Cultural transmission modes of music sampling traditions remain stable despite delocalization in the digital age
CCultural transmission modes of music sampling traditions remain stabledespite delocalization in the digital age
Mason Youngblood a,b,1aDepartment of Psychology, The Graduate Center, City University of New York, New York, NY, USA b Department of Biology, Queens College, City University of New York, Flushing, NY, USA [email protected] Abstract
Music sampling is a common practice among hip-hop and electronic producers that has played a critical role in thedevelopment of particular subgenres. Artists preferentially sample drum breaks, and previous studies have suggested thatthese may be culturally transmitted. With the advent of digital sampling technologies and social media the modes ofcultural transmission may have shifted, and music communities may have become decoupled from geography. The aimof the current study was to determine whether drum breaks are culturally transmitted through musical collaborationnetworks, and to identify the factors driving the evolution of these networks. Using network-based diffusion analysis wefound strong evidence for the cultural transmission of drum breaks via collaboration between artists, and identified severaldemographic variables that bias transmission. Additionally, using network evolution methods we found evidence that thestructure of the collaboration network is no longer biased by geographic proximity after the year 2000, and that genderdisparity has relaxed over the same period. Despite the delocalization of communities by the internet, collaborationremains a key transmission mode of music sampling traditions. The results of this study provide valuable insight intohow demographic biases shape cultural transmission in complex networks, and how the evolution of these networks hasshifted in the digital age.
Keywords – cultural evolution, transmission modes, music sampling, delocalization, network analysis
Introduction
Music sampling, or the use of previously-recorded mate-rial in a new composition, is a nearly ubiquitous practiceamong hip-hop and electronic producers. The usage of drumbreaks, or percussion-heavy sequences, ripped from soul andfunk records has played a particularly critical role in thedevelopment of certain subgenres. For example, “Amen,Brother”, released by The Winstons in 1969, is widely re-garded as the most sampled song of all time. Its iconic 4-bardrum break has been described as “genre-constitutive” [1],and can be prominently heard in classic hip-hop and junglereleases by N.W.A and Shy FX [2]. Due to the consistentusage of drum breaks in particular music communities andsubgenres [1–5] some scholars have suggested that they maybe culturally transmitted [6], which could occur as a directresult of collaboration between artists or as an indirect effectof community membership.Before the digital age, artists may have depended uponcollaborators for access to the physical source materials andexpensive hardware required for sampling [7]. In the 1990s,new technologies like compressed digital audio formats anddigital audio workstations made sampling more accessibleto a broader audience [8]. Furthermore, the widespreadavailability of the internet and social media have delocal-ized communities [9], and allowed global music “scenes” toform around shared interests beyond peer-to-peer file shar- ing [10, 11]. Individuals in online music communities nowhave access to the collective knowledge of other members[12, 13], and there is evidence that online communities play akey role in music discovery [14]. Although musicians remainconcentrated in historically important music cities (i.e. NewYork City and Los Angeles in the United States) [15, 16],online music communities also make it possible for artiststo establish collaborative relationships independently of ge-ographic location [17]. If more accessible sampling technolo-gies and access to collective knowledge have allowed artiststo discover sample sources independently of collaboration[18], then the strength of cultural transmission via collabo-ration may have decreased over the last couple of decades.Similarly, if online music communities have created opportu-nities for interactions between potential collaborators, thengeographic proximity may no longer structure musical col-laboration networks.Studies of the cultural evolution of music have primar-ily investigated diversity in musical performances [19] andtraditions [20], macro-scale patterns and selective pressuresin musical evolution [21–24], and the structure and evolu-tion of consumer networks [14, 25, 26]. Although severaldiffusion chain experiments have addressed how cognitivebiases shape musical traits during transmission [27–29], fewstudies have investigated the mechanisms of cultural trans-mission at the population level [30, 31]. The practice ofsampling drum breaks in hip-hop and electronic music is1 a r X i v : . [ s t a t . A P ] J a n n ideal research model for cultural transmission becauseof (1) the remarkably high copy fidelity of sampled mate-rial, (2) the reliable documentation of sampling events, and(3) the availability of high-resolution collaboration and de-mographic data for the artists involved. Exhaustive onlinedatasets of sample usage and collaboration make it possibleto reconstruct networks of artists and track the diffusionof particular drum breaks from the early 1980s to today.Furthermore, the technological