Cover versions as an impact indicator in popular music: A quantitative network analysis

José Luis Ortega

doi:10.1371/journal.pone.0250212

. 2021 Apr 19;16(4):e0250212. doi: 10.1371/journal.pone.0250212

Cover versions as an impact indicator in popular music: A quantitative network analysis

José Luis Ortega ^1,^2,^*

Editor: Nicola Perra³

PMCID: PMC8055012 PMID: 33872337

Abstract

In contemporary popular music, covering or adapting previous songs signals a tribute or reworking of popular hits, which implies acknowledgement of the original musicians. This connection can be interpreted as a form of musical impact among artists. A network graph with more than 106k artists and 855k cover versions extracted from the web site SecondHandSongs was created. The objective is to explore the shape of this network, identify the most relevant artists according to different impact measurements and to visualize connections between music genres. This analysis is done from a longitudinal perspective with the aim of understanding how cover versions can inform us about the history of the contemporary popular music. Results show that the number of covers by artist is skewed distributed, diminishing gradually since the 1950s. Different network metrics have allowed to identify the most covered (weighted indegree), the most influential (PageRank) and the most crossover artists (weighted betweenness centrality). The network graph also shows that genre affinity is the main criterion for covering songs between artists, language being the second. Remakes from other genres reflect that Jazz and Pop/Rock are the most influential because they emerge stronger and form the core of their respective sub-networks. Cover songs describe two cycles. In a first phase from 1900s to 1950s, dominated by Jazz and Vocal artists, the covers are more frequent and associated with the notion of reworking (e.g. jazz standards); in a second stage, since the 1950s, when the Pop/Rock emerges, cover songs are less common and seen as tribute.

Introduction

Popular music has become one of the most important cultural phenomena in the 20th and 21st centuries [1]. The invention of broadcasting media, new recording systems and the establishment of the middle class have favoured the mass consumption of recorded music, making it one of the most fruitful entertainment businesses today [2, 3]. Since the early 20th century, artists and bands have emerged over the years writing and performing thousands of popular songs that reflect the worries, hopes and feelings of society. Local traditions (Flamenco, Samba, Celtic music, etc.) and new technological advances (amplifiers, synthesisers, etc.) have led to the proliferation of genres and styles that have enriched musical expression and enabled the transmission of new musical forms to the next generation [4].

One element that reveals the hereditary nature of popular music is the cover, adaptation or remake of songs. A cover is a new recording of a piece of music originally written or performed by another musician. These acknowledgements lead to links between artists and illustrate the musical influences of soloists or bands, while assessing the impact of artists’ careers in specific musical communities.

The advent of the World Wide Web ushered in the appearance of specialized music databases that began collecting and publishing information about artists, discs, songs and styles [5]. The main purpose of these services was to promote disc catalogues and to preserve intellectual rights [6]. Allmusic, Discogs and MusicBrainz are just a handful of open music resources that provide detailed information about commercial music recordings. SecondHandSongs (secondhandsongs.com), with close to one million covers, is the most complete database collecting information about cover songs and their performers. All these platforms provide a great volume of data that can foster our knowledge about the contemporary popular music from a quantitative perspective.

This study therefore attempts to address this research gap analysing how cover songs between artists allows us to know the evolution of the popular music over a century, identifying their most influential artists and describing the emergence of genres and styles.

Related research

Until recently, music studies have been distinguished by the use of qualitative methods such as case studies and textual criticism. The appearance of many web-based music databases and advances in music data processing and analysis have led to the emergence of a new quantitative paradigm in this discipline [7, 8]. From this perspective, Rose et al. [9] used digitized documents to study the rise and fall of music printing in the 16th and 17th centuries, tracking specific composers and observing changing trends in genres. Park et al. [10] studied the co-authorship network of western classical music composers, finding preferential attachment as a growth mechanism. And McAndrew and Everett [11] analysed encounters between British classical composers to detect hotspots of creativity.

Within the scope of contemporary popular music, many papers have studied collaboration patterns between Jazz musicians [12], music industry agents [13] and Finnish heavy metal artists [14]. Other studies have focused on lyrics content. Napier and Shamir [15] examined the sentiment of 6.1k songs from 1951 to 2016, finding that anger and fear have increased significantly. Ruth [16] observed that love is the most relevant topic in German rock music. However, the most common research theme is audio features analysis to observe changes and tendencies. Collins [17] studied the influence of early electronic dance music through audio content analysis. Mauch et al. [18] analysed the harmonic and timbral properties of 17k Pop and Rock songs to detect the evolution of styles. Klimek et al. [19] compared the musical styles of eight million music albums to test whether new symbols are introduced by current elite members. Finally, Youngblood [20] was based on Whosampled site to test the conformity bias in the cultural transmission of music samples.

Nevertheless, an important gap in the quantitative studies on the contemporary popular music has been the total absence of analysis on cover versions. In most of the cases, they have been observed from a theoretical perspective, figuring out the cultural meaning of this type of songs. For Plasketes [21], covers are a symptom of standardization and loss of originality, while others [22, 23] defend the importance of covers in the transmission of styles and ideas. In this context of cultural studies, many papers have explored the meaning of cover versions using specific cases studies. Doktor [24] analysed cross-gender adaptations of the song Satisfaction. Chu and Leung [25] studied the role of cover songs in the decline of Cantopop. The only exception is the work of Anderson [26] who quantified the ‘adaptations’ wave in French Pop.

Objectives

The principal objective of this study is to visualize and analyse the cover songs network between artists in order to describe its topology and evolution throughout the time. In this form, it attempts to understand how the use of cover versions has changed over the decades, what artists have been more influential in each moment and how the different music trends have been associated between them. To that end, several network indicators (weighted indegree, PageRank and weighted betweenness centrality) are proposed to assess distinct forms of impact on the artists. Three research questions were formulated:

What are the topological characteristics of the cover songs network and how do they evolve over time?
Which musical artists are the most influential in the network and how have they reached that position? What information does each network metric provide about the impact?
Can cover songs inform us about the context of each music genre? Is it possible to observe the evolution of the preferences for each genre?

Methods

Data sources

SecondHandSongs

This database records data about covered songs and their performers and was created by Bastien De Zutter, Mathieu De Zutter and Denis Monsieur in 2003. Since then, the database has been supplied by a group of volunteer editors that manually add and update the information about songs and artists. As a result, the database coverage may not be exhaustive or uniform. In January 2020, the site reported 788k covers, 143k artists and 96k original songs. According to SecondHandSongs, original performers are those that perform, record or release a song for the first time, and a cover is when subsequent artists perform, record or release that same song again. This definition excludes songwriters, who could be the original authors, but not the original performers. This database is limited to covered songs only and does not provide a comprehensive list of recorded songs by artists [27].

Allmusic

An exhaustive music database released on the Web in 1994 and owned by TiVo Corporation. At the last estimate it provided 30 million tracks and three million albums [28]. Allmusic’s most valuable information is the classification scheme. Each artist is arranged according to 21 genres and over 100 styles. Styles are not considered as sub-categories of genres, therefore, they are labels or keywords freely assigned to an artist.

Data extraction

Web scraping techniques were used to obtain data from SecondHandSongs and Allmusic. WebQL software was used to write ad-hoc scripts that crawl and retrieve specific information from each site. Details about songs, covers and artists were retrieved from SecondHandSongs. Each performer is identified by a sequential number. Then, numbers from 1 to 175000 were automatically generated to retrieve the list of covered songs and artists by each performer. (This process was carried out in January 2020 and the data are accessible in: https://osf.io/6j2ru/).

Because SecondHandSongs only includes information about covered songs, Allmusic was used to broaden information about artists. Concretely, the complete number of songs recorded by each artist and the genres and styles of each performer. The total number of songs allows us to measure an artist’s production, while the genres and styles enable us to group artists. As with SecondHandSongs, sequential codes from mn0000000001 to mn0003600000 were queried to retrieve artist profiles in Allmusic, obtaining 176k.

Data cleaning

SecondHandSongs assigns an id to each cover, but not to a song because the title of the cover could be different from the original (i.e. translations, reworking of lyrics). Links between artists were therefore established according to the number of times an artist covers another, regardless of the title of the song. In the event that an artist covers the same song more than once, it was counted as a different cover.

The name of the artist or band was used to match SecondHandSongs and Allmusic records. This element was carefully cleaned to avoid mismatches between slight name variations. Thus, some characters such as ampersands and punctuation marks were removed or replaced by normal letters (Simon & Garfunkel by Simon and Garfunkel, “Big Mama” Thornton by Big Mama Thornton). In other cases, names with different variations were normalized (Booker T. and the Memphis Group’s by Booker T. and the M.G,’s, Johnny Burnette Trio by Johnny Burnette Rock and Roll Trio); synonymy between different artists (Tom Jones, 60s vocal artist and Tom Jones, 50s lyricist) was solved adding a digit to the name.

Medleys pose another problem because they include parts from more than one song. SecondHandSongs includes this type of song but does not properly identify which performer is the author of each different part. These songs (11,112) were removed because they were not significant (1%) and could introduce a lot of noise.

When a song is traditional or the original performer is unknown, SecondHandSongs assigns it to a generic unknown (20,428). These covers were removed because there is not a connection between artists. In addition, SecondHandSongs joins the code of two or more artists when they perform together (e.g. Sinatra and Bono, Elvis Presley and the Jordanaires). These cases were duplicated, assigning the same song to each artist. After all these transformations the final sample included 106k artists and 855k cover songs.

Social network analysis

Covering a song implies a cultural and meaningful connection between the original artist and subsequent performers. The best way to analyse these connections is through Social Network Analysis, the branch of sociology specializing in the structural analysis of social relationships. Social network metrics are used to identify different type of impact:

Weighted Indegree centrality (kⁱⁿ) measures the weighted number of acs incident to a node [29]. In our case, it is equal to the total number of cover songs of one artist. This indicator measures the accumulated impact of an artist in the music community.
Weighted Outdegree centrality (k^out) works in the opposite direction and counts the weighted number of arcs directed out of a node. That is, the total number of covers performed by an artist. This metric allows us to assess artists’ ability to praise and acknowledge their influences.
PageRank (PR) is the original algorithm used by Google to rank their search results [30]. It is used to assign a relative weight to each node, according to the value of the linking nodes, which transfer a proportional value to the cited node. It was calculated taking weights into account. The PageRank of an artist is an indicator of the importance of the performers that cover their songs. Thus, an artist whose songs are covered by performers with many versioned songs is more important than artist covered by unknown performers.
Weighted betweenness centrality (b) is an indicator that counts the number of weighted shortest paths that pass through a node [31]. The importance of this metric is that it allows the detection of nodes at an intermediate position between distant nodes and groups. According to song remakes, weighted betweenness centrality scores artists that cover or are covered by other artists belonging to different groups. To a certain extent, it identifies crossover artists that inspire or are inspired by different artists from different styles, countries and epochs.

Results

General characteristics of the covers network

A network graph was constructed with 106,095 nodes (artists) and 855,650 arcs (cover versions) extracted from SecondHandSongs. The very low density of the graph (d<0.01%) shows few connections between artists, suggesting that covering songs is highly selective. This interpretation could be confirmed by the cumulative distribution of covers performed (outdegree) and received (indegree) by artist. They follow a power-law distribution (Fig 1), with an exponent (α) of 2.882 for outdegree and 2.937 for indegree. R package poweRlaw was used to create the cumulative distribution functions (CDF) and the maximum likelihood estimation (MLE) method was used to fit the distributions. This type of distributions suggests a considerable imbalance in the number of covers performed/received by artist. For example, the outdegree distribution shows that only 7% of artists perform no cover and that 48% of them cover the songs of at least two artists. These percentages indicate that covering songs is relatively common. However, the indegree distribution reveals that only 22% of the artists have been covered, whereas just 100 artists (0.1%) encompass 41% of all the versions. This high disproportion reveals that while covering songs is a usual way of doing music, these versions are concentrated on songs belonging to a very reduced number of artists who make up the star system of contemporary popular music.

