Dépôt Institutionnel de l’Université libre de Bruxelles / Université libre de Bruxelles Institutional Repository Thèse de doctorat/ PhD Thesis Citation APA:
Ceulemans, C. (2013). Three essays in the economics of music: reputation and success of musicians (Unpublished doctoral dissertation). Université libre de Bruxelles, Faculté Solvay Brussels School of Economics and Management, Bruxelles. Disponible à / Available at permalink : https://dipot.ulb.ac.be/dspace/bitstream/2013/209455/6/36755c7b-91f1-4c05-9ea2-dd5fc9376cca.txt
(English version below) Cette thèse de doctorat a été numérisée par l’Université libre de Bruxelles. L’auteur qui s’opposerait à sa mise en ligne dans DI-fusion est invité à prendre contact avec l’Université ([email protected]). Dans le cas où une version électronique native de la thèse existe, l’Université ne peut garantir que la présente version numérisée soit identique à la version électronique native, ni qu’elle soit la version officielle définitive de la thèse. DI-fusion, le Dépôt Institutionnel de l’Université libre de Bruxelles, recueille la production scientifique de l’Université, mise à disposition en libre accès autant que possible. Les œuvres accessibles dans DI-fusion sont protégées par la législation belge relative aux droits d'auteur et aux droits voisins. Toute personne peut, sans avoir à demander l’autorisation de l’auteur ou de l’ayant-droit, à des fins d’usage privé ou à des fins d’illustration de l’enseignement ou de recherche scientifique, dans la mesure justifiée par le but non lucratif poursuivi, lire, télécharger ou reproduire sur papier ou sur tout autre support, les articles ou des fragments d’autres œuvres, disponibles dans DI-fusion, pour autant que : - Le nom des auteurs, le titre et la référence bibliographique complète soient cités; - L’identifiant unique attribué aux métadonnées dans DI-fusion (permalink) soit indiqué; - Le contenu ne soit pas modifié. L’œuvre ne peut être stockée dans une autre base de données dans le but d’y donner accès ; l’identifiant unique (permalink) indiqué ci-dessus doit toujours être utilisé pour donner accès à l’œuvre. Toute autre utilisation non mentionnée ci-dessus nécessite l’autorisation de l’auteur de l’œuvre ou de l’ayant droit. ------English Version ------This Ph.D. thesis has been digitized by Université libre de Bruxelles. The author who would disagree on its online availability in DI-fusion is invited to contact the University ([email protected]). If a native electronic version of the thesis exists, the University can guarantee neither that the present digitized version is identical to the native electronic version, nor that it is the definitive official version of the thesis. DI-fusion is the Institutional Repository of Université libre de Bruxelles; it collects the research output of the University, available on open access as much as possible. The works included in DI-fusion are protected by the Belgian legislation relating to authors’ rights and neighbouring rights. Any user may, without prior permission from the authors or copyright owners, for private usage or for educational or scientific research purposes, to the extent justified by the non-profit activity, read, download or reproduce on paper or on any other media, the articles or fragments of other works, available in DI-fusion, provided: - The authors, title and full bibliographic details are credited in any copy; - The unique identifier (permalink) for the original metadata page in DI-fusion is indicated; - The content is not changed in any way. It is not permitted to store the work in another database in order to provide access to it; the unique identifier (permalink) indicated above must always be used to provide access to the work. Any other use not mentioned above requires the authors’ or copyright owners’ permission.
Université libre de Bruxelles
FACULTÉ SOLVAY BRUSSELS SCHOOL OF ECONOMICS AND MANAGEMENT
Three Essays in the Econonnics of Music: Réputation and Success of Musicians
Thèse de Doctorat présentée en vue de l'obtention du titre de Docteur en Sciences Economiques et de Gestion
Par Cédric Ceulemans
Directeur: Professeur Patrick Legros - Université libre de Bruxelles Co-directeur: Professeur Victor Ginsburgh - Université libre de Bruxelles Membres du jury: Professeur Antonio Estache - Université libre de Bruxelles Professeur Kathryn Graddy - Brandeis University Professeur Sébastien Van Bellegem - Université catholique Université libre de Bruxelles
FACULTÉ SOLVAY BRUSSELS SCHOOL OF ECONOMICS AND MANAGEMENT
Three Essays in the Economies of Music: Réputation and Success of Musicians
Thèse de Doctorat présentée en vue de l'obtention du titre de Docteur en Sciences Economiques et de Gestion
Par Cédric Ceulemans
Directeur: Professeur Patrick Legros - Université libre de Bruxelles Co-directeur: Professeur Victor Ginsburgh - Université libre de Bruxelles Membres du jury: Professeur Antonio Estache - Université libre de Bruxelles Professeur Kathryn Graddy- Brandeis University Professeur Sébastien Van Bellegem - Université catholique de Louvain
Université Libre de Bruxelles
00; 53^640
Académie Universitaire Wallonie-Bruxelles Acknowledgments
First and Foremost I want to thank my advisors Prof. Victor Ginsburgh and Prof. Patrick Legros. Without their support I would not hâve been able to start, pursue, or finish my thesis. Victor is an amazing advisor, a great scholar, and a passionate intellectual. The scope of his knowledge and interests are so vast that it could hâve discouraged me. Instead, his appetite for knowledge revealed to be contagious and a constant source of inspiration. My ambition for the future is to match Victor’s willingness to learn and ability to keep moving forward.^ For the motives mentioned above, for his kindheartedness and for many others reasons, Victor’s selfless time and encouragements were sometimes ail that kept me going. I was fortunate to be aiso advised by Patrick who let me work on my own when needed but whose door was aiways open when I had questions or needed feedback. As a co-author, I had the opportunity to witness and learn from his great intellectual creativity and his Sharp mind. An unforgettable lesson and expérience I hope to repeat in the future. Above his or her inputs in my thesis, each member of the jury contributed to my professional growth in one way or the other. I am sincerely thankfui to Prof. Antonio Estache, Prof. Kathryn Graddy, and Prof. Sébastien Van Bellegem for their extra time and considération. I thank Prof. Marjorie Gassner. She is an exemplary teacher and, besides her gigantic work for the university, she still takes the time to support and advice teaching assistants. From my “T.A.” period, I want to thank Prof. Abdul Noury and my “Micro” colleagues: Nastassia, Nicky, and Mohamed. I enjoyed working and laughing with you.
^ I know it will be hard for me to match Victor in those areas. However, the certainty of failure should not be an excuse not to try. “Success is not final, failure is not fatal: it is the courage to continue that counts.” (Winston Churchill) From my “DEA” period, I want to thank Claude Adan, who warmiy welcomed me at EGARES and has been so helpfui ever since, as well as my “companions in crime": Sadibou and Juan Manuel. Finally, I thank my closest friends, Jeanine D.M., my parents-in-law and my family. Very spécial thanks are due to my parents and “big” brothers. Last but not least, I am infinitely thankfui to my wife, Julie, and our little “pumpkin” Estelle.^
^ Estelle probably does not care about my thanks right now. My guess is that she is much more preoccupied by the following metaphysical question: “Will I eat strawberries or raspberries for desert tonight?”
2 Table of Contents Introduction 4 Chapter 1 : Rock Bands: Matching, Recording & Work Organization 12 1. Introduction 13 2. Rock Mu sic: Records and Musicians 14 2.1. Records 14 2.2. Musicians: song-makers, entertainers and “rentiers” 20 3. Data 24 4. Behavior and Work Organization of Rock Bands 25 4.1. Bands 26 4.2. Solo artists 32 4.3. Producers 34 5. Rock and Roll Bands, (In)complete Contracts and Creativity* 35 5.1. Specifying matchings 37 5.2. Data 41 5.3. Results and conclusions 41 6. Conclusion 43 Chapter 2: Musical Characteristics and Success in Popular Music 46 1. Introduction 47 2. Charts 49 3. Data 50 3.1. Choice of songs included in the database 51 3.2. Measures of success 54 3.3. Musical characteristics and control variables 55 4. Results 59 4.1. Commercial success 60 4.2. Critics’ and music lovers’ success 65 5. Discussion and Conclusion 67 Chapter 3: The Formation of the Canon of Baroque Music** 70 1. Introduction 71 2. Method and data 72 3. Results 74 3.1. Does a consensus exist between musicologists? 74 3.2. Persistence of composers’ réputations 78 3.3. Do évaluations follow an autoregressive process? 82 3.4. National bias and composer spécifie effects 87 4. Discussion and Conclusions 88
* Section 5 of Chapter 1 was published as C. Ceulemans, V. Ginsburgh and P. Legros (2011), "Rock and Roll bands; (In)complete contracts and creativity". American Economie Review, Papers and Proceedings 101(3), 223-242.
** A slightly different version of Chapter 3 was published as C. Ceulemans (2010), "The réputation of Baroque composers 1790-2000". Empirical Studies of the Arts 28(2), 223-242.
3 Introduction The odds of success in the arts are and hâve aiways been extremely low. In his “Lettre sur le commerce des livres”, Diderot (1763) stated that “one out of ten books is a success, four will sell well enough to recoup the costs in the long run and the remaining five books are produced at a loss.”^ The probability of success is even lower in the contemporary music industry. Roughiy 80 percent of the albums released fail to cover their costs (Caves, 2000, p.61). Out of 97,751 albums released in 2009 only two percent sold 5,000 copies or more and only 12 albums reached one million units (Peoples, 2010). The disparity in success finds an écho in terms of earnings. The top one percent musicians took in 56% of total concert revenues in 2003 (Connolly and Krueger, 2006). These figures indicate that the arts and entertainment industry is a market of superstars, that is a market where a relatively small number of people earn enormous amounts of money and dominate the activity in which they engage (Rosen,1981). Scholars hâve long shown an interest in understanding what makes a work of art or an artist a superstar. According to Diderot (1763), the probability of success is fixed and thus impossible to increase or decrease. Even if a book is excellent, success dépends on “an infinity of reasonable or odd circumstances” which makes it hardiy impossible to predict what work will be successful. In other words success is a lottery. Some features, like the author's réputation, might limit the risk of making a loss, however.^ More than two centuries after Diderot's letter, Rosen (1981) laid down the first theoretical framework that explains the existence of superstars. He developed a model where artists differ in quality in the eyes of the public. For consumera, a low quality artist is a poor substitute for a highiy talented one. Consumera prefer to consume one highiy talented artist rather than many low quality ones. Hence, every consumer wants to consume the “beat.” In situations in which art is nearly perfectiy reproducible, the “beat” artist becomes a monopolist who chooses to either fix a high
« Ajoutez que, de compte fait, sur dix entreprises, il y en a une, et c'est beaucoup, qui réussit, 2quatre dont on recouvre ses frais à la longue, et cinq où l'on reste en perte. » (Diderot, 1763, p. 41) « Quel rapport y a-t-il, s'il vous plaît, entre son espérance et ses risques ? Voulez-vous connaître précisément la valeur de sa chance ? Elle est comme le nombre de livres qui durent, au nombre de livres qui tombent, on ne peut ni la diminuer ni l'accroître ; c'est un jeu de hasard, si l'on en excepte les cas où la réputation de l'auteur, la singularité de la matière, la hardiesse ou la nouveauté, la prévention, la curiosité, assurent au commerçant au moins le retour de sa mise. » (Diderot, 1763, p. 41)
5 price and serve a small number of consumers or a low price and serve the entire market. The recording industry is a good example of the superstar model. The art is perfectiy reproducible (CDs or MP3s perfectiy reproduce the music recorded in the studio) and singers are endowed with different amounts of talent. The “best” artist can serve the entire market at a low price, since ail new releases are sold at almost similar prices, regardiess of the musicians' quality. In theory, the “best” artist should capture the entire market and earn an enormous amount of money. Artists who are just a little less talented will earn much less (if anything). The model can be summarized as follows: “small différences in talent will translate in large différences in earnings.” According to Adler (1985), talent is not the cause of the existence of superstars. Many artists hâve the required qualities to achieve celebhty; what makes a superstar is “the need on the part of consumers to consume the same art that others do” (Adler, 2006). This need arises form the fact that consomption of art is a dynamic process. The utility that consumers get from an artist increases with their knowledge of the artist; the more you know the more you enjoy. This knowledge can be acquired from the exposure to the art itself, discussions with friends and relatives, and gathering information from the media. The more popular the artist is, the easier it is be to be knowledgeable about him (popular artists attract médias' attention for example). Adler's model implicitly underlines that art consomption is an expérience. The consomption of the art itself is only a part of the expérience. To be able to share it with others is equally important. Because the probability of “sharing” is higher, consumers want to consume popular artists. This consomption behavior is similar to the bandwagon effect which is defined as “the case where an individual will demand more (less) of a commodity at a given price because some or ail other individuals in the market aiso demand more (less) of the commodity” (Leibenstein, 1950, p. 190). In the arts, the bandwagon effect produces superstars by making popular artists increasingly more popular. Because the number of successfui artists is limited at any given time, not ail artists can become superstars. Who will emerge from the large pool of equally talented artists? The lucky ones! Consumers choose randomly the new artist they add to their consomption bundle. The selected artist therefore has by pure luck an initial advantage of being
6 (slightiy) more popular than her competitors. Thanks to the bandwagon effect, this initial advantage will make her the new superstar. A different explanation to Alder's theory of the emergence of stars exists. McDonald (1988) develops a dynamic model where emerging and established artists are présent on the market of the performing arts. AU artists can perform a bad or a good show. However, artists differ in talent; the more talented the artist the higher the probability of producing a good show. Emerging artists do not know whether they are talented or not. Therefore they ail take their chance and enter the market hoping that the public will recognize their talent. On its part, the public believes that the probability of attending a good show is higher for established artists. This belief arises because established artists hâve a good track record while emerging artists do not. Attending one of the emerging artist's show is therefore more Yisky'. Established stars hâve thus an advantage over emerging ones. They can charge the same ticket price and attract a larger audience, or charge a higher price and attract a similar audience. Hence, stars' revenues will be much higher than those of emerging artists. The income différence will increase with the audience's risk aversion. If consumers are highiy risk averse, they will hâve a strong preference for established artists and will pay a high price to attend their performances. Emerging artists can access stardom, however. Only the most talented artists - i.e. those who produce consistently good shows - will become stars. In that sense, McDonald's and Rosen's model are similar: Différences in talent is the main feature explaining an artist's success. Théories on the économies of superstars hâve been tested empirically. The most difficult issue concerns the quantification of talent (how to measure talent?). Hamien (1991, 1994) analyzes the popular music market and uses the quality of the voice as a proxy for talent. He finds that record sales increase less than proportionally with talent and concludes that his results contradict Rosen’s theory. Schuize (2003) argues that the quality of the voice might not be the relevant measure for artistic quality in popular music. Factors such as the quality of the show on stage, sex appeal, or the contents of the lyrics are aiso important; “Hamien’s approach suffers from an omitted variable bias” (Schuize, 2003, p. 434).
7 Chung and Cox (1994) aiso use the popular music market to test Rosen’s and Adler’s models but their approach is different. They show that the distribution of gold records^ among artists follows a Yule distribution. In the présent case, this stochastic model can be described as the following consumption pattern: Consumers buy a CD in sequential order. Each consumer buys a spécifie CD with a probability that increases with the number of previous sales of the CD. Moreover, the probability that a consumer chooses a new CD is constant and close to zéro. The fact that the Yule distribution is a good fit of the distribution of gold records supports Adler’s theory that a few lucky artists will benefit from the bandwagon effect. However, Chung and Cox’s results do not hold with a different measure of success. Instead of gold records, Giles (2006) uses either the lifetime or quantity of number one hits as success measures. In that case, the “Yule version” of superstardom is rejected. The results “leave open the possibility that Rosen’s (1981) explanation of superstardom (i.e., small différences in ability are translated into large différences in success) is relevant to this industry” (Giles, 2006, p.73). Théories on the superstars phenomenon suggest thus that luck and/or talent are the driving forces behind success. The controversial results of the empirical studies show that talent and luck are vague concepts. Their importance is therefore hard to measure statistically and might be overestimated as suggested by the two following quotes: “I can recognize talent but I don’t know what to do with it sometimes. [...SuccessfuI artists] might not be the best songwriters, they might not be the best players, they might not even be the best performers” (Bruce Allen, manager of pop stars such as Bryan Adams and Michael Bublé). “In the music business, people get lucky, but I don’t see the lucky ones hanging around for a long time. You can get lucky once. You don’t get lucky over and over and over again” (Peter Patemo, attorney of Metallica and Dr. Dre). If talent and luck are not sufficient conditions to make a career in the music business, what eise might détermine artists’ success? The “superstars models” left performers with no “active” rôle: successfuI artists are either endowed with an innate talent far above the average or are extremely lucky. However, ail musicians (talented
^ The Recording Industry Association of America awards the Gold plaque to records selling 500,000 copies or more in the U.S.A.
8 or not; lucky or not) take continuousiy decisions that affect their career. Chapter 1 and 2 of this dissertation analyze in details some of these decisions and their influence on success. Chapter 1, Rock Bands: Matching, Recording & Work Organization, ^ investigates the impact of partnerships, matching, and work organization on the success of rock musicians using a unique database of 1,494 albums released between 1970 and 2004. We show that rock bands differ in their work organization because the agreements between the members of band are different. These agreements can be seen as implicit contracts. Drawing on this observation, we develop a model where agents (musicians) with different levels of creativity match (to form a band) and produce a joint output (a song). We show that the way agents match (positively or negatively) is correlated with success and dépends on the (in)completeness of contracts. The theoretical results are supported by the data. Chapter 2, Musical Characteristics and Success in Commercial Music, analyzes the relationship between musical characteristics, that can objectively be measured, and different types of success (commercial success, critical success, and success assessed by music lovers). We show that the strength and the direction (positive or négative) of the relationship between success and musical characteristics vary with the measure of success. Though, Hendricks and Sorensen (2009), MacDonald (1988) and Diderot (1763) hâve shown that réputation and success are related, the third chapter goes in a slightiy different direction than the two others as it deals with long term réputation of composers rather than commercial success of pop-rock musicians. Chapter 3, The Formation of the Canons of the Baroque Music,^ analyzes the réputation of baroque composers over time. The dataset makes it possible to describe the évolution of composers’ réputation and of the baroque canon. The entries in seven important musical dictionaries written between 1790 and 2000 are used to measure réputations. We provide evidence that a consensus exists between musicologists, who often rely on their predecessors’ work.
The results presented in Section 5 of this chapter hâve been published in “Rock and Roll Bands, (In)complete Contracts and Creativity” (Ceulemans et al., 2011) ® A slightiy different version of this chapter has been published under the title “The Réputation of Baroque Composers 1790-2000” (Ceulemans, 2010)
9 References
Allen, B., What makes an artists successfui?, http://www.artistshouse.org.
Adler, M. (1985). Stardom and talent. American Economie Review, 75, 208-211.
Adler, M. (2006). Stardom and talent, in Handbook of Economies of Art and Cuiture, edited by V. Ginsburgh and D. Throsby, Amsterdam and New York: EIsevier, 895- 906.
Caves, R. (2000). Creative Industries: Contracts Between Art and Commerce. Cambridge and London: Harvard University Press.
Chung, K., Cox, R. (1994). A Stochastic model of superstardom: an application of the Yule distribution. The Review of Economies and Statistics, 76, 77^-775.
Ceulemans, C. (2010). The réputation of Baroque Composera 1790 - 2000, Empirical Studies ofthe Arts, 28(2), 223-242.
Ceulemans, C., Ginsburgh, V., Legros, P. (2011). Rock and Roll Bands: (In)Complete Contracts and Creativity. Amercian Economie Review, Papers and Proceedings, 101(3), 217-221.
Connolly, M. and A. Krueger (2006). Rockonomics: the économies of popular music, in Handbook of the Economies of Art and Culture, edited by V. Ginsburgh and D. Throsby, Amsterdam and New York: EIsevier, 667-719.
Diderot, D. (1763). Lettre sur le commerce des livres. Texte annoté par Christophe Paiilard, Université du Québec à Chicoutimi, http://classiques.uqac.ca/classiques/Diderot_denis/lettre_commerce_livre/lettre_com _livres.pdf.
Giles, D. (2006). Superstardom in the US popular music industry revisited. Economies Letters, 92, 68-74.
Hamien, W. (1991). Superstardom in popular music: empirical evidence. The Review of Economies and Statistics, 73, 729-733.
Hamien, W. (1994), Variety and superstardom in popular music. Economie Inquiry, 32, 395-406.
Hendricks, K., Sorensen, A. (2009). Information and the skewness of music sales. Journal ofPolitical Economy, 117, 324-369.
Leibenstein, H., Bandwagon, Snob and Veblen effects in the theory of consumer demand, The Quarterly Journal of Economies, 64, 183-207.
MacDonald, G., (1988). The économies of rising stars. American Economie Review, 78, 155-161.
10 Paterno, P., Seing a successfui artist, http://www.artisthouse.org.
Peoples, G. (2010). Analysis ; Important Sales Trends You Need To Know, billboard.biz, http://www.billboard.biz/bbbiz/content_display/industry/news/ e3i4ad94ea6265fac02d4c813c0b6a93ca2.
Rosen, S. (1981). The économies of superstars, American Economie Review, 71, 845-858.