changes that have occurredover the same time period provide a natural experiment forhow the digital age has impacted cultural transmission morebroadly [32].The aim of the current study was to determine whetherdrum breaks are culturally transmitted through musical col-laboration networks, and to identify the factors driving theevolution of these networks. We hypothesized that (1) drumbreaks are culturally transmitted through musical collabo-ration networks, and that (2) the strength of cultural trans-mission via collaboration would decrease after the year 2000.For clarification, the alternative to the first hypothesis iscultural transmission occurring outside of collaborative re-lationships (i.e. independent sample discovery via “crate-digging” in record stores or online). Previous studies haveinvestigated similar questions using diffusion curve analy-sis [30], but the validity of inferring transmission mecha-nisms from cumulative acquisition data has been called intoquestion [33]. Instead, we applied network-based diffusionanalysis (NBDA), a recently developed statistical methodfor determining whether network structure biases the emer-gence of a novel behavior in a population [34]. As NBDAis most useful in identifying social learning, an ability thatis assumed to be present in humans, it has been primar-ily applied to non-human animal models such as birds,whales, and primates [35–37], but the ability to incorpo-rate individual-level variables to nodes makes it uniquelysuited to determining what factors bias diffusion more gen-erally. Additionally, we hypothesized that (3) collaborationprobability would be decoupled from geographic proximityafter the year 2000. To investigate this we applied separabletemporal exponential random graph modeling (STERGM),a dynamic extension of ERGM for determining the variablesthat bias network evolution [38]. Methods
All data used in the current study were collected in Septem-ber of 2018, in compliance with the terms and conditions ofeach database. For the primary analysis, the three mostheavily sampled drum breaks of all time, “Amen, Brother”by The Winstons, “Think (About It)” by Lyn Collins, and“Funky Drummer” by James Brown, were identified usingWhoSampled . The release year and credits for each songlisted as having sampled each break were collected usingdata scraping. In order to avoid name disambiguation, only artists, producers, and remixers with active Discogs linksand associated IDs were included in the dataset. In or-der to investigate potential shifts in transmission strengtharound 2000, the same method was used to collect data forthe eight songs in the “Most Sampled Tracks” on WhoSam-pled that were released after 1990 (see ?? ). One of these,“I’m Good” by YG, was excluded from the analysis becausethe sample is primarily used by a single artist. Each set ofsampling events collected from WhoSampled was treated asa separate diffusion. All analyses were conducted in R (v3.3.3).Collaboration data were retrieved from Discogs , a crowd-sourced database of music releases. All collaborative re-leases in the database were extracted and converted to amaster list of pairwise collaborations. For each diffusion,pairwise collaborations including two artists in the datasetwere used to construct collaboration networks, in whichnodes correspond to artists and weighted links correspondto collaboration number. Although some indirect connec-tions between artists were missing from these subnetworks,conducting the analysis with the full dataset was compu-tationally prohibitive and incomplete networks have beenroutinely used for NBDA in the past [35, 36, 39].Individual-level variables for artists included in each col-laboration network were collected from MusicBrainz , acrowdsourced database with more complete artist informa-tion than Discogs, and Spotify , one of the most popularmusic streaming services. Gender and geographic locationwere retrieved from the Musicbrainz API. Whenever it wasavailable, the “begin area” of the artist, or the city in whichthey began their career, was used instead of their “area”,or country of affiliation, to maximize geographic resolution.Longitudes and latitudes for each location, retrieved usingthe Data Science Toolkit and Google Maps, were used to cal-culate each artist’s mean geographic distance from other in-dividuals. Albunack , an online tool which draws from bothMusicbrainz and Discogs, was used to convert IDs betweenthe two databases. Popularity and followers were retrievedusing the Spotify API. An artist’s popularity, a proprietaryindex of streaming count that ranges between 0 and 100,is a better indicator of their long-term success because itis calculated across their entire discography. Followers is abetter indicator of current success because it reflects userengagement with artists who are currently more active onthe platform. Discogs IDs are incompatible with the SpotifyAPI, so artist names were URL-encoded and used as textsearch terms.In order to identify whether social transmission betweencollaborators played a role in sample acquisition, order ofacquisition diffusion analysis (OADA) was conducted us-ing the R script for NBDA (v 1.2.13) provided on the La-land lab’s website . OADA uses the order in which indi- https://musicbrainz.org/ https://lalandlab.st-andrews.ac.uk/freeware/ c ). Models with a∆AIC c < <
2) with the most individual-level variableswere run separately to assess the effects of each variable onnetwork evolution.