In order to study the evolution of the network, artists were classified by the decade in which they started their musical career. During this process, 4,983 artists were discarded because they did not have that information. Then, several cumulative networks by decade were built. This aggregate approach is because the cover versions are made of current or precedent artists and each network rests on the previous one. The growing pattern of the network is not constant (Fig 2). The average of covers (weighted average degree) strongly increases from 1910s (k = 1.07) to 1950s (k = 18.18), but since 1960s (k = 16.85) this average gradually descends until 2010s (k = 9.68) (Table 1). This same occurs with the clustering coefficient, which increases up to.093 in 1940s and then slowly drops to.066 in 2010s. This result suggests that during the 1910s-1950s, the network is growing in number of covers, with artists that tend to sing previous songs. However, since 1960s, the pattern changes and the artists that joined the network cover less songs.

Table 1. Network indicators evolution by decade.

Network indicators	1910s	1920s	1930s	1940s	1950s	1960s	1970s	1980s	1990s	2000s	2010s
Nodes	106	1353	2732	4647	9735	19152	27737	35502	53398	80033	101112
Arcs	82	5260	18974	47800	138949	256180	344714	419488	555932	698403	792509
Average degree	.774	3.888	6.945	10.286	14.273	13.376	12.428	11.816	10.411	8.726	7.838
Weighted average degree	1.066	4.894	9.035	13.274	18.183	16.85	15.578	14.752	12.94	10.805	9.678
Average path length	3.07	3.707	3.954	3.899	3.97	4.098	4.341	4.493	4.696	4.888	4.968
Clustering coefficient	0.046	0.079	0.087	0.093	0.09	0.086	0.083	0.081	0.074	0.068	0.066

Open in a new tab

The covering of songs between generations of musicians can illustrate with more detail this changing pattern. Fig 3 shows how artists that started their career in the 1900s-1950s cover more frequently songs of 1920s artists. This fact would indicate that during the first middle of the 20th Century, artists from the 1920s exercised a great influence on their partners until the 1950s. However, from this decade, it is observed that artists start to cover songs mainly from the 1950s-1960s, suggesting a generational change in the preferences of the artists. This pattern is maintained until the 2000s, when a new change is surmised. One of the possible explanations of these variations would be the emergence of new music genres in specific moments (see below, Genres and styles section).

Top covered artists

The cumulative distribution of covers has showed that just a small portion of artists are profusely covered. This section analyses who are these figures and how they have evolved. Supplementary material S1 File includes the full list of artists and impact indicators. Table 2 presents the ranking of the first ten most versioned artists and bands. Because an artist may cover different songs from the same performer, weighted indegree is used to compute the strength of this association. The purpose is to show who shapes the core of the network and what features they have in common. The number of songs (supplied by Allmusic) contains the total number of musical track recorded and released by artist, including different recordings of the same song. Due to this and the skewed distribution of cover songs (Fig 1), this data was not used as normalization variable. The Beatles is the most covered group with 16,825 versions, followed by the Jazz composer Duke Ellington with 10,564 and the singer Bing Crosby with 7,011. All the performers belong to the English-speaking culture, revealing the strong and early development of the music industry in these countries [32]. Another particularity is that, apart from The Beatles, Bob Dylan and Elvis Presley, they all started their music careers in the 1920s, 1930s and 1940s, which could indicate that there is a time-dependence in the covering of songs. In addition, these old artists belong to genres with a long-standing tradition such as Jazz and Vocal, while The Beatles, Bob Dylan and Elvis Presley come from Pop/Rock.

Table 2. Top 10 artists by number of covered songs (weighted indegree).

Artist	Active period	Songs	Main genre	Style	Weighted indegree	Weighted outdegree
The Beatles	1960s	2096	Pop/Rock	British Invasion	16825	112
Duke Ellington	1920s–1970s	7447	Jazz	Big Band	10564	262
Bing Crosby	1920s–1970s	8864	Vocal	Traditional Pop	7011	676
Bob Dylan	1960s–2010s	2378	Pop/Rock	Folk/Country Rock	5986	215
Thelonious Monk	1940s–1980s	1526	Jazz	BeBop	4860	76
Elvis Presley	1950s–1970s	5175	Pop/Rock	Rock & Roll/Roots	4692	516
Frank Sinatra	1930s–1990s	8052	Vocal	Traditional Pop	4556	820
Miles Davis	1940s–1990s	4323	Jazz	BeBop	3676	193
Fred Astaire	1920s–1980s	1724	Vocal	Show tunes	3621	109
Judy Garland	1920s–1960s	3278	Vocal	Cast Recording	3468	195

Open in a new tab

It includes number of songs from Allmusic and weighted outdegree.

A more detailed view about the evolution of the impact of these artists is plotting the number of covered songs throughout the decades (Fig 4). Performers from the 1920s–1930s depict a constant growth, headed by Duke Ellington and Bing Crosby, pillars of the Jazz and Vocal music respectively. It is also possible to appreciate the strong emergence of Thelonious Monk and Miles Davis, two jazz figures representative of the Bebop in the 1940s. However, the most significant increase is seen in The Beatles and Bob Dylan, who start their career in the 1960s and in just three decades they climb to the first and fourth position respectively. In general, this result suggests that as the cover network evolves over time, the impact of new figures is greater and faster. This would cause that the covering of songs is more and more concentrated in few artists, increasing extremely their impact over time. In addition, the sudden appearance and success of The Beatles and Bob Dylan would be representative of the generational change in the 1960s, as previously observed (Fig 3).

However, weighted indegree shows a cumulative impact that increases throughout the decades. A way to know the influence of artists in each moment is by calculating its PageRank. This metric shows the importance of an artist according to who their songs are covered by, rewarding covers from popular performers (who in turn are also very versioned) and undervaluing remakes from little-versioned musicians. Now, Bob Dylan climbs to the second position and new Pop/Rock classical figures, such as David Bowie, The Rolling Stones and Chuck Berry, emerge (Table 3). Other pioneers such as Prince’s Band, an original jazz band and author of some of the most important Ragtime jazz standards, and Ethel Merman, a central figure of 1930s’ musical theatre, acquire more relevance.

Table 3. Top 10 artists by PageRank.

Artist	Active period	Songs	Main genre	Style	PageRank
The Beatles	1960s–1970s	2096	Pop/Rock	British Invasion	0.00887
Bob Dylan	1960s–2010s	2378	Pop/Rock	Folk/Country Rock	0.004559
Bing Crosby	1920s–1970s	8864	Vocal	Traditional Pop	0.003301
Duke Ellington	1920s–1970s	7447	Jazz	Big Band	0.002846
David Bowie	1960s–2010s	1715	Pop/Rock	Hard Rock	0.002725
The Rolling Stones	1960s–2010s	1359	Pop/Rock	Hard Rock	0.002607
Prince’s Band	1900s–1920s	7	Jazz	Ragtime	0.002574
Chuck Berry	1950s–2010s	1222	Pop/Rock	Rock & Roll/Roots	0.002556
The Velvet Underground	1960s–1990s	386	Pop/Rock	Experimental Rock	0.002339
Ethel Merman	1930s–1980s	698	Vocal	Traditional Pop	0.002236

Open in a new tab

Fig 5 illustrates with more detail how the influence of these top artists occurs in different periods. We observe thus that the Prince’s Band acquires great prominence in the initial period of the 1900s-1920s, being a key figure in the development of the early jazz. Then, some figures of the jazz (Duke Ellington) and vocal music (Bing Crosby and Ethel Merman) dominate the network in the 1920s-1950s, coinciding with the height of the Jazz Age. However, since the 1950s, all these artists experience a decline in favour of the new Pop/Rock stars. These new performers (The Beatles, Bob Dylan, The Rolling Stones), raised from the 1960s, increase their importance throughout the following decades until today.

Another way to select important artists on the music scene is through weighted betweenness centrality, which measures musicians’ tendency to cover songs from distant and different artists, either by genre, language or epoch, and indicates the artist’s degree of crossover. However, this metric cannot be interpreted according to the direction of the arc, making it impossible to know whether the position of an artist is due to the cover received or made [33]. To aid interpretation, the weighted indegree and weighted outdegree are also considered (Table 4). The first positions of the ranking (1st–6th) are occupied by performers who are covered far more frequently than they cover, which suggests that these artists (Frank Sinatra, Bing Crosby, Bob Dylan, The Beatles) are bridges that influence later performers from a broad range of spheres. Conversely, as recognized composers and arrangers of numerous popular hits, James Last and Ray Conniff covered more songs than they were covered. They are, then, bridges that are influenced by classical artists from different fields. Acker Bilk follows the same behaviour because he was a representative of Trad Jazz, a revival of Dixieland and New Orleans Jazz.

Table 4. Top 10 artists by weighted betweenness centrality.

Artist	Active period	Songs	Main genre	Style	Weighted indegree	Weighted outdegree	Weighted Betweenness centrality
Frank Sinatra	1930s–1990s	8052	Vocal	Traditional Pop	4556	820	348112988
Bing Crosby	1920s–1970s	8864	Vocal	Traditional Pop	7011	676	224973559
Bob Dylan	1960s–2010s	2378	Pop/Rock	Folk/Country Rock	5986	215	211717237
The Beatles	1960s–1970s	2096	Pop/Rock	British Invasion	16825	112	132659702
Elvis Presley	1950s–1970s	5175	Pop/Rock	Rock & Roll/Roots	4692	516	124172981
Johnny Cash	1950s–2000s	4946	Country	Country Gospel	1224	468	98411237
Acker Bilk	1950s–2000s	1202	Jazz	Trad Jazz	120	282	39169103
Ray Conniff	1930s–1990s	1805	Easy Listening	Orchestral/Easy Listening	503	755	33349589
David Bowie	1960s–2010s	1715	Pop/Rock	Hard Rock	2360	91	32278473
James Last	1960s–2010s	2757	Easy Listening	Mood Music	79	453	21715877

Open in a new tab

It also includes weighted indegree, weighted outdegree and total songs.

Genres and styles

This section attempts an in-depth exploration of the musical impact of adaptations between genres over time, by observing when new trends emerged and what their main influences were.

Fig 6 depicts the full network of artists (nodes) connected by covers versions (arcs). Supplementary material S1 Fig includes a full-size version of the Fig 6. The colour of the artists represents their main genre, the size of nodes the number of recorded songs and the label size the number of covers received. Due to the great size of Pop/Rock genre, it was subdivided in styles (Rock & Roll/Roots, Pop/Rock, British invasion, Soft Rock, Psychedelic/Garage, Hard Rock, Country Rock, Foreign Language Rock, Art-Rock/Experimental, Punk/New Wave, Heavy Metal and Alternative/Indie Rock) to observe groups and connections between them. A first impression reveals that the artists tend to be grouped by genre, resulting in compact clusters such as Jazz, with 73% of internal covers, Country, with 43%, or Blues, with 37%. This high cohesion would indicate conservative genres with long traditions and few external influences. Closely connected genres are also visible: Jazz and Vocal, with a 24% of Vocal singers covered by Jazz musicians and 20% in reverse. The reason could be that these genres were very popular from the 1910s–1940s, when many Jazz bands (e.g. The Dorsey Brothers, Duke Ellington’s Orchestra) included singers or crooners (e.g. Bing Crosby, Louis Armstrong) who interpreted their pieces [34]. Less perceptible is the influence of language on the connection between artists due mainly to the predominance of English-speaking musicians (63%). The lower right section features a compact cluster of Brazilian musicians around Antonio Carlos Jobim and a less dense cluster of French singers near to Edith Piaf and Serge Gainsbourg.