Schuize, G.G. (2003). Superstars, in Towse, R. (Ed.), Handbook of Cuitural Economies, Cheltenham: Edward Elgar, 431-436.
11 Chapter 1
Rock Bands: Matching, Recording & Work Organization
12 1. Introduction
According to Serge Gainsbourg, French singer and songwriter, “La chanson est un art mineur contrairement à la musique classique qui demande une initiation.”^ This quote suggests that a very limited knowledge is required to write and produce popular music and implies that musical awareness is not a must for listeners to enjoy this type of music. Gainsbourg’s daim may look strong, but it is not far from being true. The popular music market is characterized by a very large number of producers (musicians) and consumers. Producers include professional and non-professional artists. Indeed, many rock musicians take non-musical jobs to make ends meet. This might be the case even if the musicien has a record deal with a label. For example, Annie Lennox was a waitress in a restaurant at the very beginning of her career with her band The Catch before becoming a world star with Eurythmies. The case of Brian May is certainly less cliché. In the late ‘60s, while recording an album with a band called Smile, he started a PhD in astrophysics at Impérial College, London. In the early ‘70s, he stopped pursuing his PhD to focus on his new musical project, Queen, a very risky decision that happened to be a good one, as Queen became extremely famous. These examples illustrate that successfui pop/rock musicians can hâve very different backgrounds and that an advanced musical éducation is not mandatory. On the contrary, in classical music, musicians hâve to follow a long educational process. A degree is, of course, not a guarantee of success but is a necessary component. For example, musicians are required to hâve a musical background and training to play in an orchestra. The director Controls the overall quality of the orchestra as well as of the work organization, and the conductor is in charge of the créative input and the musician’s motivation. In popular music, labels, which are artists’ employers, are not involved in their organization, nor do they select musicians based on their musical training or quality. Rather, they choose artists based on their commercial potential. In other words, anybody could enter the market as long as a record label feels they could be
^ “Popular song is a minor art, contrarily to classical music which requires to be initiated.” Serge Gainsbourg, in “Apostrophe” (French télévision show), October 1986.
13 profitable. Hence, it is up to artists themselves to select collaborators (band members, producers, lyricists, composers, etc.) and to define their organization. Cameron and Collins (1997) were the first economists to focus on the internai structure and arrangement of rock groups. Drawing on a limited sample of bands, Cameron and Collins describe how informai contracts among members explain some characteristics of the behavior of such groups (partnership durability, opportunistic appropriation of band name,^ governance structure inside the band, etc.). In this chapter, we investigate the impact of partnerships, matching, and work organization on the success of rock musicians using a unique database of 1,494 albums released between 1970 and 2004. Section 2 summarizes the évolution of rock and roll, the recording industry, and musicians’ work and earnings. We point out that rock music and the recording industry are bound together. We aiso suggest that although album sales hâve decreased since 2000, authorship and royalties are still an important issue for bands. Section 3 describes the data. Drawing on a subsample of our database, Section 4 investigates in detail the internai organization of bands and solos artists. In Section 5, we develop a model where agents (musicians) with different levels of creativity match (to form a band) and produce a joint output (a song). We show that the way agents match (positively or negatively) is correlated with success and dépends on the (in)completeness of contracts. The theoretical results are supported by the data. Finally, Section 6 concludes.
2. Rock Music: Records and Musicians
2.1. Records
What is rock music, and when did it start? Like any other musical genre, the birth of rock and roll is impossible to date with précision. “Rock 'n' roll was an evolutionary process. To name any one record as the first would make any of us look a fool.”^
^Band name like Rolling Stones, or Aerosmith hâve an économie value. Hence Cameron and Collins (1997) “expect opportunistic individuals to appropriate band name and use it to market bands which bear little relationship to the original membership.” ^ Billy Vera, Foreword to What Was the First Rock ’n’Roll Record by Jim Dawson and Steve Propes (1992).
14 However, 1954 is accepted by every expert to be a milestone in the history of rock. That year, Bill Haley and His Cornets recorded “Rock Around the Clock,” which is the first rock and roll song to chart at Number 1. The same year, Elvis Presley, the icon of the genre, recorded his first single “That's AN Right”. It is easier to define what rock and roll was in the beginning: “a type of popular dance music originating in the 1950s, characterized by a heavy beat and simple mélodies. Rock and roll was an amalgam of black rhythm and blues and white country music, usually based on a 12-bar structure and an instrumentation of guitar, bass, and drums.”"* Rock and roll music was rapidiy renamed as rock music to include the many new styles that emerged in the ‘60s. Today, the term encompasses so many subgenres and forms of music that any attempt to define it would be vain. Indeed, rock music is very vaguely defined as “a form of popular music which evolved from rock and roll and pop music during the mid and late 1960s. Harsher and often self-consciousiy more serious than its predecessors, it was initially characterized by musical expérimentation and drug-related or anti-establishments lyrics.”® Music critics aiways needed to find a new stylistic dénomination for new artists' styles, from psychedelic rock in the ‘60s to electro-pop in the recent years. By the ‘70s, the rock music genre included very different types of music. For example, what do the progressive rock of Genesis and punk rock of the Sex Pistols hâve in common? On the one hand, Genesis, which used complex rhythms, instrumentation, and song structure, was influenced by classical music and was composed of highiy skilled musicians, composing songs that lasted 10 minutes or longer. On the other hand, the Sex Pistols were poorly skilled musicians,® played short, aggressive songs with stripped-down instrumentation, and deliberately wanted to do and represent exactiy the opposite of progressive rock. Both Genesis and Sex Pistols are categorized as “rock music,” but what do they hâve in common except their influential rôle on popular music? Maybe the answer is to be found in the statement made by Michael Lydon, who claimed that “rock was never an art form
'' Oxford Dictionaries Online; http://oxforddictionaries.com. ® Oxford Dictionaries Online: http://oxforddictionaries.com. ® “When Sid joined [the band] he couidn't play guitar but his craziness fit into the structure of the band” John McLaren quoted by Charlotte Robinson (2001).
15 that just happened to make money nor a commerdal undertaking that sometimes became art. Its art was synonymous with its business.”^ Punk and progressive rock can be seen as art forms for different reasons: progressive rock for its musical value and punk rock because of its political and anti establishment lyrics. In both cases, to exist, it had to be commercially profitable for the recording industry. The link between rock music and labels is a two-sided story. We will first argue that rock is “recorded music” that developed thanks to the recording industry, and we will then show through a brief history of the recording industry that the growth of the music market during the second half of the 20‘^ century was boosted by rock music.
Rock Music: A Recording Tradition
Before rock and roll, “western” music was transmitted in two different ways: by musical scores or from one person to another. Folk, jazz, and blues belong to the oral tradition. Musicians learn songs by hearing them played by other musicians. Each génération of musicians interprets those standards. On the contrary, classical music and its “dérivatives” (Broadway musicals, for example) belong to the written tradition. Music is transmitted through scores written by composers. With rock, a new way to transmit music was born: records. Elvis Presley, nicknamed the King of Rock ‘n’ Roll, never toured in Europe. “The King” played only three gigs outside the U.S., ail of them in Canada. Though Presley sold sheet music, it was a spin-off product rather than a “transmission tool” for his music; he owed his famé to his records. He influenced générations of European musicians who never saw him perform live and never played his music by reading a score. The impact of records on musicians was so important that the génération following Presley thought more in terms of records than concerts. The Beatles even stopped touring to focus on the production of records. Kania (2006) summarizes this characterization by saying, “Rock music is primarily a recording tradition, and its live performances dépend partiy on that tradition for their value. Thus, live rock performances, while undeniably an important
Quoted by Garofalo (1999)
16 part of the rock world, are not the primary focus of critical attention in that tradition.” Rock is influenced by music belonging to oral and written traditions, and the way rock is transmitted is a mix of both traditions. The “recording tradition" uses a medium similar to the written tradition; it is no longer a sheet of music but an LP or a CD (or any other format), and this medium delivers sound like in the oral tradition. Musicians learn music by listening to it, like in the oral tradition, not from a spécifie musicien in “live circumstances” but by the intermediary of a record, which is the same for everyone around the world. To record the music and produce the medium, advanced technology is needed. Hence, rock musicians hâve to work with “for profit” companies that can produce quality records in quantity. Since the very beginning, rock music has been bound to the technology and business of the recording industry.
Rock and Recording industry: A Brief History^
The éruption of rock and roll not only changed the musical tradition permanently but aiso propelled the music industry to a new level of profits. Revenues from record sales (ail musical genres included) climbed from $213 million in 1954 to $603 million in 1959. This development was caused by three factors. First, the économie growth between 1951 and 1970 was so spectacular that it is now often referred to as the “Golden Age.” The music industry clearly benefited from this favorable économie situation. Second, the demand for rock and roll music exploded, partiy due to the first cohort of baby boomers who were, in the late ‘50s, listening to teenage music. Between 1955 and 1959, rock and roll jumped from a 16% share of the pop market to 43%. Third, Columbia and RCA, two major recording companies, were innovative on the technological front. In 1931, RCA invented the long playing (LP) format. As the name suggested, LPs (aIso named vinyls) could contain more music than the previous formats and represented a big improvement in sound quality. Because of the Great Dépréssion and the huge popularity of radio, vinyls did not enjoy commercial success until the ‘50s when Columbia (re)invented the long playing records and RCA the so-called “45” format. For the first two years, there was an
® Uniess otherwise stated, figures in this subsection are taken from Garofalo (1999).
17 uncertainty about the “winning” format. Finally, Columbia won the battle for albums, and the RCA 45 format was used for “singles.” At the end of the ‘50s, vinyl became a mass product and stayed the leading music format until the mid-‘80s. Although some major music companies were at the forefront of the technological changes, they ail missed the rock and roll musical révolution. Majors (the biggest music recording companies) were slow in integrating rock music into their catalogs. As a conséquence, the number of independent labels sharpiy rose from a 22% share of the pop market in 1955 to 66% in 1959. The décliné of the majors can aiso be observed in weekiy top ten U.S. Charts. In 1949, the four majors produced 90% of the top ten songs. Ten years later, this share had decreased to 34%.^ Since the 1960s, rock/pop music and ail the subgenres hâve led ail activity in the industry. The domination of rock increased throughout the years. Today, pop and rock music account for 55% of global retail sales, followed by classical and jazz music at 5% and 2.5%, respectively. Throughout the ‘60s many labels were created, providing the opportunity for many musicians to enter the business and hence for consumers to increase the probability of finding a suitable “product." The Creative environment in the rock music industry had a positive impact on the number of records sold, which steadily increased until 1978. During the second half of the ‘70s, we observed a merger mania and a décliné of independent labels. Garofalo (1999) describes the new rôle of independent labels: “While the indies [independent labels] may still hâve entered the business to fill a void in the market, their larger fonction became providing research and development for the majors. When Robert Stigwood’s RSO label and Neil Bogart’s Casablanca demonstrated from the bottom up that a fortune could be made in disco, PolyGram (Philips) simpiy stepped in and acquired a controlling interest in both labels... Far from being in compétition with the majors, the independents had now become part of the same corporate web.”
® Peter Tschmuck (2009) 2010 Figures. Source; Music & copyright, http://musicandcopyright.wordpress.eom/2011/06/01/
18 At the end of the ‘70s, the industry, as well as the whole economy, went through a recession. The sales figures from 1978 to 1982 decreased by 40%.^^ The IFPI (International Fédération of the Phonographic Industry) argued that the décliné in industry revenue during the recession was directiy related to the rising sales of cassette tape recorders and blank tapes. The campaign against home taping was symbolized by the slogan of the British Phonographic Industry (BPI): “Home Taping is Killing Music.” Among other bands, the Dead Kennedys did not agréé with the anti home taping campaign and parodied the slogan in their 1981 record In God We Trust, Inc. On the original cassette version, Side A contained ail songs, with Side B left blank. On the cassette, buyers could read, “Home taping is killing record industry profits! We left this side blank so you can help.” Although home taping did not stop, the industry went through a new growth phase from 1983 through 1999 thanks to two technological shocks: the Walkman (a portable music cassette player) and the compact dise. Pre-recorded music cassettes became popular in the early ‘80s thanks to the commercialization of the Walkman. Consumers were ready to sacrifice sound quality (vinyl has a much better sound quality than cassettes) to gain mobility (the Walkman allows people to listen to music at school, in the métro, while jogging, etc.). In 1981, the audio compact dise was launched on the market. As usual, it took time before the new medium was adopted by a large community of consumers. By 1992 it was the most popular medium and marked the beginning of a very profitable period for record companies. The sales increase was largely due to the reissue of old albums. This is what the industry calls the “back catalog.” Consumers bought albums they aiready had on LP and/or music cassettes for the second or third time. By the early ‘90s, “catalog sales were estimated at 40% of ail album sales, making the back catalog, for many top-selling artists, their most valuable asset” (Garofalo, 1999, p.344). Back catalogs are aiso very profitable for labels. According to Vogel (2010, p.270), the catalog can often account for half of the revenue and three-fourths of the profit for a major label. Sales rose consistently during the ‘90s to achieve an all-time high in 1999. Since then, the industry has been in crisis. The computer and MP3 révolution of the 2000s hit the industry hard. Labels hâve not yet found a winning business model to return to previous profits and growth rates.
Peter Tschmuck (2009)
19 This brief history shows that rock music is the best ally of the recording industry. From the mid-‘50s to the mid-70s, the creativity of the rock scene encountered huge success. Since then, the recording industry has capitalized on several stars. In the ‘80s, they heavily promoted new work from stars of the ‘70s (Michael Jackson, Bruce Springsteen, Phil Collins, etc.) and newer acts (Madonna, Whitney Houston, etc.) to make superstar albums in the line of Thriller.^^ In the ‘90s, labels made profits with the back catalog through old albums or new compilation albums of stars from the 1950s to the 1980s, from Presley to Guns N’ Roses. Still today, rock music is a label’s best asset.
2.2. Musicians: song-makers, entertainers and “rentiers”
The many changes encountered in the music industry dictated the évolution of musicians’ work and their sources of revenue. Until the early ‘60s, rock and roll was a market for singles rather than albums. Elvis Presley released 17 Number 1 singles between 1959 and 1962. Bill Haley and His Cornets are known for hit singles like “Rock Around the Clock” and “Shake, Rattle and Roll,” but not for albums. In the mid-‘60s we observe a shift from a culture of “singles” to a culture of albums. Rock and roll musicians were proponents of singles, whereas the new générations of rock musicians wanted to be more involved in the créative process and make the album a work of art rather than a compilation of singles. “Pet Sounds” (1966) by the Beach Boys and “Sgt. Pepper’s Lonely Hearts Club Band” (1967) by The Beatles are two albums that greatly influenced and accelerated that shift. In the 1970s, bands like Led Zeppelin and Pink Floyd released only one or two singles per album (which usually did not chart high) but sold millions of LPs. With the birth of MTV in 1981, the power of video clips greatly increased. Because of the cost, labels only made videos for singles. However, sales of singles consistently decreased from 228 million units in 1973 to 108 million in 1993 and a mere 12 million in 2003. Instead, singles became a significant promotional tool for albums and emphasized the rôle of Visual content. Michael Jackson perfectiy
“Thriller” is an album by Michael Jackson that was released in 1982. It is considered as the worldwide bestselling album of ail times.
20 integrated this element for the album “Thriller.” The video for the song “Thriller” was a 14-minute short movie that set new standards for production. Video clips for “Billie Jean” and “Beat It” aiso greatly boosted album sales. Most rock artists followed the trend and wanted to impress audiences, as evidenced by Dire Straits using computer animation in the clip of “Money for Nothing” (1985) or Peter Gabriel’s “SIedgehammer” video (1986) use of spécial effects. Rock artists hâve aiways paid attention to visuals, especially when performing on stage. Famous examples are the leg movements of Elvis Presley, theatrical performances of David Bowie playing “Ziggy Stardust” in the early 1970s, or the epic Project of “The Wall” by Pink Floyd in 1980-1981. Those examples aIso show the trend of live performances shifting from concerts in the early 1960s where you could barely hear the music if the audience screamed too loud to mega-shows in the 1980s with light shows, visuals, and powerfui sound. Rock artists are song-makers and performers and, since the ‘80s, aiso entertainers.^^ Until the recording industry crisis in the 2000s, musicians earned money by selling records, mainly albums. The digital révolution changed this situation rapidiy and radically. Album sales decreased drastically (approximately 1 billion units in 2000 in comparison with 310 million in 2010). In contrast, singles had never sold so well: 12 million units in 2003, 146 million units in 2004, and more than 1 billion units in 2010.^'' Rock music is again a singles culture like in the 1950s. This cultural shift had a négative impact on artists’ royalties. However, bands produce not only records but aiso concerts. The two are complementary. As long as greater attendance raised record sales, bands kept concert prices relatively low. This was the case until 1999. To compensate for the décliné in income from records, artists increased the prices of concerts (Krueger, 2005). The shift of culture in rock music generates a shift in artists’ revenue.
Queen were well aware of their rôle as entertainers. Freddie Mercury composed a song called “Let Me Entertain You” in 1978. Lyrics are directiy addressed to the audience and are self-deprecating and somewhat cynical about concert tours of rock bands. The huge increase from 2003 and 2004 was due to the création of the iTunes store in April 2003,the first store where consumers could legally and relatively easily download digital singles and albums. RIAA computes digital sales since 2004.
21 Table 1. Artists’ Ranking Based on U.S. Earnings in 2010^^ Total Earnings Tour Earnings Rank Artist (in millions of (in millions of dollars) dollars) 1 Lady Gaga 30.6 23.8 2 Bon Jovi 30.4 28.3 3 Roger Waters 24.5 23.9 4 Dave Matthews Band 23.9 23.3 5 Justin Bieber 22,4 13.7 6 Taylor Swift 20,7 10.4 7 Michael Bublé 19.8 16.6 8 Eagles 18.3 17.3 9 The Black Eyed Peas 16.8 11,6 10 Paul McCartney 14.2 12.9
Table 1 suggests that nowadays superstars’ earnings corne from concerts. In the following sections we study the rôle of bands’ work organization that relies heavily on royalties for those who compose or write (“writing” royalties). Are writing royalties a relevant issue for today’s musiciens? We daim that it still matters for the same reasons as before: money and art.
Money
Rock musiciens' revenue cornes from three different sources: records (digital, physical, DVDs, etc.), live performances (mainly concerts), and merchandising (sheet music, t-shirts, posters, commercials, etc.). In this category, songwriters will earn more money from writing royalties on records, sheet music, commercials, and tours. Indeed, when a band performs its own songs it has to pay writing royalties to a “performance rights organization.” The calculation of the payment differs among countries, but writing royalties are often based on ticket price and attendance and
Source: Billboard. Revenue from merchandise sales, [...], and songwriter performance royalties from terrestrial radio play, DVDs and ringtones are not included in this calculation. “Why? There just isn’t enough of that kind of data available across the whole board” (Bilboard.com, “Top 40 Money Makers"). Including this revenue might decrease the différence between concert income and other sources of income.
22 therefore on concert revenue. Eventually, this money will go back into the pocket of the songwriter(s). When not touring, it is obvious that songwriters will earn more than the “non- songwriters” members, thanks to the author's rights each time an album, song, or music sheet is sold and each time a song is played on the radio, in public spaces or (legally) streamed on the internet. Writing royalties aiso ensure a stream of revenue throughout artists’ lives. Even when the band breaks up and stops touring, writing royalties might be earned without doing anything, making the rock musicien a rentier. Finally, in Ireland, songwriters were not required to pay taxes on earnings from music publishing until January 2006. From that point on, artists that make more than 500,000€ hâve to pay taxes on half of their “créative” income. That year, U2 moved their music publishing company from Ireland to the Netherlands because royalties remain nearly tax-free.^^ Though this move was poorly received by fans and hurt the réputation of the band, U2 defended their decision, showing that songwriters’ royalties are not negligible. This is especially true if the whole life span of a musician is considered (as it is the case in Table 1).
M
Musiciens joining a band embark on a créative process. Rock is a recording tradition. Therefore, creativity takes place during composition and in the recording studio^'* so a musician being credited as a songwriter can prove that he is not just following the leader. It shows that his créative input is rewarded and acknowledged by other band members. Because performing his own work is gratifying and a musician wants his song to sound as best as best possible, a “songwriter’s musician” will exert maximum effort during the recording sessions of his songs. If reciprocity and cohésion among members exists, every songwriter in the band will work just as hard when recording other members’ songs. The induced émulation and effort might lead to an increase in the artistic quality of the band’s output.
Fergal O’Brien, Sono, Preacher on Poverty, Tarnishes Halo With Irish Tax Move, October 15, 2006. Bloomberg.com. This is not the case, for example, in jazz music where improvisation leaves a lot of space for creativity during concerts and recording sessions whether you are a composer or not.