Results
The three most heavily sampled drum breaks of all time werecollectively sampled 6530 times ( n = 2966, n = 2099, n = 1465). 4462 (68.33%) of these sampling events were asso-ciated with valid Discogs IDs, corresponding to 2432 uniqueartists (F: n = 143, 5.88%; M: n = 1342, 55.18%; Other orNA: n = 947, 38.94%), and included in the primary OADAand STERGM. The eight samples released after 1990 werecollectively sampled 1752 times ( n = 284, n = 260, n =248, n = 198, n = 194, n = 193, n = 192, n = 182).1305 (74.53%) of these sampling events were associated withvalid Discogs IDs, corresponding to 1270 unique artists, andincluded in the additional OADA. NBDA
The best fitting model from the primary OADA, which wasmultiplicative and included all four individual-level vari-ables, can be seen in Table 1. In support of our first hy-pothesis, a likelihood ratio test found strong evidence forsocial transmission over asocial learning (∆AIC c = 141; p < p < p < p < p = 0.89). The diffusionnetwork and diffusion curve for all three drum breaks in-cluded in the primary OADA are shown in Figures 1 andS1, respectively. All other models fit to the primary OADAcan be found in the supporting information.The results of the additional OADA, conducted using theseven diffusions from after 1990, can be found in the sup-porting information. A likelihood ratio test found strongevidence for social transmission overall (∆AIC c = 88; p < = 0.20, p = 0.31) or median year of diffusion and social transmissionestimate (R = 0.17, p = 0.36) (see Figure S2). STERGM
For both time periods the second best fitting STERGMmodels (∆AIC <
2) included all four individual-level vari-ables, the results of which can be seen in Table 2. All othermodels, including those assuming different transition years,can be found in the supporting information. Across bothperiods there appears to be homophily based on popularity( p < p < p s < ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ●●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ●● ● ● ●●● ● ● ●● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ●● ● ● ●● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●●● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ●●● ● ● ●● ● ●●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ●● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ●● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●●● ● ●●● ●● ●● ● ●● ● ● ● ●● ● ●● ●● ● ● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●●● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●●● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ●● ●● ● ● ●● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ●●● ● ● ●● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● Uninformed Newly informed Previously informed
Figure 1: The diffusion of all three drum breaks through the combined collaboration network. At each time pointindividuals who have not yet used one of the drum breaks (informed) are shown in white, individuals who first used oneof the drum breaks in a previous time step (previously informed) are shown in blue, and individuals who first used one ofthe drum breaks in the current time step (newly informed) are shown in red.4 ultiplicative Model - Order of AcquisitionEstimate Effect size p Gender -0.11 0.81 < < < c p With social transmission 14719 < Table 1: The results of the multiplicative model for theOADA including all individual-level variables. The toppanel shows the model estimate, effect size, and p -value foreach individual-level variable. The bottom panel shows theAIC c for the model with and without social transmissionand the p -value from the likelihood ratio test.support of our third hypothesis, mean distance negativelypredicts link formation only before 2000 ( p < p < = 0.048, p < =0.090, p < Discussion
Using high-resolution collaboration and longitudinal diffu-sion data, we have provided the first quantitative evidencethat music samples are culturally transmitted via collab-oration between artists. Additionally, in support of thewidespread assertion that the internet has delocalized artistcommunities, we have found evidence that geographic prox-imity no longer biases the structure of musical collabora-tion networks after the year 2000. Given that the strengthof transmission has not weakened over the same time pe-
STERGM 1984-1999 2000-2017Effect size p Effect size p Gender (F) 6.86 < < < < < < < < Table 2: The results of the STERGM analyses for beforeand after 2000. The table shows the effect size and p -valuefor gender, popularity, followers, and mean distance duringeach time period.riod, this finding indicates that collaboration remains a keycultural transmission mode for music sampling traditions.This result supports the idea that the internet has enhancedrather than disrupted existing social interactions [9].Gender appears to play a key role in both network struc-ture and cultural transmission. Across the entire time pe-riod, collaborations were more likely to occur between indi-viduals of the same gender. Additionally, the probability ofcultural transmission appears to be much higher for femaleartists. This effect could be a result of the much higher levelsof homophily among women before 2000. Previous work hassuggested that high levels of gender homophily are associ-ated with gender disparity [43–45], which is consistent withthe historic marginalization of women in music productioncommunities [1, 10, 46]. Although the proportion of femaleartists in the entire dataset is extremely low ( ∼ Acknowledgments
I would like to thank David Lahti and Carolyn Pytte, as wellas all members of the Lahti lab, for their valuable conceptualand analytical feedback.
Data Availability Statement
All R scripts and data used in the study are available inthe Harvard Dataverse repository: https://doi.org/10.7910/DVN/Q02JJQ . References
1. Whelan, A. M. in
Gender and Music
M/C Journal (2007).3. Frane, A. V. Swing Rhythm in Classic Drum Breaks From Hip-Hop’sBreakbeat Canon. Music Perception ® ? Remixing musical fu-tures. British Journal of Music Education
Organised Sound Organised Sound
Journal of Intellectual Property Law
Sharing Sounds: Musical Innovation, Collabora-tion, and Ideological Expression in the Chilean Netlabel Movement
PhD thesis (University of Michigan, 2017).9. DiMaggio, P., Hargittai, E., Neuman, W. R. & Robinson, J. P. SocialImplications of the Internet.
Annual Review of Sociology
The Computerization of Practice in Peripheral MusicCommunities
PhD thesis (Simon Fraser University, 2003).11. Alexandraki, C. & Valsamakis, N. in
Virtual Community Practicesand Social Interactive Media: Technology Lifecycle and WorkflowAnalysis .12. Salavuo, M. Open and informal online communities as forums of col-laborative musical activities and learning.
British Journal of MusicEducation
Social Considerations in Online Communi-ties: Usability, Sociability, and Success Factors
PhD thesis (2002),1–46.14. Garg, R., Smith, M. D. & Telang, R. Measuring Information Diffu-sion in an Online Community.
Journal of Management InformationSystems
Journal of Planning Education andResearch
Sounds of the Underground .17. Kruse, H. Local Identity and Independent Music Scenes, Online andOff.
Popular Music and Society
Or-ganised Sound et al.
Identifying style-types in a sample of musical im-provisations using dimensional reduction and cluster analysis.
Psy-chology of Aesthetics, Creativity, and the Arts
Ethnomusicology
Royal SocietyOpen Science doi: (2015).
2. Percino, G., Klimek, P. & Thurner, S. Instrumentational complexityof music genres and why simplicity sells.
PLoS ONE Proceedingsof the National Academy of Sciences
Proceedings of the NationalAcademy of Sciences
Mining the Dynamics of Music Prefer-ences from a Social Networking Site in (Athens,Greece, 2009). doi: .26. Monechi, B., Gravino, P., Servedio, V. D., Tria, F. & Loreto, V. Sig-nificance and popularity in music production. Royal Society OpenScience doi: (2017).27. Verhoef, T. The origins of duality of patterning in artificial whistledlanguages. Language and Cognition Nature Human Behaviour doi: (2017).29. Lumaca, M. & Baggio, G. Cultural transmission and evolution ofmelodic structures in multi-generational signaling games. ArtificialLife
Sociological Methodology
ArXiv e-prints. arXiv: (Sept.2018).32. Acerbi, A. A cultural evolution approach to digital media.
Frontiersin Human Neuroscience
636 (2016).33. Laland, K. N. & Kendal, J. R. in
The Biology of Traditions: Modelsand Evidence (eds Fragaszy, D. M. & Perry, S.) 33–55 (CambridgeUniversity Press, Cambridge, 2003).34. Franz, M. & Nunn, C. L. Network-based diffusion analysis: a newmethod for detecting social learning.
Proceedings of the Royal So-ciety B: Biological Sciences et al.
Experimentally induced innovations lead to per-sistent culture via conformity in wild birds.
Nature
Science
PLoS Biology e1001960(2014).38. Krivitsky, P. N. & Handcock, M. S. A separable model for dynamicnetworks.
Journal of the Royal Statistical Society
Proceedings of the Royal Society B: Biological Sciences
Journal of Theoretical Biology
Model selection and multimodelinference: a practical information-theoretic approach isbn : 0387953647 (Springer, 2002).42. Hunter, D. R., Goodreau, S. M. & Handcock, M. S. Goodness ofFit of Social Network Models.
Journal of the American StatisticalAssociation
Human Resource Management,
Environment and Planning A: Economy andSpace
Advances in Complex Systems doi: (2018). 46. Baker, S. & Cohen, B. M. Z. From snuggling and snogging to sam-pling and scratching: Girls’ nonparticipation in community-basedmusic activities.
Youth & Society
American SociologicalReview
International Journal of Research in Marketing
Club Cultures: Music, Media and Subcultural Capital (Polity Press, Cambridge, 1995).50. Hesmondhalgh, D. The British Dance Music Industry: A Case Studyof Independent Cultural Production.
The British Journal of Soci-ology
Popular Music
Poetics
Culture and the Evolutionary Process (The University of Chicago Press, Chicago, 1985).54. Smith, J. A. & Moody, J. Structural effects of network samplingcoverage I: Nodes missing at random.
Social Networks
Social Networks
Frontiers in Psychology doi: (2016).57. Hoppitt, W. The conceptual foundations of Network-Based DiffusionAnalysis: choosing networks and interpreting results. PhilosophicalTransactions of the Royal Society B doi: (2017).58. Wild, S. & Hoppitt, W. Choosing a sensible cut-off point: assessingthe impact of uncertainty in a social network on the performance ofNBDA.
Primates. doi: (2018).59. Franz, M. & Nunn, C. L. Investigating the impact of observation er-rors on the statistical performance of network-based diffusion anal-ysis.
Learning & Behavior
Computational Strategies forBreakbeat Classification and Resequencing in Hardcore, Jungleand Drum & Bass in Proc. of the 18th Int. Conference on DigitalAudio Effects (DAFx-15) (2015). upporting information NBDA
The results of the multiplicative NBDA model fit to the primary OADA with all four individual-level variables are shownbelow.
Summary o f M u l t i p l i c a t i v e S o c i a l Transmission ModelOrder o f a c q u i s i t i o n dataUnbounded p a r a m e t e r i s a t i o n .C o e f f i c i e n t s : Estimate Bounded se z pS o c i a l t r a n s m i s s i o n 1 1.334607 e −
01 0.1177462 NA NA NAgender − −
01 NA 4.017186 e − − −
03p o p u l a r i t y − −
02 NA 1.465003 e − − − −
08 NA 1.956053 e − − − − −
09 NA 1.361273 e − − −
01L i k e l i h o o d Ratio Test f o r S o c i a l Transmission :Null model i n c l u d e s a l l other s p e c i f i e d v a r i a b l e sS o c i a l t r a n s m i s s i o n and a s o c i a l l e a r n i n g assumed to combine m u l t i p l i c a t i v e l yDf LogLik AIC AICc LR pWith S o c i a l Transmission 5 7354.4 14719 14719 143.09 0Without S o c i a l Transmission 4 7425.9 14860 14860
The results of all NBDA models fit to the primary OADA. In the “Additive?” column TRUE means the model wasadditive, FALSE means the model was multiplicative, and NA means the model was asocial. In the “ILVs”, or individual-level variables, column the numbers correspond to the variables included in the model (1: gender; 2: popularity; 3:followers; 4: mean distance).
Additive ? ILVs S o c i a l ? AICc deltaAICcFALSE 1 2 3 4 s o c i a l 14718.7305939155 0FALSE 2 3 4 s o c i a l 14723.2581626229 4.53FALSE 1 2 4 s o c i a l 14749.7210576547 30.99FALSE 2 4 s o c i a l 14755.0075144196 36.28FALSE 1 3 4 s o c i a l 14795.6248094134 76.89FALSE 3 4 s o c i a l 14796.5052668663 77.77FALSE 4 s o c i a l 15956.6213244064 1237.89FALSE 1 4 s o c i a l 15957.1346280933 1238.4FALSE 1 2 3 s o c i a l 32229.4897515171 17510.76FALSE 1 2 s o c i a l 32260.0747802103 17541.34FALSE 2 3 s o c i a l 32267.3112186711 17548.58FALSE 2 s o c i a l 32298.2544538564 17579.52FALSE 1 3 s o c i a l 32356.8941850595 17638.16FALSE 3 s o c i a l 32389.9528866174 17671.22FALSE 1 s o c i a l 43609.4787479712 28890.75TRUE 1 s o c i a l 43612.45988987 28893.73NA 0 s o c i a l 43662.7306996825 28944NA 1 a s o c i a l 43779.4569488198 29060.73NA 0 a s o c i a l 43815.27922266 29096.55TRUE 2 s o c i a l I n f I n fNA 2 a s o c i a l I n f I n fTRUE 1 2 s o c i a l I n f I n fNA 1 2 a s o c i a l I n f I n fTRUE 3 s o c i a l I n f I n fNA 3 a s o c i a l I n f I n fTRUE 1 3 s o c i a l I n f I n fNA 1 3 a s o c i a l I n f I n fTRUE 2 3 s o c i a l I n f I n fNA 2 3 a s o c i a l I n f I n fTRUE 1 2 3 s o c i a l I n f I n fNA 1 2 3 a s o c i a l I n f I n fTRUE 4 s o c i a l I n f I n fNA 4 a s o c i a l I n f I n fTRUE 1 4 s o c i a l I n f I n fNA 1 4 a s o c i a l I n f I n fTRUE 2 4 s o c i a l I n f I n fNA 2 4 a s o c i a l I n f I n fTRUE 1 2 4 s o c i a l I n f I n fNA 1 2 4 a s o c i a l I n f I n fTRUE 3 4 s o c i a l I n f I n fNA 3 4 a s o c i a l I n f I n f RUE 1 3 4 s o c i a l I n f I n fNA 1 3 4 a s o c i a l I n f I n fTRUE 2 3 4 s o c i a l I n f I n fNA 2 3 4 a s o c i a l I n f I n fTRUE 1 2 3 4 s o c i a l I n f I n fNA 1 2 3 4 a s o c i a l I n f I n f
Year P r opo r t i on I n f o r m ed Figure S1: The combined diffusion curve for all three drum breaks included in the primary OADA. The proportion ofinformed individuals is on the y -axis, and the year is on the x -axis. Although recent research suggests that inferringacquisition modes from diffusion curves is unreliable, it appears that the curve may have the S -shape indicative of socialtransmission prior to the early-2000s. The eight songs in the “Most Sampled Tracks” on WhoSampled that were released after 1990 are shown below. The fifthsong, “I’m Good” by YG, was excluded from the additional OADA because it is a producer tag used by a single artist.1. “Crash Goes Love (Yell Apella)” by Loleatta Holloway (1992)2. “Shook Ones Part II” by Mobb Deep (1994)3. “C.R.E.A.M.” by Wu-Tang Clan (1993)4. “Sound of Da Police” by KRS-One (1993)5. “I’m Good” by YG (2011) [excluded producer tag]6. “Juicy” by The Notorious B.I.G. (1994)7. “Sniper” by DJ Trace and Pete Parsons (1999)8. “Who U Wit?” by Lil Jon and The East Side Boyz (1997)The results of the additive NBDA model fit to the additional OADA are shown below. Remember that the fifth song wasexcluded, so the transmission estimates for five, six, and seven here are actually for six, seven, and eight.
Summary o f Additive S o c i a l Transmission ModelOrder o f a c q u i s i t i o n dataUnbounded p a r a m e t e r i s a t i o nC o e f f i c i e n t s Estimate BoundedS o c i a l t r a n s m i s s i o n 1 0.13558602 0.11939740S o c i a l t r a n s m i s s i o n 2 0.28805974 0.22363849S o c i a l t r a n s m i s s i o n 3 0.05184340 0.04928814S o c i a l t r a n s m i s s i o n 4 0.60154771 0.37560399S o c i a l t r a n s m i s s i o n 5 0.06816578 0.06381573S o c i a l t r a n s m i s s i o n 6 0.07600555 0.07063677S o c i a l t r a n s m i s s i o n 7 0.01547410 0.01523830L i k e l i h o o d Ratio Test f o r S o c i a l Transmission :Null model i n c l u d e s a l l other s p e c i f i e d v a r i a b l e sS o c i a l t r a n s m i s s i o n and a s o c i a l l e a r n i n g assumed to combine a d d i t i v e l yDf LogLik AIC AICc LR pWith S o c i a l Transmission 7 6450 12914 12914 101.99 0Without S o c i a l Transmission 0 6501 13002 13002
005 2010 201500.10.20.30.40.50.6 2005 2010 2015
Mean Year Median Year S o c i a l T r an s m i ss i on E s t i m a t e Figure S2: The relationship between diffusion years and transmission strengths for all seven diffusions included in theadditional OADA. The mean (left) and median (right) years of diffusion are on the x -axis, and the social transmissionestimates from the additive model are on the y -axis. Linear regression found no significant relationships between eithermean year of diffusion and social transmission estimate (R = 0.20, p = 0.31) or median year of diffusion and socialtransmission estimate (R = 0.17, p = 0.36). STERGM
The results of all formation models of the STERGM fit to the data from 1984-1999. In the “ILVs”, or individual-levelvariables, column the numbers correspond to the variables included in the model (1: gender; 2: popularity; 3: followers;4: mean distance).
ILVs AIC deltaAIC1 2 4 7676.380 0.000001 2 3 4 7678.174 1.793481 2 7693.425 17.044731 2 3 7695.203 18.822401 4 7702.003 25.623001 3 4 7702.160 25.779212 4 7710.403 34.022562 3 4 7711.909 35.528981 7718.147 41.766951 3 7718.390 42.009532 7727.194 50.813232 3 7728.701 52.320174 7737.373 60.992493 4 7738.094 61.713190 7753.228 76.847423 7753.985 77.60457
The results of the best-fitting formation model of the STERGM with the most individual-level variables fit to the datafrom 1984-1999. ==========================Summary o f model f i t==========================Formula : y . form ˜ edges + nodecov (” meandist ”) + a b s d i f f (” p o p u l a r i t y ”) +a b s d i f f (” f o l l o w e r s ”) + nodematch (” gender ” , d i f f = TRUE) < environment : 0x1cb31da8 > I t e r a t i o n s : 11 out o f 20Monte Carlo MLE Results :Estimate Std . Error MCMC % z value Pr( > | z | )edges − −
01 0 − < − ∗∗∗ nodecov . meandist − −
07 3.371 e −
08 0 − ∗∗∗ a b s d i f f . p o p u l a r i t y − −
02 3.284 e −
03 0 − < − ∗∗∗ a b s d i f f . f o l l o w e r s 1.194 e −
08 2.579 e −
08 0 0.463 0.643396nodematch . gender . − −
01 0 5.851 < − ∗∗∗ nodematch . gender . 0 3.478 e −
01 1.965 e −
01 0 1.770 0.076773 .nodematch . gender . 1 5.313 e −
01 1.117 e −
01 0 4.757 < − ∗∗∗ −−− S i g n i f . codes : 0 ’ ∗∗∗ ’ 0.001 ’ ∗∗ ’ 0.01 ’ ∗ ’ 0.05 ’ . ’ 0 . 1 ’ ’ 1 ull Deviance : 6349494 on 4580192 d e g r e e s o f freedomResidual Deviance : 7664 on 4580185 d e g r e e s o f freedomAIC : 7678 BIC : 7772 ( Smaller i s b e t t e r . ) The results of the goodness-of-fit analysis of the formation model of the STERGM with the most individual-level variablesfit to the data from 1984-1999 are below.
Goodness − of − f i t f o r degreeobs min mean max MC p − value0 10698 10388 10484.50 10553 0.001 801 1032 1101.97 1200 0.002 148 89 104.41 126 0.003 43 16 20.71 29 0.004 15 5 8.01 11 0.005 0 0 0.38 2 1.006 6 0 0.95 2 0.007 1 0 0.07 1 0.148 12 4 6.55 7 0.009 1 0 0.42 3 0.6210 1 0 0.03 1 0.0611 2 0 0.91 1 0.0012 2 0 1.00 2 0.1413 0 0 0.08 1 1.0014 0 0 0.01 1 1.00Goodness − of − f i t f o r edgewise shared partnerobs min mean max MC p − valueesp0 533 608 645.34 700 0esp1 92 43 43.01 44 0esp2 42 19 19.01 20 0esp3 6 3 3.00 3 0esp7 71 36 36.00 36 0esp8 1 0 0.00 0 0Goodness − of − f i t f o r minimum g e o d e s i c d i s t a n c eobs min mean max MC p − value1 745 709 746.36 801 0.922 421 186 212.11 249 0.003 243 46 67.21 102 0.004 138 14 22.30 45 0.005 60 6 8.67 22 0.006 19 0 1.07 7 0.007 5 0 0.06 3 0.00I n f 68788954 68789399 68789527.22 68789605 0.00Goodness − of − f i t f o r model s t a t i s t i c sobs min mean max MC p − valueedges 745.00 709.00 746.36 801.00 0.92nodecov . meandist − − − − − The results of all formation models of the STERGM fit to the data from 2000-2017. In the “ILVs”, or individual-levelvariables, column the numbers correspond to the variables included in the model (1: gender; 2: popularity; 3: followers;4: mean distance).
ILVs AIC deltaAIC1 2 3 14194.16 0.000001 2 3 4 14196.11 1.946882 3 14365.84 171.677872 3 4 14367.30 173.139431 2 14409.22 215.061681 2 4 14410.98 216.821451 3 14563.71 369.553511 3 4 14565.71 371.552122 14599.55 405.392752 4 14600.59 406.432841 14636.97 442.808221 4 14638.95 444.787633 14748.97 554.807633 4 14750.71 556.553460 14836.01 641.846664 14837.57 643.40921 ==========================Summary o f model f i t==========================Formula : y . form ˜ edges + nodecov (” meandist ”) + a b s d i f f (” p o p u l a r i t y ”) +a b s d i f f (” f o l l o w e r s ”) + nodematch (” gender ” , d i f f = TRUE) < environment : 0 x1ad107308 > I t e r a t i o n s : 10 out o f 20Monte Carlo MLE Results :Estimate Std . Error MCMC % z value Pr( > | z | )edges − −
02 0 − < − ∗∗∗ nodecov . meandist − −
09 1.702 e −
08 0 − − −
02 2.799 e −
03 0 − < − ∗∗∗ a b s d i f f . f o l l o w e r s 1.726 e −
07 9.188 e −
09 0 18.786 < − ∗∗∗ nodematch . gender . − −
01 3.605 e −
01 0 2.227 0.02592 ∗ nodematch . gender . 0 − −
01 2.582 e −
01 0 − ∗∗ nodematch . gender . 1 8.877 e −
01 7.878 e −
02 0 11.269 < − ∗∗∗ −−− S i g n i f . codes : 0 ’ ∗∗∗ ’ 0.001 ’ ∗∗ ’ 0.01 ’ ∗ ’ 0.05 ’ . ’ 0 . 1 ’ ’ 1Null Deviance : 7195553 on 5190494 d e g r e e s o f freedomResidual Deviance : 14182 on 5190487 d e g r e e s o f freedomAIC : 14196 BIC : 14290 ( Smaller i s b e t t e r . ) The results of the goodness-of-fit analysis of the formation model of the STERGM with the most individual-level variablesfit to the data from 2000-2017 are below.
Goodness − of − f i t f o r degreeobs min mean max MC p − value0 11179 10679 10811.81 10930 0.001 1507 1878 2005.54 2115 0.002 371 313 351.05 398 0.243 130 76 88.47 104 0.004 68 17 25.10 31 0.005 26 3 7.85 13 0.006 5 0 1.94 6 0.067 5 0 1.94 3 0.008 0 0 0.26 2 1.009 2 0 0.04 1 0.0010 1 0 0.00 0 0.00Goodness − of − f i t f o r edgewise shared partnerobs min mean max MC p − valueesp0 1176 1324 1379.71 1458 0esp1 294 144 144.42 147 0esp2 75 35 36.03 37 0esp3 22 10 10.02 11 0Goodness − of − f i t f o r minimum g e o d e s i c d i s t a n c eobs min mean max MC p − value1 1567 1514 1570.18 1648 0.902 1189 607 677.37 735 0.003 1107 320 373.55 436 0.004 950 177 225.67 286 0.005 658 83 123.00 177 0.006 380 37 69.22 125 0.007 182 15 39.00 82 0.008 79 3 19.09 47 0.009 16 1 8.26 34 0.2210 1 0 2.59 14 1.0011 0 0 0.58 7 1.0012 0 0 0.12 4 1.0013 0 0 0.02 1 1.00I n f 88352442 88355141 88355462.35 88355763 0.00Goodness − of − f i t f o r model s t a t i s t i c sobs min mean max MC p − valueedges 1567.00 1514.00 1570.18 1.648000 e+03 0.90nodecov . meandist − − − − odematch . gender . 1 1012.00 956.00 1014.21 1.069000 e+03 0.94 −
50 0 500102030405060 0 5M 10M 15M 20M
Popularity Followers N u m be r o f C o ll abo r a t i on s Figure S3: The relationship between popularity and followers and the number of collaborations for each artist in thedataset. Popularity and followers are on the x -axis, and number of collaborations is on the y -axis. Linear regressionfound significant positive relationships between both popularity and number of collaborations (R = 0.048, p < = 0.090, p < Pre − − − − − value Estimate p − value Estimate p − value Estimate p − valueMean Distance − −
07 9 . 0 e − − −
07 1 . 1 e − − −
07 3 . 2 e − − −
07 1 . 3 e − − −
03 2 . 9 e − − −
03 1 . 6 e − − −
02 1 . 5 e − − −
02 8 . 4 e − − −
07 8 . 0 e − − −
07 1 . 9 e − − −
08 5 . 5 e −
02 1 . 2 e −
08 6 . 4 e − −
02 1 . 3 e+00 3 . 1 e −
02 1 . 3 e+00 1 . 3 e −
02 1 . 9 e+00 4 . 9 e − −
02 7 . 3 e −
01 2 . 6 e −
01 7 . 4 e −
02 4 . 7 e −
01 1 . 4 e −
04 5 . 3 e −
01 2 . 0 e − − − − − value Estimate p − value Estimate p − value − −
07 1 . 6 e − − −
07 1 . 2 e − − −
07 2 . 0 e − − −
02 1 . 7 e − − −
02 9 . 9 e − − −
02 4 . 4 e −
172 . 8 e −
08 2 . 0 e −
01 3 . 9 e −
08 4 . 3 e −
02 4 . 6 e −
08 8 . 0 e −
031 . 8 e+00 5 . 1 e −
09 1 . 7 e+00 6 . 1 e −
08 1 . 5 e+00 1 . 6 e −
065 . 7 e −
01 2 . 0 e −
08 6 . 7 e −
01 5 . 8 e −
13 6 . 7 e −
01 7 . 5 e − − − − − − value Estimate p − value Estimate p − value Estimate p − valueMean Distance − −
08 2 . 1 e − − −
08 1 . 7 e − − −
09 5 . 8 e − − −
09 8 . 2 e − − −
02 9 . 3 e − − −
02 2 . 4 e − − −
02 1 . 7 e − − −
02 5 . 7 e − −
07 2 . 6 e −
74 1 . 6 e −
07 1 . 6 e −
77 1 . 7 e −
07 1 . 1 e −
75 1 . 7 e −
07 9 . 9 e − −
07 1 . 3 e+00 9 . 2 e −
07 8 . 3 e −
01 1 . 4 e −
02 8 . 0 e −
01 2 . 6 e − −
01 7 . 3 e −
37 9 . 0 e −
01 2 . 0 e −
35 8 . 6 e −
01 2 . 8 e −
30 8 . 9 e −
01 1 . 9 e − − − − − value Estimate p − value Estimate p − value1 . 2 e −
09 9 . 4 e −
01 1 . 7 e −
08 3 . 4 e −
01 2 . 2 e −
08 2 . 6 e − − −
02 3 . 4 e − − −
02 2 . 4 e − − −
02 3 . 1 e −
521 . 8 e −
07 3 . 2 e −
77 1 . 9 e −
07 1 . 9 e −
81 2 . 0 e −
07 1 . 2 e −
838 . 7 e −
01 1 . 5 e −
02 9 . 7 e −
01 7 . 6 e −
03 1 . 2 e+00 1 . 4 e −
038 . 6 e −
01 1 . 5 e −
25 8 . 4 e −
01 4 . 7 e −
21 9 . 2 e −
01 5 . 6 e −21