We created a genre network connected by the number of covers between them (Fig 7). The arc weights were normalized per each node, and any weight below 0.1 was removed. This is the same as only including arcs where at least 10% of all the cover songs performed (outdegree). For example, Jazz adapts 40% of Jazz pieces and 24% of Vocal songs and is an inspiration for 65% of Jazz and 10% of Vocal songs. This pruning criterion allows to select only the most important relationships among genres. Based on these relationships, five modules were identified according to the Louvain method (Resolution = 0.75, modularity = 0.279). The lower half of the picture includes all pre-Pop/Rock genres; early urban music (i.e. Jazz, Vocal, Stage & Screen, etc.) is shown in violet; and rural genres, headed by Country, in dark green. The upper half gathers the post-Pop/Rock genres. The core of Pop/Rock is highlighted in green and linked to the genres directly associated with it: R&B (Rhythm and Blues) and Blues and contemporary rock led by Alternative/Indie Rock. The blue cluster groups music genres from languages and cultures other than English (i.e. Latin, International and Foreign Language Rock), again revealing a language pattern in versioned songs. Thus, the genre network seems to indicate that there are two clear groups of genres: a cluster of old styles headed by Jazz and Vocal, and all the genres resulting from the advent of Pop/Rock.

Fig 8 depicts the proportion of cover songs in each decade grouped by the music genre of the covered artists. For instead, artists from the 1900s cover 55% of songs belonged to Vocal artists and 17% to Classical authors. Whereas artists appeared in the 2010s prefer to cover songs from Pop/Rock (39%) and Vocal (11.7%) artists. Notice that Pop/Rock styles have been merged to reduce lines and make clear the growth of the Pop/Rock music. This picture allows to appreciate how the music preferences change over time and the impact of new genres. Thus, the image shows how the Vocal and Jazz music describe high covering percentages until the 1960s, the moment in which Pop/Rock and their styles strongly emerge to dominate the music scene until today. It is interesting to notice the high peak of cover versions of R&B artists in the 1960s, which could indicate the great adaptation of black music by white performers through the Rock and roll and British rhythm and blues [35, 36].

Discussion

The use of song adaptations between artists as an impact measure has made it possible to observe how the influence and appreciation of artists are distributed and how this has affected the creation and evolution of music genres. Firstly, the versioning of songs is a very unbalanced phenomenon, in which a small percentage of artists (less than 1%) are repeatedly versioned, while the vast majority (78%) are not. This unbalanced environment could be due to the low density (d<0.01%) and poor frequency of cover songs (9.8 covers per artist against 79 songs per artist), which would determine that musicians prioritise to cover the most popular and prestigious artists [20].

Perhaps, one of the most interesting results regarding the evolution of the whole network is that the average of covers (weighted average degree) does not constantly grow, but since the 1950s decade, it gradually descends. This inflection point suggests an important transformation in that time, in which post-1950s generations of musicians stop doing remakes of previous songs. This perception is reinforced when cover versions between generations are observed (Fig 2). Then, it is possible to appreciate that in the 1910s-1950s period, artists have a predilection for covering songs from 1920s musicians, but this trend is broken since the 1950s, when new performers prefer to cover songs from 1960s artists. These two marked periods coincide with the raising of the Jazz Age (1920s) and the Rock Revolution (1950s), which suggests that the appearance of these music genres would imply something more than just different tastes [18, 37, 38]. Both music genres could be consequence of deep transformations in the way of making music, including their perception on cover songs. During the 1900s-1950s period, when jazz bands and vocal singers dominated the music scene, the covering of songs was more frequent because they would compete in the performance of popular hits. The frequent use of standards in Jazz could reflect a greater predilection for the interpretation and improvisation [39]. However, with the advent of the Pop/Rock in the 1950s, the cover song lost prominence because musicians compete now in originality and creativity, considering cover versions as an appropriation [21, 35, 36]. In this context, cover songs tend to be used as a tribute to significant artists that serve as inspiration [40, 41].

This evolution in the use and meaning of cover versions can be also appreciate in the top covered artists. Both weighted indegree (Fig 3) and PageRank (Fig 4) have showed that musicians from the pre-Pop/Rock era (e.g. Duke Ellington, Bing Crosby) accumulate a constant number of covers, while subsequent performers from the 1960s (e.g. The Beatles, Bob Dylan) quickly overtake the number of covers of the previous generations. Taking into account that the covering of songs descends since 1950s, the fast success of these figures could illustrate an extreme concentration of the impact in special artists thereafter. PageRank, that avoids this cumulative effect, describes with more detail how Jazz and Vocal performers slowly lose their influence from the 1950s in favour of the new Pop/Rock stars, who rapidly increase their influence from the 1960s onwards. Furthermore, these network metrics have demonstrated great utility for describing different types of impacts such as cumulative (weighted indegree), influence (PageRank) and crossover (weighted betweenness), improving our appreciation on the role of different artists in the evolution of popular music.

The network graph visualization reports that genre affinity is the main criteria for covering songs among musicians, secondary reasons being language and nationality (e.g. Brazilian bossa nova, French chanson). As we have previously discussed, Jazz and Pop/Rock form the axes on which contemporary popular music lies [18, 37, 38]. The genres network shows that Jazz and Vocal are the most influential genres for pre-Pop/Rock styles, while Pop/Rock is the touchstone for subsequent genres. From a longitudinal perspective, the percentage of cover versions by genre in different decades (Fig 8) evidenced again how pre-Pop/Rock genres, mainly Jazz and Vocal, are the most covered genres until the 1950s. The moment in which Pop/Rock and associated styles grow up to be the most dominant genre at the beginning of the 21^st Century.

Limitations

However, the results of this study are limited to the sources used. Although the site SecondHandSongs collects most covers, it is not always completely comprehensive and certain minority artists and local styles (i.e. flamenco, raï, afrobeat) may not be well represented. Genre classification is according to Allmusic.com and some categories, such as the arc between Pop and Rock, are debatable. These limitations can influence the results relative to the relationship among genders.

Another point of discussion could be the normalization of covers by the number of recorded songs because one could argue that as more songs are released by an artist more chance to be covered. However, these normalizations in skewed distributions could produce some artefacts, in which one-hit wonders would accumulate a disproportionate impact with regard to the real influence of these artists (see an example in Supplementary materials S2 File). An added problem is that the number of total songs in Allmusic is equivalent to the total number of different recordings, even when these different sessions are about the same song. This introduces a bias against the most popular singers, because they have more published material than others (i.e. greatest hits, live discs, hidden sessions) and their relative impact could be reduced.

Conclusions

The obtained results let us to conclude that the cover versions network experiences a strong growth of remakes during the first decades until the 1950s, and from that moment the average of cover versions by artist gradually descends. This trend has led to confirm the existence of generational differences when the musicians cover songs. These patterns have been able be explained by the emergence of Jazz in the 1920s and the Pop/Rock in the 1950s.

The use of different network metrics has allowed to rank the most important artists in each moment, identifying the great impact of important musical figures such as Duke Ellington, Bing Crosby, The Beatles and Bob Dylan and their role in the height of the Jazz and the arrival of the Pop/Rock. These metrics have described different types of impact, providing significant nuances in the assessment of musicians.

Finally, cover versions network has shown that the affinity among genres and styles is the main criteria to cover songs. The genre network has illustrated how Jazz music and Pop/Rock shape the core of the popular music in the 20^th and 21^st centuries. Jazz constitutes the axis of old genres such as Vocal and Stage & Screen, whereas Pop/Rock is central for Rock styles such as R&B and Alternative/Indie Rock. This analysis at the genre level has also made possible to appreciate the change in the preferences for covering songs, confirming the transition of the Jazz Age in the 1920s to the Rock Revolution in the 1950s.

Supporting information

S1 File. Full artists indicators.

(CSV)

Click here for additional data file.^{(7.6MB, csv)}

S2 File

(DOCX)

Click here for additional data file.^{(14.9KB, docx)}

S1 Fig

(TIF)

Click here for additional data file.^{(35.6MB, tif)}

Data Availability

All Cover songs files are available from the Open Science Foundation database (accession URL https://osf.io/6j2ru/).

Funding Statement

The author(s) received no specific funding for this work.

References

1.Shuker R. Understanding popular music culture. 5th edition. London: Routledge; 2016. [Google Scholar]
2.Burnett R. The global jukebox: The international music industry. Psychology Press; 1996. [Google Scholar]
3.Wall T. Studying popular music culture. London: Sage; 2013. [Google Scholar]
4.Borthwick S, Moy R. Popular music genres: An introduction. London: Routledge; 2020. [Google Scholar]
5.Schroeder E. Music Databases. Encyclopedia of Library and Information Science: Lib-Pub. 2003:1931.
6.Cohen M, Baker G. Takin’Care of Business: Music Business Reference. Music Reference Services Quarterly. 2015, 18(3–4): 157–163. [Google Scholar]
7.Weyde T, Cottrell S, Dykes J, Benetos E, Wolff D, Tidhar D, et al. Big data for musicology. In: Proceedings of the 1st International Workshop on Digital Libraries for Musicology; 2014. p. 1–3.
8.Cottrell S. Big Music Data, Musicology, and the Study of Recorded Music: Three Case Studies. The Musical Quarterly. 2018, 101(2–3): 216–243. [Google Scholar]
9.Rose S, Tuppen S, Drosopoulou L. Writing a Big Data history of music. Early Music. 2015, 43(4): 649–660. [Google Scholar]
10.Park D, Bae A, Schich M, Park J. Topology and evolution of the network of western classical music composers. EPJ Data Science. 2015, 4(1): 2. [Google Scholar]
11.McAndrew S, Everett M. Music as collective invention: A social network analysis of composers. Cultural Sociology. 2015, 9(1): 56–80. [Google Scholar]
12.Gleiser PM, Danon L. Community structure in jazz. Advances in complex systems. 2003, 6(04): 565–573. [Google Scholar]
13.Budner P, Grahl J. Collaboration networks in the music industry. arXiv:1611.00377 [Preprint]. 2016 https://arxiv.org/abs/1611.00377
14.Makkonen T. North from here: The collaboration networks of Finnish metal music genre superstars. Creative Industries Journal. 2017, 10(2): 104–118. [Google Scholar]
15.Napier K, Shamir L. Quantitative sentiment analysis of lyrics in popular music. Journal of Popular Music Studies. 2018, 30(4): 161–176. [Google Scholar]
16.Ruth N. “Where is the love?” Topics and prosocial behavior in German popular music lyrics from 1954 to 2014. Musicae Scientiae. 2019, 23(4): 508–524. [Google Scholar]
17.Collins N. Influence in Early Electronic Dance Music: An Audio Content Analysis Investigation. ISMIR. 2012: 1–6.
18.Mauch M, MacCallum RM, Levy M, Leroi AM. The evolution of popular music: USA 1960–2010. Royal Society open science. 2015, 2(5): 150081. 10.1098/rsos.150081 [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Klimek P, Kreuzbauer R, Thurner S. Fashion and art cycles are driven by counter-dominance signals of elite competition: quantitative evidence from music styles. Journal of the Royal Society Interface. 2019, 16(151): 20180731. 10.1098/rsif.2018.0731 [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Youngblood M. Conformity bias in the cultural transmission of music sampling traditions. Royal Society open science. 2019, 6(9): 191149. 10.1098/rsos.191149 [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Plasketes G. Re-flections on the Cover Age: A Collage of Continuous Coverage in Popular Music. Popular Music and Society. 2005, 28(2): 137–161. [Google Scholar]
22.Cusic D. In defense of cover songs. Popular Music and Society. 2005, 28(2): 171–177. [Google Scholar]
23.Solis G. I did it my way: rock and the logic of covers. Popular Music and Society. 2010, 33(3): 297–318. [Google Scholar]
24.Doktor SD. Covering the Tracks: Exploring Cross-gender Cover Songs of the Rolling Stones "Satisfaction" [Doctoral dissertation]. University of Georgia; 2008.
25.Chu YW, Leung E. Remapping Hong Kong popular music: covers, localisation and the waning hybridity of Cantopop. Popular Music. 2013, 32(1): 65–78. [Google Scholar]
26.Anderson K. Song ‘adaptations’ and the globalisation of French pop, 1960–1970. French Cultural Studies. 2015, 26(3): 330–342. [Google Scholar]
27.SecondHandSongs [Internet]. Introduction to the database. 2020. https://secondhandsongs.com/page/Introduction
28.Smith E. Motherboard. The Story of AllMusic, the Internet’s Largest, Most Influential Music Database. 2016. https://www.vice.com/en_us/article/53djj8/the-story-of-allmusic-the-internets-largest-most-influential-music-database
29.Freeman LC. Centrality in social networks conceptual clarification. Social networks. 1978, 1(3): 215–239. [Google Scholar]
30.Brin S, Page L. The anatomy of a large-scale hypertextual Web search engine. Computer networks and ISDN systems. 1998, 30(1–7): 107–117. [Google Scholar]
31.Opsahl T, Agneessens F, Skvoretz J. Node centrality in weighted networks: Generalizing degree and shortest paths. Social networks. 2010, 32(3): 245–251. [Google Scholar]
32.Gronow P. The record industry: the growth of a mass medium. Popular music. 1983, 3: 53–75. [Google Scholar]
33.White DR, Borgatti SP. Betweenness centrality measures for directed graphs. Social networks. 1994, 16(4): 335–346. [Google Scholar]
34.Pitts M, Hoffmann F, Carty D, Bedoian J. The Rise of the Crooners: Gene Austin, Russ Columbo, Bing Crosby, Nick Lucas, Johnny Marvin and Rudy Vallee. Scarecrow Press; 2001. [Google Scholar]
35.Garofalo R. Crossing over: From black rhythm & blues to white rock ‘n’roll. In Kelley N, editor. Rhythm and business: The political economy of black music. New York: Akashit Books; 2002, pp. 112–137. [Google Scholar]
36.Kellett A. The British Blues Network: Adoption, Emulation, and Creativity. Ann Arbor: University of Michigan Press; 2017. [Google Scholar]
37.Scheuer TE. American Popular Music: Readings from the Popular Press. Bowling Green: UP; 1989. [Google Scholar]
38.Nicholls D., editor. The Cambridge history of American music (The Cambridge History of Music), Cambridge, UK. Cambridge University Press, 653; 1998. [Google Scholar]
39.Kania A. All Play and No Work: An Ontology of Jazz. The Journal of Aesthetics and Art Criticism. 2011, 69(4): 391–403. [Google Scholar]
40.Dineley S. Covers Uncovered: A History of the" Cover Version," from Bing Crosby to the Flaming Lips. [Doctoral dissertation]. Wester University; 2014. https://ir.lib.uwo.ca/cgi/viewcontent.cgi?article=3630&context=etd
41.Homan S. ’You’ve Got to Carry that Weight’: Tribute Acts in the Entertainment Supermarket. In Homan S, editor. Access All Eras: Tribute Bands and Global Pop Culture. New York: Open University Press; 2006, pp. 32–51. [Google Scholar]