23 3. Data
The database consista of albums created by the 151 bands listed in Dodd’s (2001) Book of Rock, which started their career between 1970 and 1979. Dodd’s définition of rock includes not only the most important artists in the genre but aiso musicians who had a significant influence on the pop/rock scene. It aIso includes very well known bands (U2, from Ireland and Aerosmith, from the US) and less celebrated ones (Big Star from the US and Alternative TV, from the UK). Larkin’s (2006) Encyclopedia of Popular Music is used to establish discographies. To treat each band equally, we imposed a 25-year limit on ail of them. That is, if a band released its first album in 1975 we tracked its discography up to 2000. In most cases, the lifetime of a group is shorter than 25 years, and we considered 25 years to be long enough to reveal a musician’s Creative output. Compilations of songs from different studio albums are excluded, as are live albums. The final database consists of 1,494 albums released by the 151 bands between 1970 and 2004. Awards conferred by the Recording Industry Association of America (RIAA) are used as proxies for success. RIAA recognizes albums that reach a certain sales threshold. Gold and platinum awards, introduced in 1970 and 1976, respectively certify sales of 500,000 and 1,000,000 albums. Multi-platinum (2 million albums sold) and diamond (10 million) awards were introduced in 1984 and 1999, respectively. To avoid “backward spillover effects” from awards given to new releases of old albums,^® the only awards taken into account are those obtained within one year of the date of the first release. That criterion yields 60 multiplatinum, 116 platinum and 164 gold awards; 1,154 of the 1,494 albums received no award. Three variables define the internai organization of a band: dispersion, outsourcing, and instability. Dispersion is defined as the Herfindahl index (based on the sharing of crédit) divided by the total number of crédits. To compute the index for each album, we collected (using the cover of the album or other sources-the band’s webpage or specialized websites such as Discogs.com or Allmusic.com) the number of times each member of the band was cited (credited) in each song. Crédits to
Hendricks and Sorensen (2009) showed that the release of a new album increases sales of old albums, and the increase may sometimes be substantiel and is permanent. Hence an album could become gold 5 or 10 years after its release thanks to the “backward spillover effect.”
24 songs that are outsourced were not taken into account. Outsourcing measures, for each album, the share of songs that are not (co-)written by any member of the band. This is computed using the same sources as those used to compute dispersion, which is album by album. On average, 11% of the production is outsourced. Instability (or membership flux) is a dummy variable that takes the value one if a change in the group’s members occurred between two albums, and zéro otherwise. On average, a change in personnel is observed every three albums. The ratio “awarded number of albums/total number of albums” is roughiy the same for soloists (21 percent) as for groups (24 percent). Soloists are more productive (11.6 albums per soloist versus 9.2 for groups) but hâve to outsource three times more than groups.
4. Behavior and Work Organization of Rock Bands
In this section we concentrate on a subsample of our database. First we select 10 superstars: the five American bands/artists and the five non-American bands/artists that hâve sold the most albums to date, according to the R.I.A.A, as well as five “half-stars.” These bands enjoyed at least one big success (i.e., gold record) during their career but could never make it to next level of stardom (i.e., platinum record). Finally, we randomly select five bands that had no commercial success. The twenty bands and solo artists that are compared are listed in Table 2. Three of the five American superstars are solo artists. AN non-American superstars are bands from English speaking countries: Australie, U.K., Ireland, and Canada. The following subsections draw solely on this subsample. When we mention superstars, half-stars, or unsuccessfui bands, we refer to those artists listed in Table 2. Though we cannot draw general conclusions based on this subsample, it is very informative on the behavior and work organization of bands.
25 Table 2. List of Artists (selected sample) Success Name Nationality Certified units in millions^® Superstars Eagles U.S.A. 100 Billy Joël U.S.A. 79.5 AC/DC Australia 71 Michael Jackson U.S.A. 70.5 Aerosmith U.S.A. 66.5 Bruce Springsteen U.S.A. 64.5 U2 Ireland 51.5 Def Leppard U.K. 35 Queen U.K. 32.5 Rush Canada 25 Half-Stars War U.S.A. 6.5 Emerson, Lake & Palmer U.K. 4.5 Emmylou Harris U.S.A. 4 Curtis Mayfield U.S.A. 1.5 Joe Jackson U.K. 1 Unsuccessfui Fanny U.S.A. 0 Hot Tuna U.S.A. 0 Nick Lowe U.K. 0 Sham 69 U.K. 0 Throbbing Gristle U.K. 0
4.1. Bands
Time to achieve success
Bands typically achieve success between their second and fifth albums. The estimated time span to achieve stardom is at most five years, starting with the release of the first record. The number of albums a band has to produce in a certain time period is an important feature of the contract between a label and a band. The pressure that labels put on new artists to deliver albums is high. The Eagles, Aerosmith, and AC/DC released their first five albums in five years, U2 released four albums in five years, Def Leppard released three albums in four years, and finally, Queen and Rush released six albums in five ears. If a band does not break through within five years or five albums, labels stop collaborating. This is illustrated by our list of unsuccessfui bands: Sham 69 released four albums in three years, Throbbing Gristle five albums in six yeas, and Fanny five
26 Source: http://www.riaa.org. Of course, unsuccessfui bands did not sell zéro albums, but their albums (studio, compilations, live) never achieved gold certifications.
26 albums in five ears. After this first rush of albums, bands broke up. If bands do not record new material after this first “rush” of albums, this might impiy that the group did not find a new label or that the band itself decided to stop because of the lack of success relative effort. Hot Tuna, which released (seven albums in eight years) might hâve had more time to record because the two lead members were aiso part of Jefferson Airplane, a famous band in the late 1960s and early 1970s that was under contract with the same label as Hot Tuna.
Composition Outsourcing
Very few bands outsourced composition. The two bands that outsourced the most were Fanny and Hot Tuna (both “unsuccessful”). Non-American superstars composed everything by themselves. AH semi-stars and superstars recorded at least one successfui album without outsourcing any song. Until the ‘60s, the outsourcing figures would hâve been very different because there was “a division of labor between singers and writers. There was aIso no formai intégration between singers and bands. At this point, singers and writers were not tied; hence vertical and horizontal intégration had not emerged. Very little of the output of acclaimed singers from the beginning of recording until the 1960s was written by anyone other than specialist songwriters...” (Cameron and Collins, 1997). The reasons for this change are to be found in the content (lyrics) and production of songs. Rock and roll was music made by the youth, for the youth. Songs still talked about love, but involved other topics as well. Rock bands rapidiy began addressing other topics spécifie to teenagers. Two hits from 1965 exemplified this idea perfectiy: “(I Can't Get No) Satisfaction” by The Rolling Stones and “My Génération” by The Who. The Rolling Stones song is about sex and anti-commercialism. One line (“trying to make some girl”) was censored when the band performed the song at the popular “Shindig!” show aired on ABC. The lyrics from “My Génération” are considered some of the most rebellious from the ‘60s and include the famous statement, “I hope I die before I get old/Talkin' 'bout my génération.” In the ‘60s, rock lyrics talked about teenagers (“My Génération”), their place in society, and their political opinions. No one was better suited than teenage musicians to write these kinds of songs.
27 Not every rock band was tagged as having political or “rebellious” lyricists; however, these bands composed their own songs as well. Rock bands changed the mode of production for songs. Until the ‘60s, an orchestra accompanied popular acts like Frank Sinatra, Bing Crosby, Glenn Miller, or Peggy Lee. Instead, rock bands usually consisted of three to five members. Composing for orchestras requires musical knowledge and therefore a skilled composer. In the case of rock, composition is much more basic and can be done within the band, assuming each member has some knowledge of his or her instrument. The quality of a rock song or album is not found in the quality of composition (as in classical music) nor in the interprétation (as in jazz) but rather, it lies in the convergence of the three éléments: songwriting (music and lyrics), interprétation, and studio work (arrangements, sound, etc.). It seems reasonable to assume that the alchemy will be harder to find if a band outsources songwriting. Outsourcing is aiso a measure of creativity. Creative bands do not need to outsource songwriting at any point in their career. This is certainly true to a certain extent. If we do not take into account any quality, originality, or success, composing a rock song does not demand great effort for a professional musician. Hence, a band might be tempted to compose its own songs to increase writing royalties and profits, even though their creativity is low. Since “nobody knows”^^ what will be successfui, bands are encouraged to compose their own songs. If they get lucky, a song will become a hit, and the band will win the jackpot. The incentive to avoid outsourcing for bands that hâve aiready been successfui is even larger. Those bands know for sure that the next album will sell a significant number of records and that singles will be played on radios and TV channels. To maximize profits, successfui bands will not outsource, even if inspiration is low. The data suggests that outsourcing tends to decrease as a band’s career progresses. Using The Eagles as an example, illustrâtes this perfectiy. They partially outsourced for the first three albums. The third album was a big success, and afterward they completely stopped outsourcing. Bands that did not outsource at ail for the first albums (Queen, U2, AC/DC, Def Leppard) continued to avoid outsourcing throughout their whole career.
As it is defined in Caves (2000, p.3): “The producer’s intimate knowledge of the good’s production process still leaves him in the dark about whether customers will like it: nobody knows."
28 Songwriter Récognition
Songwriter’s crédits are another important issue of the internai organization of a band. A musician who is credited as songwriter, whether it is for the lyrics or the music, will hâve enhanced revenue immediately after the release of an album and throughout his or her life, if the album is successful. Bands hâve different possibilities in attributing crédits. A quite obvious scénario is the following: musiciens propose songs to the other members of the band. The band members décidé on the songs they want to record. The musician(s) who actually wrote the recorded songs and /or lyrics get the crédit. This is a fair solution: Musiciens are rewarded according to effort and quality of their work. However, this might not be the most effective solution. Because a record is a the outcome of joint production, it is hard to clearly identify the contribution of each musician and even harder to détermine how this contribution should be rewarded. We will illustrate this problem with an “everyday studio life” example. Suppose we hâve a four-member band: a bassist, a guitarist, a singer, and a drummer. The bassist cornes up with the music and lyrics for a song. For some reason the singer rewrites one line of the lyrics. In the middie of the song a 16-bar guitar solo was planned, so the guitarist has to invent a solo. Finally, the producer suggests beginning the song with a two-bar drums introduction created by the drummer. How do you share crédits and rewards in this very common case? How do you choose which songs to record? It is interesting to see how superstars hâve managed this issue. Each band has its own solution. The Eagles, a five-member band, hâve no clear ruie of crédit sharing. However, we observe that each song is credited to two or three members of the band, and ail members are credited at least once per album. This means that those who take part in the création of a song are credited, the others are not, but the band ensures that every member is recognized for her contributions. This organization ensures that those who work more and/or produce higher quality work gain more récognition (they will be involved in the création of more songs) but aiso that everybody is involved in the Creative process. This ensures the participation and motivation of every member. Aerosmith, aIso a five-member band, clearly has two leaders: Steven Tyler (singer) and Joe Perry (lead guitar). Having the singer and guitarist as dual leaders is the archétype of rock bands. This formula was popularized—\nb could even say
29 mythifiée!—in the ‘60s by the duo Mick Jagger (singer) and Keith Richards (guitarist) from The Rolling Stones. Though there are two leaders, in seven out of twelve albums other members are credited as songwriters and in two of them ail members are credited as songwriters. AC/DC has a clear scheme of crédit allocation. Three members out of five are credited. The Young brothers, Angus (lead guitarist) and Malcolm (rhythm guitar), compose the music, and the singer writes the lyrics. AC/DC is the only superstar band to hâve some members that hâve never been credited. The case of U2 is aiso very simple. The four members are credited as music composers for ail songs on every album. Regarding the lyrics, most are attributed to Bono (singer) but aIso sometimes to The Edge (guitarist). Though it is often said that Bono and The Edge bring most of the basic ideas to the songs,the band decided to adopt an equal sharing ruie. Def Leppard, a five-member band, worked in the same way as the Eagles. Several members are credited for each song on the album, and every member is credited at least once. The only différence is that one member, the drummer, is only credited as a songwriter on some albums. Queen offers yet another answer to this issue. The sharing ruIe is very original and very clear right from the band’s beginning. Each member composes a certain number of song(s). Each song is attributed to only one member (instead of several members for the other bands). The typical Queen album from the early days consisted of one song by the bass player (John Deacon), one song by the drummer (Roger Taylor), four songs by the singer (Freddie Mercury), and four songs by the guitarist (Brian May). After the fifth album, the drummer and bass player composed two songs each. This distribution is not by chance. It is clear that the four members signed an implicit contract that reflects the skills and efforts made by every member. It is worth noting that for the two first albums, John Deacon wrote no song. The other members of the band insisted that he composes as well. He finally did so in the third album of the band. This album became the first hit for Queen. It is hard to détermine if the album became a success partiy because John Deacon had a higher incentive and therefore worked harder or if it was just a coincidence. Either way, the members
Robert Hilburn wrote “Bono and guitarist the Edge bring ideas into the studio-a title, the trace of a melody or a catchy riff Robert Hilburn, Los Angeles Times, August 08, 2004. However, it is impossible to State this with certainty as the four U2 members are equally credited for music writing.
30 of Queen were right to encourage Deacon to become a songwriter. Indeed, he composed several hits for the band. One of them, “Another One Bites the Dust,” became the band's biggest hit in the U.S.A. Ail four members of Queen individually wrote hit singles, which is quite unique. Finally, the trio band Rush bas a very clear and unchanging ruie of sharing. Neil Peart, the drummer who joined the band on the second album, writes the lyrics, and Geddy Lee and Alex Lifeson compose the music. We can classify superstar bands into three different categories: (1) Bands that hâve a predefined sharing ruIe where every member is a songwriter (Rush, U2, and Queen): (2) bands that adopt no clear ruie but manage to hâve ail members credited as songwriters (The Eagles and Def Leppard) and (3) bands in which members are not aiways or never credited and adopt a clear sharing ruie (AC/DC and Aerosmith). The half-star bands War, and Emerson, Lake and Palmer fall respectively into Category 1 and 3.
Membership flux
The sharing ruie seems to hâve an impact on the stability of the lineup (membership flux) and the longevity of the band. U2, Rush, and Queen, which had very clear sharing ruies where every musicien was credited on every album, are the only bands to hâve had no problems with membership flux; the original lineup never changed. On the contrary, ail other bands had at least one membership change due to Personal conflicts. Those three bands (U2, Rush, and Queen) aiso enjoyed long careers. U2 and Rush are still active, and Queen stopped after 20 years following the death of lead singer Freddie Mercury. AC/DC aIso had to manage the death of its lead singer, but at a much earlier stage in their career, so the remaining members decided to hire a new singer. The band became even more popular with this new lineup, but it is important to note that the Young brothers composed ail the music before and after the death of the first lead singer. Def Leppard aiso had to cope with death. Guitarist Steve Clark died from an accidentai overdose on a mixture of alcohol, codeine, and valium. Fie was replaced by a new musicien, and the band enjoyed the same success as before. A couple of
31 years before this sad event, the band had to fire one of its members. The official charge was “excessive alcohol consumption in the workpiace.” As this small sample illustrâtes, the list of successfui musicians with drug and alcohol problems or dying at a young âge is, unfortunately, quite long. Bellis et al. (2007) show that pop stars expérience significantly higher mortality rates than demographically matched population.
Table 3. Bands’ Work Organization (selected sample) Variables Description Superstars Half-Stars UnsuccessfuI Average Share of songs that are not written 0.027 0.12 0.17 Outsourcing by any member of the band
Average “Normalized” Herfindahl index 0.017 0.022 0.08 Dispersion (based on the sharing of crédit)
Average 1 if a change in group’s members 0.16 0.3 0.42 Instability occurred between 2 albums
Though each band organizes differently, Table 3 suggests that superstars outsource less and tend to share crédits in a more egalitarian way than average bands. Only one group in our database, Dire Straits, became very famous despite the lack of crédit sharing among members, but they aiso recorded few albums (six albums in 14 years).
4.2. Solo artists
The organization of solo artists (soloists) is quite different from that of bands. They do not face the problem of crédit sharing, nor do they hâve to deal with the possibility of members leaving the band or breaking up the band. However, some similarities exist between solo artists and bands.
Time to achieve success
Soloists aIso hâve a limited time to break through to success. Three to five albums are necessary; this is similar to what happens for groups; however, if success is not achieved, the soloist might continue his or her career, while a band is not likely to do so. Nick Lowe was acclaimed by critics, but he never enjoyed
32 commercial success. He is still active today after more than 30 years. Soloist half- stars typically hâve longer careers than half-star bands, which usually break up if they do not achieve success. Joe Jackson and Emmylou Harris hâve been commercially unsuccessfui since 1982 and 1981, respectively, but are still active today. This suggests that labels are more generous with soloists than with bands. In fact, labels might be more likely to extend a contract for a soloist than for a band because it is less risky. With bands, labels run the risk that the group breaks up and hence lose their initial investment. If a band is unsuccessfui, the probability of breaking up increases, and the label will be less likely to sign a new deal with that band. Of course, break-ups do not exist with soloists.
Outsourcing
In general, soloists outsource more than bands, but like superstar bands, solo superstars outsource little. Billy Joël and Bruce Springsteen, two of the three superstars in our sample, never outsourced. The case of Michael Jackson is more complex but clearly reveals the value of artistic incentive provided by songwriter crédits. Before starting his solo career, Michael Jackson was aiready a star known as the lead singer of The Jackson 5. From 1971 to 1980,^® he conducted both careers at the same time. Although Michael enjoyed success with The Jackson 5, the first four solo albums did not sell very well. He had no Creative control, and ail songs were outsourced. For his fifth album, entitled Off the Wall, he changed labels and for the first time had some control over the musical and créative aspects of the album. Off the Wall instantly became a huge success. Outsourcing was still high (70%), but for the first time Michael Jackson was creatively involved in the album. Moreover, he did compose the first single and most successfui song of the album, “Don't Stop Till You Get Enough." Drawing on this success, Michael outsourced less and less throughout his career. Only 14% of the 1991 album Dangerous was outsourced.
He contributed to two more Jackson 5 albums, one in 1984 and the last one on 1989, but he clearly focused on his single career beginning in 1980.
33 4.3. Producers
Cameron and Collins (1997) define the producer as “the external allocator [...] who supervises the process of création and the nature and composition of the final product. The producer aiso serves an agency function of providing an objective response ear, from outside the band process, to judge the work. Generally the producer is only hired on a spot basis by the band and is not necessarily the sole arbiter over the finished product but his presence clearly is a means of reducing conflict.” The producer has been given a key rôle since the collaboration between George Martin and The Beatles. Martin’s impact on The Beatles’ work was so crucial that he is aiso known as the “fifth Beatle.” The real impact of a producer on the quality of music is difficult to analyze. What we will explore in this subsection is his impact on the success of musiciens. Except for Fanny (and some records by Hot Tuna), unsuccessfui bands produced their album themselves. Half-stars worked with (co-)producers on their successfui albums, except for Emerson, Lake and Palmer. AH superstars relied on a (co-)producer. In many cases, superstars’ access to stardom came in collaboration with a spécifie producer. After working unsuccessfully with one or several producers, the first collaboration with a subséquent producer generated a hit. For example, Billy Joel’s first four albums were unsuccessfui.In his fifth album, he worked with Phil Ramone for the first time, and the album was a huge success. The other bands (associated producers are between brackets) that hâve a similar story are Def Leppard (Robert John “Mutt” Lang), the Eagles (Bill Szymczyk), Michael Jackson (Quincy Jones), Aerosmith (Jack Douglas), Bruce Springsteen (Jon Landau), AC/DC (Robert John “Mutt” Lang), and Billy Joël (Phil Ramone). AH these collaborations lasted for at least three albums, and they were aiways successfui. AH successfui albums by soloists hâve been (co-)produced by an external allocator.^^ Without producer, soloists hâve only “employées” to work with, With a producer, the relationship is different. It is the producer’s job to make decisions about the structure of a song, arrangements, sound, etc. In this case, producers can be
The fourth album produced by Billy Joël was especially commercially disappointing. One successfui album has been produced by Bruce Springsteen without co-producer. This album was the less successfui during the period 1975-1995.
34 seen as a source of émulation for singers. This suggests that soloists need some émulation and teamwork to be at their best. In bands, the émulation is aiways présent because each member bas a voice; some members more than others, but they ail bave more to say tban ordinary employées. In tbat case, producers are présent to ease potential conflicts among members. In tbe few instances wben superstars did not employ an external (co-)producer, tbe album sold poorly relative to its predecessors. We can conclude tbat tbe intervention of a producer is crucial in belping a band to become successful. However, bis rôle sbould not be overestimated: despite its external and renowned producer, Fanny never became successful. Half-stars soloists, and bands released successful and unsuccessfui albums witb tbe same producers. Finally, superstars (witb the exception of the Eagles^^) became renown using several producers. In this section, we showed that bands adopt different crédit sharing ruies. These agreements between band’s members can be seen as implicit contracts. Section 5 analyzes the conséquences of the (in)completeness of contracts on the internai organization of rock bands.
5. Rock and Roll Bands, (In)complete Contracts and Creativity^^ (With Victor Ginsburgh and Patrick Legros)
In bis analysis of the “battle” between the Beatles and the Beach Boys, Clydesdale (2006) suggests that the Beatles “sbould not be seen as Creative geniuses but as a créative process, [behind which] were two dominant forces. First was the importance of rivalry with the Beach Boys and [second] the nature of the working team that possessed high levais of exchange and complementary blends of expertise and thinking styles." Clydesdale aiso suggests “that the structure of incentives is important in determining the nature of the créative output.”
