word of mouth. The dependent variable was operationalized through count measure of Spotify’s Viral Charts across all 60 worldwide territories where Spotify is available. Spotify’s Viral Charts are aggregates of songs’ sharing and engagement activity on main social media platforms (eg. Facebook, Twitter, Instagram). All Viral charts include a top 50 virality ranking yet as stated this study used only the count measure. Unfortunately, due to complications it was not possible to collect the data for Viral Charts on weekly intervals but only in aggregate for the whole four week period. Meaning, the study was able to register the number of viral charts a song entered during the 4 week period in total but not identify the number of charts for each
31
week individually. Effectively, larger number of viral charts during the 4 week period indicated a larger amount of word of mouth around the song in any of the Spotify’s 60 territories.
3.4 Independent variables
Innovativeness and category spanning were the 2 independent variables used in this study. Both variables were measured in count through number of Spotify playlists, which were defined as either innovative or category spanning playlists. Often throughout the paper the two playlists together will be referred to as expert selectors playlists.
The first independent variable innovativeness captured innovativeness of the song determined by Spotify playlist curators. Innovativeness was operationalized through the number count of all Spotify curated playlists that contain ‘new’ in their title. Spotify features these new music playlists on their platform under ‘New’ section and for users the playlists describe to feature the best and most innovative new music. This is done worldwide on every territory and therefore was decided to be a valid measure of innovativeness. Naturally, such operationalization relies heavily on Spotify music experts’ judgement of innovativeness. There is a bias for platform specific effects, which can affect the generalizability of the results. Yet, since Spotify is the most popular on demand music streaming platform worldwide the platform’s experts can be considered to have relevant expertise in music and established a familiarity with categories among its’ users.
The second independent variable category spanning captured the category spanning ability of a song determined by Spotify playlist curators. Category spanning ability was operationalized through all Spotify curated ‘mood’ playlists. These playlists are under Genres & Moods section on Spotify and are comprised of songs that bridge multiple genre categories. At time of this study it was not yet possible to identify all genre playlists via Spotify’s API, which can be fruitful for future studies. Taken together, with
32
such division in Spotify experts’ playlists the study was able to determine the innovativeness and category spanning ability of the songs.
3.5 Moderating variables
This study used as the only moderating variable the producer of experience goods (ie. songs). A specific distinction was made between major and independent producer of songs. Producer type was used as a moderator because it is thought to moderate Spotify playlist curators support for songs and if either producer’s songs are determined as more innovative or to span more categories.
Major producers’ songs were operationalized as all songs, which had a distribution affiliation with any of the music industry’s three largest record labels: Sony, Universal, Warner or their subsidiaries. Independent producers’ songs were operationalized as all songs, which did not have any affiliations with the three major labels or their subsidiaries.
3.6 Control variables
The study first controlled for playlists owned by the three major labels. Major labels have a wide playlist infrastructure of their own, which makes it fair to assume they can exert additional influence on major owned songs. On Spotify the playlists from the three major labels Sony, Universal, Warner are under the brands Filtr, Digster and Topsify respectively. Lastly, the majors owned playlists data was collected in the same method as for Spotify curated playlists.
Second control variable of the study was star power as used by other selectors researchers in the past (Basuroy et. al 2003; Gazley, Clark, Sinha 2010; Tsao 2014). For this research star power was used in order to control for highly popular blockbuster artists who by default were expected to have more support from curators and to also have more fans. Stars can be more easily added to more playlists and have an active fan base who can listen to the song on their personal playlists. In order to identify stars the study
33
collected weekly Global Top Chart data from Spotify from January 2016 until the beginning of this research in February 2017. The weekly Global Top Chart features 200 most played artists in the world week by week. Hence, if a producer’s song was on the chart during that time he would qualify as a star since to reach such global top chart requires a high number of popularity.
3.7 Methods
The study used Repeated measures ANOVA and hierarchical multiple regression analysis as the two main statistical methods. The methods were used to investigate both hypothesis 1 and 3. Firstly, the study used Repeated measures ANOVA to validate if successive selection occurs. This was done based on the weekly collected playlists and performance data. Multiple selection periods and successive selection were confirmed to occur if there was any increase or decrease in the number of Spotify playlists after the first selection period (day of song’s release). Secondly, after validating that successive selection occurs the study used hierarchical multiple regression analysis in order to analyze the predictive power of Spotify expert playlists for the dependent variables in successive selection. However, since it was not possible to collect Viral Charts data on weekly intervals all data was aggregated into four week totals. As a result, the study was not able to create weekly regression models but two models, which reflected the songs’ performance and word of mouth worldwide for the whole four week period. In this way, the research could analyze the total predictive power of Spotify expert playlists in successive selection but not determine how much the expert selectors playlists affected performance differently on each weekly selection period. Lastly, for hypothesis 2 independent samples t-tests were conducted under the normal distribution rule (N>30).