PLoS One. doi: 10.1371/journal.pone.0250212.r001

Decision Letter 0

Nicola Perra

6 Nov 2020

PONE-D-20-26285

Cover versions as an impact indicator in popular music: a quantitative network analysis

PLOS ONE

Dear Dr. Ortega,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Dec 21 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

We look forward to receiving your revised manuscript.

Kind regards,

Nicola Perra

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

Additional Editor Comments (if provided):

Dear authors,

as clear from the reviewers comments the work has merit and deserve to be published.

Please take into considerations all the comments raised by both reviewers. I believe that addressing them would really help and strengthen the paper.

Best regards,

Nicola Perra

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The manuscript “Cover versions as an impact indicator in popular music: A quantitative network analysis” uses a large crowdsourced database of musical covers (the performance of an already existing song by a different artist) to look at the network of covers. The analysis extracts the most central artists according to three measures of centrality. It also examines the connections between genres of covers. Finally, it provides some temporal analysis of how covers have evolved over the last century.

I think that the question of how covers result in social and musical connectivity is interesting and that networks are an appropriate method for parsing this phenomenon. However, I am not sure that the analysis in the manuscript has used networks to their fullest extent, and many of the analyses seem to be confounded by time. The question of time, as in how long an artist has existed, seems to be central to many of the analyses---it is the motivation for introducing PageRank, and at the center of many of the genre analyses. However, for how much it comes up, I do not think it has been directly and adequately addressed.

For example, the analysis observes that ranking artists by in-degree results in a number of older bands. It then states “a more precise way to reduce the oldness effect on the musical impact is by calculating its PageRank.” I do not think that PageRank is a sufficient answer here---it does explicitly not do anything to guarantee that time is properly accounted for in the analysis. A number of the artists in Table 2 are still from the 1920s and 1930s (brief aside: I would like to see more than the top 10 artists, perhaps as an appendix table). Instead, I think time should be tackled directly: we are interested in who the most central artists are in the cover network, but an artist is more likely to be covered simply if they are around longer. A simple analysis might be normalizing the in-degree by the number of years since the artist first became active. A more comprehensive analysis, that I would like to see, would treat the cover network as a temporal network and use temporal centrality measures to understand how artists have become more or less central over time. The individual artists section culminates in the statement that “musical oldness of artists is a key factor to improving the likelihood of being covered.” This seems more like the analysis hasn’t properly accounted for time, rather than a result. That is why I would like to see time handled more directly.

The genre analysis does a bit more to look at temporal dynamics, but it also avoids probing too deeply. I think Figure 6 is interesting. I think that it would work better as a heatmap where the rows and columns are years, and the color is Figure 6’s y-axis. I would like to know more about what is going on here. What genres are driving these temporal lags in covers? How does the centrality of genre covers change as a result in these lags?

I think that the manuscript struggles to pinpoint particular temporal questions to answer because its research questions are relatively broad and the main motivation is to do a quantitative analysis. I do not think doing a quantitative analysis necessarily tells us anything new, and so I don’t find the “total absence of quantitative studies” to be that compelling (though isn’t that also contradicted by the previous two paragraphs in literature review?). However, the manuscript mentions that qualitative studies have led to theorizing about the cultural contribution of covers (references 20-22). I think there is a missed opportunity here to use the large-scale network analysis to help clarify and expand upon what has been noticed in the theoretical/qualitative studies. I think by engaging more closely with the prior literature and synthesizing it into hypotheses that have not been tested would help generate a more productive analysis, and help guide questions about how time affects cover centrality.

Overall, I would like to see the issue of time addressed more clearly and more directly in the analysis. I think it confounds how we interpret both the individual and genre analyses. To this end and deriving questions related to how time affects the cover network, I think the manuscript would benefit from engaging with the hypotheses of the prior literature, and building from them, rather than doing a general descriptive analysis of the network.

Other comments

- The introduction should tell me what the questions are and how the analysis proposes to answer them. It ends abruptly before going to the related research. Similarly, the manuscript says that the objective is “to visualize and analyze the cover songs network between artists in order to describe its topology and identify the main nodes.” But why do that? What do we learn from making a network and identifying the main nodes that we didn’t know before? The final conclusion is “These results further strengthen the quantitative approach in studies on the history of music.” But what do they strengthen? What are they speaking to?

- I am worried about duplications in the data. In data cleaning, it is mentioned that if an artist covers the same song more than once, it is counted as a different cover. Has there been any validation to ensure that those covers of the same song are not exactly the same cover? similarly, I am worried that the number of songs in the Tables are extremely high. Duke Ellington was prolific, but I think it is unlikely that he wrote (or performed?) 7,447 songs. Similarly for the other artists, though there is then a huge gap with Franz Gruber and Joseph Mohr. I would like ot know what is going on here better. Why are the numbers so large?

- These analyses do not establish that the cover network is scale free and I think that the language of “scale free” should be left out of the paper. The degree distributions do not look scale free to me visually, and no test has been reported to indicate that they follow a power law. What is alpha? Relatedly, the discussion of “preferential attachment” should be dropped because no analysis establishes that there is preferential attachment in the network (exactly because there are no analyses that directly model the time of artists and receiving covers). I think that an analysis of preferential attachment would be interesting, and it can be done (for example, see Laszlo Barabasi’s textbook for measuring preferential attachment), but it is not done here. Even if it is done, I think the language of “scale free” should not be included---preferential attachment in a real world system does not guarantee a power law degree distribution.

- How PageRank is framed through the paper is inconsistent. In the introduction and abstract, it is described as measuring the “most influential.” In the analysis, it is described as accounting for time (which I don’t think it does). The intuitive description of PageRank when it is introduced as a metric is correct, and that should be used to describe in the other parts of the paper.

- I don’t believe there is a sentence that explicitly says what the nodes and edges are in the network. Specifically, I want to know if the nodes are artists from SecondHandSongs, or from Allmusic.

- In the SecondHandSongs dataset, what does it mean for a song to be labelled as “traditional.” In the Allmusic dataset, what is meant by “genres and styles are not interconnected”? Isn’t a network made of them later?

- The SecondHandSongs dataset section first says that there are 96 thousand songs. The data extraction paragraph says that 155 thousand songs were extracted. Why are they not the same number?

- Details about song and artist IDs do not need to be included in the manuscript.

- I do not understand the inclusion criteria for the genre network: “Only the most important influences were highlighted through percentages of covers equal to or higher than 10% the total number of covers made.”

- The names of artists and bands should not be italicized.

Reviewer #2: The manuscript ‘Cover versions as an impact indicator in popular music: a quantitative network analysis’ deals with the analysis of a database of cover songs through the lens of networks, where two artists are linked if one covers a song of the other. The resulting network is then used to measure the influence between artists and to exploit the connections between their respective genres.

The idea is (to the best of my knowledge) new, the results are interesting, and the presentation clear (the figures not so much). Overall, I believe that the manuscript deserves publication in PLoS One, but there are some points that I would like the authors to discuss first:

1) As far as I understand the network representation adopted by the authors, there is a directed link between artist i and artist j if i covers a song of j. The authors also write "Links between artists are therefore established according to the number of times an artist covers another", so this makes me think that links are weighted. However, all the measures used do not take weights into account. This is an important point that needs to be clarified. In case the measures take weights into account this should be explained better. Otherwise, why not considering weighted measures (in-/out-strength instead of degree, betweenness centrality defined on the shortest paths that consider weights etc.)?

2) Is there a way to account for the size of the production of different artists to make the cover flows comparable? For example, an artist with a huge production that gets covered a lot would could in principle be comparable with an artist with a smaller production that gets covered a bit less. The same argument holds for genre-to-genre comparisons, since the production in each year and for each genre can change. I believe that counters should be somehow normalised to be comparable. See for example what is usually done for citation networks [Radicchi, F., & Castellano, C. (2011). Rescaling citations of publications in physics. Physical Review E, 83(4), 046116.] or [Waltman, Ludo. "A review of the literature on citation impact indicators." Journal of informetrics 10.2 (2016): 365-391.] to assess the influence of authors and temporal flows of knowledge across different fields [Sun, Y., Latora, V. The evolution of knowledge within and across fields in modern physics. Sci Rep 10, 12097 (2020)].

Below a list of minor points:

1) Abstract: the sentence "However, no studies have explored the meaning of covers from a quantitative and structural perspective" sounds a bit vague to me. What is the meaning of a cover? I would suggest rephrasing it. (the paper is not about the meaning of covers as far as I see it). Also, the structural perspective might be misleading here, since it's not on the structure of the cover, but covers are used to create a structure between artists.

2) Introduction: "These acknowledgements lead to (...) while assessing the impact of artists’ careers". As this sentence is written at the moment is seems that covers access the impact of the careers. It should be rephrased.

3) Introduction, last sentence: "who their songs were versioned by" sounds a bit weird.

4) Related research, third paragraph: "Nevertheless, an important gap in the analysis of cover songs is the total absence of quantitative studies." The second paragraph presents some quantitative studies on music, not on covers. Given that the third paragraph starts with 'nevertheless', I would swap the writing to emphasise the lack of analyses of covers among the quantitative studies rather than the lack of quantitative studies among the analyses of covers.