This exception is most probably due to the short discography of the band. This section has been published in “Rock and Roll Bands, (In)complete Contracts and Creativity” (Ceulemans et al., 2011)
35 Indeed, when production is joint, the characteristics of partners and the nature of contracts are crucial in explaining the success or failure of the partnership. More talented partners increase the probability of success but may aiso daim a larger share of the pie. If contracts are complété —as they are when partners choose an output-contingent sharing ruie ex ante, then they will match efficiently in partnerships. It will not be possible to rematch agents in such a way as to increase the total surplus in the industry. However, if contracts are incomplète —that is, if the partners cannot agréé on sharing ruies that reflect the varying levels of creativity— the way agents match will not necessarily be surplus maximizing and may be quite different from the matching observed under complété contracting.^ This suggests that the pattern of matching can be an indicator of the degree of contract completeness. Creativity is not observable directiy but can be indirectiy measured by the crédit that members of the band receive for composing and writing songs (as well as for other skills). We develop a model where agents with different levels of creativity match and produce a joint output. When creativity within the group fails, the partnership can purchase songs created by others (outsourcing). But songs created within the group are more likely to succeed (think of them as spécifie to the group members’ characteristics) than those created by outsiders. We consider two spécifications, one in which the members of the group sign complété contracts, where, in case of success, each partner’s share is freely specified, using an output contingent ruIe, and another in which members are limited to incomplète contracts and use a “gentlemen’s agreement” to share equally the returns from their activity. The composition of the group affects the probability of creating songs within the group and the probability of outsourcing them. We show that when contracts are complété, musicians match in a négative assortative way: the most créative match with the least Creative. Under incomplète contracting, musicians match in a positive assortative way: more créative musicians match with similarly Creative musicians. This différence in the matching pattern aIso has conséquences for the relationship between an index that measures the “dispersion of creativity” within the group (and is directiy related to the matching pattern) and the probability that the group will hâve a hit.
See Legros and Newman (2007) for a general analysis of matching models with nontransferability.
36 In the complété contract spécification, there is a positive covariation between dispersion and success; when contracts are incomplète, this relation is négative. The data show that the covariation between dispersion and success is significantly négative, and that rock bands therefore appear to hâve a tendency to enter into incomplète contracts. This gives theoretical and empirical support to what was merely an hypothesis based on very good intuition in Clydesdale’s (2006) paper.
5.1. Specifying matchings
We consider two-member bands whose members are jointly involved in the création and production of songs. Musicians hâve a créative type that is distributed with distribution F [a] on [0,1]; to simplify, we restrict attention to symmetric distributions.^® Each band tries to create one song and produce it. A song that is
“normal” brings a profit jt,^ while a “hit” brings a profit >jt^.
For a given group (^a,b), the process of création is such that
• with probability (1 - a)(l- b)r\o member succeeds in creating a song. The group can then buy a song at market price q. This song will become a hit with a low probability
• with probability a{\ - è) member a créâtes the song and gets the crédit while with probability member b créâtes it and gets the crédit. Because the song is created within the band, it becomes a hit with probability, where p„ > Pi^',
• with probability ab both members succeed in creating the song, which then becomes a hit with probability p^. Because the création is joint in this case, each musicien receives crédit.
The matching patterns are indépendant of the symmetry of the distribution and the covariation between the variables is the same as in the paper for the incomplète contracting case. When contracts are complété, the covariation between the different variables is affected by the distribution F but is still different from that obtained under incomplète contracting.
37 A. Matchings and success
Let
^ = Pl^h+(^-Pl)^l ^ = Ph^h+(^-Ph>l represent the expected profits when the band buys a song from an outsider(fF) and when it produces a song created by one of its members(F). Clearly V >W since
Ph > Pl and At the time of création of the band, the expected total profit is
n ( iï, 6 ) = (1 - a){\ - b){W -q) + [a{\ - é) + b{\ -a) + ab]V.
In the complété contracting case, profits are fully transférable between members. In the incomplète contracting case, profits are imperfectiy transférable; we consider the extreme situation where profits are shared equally.^® This case corresponds to what the industry refers to as a “gentlemen’s agreement.” The set of feasible payoff allocations within a group reflects contract completeness or incompleteness. With full transferability, any allocation (w,n(«, ^)-w) between the two partners is on the Pareto frontier; with limited transferability, the Pareto frontier reduces to the pair(f](a, b)l2,Y[{a, b)l2). An equilibrium spécifiés a matching fonction and a payoff allocation in such a way that two matched agents hâve a feasible allocation for this match and there exist no feasible payoffs for any two agents that are strictiy greater than their equilibrium payoffs. As is well known, in the complété contracting case, the ex ante formation of groups will maximize total profit in the band, and the way musicians match reflects their comparative advantages. Here, becausea^n(‘^> = ^be marginal productivity of a given type of partner decreases with the créative type of the partner. There is therefore négative assortative matching in equilibrium, and if
Imperfect transferability arises if it is too difficult to agréé on shares of profits as a fonction of the characteristics of the agents. One explanation for this is the difficulty of preventing renegotiation and “hold-up” (Grossman and Hart, 1986): Once a song is created, the other musicians may threaten to leave the group or not to produce the song if they do not get a higher share of the surplus. If the song created within the group has no value outside the group, this leads to equal sharing.
38 m(a) is the match of a, then, by measure consistency, F{a) + F(/w(a)) = l . Since we
assume that F is symmetric, it must be that m{a) = l-a. In this case, the expected
probability of success is
S{a) = PH - Pl)’ which increases with a if a & 1 / 2 and decreases if a < 1 / 2. Because there is négative assortative matching, the variance of types in the group varies with a, and the amount of crédit that each member receives aiso varies with a. Note that the total
amount of crédit in the band is a + m[a) = \. Because that total is indépendant of a,
the shares of crédit received by the partners in equilibrium are(o,l- a).^^
By contrast, in the incomplète contracting case, each musician a wants to
match with the musician 6 for whom yiY\{a, b) is maximum: the process of matching
is no longer governed by comparative advantage but by absolute advantage. Since n(<3, b) increases strictiy with b , ail musiciens want to match with the highest
possible type, and this leads to positive assortative matching: now,/w(a) = a.^® The
probability of success is then
(«) = - (1 - ^)\Ph - Pl) >
which increases with a.
B. Matchings, sharing, and outsourcing
The “dispersion of creativity” measure that we use is a normalized Herfindahl index, equal to the sum of the squares of the shares of crédit divided by the total expected amount of crédit (or number of crédits) in the group. In the complété contracting case, this index is
The only one to receive crédit with probability a" is a. However, he shares crédit with the other member with probability a{\ - a), and therefore he has an expected number of crédits of a. This is true for any distribution.
39 D{a) = a^+{\ - af, which increases with a when o a and decreases when a < K- There is therefore a positive covariation between S and D in the complété contracting case. If contracte are incomplète, the crédit that goes to each member is a, while the total crédit is2o. Each partner has an equal share of crédit, yielding
which decreases with a. There is therefore a négative covariation between 5'andD. There aiso existe a covariation between outsourcing (buying a song instead of creating it) and dispersion. In the case of complété contracting, outsourcing is equal to
0{a) = a(l - a),
which increases for a < >^, and then decreases.
By contrast, if contracte are incomplète
which decreases with a. This leads to the following proposition, which will guide our empirical strategy.
PROPOSITION 1: In the complété contracting case, there is a positive covariation between the expected probability of a hit and the dispersion of crédit within the group. This covariation is négative in the incomplète contracting case. The covariation between outsourcing and dispersion is négative in the first case and positive in the second.
The model developed here deals with “singles” produced by two-member bands. In reality, bands are larger and the number of members who are credited is sometimes greater than two, but the basic insight concerning matchings and (in)completeness remains valid.
40 5.2. Data
We use the database described in section 3. Because we are interested in bands in which, most of the time, several members are active creators (though it may happen that crédit goes to only one member in some albums), we excluded albums in which ail crédit aiways goes to a single musicien (“soloists,” such as Michael Jackson). This reduced the database to 107 bands and 982 albums. That is 110 platinum (and multiplatinum) and 123 gold awards; 749 of the 982 albums received no award. Two reasons led us to consider albums instead of bands. First, bands are often unstable. Though the name of the band may remain the same, members change, and it would hâve been difficult to deal with such changes. Second, the number of albums is much larger than the number of bands, which is important for empirical analysis. In essence, we assume that each album is produced by a different band. Following our theoretical model, the variables dispersion and outsourcing define the internai organization of a band or, here, of an album. Dispersion is defined as the Herfindahl index (based on the sharing of crédit) divided by the total number of crédits. Outsourcing measures, for each album, the share of songs that a band buys on the market for songs. Success is represented by a dummy variable that takes the value 1 if the album received at least a gold award, and 0 otherwise.^®
5.3. Results and conclusions
Proposition 1 provides an easy way to test which model (complété or incomplète contracting) applies, since the sign of the corrélation between success and dispersion and between outsourcing and dispersion tells us which type of contract has been entered into. Results are summarized in Table 1. Since success is a dichotomous variable, we simpiy test whether the différence in mean dispersion varies between albums with and with no awards. The test shows that the différence is significantly négative and has a very low probability (0.0002) of being positive.
Separating gold from platinum and multiplatinum does not change the results.
41 Similar results are obtained with logit régressions, whether or not we introduce exogenous control variables that may affect sales, and thus awards, but not dispersion: (a) a dummy variable equal to 1 if the band is American, and 0 otherwise (essentially British bands, but aiso from Canada, Australia and Europe)-American bands do significantly better than others; (b) a dummy variable equal to 1 if the label is from one of the major recording studios, and 0 otherwise-it significantly helps to be produced by a major; (c) a piracy variable equal to 0 before 1999, and to 1 afterwards, to take into account that sales may hâve decreased as a resuit of piracy, making success more difficult to attain-the estimated parameter is négative as expected, though it is not significantly different from 0 at the 5 percent probability level. The coefficient of success on dispersion is significantly négative in ail cases. The corrélation coefficient between outsourcing and dispersion is equal to 0.09, which is significantly different from 0 at the 0.5 percent probability level.'’° Both results point to the conclusion that contracts are incomplète and there is positive assortative matching of partners in a band.
Table 1. Estimation Results______Panel A. Comparison ofmeans, dispersion index Albums with no 0.050 award Albums with award 0.026 Différence of means -0.024 (f = -3.56) H 0 : Différence <0 Pr = 0.9998
Panel B. Logit régressions, dépendent variabie is success (z-values between brackets) Dispersion only -8.46 (-3.81) Intercept -0.88 (-8.99) Dispersion -10.77 (-3.88) US group 0.91 (5.52) Major label 2.10(8.25) Piracy -0.63 (-1.11) Intercept -2.80 (-10.90) Observations 982
Qualitatively identical results are obtained if “soloists” are included in the calculations.
42 6. Conclusion
We show that rock music and the recording industry are bound together. First, the birth of rock and roll increased the size of the music market and the recording industry’s revenue. During the years that followed, and still today, rock music and its sub-genres lead record sales. Secondly, rock music is a recording tradition (Kania, 2006). Recording is what matters the most for rock musicians. Therefore, to analyze artists’ output and their possible success, it is essentiel to focus on albums. Moreover, we show that the internai organization of bands (and of soloists) often varied between albums. Drawing on a subsample of our database, we analyze the organization of bands and soloists by looking at four building blocks: (1) outsourcing, (2) songwriter crédit sharing, (3) instability, and (4) producers. We observe that superstars outsource very little, share crédits among members, are (relatively) stable, and make use of an external (co-)producer. UnsuccessfuI bands do not respect at least one of these “ruies.” The rôle of producers seems to be of particular relevance. Even superstars underperform when a (co-)producer is absent. We suggest that in the case of bands, the producer eases the conflicts between members, and in the case of soloists he or she represents a necessary source of émulation for the artist. When production is joint, like rock albums, the characteristics of partners and the nature of contracte are crucial in explaining the success or failure of the partnership. In the case of rock bands, partners are endowed with different level of creativity and can either sign complété or incomplète contracte. The theoretical model and the empirical results show that bands tend to enter into incomplète contracte and hence match in positive assortative way. The data aiso provides evidence that American artists perform better than others, as do albums released by a major label.
43 References
Billboard, “Music’s Top 40 Money Markers 2011”, http://www.billboard.com.
Bellos, M., Hennell.T., Lushey, C., Hughes, K., Tocque, K., Ashton, J., 2007. Elvis to Eminem: quantifying the price of famé through early mortality of Européen and North American rock and pop stars. Journal of Epidemiology and Community Health 61, 896-901.
Cameron, Samuel, and Olivier Collins, 1997. “Transaction costs and partnerships: The case of rock bands,” Journal of Economie Behavior & Organization 32, 171-183.
Caves, Richard, 2000. Creative Industries: contracts between art and commerce. Cambridge and London: Harvard University Press.
Ceulemans, C., Ginsburgh, V., Legros, P. (2011). “Rock and Roll Bands: (In)Complete Contracts and Creativity,” Amercian Economie Review, Papers and Proceedings, 101(3), 217-221.
Clydesdale, Greg, 2006. “Creativity and compétition: The case of the Beatles,” Creativity Research Journal, 18, 129-139.
Dawson, Jim and Steve Propes, 1992. What Was the First Rock ‘n’ Roll Record. London: Faber & Faber.
Dodds, Philip, 2001. The Book of Rock. London: Pavillon Books.
Fergal O’Brien, 2006 (October 15). “Bono, Preacher on Poverty, Tarnishes Halo With Irish Tax Move,” Bloomberg.com.
Garofalo, Reebee, 1999. “From music publishing to MP3: Music industry and industry in the twentieth Century,” 17(3), Amercian Music, 318-354.
Gainsbourg, Serge, October 1986, in Apostrophe (French télévision show).
Grossman, Sandy, and Oliver Hart, 1986. “The cost and benefits of ownership: A theory of vertical and latéral intégration,” Journal of Political Economy 94(4), 691- 719.
Hendricks and Sorensen, 2009. “The information and skewness of music sales,” 117(2), 324-369.
Hilbrun, Robert, 2004 (August 8). “Where craft ends and spirit begins,” Los Angeles Times, retrieved online, http://artciles.latimes.eom/2004/aug/08/entertainment/ca- hilburn8.
Kania, Andrew, 2006. “Making tracks: The ontology of rock music.” Journal of Aesthetics and Art Criticism”, 64(4), 401-414.
44 Krueger, Alan B., 2005. “The économies of real superstars: the market for rock concerts in the material world,” Journal of Labor Economies 23, 1-30.
Larkin, Colin, 2006. The Encyclopedia of Popular Music, 4th édition. New York and Oxford: Oxford University Press.
Legros, Patrick, and Andrew Newman, 2007. “Beauty is a beast, frog is a prince: Assortative matching in a nontransferable world,” Econometrica, 75(4), 1073-1102.
Music & Copyright, http://musicandcopyright.wordpress.com.
Oxford Dictionaries Online, “Rock ‘n’Roll,” http://oxforddictionnaries.com/definition/ rock+and+roll.
Oxford Dictionaries Online, “Rock Music,” http://oxforddictionnaries.com/definition/ rock+music.
Recording Industry Association of America, http://www.riaa.org.
Robinson, Charlotte, 2001. "So Tough: The Boy Behind the Sid Vicious Myth," Pop Matters,” retrieved online, http://www.popmatters.eom/pm/feature/060809-sidvicious.
Tschmuck, Peter, 2009. “Creativity and innovation in the music industry,” manuscript.
Vogel, Harold L., 2010. Entertainment industry économies: A guide for financial analysis, Cambridge: Cambridge University Press.
45 Chapter 2
Musical Characteristics and Success in Popular Music (wlth Uonel Detry)
46 1. Introduction
With the introduction of rock and roll and the 45 rpm record in the 1950's, singles became a key product for the recording industry. A single is a song that is released separately from an album in addition to usually appearing on the album. The recording label typically has the contractuel right to choose which songs from the album will be promoted as singles. This decision is crucial because it often drives the économie viability of an album. If a single sells well, it will boost the album’s sales, making it a profitable product. The pattern of a single’s life can be summarized in four steps. First, labels présent a new single to radio program planners and try to convince them to play it as often as possible. Second, radio stations play the song, and consumera buy the single. Third, (success) charts are compiled based on radio airplay and sales figures. Fourth, radio stations increase or decrease the frequency of airplay according to the charts’ rankings. The cycle repeats itself until the market is saturated. Hence, the purpose of a single is to captivate the audience of radio and télévision musical programs. The question of how to achieve this goal has aiways puzzled the recording industry: "Can the ability to achieve success be attributed to a more or less innate sixth sense? [...] Is success achieved through bribery, through massive 'plugging,' through a dulling of the senses or through conformism, as the ritual daims of the press would hâve it?" (Hennion, 1983, p.159). To reduce this uncertainty, labels use several tools: quality and quantity of promotional kits, the label’s network and social influence, press gatekeeping, and payola. Payola is defined as “undisclosed payments (or other inducements) which are given to bring about the inclusion of material in broadeast programs” (Coase, 1979). In the music industry, payola is mainly used to boost a single’s sales. Since 1960, payola has been illégal in the United States, but the practice still exists. For example, in 2002, Sony gave KHTS-FM a plasma screen télévision worth $3,325 in exchange for the station to play “Shut Up” by Kelly Osbourne, “Pandémonium” by B2K, and “This is Me...Then” by Jennifer Lopez (Rossman et al., 2008).
47 Payola is very likely to continue, as current music consumption is dominated by downloaded singles (1.1 billion units) as opposed to CDs (226 million units) and downloaded album (83 million units)^ Consumers hâve access to millions of songs and can preview each song (in 30-second excerpts on iTunes or Amazon, or the entire song on websites such as YouTube, Grooveshark, etc.). Consumers choose which songs they wish to purchase, meaning that consumers are not forced to buy the single or the entire album, as was the case before the digital era. Every song is sold at the same price^ and is equally easy to download (consumers hâve the same access to every song). Although such a market could smooth consumption across artists and songs, there are still huge inequalities in the success of different songs. Among the most reasonable explanations are social influence and product characteristics. Banerjee (1992) developed a model of social influence, known as herd behavior, which explains that people choose a product because other people hâve chosen it, even if they would not normally choose the product based on their own knowledge. The experiment of Salganik et al. (2006) tested social influence on the inequality and unpredictability of success in an artificial music market. The study revealed that even a weak indication of others’ preferences is enough to increase inequality in an artificial market. In other words, participants downloaded songs only because others had downloaded them before. In the downloadable music market of the “real world,” it is nevertheless interesting to analyze the impact of a product’s characteristics on sales. Consumers face the same price, packaging, “transportation” costs, and quality^ for ail songs; however, différentiations are found in song characteristics. In this paper, we investigate if success is driven by musical characteristics that can be measured objectively. The next section is devoted to so-called charts, which are used to measure commercial success. These are essentially rankings based on realized sales. Section 3 describes the unique database that we build and then used to
Vigures are taken from the R.I.A.A.’s website (http://www.riaa.org) and describe the U.S. market in 2010. ^The database used in the paper includes songs released in 2009. At that time, iTunes was by far the largest digital store. AH songs were sold at $0.99 until April 2009. Since then, three prices hâve become available: $0.69, $0.99, and $1.29. In practice, very few songs are sold at $0.69. New releases and popular songs are sold at $1.29, and the majority of other songs at $0.99. ^In this case, quality means sound quality. AH songs are in an AAC format and compressed at 256 kb/s.
48 estimate the influence of musical characteristics on various types of charts. Sections 4 and 5 présent the results and conclusions of our research, respectively.
2. Charts
The first music chart appeared almost a century ago in the UK. A few years later, in 1934, the Billboard chart appeared in the US and led to a sériés of radio “Hit Parade” programs (Parker, 1991). In 1958, the accuracy of sales transformed the Billboard charts into the Billboard Hot 100 as we know it today: a ranking of songs’ popularity. Nowadays, Billboard computes dozens of charts, but the Billboard Hot 100 remains by far the most popular chart for the singles’ market. This ranking is obviousiy very important for the music industry (labels, radio stations, etc.) to know which songs sell and which do not. Not only professionals are interested in the charts; consumers are aiso obsessed with sales figures and charts (Parker, 1991). Since radio stations and consumers use the charts to décidé what is worth airing and listening to, songs from the best-selling artists will continue to sell until the market has been saturated. The music chart can thus be thought of as a product and a mirror of public consumption at the same time (Attali, 1977, p.201). If a chart is a product, it can be (illegally) purchased. For example, Michael Treacy and Fred Wiersema are suspected to hâve bought The New York Times book chart, (aIso known as the “Best Sellers” list). As the authors of The Discipline of Market Leaders, they were convinced that a high position on this chart would lead to a substantial increase in sales. They secretly bought 50,000 copies of their own book from stores whose sales are monitored for The New York Times Best Sellers list. Their book made the Best Sellers list and sold well enough to continue as a bestseller without further intervention by the authors (Bikhchandani et al., 1998). This anecdote suggests that being in the charts “artificially” boosts product sales. The existence of charts might thus increase the inequality in cultural markets, and contribute to creating superstars. The study of Salganik et al. (2006) tests this hypothesis. They created an artificial music market and divided participants into three groups. Group 1 had no information about others’ preferences. Group 2 had information about the number of previous downloads, with the songs presented in a rectangular grid and the positions
49 of the songs were randomly assignée! among participants. Finally, Group 3 had the same information as Group 2, but songs were presented in descending order of popularity (i.e., ranking/chart présentation). The experiment demonstrated that the ranking présentation increased success inequalities. This holds true whether the results of Group 3 are compared with those of Group 1 or of Group 2. Berns et al. (2010) obtained a similar resuit: song popularity had a significant effect on the participants’ likability ratings of the songs. In other words, participants said they liked a song because the song was popular. The study suggested that the anxiety generated by the mismatch between one’s own preferences and the preferences of others is a principal mechanism for how rankings affect consumer choice. A study by StrobI and Tucker (2000) on chart success of albums in the UK indicated that charts were highiy skewed, whether measured by the total number weeks of success per artist or the total number of albums listed per artist. The length of survival on the charts was positively correlated with album type (greatest hits and soundtracks performed better) and initial popularity, while it was negatively correlated with seasonal demand (albums entering the charts during the pre- Christmas period remained in the charts for a shorter period of time). Charts are so important for music consumers that nearly every specialized magazine or website now provides its own rankings. Those charts are aiso included in our database, which is described in the next section.