34
4. Results
In this section the results of the study are presented. First, descriptive statistics are presented followed second by correlations between all variables. Thirdly, the hypotheses are discussed. Lastly, robustness checks for the regression and general analysis are discussed.
4.1 Descriptive statistics
Table 1 presents the sample and distribution of populations of the study. From total sample (N=253) Majors owned songs made up 150 and independent owned songs 103. In percentages these two populations were 59% and 41% respectively. Among the sample 53 stars were found in total. Distribution of stars however consisted mainly of major producers with 74% and only 24% independent producers respectively.
Table1. Full sample distribution table
Table 2 presents through descriptive statistics the number of expert selectors playlists the songs were featured in. Moreover, the table differentiates between innovative and category spanning playlists by the experts and also shows the weekly total number of expert playlists. The analysis shows that during the study’s 4 week period the number of playlists songs were featured in varied on all selection periods. On day of the song’s official release date the total number of expert playlists was (mean=10.15; SD = 13.19), consisting of innovative playlists (mean=6.55; SD=6.96) and category spanning playlists
Population Major Independent
Sample size (N=253) 150 103 Sample % 59% 41% Stars (N=54) 41 13 Star % 76% 24%
35
(mean=3.60; SD=6.23). 1 week after the official release date the total number of expert selectors playlists was (mean=14.06; SD=18.34) with innovative playlists (mean=6.94; SD=7.37) and category spanning playlists (mean=7.12; SD=10.97). More differences can be found between other weeks' descriptives and for instance on the last selection period ‘4 weeks after the release’ where total number of playlists was (mean=6.24; SD=10.07), consisting of innovative playlists (mean=0.84; SD=1.49) and category spanning playlists (mean=5.40; SD=8.58). In sum, Table 2 illustrates that the number of expert selectors playlists differed on all weekly selection periods, indicating that successive selection occurs. However, more analysis was required to investigate if the weekly differences in selection periods were statistically significant. To achieve this, a repeated measures ANOVA model was created.
Table 2 - Descriptives – Spotify expert selectors playlists per songs
N= 253
Expert Selectors Playlists
Week 0* Week 1 Week 2 Week 3 Week 4 Mean SD Mean SD Mean SD Mean SD Mean SD
Innovative 6.55 6.96 6.94 7.37 1.25 1.90 1.07 1.73 0.84 1.49 Category Spanning 3.60 6.23 7.12 10.97 5.21 7.77 3.68 5.52 5.40 8.58 Total # of expert playlists per song 10.15 13.19 14.06 18.34 6.46 9.67 4.75 7.25 6.24 10.07
*on day of song's release
Repeated measures ANOVA was conducted between all the five selection periods (Week 0 - Week 4) and the two expert selectors playlists innovative and category spanning. Final analysis involved multiple statistical tests out of which Box’s test of equality of covariance and multivariate tests were used. Tables with results can be found in Appendix A (Table1).
36
In the first level of analysis, Box’s test was found to be significant at (F=104.86; p=0.001) indicating that covariance within both playlist groups is not equal. Therefore, the study decided to use Pilai’s trace criterion as a valid measure for analyzing the interaction effect between change in number of playlists and time. Test results show that Pilai trace was found to be significant for both time F(4,501)= 64.36 ; p<0.001 and when analyzing the interaction effect between time and playlists (p<0.001). Moreover, partial ETA squared equaled (hp2= 0.339; hp2= 0.348) for time and interaction effects respectively, indicating a large
effect since (hp2>0.25), whereas (hp2>0.09) and (hp2>0.01) would indicate medium and small effect sizes
respectively. In sum, these results confirm that successive selection can occur since the weekly differences in number of Spotify experts’ playlists (Table 2) for songs were largely significant in all of the weekly selection periods. Therefore, the analysis confirmed the main premise of this paper that successive selection can occur by expert selectors. Effectively the study had then collected enough evidence and could move on to perform further analysis and investigate the hypotheses.