5) Objectives. The second point reads "Could social network measures be used to detect different types of impact?". I understand this is precisely what the manuscript is about, but I do not think that the results of the paper actually answer this question. The manuscript does not compare the ranking induced by the measures with some "ground truth" influence of artists, since it obviously does not exist (maybe some proxy for popularity could be used). Thus, since the authors are not testing the ability of the measures to capture the influence, but are proposing to use the measures as a proxy for influence, I would change/remove the question.

6) A clear definition of the network representation and nomenclature is missing. This would help the reading of the following 'social network analysis' section, where nodes and lines are mentioned (also, why not calling them links?).

7) Degree distribution section: the \\alpha is not defined. Also, which method is used to fit the distribution? This should be specified. The fit does not seem to capture well the tail of the out degree distribution.

8) The authors focus on the Page Rank and the betweenness centrality, but there is obviously a long list of additional measures that could be used instead to produce rankings in -evolving- networks (Liao, Hao, et al. "Ranking in evolving complex networks." Physics Reports 689 (2017): 1-54.). For example, the alpha-centrality could be a good proxy for influence in static graphs. Why did the authors restricted their attention to the selected measures?

9) Just before Genres and styles section: "This result confirms that the musical oldness of artists is a key factor to improving the likelihood of being covered". This makes sense, but isn't there already a bias towards the "old" artists that survived against the new ones? There would be "old" forgotten ones that did not even appear in the database, so oldness in this sense already encodes an inner survivorship bias. The authors could maybe expand on this.

10) Fig 3 is not very readable.

11) I really like the idea behind Fig. 4. Would it be possible to visualise it also with a heatmap instead of a network? It would make possible to directly compare flows within and across genres. Ideally, it would be interesting to see how these flow evolved in time and reorganise due to the appearance of new genres.

12) Fig. 5 and Fig. 6 do not have labels.

13) Fig. 6 is not easily to interpret, since lines typically denote an evolution, while here each point is a percentage. Also, I noticed something that I do not understand. As I understand the chart, there are songs from artists of the 90s that are covered by artists that started in the 1910...(the lightest blue peak at 1990). How long can a career be? This should be checked, and maybe a simple stacked bar chart would facilitate the visualisation? Otherwise, I would suggest to use a heatmap with the years of the two artists on the axes and the proportion of covers as color intensity. This would make the comparison easier.

14) In the conclusions the authors state that "Most interestingly, perhaps, these influences are the result of cycles." The connection between influence and cycles is not clear to me and could be further explained.

15) Here some related papers that might be of interest:

- Serrà J, Corral Á, Boguñá M, Haro M, Arcos JL (2012) Measuring the evolution of contemporary western popular music.Sci. Rep.2:521.

- Zivic PHR, Shifres F, Cecchi GA (2013) Perceptual basis of evolving western musical styles. Proc. Natl. Acad. Sci. U.S.A.110(24):10034–10038.

- Youngblood, M. (2019). Conformity bias in the cultural transmission of music sampling traditions. Royal Society open science, 6(9), 191149.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Iacopo Iacopini

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2021 Apr 19;16(4):e0250212. doi: 10.1371/journal.pone.0250212.r002

Author response to Decision Letter 0

16 Dec 2020

5. Review Comments to the Author

First of all, I would like to thank to the reviewers their comments because their suggestions have led me to detect important gaps in the manuscript and explore the evolutionary aspect of the data with more detail. Great part of the manuscript was then rewritten to centre the manuscript in how the network has evolved in successive decades.

Section “Degree distributions” has been rename and broaden as “General characteristics of the covers network”. Now, the global network has been split by decade in which the artists start their music career, covering since 1910s to 2010s. This allows us to study how the network has evolved as new artist were aggregated. Apart from the indegee and outdegree distributions (Figure 1), this section now presents several global network indicators (Figure 2 and Table 1) such as Weighted average degree and Clustering coefficient that describe the characteristics of the network in each decade. This has allowed to observe different growing patterns. In addition, Figure 3 (former Figure 6) has been added to explain the possible causes of the irregular growth of the network.

A table with the network indicators of the artists and a large-size version of Fig 6 was added as supplementary material.

Figure 6 (Now Figure 3) has been redone as a heatmap. Now, differences between generations are better appreciated. Section about genres has been remade to emphasize how the preferences covering genres have changed (Figure 8). This result attempts to explain the structural changes observed in the growth of the global network.

The objectives of the paper have been changed due to the new approach of the manuscript and all the objectives include now an evolutionary perspective.

Results section has been enriched with longitudinal analyses that respond to the research questions. Now, the discussion about these results has been improved adding more previous studies that support the interpretations

Other comments

A paragraph was added at the end of the introduction to define the principal purpose of the paper. As I have commented, objectives have been changed and now more accurate research queries were formulated. Conclusions section was added to give more relevance to this part and new conclusions were written to precise the main findings.

This only occurs with songs that share the same title. It is very infrequent that an artist covers a same song from the same artist in different moments. In spite of this, I think that each cover versions shall be counted as different because it is created in a different context and they express the strength in the connection between artists.

similarly, I am worried that the number of songs in the Tables are extremely high. Duke Ellington was prolific, but I think it is unlikely that he wrote (or performed?) 7,447 songs. Similarly for the other artists, though there is then a huge gap with Franz Gruber and Joseph Mohr. I would like ot know what is going on here better. Why are the numbers so large?

The number of songs are obtained from Allmusic.org page. This site gathers all the songs recorded by an artist. This link shows the current number of Duke Ellington songs, 7621.

https://www.allmusic.com/artist/duke-ellington-mn0000120323/songs/all

This amount is only used to express the size of the node in the graph and they are used to illustrate the production of musicians. They are not relevant in the analysis.

However, in my opinion that volume of recorded songs is common in important music starts. For example, Bing Crosby has close to 9000 songs

https://www.allmusic.com/artist/bing-crosby-mn0000094252/songs/all

The Beatles, 2000 songs

https://www.allmusic.com/artist/the-beatles-mn0000754032/songs/all

Allusions to scale free have been avoided. Due to the new perspective of the paper, focused on the evolutionary aspects, any issue on the type of network is irrelevant and it has been removed.

Alpha is the exponent of the curve. This has been included in the text (p. 8)

That reference about the use of Page Rank for counteracting time effects is wrong and was removed. This metrics is now only used as influence indicator.

- I don’t believe there is a sentence that explicitly says what the nodes and edges are in the network. Specifically, I want to know if the nodes are artists from SecondHandSongs, or from Allmusic.

This sentence was added to General characteristics of the covers network in p. 8: “A network graph was constructed with 106,096 nodes (artists) and 855,650 arcs (cover versions) extracted from SecondHandSongs”

Traditional means that the first performer of the song is unknown, so then it is not possible to create a link between covered artist and subsequent performers.

That sentence about the interconnection of genres and styles is a mistake and that idea is wrongly expressed. A new text was added: “Styles are not considered as sub-categories of genres, then they are used as labels or keywords that are freely assigned to an artist.”

- The SecondHandSongs dataset section first says that there are 96 thousand songs. The data extraction paragraph says that 155 thousand songs were extracted. Why are they not the same number?

There is a mistake in the count of original songs. The sample only include titles of songs that have been covered, so 155k is actually the list of titles included in the sample, both covers and original songs. I have decided to remove that figure because it would be confused.

- Details about song and artist IDs do not need to be included in the manuscript.

I think that this information is very important to reproduce the sample. This detail about the IDs allows other researchers to extract data from SecondHandSongs and Allmusic.

This is a pruning process that selects only the most important links. The reason is that all the genres have links among them, resulting a ball-shape network. This pruning criterion ensures that only the most significant relationships are drawn in the picture, producing a more interesting graph. This text was added in p. 18: “The graph only includes links that are equal or higher than 10% of all the cover songs of other genres (outdegree). For example, Jazz adapts 40% of Jazz pieces and 24% of Vocal songs and is an inspiration for 65% of Jazz and 10% of Vocal songs. This pruning criterion allows to select only the most important relationships among genres.”

- The names of artists and bands should not be italicized.

Corrected

I agree with you, and it was a mistake to use those metrics for a weighted network. Degree was replaced by the Weighted degree. Now, the results are more consistent.

This is an interesting point because one could assume that some type of normalization is necessary to avoid size-effects. However, this issue is problematic from a conceptual perspective. Firstly, we are talking about musicians, not about scientists. The concept of production (number of papers vs. number of songs) and impact (number of citations and number of covered songs) cannot be literally translated. In Science, any paper includes bibliographic references (it is almost impossible to publish a research article without references), then as more articles more citations. In other words, citations are dependent of the publications.

However, songs do not include any reference to previous releases, musicians are not determined to cover songs of other artists. Therefore, more songs do not imply more cover versions. In fact, there are musicians with a large trajectory and very low number of cover versions, and the contrary (e.g. tribute bands). This is confirmed by the low correlation between the number of songs recorded and covers done (weighted outdegree) (r=.465) and received (weighted indegree) (r=.443). Then, the use of a normalized count would cause numerical artefacts, undervaluing great musical figures with an extensive career and large production (e.g. Duke Ellington, Bing Crosby) and overvaluing unknown artists whose exclusive merit is to firstly interpret a popular song such as Silent night by Franz Gruber and Joseph Mohr, or Summertime by Abbie Mitchell. In my opinion, the use of normalized counts would cause confusing results.

Below a list of minor points:

Abstract was rewritten and that sentence was removed due to its ambiguity. Now, the abstract is focused on the new results and their conclusions.

The sentence was rewritten: “These acknowledgements lead to links between artists and illustrate the musical influences of soloists or bands, while enabling to assess the impact of artists on specific musical communities.”

3) Introduction, last sentence: "who their songs were versioned by" sounds a bit weird.

The sentence has been rewritten: “…including information about cover songs and their performers.”

I agree with your observation. The text was rephrased: “Nevertheless, an important gap in the quantitative studies on the contemporary popular music is the total absence of analysis on cover versions. In most of the cases, they have been observed from a theoretical perspective, figuring out the cultural meaning of this type of songs.”

That research question was reformulated: “Which musical artists are the most influential in the network and how have they reached that position? What information does each network metric provide about the impact?”

This information is now included at the beginning of the General characteristics of the covers network.

This text was added to explain the power-law distributions: “This interpretation could be confirmed by the cumulative distribution of covers performed (outdegree) and received (indegree) by artist. They follow a power-law distribution (Figure 1), with an exponent (α) of 2.882 for outdegree and 2.937 for indegree. R package poweRlaw was used to create the cumulative distribution functions (CDF) and the maximum likelihood estimation (MLE) method was used to fit the distributions.”

The study is focused on three metrics: Weighted indegree, PageRank and Betweeeness centrality. These metrics provide three different points of view about the impact: Weighted indegree offers information about how the impact is accumulated, PageRank about the influence capacity of artists, and the betweenness centrality measures the ability to be covered by different and distant artists. These metrics were selected by several reasons: (1) they are well known by a generalist audience (PLOS ONE is a multidisciplinary journal), (2) their meaning and calculation is easy to understand, then it is easy to translate their meaning to a different environment such as the popular music, and (3) they are implemented in many network analysis software packages (Gephi, Pajek, VOSviewer, etc.).

I agree that there much more network metrics, but this is a study about cover songs and I think that three metrics is enough to have a good view about the impact of the musicians. I think that the inclusion of more metrics would not improve the results nor provide additional information. For example, alpha-centrality belongs to the eigenvector family (PageRank is a version of eigenvector for directed networks). It is very likely that the result would be similar than the observed with PageRank.

That paragraph was removed due to that result was obvious and to make way to other relevant results.

10) Fig 3 is not very readable.

Figure 3 (now Figure 6) was changed by a more complete figure. The problem of this figure is the size and resolution. I think that the journal would publish a large-size version of the figure that allows to explore the connections and groups.

Figure 8 was added to visualize the evolution of cover songs by music genre. This figure illustrates the preference by certain genre in each decade.

12) Fig. 5 and Fig. 6 do not have labels.

