Beauty is in the ear of the beholder : the effect of contrast on the relation between category spanning and performance

(1)

1

Beauty is in the Ear of the Beholder:

The Effect of Contrast on the Relation between

Category Spanning and Performance

Lucas Jan van der Graaf 10006036

June 29, 2015 – Final

MSc. in Business Administration - Entrepreneurship and Management in the Creative Industries Track

Supervised by dhr. prof. dr. N.M. Wijnberg Word Count: 11.584

(2)

2 Statement of Originality

This document is written by Student Lucas Jan van der Graaf 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)

3

Abstract

The creative industries are characterized by extreme uncertainty. To counter this uncertainty categories are used. Research shows that when a product spans categories performance suffers. In addition, the contrast of categories (how distinct the boundaries of a category are) influences this relationship positively. Insufficient research on contrast has been conducted what has led to an inconsistent operationalization of contrast. Furthermore, the causal-model view indicates that the perception of consumers can influence the effect of contrast. Therefore this research focused on creating an objective and subjective contrast measure to answer the question how contrast influences the relation between category spanning and performance. Based on previous research, performance was divided into survival and popularity. It was hypothesized that both objective and subjective contrast influenced the negative effect of category spanning on both popularity and survival negatively. The research was quantitative and set in the popular music industry. Performance data was collected from Billboard 200, data on genres was collected from the Rovi Corporation and subjective contrast data was gathered through a questionnaire with 75 participants. The only significant results contradicted the expectations; contrast had a negative effect on the negative influence of category spanning on popularity. Thus category spanning albums benefit from having high contrast and non-category spanning albums benefit from having low contrast. Additional analyses where different genre levels were considered showed that subjective contrast had a distinct effect on popularity. Explorative analyses also showed that there were differences between subjective and objective contrast for certain genres. In conclusion this research constructed a contrast measure for the music industry and made the first steps in incorporating subjective measures of contrast in category research. Hereby this research has paved the way for future research to use objective and subjective contrast to explain the effects of category spanning.

(4)

4

1. Introduction

“Good Lord Boyet, my beauty, though but mean, Needs not the painted flourish of your praise: Beauty is bought by judgement of the eye, Not utter'd by base sale of chapmen's tongue”. This quote from Shakespeare resembles a truth in the creative industries: It is

impossible to objectively know what is beautiful because it is subjective. The fact that creative goods can’t be valued prior to consumption leads to uncertainty; you do not know the value of a product until it is consumed (Caves, 2000). To counter the uncertainty that arises for both producers and consumers, classification systems are used. Classification systems are extremely important in the creative industries. Artists, producers and customers use it to communicate and understand the products that are produced and consumed (DiMaggio, 1987). Categories function to summarize, classify and label product attributes, uses, and benefits (Rosa, Judson & Porac, 2005) and arise from “social agreements about the meanings

of labels assigned to sets of objects” (Negro, Hannan & Rao, 2010, p. 1399). Common use of

categories are restaurant labels (e.g. Indonesian, fast-food), labels of industries to classify companies (e.g. oil- and gas industry) or genres in film or music.

Research shows that performance suffers when the communication of genres fails. If consumer’s expectations created by genre classification do not match with the actual product, quality judgment will be low (Hsu, Hannan & Koçak, 2009). This mismatch of expectations and reality happens most often when a product spans categories. Category spanning occurs when a product (or producer) belongs to more than one category. The first research on category spanning proposed that the negative influence of category spanning on performance resulted from the mismatch of the audience’s expectations and reality (Zuckermann, 1999). After these initial findings it became apparent that the boundaries of a category affected this relationship. Research found that when category boundaries are less distinct, the negative effect of category spanning is smaller.

(6)

6 The distinctiveness of the boundaries of a category is also called category contrast. Category contrast can be defined as “the average typicality among those with the label who

have a nonzero typicality.”, (Kovács & Hannan, 2014 ,p. 16). Thus the average category

typicality of producers or a product with that category label make up the contrast; it is the degree to which a category stand outs from the background (in this case other categories). The higher the contrast the clearer the boundaries of the category are, and the easier it is to classify products in or out of this category. The presence of atypical members lowers contrast. Research on contrast has been scarce and lead to different operationalizations of contrast (Van Venrooij & Schmutz, 2013; Alexy & George, 2013; Negro, Hannan & Rao, 2010; Kovacs & Hannan, 2010). I will add to this discussion by examining how contrast can be operationalized in the music industry in light of recent literature.

The theoretical basis as used in this article stems from the categorical imperative view (Zuckermann, 1999) and mainly the causal model view (Durand & Paolella, 2013). A question that arose in the early stages of category spanning research was why producers or products performed worse when they spanned categories. Broadly two explanations were given. Firstly producers have to divide their attention and resources across categories what results in suboptimal performance in a single category. Secondly consumers believe that a producer that spans categories performs worse while there is no difference between them and a non-category spanning equivalent. Research supported the latter explanation: consumers believe that category spanning products are worse than non-category spanning products while there is no actual difference in quality (Negro et al., 2010). The causal-model view built upon the importance of audience perception (Durand & Paolella, 2013). The model states that differences in audiences can explain the effects of category spanning. An audience can differ in what they think are crucial components for a product to belong to a certain category. People with for example more knowledge on a subject as music may focus on different features of a

(7)

7 genre to classify a song. This subjective perception of contrast has received very little attention in the category research (Vergne & Wry, 2014) and will therefore be the focus in this study alongside the construction of an objective measurement of contrast. Herby I try to answer the following question: How does contrast influence the effect of category spanning

on performance?

This study will be focusing on the music industry. The music industry makes intensive use of categories (genres) to communicate to consumers and producers (Lena & Peterson, 2008). Genres are defined as a tool to classify music, “it describes a manner of expression

that governs artists work, their peer groups, and the audiences for their work,” (Lena &

Peterson, 2008, p. 697).The music industry is a high-profit and dynamic industry (Strobl & Tucker, 2000) but has been underexposed in the category literature (Van Venrooij & Schmutz, 2013; Lena & Peterson, 2008). For the past years the music industry's total revenue has been decreasing because of (illegal) digital downloads and consequently dropping hard-copy sales (Bhattacharjee, Gopal, Lertwachara, Marsden & Telang, 2007). But in 2013 the total revenue was still around 15 billion dollar (Smirke, 2014). This extremely competitive business pushes producers to constantly gain a competitive advantage (Mol, Wijnberg & Carroll, 2005; Giorgi, Lockwood & Glynn, 2015). To see how category spanning and contrast function in this dynamic and competitive industry serves as basis for other industries as well.

This study will look at how objective contrast and subjective contrast influences the relation between category spanning and performance. This will be done in a quantitative research setting. I will create an objective contrast measure from performance data collected from the Billboard charts and genre data collected from the Rovi corporation. Besides this I will create a subjective contrast measure by means of a questionnaire. This questionnaire will furthermore serve as the basis for explorative research concerning the role of audience differences in the construction of subjective contrast.

(8)

8 In the next section I will discuss literature concerning category spanning, the causal-model view and contrast after which I will present my hypotheses. Thereafter I will explain the data collection and methodology. After that I will present the results of this study. I will conclude with a discussion on the results, limitations and directions for future research.

2. Literature Review

Category spanning occurs when a product (or producer) belongs to more than one category (for example a reggae and hip-hop album). Category spanning is in other words what happens when a product (or producer) crosses category boundaries. “In absolute terms, category

boundaries define what lies inside and what lies outside a category. In relative terms, category boundaries help distinguish between different categories. Category boundaries often consist of lists of category attributes that offerings must possess to fall within the category's boundaries”, (Vergne & Wry, 2014, p. 69). Category research looks at producers or products.

For my research I will focus on product categories: “A product category is recognized as such

when similar socio-technical artefacts come to be exchanged as products within a distinct market segment that serves as a basis for interaction between producers, buyers, and external audiences (e.g., critics).”(Vergne & Wry, 2014, p. 68). As mentioned earlier, category

spanning has been a big topic in the category research. Findings from previous research will be discussed in the next section.

2.1 Effects of Category Spanning

Research found that spanning boundaries has a negative effect on performance (e.g. Zuckermann, 1999; Hsu, 2006; Kovàcs & Hannan 2010). Early research that examined the negative impact of category spanning on performance is the research by Zuckermann (1999). He introduced the categorical imperative view which states that organizations that span multiple categories are overlooked by the audience. This is because they believe those

(9)

9 organizations have less expertise. When a producer has to disperse his attention across multiple categories less expertise can be acquired for a single category. This theoretical framework inspired more research that supported the hypothesis that organizations that span categories perform worse.

Hsu, Hannan and Koçak (2009) found support for the categorical imperative with two empirical tests on specialists (producers that focus on one category) and generalists (producers that focus on multiple categories) in the feature-film industry. They found that when products incorporate features from different categories, the audience will consider the product to be poorly fitting with the category. Category fit is in this case the extent to which a product falls into the category boundaries; therefore a poor category fit occurs when a products has little of the features that belong to the category, where a good category fit occurs when a product has a lot of the features that belong to the category. Taken together, category spanning products are less likely to meet the category's expectations and therefore receive sanctions from the audience. They also found that producers will be less able to focus their resources on specific categories/audiences. As such, if they want to incorporate multiple categories they will have to disperse their attention across audiences. This would subsequently lead to lower performance in a specific category.

In another empirical study in the movie industry Hsu (2006) found that audiences have a negative reaction concerning the movie’s quality when a movie spans multiple categories. This was ascribed to the fact that producers cannot produce products that show features of several categories as well as products that clearly fall in one category. Furthermore, Hsu (2006) found that audiences were larger for movies that spanned multiple categories. Even though the quality judgment decreases, which in turn leads to lower playability (and so a shorter runtime), the marketability is higher for category spanning products. That is, audiences are attracted to one of the multiple categories. In sum audiences were more

(10)

10 interested in movies that spanned categories (higher attention), but found them of less quality (lower appeal). Because of this, category spanning movies had a good opening weekend, but had short survival (the weeks the movie played in the cinema). This conclusion illustrates that the behavior of audiences is shaped in different ways by category spanning. I will therefore also look at attention and appeal as two different performance outcomes.

Another explanation for the negative effect of category spanning are that audiences experience less meaningful identities with category spanning products (Negro et al., 2010). This may stem from the fact that audiences cannot determine in what category a producer belongs when a producer spans multiple categories (Zuckermann, 1999). In summation the previously discussed research comes down to two general explanations. Firstly, category spanning products (or producers) are of lower quality than non-category spanning products. Secondly, the perception of quality from the audience results from the perception of category spanning. This subjective view of quality consequently results in lower performance.

Negro and Leung (2013) examined these two possible explanations further: producers actually perform worse when spanning categories or audiences perceive the producer as a bad category fit and therefore give them poorer ratings. They researched this question by measuring blind and non-blind expert ratings on wines. The ratings in the blind condition were seen as an objective quality measure; the experts did not know which wines spanned categories and which ones did not. By controlling for the scores given in the blind condition they found that producers that spanned categories still received lower scores. This means that the perceptual categorical fit determines the effect of category spanning on quality ratings. These conclusions were supported by research in crowd-funding by Leung and Sharkey (2013). In this research producers with the same quality signals were measured on the support they received. The results were that producers received less support when spanning categories. Also from social research on boundaries Lamont and Molnar (2002) conclude that

(11)

11 the perception of boundaries differs per audience. These findings indicate that audience perception is key in explaining the effects of category spanning and maybe less so the actual performance of the producer.

2.2 Causal-Model View

The fact that the audience believes that a “multiple-category” product is inferior to a “one-category” product while there is no actual difference raises the question how this perception comes to be. In their overview article Durand and Paolella (2013) differ from the categorical imperative view to explain why there are still hybrid or category spanning organizations and how audience perception is indeed important to explain the effect of category spanning. To illustrate this they use the causal-model view. The causal-model view states that certain features of a category are more important (have higher valence) then other features to determine to which category someone or somewhat belongs to. This view stresses that what is considered an important category feature differs per audience. An audience with, for example, more knowledge on a subject will think certain features are more important in defining prototypical category membership. An audience with less knowledge may focus on obvious but less important features. An example: Someone who doesn’t listen to jazz might say something is a typical “cool-jazz” album because it features slow tempos what clearly can be heard in the recordings. But someone who considers himself a jazz expert might consider something a typical “cool-jazz” album because of the intricate harmonies, which is harder to pick up on and requires more expertise in the jazz field to know. Thus, differences in the audience (like knowledge) result in how they think something is a typical category member. This is than determinative for the construction of category boundaries and determining contrast. This means that the effects found in the previously discussed literature are not that simple and could heavily depend on audience perception.

(12)

12 Vergne and Wry (2014) wrote an overview article on categorization where multiple issues and directions are discussed for future research. One important underexposed issue is the difference in audiences and the effect on category spanning. Resulting from social research Lamont and Molnar (2002) found that audience characteristics determine how boundaries are seen. Furthermore, there is some research on differences in audience perception, but mostly not related to category spanning. Research for example found that people with high expertise rated products differently and also responded differently to reviews than people with lower expertise (Zhu & Zhang, 2010). Pontikes (2012) also found that there was a different reaction between audiences concerning category spanning. They tested the difference in reaction to category spanning between makers (producers) and market-takers (consumers). They found that for producers, category spanning can look like innovation, whereas this ambiguity for consumers just makes the categories vague. Less related to category spanning but underlining the importance of audience characteristics is the research of Roose, Van Eijck and Lievens (2012). They identified three main consumer characteristics (cultural capital, economic capital and age) that influence openness for certain (more innovative) art forms. The research stated the more cultural capital one has the more one is open for new art forms. This is not directly related to category spanning, but it clearly shows that appreciation of new forms of art is related to the audience’s characteristics.

This research will look at audience perception as an determining factor in the relation between category spanning and performance. Expertise is mentioned as a determinant in audience perception. Audience expertise refers to the amount of knowledge one has on a certain topic (in this case musical genre). Research showed that experienced listeners and musicians experience music differently in contrast to non-experienced listeners (Besson, Schön, Moreno, Santos & Magne, 2007; Bigand & Poulin-Charronat, 2006; Bigand, Vieillard, Mardurell, Marozeau & Dacquet, 2005). They have different brain activity and responses

(13)

13 during listening to music due to different processing. Expertise influences how music is perceived and will therefore influence the way audiences see category boundaries. This research will therefore investigate audience expertise as a possible determinant of audience perception.

In summation, the previous literature on category spanning has moved to a view where perception of quality by audience determines the effects of category spanning; not actual performance of the producer. To explain this effect we have to look at the subjective audience perception of a category but also at the “objective” properties of a category. One of the most important properties of a category is the boundary: Which products fall in between boundaries of a category and which don’t? Research on category boundaries focusses on the contrast of a category (Negro et al., 2010; Kovács & Hannan, 2014).

2.3 Category Contrast

As mentioned category contrast refers to the extend a category stands out from its background. Contrast is usually measured by calculating the average grade of membership of all the members of the category (Carroll, Feng, Le Mens & McKendrick, 2010; Negro et al., 2010; Kovács et al., 2010). Some research uses the term “fuzziness” to indicate the distinctiveness of the boundaries. Contrast and fuzziness of boundaries are two related concepts, but I will mainly use the term contrast in this research. A category has fuzzy boundaries when it is has low contrast. Vice versa, a product with high contrast does not have fuzzy boundaries. Bogaert, Boone and Carroll (2010) define contrast of a product as how distinct the product is from other products.

Research found that the less contrast a category has, the less negative the effect of category spanning is on performance (Kovács & Hannan, 2010). This was researched by looking at restaurants and their categories (e.g. French, Kosjer, Indonesian). Contrast was constructed by measuring per category how many of the category members had multiple

(14)

14 category memberships. When a category (for example Indonesian) contained a lot of members who belonged to multiple categories, the contrast of that category would be low. Whereas when that category had only members that belonged to that category and none other, the contrast of that category would be high. So when a category had a lot of members spanning categories the negative effect for a producer spanning that category would be less severe. Other research also in the restaurant industry from Kovács and Hannan (2014) found that the more salient (high-contrast) the boundaries are, the more severe the negative effect of category spanning is on performance.

In line with the above-mentioned findings Alexy and George (2013) found a positive effect of blurring the boundaries between categories. Their research spun over a period of ten years which was set in the IT industry. Contrast was measured by counting press releases where a firm was mentioned together with category-specific activities. If a category was often accompanied by other categories in these activities, the boundaries for that category would be considered blurry. The categories with blurry boundaries (low contrast) had a smaller negative effect than categories with clear boundaries (high contrast) when spanned. Negro et al. (2010) looked at contrast in the wine-making industry where contrast for wineries was constructed by looking at previously produced wines. When the categories that the previously produced wines had a large niche-width (niche width is how big the category is, i.e. how many different wines could belong to that category) the contrast for that producer would be low. This research showed a negative effect of having high contrast when a product spans categories, but a positive effect of contrast when a product does not span categories.

In summary, research from different fields provide support for a positive influence of contrast on the negative effect of category spanning on performance. In this research I will focus on the contrast of a product. Furthermore, I will make a divide between objective contrast and subjective contrast. Objective contrast will be constructed per category as

(15)

15 discussed in Kovacs and Hannan (2010) by measuring average typicality. Subjective contrast will be constructed by means of a questionnaire. I will use these category contrast measures to construct the objective and subjective contrast of an album. This will be further explained in the methodology.

A proposed measure of subjective contrast has never been constructed nor has a proposed measure of objective contrast has been constructed in the music field. In the next section I will present my hypotheses before moving to the methodology.

2.5 Hypotheses

To answer the question on how contrast influences the relation between category spanning and performance I propose two hypotheses which will be defined in this section. The first hypothesis concerns the influence of objective contrast in the relation between category spanning and performance. The second hypothesis concerns the influence of subjective contrast on the relation between category spanning and performance.

The research of Van Venrooij and Schmutz (2013; in press) overlaps partly with this study. In one of their articles contrast was defined (in their case fuzziness) in the music industry where it was linked to popularity (Van Venrooij and Schmutz, 2013; in press). Fuzziness was constructed by looking at the co-occurrence of genres in reviews. This resulted in a matrix with contrast measures per genre dyad. The fuzziness of a single album was constructed by using a formula to calculate the fuzziness of all the possible combinations of genre per album. So category fuzziness was co-occurrence of genres and album fuzziness was all possible “fuzziness-measure” dyads taken together. The problem with this operationalization is that co-occurrence is used as fuzziness. Co-occurrence, or distance, is closely related to contrast but treated differently in this paper. I define distance as the closeness of a set of categories/genres and contrast as a measure of one genre where that genre is compared with its complete background (Kovács & Hannan, 2010; 2014). It is often

(16)

16 the case that two categories with high distance have less contrast with each other (and vice versa). Unfortunately for reasons related to time and resources distance could not be taken into account in this research.

The study by Van Venrooij and Schmutz (2013) looked at expert reviews as a performance measure. In line with the results of Hsu (2006) they found that the number of reviews increased by fuzziness, but review scores were lower. A big problem with this research is that they used contrast in the construction of the independent variable (fuzziness). They also based both their independent and dependent variable on expert reviews, what results in a single source bias. However, their research setting to a large extent represents the research setting in this article. While category spanning and contrast are used differently it does show that fuzziness or contrast affects the relation between category spanning and performance. Both in this article and in Hsu (2006) there are two different performance outcomes, appeal (do people actually like it) and attention (how many people are interested in going). In the case of popular music you can make that distinction by looking at the most popular an album has been and how long an album stays popular. This would respectively translate in popularity and survival.

The discussed research shows that higher contrasting categories result in more negative effects of category spanning (Kovács, Balázs & Hannan, 2010; Negro et al., 2010; Kovács & Hannan, 2014). Alexy and George (2013) also found a positive effect of blurring the boundaries (reducing contrast) between categories. The research of Van Venrooij and Schmutz (2013) suggests that also in the popular music higher contrast leads to increased negative effects of category spanning, in this case lower popularity and lower survival. This leads to the following hypotheses (See also Figure 1):

(17)

17

Hypothesis 1a: Objective contrast moderates the relation between category spanning and

popularity so that when the maximum contrast of a product is higher the negative effect of category spanning is higher.

Hypothesis 1b: Objective contrast moderates the relation between category spanning and

survival so that when the maximum contrast of a product is higher the negative effect of category spanning is higher.

Resulting from the causal-model approach, audience perception seems key in explaining the effect of category spanning on quality judgment. Previous research shows that differences in an audiences in the creative industries leads to different behavior (Roose, et al., 2012) and also to different reactions to category spanning (Pontikes, 2012).

Van Venrooij and Schmutz (in press) researched the difference in effect of categorical ambiguity (operationalized as fuzziness of the genres) on performance between two popular music fields. They researched how categorical ambiguity influenced critical acclaim (as measured by looking at expert reviews) and commercial acclaim (measured by looking at Billboard rankings). The different fields were the commercial field and artistic field, both subfields in the pop-music field. These sub-fields have different audiences; the commercial field is mainstream oriented and the artistic field is independent oriented. They found that the negative influence of categorical ambiguity was lower in the artistic field than in the commercial field. This shows that a difference in audience results in a different reaction to category spanning. It underlines the importance of taking audience perception into account when looking at the relation between category spanning, contrast and performance.

Because there has not been any research on the effect of subjective contrast I assume that the effect of subjective contrast will be the same as that of objective contrast. This results in the following hypotheses (see also Figure 1):

(18)

18

Hypothesis 2a: Subjective contrast moderates the relation between category spanning and

popularity so that when the maximum contrast of a product is higher the negative effect of category spanning is higher.

Hypothesis 2b: Subjective contrast moderates the relation between category spanning and

survival so that when the maximum contrast of a product is higher the negative effect of category spanning is higher.

It seems that contrast experienced by an audience (subjective contrast) will differ from contrast as measured through genre-data (objective contrast). How this will differ is rather unclear. It is possible that contrast will be lower because people do not know genres that well and do not know the exact genre boundaries. Alternatively, people could experience higher contrasts because they listen to a specific genre very often and create an own image of where boundaries of that genre are. It seems that in the case of music appraisal audience expertise is an important factor. As discussed audiences with higher expertise have more knowledge on a certain genre. According to the causal-model view that difference in knowledge may result in a difference in view of category boundaries. Therefore to explain subjective contrast I will exploratively look at the influence of audience expertise on subjective contrast.

Figure 1 Hypotheses one and two

Category Spanning

Objective Contrast Subjective Contrast H2: + H1: + - H2 Popularity Survival a b

(19)

19

3. Methodology

In this section I present the methodology used in this research. I will begin with the research setting after which I will discuss the data collection and sample. Hereafter the operationalization of the variables is discussed.

3.1 Research Setting

The research was quantitative and set in the popular music industry. As with all creative industries the music industry is characterized by a “winner takes all” system (Giorgi, et al. 2015). One of the big indicators of success in popular music is placement on popularity charts (for example the Billboard top 200). As discussed genres are extensively used in the music industry (Lena & Peterson, 2008). The importance of the Billboard charts as an indicator for success and the extensive use of categories makes the music industry an interesting field for category research.

This research consisted of database research and a questionnaire. The database consisted of Billboard 200 rankings from the past 5 years and genre data from the Rovi Corporation. A questionnaire measured subjective genre contrast as perceived by consumers.

3.2 Data collection and Sample

Billboard.com Data on album performance was collected from the Billboard 200 chart from

March 2010 to March 2015. The Billboard top 200 is the main source of data on commercial success of albums (see also Van Venrooij & Schmutz, 2013). The performance data from the Billboard 200 is based on sales from hardcopy and digital sales. I wanted to measure the performance of albums that were released in the period in which I have measured the performance. It is the case that previous successful albums tend to linger in lower regions of the top 200. This has led to “classic” albums appearing for very long periods of time in the top 200. I also do not think that one week on the 200st spot resembles the same as a week on the 1st or even 50th spot. I therefore made the decision to limit the list to the top 50 to only have

(20)

20 truly successful albums. I have also deleted albums that were produced before the period of which I have collected Billboard data. The sample from Billboard.com consisted of 2751 albums that were featured in the top 50 of the Billboard 200 from the first week of March 2010 to the second week of March 2015.

Rovi The Rovi Corporation collects metadata on albums which is presented on

Allmusic.com. This site features extensive data on genres of albums. It also features data on the label, artist and other features like collaborating artists. Allmusic.com has an extensive database that could be accessed and is also used in other research to collect data on genres (see Van Venrooij & Schmutz, 2013). With a search script data was collected from allmusic.com. This search returned data on 914 albums. The low number of returned data on albums is partly due to the fact that soundtracks and compilation albums did not have a clear artist and were therefore not found. The majority of unfound albums was due to a misalignment of Billboard and Rovi data. However with a final sample of 914 albums the analyses could be performed.

Questionnaire participant The sample for the questionnaire consisted of 92

participants. Participants were gathered through my personal network and via social networks. The questionnaire took approximately five minutes to complete. Participants received no reward for their participation.

3.3 Dependent Variables

Performance Performance was divided in survival and (maximum) popularity respectively

measured with “weeks in top 50” and "peak position". Weeks in top 50 was measured as the sum of weeks an album was in the top 50 of the Billboard 200.The peak position was measured by looking at the highest position an album occupied during that period. For additional analyses weeks in the top 200 was taken as a performance measure and operationalized as the sum of weeks an album was in the Billboard 200.

(21)

21

3.4 Independent Variables

Genres A category or genre in music is based on multiple aspects like playing style, pitch,

timbre, lyrics, and dynamics (e.g. Lee, Shih, Yu & Su, 2007; Scaringella, Zoia & Mlynek, 2006; Lena & Peterson, 2008). Previous research makes use of genre taxonomies; a hierarchical organization of genres in several levels; main-genres, sub-genres and "sub-sub-genres", called styles. Using hierarchical taxonomies in music categorization improves usability and also shows the relationships in between (sub-)genres (Li & Ogihara, 2005). A similar hierarchical distinction in genre levels has not been used in previous research and was therefore used in this research.

The genre taxonomy from allmusic.com is hierarchically from the top Main-Genre (e.g. Pop/Rock), Sub-Genre (e.g. Heavy Metal) and Style (e.g. Goth Metal). Data from allmusic.com contained main genres and styles, but no sub-genres. Using computer script styles were converted to sub-genres by replacing each style with the accompanying sub-genre. This resulted in a total of 63 sub-genres of which there was at least one member with non-zero typicality (see appendix A). The data on genres from allmusic.com also included a typicality measure per main-genre and style which represented the amount that genre was represented in that album. This typicality or grade of membership (GoM) ranged from 5 to 9 for styles and from 5 to 10 for main-genres. When an album had two or more styles of a sub-genre the highest possible GoM was used to indicate the GoM of the sub-genre.

Category spanning Category spanning was measured by adding the number of genres

that were spanned. This was done on sub-genre level. Category spanning on main-genre level would leave out important data, for example an album that belongs to the heavy metal genre and singer-songwriter genre would be classified as pop/rock, while it clearly spans genres. Category spanning on style level would create less accurate data, for example an album that belongs to the Goth Metal, Heavy Metal, Post-Grunge and Alternative Metal styles would be spanning four genres while it clearly does not. Although category spanning was mainly

(22)

22 measured on sub-genre level a category spanning measure on main-genre was included for explorative analyses. This was to see if this “more extreme” form of category spanning yielded different results.

3.5 Moderating Variables

Objective Category Contrast Category contrast was measured on sub-genre level. Using the

same methods as Kovács and Hannan (2010; 2011; 2014) contrast per sub-genre was constructed by calculating the average GoM of category members with a non-zero typicality for that genre. To establish the weighted contrast per album first the percentile of each genre contained by that album was calculated by dividing the GoM of a genre by the total GoM the album contained. Then the percentile of each genre of that album was multiplied by the average contrast of the accompanying sub-genre. Besides this measure of average contrast the maximum contrast was also constructed for explorative analyses. Maximum contrast was calculated by using the contrast of the genre that contained the highest contrast in that album, regardless of the highest GoM. Contrast per genre can be found in the appendix A.

Subjective Category Contrast Subjective category contrast was measured with an

online questionnaire. After a statement of anonymity participants were asked to rate each of the sub-genres of the pop/rock main genre on contrast. Only the pop/rock subgenres were selected to keep the questionnaire short to attract participants. Pop/rock was the largest main-genre and was therefore selected. This resulted in 17 subjective contrast measures per subgenre (see appendix A). After a brief explanation of the concept of genre boundaries participants were asked “How clear are the boundaries of this genre for you?”. The particpants then had to rate each sub-genre on a scale from 1 (very low contrast) to 10 (very high contrast). There were 3 different versions both in Dutch and English of the test with each another random presentation order of the sub-genres (see Appendix B for full questionnaire).

(23)

23 After the contrast measures people were asked three questions on expertise (an example question is “How often do you give music recommendations?”) that they scored on a Likert scale from 1 (Never) to 7 (Very Often). A reliability analysis of the three expertise questions resulted in an α of .891 for the average measure of expertise. The questionnaire concluded with a question on age and sex. The test took approximately five minutes.

3.6 Control Variables

Second album Previous success is an important determinant for latter success (Battacharjee,

Gopal, Lertwachara, Marsden & Telang, 2007). This previous success was operationalized by looking if there was an album by the same artist featured in the top 50 before. This resulted in a dummy variable for “second album” with the score 1 meaning no (it is a first in the top 50) and 2 meaning yes (it is a second album of that artist in the top 50). I did not look at the difference between a second, third or fourth album. This was done because I only use this variable as a control measure which I wanted to keep dichotomous and simple.

Major Label An album has more resources when released on a major label. Besides a

larger network there is a larger budget for marketing and production. To account for this dissimilarity between a release on a major label or on an independent label, major label was taken as control variable. It was operationalized as a dummy variable with 0 meaning independent label and 1 meaning major label.

4. Results

In this section the results are discussed. I will first discuss the descriptive statistics including missing variables and deleted cases. Afterwards I will discuss the correlations. Then I will test my hypotheses using an OLS regression. Lastly I will discuss the additional analyses.

(24)

24

4.1 Descriptive statistics

From the first sample of 914 albums 15 contained missing data and were therefore removed. One other album was a soundtrack to a film. Because of limited resources I couldn’t account for the film’s success as a co-variate, therefore this entry was removed. The release date of the albums ranged from 1969 to 2015 (See Table 1). As expected the majority was released between 2010 and 2015; the years from which Billboard data was collected. In total 34 albums were not produced in this period and were considered “classics”. These albums are so successful they have a continuing success that goes far beyond their initial release success. This continuing success is not only based on “normal” factors that contribute to success (like marketing during release) but also to “superstar” or classic status of an album or artist. This recurring success can also come from other factors, for example the death of an artist (e.g. Amy Winehouse). There were too many different variables to account for and therefore these albums were removed from the sample. Lastly two albums were considered outliers because of their category spanning scores which were far above the rest (SD>4). This resulted in a final sample of 862 albums. A correlation analysis showed that release date was not correlated with any of the key variables: Peak position, r=-.050, p=.139, Weeks in Top 50, r=.053,

p=.122, Objective Contrast r=.008, p=. 812, Subjective Contrast, r=.009, p=. 843, Category

(25)

25 Table 1 Release Dates of Albums

Release N % 1969 1 0,1 1977 1 0,1 1978 1 0,1 1979 1 0,1 1993 1 0,1 1994 1 0,1 1999 1 0,1 2000 1 0,1 2003 3 0,3 2004 4 0,4 2005 1 0,1 2008 7 0,8 2009 11 1,2 2010 151 16,8 2011 154 17,2 2012 154 17,1 2013 168 18,7 2014 211 23,5 2015 26 2,9 Total 898 100,0

Seventeen respondents from the questionnaire missed crucial data and were therefore removed from the data. This resulted in a final sample of 75 respondents (35 male, 35 female, 3 unknown) with an average age of 29.76 (SD: 12.75). Hereafter objective and subjective contrast were calculated per album.

4.2 Correlations

A correlation analysis was performed with Peak Position, Weeks in Top 50, Category Spanning on sub-genre level, objective- subjective contrast, second release (yes or no) and major label (yes or no). Peak position has been reverse scored for all further analyses so that a higher score means a better performance. The results together with the mean and standard deviation are presented in table 2.

Support for splitting performance in popularity and survival can be found by looking at the lack of correlation between peak position and weeks in top 50. This suggests that the

(26)

26 peak of popularity does not necessarily relate to the survival or life span of an album. The correlation between distribution through a major label and peak position, but not with weeks in top 50, also links to this; there is a larger budget for production and marketing what leads to more attention, but does not necessarily lead to better quality. This in turn leads to high sales for a short period during release. The same correlation is found between second release and peak position, but not between second release and weeks in top 50. Here popularity also increases, but survival does not; a second release builds on the success of a first successful album to create attention.

Another interesting correlation is found between category spanning and contrast; more category spanning is related to lower contrast. This is in line with the before discussed literature; spanning categories leads to less clear boundaries. This correlation can indicate multicollinearity between these variables. A same correlation for subjective contrast can be found with main genre spanning, but not with sub-genre spanning. This could mean that people see an album as having less contrast when it crosses main genres. But because subjective contrast was only measured on subgenre level it is hard to truly say something about the relation between main genre spanning and subjective contrast. This significant correlation should however be interesting for further research. I elaborate more on this in the explorative analyses.

Subjective contrast and objective contrast are correlated. I will look further into the difference between objective contrast and subjective contrast in the explorative analyses. Lastly subjective contrast and major label had a positive correlation. A small correlation but it is apparent that major labels produce genres that are more distinctive according to the public.

(27)

27 Table 2 Correlation Matrix

M SD 1 2 4 6 7^ 8 9

Performance

1Peak Position 19.05 14.93 1

2 Weeks In Top 50 4.92 11.12 .036 1

Category Spanning

4 Sub Genre Level 2.03 1.00 -.009 .008 1

Contrast

6 Objective Contrast 7.40 .35 -.045 .020 -.366** 1

7 Subjective Contrast^ 5.93 1.03 .041 .042 -.059 .257** 1

Control

8 Second release (1=no, 2=yes) 1.18 .386 .140** -.036 -.037 .033 .087 1 9 Major Label (0=no, 1=yes) .81 .395 .245** .001 .079* -.058 .101* .049 1

N=862, ^N=470,*p<.05, **=p<.01

4.3 Hypothesis testing

Hypothesis 1. A Shapiro-Wilk test was significant for the independent - and moderator

variables (category spanning on sub-genre level, S-W=.243 p<.001, contrast on sub-genre level, S-W=.932, p<.001), so normality was assumed. As can be seen from Table 3 subjective contrast and category spanning are correlated. To account for this multicollinearity both these and all other variables were centered. Homoscedasticity was confirmed by examining the Q-Q plots of the variables. The Durbin-Watson statistic for all variables in this model was 1.980 so independence was assumed.

A hierarchical regression was performed with second release and major label as control variables, category spanning and contrast on subgenre level as independent variables and the interaction of these latter two variables on peak position (See Table 3). The first model with second release and label as predictors and peak position as performance was significant F(2, 861)=35.601, p<.001. As expected the effect of major label and second release was very large; together they explained 7.7% of the variance found in peak position. The second model included category spanning and was also significant, F(3, 860)=23.882,

p<.001, but did not lead to a significant increase in R², ∆F(1, 858)=.487, p=.485. Also when

(28)

28 857)=18.454, p<.001, but the increase in R² was not, ∆F(1, 857)=2.080, p=.150. The fourth model included the interaction between category spanning and contrast. This model was also significant F(5, 856)=16.841, p<.001 and the increase in R² of 1% was too, F(1, 856)=9.645,

p=.002. Because of this significant change in R² the model explained 9% of the variance in

peak position. While this model is significant, the interaction is the opposite as expected based on hypothesis 1a. So hypothesis 1a was not confirmed: Contrast has a negative effect on the negative influence of category spanning on popularity (see figure 2).

The main-effect of category spanning on subgenre level was only significant when the interaction with contrast was added to the model. This means that the negative effect of category spanning is heavily dependent on the measurement of contrast.

Table 3 Peak Position regression effects (objective contrast)

R R² ∆R² B SE β t Model 1 .277 .077** - Second Release 4.973 1.269 .129 3.901** Label 9.010 1.239 .239 7.294** Model 2 .278 .077** .001 Second Release 4.936 1.270 .128 3.883** Label 9.080 .1.243 .241 7.311** Category Spanning -.343 .491 -.023 -.544 Model 3 .282 .079** .002 Second Release 4.977 1.270 .129 3.920* Label 9.022 1.243 .239 7.257** Category Spanning -.618 .527 -.041 -1.173 Contrast -2.141 1.485 -.051 1.442 Model 4 .299 .090 .010 Second Release 5.057 1.264 .131 4.002** Label 8.725 1.241 .231 7.032** Category Spanning -30.073 9.499 -.2.015 -3.166** Contrast -10.524 3.077 -.250 -3.420* Interaction 4.026 1.296 1.913 3.106* N=862, **p<.001. *p<.005

(29)

29

Figure 2 Interaction between category spanning and objective contrast on peak position

In Table 4 the results from a hierarchical regression with second release, major label, category spanning, contrast and the interaction between category spanning and contrast on weeks in top 50 are presented. Both the first model, F(2, 859)=.557, p=.573 and the second model, F(3, 858)=.381, p=.766, were not significant. The third model that included subjective contrast was also not significant, F(4, 857)=.423, p=.792.With the interaction added the model still was not significant, F(5, 856)=.363, p=.874. None of the changes in R² were significant (the most significant ∆R² was when subjective contrast was added, F(1, 857)=.549,

p=.459) (see figure 3).This means that hypothesis 1b is not supported.

Low Category Spanning Average Category Spanning High Category Spanning Peak Po si tion Low Contrast Average Contrast High Contrast

(30)

30 Table 4 Weeks in top 50 regression effects (objective contrast)

R R² ∆R² B SE β t Model 1 .036 .001 .001 Second Release -1.038 .983 -.036 -1.055 Label .072 .960 .003 .075 Model 2 .036 .001 .000 Second Release -1.031 .985 -.036 -1.047 Label .058 .964 .002 .060 Category Spanning .067 .381 .006 .177 Model 3 .044 .002 .001 Second Release -1.047 .985 -.036 -1.063 Label .081 .965 .003 .084 Category Spanning .177 .409 .016 .433 Contrast .854 1.152 .027 .741 Model 4 .046 .002 .000 Second Release -.040 .986 -.036 -1.055 Label .055 .986 -.036 -1.055 Category Spanning -2.434 7.411 -.219 -.328 Contrast .111 2.401 .004 .046 Interaction .357 1.011 .228 .353 N=862

Figure 3 Interaction between category spanning and objective contrast on weeks in top 50.

Hypothesis 2. To test the interaction as proposed in hypothesis two a moderated

regression analysis was conducted. A Shapiro-Wilk test for subjective contrast was significant, S-W=.941, p<.001 so normality was assumed. The Durbin-Watson test

Low Category Spanning Average Category Spanning High Category Spanning Wee ks in t o p 50 Low Contrast Average Contrast High Contrast

(31)

31 approached 2 (2.125), therefore independence of variables was assumed. Homoscedasticity was confirmed by examining the Q-Q plots of the variables.

The results of the hierarchical regression with second release and major label as control variables, category spanning and subjective contrast on subgenre level as independent variables and the interaction of these latter two variables on peak position can be seen in Table 5. The first model with second release and major label as predictors explained 6% of the variance in peak position, F(2, 467)=14.845, p<.001. Model 2 with category spanning added was also significant, F(3, 466)=10.280, p<.001, but R² didn’t change ∆F(1, 466)=1.141, p=.286. Also model 3 was significant, F(4, 465)=7.716, p<.001, but there no significant change of R², ∆F(1, 465)=.083, p=.773. With the interaction added model 4 was still significant, F(5, 464)=6.256, p<.001 (see figure 4), but the change in R² was not, ∆F(1, 464)=.455, p=.500. This means that hypothesis 2a is not supported.

Table 5 Peak Position regression effects (subjective contrast)

R R² ∆R² B SE β t Model 1 .244 .060 - Second Release 5.011 1.749 .129 2.865 Label 7.257 1.626 .201 4.464 Model 2 .249 .062 .002 Second Release 5.173 1.755 .133 2.947* Label 6.992 1.644 .193 4.252* Category Spanning .885 .828 .049 1.068 Model 3 .249 .062 .000 Second Release 5.134 1.762 .132 2.913* Label 6.941 1.655 .192 4.193* Category Spanning .901 .831 .050 1.084 Subjective Contrast .190 .657 .013 .289 Model 4 .251 .063 .001 Second Release 5.265 1.774 .135 2.968* Label 7.003 1.659 .194 4.222* Category Spanning -2.668 5.354 -.147 -.498 Subjective Contrast -.759 1.552 -.052 -.489 Interaction .609 .902 .205 .675 N=470, *p<.001

(32)

32

Figure 4 Interaction between category spanning and subjective contrast on peak position

To test hypothesis 2b a moderated regression was executed (Table 6). None of the models were significant: model 1, F(2, 467)=1.511, p=.222, model 2, F(3, 466)=1.094,

p=.351, model 3, F(4, 465)=1.128, p=.343 and model 4 including the interaction, F(5,

464)=.901, p=.480 (see figure 5). The highest ∆R² was found from model 2 to 3 (adding subjective contrast), ∆R² =.002, ∆F(1, 465)=1.227, p=.226. So hypothesis 2b was not supported. Low Category Spanning Average Category Spanning High Category Spanning Peak Po si si to n Low Contrast Average Contrast High Contrast

(33)

33 Table 6 Weeks in top 50 regression effects (subjective contrast)

R R² ∆R² B SE β t Model 1 .080 .006 - Second Release -1.230 1.170 -.052 -1.129 Label -1.365 1.087 -.058 -1.256 Model 2 .084 .007 .001 Second Release -1.372 1.175 -.054 -1.168 Label -1.280 1.101 -.054 -1.163 Category Spanning -.258 .554 -.024 -.514 Model 3 .098 .010 .003 Second Release -1.471 1.178 -.058 -1.249 Label -1.410 1.107 -.060 -1.274 Category Spanning -.243 .555 -.021 -.438 Subjective Contrast .486 .439 .052 1.108 Model 4 .098 .010 .000 Second Release -1.481 1.186 -.058 -1.248 Label -1.414 1.109 -.060 -1.275 Category Spanning .019 3.581 .002 .005 Subjective Contrast .556 1.038 .059 .536 Interaction -.045 .603 -.023 -.074 N=470

Figure 5 Interaction between category spanning and subjective contrast on weeks in top 50

4.4 Robustness checks and explorative analyses

For these additional analyses I included different variables. I looked at category spanning on main-genre level, maximum subjective contrast per album , maximum objective contrast per album and weeks in top 200 (Table 7).

Low Category Spanning Average Category Spanning High Category Spanning Wee ks in t o p 50 Low Contrast Average Contrast High Contrast

(34)

34 Table 7 Descriptive statistics additional variables

N M SD

Main Genre Spanning 862 1.28 .558

Max Objective Contrast 862 7.591 .283

Max Subjective Contrast 470 6.608 .920

Weeks in Top 200 862 15.577 24.849

Robustness Checks

To test for robustness the analyses with weeks in top 50 as outcome variable were done with weeks in top 200 as outcome variable. Both models were still insignificant, objective contrast: R²=.004, F(5, 856)=.704, p=.621, due to interaction ∆R² =.000, ∆F(1, 856)=.384, p=.535, subjective contrast: R²=.003, F(5, 464)=.3047, p=.909, due to interaction ∆R² =.001, ∆F(1, 464)=.488, p=.485.

Moderated regression analyses with major label and second album as controls and category spanning as independent variable were conducted with maximum objective contrast and maximum subjective contrast as moderators for both peak position and weeks in top 50. The full model with maximum objective contrast was significant, F(5, 856)=16.3516, p<.001, but the addition of the interaction was not significant, ∆R²=.001, ∆F(1, 856)=.070, p=.791. This supports the initial operationalization of contrast per album contrast through weighted contrast. With weeks in top 50 as outcome variable and maximum contrast as moderator results were insignificant, F(5, 856)=.585, p<.7122, the addition of the interaction was also insignificant, ∆R²=.000, ∆F(1, 856)=.009, p=.924. With maximum subjective contrast as moderator both regression models with peak position and weeks in top 50 as outcome variable were insignificant.

Explorative Analyses

To see if category spanning on a different level yielded different results multiple moderated regression analyses were conducted, all with second release and major label as control variables. Moderated regression analyses for both peak position and weeks in top 50

(35)

35 were conducted where category spanning on subgenre level was replaced with category spanning on main-genre level. The regression models were insignificant for weeks in top 50. With peak position as outcome variable however, the model with objective contrast as moderator was significant, R²=.0872, F(5, 856)=16.3529, p<.001, but not due to the interaction ∆R² =.003, ∆F(1, 856)=3.120, p=.074. With subjective contrast as moderator the model was significant, R²=.069, F(5, 464)=6.909, p<.001, due to interaction ∆R² =.009, ∆F(1, 464)=4.271, p<.05 (see figure 5). This interaction is as proposed in the hypothesis 2; having high subjective contrast has a positive effect when not spanning categories, having low subjective contrast has a positive effect when spanning categories. Besides this there was also a negative main-effect of category spanning on main-genre level on peak position with second release and major label as covariates, R²=.081, F(3, 861)=25-267, p<.001, where category spanning on main genre level had a ∆R²=.005, ∆F(1, 858)=4.324., p=<.05. It is apparent that category spanning on main-genre level yields different result than category spanning on sub-genre level as can be seen in the main effect and the interaction with subjective contrast.

Figure 6 Interaction between category spanning on main-genre level and subjective contrast

on peak position. Low Category Spanning Average Category Spanning High Category Spanning Peak p o si tion Low Contrast Average Contrast High Contrast

(36)

36 I also examined the differences found between the average objective contrast and subjective contrast per genre. I first centered the contrast scores so they could be compared. This resulted in the Z-score for both objective and subjective contrast. I than sorted them by biggest Z score difference (Table 8). The biggest difference is found in the psychedelic/garage genre. The objective contrast for this genre was however established on the basis of two albums, so the objective contrast measure for this genre is not very reliable. But if we look at the second and third biggest difference scores (Art-Rock/Experimental and Punk/New Wave) we see that those genres, as well as the psychedelic/garage genre, are hybrids. Hybrid genres are genres constructed from two seemingly different genres like punk and new wave that are aggregated with a forward slash. The large difference scores for hybrid genres imply that the use of hybrid genres constructed by allmusic.com do not communicate the intended genre features to the consumers.

Table 8 Objective and subjective contrast scores of the pop/rock main genre per genre

Sub Genre N Objective

Contrast Subjective Contrast Z Objective Contrast Z Subjective Contrast Z Difference Psychedelic/Garage 2 4 5.831 -2.929 -0.009 2.920 Art-Rock/Experimental 23 7.435 3.655 0.589 -1.498 -2.087 Punk/NewWave 12 5.5 6.746 -1.393 0.617 2.010 British Invasion 19 6.842 3 -0.018 -1.946 -1.928 Soft Rock 71 7.155 4.155 0.302 -1.156 -1.458 Alternative Indie Rock 341 7.636 5.063 0.795 -0.534 -1.329 HipHop Urban 70 7.386 7.956 0.538 1.446 0.908 EuroPop 3 6.667 4.521 -0.198 -0.905 -0.708 Country-Pop 38 6.500 6.324 -0.369 0.328 0.697 Pop Rock 219 7.288 5.704 0.438 -0.096 -0.533 Rock Roll 54 7.222 7.099 0.371 0.858 0.488 Dance 71 7.085 6.775 0.230 0.637 0.407

Folk Country Rock 15 6.133 5.282 -0.744 -0.385 0.359

Heavy Metal 117 7.880 7.803 1.045 1.340 0.296

Hard Rock 65 7.615 6.803 0.773 0.656 -0.118

(37)

37 The causal model view states that expertise can be an important determinant for contrast. To examine this the participants were divided based on their expert scores (M=4.812, SD=1.344) in a low expertise group (N=34) and high expertise group (N=39) . Between the two groups there was no significant difference in sex χ2(1)=3.660, p=.056. The high expertise group however had a significant lower age, t(71)=3.065, p<0.01. This however did not affect further analyses. A t-test was conducted to test between contrast scores given for each genre. The only significant results could be found for the difference contrast scores given for British Invasion (equal variances not assumed, F(2, 69)=10.142, p<.01, t(65.579)=-3.466, p<.001), Psychedelic/Garage (equal variances assumed F(2, 69)=.348, p=.557, t(71)=-2.527, p<.001). To see if these genres with high differing contrast scores yield different effects regressions were conducted where only albums that contained genre labels of the genres with the three largest difference scores (Psychedelic/Garage, Art-Rock/Experimental and Punk/New Wave). Multiple moderated regressions were performed with major label and second release as covariates, category spanning as independent variables and objective- or subjective contrast as moderator on peak position. With objective contrast this model was not significant, N=55, R2=.026, F(5, 49)=.258, p=.934. The model with subjective contrast however was marginally significant, N=39, R²=.220,F(5, 33)= 1.857, p=.067. While hypothesis two was not confirmed, these explorative results show that audience perception influences contrast, what in turn can explain album performance. In the next section these and the other results are discussed.

5. Discussion

In this section I will first elaborate on the findings of this research. Thereafter I will discuss limitations and directions for future research. I will conclude with practical implications.

(38)

38

5.1 Discussion of Results

In this paper the influence of category contrast on the relation between category spanning and performance was researched. Thus far there has been almost no research on category spanning set in the music industry. The research that has been done on popular music is scarce and has its limitations concerning the definition of contrast (Van Venrooij & Schmutz, 2013; in press). This study tried to create a contrast measure that could explain the relation between category spanning and performance. By also including the consumers as determining factor for contrast this research has tried to fulfill an unexplored area in the category spanning literature (Vergne & Wry, 2014). Thus far there has been no research in the music industry that incorporated the measure of contrast I used, nor has there been any attention for the perception of genres and their features. Based on category research in other industries and the causal model view I expected that the negative effect of category spanning would increase when contrast is high both for objective and subjective contrast (Hsu, 2006; Negro et al., 2010; Kovacs & Hannan, 2010). Because of previous findings I have split performance into popularity and survival. The only significant result was found testing hypothesis 1a. This result was however in the opposite direction as was expected. Despite these results there were interesting insights discovered also through the exploratory analyses. In the next section I will discuss and explain these findings.

Firstly there was no immediate negative effect found of category spanning on performance when contrast was not taken into consideration. There was however a negative main effect of category spanning on subgenre level when contrast was included in the model. This finding underlines the importance of contrast as a determinant for the effect of category spanning. Besides this there was a negative main-effect of category spanning on performance when testing category spanning on main-genre level. This finding supports the decision to look at genes as a taxonomy with different levels. While this is intuitively true for the music industry (the differences between main-genres and subgenres are often considered [Lena and

Beauty is in the ear of the beholder : the effect of contrast on the relation between category spanning and performance