3. Data
We constructed the database in three steps explained in each of the sub sections. We selected the songs to be included in the dataset based on critics and music levers’ “best songs of the year” lists as well as on billboard charts. We observe that critics’ rankings and billboard charts consist of different songs and hence that a song’s success might dépend on the criteria used to evaluate it. To take this observation into account, we define several measures of success. Finally, we analyze the musical content of each song. This led to a database consisting of 514 songs which makes it possible to analyze the effect of musical characteristics on success.
50 3.1. Choice of songs included in the database
We looked at several 2009 year-end charts. Songs that appeared in at least one of these charts constitute our database. Three types of charts were taken into account in order to encompass ail aspects of the musical market: commercial charts, critics’ charts, and music lovers’ charts.
Commercial Charts
Commercial charts consist of the top-selling tracks of 2009 in the United States, and four sources were used. The first chart, the Billboard Hot 100, was selected because it is the most famous and one of the oldest music charts. At the end of each year, Billboard releases a list of the top 100 most successfui songs during the year. Like their weekiy charts, it represents songs with “the greatest airplay and sales gains.”'* The Billboard Hot 100 “indicates sales in the largest recorded music market in the world [and] it aiso takes into account radio airplay. This is largely because of the greater importance that commercial radio has in promoting records in the US market.” (Parker, 1991, p.207). The second chart selected was iTunes, because it is the leading store in the digital music market with approximately 70% of the market share.® At the end of every year, a list of the 10 songs that realized the largest number of sales is released. The third list cornes from Amazon, which publishes a list of the 100 best- selling MP3s from their website. Their MP3-selling platform ranks second in the digital music market in the US.® The reason for analyzing these digital charts is obvious. Over the last few years, the singles market has been evolving into a fully- digital market. In the US, during 2009, 99.9% of the singles sold were digital files.^ Finally, in order to consider the latest trends in musical consumption behavior, we included the chart of Spotify, which is a service that allows users to stream music
‘*http://www.billboard.com. ®According to the NPD Group, “Digital Music Increases Share of Overall Music Sales Volume in the U.S," http://www.npd.eom/press/releases/press_090818.html. ®According to the NPD Group, “Digital Music Increases Share of Overall Music Sales Volume in the U.S,” http://www.npd.com/press/releases/press_090818.html. ^According to the R.I.A.A. shipment database, http://www.riaa.com.
51 on the web. It bas rapidiy become one of the most-used music streaming services and accounts for more than 10 million users across Europe.® The chart we used is a list of the top 100 of most-played tracks in 2009.
Critics’ Charts
Critics’ charts consist of “best songs of the year” lists presented by eight magazines and one radio program: Pop Matters (Top 50), Spinner (Top 25), Conséquence of Sound (Top 50), NME (Top 50), Pitchfork (Top 100), Rolling Stone (Top 25), Slant (Top 25), Spin (Top 20), and NPR (“Song of the Day” Top 10). These magazines/websites were selected due to their famé, prestige, and/or readership. A brief description of each magazine follows, along with the website’s traffic ranking in the US (enclosed in parenthèses) according to Alexa.com.® Alexa.com ranks ail websites according to the number of page views and average daily visitors over the previous three months. We used this ranking as a rough measure of magazine’s popularity. Pop Matters (3,821) is an online magazine that was launched in 1999. It discusses movies, comics, books, and music in the form of regularly-written reviews, interviews, and academie essays. Sp/nner (2,167) is a blog that diseuses only about music but in a much more commercial manner than Pop Matters. Conséquence of Sound (6,901) is an online magazine that was launched in 2007 and has received several awards for its quality.^® The New Musical Express (3,921), better known as NME, is the “world’s longest-running music weekly.”^^ It started publishing in March of 1952, and it was the first British magazine to include a singles chart, in November of 1952. Today, NME aiso provides music news through its website. Even though NME is based in the UK, we decided to include it because of its prestige. Pitchfork (913) is known as the leading media source for independent music in the United States and abroad. It is a website, established in 1995, that publishes music criticism, news and artists’ interviews. Since 1967, Rolling Stone (641) has
®http://eu.techcrunch.com/2010/09/15/spotify-10-million/. Although it is a European website, 71 songs ranked in the Spotify Top 100 has been ranked in the Billboard Hot 100. ®Ranking retrieved July 4, 2011. ^°http://consequenceofsound.net/advertising. ^^http://www.nme.com/about.
52 been one of the most important music magazines in the world. Although it covers other topics like politics, the magazine is best known for its music critiques and interviews. On the internet, it is aiso one of the leading websites in the music market. Though less visited, Slant (8,272) enjoy a good réputation. For example, The New York Times called Slant “a repository of passionate and often prickly pop-cultural analysis.Spin (3,790) was established in 1985 and rapidiy became an alternative to Rolling Stone. Since 2008, Spin aIso releases a digital édition of the magazine. National Public Radio (241), aiso known as NPR, broadcasts through 900 stations in the US. In total, it reaches an audience of more than 25 million listeners each week.^^ NPR’s program, “Song of the Day,” releases a top 10 list at the end of the year.
Music Lovers’ Charts
Music lovers’ charts consist of what listeners like, share, rate, or simpiy listen to. Three charts were selected based on three criteria: availability of the chart, number of users, and quality of the chart. The first one, “Last.fm”, is a website that enables users to share what they are listening to. In 2008, it reported a figure of 21 million unique users each month.^"* As each track is listened to by a user, it is sent through a widget to the internai server, so charts are updated continuousiy. At the end of each year, a list of the 40 most- played artists is computed. For each artist, we included the most-played track. The second chart, “We Are Hunted,” includes the top 99 hottest singles. It focuses on how music is spreading rather than how music is selling. Its aim is to “listen to what people are saying about artists and their music on blogs, social networks like Facebook and MySpace, message boards and forums, Twitter and P2P networks to chart the top songs online every day.”^^ This process results in a ranking that includes more “alternative" songs than commercially successfui ones.
^^Scott, A.O., “Say ‘Brian De Raima.’ Let the Fighting Start,” September 17, 2006, The New York Times retrieved online; http://www.nytimes.com/2006/09/17/movies/17scot.html' ^^http://www.npr.org/about/aboutnpr. ^''The Guardian : http;//www.guardian.co.uk/media/2008/feb/22/digitalmedia1. ^®http;//wearehunted.com/a/#/about/.
53 Finally, the third chart is “Rate Your Music,” which is a user-driven music database where users can rank and review albums. It contains more than two million releases, 20 million rankings, and 1.3 million reviews. The website offers a continuousiy-updated users’ chart that is based on songs’ release dates. We extracted the ranking (top 100 songs released in 2009) at the beginning of 2010. Though ail the selected charts concern the year 2009, some of them include songs released before that year. Older songs would only be listed in the charts because, for instance, they were made popular again by a movie or TV show or because of the “death factor” (when a star dies, we often observe a ravivai of his or her popularity). This is what happened with the death of Michael Jackson on June 25, 2009. Many people gathered to listen to Jackson’s music, and radios aired his music more than ever. The charts from Last.fm show that millions of his songs were played just after his death and in the month that followed. That put him well above other artists, with twice as many listeners as the second-ranked artist.^® Sales of Jackson’s songs aiso increased, which resulted in his higher ranking on the Billboard after his death. We deleted those older songs from our database because the purpose of this paper is to elicit a link between musical characteristics (which is likely to be a fonction of the spirit of time) and success.
3.2. Measures of success
For the 514 songs in our database, we recorded their highest rank (or peak position) in the Billboard Mot 100 chart and the number of weeks they stayed in that chart. These two variables describe intensity and length of success, and were therefore used to define two variables that measure a song’s commercial success in the United States. A song ranked number one in the Billboard Mot 100 will be given the value 100, and a song ranked 100 will receive the value 1. Songs that are not ranked are given the value of 0. This is for ease of interprétation and to maintain consistency with the other “success” variables, which are ascending. The variable is called “Billboard Rank.”
^®See, for instance, the weekiy chart that follows his death: http://www.lastfm.fr/charts/artist?charttype=weekly&subtype=artist&range=1245585600-1246190400.
54 “Survival” is the number of weeks a song remains in the Billboard Hot 100. The information were retrieved until April 2011 so that every song had a complété “singles’ cycle” (i.e., enter the charts, reach its peak position, and exit the charts). This late retrieval avoids discrepancies between songs released in the beginning of 2009 and those released at the end of that year. In addition to these two “commercial success” indicators, we constructed two other “success” variables, one that measures critics récognition, and the other that représenta music levers’ preferences. “Critics” is the number of times a song is mentioned in critics’ year-end charts. Since there are eight such charts, the variable “critics” takes values 0, 1, ..., 8. Finally, “Music Levers” is the number of times a song is présent in “music levers’” year-end charts. It takes values 0, 1, 2 or 3.
3.3. Musical characteristics and control variables
In order to analyze musical attributes of these songs, we used a tool called Analyze, which was developed by The Echo Nest Corporation and consiste of an application-programming interface (API) that produces an analysis of various song aspects.^® It is the only musical analysis tool that incorporâtes music perception principles to produce objective measurements. After uploading a song, Analyze delivers résulté for a variety of attributes that are directiy stored in a spreadsheet. For our research, we focus on the following factors delivered by Analyze: “Key,” “Duration,” “Mode,” “Tempo,” and “Time Signature.” Although Analyze was usefui, it made many mistakes for some variables. Moreover, the API did not deliver résulté for three variables that we wanted to include in the study: “Type of Production,” “Gender,” and “Triplet.” This required us to go through the database again and listen to the songs to correct the mistakes and add the new variables. My co-author analyzed the “harmonie” variables and I took care of the “rhythmic” variables song by song. Our database consiste of 514 songs and the average length is 4 minutes and 3 seconds. That is my co-author and I both had to listen to more than 34 hours of music to compile ail the necessary information. Table 1 describes the dataset in some detail and gives the number of observations (frequencies) for each variable.
’^We excluded NME because its British origin would make the impact of artists’ nationality (one of the control variables that are described later) on “Critics" undetectable. ’®The API is available for free on The Echo Nest’s website: http://the.echonest.com.
55 This section describes the variables used to measure the musical characteristics of the songs and discuss the information contained in Table 1. (1) Type of production: each song is classified as “electronic,” “acoustic,” or “mixed.” Vocals were not considered in this différentiation. In an electronic song, one can hear (almost) only electronic sounds: drum machines, synthesizers, etc. An acoustic song is played with acoustic instruments: guitar, piano, bass, drum kit, etc. Guitars may or may not be electric. A mixed song is a blend of acoustic and electronic sounds. Though electronic music is popular since the 1980s, most songs (295 out of 514) in 2009 are still played with acoustic and electric instruments.
(2) Gender: three possibilities can characterize the singer(s): male, female, or duo (i.e., one male and one female lead singer). We did not include in the database instrument-only songs (i.e., songs without a singer). It represents eleven observations, none of which is charted in the Billboard Hot 100. The commercial music industry is known to be dominated by male artists. The data support this statement: male singers (327 songs) outnumber female singers (140 songs).
(3) Tempo: the speed of the song, expressed in beats per minute. The average tempo is 114 beats/minute and the standard déviation is 31.4 beats/minute.
(4) Duration: length of the song expressed in seconds. The average length of a song is 243 seconds and the standard déviation is 71 seconds.
(5) Key: the overall musical key in which the song is written. C is noted KeyO; C# is noted Keyl,..., B is noted Keyll. Table 1 shows that ail keys are used by composers but not with the same frequency. D and C are the most common keys as opposed to G# and A#. The data does not make it possible to establish if this is a trend in the commercial music genres or if it is particular to the year 2009.
(6) Mode: the mode is either minor or major. Although many different modes exist, these are the only two used in popular music. Our data indicates that musicians compose evenly in minor and major modes.
56 (7) Time signature: the overall time signature détermines the number of beats in each measure. It is aiso called a meter. In popular music, most of the songs are in 4/4. In our case 97% of the songs hâve a “4/4” time signature. (8) Triplet: when each beat is composed of three sub-beats, it can create a different groove, even though the song may still be in a 4/4 time signature. Only 45 compositions use the “triplet” rhythmic.
Finally, we added three control variables:
(1) Label: we classified the songs into two label groups, major or indépendant, depending on which label they were signed with at the time of the release. This classification reflects the organization of the actual music industry, which is divided into a large number of small indépendant labels and four major corporations (Universal, Sony, Warner, EMI) that account for approximately 75% of the market.^® Our database is even split between musiciens signed by majors and Indépendant labels.
(2) Nationality: a dichotomous variable depending on the US or non-US origin of the artist. Sixty percent of the musiciens in our data are American.
(3) Ouest: a dichotomous variable signaling the appearance of a guest star in the song. It is common to market a song or an artist by including a well-known singer as being “featured” (or as a guest) in the song. There are nearly as many “duos” as there are “guests” songs in our database suggesting that in order to broaden artists’ audience even more “guest” artists are often from the opposite sex than the “main” artist.
^^According to Music & Copyright (figures for 2009): http://musicandcopyright.wordpress.eom/2010/04/21/sony-music-makes-gains-on-dominant-universal-in-2009/
57 Table 1—Descriptive Statistics No. Of Variable observations Acoustic 295 Type of production Electronic 119 Mixed 100 Female 140 Gender Male 327 Duo 47 Tempo see text Duration (in seconds) see text 0(C) 58 1 (C#) 30 2(D) 80 3(D#) 27 4(E) 56 5 (F) 35 Kev 6 (F#) 29 7 (G) 43 8 (G#) 22 9 (A) 56 10(A#) 22 11 (B) 56 Major 269 Mode Minor 245 4/4 499 Time Signature Non-4/4 15 Yes 45 Triplet No 469 Major 263 Label Indépendant 251 U.S.A. 315 Nationality Non-U.S.A. 199 Yes 41 Song features a guest No 473 Total Number of Observations
58 4. Results
We deal with two sets of variables, summarized in Table 2, one of which contains several measures of success (“Billboard Rank”, “Survival”, “Critics”, “Music Lovers”), while the other contains musical characteristics and control variables. We considered using a statistical technique referred to as canonisai corrélation (see for example Anderson, 1984), which allows “regressing" one set on the other, in fact finding two linear combinations, one for the success variables the other for the characteristics and Controls, such that the corrélation between the resulting two linear combinations is the largest possible. We thought that this multivariate technique could shed some light on where the strong corrélations would lie. The analysis unfortunately led to results that we were unable to interpret. Since the results of canonical corrélations were difficult to interpret, we turn to regressing each “success” variable on musical characteristics and control variables, distinguishing between commercial success (“Billboard Rank” and “Survival”) and success as judged by critics and music lovers. Before analyzing the results, we should make two preliminary remarks. First, some “Key” coefficients are statistically significant but not reliable because the “Key” variables are not jointly significant, even when the less frequently used keys are deleted from the régression. Second, observations on songs constitute the database, but one artist might be the author of several songs. This means that we hâve “groups” or “clusters” in our data and we suspect that intragroup observations are correlated: observations are independent across groups (clusters) but not necessarily within groups. Therefore we aiways use a robust variance estimator for cluster-correlated data (Williams, 2000 and Rogers, 1993).
^°ln fact, one can compute several such combinations. For details, see e.g. Anderson (1984).
59 Table 2—Variables’ Description Type of variables Variables Description “Inverted” peak position of the song in the Billboard Rank Billboard Mot 100 (a number between 0 and 100). Number of weeks a song remained in the Survival Billboard Mot 100 (a number between 0 and 76). Success Number of times a song is présent in critics’ Critics year-end charts ( a number between 0 and 8). Number of times a song is présent in music Music Lovers lovers’ year-end charts (a number between 0 and 3). Electron 1 if song is electronic; 0 othen/vise. Type of ic Productio 1 if song is a mix of electronic and acoustic; Mixed n 0 otherwise. Acoustic 1 if song is acoustic; 0 otherwise. Male 1 if lead singer is a male; 0 otherwise. Female 1 if lead singer is a female; 0 otherwise. Gender 1 if lead singers are female and male; 0 Duo Musical otherwise. Characteristics Tempo Beat Per Minute. Duration Length expressed in seconds. Squared Duration Square of duration. Coded as 12 binary variables. One for each Key key. Mode 1 if song is in major mode; 0 otherwise. Time Signature 1 if song is in 4/4; 0 otherwise. 1 when each beat is composed of three Triplet sub-beat ; 0 otherwise. 1 if song was released by a major; 0 Label otherwise. 1 if song was made by an American artists; 0 Control Nationality otherwise. Guest 1 if quest appearance; 0 otherwise.
4.1. Commercial Success
The “Billboard Rank” variable contains many zéros since many songs did not chart in the Billboard Mot 100. In these cases, we do not observe the level of success (i.e., whether they were totally unsuccessfui or very close to enter the Billboard chart). These observations are thus left-“censored” because we do not know whether an observation should take the value zéro or be négative. We first ran a tobit régression (see Tobin, 1958). However, the dépendent variable “Billboard Rank” is not continuous but rather a ranking (with many ties possible). Consequently, we aiso used ordered logit and logit models. In the ordered logit case, we grouped
60 the observations in eight categories according to the billboard rank. Hence, the variable “Billboard Rank” takes the value 0 (for unranked songs), 1, 2..., and 7(for highiy ranked songs). In the logit case, “Billboard Rank” is a dichotomous variable that takes value zéro for non-ranked songs and one otherwise. The logit régression estimâtes how musical characteristics influence the probability of entering the Billboard chart. Results are shown in Table 3. In ail three régressions, the coefficients of “Electronic” and “Female” are positive and statistically significantly different from zéro at the five and ten percent probability level, respectively. Electronic songs perform better and female singers are more likely to enter the charts. The control variables play an important rôle: “Nationality” and “Label” are the most significant ones (included in our database) that explain a song’s position in the charts. This suggests that payola, mainly used by majors,^^ and other promotional techniques are efficient. Having a “Guest” (a trick used by some important artists) might aiso boost sales. There is an optimal duration for a song when the coefficients of “Duration” and “Squared Duration” are positive and négative, respectively. Grain and Tollison (1997) show that the duration of hit songs fluctuâtes over time. They analyze the average length of number 1 songs and distinguish three periods: 1940-1955, 1956-1964, and 1965-1988 with an average duration of 166 seconds; 146 seconds, and 235 seconds, respectively. Songs were shorter before the 1970s because the best selling format at the time was the 45rpm vinyl dise that could not hold more than three minutes of music. Hence, record labels promoted only songs that fit into that spécifie format. When the technical restrictions disappeared, the length of hit songs increased up to the “four minutes format” which exists for the last thirty years. For 2009, we estimate the optimal duration of a song to be equal to 246 seconds. The length is likely to be influenced by radio stations since long songs might be excluded from their playlists.^^
^^Independent labels do rarely hâve the financial asset to use payola on a large scale like majors can do. ^^When a song on an album is too long and that a label wants to promote it as a single, a “ Radio edit” (i.e., a shorter version) is oflen produced.
61 Table 3—Commercial Success Estimation Results: Billboard Rank Tobit______Logit______Ordered Logit Coeff. t-value Coeff. z-value Coeff. z-value Electronic 31.43*** (2.91) 0.93** (2.31) 1.12*** (3.2) Mixed -5.40 (-0.43) -0.27 (-0.58) 0.11 (0.28) Female 16.76* (1.82) 0.57* (1.65) 0.62** (2.03) Duo 4.21 (0.34) -0.05 (-0.11) 0.50 (1.09) Tempo -0.12 (-1.09) -0.003 (-0.76) 0.00 (-0.9) Duration 1.58*** (2.8) 0.04** (2.44) 0.05** (2.53) Squared -0.003*** (-2.83) -9E-05** (-2.55) -10E-05^** (-2.47) Duration Key1 -18.01 (-1.29) -0.85 (-1.58) -0.25 (-0.55) Key2 -26.36 (-1.56) -1.36** (-2.1) -0.57 (-1.02) Key3 -16.11 (-0.96) -0.99 (-1.5) -0.45 (-0.86) Key4 -4.46 (-0.33) -0.32 (-0.6) 0.17 (0.37) Key5 -16.33 (-1.01) -0.7 (-1.06) -0.30 (-0.62) Key6 -7.61 (-0.45) -0.63 (-0.9) 0.39 (0.66) Key7 8.67 (0.62) 0.23 (0.42) 0.55 (1.26) Key8 -32.68 (-1.61) -1.35* (-1.77) -0.91 (-1.45) Key9 -17.28 (-1.13) -0.88 (-1.43) -0.22 (-0.43) Key 10 -27.07* (-1.72) -1.07* (-1.71) -0.56 (-1.26) Keyll -26.16 (-1.64) -1.31** (-2.1) -0.47 (-0.86) Mode 1.07 (0.12) 0.15 (0.45) 0.05 (0.18) Time Signature 26.30 (0.54) 1.49 (0.86) 1.29 (0.81) Triplet -3.22 (-0.24) -0.14 (-0.31) -0.07 (-0.18) Label 117.47*** (11.57) 3.67*** (7.97) 3.50*** (8.55) Nationality 59.72*** (5.67) 1.88*** (5.09) 1.80*** (4.7) Guest 24.29* (1.88) 1.29** (2.1) 0.52 (0.97) Intercept -332.6*** (-3.89) -10.5*** (-3-52) / / Intercept (cut)1 11.72697 Intercept (eut) 2 11.96665 Intercept (eut) 3 12.13882 Intercept (eut) 4 12.82398 Intercept (eut) 5 13.54116 Intercept (eut) 6 14.12505 Intercept (eut) 7 14.51874 Observations Left- 353 Censored 161 Uncensored Total 514 514 514 Notes: 1) *, **, *** indicate that the coefficients are different from zéro at a 10, 5 and 1% probability level, respectively. 2) Omitted dummy variables are: “Acoustic" for “Type of Production", “Male” for “Gender”, “KeyO” (i.e. Key C) for “Key”, “Minor” for “Mode”, “non-4/4” for “Time Signature”, “no triplet” for the variable “triplet”, “Non-U.S.” for “Nationality”, and “No Guest” for “Guest.”
62 Our second measure of commercial success is survival in the charts. “Survival” can be thought as a count variable (a song can stay 0, 1,2, etc. number of weeks in the charts). Therefore we use a count model and because “Survival” exhibits over-dispersion — the variance is much larger than the mean — the négative binomial should perform better than the Poisson model. As shown in Table 4, we estimate “Survival” using a zero-inflated négative binomial régression, since the data include an excess of zéros (i.e. a large number of non-ranked songs). When we restrict the data to the songs that hâve been charted, the number of observations falls to 161. In this case, the dépendent variable takes only strictiy positive values, though zéro is a possible outcome. Therefore we use a zero- truncated négative binomial model. Zero-inflated and zero-truncated négative binomial régressions lead to similar results. They show that electronic songs stay longer in the “Billboard Hot 100” and duration increases survival in the charts up to a certain point, after which it decreases. Contrary to “Billboard Rank”, “Survival” does not dépend on “Label”, “Nationality”, or “Guest” appearance. This suggests that major labels can buy — literally for songs that are promoted via payola, though it is prohibited — a song’s entry in the charts but not its survival. It could be that once a song enters the “Billboard Hot 100”, it has sufficient media attention to compete fairly with ail other charted songs. If this is the case, it seems reasonable that being American, promoted by a major label, or having a guest would not significantly improve chart survival. The coefficient of “Mode” is positive and significantly different from zéro at the five percent probability level. “Happy” songs are more likely to stay in the charts. Indeed, the major and minor modes are associated with happiness and sadness, respectively (see e.g. Crowder, 1985). There is no definite explanation for why the association exists but similarities between the spectra of voiced speech uttered in different emotional States and the spectra of particular minor and major intervals hâve been observed (Bowling et al., 2010). For example, the minor third communicates sadness in speech, mirroring its use in music (Curtis and Bharucha, 2010).
^^Note that in this case, one song only has a “time signature” different form 4/4. We therefore excluded the variable from the zero-truncated négative binomial régression.
63 Table 4—Commercial Success Estimation Results: Survival Zero-Truncated Zero-Inflated Négative Binomial Négative Binomial Coeff. z-value Coeff. z-value Electronic 0.32** (2.57) 0.30** (2.34) Mixed -0.02 (-0.13) -0.02 (-0.16) Female -0.01 (-0.13) -0.02 (-0.16) Duo 0.01 (0.09) 0.01 (0.08) Tempo -0.001 (-0.53) -0.001 (-0.42) Duration 0.03*** (4.57) 0.03*** (4.30) Squared -5E-05*** (-4.72) -5E-05*** (-4.42) Duration Key1 0.05 (0.34) 0.04 (0.34) Key2 0.02 (0.13) 0.02 (0.11) Key3 -0.04 (-0.29) -0.04 (-0.28) Key4 -0.02 (-0.13) -0.02 (-0.14) Key5 -0.34** (-2.35) -0.35** (-2.35) Key6 0.13 (0.67) 0.13 (0.67) Key7 0.05 (0.25) 0.04 (0.22) Key8 -0.07 (-0.29) -0.07 (-0.27) Key9 -0.03 (-0.18) -0.03 (-0.20) Key10 -0.38* (-1.69) -0.38* (-1.68) Key11 -0.01 (-0.03) 0.02 (0.09) Mode 0.29** (2.4) 0.28** (2.26) Time Signature // 0.54 (1.55) Triplet 0.001 (0) -0.01 (-0.04) Label 0.20 (0.94) 0.17 (0.77) Nationality 0.10 (0.74) 0.10 (0.79) Ouest 0.02 (0.15) 0.01 (0.12) Intercept -0.84 (-0.99) -1.21 (-1.53) Observations Zéro 353 Non-Zero 161 Total 161 514 Notes: 1) *, **, *** indicate that the coefficients are different from zéro at a 10, 5 and 1% probability level, respectively. 2) Omitted dummy variables are: “Acoustic" for “Type of Production”, “Male" for “Gender”, “KeyO" (i.e. Key C) for “Key”, “Minor” for “Mode”, “non-4/4” for “Time Signature”, “no triplet” for the variable “triplet”, “Non-U.S.” for “Nationality”, and “No Ouest” for “Ouest.”
64 Our results show that there is no magical formula to compose a hit song but that the “spirit of time” and the promotional force of majors can hâve a significant influence. The stéréotypé of a hit song in 2009 is sung by a female American artist who is under contract with a major label, sings a four-minute song in major mode with the support of a guest. A perfect example is the American singer Lady Gaga, who is under contract with Interscope (a subsidiary of Universal Music). Her song, “Just Dance” featuring Colby O’Donis (4:02 in length and composed in major mode) was one of the biggest successes of 2009.
4.2. Critics’ and Music Lovers’ Success
We now analyze how professionals (variable “Critics”) and amateurs (variable “Music Lovers”) are influenced by musical characteristics. “Critics” is a number between zéro and eight and “Music Lovers” a number between zéro and three (both numbers represent the number of charts). In both cases, we use the ordered logit estimation method. Very few songs appear in more than two charts, and we decided to aggregate into one group ail the songs présent in two or more than two charts. We aiso aggregated in two groupe only (0 and positive) and ran logit régressions. The results, presented in Table 5, show that musical characteristics and control variables hâve a different impact on Critics charts than on Billboard charts (in Table 3) We might hâve expected this resuit for three reasons: (1) critics are professionals and listen to much more music than the “average consumer.” This should hâve an effect on music preferences, (2) critics target “music lovers” who are a sub-sample of music consumers, and (3) critics are not “corrupted” by payola. The coefficients picked up by “Mixed”, “Duo”, and “Nationality” are positive and statistically significantly different from zéro at the five percent probability level for the first one and at the ten percent probability for the other two variables. These results suggest that critics might pay attention not to hurt their potentiel readerships. Indeed with “Mixed” songs, critics favor electronic and acoustic sounds at the same time; the same argument holds for “Duo” (one female and one male lead singer). Online magazines hâve mostly an American readership. This might explain their tendency to favor American artists. Finally, the négative sign of the coefficient picked by the “Label” variable might be seen as a signal sent to readers that magazines are
65 independent with respect to majors. However, the hypothesis that critics are influenced by their potentiel readership can unfortunately not be tested with our data. Critics prefer songs composed in major mode. As a significant part of their job is to listen to music, critics might be biased in favor of “happy” songs.
Table 5—Critics and Music Levers Estimation Panel A. Critics Panel B. Music Levers Logit Ordered Logit Logit Ordered Logit Coeff. z-value Coeff. z-value Coeff. z-value Coeff. z-value Electronic 0.09 (0.33) 0.07 (0.27) -0.71** (-2.31) -0.70** (-2.34) Mixed 0.72“ (2.57) 0.71*** (2.8) -0.07 (-0.26) 0.004 (0.02) Female 0.10 (0.45) 0.06 (0.25) 0.23 (1.03) 0.30 (1.37) Duo 0.62* (1.79) 0.59* (1.77) -0.08 (-0.23) -0.15 (-0.46) Tempo 0.002 (0.59) 0.002 (0.68) 0.004 (1.10) 0.003 (0.99) Duration 0.007 (1.08) 0.005 (1.13) 0.003 (0.57) 0.003 (0.54) Squared -3E-06 (-0.23) -7E-07 (-0.13) -7E-06 (-0.91) -7E-06 (-0.91) Duration Keyl -0.28 (-0.52) -0.28 (-0.52) -0.27 (-0.51) -0.37 (-0.7) Key2 0.03 (0.07) 0.04 (0.09) 0.19 (0.49) 0.12 (0.31) Key3 0.41 (0.76) 0.10 (0.22) -0.12 (-0.23) -0.14 (-0.25) Key4 -0.10 (-0.24) -0.14 (-0.35) -0.13 (-0.33) -0.21 (-0.53) Key5 0.74 (1.58) 0.46 (1.07) 0.46 (0.99) 0.56 (1.15) Key6 0.21 (0.37) 0.24 (0.46) 0.04 (0.08) -0.09 (-0.18) Key7 -0.82 (-1.59) -0.83* (-1.77) 0.52 (1.14) 0.49 (1.09) Key8 0.82 (1.52) 0.82 (1.45) 0.01 (0.01) 0.06 (0.09) Key9 0.49 (1.07) 0.52 (1.24) 0.17 (0.42) 0.18 (0.44) KeylO -0.33 (-0.58) -0.48 (-0.94) -0.14 (-0.23) -0.26 (-0.44) Keyll 0.13 (0.28) 0.03 (0.08) 0.62 (1.47) 0.50 (1.21) Mode 0.42* (1.90) 0.37* (1.82) -0.34 (-1.50) -0.32 (-1.47) Time signature 0.22 (0.30) 0.04 (0.05) -0.47 (-0.76) -0.33 (-0.62) Triplet -0.29 (-0.82) -0.27 (-0.78) 0.49 (1.52) 0.45 (1.47) Label -1.25*** (-5.51) -1.06*** (-4.74) -0.82*** (-3.89) -0.75*** (-3.56) Nationality 0.41* (1.88) 0.42* (1.94) -0.62*** (-2.94) -0.60*** (-2.87) Ouest -0.43 (-0.94) -0.38 (-0.85) -1.51** (-2.39) -1.52** (-2.44) Intercept -2.51* (-1.85) 0.21 (0.17) Intercept (eut) 1 1.87 -0.13 Intercept (eut) 2 4.06 2.67 Observations 514 514 514 514 Notes; 1) *, **, *** indicate that the coefficients are different from zéro at a 10, 5 and 1% probability level, respectively. 2) Omitted dummy variables are: “Acoustic” for “Type of Production”, “Male” for “Gender”, “KeyO” (i.e. Key C) for “Key”, “Minor” for “Mode”, “non-4/4” for “Time Signature”, “no triplet” for the variable “triplet”, “Non-U.S.” for “Nationality”, and “No Ouest” for “Ouest.”
For “Music Levers,” the coefficients of “Electronic,” “Label,” “Ouest,” and “Nationality” are negatively signed and statistically significantly different from zéro at the five percent probability level. Since our “Music Levers” ranking is made by internet users from around the world, the “Nationality” coefficient is less relevant for our analysis. The coefficients hâve the opposite sign of those obtained for the
66 “Billboard Rank" régression. This might be contradictory because music levers are aiso music consumers. But music levers seem te represent a too small fraction of the music consumers’ population to bave a significant impact on the Billboard charts.
5. Discussion and Conclusion
Several experiments in the field of psychology (for reviews, see Jusiin and Laukka, 2003) hâve demonstrated that objective properties of music hâve an impact on listener responses. For example, tempo has an effect on arousal and pleasure (Kellaris and Kent, 1993). The présent study évaluâtes whether such musical characteristics are aIso correlated with consumption behavior and critic ratings. Our results suggest that only the mode impacts songs’ survival in the commercial charts as well as critics’ rankings. The influence of the other characteristics dépends on the population sample (specialists versus non-specialists), suggesting that public and critics’ tastes are different. Comparing the three logit régressions (“Billboard Rank”, “Critics”, and “Music Levers”), one can observe that the “Label” coefficient is aiways statistically significantly different from zéro at the one percent probability level but it is positive with “Billboard Rank” and négative otherwise. This différence may be due to payola which hides the musical content to non-specialists. In other words, non-specialists are more likely to pay attention to the media hype and specialists to the music itself. This hypothesis is supported by an experiment on the influence of popularity on ratings of music: “Conformists [participants who are heavily influenced by other’s preferences, like the non-specialists] had lower activity across the whole song period relative to non-conformists, indicating that their sensitivity to popularity was aiso related to the degree to which they may hâve paid attention to the musical semantics of the song itself, which includes chord progressions, rhythm and lyrics.” (Berns et al., 2010, p. 2695). To be commercially successfui, a song must be largely promoted and be in the “spirit of time,” that is being a formatted product. Crain and Tollison (1997) show that duration and tempo of popular songs hâve changed over time. Further research could include expanding the scope of our analysis to incorporate additional years in order to measure the possible shifts in musical characteristics over time.
67 Finally, our analysis of the musical content of songs is of course incomplète. Variables that describe mélodie complexity, content of the lyrics, etc. are not included. Finding a method to measure them objectively might be another way to extend the présent research.
References
Anderson, T., 1984. An Introduction to Multivariate Statistical Analysis, New York: Wiley.
Attali, J., 1977. Bruits, Paris: Fayard.
Banerjee, A. V., 1992. “A simple model of herd behavior," The Quarterly Journal of Economies, 107(3), 797-817.
Berns, G. S., Capra, C. M., Moore, S., & Noussair, C., 2010. “Neural mechanisms of the influence of popularity on adolescent ratings of music,” Neuroimage, 49(3), 2687- 2696.
Bikhchandani, S., Hirshieifer, D., Welsh,!., 1998. “Learning from the behavior of others: Conformity, fads, and informational cascades,” Journal of Economie Perspectives, 12(3), 151-170.
Biliboard Magazine, http://www.billboard.com.
Bowling, D., Gill, K., Choi, J., Prinz, J., Purves, D., 2010. “Major and minor music compared to excited and subdued speech,” Journal of the Acoustical Society of America, 127(1), 491-503.
Coase, R.H., 1979. “Payola in radio and télévision broadeasting,” Journal of Law and Economies, 22(2), 269-328.
Grain, W. M. & Tollison, R. D., 1997. “Economies and the architecture of popular music,” Journal of Economie Behavior and Organization, 901, 185-205.
Crowder, R.G. (1985). “Perception of the major/minor distinction: III. Hedonic, musical, and affective discriminations,” Builetin of the Psychonomic Society, 23(4), 314-316.
Curtis, M. E., Bharucha, J. J. (2010). “The minor third communicates sadness in speech, mirroring its use in music,” Emotion, 10, 335-348.
Hennion, A. ,1983. “The production of success: an anti-musicology of the pop song,” Popular Music, 3, 159-193.
68 Jusiin, P. N., and Laukka, P., 2003. “Communication of émotions in vocal expression and music performance: different channels, same code?” Psychological Bulletin, 129, 770-814.
Kellaris and Kent, 1993. “An exploratory investigation of responses elicited by music varying in tempo, tonality, and texture,” Journal of Consumer Psychology, 2, 4, 381- 401.
Parker, M., 1991. “Reading the charts - Making sense with the hit parade,” Popular Music, 10(2), 205-217.
Rogers, W., 1993. “Régression standard errors in clustered samples,” Stata Technical Bulletin, 13, 19-23.
Rossman, G., Cjiu, M. M. , Mol, J., 2008. “Modeling diffusion of multiple innovations via multilevel diffusion curves: Payola in pop music radio,” Sociological Methodology, 38, 201-230.
Salganik, M. J., Dodds, P. S., & Watts, D. J., 2006. “Experimental study of inequality and unpredictability in an artificial cultural market,” Science, 311(5762), 854-6.
StrobI, E. and Tucker, C. 2000. “The dynamics of chart success in the U.K. pre- recorded popular music industry,” Journal of Cultural Economies, 24, 113-134.
Tobin, J., 1958. “Estimation of relationships for limited dépendent variables,” Econometrica, 26, 24-36.
Williams, R., 2000. “A note on robust variance estimation for cluster-correlated data,” Biométries, 56(2), 645-646.
69 Chapter 3 The Formation of the Canon of Baroque Music"^
’ A slightly different version of this chapter has been published under the title “The Réputation of Baroque Composers 1790-2000” (Ceulemans, 2010).
70 1. Introduction
The purpose of this paper is to analyze how baroque composers fared over time. The end of the baroque era is traditionally associated with Johann Sébastian Bach’s death in 1750. Forty years later Gerber published the first biographical dictionary of music.^ There is thus a period of 200 years (1800-2000) which can be analyzed to follow the évolution of the réputation of baroque composers. Réputation is measured by the length of entries devoted to composers in seven music dictionaries that span these 200 years. Ranking of composers or musiciens is a délicate question. Rating is as old as music. Legend holds that the Pythian Games at Delphi began as a musical contest before the athletic events were introduced in 582 BC. In Athens, musical events occurred during the Panathenaic Games. According to Young (2004, p.100), the lyre playing with singing was the most popular musical contest. Only the winners of chariot races earned larger prizes. AIso during the Baroque era compétitions were held: In 1708, Domenico Scarlatti and Georg Friedrich Handel competed as organists in Rome. Today musical contest are organized everywhere with the participation of prestigious musicians in the juries, and books rating the greatest composers are published. And in a way, even if this is not their main goal, dictionaries implicitly rank composers and musicians by devoting a certain number of Unes to each of them. On the other hand, there is no accounting for taste and therefore rankings could be regarded as meaningless. Several composers and musicians stand up against musical contests. Such was the case of Bêla Bartok who wrote that compétitions were made for horses not for musicians. Claude Debussy was very critical of the famous Prix de Rome, a French contest founded in 1663, which rewards painters, sculptors, architects and musical composers. He qualified the prize as “uselessly grotesque...It could even become dangerous, since the favors attached to it force us to listen to lots of bad music, and familles anxious for the
^Matthias Schacht compiled the first biographical dictionary devoted solely to musicians, Musicus danicus, in 1687, though it was not published until 1928. During the 18*^ century four dictionaries were prepared before Gerber’s Lexicon (1790-1792) but none of them are as important as Gerber’s. That is the reason for which Coover (2001, p. 313) writes that Gerber’s Lexicon stands as “the first independent dictionary of musical biography and a model for many successors.”
71 children’s future find encouragement in this prize, at a time in which there are too many engineers.” (Claude Debussy, 1962, p. 43). The fact that Debussy was a winner of the Prix de Rome shows ail the ambiguity about musical contests and rankings. Another criticism of musical compétitions is made by Flores and Ginsburgh (1996). They show that factors other than the ability of the musician interfère in the judgment. The final ranking in The Queen Elisabeth Musical Compétition, a prestigious Belgian compétition for violin and piano, is influenced by the day in which the candidates appear. Those who perform during the last days hâve a higher probability of being ranked among the first than do those who appear in the beginning. Since the order in which musicians perform is drawn at random in the beginning of the compétition, a completely exogenous factor that has nothing to do with talent or previous results, plays a rôle in the judgment and has conséquences on the final ranking. Intuitively, one of the main factors that makes rankings subjective is the time in which they are established. For example, musicologists often remind us that Johann Sébastian Bach was neglected during the second half of the 18**^ century. Nowadays nobody would dare not ranking Bach among the 10 greatest composers of ail times. But do composers’ réputations really change from one génération to another or is there a consensus through time and musicologists? Our results indicate that such a consensus exists and that musicologists rely on their predecessors’ work.
2. Method and data
We started with ail composers listed in Sadie’s Companion to Baroque Music (1990). We kept ail the composers to whom Gerber’s Historisch-biographisches Lexicon der Tonküstler {^790-^792), Riemann’s Musik-Lexikon (Riemann & Einstein, 1929) and The New Grove’s Dictionary of Music and Musicians (2001) devoted at least 19, 17 and 292 lines, respectively (which represent 3 percent of the maximum number of lines devoted to the first ranked composer in each dictionary). This led to a list of 366 artists for which we went through Fétis’ Biographie Universelle des musiciens (1837-1844), Baker’s Biographical Dictionary of Musicians (1900), Die
72 Musik in Geschichte und Gegenwart, aiso known as MGG (1949-1986) and Honegger’s Dictionnaire de la musique - les hommes et leurs oeuvres (1970). The list eventually includes artists who are better known as theorists, instrumentalists, instrument makers or even as political personalities (Frederick II of Prussia for example) than as composers, but ail of them hâve written scores. The list aIso includes composers whose affiliation to the baroque movement is still discussed by scholars (Luily and J.-P. Rameau for example^). However, ail listed musiciens worked entirely or partially during the Baroque era. The choice of Gerber was quite obvious. It is the first independent biographical dictionary of music (Coover, 2001) and his Lexicon formed the base of several biographical dictionaries of the lO**^ century. “Gerber’s work is still indispensable, especially concerning personalities of the 18*^ century" (Wessley, 2001, p. 685). Fétis was aiso an undisputable choice. Though Fétis is known to deliver some Personal opinions, his Biographie Universelle des musiciens was a landmark in the discipline of musicology. “In the long view of history, his methodology is much more significant than his Personal biases. The Biographie Universelle achieves a leading position in the tradition of scholarly music historiography” (Coover, 2001, p. 314). The work (actually, the second édition that came out in 1869) even enjoyed a reprinted version in 1972. At the cross-roads of the 19‘^ and 20*'^ century, the first éditions of Grave (1890) and of Baker (1900) were two important contributions. The Grave is a more general work dealing with musicians, instruments and the musical terminology. But the 1890 édition is not at ail as complété as it is today. Baker instead focused on biographies. We went through both dictionaries but Baker’s work appeared more appropriate. Riemann is “one of the founders of modem musicology and the pre-eminent scholar and teacher of his génération” (Hyer, 2001, p. 362). One of his famous works is the Musik-Lexikon (1882) which was often reedited and even translated into several languages (French, Danish, English and Russian). After Riemann’s death in 1919, Alfred Einstein made several révision of the Musik-Lexikon. The one used here is the édition published by Riemann and Einstein in 1929.
^See the dicussion by Maillard (2004) on the French musical style in the l?'*' century.
73 In France, after World War II, the most significant dictionary is the Dictionnaire de ia Musique compiled by Marc Honegger in 1970. Finally, the MGG and the New Grave Dictionary are, at this time, the most extensive musical encyclopedias of music. Obviousiy, other dictionaries exist and could hâve been used. Therefore the study may dépend on the authors chosen. But this remark remains valid whatever the Works selected. Scholarship guided our list of authors and we tried to select undisputed works. We aiso paid attention to publication date (so that the 200 years are carefully spanned: 1790, 1840, 1900, 1930, 1950, 1970, 2000) and nationality of the authors (three German, two French, including Fétis, who was born in Belgium, but was active in France, one American, Baker, and one British, Grove’s Dictionary). We are reasonably convinced that a different set of reference works would lead to results that are roughiy identical in terms of the canons that are emerging. However, including Italien dictionaries such as Basso (1983-1990) or Bertini (1814-1815) in our database could probably change some of the conclusions, esprcially on national biases.
3. Results
3.1. Does a Consensus Exist between Musicologists?
We start by ranking composera for each musicologist. The rankings are simpiy derived from the length of entries in each dictionary. Though this is a common method, it has often been criticized. Woods (1911, p. 573) argued that the “adjective method,” which uses évaluative statements rather than line counts, is a better way of “measuring small différences” between personalities. Fortunately our goal is not to judge whether Bach is a greater composer than Handel but rather to analyze the emergence of canons. In other words, it is not so important to know whether Bach is first or second but it matters to know whether he is among the top ten or not. Therefore we think that in our case the “space count method” is sufficient. Two lists, one restricted, the other complété are used in the analysis. The “restricted” list consists of the 54 composers (we aimed at 50, but ended up with 54 given ties) ranked first by each musicologist, which leads to a total of 123 names. The “complété” list contains ail 366 composers.
74 We start with a rough measure of convergence between dictionaries using the restricted list. Table 1 gives the number of composers common to ail pairs of dictionaries. This number is aiways greater than or equal to 27, except for the pairs Grove-Gerber and Grove-Fétis, and shows that 50 percent or more of the top 54 names are common between musicologists. In most cases, this number decreases as the time intervals increase. This can be seen very clearly with Gerber (first row in Table 1): The number of common names decreases monotonically from Fétis (33 names in common with Gerber) to Grove (only 20 common names are left).
Table 1 Number of Common Composers between Pairs of Dictionaries (Restricted List) Gerber Fétis Baker Riemann MGG Honegger Grove (1790) (1840) (1900) (1930) (1950) (1970) (2000)
Gerber - 33 32 31 27 27 20 Fétis 33 - 39 35 32 27 24
Baker 32 39 - 39 34 32 29
Riemann 31 35 39 - 35 33 29
MGG 27 32 34 35 - 32 31
Honegger 27 27 32 33 32 - 34
Grove 20 24 29 29 31 34 -
As a complément and a refinement of the first measure, we calculate the (usual) corrélation coefficients between ail pairs (rank corrélations could not be calculated since we use a restricted list and therefore some ranks are missing). The results are shown in Table 2. Ail coefficients are positive and only three are not significantly different from zéro (Fétis-Gerber, Fétis-Honegger and Riemann- Honegger) at a 10% probability level. This indicates that, in general, some consensus exists on the ranking. This is even more relevant when we look at the complété list of names. In that case we can compute Spearman rank corrélations (Siegel & Castellan, 1988, pp. 235-244), shown in Table 3. Ail coefficients are positive and statistically different from zéro at a 0.001 probability level. With the exception of Gerber, corrélations are very large. Note that corrélations are often larger with the immédiate predecessor or successor. For example, Baker-Fétis and Baker-Riemann are the largest coefficients for Baker.
75 Table 2 Corrélation Coefficients between Ranks (Restricted List) Gerber Fétis Baker Riemann MGG Honegger Grove (1790) (1840) (1900) (1930) (1950) (1970) (2000)
Gerber - 0.34 0.39 0.50 0.24 0.39 0.33 Fétis 0.20 - 0.40 0.48 0.39 0.22 0.41 Baker 0.34 0.48 - 0.49 0.40 0.28 0.49 Riemann 0.38 0.42 0.47 - 0.49 0.20 0.38 MGG 0.33 0.27 0.44 0.41 - 0.35 0.25
Honegger 0.24 0.33 0.49 0.50 0.53 - 0.38
Grove 0.24 0.28 0.42 0.45 0.57 0.47 - Note: Every row considers the 54 artists ranked first by the art historian whose name appears in the row and gives the corrélation with the ranking of the historian whose name appears in the column. Therefore, the table of corrélation coefficients is not symmetric.
Table 3 Spearman Rank Corrélations (Complété List) Gerber Fétis Baker Riemann MGG Honegger Grove (1790) (1840) (1900) (1930) (1950) (1970) (2000)
Gerber - 0.41 0.35 0.17 0.20 0.24 0.19 Fétis 0.41 - 0.71 0.55 0.45 0.49 0.47 Baker 0.35 0.71 - 0.67 0.53 0.53 0.55
Riemann 0.17 0.55 0.67 - 0.59 0.68 0.54 MGG 0.20 0.45 0.53 0.59 - 0.65 0.63
Honegger 0.24 0.49 0.53 0.68 0.65 - 0.63
Grove 0.19 0.47 0.55 0.54 0.63 0.63 -
Figure 1
Spearman rank corrélations as a function of time interval between two dictionaries
0.8 1
c 0.7-
= 0.6 - (0 g 0.5 - O 0.4 - g 0.3 ■ S 0.2 - w 0.1 -
0 -- —1----- 1------1------1------1------1---- 1------1------1------1------1 0 20 40 60 80 100 120 140 160 180 200 220 time interval
76 Figure 1 shows that coefficients tend to decrease as the time interval between two dictionaries increases. When time intervals are larger than 100 years the coefficients are strictiy smaller than 0.5. On the contrary when time intervals are equal to or smaller than 40 years the coefficients are strictiy larger than 0.58. Next we compute the Kappa statistic (Siegel & Castellan, 1988, pp. 284-291) for the restricted list. This coefficient measures the overall agreement between ail seven dictionaries. The statistic varies between 0 (no agreement) and 1 (perfect agreement). The Kappa statistic is equal to 0.247 and is highiy significantly different from zéro at the one percent probability level (z-value is equal to 11.55). This shows that the agreement is positive, even if it is not very large. Using the complété list of composers, we can compute a statistic that has a similar interprétation: Kendall’s coefficient of concordance W (Siegel & Castellan, 1988, pp. 262-272). It is equal to 0.56 (xMest = 1431.66) and is statistically different from zéro at the 99.9% confidence level. (Note that 14^ is equal to 0.648 (xMest = 1419.12) and is statistically different from zéro at a 0.001 probability level if we do not take into account Gerber’s dictionary). This is a very strong resuit but the coefficient of concordance assumes that the rankings are made independently from each other. The Spearman rank corrélations and the dynamic model, estimated later, indicate that this may not be the case. Musicologists seem to be influenced by their predecessors. Therefore we submitted the Spearman rank corrélation matrix to a principal component analysis which shows that the first two principal components accounts for 53% and 25%, respectively, of the total variance. The first component is negatively correlated with Gerber and positively with ail other musicologists. This resuit combined with the Spearman rank corrélations and Kendall’s coefficient suggests that Gerber was a Pioneer in musicology but that Fétis really set the tone for biographical dictionaries. Our results are consistent with those of Ginsburgh and Weyers (2006) who did a similar analysis for Italian painters of the Renaissance. The Kendall coefficient, the Kappa statistic and the corrélations coefficients are comparable in magnitude. The main différence seems to be found in the time pattern. In our case, the Spearman rank corrélation coefficients (Table 3) and common composers (Table 1) tend to decrease as the time interval increases. This is not the case in the study by Ginsburgh and Weyers.
77 3.2. Persistence of Composers’ Réputations
The measures discussed so far indicate that a general consensus exists between musicologists. In this section, we test for the persistence of composers’ réputation by going into the details of names. Twenty-four composers are présent at least once in a top ten list. The évolution over time varies in many ways. See Appendix Table A. Johann Sébastian Bach is the only musicien to be aiways présent. Handel, Monteverdi and Rameau appear among the top ten in Fétis (1840) and never leave afterwards. The same holds for Schütz but he entered the top ten list a little later with Baker (1900). Tartini’s faring is completely different. He was very appreciated until 1930 (Riemann) but he later disappeared from the first ranks. Marc-Antoine Charpentier was really neglected. His rankings are very bad before 1970 (Honegger). Today, he is considered to be as important as his long time celebrated countrymen: Rameau and Luily. Finally, the surprise cornes from Antonio Vivaldi, nowadays, one of the most popular baroque composer, who is ranked only twice among the top ten (in MGG and Grove). This would probably hâve been different if Italiens were présent in the list of musicologists. We aiso need to look at the restricted list of names to hâve a broader view on the évolution of composera’ réputation. Thirteen composers hâve aiways been ranked among the first 54. This group clearly passes the test of time and can be seen as part of the canon. Their names (and ranks attributed by each musicologist) appear in Table 4. We could add Corelli, Frescobaldi and Purcell to this list since they are aiways présent after Gerber. Antonio Vivaldi and Wilhelm Friedrich Bach are poorly ranked by Fétis and Baker only, suggesting that they were moderately appreciated during the 19*^ century. Table 4 shows that only four nations made it to the history of baroque music: Germany, France, England and Italy. Looking at the canon of baroque music, there are six German, six Italian, two French, and two English composers. This domination is aIso verified when we look at the complété list of composers where 35 percent are Italian, 31 percent German, 14 percent French, 11 percent English and only 9 percent are coming from other countries.^
^We follow Grove (2001) for nationalities.
78 Table 4 The Baroque Canon (Restricted List) Name Gerber Fétis Baker Riemann MGG Honegger Grove (1790) (1840) (1900) (1930) (1950) (1970) (2000) Bach (j.-s.) 10 3 2 1 1 1 1 Handel 15.5 1 1 2 2 3 2 Schütz 30 31 7 4 4 5 3 Telemann 6 30 13 8 3 6 4 Monteverdi 45.5 10 9 3 7 7 5 Rameau 14 2 4 5 6 2 6 Scarlatti (a.) 44 9 14.5 12 5 30 9 Bach (C.p.E.) 31 29 16.5 18.5 11 14 11 Masse 3 8 12 27.5 13 23 12 Luily 11 4 5 9 17 10 19 Pergolesi 21 27 6 11 14 16 33 Tartini 7 7 8 7 42 19 39 Keiser 15.5 19 32 14 12 46 40.5 Note: The numbers are ranks. In case of ties, each composer is assignée! the mean rank. This explains non integer ranks.
Obviousiy, Germany and Italy are the leaders in baroque music, but why is this so? A possible explanation is given by Elias (1991) and Baumol and Baumol (1994) who both point to the “competing noble courts” hypothesis. Elias (1991, pp.43-46) argues that the relatively large number of musicians in Germany and Italy during Mozart’s period was due to political fragmentation. In those countries, dozens of courts and cities competed for prestige, and thus for musicians. This rivalry involved the création of numerous musical positions. On the contrary, in France and England, musical positions were concentrated in Paris and London as a resuit of centralization. There were no competing courts that could rival with the King’s power, weaith and prestige. A musicien fallen from the King’s favour had nearly no chance of finding another employer. Baumol and Baumol (1994) made the same hypothesis for the Holy Roman Empire and the Habsburg possessions at the end of the 18*^ century. Their secondary argument is that “in the Empire, as in England and France, the rising prosperity of the eighteenth century - the first stirrings of the industrial révolution and associated developments such as the rise in the weaith and position of the small body of the bourgeoisie as consumers of culture -- contributed to the demand that underlay the création of a free market in musical composition” (Baumol & Baumol, 1994, p. 172). Scherer (2004, pp.130-132) makes a distinction between what he calls the “strong form” and the “weak form" of the “competing noble courts” hypothesis. The
79 strong form implies that only Germany, the most politically fragmented State, qualifies. The weak form ignores imperia! oversight from Vienna and includes the four main remnants of the Holy Roman Empire (Germany, Italy, Austria and Czechoslovakia). Scherer tests both hypothèses: the strong form is discredited but the weak form delivers a positive resuit. He aiso suggests that the rising middie-class prosperity (measured by Gross domestic product, adjusted in constant 1990 dollar terms) had a positive impact on the demand for musicians, confirming the intuition of the Baumols. Furthermore, Scherer demonstrates that “magnet cities” (primarily Paris and London) attracted, thanks to advantageous wages, musicians from ail over Europe: increasing the population of musicians in those countries. Vaubel (2005) aIso studies the “competing noble courts” argument. However, he focuses on Baroque music'^ and he restricts his analysis to four countries: Italy and Germany, as “fragmented States” and France and England, as “centralized States.” Using the average duration of employment as a proxy for compétition, he shows that the most famous Italian and German composers of the Baroque period shifted between employers significantly more often than their French and British counterparts. Thus, political fragmentation has promoted compétition on the demand side. This explains, at least partiy, that the rise of baroque music took place in Germany and Italy rather than in France or England. Vaubel’s results combined with Scherer’s “magnet cities” argument fit perfectiy well our database and results. Germany and Italy offer a large number of baroque musicians due to the “competing noble courts,” but England and France manage to get their share in baroque music thanks to their “magnet cities” (London and Paris). Another aspect of the nationality is its effect on the ranking of composers. How does nationality influence musicologists? In the econometric section of the paper, we will test statistically whether we observe a national bias, though some interesting facts can be pointed out here. Sixty-three percent of English composers obtain their best ranking in an English dictionary. Morley, Arne, Boyce, Greene, Weelkes, Lawes, Flumfrey, Clarcke, Wilbye, Eccles, and Avison seem to benefit from this national bias since their rankings are high in both Baker and Grove. No other
^Baumol and Baumol (1994) and Elias (1991) focused on Mozart’s period. Scherer (2004) worked with a database of 646 composers bom between 1650 and 1850.
80 dictionary gives them as much importance. Fifty-four percent of French composers obtain their highest ranking in French dictionaries. This includes Mondonville, Mauduit, Nivers, Lambert, Rameau, and Guillemain. Finally, 70% of German composers are ranked highest by a German dictionary. Quite surprisingly, the percentage of French composers favoured by Fétis and Honegger rises to 80% when Gerber’s ranking is ignored. It is hard to say if the increase is due to the fact that Gerber especially appreciated the French répertoire or to a lack of information concerning English and Italian musicians. Indeed, Gerber neglected some important Italian composers such as Frescobaldi, Corelli, Gabrielli, Stradella, and Durante, ail “discovered” by Fétis who appeared to like Italian music, placing seven transalpine musicians in his top 15 group. At the end of the 19*^ century, Baker introduced 12 composers who appear for the first time in the top 54 group. Among this list, we find several English composers (Morley, Byrd, and Boyce) but aiso some musicians whom will remain in good positions during the 20‘^ century (Domenico Scarlatti, Couperin, Schein, and Campra). Despite the specificities of each musicologist, 29 composers (13 composers who are aiways présent in the 54 top list, see Table 4; the 16 composers who are spécifie to the 19“^ century are listed in Appendix Table B) are ranked among the 54 first in Gerber, Fétis, and Baker. Some names like Porpora or Praetorius are still appreciated today; others hâve completely disappeared (e.g., Schrôter and Sorge). Concerning the 20‘^ century, only 8 composers are présent among the 54 first places in Riemann, MGG, Honegger and Grove. Frescobaldi, Purcell, and Corelli where aiready présent in both Fétis and Baker. So, only five composers are really breaking through during the past century (Domenico Scarlatti, Buxtehude, Vivaldi, Schein, and Wilhelm Friedman Bach). With some irony, one could argue that this ranking looks like a family reunion. Indeed, after his father Johann Sébastian and his brother Cari Phillip Emanuel, Wilhelm Friedman Bach enters the circle of the greatest composers. And the same holds for Domenico Scarlatti, son of Alessandro. Consequently, only 19 out of 366 artists (16 artists corne from the canon of baroque music of Table 4, to which we add Corelli, Frescobaldi, and Purcell who are présent among the top 54, except in Gerber) can be considered as the canon of the baroque era. Vivaldi, Buxtehude and Domenico Scarlatti enjoy a very good réputation during the 20“^ century but will this situation persist in the future or will they return to oblivion? As we show now, if they disappear, they will likely do so only gradually.
81 3.3. Do Evaluations follow an Autoregressive Process?
In the previous sections, we hâve shown that there exists a consensus between dictionaries and that some composers are appreciated by ail générations of musicologists. This implies that, at least to some extent, dictionaries rely on their predecessors. We aiso suggested that the nationality of composers could hâve an influence on the judgment made by musicologists. In this section, we use a dynamic panel data model to test whether or not, and to what extent, these two variables can explain the importance of entries in the sequence of dictionaries. The data used is the complété list of 366 composers in six dictionaries. Since we use a lagged variable in our model and that the publication dates of Honegger and MGG overlap,^ we kept the MGG since it is more important than Honegger. Given that entries in the dictionaries are of different lengths, we normalize by Z, computing shares T, = , where U is the number of Unes devoted to composer i in
t dictionary t and U is the total number of Unes devoted to ail 366 composers in dictionary t. Note that a share of 0 (i.e., the composer is not présent in the dictionary) is not considered as a missing value. Therefore we hâve a data set that consiste of a balanced panel. The auto-regressive réputation équation can be written
5,, =c + or5,(,_,) + ^'A:„+f,+^,_ (1) where Sit is the square root of lu (i = 1, 2, ..., 366; t = 2, 3, ..., 6). We use the square root transformation which smoothes the data and makes them less sensitive to outliers.® However, since this transformation is less common we will compare the
®The MGG was initially published between 1949-51 and 1968. But an important supplément of two volumes, used in our database, was published in the seventies. The first volume was in print in 1973 and the second in 1979. Finally, the registerwas available in 1986. ®The presence of outliers in our data is not surprising. For example, Bach is considered as one of the greatest composers of ail times. We should therefore expect that the length of the entries devoted to him will be much larger than those devoted to the average baroque composer. The logarithmic transformation is more frequently used, but here it would cause problems with the many zéros. This could be overcome by multiplying ail the values by 1000 and replacing the zéro values by ones. Doing that would generate a distribution that is negatively skewed, which is not consistent with the original data. Instead, the square root transformation skews in the same proportion as the log transformation, but to the right. Secondly, the square root transformation does a better job in smoothing the data. Because of the zéros, the log transformation produces a range of values for which the density is zéro. This is not the case for the square root transformation. Ideally, one could use the Box-Cox method to
82 results with those obtained by a model in levais and in logarithms (which résulta in a loss of data but avoids the problems discussed above). The Xit vector contains six dummy control variables that capture the possible national biases of musicologists. Each of these takes the value 1 if composer and musicologist hâve the same nationality and zéro otherwise. Some values are missing since the model is estimated with lags. Their effects will be discussed later in a static version of the model. The parameters to be estimated are c, a and p. The spécification aiso contains a permanent but unobservable composer-specific effect £j and an error term Uit. The are assumed to hâve finite moments and in particular E(Uit)= E(Uit Uis)= 0 for t s. We first estimate (1) using ordinary least-squares (OLS). Results are shown in Table 5. The coefficient picked by the lagged variable shows clearly that musicologists rely on their predecessors’ work and that there is some persistence in the réputation of composers.
Table 5 Composer Réputation Equation (OLS) Coefficient Standard Error Lagged composer 0.664 0.016 réputation Fétis bias 0.030 0.029 Baker bias 0.092 0.032 Riemann bias 0.095 0.020 MGG bias 0.055 0.020 Grove bias 0.134 0.032 Intercept 0.139 0.008 Numberof observations: 1830 f?-squared: 0.503 Adjusted R-squared: 0.501
However, in this dynamic model, ordinary least squares results are biased and inconsistant even if the Uu are not serially correlated. As shown by Baltagi (1995, pp. 125-126), this problem cornes from the fact that since sitis a fonction of £i, Si(M) is aIso a fonction of £i. Therefore, Si(n), the right-hand regressor in (1), is correlated dérivé the optimal exponent (instead of 'A) when estimating by ordinary least squares. This would hardiy be tractable using the Arellano and Bond (1991) method.
83 with the error term. This leads to an upward bias in the coefficient of the lagged dépend variable (see Nerlove, 1967 and Hsiao, 1986, pp. 76-78). The solution is to use a method suggested by Arellano and Bond (1991). In order to get a consistent estimate of a as N—and T is fixed (N=366 and T=6 in our case), we différence équation (1) to eliminate the individual effects (Si). This leads to
-^,(,-1) = c + «(■^,(,-1) -^,(,-2)) + P V., -^,(,-1)) + (^,7 -(2)
Now, Si(t-2) is a valid instrument for (Sit-Si(t.i)) since it is uncorrelated with (Uit - Ui(n)) and highiy correlated with (Si(t-i)-Sj(t-2)). Assuming, that the Xj are strictiy exogenous,
Xii, Xi2,...,Xi6 are aiso valid instruments (For more details, see Arellano and Bond or Baltagi, pp. 126-132). The results of the one-step Arellano and Bond (1991) estimator are shown in Table 6.
Sargan’s test checks the null hypothesis that the instrumental variables are uncorrelated with the residuals and can, therefore, be considered as valid instruments. This is the case here. Arellano and Bond (1991) argue, however, that the one-step estimator tends to over-reject the null hypothesis in presence of heteroskedasticity, and suggest using a two-step estimator, which is more robust. The results of Table 6 confirm that the model is well specified, and that the instruments are valid. The drawback of this estimator is that the standard errors tend to be biased downward in small samples. Therefore, we prefer to use the one-step results for inference on the coefficients. The coefficient picked by the lagged variable is equal to 0.2. As expected this is much smaller than in the OLS régression but the value is still highiy significantly different from zero.^ We aIso estimated équation (1) using the share in level (lit) and in logs (In(lit)) instead of square roots (Sit). Results are very similar: Ail coefficients are significantly different from zéro and positive. The lagged réputation coefficient is equal to 0.205 in the case of ht and 0.196 in the case of In(ht). A régression using robust standard errors leads to comparable results.
^ Since section 3.1 suggests that Gerber was a pioneer in musicology but that Fétis really set the tone for biographical dictionaries, we aiso run the régression without “Gerber”. See Appendix Table C. The “Lagged composer réputation" coefficient is equal to 0.22, which is very close to the value (0.20) when Gerber is included (Table 6).
84 A coefficient equal to 1 would suggest that each dictionary is a copy of the previous, while 0 would mean that successive évaluations are completely independent from each other. In our case, the importance devoted to a composer in a dictionary is for 20% a réminiscence of the previous dictionary. This confirms that musicologists rely on their predecessors’ work but the effect vanishes fast, suggesting that they are mostly influenced by their direct predecessor. This resuit is consistent with the Kappa statistic discussed above and the rank corrélation coefficients which tend to decrease as time between dictionaries increases.
Table 6 Composer Réputation Equation (Arellano and Bond Estimators) One-Step Two-Step Coefficient Standard Coefficient Standard Error Error Lagged composer 0.204 0.030 0.177 0.041 réputation Baker bias 0.072 0.028 0.078 0.030 Riemann bias 0.113 0.017 0.114 0.015 MGG bias 0.126 0.017 0.133 0.013 Grove bias 0.141 0.033 0.135 0.033 Intercept -0.007 0.003 -0.007 0.003 Sargan test 16.37 (15) 14.18 (15) [P-value= 0.36] [P-Value= 051] Number of observations; 1464 Number of groups: 366______
The relationships between past and présent judgments are cast in a statistical framework by Simonton (1998), who distinguishes five possible configurations:
(a) Transhistorical stability: The relationship among the judgments is relatively constant. They may vary over time but not in a systematic way, suggesting that each génération applies largely the same set of criteria in evaluating works.
(b) Exponentiel decay: Judgments made by a génération are influenced by their direct predecessors. The consecutive évaluations are ruied by a first-order autoregressive process, implying that the corrélations between past and présent judgments should decrease rapidiy over time and eventually disappear (zéro corrélation).
85 (c) Graduai attrition: Corrélation between past and présent évaluations décliné in a graduai and linear way. Given sufficient time it may disappear.
(d) Cyclical fashion: Judgments fluctuate (quasi-)periodically overtime.
(e) Complété transhistorical instability: There is no consistency in the criteria applied by consecutive générations. Corrélations between successive évaluations will be close to zéro.
Our analysis shows that configurations (c), (d) and (e) do not appiy to baroque music. The results of the dynamic model suggest that configuration (b) of exponential decay applies. The coefficient of 0.2 picked by the lagged variable suggests that musicologists are almost only influenced by their direct predecessor. However, Tables 1, 2 and 3 show that corrélations do not fall to zéro, suggesting that “exponential decay” does not fit exactiy either. Most results are reasonably consistent with a combination between transhistorical stability and exponential decay. How do our results compare with what others hâve found? Rosengren (1985) suggests that the stability of literary famé is only due to an autoregressive process. Simonton (1991) argues that posthumous réputation of philosophers, American presidents, artists and classical composera is ruied by a single-factor model. In other words, réputation is explained by the genius of each personality and not by an autoregressive process. Our results lie somewhere between Rosengren’s and Simonton’s conclusions. The lagged réputation coefficient is small but statistically significant. A possible interprétation of this is that the réputation of famous composera is driven by a single latent variable (genius), while the réputation of less talented composera is the resuit of autoregression, but our data do not make it possible to disentangle the two effects.
86 3.4. National Bias and Composer Spécifie Effects
The dynamic model does unfortunately not make it possible to estimate nationality biases for ail musicologists and it is impossible to estimate composer- specific effects, since the dummy variables used to represent them are constant over time. We therefore estimated the following (static) model:
= c + y,P, + j/j/* + y,t- +e,+ v, _ (3) where the s» and Xjt are defined as above, while the Pi, n and t are dummy variables with value equal to 1 if composer i wrote opéra or stage music (pi), an oratorio or a passion (n), or some theoretical work (ti). They are equal to zéro otherwise. We can now test whether the fact that a composer wrote an opéra and/or an oratorio and/or a theoretical essay will increase his importance in dictionaries. Equation (3) aiso contains a permanent, but unobservable, composer-specific effect £i and an error term Uit and is estimated using the full list of 366 composers in ail seven dictionaries. The parameters can be estimated using either a fixed or a random effects model. The question is whether £i should be treated as a random variable (hence the name “random effect”) or a parameter to be estimated (“fixed effect”). According to Mundlak (1978) and Wooldridge (2002, pp.251-252), the key issue is whether or not £i is uncorrelated with the observed explanatory variables (xu, pi, n and f). If one can assume that there is no corrélation, then a “random effect” model should be used. Otherwise, one should turn to “fixed effect” estimation. The Hausman test (1978) can be used to discriminate between the two possibilities. Under the null hypothesis (Ho = no corrélation), both estimâtes are efficient. Otherwise, random effect estimâtes are inconsistent. Ho should be rejected if there is a statistically significant différence between the fixed and the random estimâtes. Rejecting Ho means that one suspects the presence of corrélations and this is interpreted as evidence against the random effects assumption. In our case, there is no statistically significant différence between the two types of estimâtes (p-value = 0.91 ) and we cannot reject Ho so it is safe to use random effects. The estimation results of équation (3) are shown in table 7.
87 Table 7 Composer Réputation Equation (Random-effects) Coefficient Standard Error Opéra 0.06 0.03 Oratorio 0.15 0.03 Theory 0.13 0.03 Gerber bias 0.07 0.02 Fétis bias 0.06 0.03 Baker bias 0.10 0.03 Riemann bias 0.09 0.02 MGG bias 0.12 0.02 Honegger bias 0.19 0.03 Grove bias 0.17 0.03 Intercept 0.32 0.02 R-squared: within = 0.06 between = 0.15 overall = 0.12 Number of observation = 2562
Number of groups = 366
Ail coefficients are positive and statistically significant at the .05 probability level. The “opéra” effect is small. The oratorio effect is larger but this may be due to Johan-Sebastian Bach who composed passions and oratorios instead of opéras. We checked for this possibility by removing Bach from our database. The results hardiy change and having composed an “oratorio” still appears to be more important. The “theory” effect is aiso significant and suggests that musicologists give more importance to artists who were aiso musical theorists or critics. Ail musicologists are biased and give more importance to their composera with whom they share their citizenship: ail the “bias” coefficients are positive and significantly different from zéro. This bias is even larger in more recent dictionaries.
4. Discussion and Conciusions
We show that some consensus exista between seven musicologists working at different times and coming from different countries. Statistics such as the Kendall coefficient, the Kappa statistic and Spearman rank corrélations demonstrate this in a quantitative way. Should these results be interpreted as a social consensus over aesthetic beauty and genius?
88 Though, our data do not make it possible to give a clear-cut answer, it would seem unreasonable to believe that réputations are only due to social consensus. Several composers enjoyed a good réputation during the 19'^ century but were forgotten or at least neglected after World War II. So why should social consensus Work for Johann-Sebastian Bach or Jean-Philippe Rameau and not for Johann- Adolph Scheibe or Georg-Andreas Sorge? Once the analysis is extended to the full set of 366 composers, we are able to show that musicologists rely on their predecessors’ work and may therefore conclude that cultural tradition matters. Indeed, one could argue that even if canonical composers such as Handel or Bach were no longer appreciated they would still be highiy ranked by musicologists. Tradition would be too strong to make such composers disappear. The dynamic model validâtes such arguments and shows that composers do not appear or disappear suddenly. They rather move up or down smoothly, but relatively fast, since the coefficient picked by the lagged variable is quite small. The econometric model shows that musicologists are aiso influenced by the nationality of composers: A German musicologist or dictionary favors German composers. This is even the case with an encyclopédie work such as the Grave though it is written by a large number of world renowned musicologists with different nationalities, but the editors still play a large rôle. We did not take into account mis- or reattributions of works. Most composers were very prolific and it is doubtfui than one or the other change of attribution could hâve a large impact on réputations. The issue could, however, explain the variations in the rankings of some composers. For example, the authenticity of several works of Carissimi is still not settled. Jones (2001, p. 137) warns that “any attempt to define Carissimi’s output must be considered provisional.” Thus, the quantity and the quality of his work could be interpreted differently across musicologists. This may perhaps explain why Carissimi’s ranking goes up and down through time. Finally, Lang and Lang (1988) argue that posthumous réputation of artists is linked to surviving examples of their original works. They distinguish four different possibilities facilitating the survival of an artists’ réputation: (a) Lifetime efforts: An artist’s own effort to distribute and preserve his créations, but aIso to keep a record that facilitâtes future identification of his work; (b) Links to posterity: After the death of the artist, relatives, close friends or admirers dedicate part of their time to the perpétuation of his memory; (c) Networks and circles: Any link to important artistic
89 and literary circles or to a political and cultural elite fosters the posthumous visibility of an artist. This aiso includes family links. For example being the son of an important artist will influence the son’s réputation; and (d) Figures in a landscape: Revitalized interest in one or another characteristic of an artist’s life or work can boost posthumous réputation. Though Lang and Lang discuss etchers, this could clearly hold for baroque composera as well. For example very little of Charpentier’s music was published during his lifetime. But, as mentioned by Hitchcock (2001, p. 507), Charpentier carefully gathered his manuscripts in numbered sheaves (cahiers) and bequeathed them to a nephew, Jacques Edouard, who was a printer and bookseller. In 1709 Jacques Edouard published a collection of 12 petits motets, and eventually sold ail Charpentier’s manuscripts to the King’s library in 1727. Without his own efforts and the help of his nephew, it may well be that Charpentier would not be part of the canon, no matter how talented he was.
Référencés
Arellano, M., & Bond, S. (1991). Some tests of spécification for panel data: Monte Carlo evidence and an application to employment équations. Review of Economie Studios, 58, 277-297.
Baker, T. (1900). Biographical dictionary of musicians. New York: G. Schrimer.
Baltagi, B. H. (1995). Econometric analysis of panel data. Chichester: John Wiley & Sons.
Basso, A. (1983-1990). Dizionario enciclopedico universale délia musica e dei musicisti. Torino: Unione Tipografico, Editrice Torinese.
Baumol, W. & Baumol, H. (1994). On the économies of musical composition in Mozart's Vienna. Journal of cultural économies, 18, 171-198.
Bertini, G. (1814-1815). Dizionario storico-critico degli scrittori di musica e de’piu celebri artisti di tutte le nazioni si antiche che moderne. Palermo.
Ceulemans, C. (2010). The réputation of Baroque Composera 1790 - 2000, Empirical Studios of the Arts, 28(2), 223-242.
Coover, J. B. (2001). Dictionaries & encyclopaedias of music. In The new Grove dictionary of music and musicians. Edited by Stanley Sadie (vol. 7, pp. 306-320). London: Macmilan.
90 Debussy, C. (1962). Trois textes. Considérations sur le Prix de Rome au point de vue musical. Une déclaration "arrangée" de Debussy en 1904. A propos de Charles Gounod. Revue de musicologie, 46, 41-48.
Elias, N. (1991). Mozart. Sociologie d’un génie. Translated by Jeanne Etoré et Bernard Lortholary. Paris : Editions du Seuil.
Fétis, F.-J. (1837-1844). Biographie universelle des musiciens et bibliographie générale de la musique. Bruxelles: Méline, Cans et Cie.
Flores Jr, R., & Ginsburgh, V. (1996). The Queen Elisabeth musical compétition. How fair is the final ranking. Journal of the Royal Statistical Society (The Statisticien), 45, 97-104.
Gerber, E.-L. (1790-1792). Historisch-biographisches Lexikon der Tonkünstler. Leipzig: J.G.I. Breitkopf.
Ginsburgh, V., & Weyers, S. (2006). Persistence and fashion in art Italian Renaissance from Vasari to Berenson and beyond. Poetics, 34, 24-44.
Grove. 2001. The new Grove dictionary ofmusic and musiciens. Edited by Stanley Sadie. London: Macmilan.
Hausman, J. A. (1978). Spécification tests in econometrics. Econometrica, 46.2, 403-426.
Hitchcock, H. W. (2001). Charpentier, Marc-Antoine. In The new Grove dictionary of music and musiciens. Edited by Stanley Sadie (vol.5, pp. 504-509). London: Macmilan.
Honegger, M. (1970). Dictionnaire de la musique - les hommes et leurs œuvres. Paris: Bordas.
Hsiao, C. (1986). Analysis of panel data. Cambridge: Cambridge University Press.
Hyer, B. (2001). Riemann, Hugo. In The new Grove dictionary of music and musiciens. Edited by Stanley Sadie (vol. 21, pp. 362-366). London: Macmilan.
Jones, A. V. (2001). Carissimi, Giacomo. In The new Grove dictionary of music and musiciens. Edited by Stanley Sadie (vol. 5, pp. 135-150). London: Macmilan.
Lang, G. E., & Lang, K. L. (1988). Récognition and renown: the survival of artistic réputation. The American Journal of Sociology, 94, 79-109. MGG (1949-1986). Die Musik in Geschichte und Gegenwart: Allgemeine Enzykiopàdie der Musik. Edited by Friedrich Blume. Kassel; Basel : Bârenreiter- Verlag.
Maillard, J.-C. (2004).Le style musical français au XVIIe siècle : doutes et certitudes. Dix-septième siècle, 224, 446-460.
91 Mundlak, Y. (1978). On the pooling of time sériés and cross section data. Econometrica, 46, 69-85.
Nerlove, M. (1967). Experimental evidence on the estimation of dynamic économie relations from a time sériés of cross-sections. The Economie Studies Quarterly, 18:3, 42-74.
Riemann, H., & Einstein, A. (1929). Hugo Riemann Musik-Lexikon. Berlin : Max Hesse.
Rosengren, K. (1985). Time and literary famé. Poetics, 14, 157-172.
Sadie, J. A. (1990). Companion to baroque music. Berkiey: University of California Press.
Scherer, F.M. (2004). Quarter Notes and Bank Notes. The Economies of Music Composition in the eighteenth and nineteenth Centuries. Princeton, NJ: Princeton University Press.
Siegel, S., & Castellan Jr, J. (1988). Nonparametric statistics for the behavioural sciences. New-York: McGraw-Hill.
Simonton, D. K. (1991). Latent-variable models of posthumous réputation: a quest for Galton’s G. Journal of Personality and Social Psychology, 60, 607-619.
Simonton, D. K. (1998). Fickie fashion versus immortel famé: Transhistorical assessments of créative products in the opéra house. Journal of Personality and Social Psychology, 75:1, 198-210.
Vaubel, R. (2005). The rôle of compétition in the rise of Baroque and Renaissance music. Journal of cultural économies, 29, 277-297.
Wessley, O. (2001). Gerber, Ernst Ludwig. In The new Grove dictionary of music and musicians. Edited by Stanley Sadie (vol.9, 685-686). London: Macmilan.
Woods, F. A. (1911, April 14). Historiometry as an exact science. Science, 33, 568-574.
Wooldridge, J. M. (2002). Econometric analysis of cross section panel data. Cambridge, Massachusetts: The MIT Press.
Young, D. C. (2004). A brief history of the Olympic games. Malden, MA: Blackwell Publishing.
92 Appendices
Table A Composers Ranked at Least Once among the Top 10 Name Rank Presence Gerber Fétis Baker Riemann MGG Honegger Grave among the (1790) (1840) (1900) (1930) (1950) (1970) (2000) top ten Bach (J.- S.) 10 3 2 1 1 1 1 7 Handel 15.5 1 1 2 2 3 2 6 Monteverdi 45.5 10 9 3 7 7 5 6 Rameau 14 2 4 5 6 2 6 6 Schütz 30 31 7 4 4 5 3 5 Telemann 6 30 13 8 3 6 4 5 Luily 11 4 5 9 17 10 19 4 Purcell (H.) 65.5 49 3 6 22 4 10 4 Tartini 7 7 8 7 42 19 39 4 Mattheson 4 5 27 13 10 33.5 66 3 Scarlatti (A.) 44 9 14.5 12 5 30 9 3 Charpentier 211 105.5 77 119.5 52 8 8 2 Masse 3 8 12 27.5 13 23 12 2 Steffani 9 43.5 18 10 43 37.5 55 2 Vivaldi 32.5 157.5 67.5 36 9 21 7 2
Buxtehude - 169.5 125 33.5 8 20 15 1 Couperin 81.5 188 20.5 56 38 9 18 1 Frédéric II 8 - 148.5 125 41 123.5 170 1 Martini 24 6 27 36 82 43 78 1 Pepusch 1 34 33.5 21 108 82.5 83 1 Pergolesi 21 27 6 11 14 16 33 1 Porpora 29 20.5 10 32 90 49 57 1 Quantz 2 25 23.5 17 51 114.5 95 1 Stôlzel 5 41.5 96 98 33 97 177 1 Note: means that the composer has no entry in the dictionary (Frédéric II in Fétis for example).
93 Table B Composers Présent among the First 54 in Gerber, Fétis and Baker, But Not Afterwards Name Gerber Fétis Baker (1790) (1840) (1900) Pepusch 1 34 33.5 Quantz 2 25 23.5 Mattheson 4 5 27 Steffani 9 43.5 18 Scheibe 13 36 47.5 Schrôter 18 18 36.5 Graun 19 22 30 Serge 20 37 30 Martini 24 6 27 Philidor 26 16 19 Porpora 29 20.5 10 Praetorius 38 28 23.5 Kerll 40 47 43.5 Marcello 42 13 52.5 Geminiani 43 38.5 52.5 Froberger 47.5 35 39
Table C
One-Step Two-Step Coefficient Standard Coefficient Standard Error Error Lagged composer 0.22 0.05 0.20 0.05 réputation Baker bias 0.08 0.03 0.07 0.04 Riemann bias 0.12 0.02 0.12 0.02 MGG bias 0.12 0.02 0.14 0.01 Grove bias 0.12 0.03 0.11 0.03 Intercept 0.32 0.02 0.31 0.03 Number of observations: 1098 Numberof groups: 366
94