Both figures has been removed

Figure 6 was change by Figure 3. I agree with you and I think that a heatmap is more suitable to represent that idea.

You comment about artists from the 1910s cover songs from 1990s artists. This is a small artefact caused by the low proportion of songs and artists from 1900s decade in the database. The reason of that weird result is that there are some old orchestras and choruses (e.g. The London Symphony Orchestra) that have a large trajectory and they sometimes cover songs from Pop/Rock stars.

Conclusions were completely rewritten and that sentence was removed. The new analysis makes more clear the connection between cycles and influence.

15) Here some related papers that might be of interest:

- Serrà J, Corral Á, Boguñá M, Haro M, Arcos JL (2012) Measuring the evolution of contemporary western popular music.Sci. Rep.2:521.

- Zivic PHR, Shifres F, Cecchi GA (2013) Perceptual basis of evolving western musical styles. Proc. Natl. Acad. Sci. U.S.A.110(24):10034–10038.

- Youngblood, M. (2019). Conformity bias in the cultural transmission of music sampling traditions. Royal Society open science, 6(9), 191149.

Many thanks for the references. I have decided to cite the last reference (Youngblood, 2019) because this study analyses music samples (a bit close to the cover versions) and it is focused on the factors that influence their selection. The remaining ones have been discarded because they are centered on musicology aspects such as the sound, melody and rhythm.

Attachment

Submitted filename: Reviewers_response.docx

Click here for additional data file.^{(27.7KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0250212.r003

Decision Letter 1

Nicola Perra

3 Feb 2021

PONE-D-20-26285R1

Cover versions as an impact indicator in popular music: a quantitative network analysis

PLOS ONE

Dear Dr. Ortega,

==============================

ACADEMIC EDITOR: Dear Authors, as acknowledged by the two reviewers the article has been drastically improved in the first revision. Please consider the minor points raised by the referees and resubmit. In doing so, please make sure to highlight the changes to the manuscript so that we can speed up the review process.

==============================

Please submit your revised manuscript by Mar 20 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

We look forward to receiving your revised manuscript.

Kind regards,

Nicola Perra

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #2: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: Partly

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: N/A

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Reviewer #1: The manuscript “Cover versions as an impact indicator in popular music: a quantitative network analysis” has improved with the revisions. I think that identifying the split in what decades are covered (1920s vs 1950s) is interesting, and I think it has helped create a more coherent story around the other analyses. I have some reservations and questions that I would still like to see addressed.

1. Table 1 and several of the figures rely on networks constructed across time. How exactly were those networks made? Is it a cumulative network, where, for example, any node and link in the 1910s network is also in the 1920s network, and any node and link in the 1910s or 1920s network is in the 1930s network? I think this is the proper way to construct the network for the analyses. But it’s not clear if instead these might be snapshot networks, where only nodes from a particular decade and who they cover are included. This should be explained in the text.

2. I am still skeptical of duplicates in the songs used to count the productivity of each artist, as seen in Table 2 for example. Looking even just at one artist, Duke Ellington, it was easy to see that several “different” songs were actually the same one. I suspect it is similar for the other artists. If this measure of productivity is used in the paper, it should do deduplication.

Also, it is not clear how artists were matched across SecondHandSongs and Allmusic. Is it just if an artist’s name matches exactly across the two datasets? This should be explained.

3. I am still unclear on what nodes ultimately end up in the networks because there are several similar but conflicting numbers about the number of nodes

- In Data Sources, it says the site “reported 788k covers and, 143k artists”

- The paper reports that “143k artists and 1.1 million covers” were extracted. This contradicts the first number. Is it because the previous stats were reported at an earlier time? If so, I think they should be dropped because they are confusing and appear to contradict the data collection.

- IDs 0 to 175000 were extracted. Why 175000 if there are only 143k artists?

- The network is reported to have 106,095 nodes. Why is it not 143k?

- The last column of Table 1 reports that there are 101,112 nodes in the 2010s network. Why is it not 106,095?

If there are data processing steps that are causing these different numbers, they should be clearly reported next to each step.

4. Figures 4 and 5 help make some of the findings clear. It is difficult to tell which curve is associated with each artist. I would suggest placing each artist’s name at the end of each curve, so that they are all aligned on the right side of the plot.

Figure 8 is interesting, but it is very difficult to tell which color lines up with which curve. I think a heatmap-style figure like Figure 3 would be more clear.

5. There are several places in the paper where it says it will do an analysis that is more than what is actually done

- In Top Covered Artists, the paper says “This section analyzes who are these figures and how they have acquired their positions.” The analysis does not say anything about *how* they acquired their positions in the network, just their importance in it.

- In Top Covered Artists, the paper says “In general, as the cover network evolves over time, the impact of new figures is greater and faster.” This has not been shown in general. In the following sentence, it also hasn’t been shown that anything is specifically “exponential.”

- In the Discussion, the paper says “Regardless of these limitations, the results could be representative of mainstream popular music.” That this is representative is not clear to me, I don’t think the paper needs to say this.

I particularly would like to see the methods questions answered. Addressing the other points would help make the paper’s story more clear and reliable.

-------

Other comments:

- The paper’s language switches tenses often. Sometimes it’s past tense, sometimes it’s present, sometimes it’s conditional. A reliable tense should be used throughout the paper.

- Network metrics (“weighted in-degree”, “clustering coefficient”) should not be capitalized throughout the paper.

- The abstract is jarring: it jumps right into explaining data after barely motivating why the structure of covers is interesting (and it is interesting).

- The density by itself does not tell us much about the centralization of the network. Many empirical networks have very low density. What is more interesting is the density over time. When the density is reported, I think the “less than” sign is backwards, it should be (d < 0.01%) if I understand correctly. Also is it that d < 0.01% or 0.01? A density of 0.0001 for 0.01% does seem very low.

- It still wasn’t clear to me immediately how the edges of the genre network were selected. If I understand correctly, it’s more clear to say something like “The edge weights were normalized per each node, and any weight below 0.1 was removed. This is the same as only including links where at least 10%...” For future reference, this is similar to but less statistically rigorous than using a backboning method like the disparity filter.

- In the discussion, it’s not clear if the “deep transformations in the way of making music” are speculations, or already known by previous work. How can “subsequent performers from the 1960s quickly overtake the number of covers of the previous generations” if “the cover song lost prominence because musicians compete now”?

-------

Typos and phrasing (since PLoS ONE doesn’t provide copy editing):

- Page 1, “network metrics have allowed to identify” is missing a word I think

- Page 2, “while enabling to assess the impact” is awkwardly phrased

- Page 4, “In this form, it attempts to know how the use of” is awkward with “it attempts to know.” The paper can be more definitive.

- Page 5, In the third research question, “possible to appreciate” is awkward since “appreciate” is usually meant in terms of subjective terms.

- Page 5, “the database has been fed,” “fed” is an awkward verb. Databases don’t eat

- Page 5, “the song is labeled as traditional,” put quotes around “traditional” to make it more clear it’s a label, and say that it’s labeled that way by the dataset (I believe?) rather than by the researcher

- Page 6, “Styles are not considered as subcategories of genres, then they are used as labels”, something isn’t right with the second half, “then they are used as”

- Be consistent with “arcs,” “lines, “edges,” “links.” Aim to use just one

- Page 12, very bottom, typo with “in favour of the new Pop/Rock starts.” Should be “stars”

- Page 15, very bottom, “For instead” doesn’t seem like the right transition

Reviewer #2: I would like to that the author for considering the suggestions given by both referees (especially those regarding the temporal aspect). I find the current manuscript much improved. The are some points that I would like to discuss before recommending the manuscript for publication.

1) I am glad that the author modified the measures as suggested and took weights into account. In fact, answering a point previously raised, the author says that the "Degree was replaced by the Weighed degree. Now, the results are more consistent." Was it the same for the other measures? At the moment it is not clear to me if all measures have been changed to their weighted counterparts or just the degree. This is important for consistency and reproducibility.

2) Regarding my previous point (2), I obviously agree that musicians are not scientists and a song does not come with a set of references. However, artists with a huge productions have a lot of material available that could be used by other artists when performing covers. In very basic terms, let us assume that an artist wants to record a cover. In the absence of real preferences (tastes or "influence"), the artist would pick a random song from the ones available, and thus would have a higher probability of covering artists with a big production (due to the mass of the production and the age). This implies that --even in the absence of a real influence mechanism-- artists with a lot of songs will end up being more covered than others. I understand the concerns of the author, but, given the central role played by 'influence as a driver for covering' in this manuscript, I believe that this is an important point that has not been addressed and should be (at least) discussed.

In addition, I find the argument 'more songs do not imply more cover versions' not really confirmed here. The correlations coefficients provided "low correlation between the number of songs recorded and covers done (weighted outdegree) (r=.465) and received (weighted indegree) (r=.443)" are not that low to me.

Minor points:

When I read 'evolutionary' I think about evolution as a change (like in biology). I thus find its use in some parts of the manuscript a bit our of context, since it reads more like a temporal analysis (how quantities change in time) rather than the analysis of how things develop (by changing into others).

In Objectives, 'indicators' is repeated.

Just before Fig.2, 'The artists that joint the network'--> joined

The text regarding Fig. 2 says "From a longitudinal point of view, the growing pattern of the network is not continuous (Figure 2)". I find the sentence 'growing pattern of the network' a bit misleading, since degree distributions are skewed (and the average is not much representative) and the real pattern I assume would be hidden in the fact that new coming nodes without many links pull down the average after the initial growing phase. The author might consider plotting also the number of new nodes/covers in time to have a clearer picture of the growing pattern of the network that generate the curves shown. In addition, I would suggest avoiding the use of 'not continuous' since it might imply the presence of a discontinuity in the curve (which is not the case).

After Fig. 4, "This fact then suggests that the covering of songs is more and more concentrated in few artists, increasing exponentially their impact over time." how is "exponentially" tested?

In Fig. 4 and 5 I find difficult to associate artist labels with curves when there are many (for example The Rolling Stones and Bob Dylan in FIg.5). Maybe arrows could help?

Fig. 8 is hard to read since blues colors are only slightly different (I can guess due to their size, but it is not straightforward).

Figure captions seem to include caption titles only. These should be improved by better explaining the variables shown.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Iacopo Iacopini

PLoS One. 2021 Apr 19;16(4):e0250212. doi: 10.1371/journal.pone.0250212.r004

Author response to Decision Letter 1

9 Feb 2021

I have introduced some lines to explain that the longitudinal analysis is done over cumulative networks:

“In order to study the evolution of the network, artists were classified by the decade in which they started their musical career. During this process, 4,983 artists were discarded because they did not have that information. Then, several cumulative networks by decade were built. This aggregate approach is because the cover versions are made of current or precedent artists and each network rests on the previous one.”

The number of songs reported by Allmusic is not exactly the total number of different musical pieces interpreted or recorded. Rather, it is the total number of releases, including different recording of the same songs. Due to this problem, that metric is not used in the study as productivity measure and it is included only for illustrative purpose. Even then, I decided to include a footnote to explain the meaning of songs:

“Data supplied by Allmusic. Number of musical tracks recorded and released by artist. It may include different recordings of the same song.”

Also, it is not clear how artists were matched across SecondHandSongs and Allmusic. Is it just if an artist’s name matches exactly across the two datasets? This should be explained.

A paragraph was added to Data cleaning section explaining how the match was made:

“The name of the artist or band was used to match SecondHandSongs and Allmusic records. This element was carefully cleaned to avoid mismatches between slight name variations. Thus, some characters such as ampersands and punctuation marks were removed or replaced by normal letters (Simon & Garfunkel by Simon and Garfunkel, “Big Mama” Thornton by Big Mama Thornton). In other cases, names with different variations were normalized (Booker T. and the Memphis Group’s by Booker T. and the M.G,’s, Johnny Burnette Trio by Johnny Burnette Rock and Roll Trio); synonymy between different artists (Tom Jones, 60s vocal artist and Tom Jones, 50s lyricist) was solved adding a digit to the name.”

3. I am still unclear on what nodes ultimately end up in the networks because there are several similar but conflicting numbers about the number of nodes

- In Data Sources, it says the site “reported 788k covers and, 143k artists”

You are right those figures are contradictory, the total number of covers reported by the site in January 2020 should not be less than the total retrieved covers. The problem is that the “1.1 million covers” refers to an unpolished data sample, which also included originals, samples and medleys. I have removed that text and I have added a new paragraph in Data cleaning explaining all the transformations of the sample and the final number of covers and artists used in the study:

“When a song is traditional or the original performer is unknown, SecondHandSongs assigns it to a generic unknown (20,428). These covers were removed because there is not a connection between artists. In addition, SecondHandSongs joins the code of two or more artists when they perform together (e.g. Sinatra and Bono, Elvis Presley and the Jordanaires). These cases were duplicated, assigning the same song to each artist. After all these transformations the final sample included 106k artists and 855k cover songs.”

- IDs 0 to 175000 were extracted. Why 175000 if there are only 143k artists?

SecondhandSongs and Allmusic use sequential IDs. This has the problem that when a record is removed (i.e. duplicate, wrong name, etc.), the gap is not filled. Then, the total number of records in the database is always less than the last IDs. In addition, and due to this problem, it is necessary a large IDs edge, to be sure that there are no more records after some empty IDs.

- The network is reported to have 106,095 nodes. Why is it not 143k?

This is explained in the new paragraph in Data cleaning.

- The last column of Table 1 reports that there are 101,112 nodes in the 2010s network. Why is it not 106,095?

This is due to 4,983 artists do not have starting period date. The reason is that they are not included in Allmusic or it does not provide that information for those artists. This line was added to the second paragraph of General characteristics of the covers network section:

“During this process, 4,983 artists were discarded because they did not have that information.”

If there are data processing steps that are causing these different numbers, they should be clearly reported next to each step.

I have decided to insert a fine line to connect each label with the trend. The idea of putting the label at the end is fine, but the problem is that some lines finish very close among them and there is no place for the labels.

Figure 8 is interesting, but it is very difficult to tell which color lines up with which curve. I think a heatmap-style figure like Figure 3 would be more clear.

Lines of the Figure 8 has been replaced by a heat map.

5. There are several places in the paper where it says it will do an analysis that is more than what is actually done

The sentence has been rewritten: “This section analyses who are these figures and how they have evolved.”

- In Top Covered Artists, the paper says “In general, as the cover network evolves over time, the impact of new figures is greater and faster.” This has not been shown in general.

The sentence has been rewritten: “In general, this result suggests that as the cover network evolves over time, the impact of new figures is greater and faster.”

In the following sentence, it also hasn’t been shown that anything is specifically “exponential.”

The sentence has been rewritten: “This would cause that the covering of songs is more and more concentrated in few artists, increasing extremely their impact over time.”

That sentence was removed and replaced by: “These limitations can influence the results relative to the relationship among genders.”

I particularly would like to see the methods questions answered. Addressing the other points would help make the paper’s story more clear and reliable.

-------

Other comments:

- The paper’s language switches tenses often. Sometimes it’s past tense, sometimes it’s present, sometimes it’s conditional. A reliable tense should be used throughout the paper.

Paper has been revised to uniform the tenses

- Network metrics (“weighted in-degree”, “clustering coefficient”) should not be capitalized throughout the paper.

Corrected

- The abstract is jarring: it jumps right into explaining data after barely motivating why the structure of covers is interesting (and it is interesting).

It is possible that the abstract quickly focuses on methods and results, but there a 250 words limitation and I think that it is more relevant to summarize the results than including sentences about the motivations of the study.

It was a typo error. Because density is the proportion of links over the total possible links, then it is common to be expressed in percentages.

Many thanks for the suggested text. I have included it in the paper. I used this criterion to prone the network because I did not find other way to do it. Thank you for your recommendation, I see that disparity filter has a R package, so I will try to use it in next studies.

The exact meaning of the sentence is a speculation mine but based on the previous references. In general, academic literature agrees that the history of contemporary popular music in the 20th century is defined by two differentiate epochs: Jazz and Rock.

This sentence was added to explain those contradictory statements: “Taking into account that the covering of songs descends since 1950s, the fast success of these figures could illustrate an extreme concentration of the impact in special artists thereafter.”

-------

Typos and phrasing (since PLoS ONE doesn’t provide copy editing):

- Page 1, “network metrics have allowed to identify” is missing a word I think

- Page 2, “while enabling to assess the impact” is awkwardly phrased

The sentence was rewritten: “while assessing the impact of artists’ careers in specific musical communities.”

- Page 4, “In this form, it attempts to know how the use of” is awkward with “it attempts to know.” The paper can be more definitive.

The sentence was rewritten: “it attempts to understand how”

- Page 5, In the third research question, “possible to appreciate” is awkward since “appreciate” is usually meant in terms of subjective terms.

Appreciate has been replaced by observe

- Page 5, “the database has been fed,” “fed” is an awkward verb. Databases don’t eat

Fed has been replaced by supplied

That sentence was removed because it is better explained in Page 7

- Page 6, “Styles are not considered as subcategories of genres, then they are used as labels”, something isn’t right with the second half, “then they are used as”

The sentence was rewritten: “therefore, they are labels or keywords freely assigned to an artist.”

- Be consistent with “arcs,” “lines, “edges,” “links.” Aim to use just one

Corrected

- Page 12, very bottom, typo with “in favour of the new Pop/Rock starts.” Should be “stars”

Corrected

- Page 15, very bottom, “For instead” doesn’t seem like the right transition

For instead is used to introduce an example. I think it is fine.

Betweenness centrality has been recalculated taking weights into account (see Table 4). This is also made explicit in the text, replacing betweenness centrality by weighted betweenness centrality.

From the first version, PageRank was calculated using weights. A sentence was added to Social network analysis > PageRank: “It was calculated taking weights into account.”

I think that the discussion about the normalization of covers is very interesting but it goes beyond to the peer-review. Let me go in depth the reasons why the normalization is not suitable in this study:

As I have commented to the first reviewer, the information about the number of songs that Allmusic provides is not entirely correct. Allmusic computes the number of different recordings, even these different sessions are about the same song. This introduce a bias in favour of the most popular singers, because they have more published material than others (i.e. Greatest hits, Live discs, hidden sessions, etc.). I think that this variable is unstable and it is not suitable to normalize the covers.

Second, the normalization in skewed distributions can produce artefacts with unrealistic results. For example, if the normalization is applied, the top 10 artists will be:

Label songs active_period starting_decade main_genre weighted indegree nor weighted indegree

Franz Gruber 2 1890s 1890s Classical 3157 1578.5

Joseph Mohr 2 1890s 1890s Classical 3157 1578.5

Emily Laurey 1 1890s 1890s Classical 1557 1557

Meister Glee Singers 1 1890s 1890s Vocal 983 983

Hollace Shaw 1 1930s 1930s Stage & Screen 975 975

Hiram Sherman 1 1930s 1930s Vocal 966 966

Gloria Grafton 1 1930s 1930s Stage & Screen 944 944

Tally-Ho! 1 1900s 1900s Vocal 919 919

Kathryn Crawford 1 1930s 1930s Stage & Screen 770 770

The result shows that these authors have a very high impact due to by only one or two songs. The first three are due to a popular Christmas song. The other one are unknown singers with only a punctual popular hit. The question is then, are these authors the most important in the contemporary popular music? I do not think so. Then, the normalization in this case provide poor results.

You say: “In the absence of real preferences (tastes or "influence"), the artist would pick a random song from the ones available, and thus would have a higher probability of covering artists with a big production (due to the mass of the production and the age).”

Correct. A random selection means that every song has the same likelihood to be elected, then the distribution will be normal or Gaussian. But the distribution is skewed, then it should be a non-random selection criterion and there is not the same likelihood to be elected. That is, it does not matter how many songs you have if they are not enough goods or interesting, they will not be covered.

I have included a footnote to explain why this normalization was not applied:

“Data supplied by Allmusic. Number of musical tracks recorded and released by artist. It may include different recordings of the same song. Due to this and the skewed distribution of cover songs (Figure 1), this data was not used as normalization variable.”

Minor points:

I have replaced evolutionary by longitudinal

In Objectives, 'indicators' is repeated.

Just before Fig.2, 'The artists that joint the network'--> joined

Corrected

Table 1 publishes the number of nodes and arcs in each time, so the readers can see the difference between each period. It is normal to assume that there are new coming nodes with a low number of covered songs. This is because precedent artists scarcely cover songs from subsequent artists. However, I do not see that new nodes pull down the average because in spite the new nodes are not covered still, the former ones are indeed being covered by the new nodes, then the average can growth (as it does happen in the 1900-1950s period).

I have replaced continuous by constant.

After Fig. 4, "This fact then suggests that the covering of songs is more and more concentrated in few artists, increasing exponentially their impact over time." how is "exponentially" tested?

Exponentially was used as a descriptive adverb. The trend was not tested. I have replace exponentially by extremely.

In Fig. 4 and 5 I find difficult to associate artist labels with curves when there are many (for example The Rolling Stones and Bob Dylan in FIg.5). Maybe arrows could help?

Those figures were edited to introduce arrows between labels and lines.

Fig. 8 is hard to read since blues colors are only slightly different (I can guess due to their size, but it is not straightforward).

Figure 8 was replaced by a heat map

Figure captions seem to include caption titles only. These should be improved by better explaining the variables shown.

Some captions have been rewritten to detail information about the variables.

Attachment

Submitted filename: Response to Reviewers.docx

Click here for additional data file.^{(24.7KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0250212.r005

Decision Letter 2

Nicola Perra

25 Mar 2021

PONE-D-20-26285R2

Cover versions as an impact indicator in popular music: a quantitative network analysis

PLOS ONE

Dear Dr. Ortega,

==============================

I would invite the authors to address the last important point raised by the reviewer number 2.

Many thanks for all the hard work in revising the manuscript

==============================

Please submit your revised manuscript by May 09 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Nicola Perra

Academic Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Reviewer #1: The author has addressed most of my concerns. I appreciate the work that was done to respond to both of the reviewers.

Reviewer #2: I would like to that the author again for implementing the suggested modifications, improving in this way the overall presentation of the manuscript. Based on the latest changes, I would recommend publication in PLoS One --even though I am still a bit puzzled by the entire normalisation argument.

In particular, there are two crucial points from the latest answer:

1) When the author writes "the information about the number of songs that Allmusic provides is not entirely correct. Allmusic computes the number of different recordings, even these different sessions are about the same song (...) I think that this variable is unstable and it is not suitable to normalize the covers." it seems to me that the non-suitability of the possessed data is used to justify the appropriateness of the employed methodology. While the possibility of doing something is obviously constrained by the data one can have access to, I think that this should not be used as a motivation --but rather as the best possible strategy given the circumstances.

2) The results with the applied normalisation are, according to the author, *poor*, since internationally acclaimed artists do not appear on the top. Again, I agree that an artist that did one single famous Christmas song should not be considered the most influential one in contemporary popular music. However, I also find a bit pointless the need to have the most famous artists we all know at the top of the ranking, where we expect them to be. If this is the case, why doing the entire analysis in the first place? I am obviously exaggerating here, but I believe that the answer lies somehow in between.

As such, I think that a little discussion on this in the conclusion is needed, and the normalised table could be put in the SI accompanied by a short text on its caveats.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Iacopo Iacopini

PLoS One. 2021 Apr 19;16(4):e0250212. doi: 10.1371/journal.pone.0250212.r006

Author response to Decision Letter 2

26 Mar 2021

6. Review Comments to the Author

Reviewer #1: The author has addressed most of my concerns. I appreciate the work that was done to respond to both of the reviewers.

In particular, there are two crucial points from the latest answer:

As such, I think that a little discussion on this in the conclusion is needed, and the normalised table could be put in the SI accompanied by a short text on its caveats.

I have added the Limitation section to discuss the normalization problem in skewed distributions and the imprecise definition of songs by Allmusic. In addition, I have added as Supplementary material a table with the top 10 ranking of artists according to the normalized impact by song and a discussion about this result.

Attachment

Submitted filename: Response to Reviewers_2.docx

Click here for additional data file.^{(13KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0250212.r007

Decision Letter 3

Nicola Perra

5 Apr 2021

Cover versions as an impact indicator in popular music: a quantitative network analysis

PONE-D-20-26285R3

Dear Dr. Ortega,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Nicola Perra

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #2: Yes

**********

6. Review Comments to the Author

Reviewer #2: All concerns and suggestions have been addressed by the author. I am happy to recommend the revised version of the manuscript for publication in PLoS One.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: Yes: Iacopo Iacopini

PLoS One. doi: 10.1371/journal.pone.0250212.r008

Acceptance letter

Nicola Perra

7 Apr 2021

PONE-D-20-26285R3

Cover versions as an impact indicator in popular music: a quantitative network analysis

Dear Dr. Ortega:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Nicola Perra

Academic Editor

PLOS ONE

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 File. Full artists indicators.

(CSV)

Click here for additional data file.^{(7.6MB, csv)}

S2 File

(DOCX)

Click here for additional data file.^{(14.9KB, docx)}

S1 Fig

(TIF)

Click here for additional data file.^{(35.6MB, tif)}

Attachment

Submitted filename: Reviewers_response.docx

Click here for additional data file.^{(27.7KB, docx)}

Attachment

Submitted filename: Response to Reviewers.docx

Click here for additional data file.^{(24.7KB, docx)}

Attachment

Submitted filename: Response to Reviewers_2.docx

Click here for additional data file.^{(13KB, docx)}

Data Availability Statement

All Cover songs files are available from the Open Science Foundation database (accession URL https://osf.io/6j2ru/).

[pone.0250212.ref001] 1.Shuker R. Understanding popular music culture. 5th edition. London: Routledge; 2016. [Google Scholar]

[pone.0250212.ref002] 2.Burnett R. The global jukebox: The international music industry. Psychology Press; 1996. [Google Scholar]

[pone.0250212.ref003] 3.Wall T. Studying popular music culture. London: Sage; 2013. [Google Scholar]

[pone.0250212.ref004] 4.Borthwick S, Moy R. Popular music genres: An introduction. London: Routledge; 2020. [Google Scholar]

[pone.0250212.ref005] 5.Schroeder E. Music Databases. Encyclopedia of Library and Information Science: Lib-Pub. 2003:1931.

[pone.0250212.ref006] 6.Cohen M, Baker G. Takin’Care of Business: Music Business Reference. Music Reference Services Quarterly. 2015, 18(3–4): 157–163. [Google Scholar]

[pone.0250212.ref007] 7.Weyde T, Cottrell S, Dykes J, Benetos E, Wolff D, Tidhar D, et al. Big data for musicology. In: Proceedings of the 1st International Workshop on Digital Libraries for Musicology; 2014. p. 1–3.

[pone.0250212.ref008] 8.Cottrell S. Big Music Data, Musicology, and the Study of Recorded Music: Three Case Studies. The Musical Quarterly. 2018, 101(2–3): 216–243. [Google Scholar]

[pone.0250212.ref009] 9.Rose S, Tuppen S, Drosopoulou L. Writing a Big Data history of music. Early Music. 2015, 43(4): 649–660. [Google Scholar]

[pone.0250212.ref010] 10.Park D, Bae A, Schich M, Park J. Topology and evolution of the network of western classical music composers. EPJ Data Science. 2015, 4(1): 2. [Google Scholar]

[pone.0250212.ref011] 11.McAndrew S, Everett M. Music as collective invention: A social network analysis of composers. Cultural Sociology. 2015, 9(1): 56–80. [Google Scholar]

[pone.0250212.ref012] 12.Gleiser PM, Danon L. Community structure in jazz. Advances in complex systems. 2003, 6(04): 565–573. [Google Scholar]

[pone.0250212.ref013] 13.Budner P, Grahl J. Collaboration networks in the music industry. arXiv:1611.00377 [Preprint]. 2016 https://arxiv.org/abs/1611.00377

[pone.0250212.ref014] 14.Makkonen T. North from here: The collaboration networks of Finnish metal music genre superstars. Creative Industries Journal. 2017, 10(2): 104–118. [Google Scholar]

[pone.0250212.ref015] 15.Napier K, Shamir L. Quantitative sentiment analysis of lyrics in popular music. Journal of Popular Music Studies. 2018, 30(4): 161–176. [Google Scholar]

[pone.0250212.ref016] 16.Ruth N. “Where is the love?” Topics and prosocial behavior in German popular music lyrics from 1954 to 2014. Musicae Scientiae. 2019, 23(4): 508–524. [Google Scholar]

[pone.0250212.ref017] 17.Collins N. Influence in Early Electronic Dance Music: An Audio Content Analysis Investigation. ISMIR. 2012: 1–6.

[pone.0250212.ref018] 18.Mauch M, MacCallum RM, Levy M, Leroi AM. The evolution of popular music: USA 1960–2010. Royal Society open science. 2015, 2(5): 150081. 10.1098/rsos.150081 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0250212.ref019] 19.Klimek P, Kreuzbauer R, Thurner S. Fashion and art cycles are driven by counter-dominance signals of elite competition: quantitative evidence from music styles. Journal of the Royal Society Interface. 2019, 16(151): 20180731. 10.1098/rsif.2018.0731 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0250212.ref020] 20.Youngblood M. Conformity bias in the cultural transmission of music sampling traditions. Royal Society open science. 2019, 6(9): 191149. 10.1098/rsos.191149 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0250212.ref021] 21.Plasketes G. Re-flections on the Cover Age: A Collage of Continuous Coverage in Popular Music. Popular Music and Society. 2005, 28(2): 137–161. [Google Scholar]

[pone.0250212.ref022] 22.Cusic D. In defense of cover songs. Popular Music and Society. 2005, 28(2): 171–177. [Google Scholar]

[pone.0250212.ref023] 23.Solis G. I did it my way: rock and the logic of covers. Popular Music and Society. 2010, 33(3): 297–318. [Google Scholar]

[pone.0250212.ref024] 24.Doktor SD. Covering the Tracks: Exploring Cross-gender Cover Songs of the Rolling Stones "Satisfaction" [Doctoral dissertation]. University of Georgia; 2008.

[pone.0250212.ref025] 25.Chu YW, Leung E. Remapping Hong Kong popular music: covers, localisation and the waning hybridity of Cantopop. Popular Music. 2013, 32(1): 65–78. [Google Scholar]

[pone.0250212.ref026] 26.Anderson K. Song ‘adaptations’ and the globalisation of French pop, 1960–1970. French Cultural Studies. 2015, 26(3): 330–342. [Google Scholar]

[pone.0250212.ref027] 27.SecondHandSongs [Internet]. Introduction to the database. 2020. https://secondhandsongs.com/page/Introduction

[pone.0250212.ref028] 28.Smith E. Motherboard. The Story of AllMusic, the Internet’s Largest, Most Influential Music Database. 2016. https://www.vice.com/en_us/article/53djj8/the-story-of-allmusic-the-internets-largest-most-influential-music-database

[pone.0250212.ref029] 29.Freeman LC. Centrality in social networks conceptual clarification. Social networks. 1978, 1(3): 215–239. [Google Scholar]

[pone.0250212.ref030] 30.Brin S, Page L. The anatomy of a large-scale hypertextual Web search engine. Computer networks and ISDN systems. 1998, 30(1–7): 107–117. [Google Scholar]

[pone.0250212.ref031] 31.Opsahl T, Agneessens F, Skvoretz J. Node centrality in weighted networks: Generalizing degree and shortest paths. Social networks. 2010, 32(3): 245–251. [Google Scholar]

[pone.0250212.ref032] 32.Gronow P. The record industry: the growth of a mass medium. Popular music. 1983, 3: 53–75. [Google Scholar]

[pone.0250212.ref033] 33.White DR, Borgatti SP. Betweenness centrality measures for directed graphs. Social networks. 1994, 16(4): 335–346. [Google Scholar]

[pone.0250212.ref034] 34.Pitts M, Hoffmann F, Carty D, Bedoian J. The Rise of the Crooners: Gene Austin, Russ Columbo, Bing Crosby, Nick Lucas, Johnny Marvin and Rudy Vallee. Scarecrow Press; 2001. [Google Scholar]

[pone.0250212.ref035] 35.Garofalo R. Crossing over: From black rhythm & blues to white rock ‘n’roll. In Kelley N, editor. Rhythm and business: The political economy of black music. New York: Akashit Books; 2002, pp. 112–137. [Google Scholar]

[pone.0250212.ref036] 36.Kellett A. The British Blues Network: Adoption, Emulation, and Creativity. Ann Arbor: University of Michigan Press; 2017. [Google Scholar]

[pone.0250212.ref037] 37.Scheuer TE. American Popular Music: Readings from the Popular Press. Bowling Green: UP; 1989. [Google Scholar]

[pone.0250212.ref038] 38.Nicholls D., editor. The Cambridge history of American music (The Cambridge History of Music), Cambridge, UK. Cambridge University Press, 653; 1998. [Google Scholar]

[pone.0250212.ref039] 39.Kania A. All Play and No Work: An Ontology of Jazz. The Journal of Aesthetics and Art Criticism. 2011, 69(4): 391–403. [Google Scholar]

[pone.0250212.ref040] 40.Dineley S. Covers Uncovered: A History of the" Cover Version," from Bing Crosby to the Flaming Lips. [Doctoral dissertation]. Wester University; 2014. https://ir.lib.uwo.ca/cgi/viewcontent.cgi?article=3630&context=etd

[pone.0250212.ref041] 41.Homan S. ’You’ve Got to Carry that Weight’: Tribute Acts in the Entertainment Supermarket. In Homan S, editor. Access All Eras: Tribute Bands and Global Pop Culture. New York: Open University Press; 2006, pp. 32–51. [Google Scholar]

PERMALINK

Cover versions as an impact indicator in popular music: A quantitative network analysis

José Luis Ortega

Roles

Abstract

Introduction

Related research

Objectives

Methods

Data sources

SecondHandSongs

Allmusic

Data extraction

Data cleaning

Social network analysis

Results

General characteristics of the covers network

Fig 1. Cumulative distribution function (CDF) of the number of received (indegree) and performed (outdegree) covers by artist (log-log), and their respective exponents (α).

Fig 2. Evolution of the weighted average degree and clustering coefficient over the decades.

Table 1. Network indicators evolution by decade.

Fig 3. Heat map with the percentage of covers performed by artists that started their career in certain decade (x) to artists from other decades (y).

Top covered artists

Table 2. Top 10 artists by number of covered songs (weighted indegree).

Fig 4. Evolution of the number of covered songs (weighted indegree) of the first ten artists in each decade.

Table 3. Top 10 artists by PageRank.

Fig 5. Evolution of the PageRank of the first ten artists in each decade.

Table 4. Top 10 artists by weighted betweenness centrality.

Genres and styles

Fig 6. Directed network graph of artists (N = 106,095) linked through covers (N = 855,650), coloured according to music genre, with node size according to number of recorded songs (Layout = Fruchterman Reingold).

Fig 7. Cover flows between music genres (N = 35) (arcs≥10% of outdegree) (Layout = Yifan Hu).

Fig 8. Heat map with the percentage of covered music genres by decade.

Discussion

Limitations

Conclusions

Supporting information

Data Availability

Funding Statement

References

Decision Letter 0

Nicola Perra

Roles

Author response to Decision Letter 0

Decision Letter 1

Nicola Perra

Roles

Author response to Decision Letter 1

Decision Letter 2

Nicola Perra

Roles

Author response to Decision Letter 2

Decision Letter 3

Nicola Perra

Roles

Acceptance letter

Nicola Perra

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases