The Persistence of : A Statistical Network Analysis of Underground Punk Worlds in and Liverpool 2013-2015

A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy in the Faculty of Humanities

2019

Joseph B. Watson

School of Social Sciences

Contents Abstract………………………………………………………………………..6

Declaration…………………………………………………………………….7

Copyright Statement…………………………………………………………....7

The Author…………………………………………………………………….8

Acknowledgements…………………………………………………………….8

Chapter One: Punk, Networks and the North West……………………………9

1.1 Introduction……………………………………………………...9 1.2 What is a Music World?…………………………………………11 1.3 What is Punk Music?……………………………………………12 1.3.1 Defining Punk Music…………………………………….13 1.3.2 My Experience of Punk Music as a Participant…………...14 1.3.3 Underground Punk in a Modern UK Context……………15 1.4 Manchester and Liverpool as Musical Cities…………………….17 1.5 Analysis………………………………………...18 1.5.1 The Network Lens……………………………………….18 1.5.2 The Analytical Toolbox…………………………………..19 1.6 Thesis Structure…………………………………………………19 Chapter Two: Music Worlds………………………………………………….22 2.1 Introduction…………………………………………………….22 2.2 Subculture………………………………………………………22 2.3 Towards Art Worlds…………………………………………….25 2.4 Collective Activity………………………………………………26 2.5 Conventions…………………………………………………….26 2.6 Resources……………………………………………………….28 2.7 Venues………………………………………………………….30 2.8 Music Making…………………………………………………...32 2.9 Underground Punk in Manchester and Liverpool……………….33 2.10 Social Networks and Music Worlds……………………………..34 2

Chapter Three: ……………………………………………...37 3.1 Introduction…………………………………………………….37 3.2 The Purpose of Network Analysis………………………………38 3.3 Ego-networks…………………………………………………...39 3.4 Unipartite Networks……………………………………………40 3.4.1 Network Phenomena…………………………………….42 3.4.2 Networks of Music Worlds………………………………45 3.4.3 Dual Projection…………………………………………..45 3.5 Bipartite Network Analysis……………………………………...47 3.5.1 Bipartite vs Multilevel…………………………………….49 3.6 Tripartite Networks……………………………………………..50 3.6.1 Bringing the Venues Back in……………………………...52 3.7 Discussion………………………………………………………53 Chapter Four: Methodology…………………………………………………..54 4.1 Introduction…………………………………………………….54 4.2 Operationalising Music Worlds………………………………….54 4.3 Research Aims…………………………………………………..57 4.4 Research Questions……………………………………………..58 4.5 Data Collection…………………………………………………58 4.6 Ethical Consideration…………………………………………...64 4.7 Data Descriptions………………………………………………66 Chapter Five: Analysis of a Tripartite Network……………………………….71 5.1 Introduction…………………………………………………….71 5.2 Motivation………………………………………………………71 5.3 Data Background………………………………………………..72 5.4 Exponential Random Graph Models……………………………73 5.4.1 Bipartite Extension………………………………………74 5.5 Bipartite Analysis………………………………………………..74 5.5.1 Musician-Band Network…………………………………74 5.5.2 Musician-Band Estimation Results……………………….75 5.5.3 Musician-Band Goodness of Fit………………………… 76

3

5.5.4 Band-Venue Network……………………………………76 5.5.5 Band-Venue Estimation Results………………………….77 5.5.6 Band-Venue Goodness of Fit……………………………79 5.6 Tripartite Analysis………………………………………………80 5.6.1 Transitive Closure………………………………………..81 5.6.2 Closure in a Tripartite Context…………………………...82 5.6.3 Recombining Bipartite Network in GOF Distribution…...83 5.6.4 Tripartite Simulation Results……………………………..85 5.7 Tripartite Results………………………………………………..85 5.7.1 Further Possibilities………………………………………86 5.7.2 Attribute Data……………………………………………86 5.7.3 Venues vs Events………………………………………...86 5.7.4 Further Configurations…………………………………...87 5.8 Discussion………………………………………………………87 Chapter Six: Bipartite Dynamic Network Analysis of Bands and Venues……...89 6.1 Introduction…………………………………………………….89 6.2 Motivation………………………………………………………90 6.3 Network Dynamics……………………………………………...90 6.3.1 Network Growth vs Network Activity…………………….91 6.4 Network Data for Analysis……………………………………...92 6.5 Stochastic Actor Oriented Models………………………………94 6.6 Modelling the Network………………………………………….95 6.6.1 Parameter Specification…………………………………..97 6.6.2 Model 1…………………………………………………..97 6.6.3 Model 2…………………………………………………..98 6.6.4 Model 3…………………………………………………..99 6.6.5 Model 4…………………………………………………..99 6.7 Descriptive Statistics…………………………………………...100 6.7.1 Model Building Procedure………………………………102 6.8 Bipartite SAOM Results………………………………………..103 6.8.1 Rate Parameters………………………………………....103

4

6.8.2 Network Parameters……………………………………104 6.8.3 Nodal Attributes………………………………………..105 6.8.4 Dyadic Covariates………………………………………105 6.9 Discussion……………………………………………………..106 Chapter Seven: Conclusion…………………………………………………..111 7.1 Introduction………………………………………………………111 7.2 Summary of Chapters……………………………………………..111 7.3 Discussion………………………………………………………...114 7.4 Further Implications……………………………………………....117 Notes………………………………………………………………………..119 References…………………………………………………………………..120 Appendix 1………………………………………………………………….128 Appendix 2………………………………………………………………….129 Appendix 3………………………………………………………………….130 List of tables and figures Figure 1: Multilevel network…………………………………………………..49 Table 1: Crosstabulation of events by city per calendar year…………………..66 Table 2: Counts of bands and venues by geographical location……………….66 Table 3: Bipartite degree centrality for venues in 2013………………………..67 Table 4: Bipartite degree centrality for venues in 2014………………………..68 Table 5: Bipartite degree centrality for venues in 2015………………………..69 Table 6: ERGM model results for musician-band network…………………...75 Table 7: Goodness of fit for musician-band ERGM………………………….76 Table 8: ERGM model results for band-venue network………………………77 Table 9: Goodness of fit for band-venue ERGM……………………………..79 Figure 2: Bipartite matrices…………………………………………………....80 Figure 3: Tripartite matrix…………………………………………………….80 Figure 4: Tripartite network of musicians, bands and venues…………………81 Figure 5: Transitive closure…………………………………………………...81 Figure 6: Tripartite network configurations…………………………………...83 Figure 7: Distributions of simulated tripartite networks………………………85 Figure 8: Cumulative advantage in tripartite closure…………………………..87 Table 10: Counts of bands by geographical location………………………….93 Table 11: Stochastic actor-oriented model results……………………………103 Final word count: 50,849

5

Abstract Scholars of social networks have, in recent years, devoted much attention to the development of methods for the analysis of increasingly complex social structures. One particular element of complexity in network methodology can be found when attempting to analyse networks with multiple types of nodes. Traditionally this data has been found and discussed as bipartite affiliation data, for example, the ties between women and social events that are found in the Southern Women dataset (Davis et. al, 1941). Simultaneously there has been a growing interest in the social networks of music worlds with scholars taking Crossley’s (2008, 2009, 2015a, 2015b) lead in using Becker’s (1974, 1976, 1982) art worlds theory as the foundation from which to analyse the networks of collective activity that form these music worlds using formal . This thesis brings together both of these areas of research in analysing the underground punk music world of Manchester and Liverpool from 2013-2015. Research on music worlds in general, has so far neglected the pivotal role of venues in sustaining these worlds, whilst largely ignoring that conventions are dynamic in nature and have the ability to be formed and reformed in the social spaces of venues. Simultaneously, methodological research on networks has so far neglected networks of more than two types of nodes. This thesis addresses both of these concerns by focusing on venues and the active choices that bands make to perform at a small group of key venues that are the foci (Feld, 1981) of the underground punk world. In an empirical analysis of an underground punk network of musicians, bands and venues, it introduces a novel statistical method for testing for the prevalence of tripartite closure patterns, concluding that venue choice is not driven by overlapping band membership ties. Through further investigation of the bipartite network of bands and venues, it is argued that core bands reaffirm their choice of venues in every time period and so the world is sustained through a small number of key venues over time and that differing choices made by bands located in Manchester and Liverpool allow them to foster their own whilst simultaneously offering them opportunities and imposing constraints on their ability to form and reform new conventions within these venues. Through this analysis, this thesis makes a unique contribution to academic understanding of music worlds, as well as introducing new methods for the analysis of tripartite structures that are generalisable to other music worlds and beyond.

6

Declaration No portion of the work referred to in the thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning. Copyright Statement The author of this thesis (including any appendices and/or schedules to this thesis) owns certain copyright or related rights in it (the “Copyright”) and s/he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes.

Copies of this thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made.

The ownership of certain Copyright, patents, designs, trademarks and other intellectual property (the “Intellectual Property”) and any reproductions of copyright works in the thesis, for example graphs and tables (“Reproductions”), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions.

Further information on the conditions under which disclosure, publication and commercialisation of this thesis, the Copyright and any Intellectual Property and/or Reproductions described in it may take place is available in the University IP Policy (see http://documents.manchester.ac.uk/DocuInfo.aspx?DocID=24420), in any relevant Thesis restriction declarations deposited in the University Library, The University Library’s regulations (see http://www.library.manchester.ac.uk/about/regulations/) and in The University’s policy on Presentation of Theses.

7

The Author Joe graduated from The University of Manchester in 2012 with a BA in Economics, specialising in Development Studies. Following this he completed an MSc in Social Research Methods and Statistics also at the University of Manchester where he was first introduced to Social Network Analysis and the community of fellow researchers at the Mitchell Centre and completed an 8000 word dissertation using stochastic actor-oriented models to analyse the network of live performance in the world in the USA over eight years from 1979-1986. Since choosing A-levels in Business Studies, Mathematics and Music Technology he has been repeatedly told that music and maths are not complimentary subject areas. He is currently employed as a research data analyst at an independent music publisher in Liverpool as well as playing lead and attempting to sing in melodic punk rock band Hummer. Acknowledgements First and foremost, I want to thank my partner Becca for her love and support throughout the course of this PhD journey and especially for her patience in this last year with a grumpy researcher in his effort to complete this thesis. Special thanks also go to both of my parents, for encouraging me to learn both the piano and guitar at a young age, which, although I never practised at the time, gave me the foundations for performing live when I decided that punk was music worth some effort. Thanks to my supervisors Nick Crossley and Johan Koskinen for their encouragement and support throughout the course of this PhD as well as their enduring patience in waiting for me to send writing to them. I also want to thank the community of researchers at the Mitchell Centre for Social Network Analysis at The University of Manchester who, in the course of many conversations helped me to shape my initial research proposal and who have provided helpful comments and insights throughout my time completing this research. Special thanks go out to all those involved in keeping the underground punk world alive, all the venues, promoters, photographers, journalists, podcasters and everyone else that selflessly puts in time and energy and money to keep things going and to all the bands that continue to create and perform music and the audiences that turn up to support underground music, without you I most likely would not have written a PhD proposal, you all created something that was interesting and engaging to explore and immerse myself in. Last but certainly not least, thanks to my bandmates Will Atkinson and Matthew Cain, for all the great times we’ve had playing punk music together. I think we’re finally starting to sound half decent. In fact, it still surprises me that after more than ten years of trying and zero singing lessons we’re on the cusp of singing in tune, a testament to perseverance if there ever was one. Long may it continue.

8

Chapter One: Punk, Networks and the North West 1.1 Introduction Punk is not dead. In fact, it never has been. The music has changed and evolved over more than forty years, bubbling under the surface of mainstream music in small venues and breaking through into public consciousness at various points in time to remind us that punk will never die. Artists identifying with the genre span across every level of the music industry, from multi-platinum1 selling bands, like Green Day (RIAA, 2013) that headline large capacity arenas and large-scale festivals, to small-scale bands performing in small local venues with a capacity of fewer than fifty people. It is at this small-scale level where I focus my attention, on the bands that perform for the love of the music, the underground bands that sustain punk music at a grassroots level. This thesis sets out to uncover the structure of the underground punk music world in Manchester and Liverpool over a period of three years from 2013 – 2015. It seeks to describe and evaluate the structure of live performance in these two major UK cities and give insight into how this music world operates. Throughout this thesis I draw on both my own experiences as a participant in the underground punk music world and combine my own knowledge and experience with rigorous empirical research in order to provide a coherent academic account and give the reader an appreciation and nuanced understanding of the social structure of performances of punk music at a local level, arguing that venues have seen a distinct lack of focus in academic work on networks of music worlds, despite their critical importance to sustaining those worlds.

In order to investigate and interrogate this world and its inner workings I draw on the tools and techniques of formal social network analysis (Wasserman and Faust, 1994; Scott and Carrington, 2011; Borgatti, Everett and Johnson, 2013). As well as using these formal social network methods, I approach the analysis of a music world from a network-driven epistemological standpoint. Through the sociological lens of a social network analyst, the world around us can be described and explained through structures created from social interactions and these interactions are the unit of analysis with which we should be concerned (Tilly, 2008; Crossley, 2011). As a quantitative-focused researcher, I draw on the growing body of literature on the use of statistical models for social network analysis in order to analyse the structure of relational ties between the musicians, bands and venues that create a network of live performance in the underground punk music world of Manchester and Liverpool, two cities located 40 miles apart in the North West of England, over a period of three years from 2013-2015. I focus on the importance of venues situated in tripartite network structures that also include musicians and bands. Academic work on music worlds has so far largely neglected

9 the role of venues in sustaining a music world and the role venues play in the realisation of bands as dual entities that form and reform conventions concerning how music both sounds and is performed. Tripartite networks, with three distinct node sets, are also significantly under-researched and so I also set out in this thesis to introduce new methods for the analysis of these types of network.

I combine Howard Becker’s (1974, 1976, 1982) theory on ‘art worlds’ with network theory on duality (Breiger, 1974) and operationalise them in my analysis of a local music world. In doing so I make an original contribution to knowledge in four ways:

Substantively: through the investigation of the structure of the activity of live performance in a local, underground punk music world in Manchester and Liverpool, two prominent industrial cities in North West England, I contribute to the growing body of empirical network analysis of music worlds.

Methodologically: through the introduction of advanced statistical methods to the sociological investigation of music worlds and cultural sociological networks more widely.

Methodologically: by developing a sound method for testing for the significance of tripartite structures in networks with three different types of nodes, applicable not only in cultural sociological contexts but across the multiple academic disciplines that use social network analysis techniques to analyse relational ties.

Theoretically: by combining theories on both music worlds and duality in networks to advocate for the investigation of networks with three types of nodes as well as specifying tripartite network phenomena of interest to network researchers more generally.

Throughout this thesis I seek to answer four key research questions:

1. Is there a tendency towards tripartite closure in the network of live performances in Manchester and Liverpool?

2. Do overlapping band-membership ties drive the formation of band-venue ties in tripartite closure?

3. What role do venues play in the underground punk world of Manchester and Liverpool?

4. How does the role of venues play out over a three-year time period from 2013-2015?

The following sections in this chapter give a comprehensive overview of the key themes of the thesis including terminology that will be used throughout. To begin with, I elaborate upon what I mean by a ‘music world’, what this encompasses and boundaries of the music world 10 that I wish to explore in this thesis; the underground punk world of Manchester and Liverpool. Next, I dedicate a section to discuss punk music, giving a brief account of its history in popular culture and explain the difficulty in defining punk music as a genre. I then move on to discuss writing on punk including both academic and journalistic sources, highlighting its continuing relevance in the musical and cultural landscape of the UK and beyond. I follow this with a discussion of my dual role as both researcher and participant in the underground punk music world and give an overview of what the modern underground punk music world in Manchester and Liverpool encompasses as well as how it operates. After discussing punk music, I describe the two cities I have chosen to focus on in this thesis Manchester and Liverpool, noting their cultural significance with particular reference to music. After clarifying my substantive choices of punk as a musical genre and the reasons for setting my focus on Manchester and Liverpool, I subsequently introduce social network analysis, both as a theoretical concept and a formal method of analysis in order to give clarity to the chapters that proceed this introduction and to make clear my epistemological viewpoint. Finally, I outline the structure of the thesis in the subsequent chapters.

1.2 What is a Music World? In defining the concept of a ‘music world’ I take the lead from Crossley (2008, 2009, 2015) and others’ work (Emms and Crossley, 2018; O’Shea, 2015; Hield, 2010; Crossley, McAndrew and Widdop, eds., 2015; Bottero and Crossley, 2011; Crossley and Bottero, 2015) adapting Howard Becker’s concept of ‘art worlds’ (1974, 1976, 1982) in order to analyse relations formed by collective activity in the form of a network (see Wasserman and Faust, 1994; Scott and Carrington, 2011; Borgatti, Everett and Johnson, 2013, for a comprehensive overview of network concepts). When defining a music world, it is important to make the distinction between the production and consumption of music. Much cultural sociological attention has been given to the consumption of culture and, by extension music. This is not my focus in this thesis. Instead I choose to cast my attention towards cultural production and assert that this is created and sustained in a network of live performance activity. A music world is formed through social interactions, and these interactions occur as the result of the need for collective activity to accomplish all the tasks associated with producing music. The broad genre of music I focus on, punk, can be considered to be a form of ‘’, that is, as opposed to classical music or jazz among other forms and I would argue that musicians who create popular music have three main forms of output; composition, musical recordings and live performances, which are all interlinked. Here I focus my attention on live performance, which, I contend is the most important of these three creative outputs in the underground punk music world and key to sustaining this world. Indeed, more generally, as sales of physical music have declined, live performance has become an increasingly important revenue source 11 for the music industry as a whole (Chapple, 2018). With a focus on live performance I explore four key concepts of a music world in chapter two; collective activity, resources, conventions and venues. It is these venues that I position as key to sustaining the underground punk music world in Manchester and Liverpool, that ensure punk is alive and kicking, that foster new conventions and reinforce established ones; where social ties are focused.

1.3 What is Punk Music? In the previous section I describe what I mean by a music world and that the music world I have chosen to focus on in this thesis is that of underground punk music in Manchester and Liverpool. This leads to the question as to why I have settled on punk music as an area of interest for research. Preceding the answer to this question, it is important to elaborate upon what punk music is and how I consider it in this thesis.

First and foremost, in this thesis punk is considered as a style of music. Punk has changed and evolved over more than forty years and has spawned an array of subgenres and styles that can now be heard in small venues across the UK, all of which broadly classifies itself as punk music. These styles develop over time with creators drawing on influences from the large body of work that exists in the form of recorded music, but also from the interaction that takes place within these venues. Punk music developed simultaneously and transatlantically in the mid-to-late 1970s, transitioning from proto-punk in the USA with and The and the pub-rock world in the UK, including Clash Frontman Joe Strummer’s former band the 101ers. These early incarnations were followed closely by simultaneous worlds in and New York with The and The developing a new more aggressive sound different to , or glam rock that was prominent at that time. In the UK punk came to symbolise anti-establishment working-class sentiments and youth culture, although as I elaborate upon in chapter two, I agree with Crossley (2008, 2015) that this is over-simplifying the matter. Punk as a musical genre has gone through multiple incarnations and various subgenres from its first wave of bands in 1975 and early 1976. In the UK it mostly disappeared from mass public consciousness with the emergence of post-punk followed by the and Indie genres in the 1990s although underground punk worlds still subsisted (Glasper, 2009, 2012, 2014a, 2014b) and in fact, the underground punk world I analyse is one that subsists beneath the mainstream music industry today.

In the USA the story follows a similar trajectory with a more underground life beyond initial mainstream emergence with a well-documented DIY hardcore punk scene from 1979-1986. (Blush, 2001, 2006; Chick, 2009; Miller, 2013; Ogg, 2014). Punk in the USA had a mainstream revival in the early 1990s with a more melodic sound that coincided with skate and surf culture 12 in the USA and beyond, as well as the CD sales boom and the music video culture established in no small part thanks to the popularity of MTV. Numerous independent labels were established in California and bands such as Green Day and The Offspring achieved global mainstream success, opening the door for distinctly ‘poppier’ sounding bands like blink-182 to go on to sell millions of albums, bringing punk into the public consciousness of millions of young people.

For the most part, punk has returned to the underground level, although many previously established bands continue to perform live and sustain a career. In 2012, Russian activist group Pussy Riot gained worldwide attention for their protest performance of their song “Punk Prayer” at Moscow’s Cathedral of Christ the Saviour Cathedral, which led to arrests and jail sentences for some members of the band (BBC, 2013). Pussy Riot continues to be politically active, claiming responsibility for a pitch invasion at the 2018 World Cup Final (Guardian, 2018). Through all these transformations punk has refused to disappear completely and has sporadically returned to public consciousness through music or activism or a combination of the two. This chequered past has made for a complex modern genre with a diverse range of protagonists creating a diverse range of sounds all under the punk umbrella. This in turn has created a diverse range of influences in the punk cannon for new bands to draw on in forming their own sound and defining their own version of punk. This phenomenon of changing and evolving sounds, I argue, does not happen in an underground music world without interaction in social spaces between bands and their audiences. The problem for the researcher then, is how to define punk within the context of sensible research questions.

1.3.1 Defining Punk Music Punk as an adjective is used in many different ways for many different reasons to describe music, politics, fashion, attitudes, or for some individuals it can be an all-encompassing philosophy for how they live their lives, even stretching so far as to define a research philosophy (Beer, 2014). As this thesis focuses on the producers of punk music and not the consumers, the fashion and ‘way of life’ aspects of punk are left well alone here. Having put these aspects to one side, the definition of punk solely in terms of music is still one that is extremely difficult to pin down. For some people the term is politically charged when associated with their music, for some people it is simply musically significant and refers to either lyrics, aesthetic quality, or more often than not, both.

Punk has evolved through many different guises from proto-punk in the early-to-mid 1970s of The Stooges, The New York Dolls and in the USA; through its so- called explosion with The Sex Pistols, and The Ramones; Street Punk in 1980s UK; 13 hardcore punk in the USA from 1979-1986 and the popularity of skate, pop and ska-punk in the 1990s and beyond. It is clear that punk has a place in the global popular music canon and a legacy that lives on in modern music (Blank Generation: Punk Rock, 2007). By referring to the music worlds in Manchester and Liverpool as punk, then, I am ascribing a term that has different meanings for different actors within these worlds. I attempt to resolve the issue methodologically by means of systematic classification based on criteria that I outline in chapter four, by including bands in the network that self-identify as punk, rather than positioning myself as arbiter of what is punk and what is not in a top-down prescriptive manner.

1.3.2 My Experience of Punk Music as a Participant In this section I highlight my own experience of punk music as an active performing musician and a participant in the underground punk world in the UK. The purpose of this is to illuminate my unique position as both researcher and participant and how I resolve to use my knowledge to make positive contributions to the thesis whilst I maintain a concerted effort to analyse the underground punk world with relative objectivity. I do not, however, believe that researcher bias can be completely removed from the research process, rather that we can strive for conclusions based on accurate and reliable data we collect, the appropriate models we choose and the epistemological viewpoint we align ourselves with in a systematic and robust manner, drawing reasonable conclusions. As might be expected then, a major part of the motivation for conducting this research is that I am interested in punk music as a consumer of the music but also as a producer of music and a participant in the underground punk music worlds in both Liverpool and Manchester. I outline my experience of punk below in an effort to be pragmatic in declaring the position from which I conduct this research; at the intersection of participant, observer and researcher; a position that I believe gives added value to rather than detracts from the conclusions I draw throughout the following chapters.

I developed a serious interest in punk music at secondary school, where I also formed my first band after first making friends with others that were also musicians or had an interest in playing music, in a school that was very encouraging of musical pursuits. I had begun to learn guitar at primary school in formal lessons around age 7 but had a strong aversion to regular practice and made slow progress until I developed an interest in guitar music, first through a I was given with Oasis’ seminal Britpop anthem Wonderwall on it, followed by a keen interest in glam rock band The Darkness. I really learned to play guitar, however, and be a rock musician, through interactions and jamming sessions with my secondary school friends. My first experience of performing live to an audience took the form of a school assembly performance of ‘’ by the Clash to stunned silence, save a few friends and one

14

History teacher who seemed to enjoy it immensely and who wondered afterwards where we had heard The Clash in the first place. Several years later, whilst completing my Master’s degree, my personal interest in punk music and academic interest in social networks, relational sociology and statistical modelling began to align and after completing a dissertation on hardcore punk in the USA, 1979-1986 (Watson, 2015) I began to reflect on my own experiences of becoming a rock musician (Bennet, 1991). On reflection, the interactions within the secondary school setting were processes that would be described by social networks scholars as social selection and social influence, choosing friends based on similar behaviours or interests as well as passing on behaviours and interests to established friends. The school environment fostered a series of social influence and selection mechanisms whereby it seemed at one stage as if most musicians in the school that played guitar, bass or drums professed to be in some sort of band. This forming of musical taste and subsequently bands as a social process was a theme that resonated with my experiences of participating in the underground punk world of Manchester and Liverpool.

It is important to note here that my three main periods of active live performance have been outside of the time periods under investigation in this thesis, from 2005-2009, 2010-2012 and November 2015 to the present, whilst the focus of this thesis is on live performances that took place in Manchester and Liverpool from 1st January 2013 up to and including the 31st December 2015. One of the bands I have performed with, Hummer, does appear in the collected network in 2015 but is excluded in the final analyses of the core network. Apart from a brief appearance as a live performer in the time period in question, I was an audience member at a large number of the gigs during this time period, engaging in conversations with fans and musicians that have shaped the analyses and theory present in this thesis and I draw on this first-hand experience in conjunction with robust empirical analyses in my explanation of the social mechanisms at play in sustaining underground punk in the North West of England. In the following section I use my experiences as participant observer to describe the underground punk in a modern UK context.

1.3.3 Underground Punk in a Modern UK Context In the previous two sections I outlined the issues with the definition of punk music and the fact that it is a constantly evolving genre, before giving an overview of my own experience of punk as a participant. Here I draw on that experience as a participant in the world and endeavour to describe the underground punk world, how it operates, the people involved, their motivations and the types of venues they perform at.

I use the term ‘underground’ here to describe the world I want to analyse because it is invariably outside of what is referred to as the ‘mainstream’ music industry. All of the 15 participants in the world I capture write and perform original music and the majority of these bands also record and release music independently or through small, independent record labels, often formed by band members as a way to release their own music as well as that of other underground bands, embracing a distinctly ‘DIY’ ethic. Similarly, the majority of the events that these bands perform at are organised by either musicians themselves or fans of punk music. Bands are often compensated with no more than petrol money and mostly operate at break even or a loss. Mostly bands are passion projects, more than perhaps a mere hobby, with hours of time and thousands of pounds spent on equipment, transport, rehearsal time and production of vinyl, CDs and merchandise, but they are by no means a commercial enterprise, and certainly not a profit making one.

Performing live is an integral part of the underground punk world and small venues in Manchester and Liverpool provide a platform for bands to express themselves and interact with audiences both collectively and as individuals. The types of venues that gigs take place in vary greatly from purpose built live venues to pubs, basements, squats and one significant pizza bar. Venues rarely exceed 100 in capacity, more often than not 50 would be considered a ‘packed house’. In two cities with large venues that attract much more popular bands, audiences at punk gigs tend to consist of a small number of regular gig-goers, passionate about new music. The overwhelming majority of these attendees tend to be involved in some way in the world, as musicians themselves, photographers, promoters, sound engineers, logo- designers or writers contributing to ‘zines’ - traditionally these are low-budget, black and white photocopied publications with a small circulation to devoted participants in a music world, but some now also find a home; alongside a small, dedicated following; online.2

Finding venues to perform at can be a struggle due to their transient nature, but there are mainstays of the world in both Liverpool and Manchester. In both cities this ecosystem is self- sustaining, and I argue in this thesis that these venues are critical to the continuation and subsistence of the underground punk world, providing a social space for conventions to be established, formed and reformed.

The music world I have analysed here is a snapshot in time, it is not intended as an account of the beginning or formation of a music world at its genesis, rather a dynamic and shifting world that changes, adapts and is sustained by what some may consider a surprisingly large and diverse number of participant bands both native to Manchester and Liverpool as well as from outside it. For the purposes of this thesis, data was collected from a total of seven hundred and twenty-five unique bands performing at three hundred and fifty-six events at forty-one venues over three years from 2013-2015. I am stating these numbers at the beginning of this thesis in the hope that it gives a small indication of the size of the network and the level of

16 activity present in the two cities. A detailed descriptive analysis of the collected dataset can be found in chapter four.

1.4 Manchester and Liverpool as Musical Cities This section aims to introduce Manchester and Liverpool as cities, with particular reference to music and their significance in the musical landscape of the UK. In the introductory paragraph above I noted one reason for focusing the network to Manchester and Liverpool; that from my experience as a participant in the underground punk world, I knew there was a significant level of live performance activity in both cities, with regular punk gigs taking place. It also happens that my hometown of Warrington lies exactly between the two cities geographically and so I am interested in both places, without necessarily a partisan allegiance to one or the other. Apart from providing a substantial pool of data and my own affinity with the North West of England, choosing Manchester and Liverpool as cities to focus on perhaps also, I hope, makes a small contribution to documenting and analysing their history as musical cities.

Both Manchester and Liverpool have played a key role in British musical history, both cities producing early professional orchestras, the Royal Liverpool Philharmonic Orchestra established in 1853 and the Halle Orchestra in 1857, both still actively performing concerts today. The two cities have also made major contributions to the world of popular music, Liverpool’s most famous global export being and Manchester’s arguably Oasis, both bands forming in their home cities and playing at local venues before signing major record deals and moving to London. They are examples of the small proportion of bands that ‘make it’. They are visible to global audiences, but there are multiple music worlds in Manchester and Liverpool that bubble under the surface of the majority of public consciousness and subsist in the social spaces of small venues and this is where I focus my attention; on underground punk.

Both cities also played significant roles in the formation of the early punk music world in the UK. The Sex Pistols’ gig at the Lesser Free Trade Hall in Manchester on 4th June 1976 was followed by a ‘critical mass’ of post-punk protagonists, purportedly audience members at the Pistols’ gig, forming The , The Fall, and Joy Division (Crossley, 2015a). As Crossley (ibid.) notes, the watershed moment in Liverpool came the following year at a Clash gig in 1977. In both cities key performance venues emerged as the focus for punk activity, The Electric Circus in Manchester and Eric’s in Liverpool. In this thesis, I will show that small venues are critical to the subsistence of the underground punk worlds of Manchester and Liverpool in a modern setting.

17

Having given a brief overview of the substantive areas of interest of this thesis, I now move to give a short introduction to social network analysis as a set of methodological tools as well as an epistemological standpoint from which I conduct this research.

1.5 Social Network Analysis This thesis is built upon the theoretical, methodological and considerable substantive research that has been conducted in the field of social network analysis (SNA). As a field of research, it can be difficult to define as it covers a multitude of traditional academic disciplines in both social sciences and beyond with scholars aligned to anthropology, sociology, psychology, social statistics, economics, mathematics and statistics departments to name a small number. In order to enlighten the reader, it is perhaps best introduced in two different guises:

1. A lens through which to view the world around us

2. A set of tools for the analysis of that world

1.5.1 The Network Lens Researchers affiliated with the field of social network analysis are concerned generally with social structure and believe that this structure can be explained through the analysis of social interactions. In this regard, I take the theory of relational sociology (Tilly, 2008; Crossley, 2011) as grounding for my own work. Relational sociology asserts that interactions are the basis on which society is built and through focusing on interaction we can uncover the mechanisms that best explain traditional sociological concerns such as power, norms, conventions, resources and social capital. Social network analysis posits that the structure formed by these interactions can be represented in a network of actors, represented by nodes, connected by defined relationships, represented by ties, in a network graph or ‘sociogram’. From an epistemological point of view, social network analysts believe that rather than focusing on individuals as discrete cases with some set of attributes, as in the case of survey methodology, we can instead focus our analyses on a set of connected individuals, whose relationships depend upon other relationships in the network of interest, as well as individual attributes simultaneously. This focus on social ties, in tandem with the belief that those ties are interdependent is what sets social network analysis apart.

These networks elicit unique research questions about local and global structures within them as well as individual roles and positions in the network. For instance, in a friendship network we are able to not only ask who is the most popular but who are brokers in the network between different groups of friends and may have more social capital, whether or not gender has an impact on who people choose as friends as well as how friendship ties form as a result of, as well as influence, common behaviours, for example, smoking. 18

1.5.2 The Analytical Toolbox Alongside a relational epistemology and a rich body of theory on social networks, simultaneously there exists a large body of methodological research, giving researchers in the discipline both qualitative and quantitative tools tailored to the analysis of relational ties. Whilst social network analysis as a research area is very much interdisciplinary with regard to traditional academic departments, it provides us with a standard set of tools for the analysis of network data. Analytically, network measures are rooted in mathematical graph theory and at a simple level, a network is characterised by a set of nodes, often individuals but not exclusively, connected by a set of defined ties. In this thesis I argue that there is a pressing need to look to methods that take into account the multiple node types that contribute to sustaining a music world, an argument I significantly elaborate upon substantively in chapter 2 and methodologically in chapter 3.

The empirical chapters in this thesis make use of statistical models in order to analyse the underground punk network. However, the goal of network researchers differs from that of the standard quantitative sociologist or social statistician in that rather than analysing a set of discrete observations, we are concerned with dependencies among the variables we measure; the ties between social actors; and it is not necessarily our aim to draw inference from a sample to a population of interest, rather we seek to study our population of interest and to say something about the way that population is structured or how that structure is formed through the creation and dissolution of ties.

1.6 Thesis Structure The following section is intended as a brief guide of the chapters that follow and to inform the reader of the structure of the overall argument. Following this introductory chapter:

Chapter Two reviews the literature in the sociological space of what I refer to as music worlds. I review the early Centre for Contemporary Cultural Studies (CCCS) literature on subculture, moving on to briefly acknowledge the theoretical work of Pierre Bordieu (1984, 1993), explaining why, similarly to other scholars of music worlds (Finnegan, 1989; Shank, 1994; Crossley, 2008, 2009, 2015; O’Shea, 2015) I prefer to use Howard Becker’s Art Worlds (1974, 1976, 1982) as a theoretical foundation for the analysis of a music world. I then elaborate upon different aspects of a music world including collective activity, conventions, resources and venues before reviewing literature that is concerned with the social networks of music worlds, contending that there is a lack of focus on the importance of venues, that can be explored through the development of new methods for tripartite networks and by using statistical modelling techniques.

19

Chapter Three reviews literature specifically related to the study of networks. It is structured sequentially to progress through explanations of different network formations. Beginning with ego networks and explaining that these types of networks are not appropriate for my research questions, I move to whole network analysis, noting key research and the role it has played in informing network research questions and methodologies. I note Breiger’s (1974) concept of duality and the key role it plays in the way we can think about musicians, bands and in particular, venues in a music world. I posit that we should think about methods of analysis that do not reduce this duality thereby losing information and elaborate upon bipartite networks, before extending this argument to advocate for the use of tripartite methods to address research questions surrounding networks with three distinct, but equally important node sets.

Chapter Four introduces methodological choices and concerns, showing how I intend to operationalise my conception of music worlds as well as the dual nature of musicians and bands through analysis of a tripartite network, subsequently outlining key research aims and four key research questions. I then present methods used for data collection, including ethical considerations and outline how the dataset I use for analysis was constructed., concluding with a brief overview of the network data including brief descriptive statistics and visual representations of the collected networks.

Chapter Five is the first of two empirical chapters in this thesis. It addresses the analysis of the tripartite network of musicians, bands and venues and how we might set about analysing a network with three different node sets; musicians, bands and venues; from a sample of the collected data. I analyse two separate bipartite networks of musicians tied to bands, and bands tied to venues and subsequently test for the prevalence of tripartite phenomena that can be explained by these two separate models. I explain in detail a novel statistical method for combining goodness of fit distributions from two separate bipartite, exponential random graph network models, concluding that these bipartite models are sufficient in the case of my network data, for explaining the tripartite phenomena. Parameters included in the models show that tie formation between bands and venues is not driven by membership ties between musicians and bands. It is shown that significant parameters pertaining to local network substructures occur at the band-venue level and that this should be the focus of further investigation.

Chapter Six is the second empirical chapter and focuses on the network of bands performing at venues in Manchester and Liverpool over three years from 2013-2015. I explain that the tripartite analysis in the previous chapter shows that the choices of bands to perform at certain venues is not driven by overlapping band-membership ties and therefore should be

20 investigated further. I posit that the temporal nature of the focused organisation of social ties (Feld, 1981) around venues can be analysed using longitudinal models for social networks, highlighting the problematic nature of discrete-time temporal exponential random graph models (Block et al., 2018) and settling on the continuous time based stochastic actor-oriented models (Snijders 2010). Subsequently, I define and estimate four separate bipartite stochastic actor-oriented models, based on Koskinen and Edlings (2012) specification, that analyse the role of venues over time and the choices made by key bands in the network. Parameter estimates of the models show that the underground punk music world is sustained by a small number of key venues that facilitate the dynamic nature of conventions that are key to a thriving music world. I also highlight the interaction between band location and venue location and compare venue choices of bands based in Liverpool with those based in Manchester.

Chapter Seven draws conclusions from the thesis as a whole, reiterating how the thesis has made original contributions to knowledge in the four ways set out in this introductory chapter. It guides the reader through the key messages in each preceding chapter and shows how the research questions have been addressed. It concludes by reviewing the key findings in the context of current research and subsequently outlines areas for further investigation.

21

Chapter Two: Music Worlds 2.1 Introduction Academic literature has variously referred to the ‘social spaces of music’ (Crossley and Bottero, 2015) as ‘scenes’ (Bennett and Peterson, 2004), cultural ‘fields’ (Bourdieu, 1993) and ‘art worlds’ (Becker, 1974, 1976, 1982).3

In this thesis I favour the music worlds approach first adapted from the art worlds theory by Crossley (2008, 2009, 2015a, 2015b), in order to describe and explain the structure of the underground punk music network in two cities in Northern England; Manchester and Liverpool, from 2013-2015. In the following chapter I will outline the theoretical foundations on which this work is based and argue how formal social network analysis can help us to understand these music worlds. The first section of this chapter discusses early work on subculture by the Birmingham Centre for Contemporary Cultural Studies (CCCS) that brought punk to the forefront of sociology and cultural studies. I explain that although some of this early work is punk-focused, it is also consumer focused, it is concerned with style, aesthetics and construction of identity and I agree with Crossley (2008, 2009, 2015a, 2015b) that this is not appropriate for the exploration of a relational social structure, rather we should focus on the production of music. I then discuss the importance of interaction in the analysis of cultural production, in particular, Crossley’s (2011) conceptualisation of relational sociology and assert that interactions are the appropriate unit of analysis for sociological phenomena. To this end I discuss the key concepts of collective activity, conventions and resources in Art Worlds (Becker, 1982), as well as music venues, and how these concepts are relevant in the context of my research. Further to this I discuss substantive, but mainly ethnographic, work on music communities that has focused on the producers of music in the UK (Finnegan, 1989; Cohen, 1991) as well as internationally (Bennett, 1980; Gilmore, 1990; Shank, 1994; Novak, 2013) and the emphasis on collective production in this work. I go on to highlight Crossley’s (2008, 2009, 2015a, 2015b) work and how he operationalises art worlds by using formal social network analysis. Finally, I discuss how my work builds on this and other work on the social networks of music worlds to make a substantive contribution to the field.

2.2 Subculture Research on punk music has often focused on the idea of ‘subculture’, from early academic work (Hebdige, 1979) to current thinking (Subcultures Network, 2014). Gelder and Thornton (1997) explore in depth this idea of subculture its use as a theoretical framework for sociological study to draw out cultural meaning in everyday activities of those often seen as ‘other’ to the mainstream, homogenous ‘mass’ that dominate society and are influenced by mass media. Subcultures are more fluid and less official than ‘communities’ or ‘societies’, often 22 bounded by “collective perception” (Thornton, 1997 p.3), and identify with a “consciousness of otherness or difference” (Thornton, 1997 p.5).

The Birmingham Centre for Contemporary Cultural Studies (CCCS) was responsible for a wealth of literature that championed the idea of subculture as a framework for the analysis of groups that were portrayed as youth cultures resisting the ‘mainstream’ culture of mass society. Clark et.al (1976), present ‘subcultures’ as an alternative way to define ‘youth cultures’; conceptualised as local substructures of a wider cultural section of society defined explicitly by social class in a top-down, prescriptive way. They assert, for instance, that criminal gangs can be identified as a sub-group of working-class culture in London’s East End. These subcultures are positioned as opposite to the mainstream, homogenous culture that is perpetuated by mass media and the ruling class. They are comprised of dissenters that are defined by their adverse relationship to the norm and their position as disrupters to society as a whole. The Birmingham CCCS can be credited for bringing British sociological focus to punk. Hebdige (1979) uses several case studies of what he terms ‘subcultures’, including reggae and Rastafarianism, beats, teddy boys and mods, but with an overwhelming focus on punk. Punk is referred to throughout as a purposeful disruption of ‘nature’, mass society and behavioural norms. Despite its chaotic reputation, Hebdige argues that the coordination between the type of clothing, the outwardly profane and disrespectful attitude and the style and lyrical content of the music is evidence of a deliberate subversion of the status quo and construction of a somewhat homogenous set of aesthetics that could be wholly identified as ‘punk’. Hebdige’s analysis, therefore, hints at the importance of collective action and convention (Becker, 1976, 1982) in cultural worlds but this is not the focus of his analysis.

Characteristics of individual participants, from the subculture perspective, are the element of interest, but under a relational framework these characteristics are important insofar as they can help us to understand the social ties that exist between actors. Studies of subculture tend to focus on external factors that lead to reactionary resistance; factors including social class, poverty and political alienation constrain young people and cause them to construct new identities outside of mainstream culture. Restrictive social structures give rise to subcultures that are the catalysts for social change. Indeed, this is the rhetoric that is often cited in journalistic accounts of the ‘punk explosion’ in the UK (Savage, 1991; Haslam, 2007; Seven Ages of Rock, 2007); that it was some grand rejection of 1970s Britain and what little it had to offer young people for their futures. Analytically, this argument does not stand up on its own; it relies too much on a ‘them versus us’ narrative of class struggle and focuses too much on deprivation and poverty. In reality, deprivation, poverty, and class struggle are constants in British society that have been well documented; Victorian Manchester by Engels (2009);

23

‘Breadline Britain’ in the 1970s (Townsend, 1979), 1980s (Mack & Lansley, 1985; Gordon and Pantazis, 1997) and 1990s (Pantazis, Gordon and Levitas, 2006); and more recently, modern concerns about the widening gap between rich and poor, the rise of the working poor and mass poverty (Lansley and Mack, 2015). Tilly (2008) notes the fallacy of searching for “constant conjunction and correlation” (p.7) and that “unchanging fundamental motives” (p.8) cannot fully explain dynamic social processes. Concluding that the punk movement, for example, was a result of deprivation, poverty and class struggle is overly simplistic. The same is true of resulting feelings of disenfranchisement, alienation and angst; all of these are contextually important in the punk story, but they are omnipresent, whilst the collective action, micro-mobilisation and resource sharing that was required to reach a critical mass for the punk world to exist were a set of distinct choices made by relatively few people (Crossley, 2015a). These interactions and the mechanisms behind them can help us to uncover the social processes that both create and sustain music worlds.

Whereas the subculture literature tends to focus on aesthetics and the construction of identity in punk music, I set out in this thesis to uncover both local and global structures that are key in the realisation of the music as live performances, how we can describe and explain these structures, and what they mean in the context of the underground punk music world and how it operates. As a sociological framework, subculture ignores the social structure of the groups themselves; the social ties and interactions between participants that are constituent parts of any collection of individuals. Simmel (1908; translated by Wolff, 1950) notes that the groups we think of as society are, at a base level, interactions between people that become cemented as “autonomous phenomena” (p.1). In a music context, Martin (1995, p.178) echoes this sentiment when he says that…

“…pattern or structure is not the cause of people’s activities, but rather their outcome”.

In addition to this, I would argue that structure also constrains interactions, a point I will return to in the following chapters.

The need for this ‘relational’ approach to sociology is outlined by Crossley (2011); for Crossley, analysis of society using either ‘atomistic’, considering individuals, or ‘holistic’, considering society as a whole, is inadequate. Instead, a ‘relational’ approach to sociology is posited as a framework for sociological investigation.4 This approach asserts that interactions are central not only to the formation of groups, organisations and institutions, but also to the construction of identities; it is through these interactions that we become social actors and the ‘mechanisms’ behind these interactions can help us to understand and explain society. Interactions, therefore, are irreducible to lone actors or whole groups. Proper sociological investigation of cultural production under the framework of Crossley’s (2011) relational 24 sociology then, needs to place the interactions between actors as central to the analysis and explanation of social phenomena. In the context of music, this theoretical approach is best embodied by Howard Becker’s (1974, 1976, 1982) theory of ‘art worlds’ and subsequently, I proceed by explaining how the key concepts of Becker’s theory play out in the context of the underground punk music world in Manchester and Liverpool.

2.3 Towards Art Worlds Both Bourdieu’s ‘fields’ (1993) and Becker’s ‘art worlds’ (1974, 1976, 1982) have long been established as seminal theories in cultural sociology, and either can serve as a lens through which cultural sociological investigation can be wrought. Before I introduce Becker’s concept of art worlds as a theoretical starting point, I will briefly explain Bourdieu’s (1993) concept of cultural fields and why I favour Becker’s approach over this. For Bourdieu, artists are constrained by a cultural “field of forces” (Bourdieu, 1993, p.64) and attempt to attain objective ‘positions’ within it, ranked by power and social capital. Consequently, the relationships that Bourdieu is largely concerned with are between artists and structural positions. What I am concerned with in this thesis, however, is interaction. Interaction between individual musicians, the bands they play in, and the venues at which they perform. Whereas Bourdieu (1984, 1993) attempts to downplay interactions and emphasise the imposition of societal structure upon them, Becker places collective action and interactions themselves at the centre of ‘art worlds’ and therefore I adopt his theory as the basis for my own work.

Below I highlight the main concepts outlined in Becker’s theory of art worlds and explain how they relate to the underground punk music world in Manchester and Liverpool. I also elaborate upon music venues; mentioned by Becker but not central to his work; and why they are important in the context of a music world.

There are four main ideas that are central to my conception of music worlds:

1. Art as collective activity – for art to exist, it requires a of cooperation between many participants. 2. Conventions – there are a set of rules, norms, perceptions and behaviours that are known and either formally or tacitly spread by participants in art worlds and these are related to the formation of network ties. 3. Resources – there are materials and personnel that are essential to the production of art and constrain the formation of ties between both musicians and bands, and bands and venues.

25

4. Venues – places of performance provide the social spaces for interaction, which I argue are not only governed by conventions, but are also the places where conventions are formed, changed and realised through performance.

Below I discuss the concept of collective activity and why it is relevant and important in a music world; the idea of conventions and how these conventions contribute to people’s ability, preference and motivation for this collective action; resources and how they are mobilised by musicians and bands; and finally, venues and their critical importance in the formation of network ties.

2.4 Collective Activity Becker emphasises the fundamental role that collective activity plays in an art world as well as

“... the complexity of the cooperative networks through which art happens” (Becker, 1982, p.1).

Collective action has to take place before any sort of defined music world can exist. People need to meet each other; they have to interact together, write songs together, learn these songs, rehearse them together and when they are happy with the standard decide to perform together at gigs. There is both intention and purpose in these actions to create music and for this music to be performed. Punk music in this way is very much a performance-focused genre of music as the energy and delivery of the live performance, often along with the volume - without being scientific in terms of decibels here, the usual intention is to be loud - is as important as the aesthetic qualities of the music; ‘art happens’ (Becker, 1982, p.1) in the punk music world through performance. In this context then, there are two stages of collective activity; the activity between the musicians to organise themselves as a band, and the interaction between the band, as both a whole entity and a group of individuals, with the venue or event at which the live performance takes place. These venues, where punk is realised in performance, are essential in the underground punk music world as they provide a platform for output for the bands involved and also provide a space for what Small (1998) terms ‘musicking’, taking part in musical performance by all those involved; the musicians, the bands as collective entities, the audience and the venue staff and it is these interactions that are essential to sustaining the underground punk world. The collective activity in the network of musicians, bands and venues is both influenced and constrained by both conventions and resources, which are now elaborated upon in turn.

2.5 Conventions Collective activity is governed by conventions, which, for Becker (1982), are sets of norms, behaviours and aesthetics that are adopted by groups of people in particular art worlds. They 26 are formed and reformed over time, constantly subject to change, but there is rarely a case when all of these conventions have to be established from scratch.5 In the underground punk music world I contend that these conventions both influence, and are formed by the interaction between musicians, bands and the venues at which they perform, and play out at these different levels of tie formation6. I will discuss the influence of conventions on tie formation in two parts; those that have bearing on ties between musicians and bands that constitute band membership; and those that have bearing on ties between bands and venues where a band has chosen to perform at a given venue.

In order to understand conventions that condition musician-band ties, it is appropriate here to briefly outline the composition of bands that are active in the underground punk music world. The overwhelming majority of underground punk bands (and punk bands in general) are guitar-based and the line-up usually consists of some combination of vocals, lead/rhythm guitar, bass (by this I mean bass guitar or less frequently, double bass) and drums. Ska-punk bands are the most common exception in that they include brass players and less often, saxophonists (although typically made of brass, they are played with a reed and are classified as woodwind), and folk-punk bands often include violinists, accordion players and mandolin players. Although they feature in many types of popular music, pianos and keyboards rarely feature in punk music. The conventions of band composition constrain the formation of membership ties between musicians and bands. If a band already consists of two guitarists, a bassist and a drummer then it is unlikely that an additional guitarist will join the band, and highly implausible that an additional drummer will join the band unless there is a vacancy created by the departure of one of these members; whereas if a band has only one guitarist, there may be scope for a second guitarist to join the band as a two guitarist line-up is considered ‘conventional’ but this again is constrained by inherent band preferences for either one or two as well as the type of music the band wants to play. It can also be constrained by resources of bands and venues. In the case of a band, members may travel to play gigs in one car, where three members and all the equipment manage to fit in that car so the band can be efficient with transport costs and additional members may disrupt this. In the case of a venue, it may only be able to accommodate a small number of musicians on a stage or may only be able to provide a certain number of microphones, microphone stands or other sound equipment that is required for live performance, thereby restricting the types or sizes of bands that are able to perform. There is therefore interaction between convention and resources in a music world and it is important to consider them as simultaneous in this way when we discuss mechanisms of tie formation. The ability for members to join different bands or form new ones can be restricted, or made easier, by conventions.

27

One simple way bands are delineated is through the sub-genre of punk that a certain band will categorise themselves as playing. These sub-genres of punk share playing techniques that are likely to be transferrable between bands of the same sub-genre. For instance, a drummer from a folk-punk band may find it difficult to join a hardcore punk band because the playing style is so different. It could be argued that the drummer may be more than capable of playing the rhythms required but it make take longer for them to learn songs or be ready for gigs or a tour, constrained by the resource of time. It may be the case that the drummer listens to hardcore punk music and is familiar with the playing techniques despite playing in a different type of band; musical tastes are not mutually exclusive. In this case the network of performance is likely to play a role in the probability of tie formation and the dependency between band membership ties and band-venue ties play out – consider the hypothetical case where a drummer is perfectly capable of playing in a hardcore punk band and could easily join and perform with a hardcore punk band, despite currently performing in a folk-punk band. Unfortunately, this drummer may only ever be seen by other musicians as a folk-punk drummer; at venues, people see the drummer performing with the folk-punk band and he projects himself as both a folk-punk drummer and is associated with that particular band. As a result, people looking to recruit a new drummer for their hardcore punk band may not consider asking that person. It may also be the case that gigs tend to be genre specific in that bands of similar sub-genres end up performing at the same gigs. In this case, the drummer may never be exposed to bands that might recruit them and only interacts with musicians that want to play folk-punk music. In reality, audience members’ musical tastes might be varied, and bands may be on more eclectic bills than this case, but this scenario serves to highlight the importance of convention and also alludes to the role of venues in tie formation, as they provide the space where interactions take place.

2.6 Resources Mobilising resources, for Becker (1982), is essential to an artist’s ability to carry out their work and they are split into two key groups; material resources and personnel. Lack of resources is a barrier to entry into the music world, but punk music has a relatively low threshold for such resources, epitomised by the synonymy of punk with the term ‘DIY’ (Do-It-Yourself). Punk, and expanding to in general, differs from classical or ‘art music’ in that it is not required of musicians to learn in a formal way through music lessons, which can be expensive and restrictive; 74% of children in the UK from higher social class have had music lessons, whilst the figure is just 58% from lower social classes (ABRSM, 2014); and as such, people become punk musicians and are recognised as so by playing and performing in a punk band much in the way Bennett (1980) ascribes to rock. Whilst I recognise that resources play a role in recording, physical media production, publicity and a number of other band activities, my 28 analysis in later chapters focuses on performance and as a result that is also my focus here. Resources contribute to the formation of relational ties between musicians, bands and venues in two contexts:

Musicians and bands – in order for a musician to become and continue to be a member of an actively performing band they first need to have access to the required resources.

Bands and venues – when bands perform, there are certain resources that are needed and expectations of who will provide these resources are governed by conventions.

In general, musicians will become members of bands in two different ways; either the band is started organically and all musicians become members of the band at the same time when it is formed; or a musician joins an existing band as a new member, either replacing a previous band member or in addition to the other members of the band, for instance an additional electric guitarist, or joining on a different instrument outside the core of guitar, bass and drums, perhaps as lead vocalist (when a guitarist or drummer previously sang lead vocals), or an additional brass player in the case of a ska-punk band. In both situations, a band is constrained by the pool of available musicians, which can be affected by a number of different factors. For instance, on a national scale, only 34% of adults in the UK population play a musical instrument (ABRSM, 2014). The number of musicians is further reduced by the number of people that play the required instruments for a punk band, and then further by the number of those people in geographical proximity to each other and subsequently by individual taste in punk and what sub-genre of punk music a given person wants to play. As a participant musician in the underground punk music world, although I was not heavily involved in performing during the time period in question, I recognise that I was somewhat lucky in that I met the other two members in my band during my time in both secondary and further education, through both music courses and extra-curricular activities in institutions that were very encouraging of musical pursuits.

Although some band members may be friends through education, the difficulty of finding like-minded musicians for those who do not have access to such friends is somewhat reduced by first being an audience member at a gig; musicians interact in venues, what Feld (1981) would refer to as ‘foci’. These phenomena of band formation as either a result of school friendships or interactions in musical spaces are documented qualitatively by Cohen (1991), in her ethnographic account of Rock Culture in Liverpool and Bennett (1980) in his qualitative analysis of the process of becoming a rock musician. For a musician to become a performing member of a band, they also have to have a basic level of equipment, their instrument; electric guitar, bass guitar or drum kit (this is slightly different in the case of a singer as microphones are usually provided at both rehearsal spaces and gig venues). Monetary resources are required 29 to purchase instruments, amplifiers and other required equipment; rehearsal time with the band and transportation for these activities, either via public transport or more likely in their own vehicle. Material resources for performance fall into two categories, those that are provided by the musicians and those that are provided by venues. Most of what is expected to be provided by musicians and venues is governed by established conventions and some are open to interpretation. Guitarists and bassists for instance, are expected to bring their own guitars, guitar leads, tuners and other pedals and amplifier heads at a minimum with speaker cabinets provided by the venue. The amplifiers have to be of certain power to be loud enough for performance as the types of venues that underground bands perform at do not always have equipment to ‘mic up’ the amplifiers and project them through the PA system.

As a minimum, venues are usually expected to provide a PA system for vocals as well as microphones. In terms of personnel the venue also usually supplies a sound engineer for the night. The set of required resources for a gig performance could also be considered a ‘convention’; there is a bare minimum of equipment a band is expected to bring and a bare minimum of equipment venues are expected to provide, as well as flexibility in some other equipment such as drum equipment and speaker cabinets. This helps the gig booking process to operate more smoothly, as bands know what to ask for beforehand and also what they are required to bring, with some equipment assumed to be the responsibility of either band or venue.

2.7 Venues It may seem obvious when discussing music venues that the main purpose for the existence of such places is for live music to be performed. Indeed, venues facilitate live music performances, but they can take a variety of forms and promote and facilitate live music for a number of different reasons. Venues might simply be pubs; with or without a purpose-built stage; purpose built live music clubs or may even be temporary or squat venues. Pubs may put on live music for commercial reasons, in order to attract patrons and increase takings, alternatively they may put on music because the owner or landlord is passionate about live music and wants to provide a local space to hear live music. More often than not venues in the underground punk music world are multi-function venues where music is not the only event and it will not be limited strictly to punk music. Venues are also subject to change, either through changing hands to new owners or new landlords with changing preferences for particular music, events, or that may not want any sort of live music; or through closure (Snapes, 2018). In addition to the financial problems currently faced by British pubs caused by high business rates, beer duty and VAT (BBC, 2018), owners also have to invest in costly equipment and are legally required to purchase annual live music licenses to cover royalty

30 payments (PRS, 2018) and are subject to noise complaints (Snapes, 2018). This precarious nature of venues appears to be a constant theme in music worlds, documented by Haslam (2015) in his book on the history of British music venues.

As a participant in the underground punk scene I have first-hand experience of the importance of venues. Venues are where conversations about music take place, where individuals can express musical tastes and opinions and meet like-minded musicians. It is this social space that facilitates interactions between musicians in close proximity that will not occur in any other social space and therefore the opportunity for interaction that will lead to the formation of new bands and new gig bookings is extremely high.

In much of the literature on music, conventions are linked inexorably with performance venues although they are often referred to as static, prescribed and established entities. Ruth Finnegan’s (1989) study of music making in Milton Keynes argues that musicians find ‘pathways’, referring to a need for folk musicians to match their playing style to a venue where it will be well received and that this can be a trial and error process whereby the audience feedback determines whether the musicians seek to return. Cohen (1991) also hints at the role of established conventions in accounts of gigs where the bands’ music and the expectation of the audience do not match up, resulting in a negative experience for band members that will influence their decision to return to particular venues.

Venues are the physical sites where punk performances take place, and venue choice, when a band makes a decision to perform at a particular venue, can be constrained by these established conventions. If we consider genre as one of these conventions, it may be the case that one venue is particularly predisposed to book ska bands and therefore more ska bands are attracted to playing this particular venue and it might be in their interest to do so because it is known that the audience responds well to that particular genre of music. I argue however, that conventions are more dynamic than this and are both formed and dynamically altered in venues through social interactions.

Becker (2004) defines ‘places’ as physical spaces where music happens, ranging from small buildings to large cities. These places, for Becker, are imbued with social meaning by the participants within them that share ideas about how places will be used and what types of people will participate but this is not investigated any further than this. I ascribe this phenomenon specifically to venues in the underground punk network and argue that they are simultaneously given and develop social meaning, whilst also allowing bands and musicians to develop and change conventions. Haslam (2015) alludes to venues as spaces that facilitate the formation of bands as well as innovations in genre and styles, with particular reference to Lennon & McCartney, and Van Morrison, as audience members at skiffle 31 performances, meeting and discussing music. Venues “carried a sense of potential within their walls” and were foci for interactions “where new ideas were born”. Venues are simultaneously both sites of innovation and catalysts for change. They are where conversations happen; interactions that lead to the formations of bands; plans are formed and enacted, and potential band members are in close proximity to one another. Collective action happens as a result of these interactions. It is also the social space where people see bands for the first time and are inspired to start their own band, or in the case of the Sex Pistols gig at the Lesser Free Trade Hall, the post-punk music world. Venues, therefore, provide the social space for these intricacies to play out and conventions to change through interaction between both musicians as individuals, and bands as collective entities with their audiences.

2.8 Music Making Literature on music-making has tended to be ethnographic and contained within defined geographical boundaries, usually individual towns or cities, for example, Milton Keynes and Liverpool in the UK (Finnegan, 1989; Cohen, 1991), and Austin, Texas in the USA (Shank, 1994). For Finnegan, ‘music worlds’ are defined explicitly by genre (p.106), for example, there are the worlds of punk and new wave, traditional folk, jazz and rock ’n’ roll among others. There are pre-ascribed sub-genres of punk contained within the underground punk music world, but it is my contention that these are not static and have the ability to be shaped by interactions between both individual musicians, and bands and their audiences in the social spaces of performance venues. Cohen’s (1991) work on Liverpool rock culture hints at this interaction between band and audience in the shaping of genre, observing that bands in Liverpool continually adapted and altered their music in order to satisfy their own tastes, but also their audiences; both people who attended their live performances, as well as potential audiences they wanted to reach and industry professionals that could help them to be commercially successful. I contend that this is a dynamic process that can happen in two distinct ways. Genre can form independently of venues; where individual members bring their preferences for certain musical genres and ways of playing to a band. It can also form in the interaction contained within the venues at which bands perform; they respond to their audiences, recruit new members with differing tastes and are influenced by styles and playing techniques of other bands and musicians that they watch. Cohen also discusses band formation and the fact that musicians tend to be in multiple different bands (see diagram in Cohen, 1991 p. 29). Band membership and how these multiple memberships affect venue choice is explored using tripartite analysis in chapter five.

32

2.9 Underground Punk Music in Manchester and Liverpool In this thesis I have chosen to focus on the punk music world of Manchester and Liverpool during a three-year time period from 2013-2015. The structure I will analyse takes the form of the live performance network of bands that have performed at particular venues and events, as well as the musicians that play in those bands. I am aware that participation in a band involves more than live performances, it also includes writing music, collaborating with other musicians, recording music, and can involve completing other tasks such as booking gigs, marketing and promotion of the band and releasing music; tasks that may be undertaken by either the bands themselves; where one person may do everything or roles may be spread out in any number of different ways; or alternatively these tasks may be completed by hired professionals, enthusiasts or friends of the band. In order to keep the geographical boundaries a little wider than has been the case in previous literature; most studies of music worlds involve one town, city or a small ‘scene’ or ‘world’ within a single town or city, often defined by music genre (Cohen, 1991; Hield, 2010; Novak, 2013); and to be able to say something meaningful about the underground punk music world in Manchester and Liverpool I have settled on a more focused approach considering only the networks of live performances including musicians, bands that they are members of and the venues at which these bands perform.

The sound of the bands involved in the underground punk music network in Manchester and Liverpool is very eclectic. There are bands that will fit strictly into certain categories and bands that take influences from a variety of other musical genres. Some will simply have a punk ethos or punk lyrics. The difficulty, as with any network study is where to draw the boundary line of what is considered punk music and what is not. For the purposes of meaningful analysis, I have decided that if the band themselves describe their music as punk then that is a sound enough justification to be considered. The fact that the bands have such a wide ranging musical taste is because there has been a wide range of punk music released since the first punk records were released, and as time elapses this results in a much larger pool of influences to draw from, especially now that music can be accessed so easily via the internet. It is also much cheaper and easier to record and release music than it was forty years ago and access to these resources means that music is much less streamlined, at least in recorded form, than it used to be.

The age structure is diverse, some people were involved in the scene at its first incarnation and some were only two or three years old when 90s American punk first sprang up. This creates a broad range of views on where punk music sits in society and to what extent it should be political or associated with politics or is even a form of resistance ascribed by the subculture

33 literature. One overarching idea that appears to permeate through all these different groups is one of DIY, the culture of bands wanting to do things for themselves and at the very least this relates to creative control over the music that they make and at the other end of the spectrum involves financing recordings, self-released music and self-booked shows and tours both in the UK and Europe. As a result of this, I expect individuals and bands as collective entities to possess different ideas about the way that punk music should sound, lyrical content of songs and underlying messages in the songs. As Ruth Finnegan (1989, p.7) puts it:

“…defined in different ways among different groups, each of whom have their own conventions supported by existing practices and ideas about the right way in which music should be realised.”

Thinking about this idea from a relational perspective we can parse out how the differences between these bands may result in different patterns of tie formation in a network. These preferences may also depend on exogenous attributes outside musical life such as pre-existing friendships or geographical proximity. From a statistical point of view, these differences provide the variability in the network that we can investigate through network modelling. Before I elaborate upon statistical modelling of networks, in the following section I discuss the literature on using formal social network analysis techniques to analyse music worlds.

2.10 Social Networks and Music Worlds More recently, sociological scholars have used formal social network analysis (SNA)7to investigate specific music worlds including punk and post punk (Crossley, 2008, 2009, 2015a, 2015b), feminist music worlds (O’Shea, 2015), Britpop (Millward, Widdop and Halpin, 2017); folk music (Hield and Crossley, 2015) and classical music (McAndrew and Everett, 2015a, 2015b).

Crossley (2008, 2015a) investigates punk as a ‘music world’ under the SNA framework when analysing the early punk music network in London. For Crossley, punk music was not simply some response to political and social conditions and did not spring from nowhere in a flash of anger and frustration. Punk was formed through social interactions that took place between key participants. Before any of the commercial actors were involved Crossley explains that a network of participants of early punk formed a ‘critical mass’ that was required to generate bands, managers, and other peripheral personnel for the punk movement to happen. In his analysis, Crossley constructs his network ties between individuals where they were; in a band together; worked together; lived together; or were in a romantic relationship. In this way ties are concrete relationships and as such, once they are formed they cannot be broken; this additive method of tie construction measures the growth of the network to reach the critical

34 mass required to form a music world. Crossley (2009, 2015a) also uses this method in later work on the music worlds of punk and post-punk in London, Manchester, Liverpool and Sheffield and this approach to tie-definition is adopted in other studies of music worlds using secondary data, including work on female composers (McAndrew and Everett, 2015a, 2015b), Britpop (Millward, Widdop & Halpin, 2017) and more loosely using collaboration to define ties in Francophone rap music (Hammou, 2015). O’Shea (2015) also uses ‘strong relationship’ ties but these were primary data collected via a face-to-face questionnaire.

Here I am concerned with network activity, rather than growth, in the underground punk network in Manchester and Liverpool over a period of three years, 2013-2015. The time period I am considering is not the start of a world or the genesis of a movement; I seek to explain how a network functions that is continuously operating and constantly changing where bands form, change composition and split up, new venues open and others cease to exist and promoters appear and disappear. It captures a particular snapshot in time. In this regard, the ties that I wish to explore pertain to the function of performance; ties between bands and venues, where a band has performed at that venue, and ties between musicians and bands, where that musician is a performing member of that band. Ties are not accumulated, but rather dynamically change between each discreet year.

Some elements of the type of data I analyse are present in previous SNA studies of music worlds; Edensor, Hepburn & Richards (2015) use data collected from Facebook; a data source I make great use of in constructing band-venue ties; to generate a network of contacts of one musician. Hield and Crossley (2015) use co-participation at events to analyse the folk world in Sheffield. For this co-participation network, Hield and Crossley (2015) mention that conventions are tied to events and that these conventions are subject to change as the composition or taste of participants that attend these events changes and that new events are created to cater for tastes that do not fit with established ones. This hints at the fact that conventions are subject to change but this argument is underdeveloped, and is not investigated using network methods; tastes are said to change independently from events and in turn, event conventions then change. It also neglects to mention the importance of the social space of the venue in the forming of these conventions and the dynamic interplay between musicians, bands and their audiences that happens through live performance activity.

It is my contention that conventions are shaped and realised in the social space of venues and that this can be investigated through analysis of network ties between musicians, bands and venues. To this end, I wish to investigate the drivers of network tie formation in a tripartite structure and push the methodological boundaries of the analysis of music networks alongside showing that statistical models are a viable method for investigation of these networks and can

35 provide valuable additional insights into their inner workings. It is my intention in the following chapter to elaborate further upon literature on network methodologies, how I intend to use existing analytical tools, and how I intend to push the boundaries of exploration and make a contribution towards the development of new methodological techniques for tripartite network analysis.

36

Chapter Three: Network Theory 3.1 Introduction In chapter two I established the need for a relational network perspective to analyse music worlds, as well as highlighting social network analysis as a set of methodological tools to investigate the relationships inherent within a music world. Subsequently, in this chapter, I expand further on both the theory and methodologies of network analysis, arguing that current methods are inadequate to investigate the research questions I pose in this thesis and that there is a gap in the current literature for the development of empirically tested tripartite methods for the analysis of social networks. In order to make a methodological contribution to the field of network analysis, this chapter constructs a theoretical and methodologically sound justification for tripartite methods to be used in this thesis, as well as advocating that they can be and should be used more generally. Arguments are made with particular reference to networks of music worlds, but the intention is to show that where tripartite structures exist, they should be analysed in a tripartite context and we should not dilute our analyses by neglecting the true structure of a network. Tripartite methods will allow researchers to investigate new types of research questions, analysed using tripartite methods that test for the prevalence of tripartite network phenomena.

I begin by outlining my perspective on network analysis and what we should aim to achieve as researchers in the discipline, drawing attention to the fact that although network literature often refers to networks as complex phenomena, there are relatively few examples of empirical research that address complex questions, with particular reference to networks with three or more types of nodes, a subject matter where literature is distinctly lacking. With a view to logically progressing towards tripartite analysis I give an overview of ego-networks and why I favour a ‘whole networks’ approach for the analysis of the underground punk music world. I explain that although ego-networks can be useful for certain types of research questions, they are not conducive to answering the questions that I pose in this thesis. Subsequently, I explore the topology of networks, distinguishing tripartite analysis from other forms, alongside explaining differences and nuances in terminology that are important but often conflated.

First, I give examples of analysis of unipartite, or what are often referred to as ‘one-mode’ networks, where ties exist between nodes of a single type. In this context I discuss the main literature that has been developed around the structure of these networks, as well as measures that have been developed for analysis. Again, I stress the importance of applying appropriate methods to answer the research questions we wish to ask and give examples of where bipartite networks, where ties exist exclusively between nodes of two different types, have been 37 projected to form unipartite networks, resulting in loss of information. I also discuss the dual projection method proposed by Everett and Borgatti (2013) and why this method is inadequate for my purposes. This is followed by literature on what Borgatti and Halgin (2011) term the ‘direct’ approach to analysing bipartite data, where the network structure remains intact, with reference to both deterministic and stochastic methods. Subsequently, I progress to tripartite methods, discussing the difference between this approach and a multilevel approach to answering my research questions, which again in the case of explicitly tripartite questions, would involve projection of ties and loss of data.

I note the efforts of Fararo and Doreian (1984) in developing mathematical formalisations of a tripartite network, Borgatti and Everett (1992) for specifying blockmodels for multimode matrices, Diani (2015) in highlighting the importance of duality and complex structures in cultural networks and Bohman (2012) in attempting to empirically test a method for tripartite analysis and for shedding light on the rich information that is lost in projecting owner level ties in the context of interlocking directorates on the corporate boards of Swedish firms. Finally, I summarise key arguments in the chapter and explain how I will proceed with analysis of a tripartite network of musicians, bands and venues.

3.2 The Purpose of Network Analysis Formal social network analysis gives us the tools to investigate the relational structure of a social world in an empirical way. In addition to this, as network researchers, we view society from a relational perspective, arguing as I have in the previous chapter, that social structure is borne out of social interaction. It follows then that if we consider these interactions as the core components of the structures we wish to investigate, then we should keep them intact in our analyses. Networks are inherently complex and our job as social network analysts is to ask questions and conduct analyses that give us an insight into these complex social structures. As social scientists, the data we deal with are inherently messy – people are highly unpredictable and do not follow universal laws akin to those found in natural sciences – and so we are faced with a challenge to place our questions and analyses in a space where we gain new understanding whilst avoiding the abstraction that comes with over simplification and the loss of purpose that comes with over complexity in the conclusions we draw. If we wish to investigate complex worlds we must represent the true structure of these worlds and ask relevant questions related to these true structures. As well as this we must ensure that our research methods are adequate to respond to our research questions. Consequently, it is my contention that new methods must be developed to allow social networks researchers to investigate networks with multiple types of actors and organisations, the ‘nodes’ in the network to avoid the loss of important network information. It serves that the more

38 information we have about a particular network, then the clarity, accuracy and insightfulness of our conclusions about the network and actors within it will be significantly improved. To illustrate the need for analyses of nodes of multiple types, in the following sections I discuss network topologies, building from ego-networks through bipartite and multilevel networks to arrive at my own conception of tripartite analysis, reasoning why this is appropriate in the context of music worlds and suggest the kinds of questions we might want to ask of other cultural worlds as well as in network research beyond the cultural field.

3.3 Ego-networks In this section I elaborate upon the reasons for preferring a ‘whole’ network approach in my analysis of the underground punk music world in Manchester and Liverpool from 2013-2015, rather than an ‘ego’ network approach. It is not intended as an argument against the use of an ego-nets approach per se, merely that it is not an appropriate method to adequately answer the questions I pose in this thesis.

All networks, from the perspective of formal social network analysis, must consist of both a defined set of nodes; these can be individuals, collective entities such as organisations, events or venues; that are connected by a defined set of ties. There can also optionally be a set of attributes for each node; for individuals these might be typical demographic information such as age or sex, but may also include other behavioural traits such as smoking or drinking; for organisations these may be, for instance, related to the type or size of the organisation. These ‘attributes’ are characteristics of the nodes that are either important in explaining network patterns or behaviours or that we suspect may impact on formation of network ties and we therefore wish to control for these factors. As the name suggests, ego-networks are ego- centric, so our set of nodes consists of one ‘ego’ and several ‘alters’ to whom the ego is tied, and optionally, ties between these alters.

Usually we want to take account of multiple ego networks to be able to infer something about a population of individuals. A common example of an ego-network is a network of social support, where the ego at the centre of the network, our individual of interest, is tied to other individuals or ‘alters’ in the network. These ties would be relationships with people that our ego may ask for advice or help from, confide in or borrow money from and we are able to compare individual networks to one another to draw conclusions.

Ego-network analysis lends itself well to the subject of social support because it provides us with the appropriate tools to investigate. In this instance, we are not necessarily interested in the overlapping ties, rather what an individual’s support network looks like, in order to compare between individuals and therefore overall network structure is not a concern. It also

39 gives us the ability to account for many different actors that are not necessarily linked. One person’s support network may not overlap with another person’s network if we are interested in a particular population, for instance, migrant women (Kornienko et al., 2018). We are able to analyse the networks separately and compare overall structures, with scope to draw inferential conclusions. If, however, we need to account for overlapping ties and our population of interest constitutes an interdependent structure, then we need to take account of the whole network and actors’ positions within it.

The effect of attributes of actors within a network are also harder to detect in an ego-based network analysis. For instance, we may wish to investigate the effect of gender on the formation of friendship ties in a school classroom and hypothesise that these ties are homophilous by gender, that boys are more likely to form friendships with other boys and girls with other girls in our school class. This effect of homophily is difficult to detect using an ego-network approach and is not possible to assess in tandem with other structural parameters. Naturally, if we are concerned with social structure then it makes sense that we would like to describe and analyse this structure as a whole, as well as wanting to measure actors’ significance as part of this structure, their roles within the network as a whole and the prevalence of certain network patterns such as the tendency towards closure.

In the case of the underground punk world, we are trying to describe and analyse a group of musicians, bands and venues that are very much inter-connected, musicians are members of multiple bands and bands perform at multiple venues. These overlaps are what make the network interesting and using ego-network analysis would be inadequate to explain these phenomena, rather we should analyse the ‘whole’ network. In lieu of this, conceptions of whole networks with nodes of one, two and three types respectively are discussed in the following sections.

3.4 Unipartite Networks

Prior to discussing literature further, it is important to be clear about definitions and terminology. I prefer, in this thesis, to use the terms unipartite, bipartite and tripartite as opposed to ‘one-mode’, ‘two-mode’ and ‘three-mode’ to describe networks. The reason behind this is that I feel the use of the ‘modes’ terminology by itself is ambiguous as it refers to the separation of nodes into groupings that make sense analytically (Everett and Borgatti, 1992), rather than concrete structural definitions. The ‘two-mode’ label, for instance, might encompass both networks where ties are found solely between different ‘modes’ or types of nodes, for instance bands and venues, as well as networks that contain within-mode ties, although these are usually referred to as ‘multilevel’ networks, which I discuss later in this

40 chapter. Whilst ‘modes’ refer to researcher defined node sets, the terms bipartite and tripartite originate from graph theory. Social networks are often represented formally as graphs in order to utilise mathematical methods to analyse and draw conclusions about social structure in a network, but we must be careful not to conflate graphs with networks. In order to be clear, here I define a network as a set of connected actors that can be represented in a graph. Below I elaborate further upon definitions of bipartite and tripartite graphs, how they differ from two-mode and three-mode networks and how I define bipartite and tripartite networks.

Consider a graph 퐺 = (푉, 퐸) where 푉 is a set of vertices and 퐸 a set of edges that connect those vertices. Where a pair of vertices is connected via an edge they are considered adjacent. A 푘-partite graph is a graph where the set of vertices 푉 can be partitioned into 푘 disjoint sets, where adjacent vertices are exclusively from disjoint sets. Graphs where 푘 = 2 and 푘 = 3 are referred to as ‘bipartite’ and ‘tripartite’ respectively. Although as researchers we can partition graphs into node sets ourselves by assigning analytical modes, for example, musicians, bands and venues, it is important to note that 푘-partite graphs can be mutually exclusive of 푘-mode networks. Bipartite graphs are not only found in representations of two-mode networks or tripartite graphs in representations of three-mode networks. In fact, graphical representations of one-mode networks can take the form of bipartite graphs. For example, we might have a network of 16 tennis players, 8 men and 8 women, from a local club who play in a mixed doubles league. Ties between the players constitute who partnered whom in league matches over one season. The players may have regular players whom they partner with, but playing partnerships change throughout the season as not all players are available for the entire season and some pick up injuries. In mixed doubles matches players are only permitted to partner other players of the opposite sex and so a graphical representation of this one-mode network will form a bipartite graph with vertices that can be split into two disjoint sets, one set of male players and another set of female players. When I describe a network as bipartite or tripartite, I describe the structure of the ties between the two or three modes of the network, where the different modes form disjoint sets of nodes in the network and ties between nodes in the network are exclusively between these disjoint sets.

The following section gives an overview of what I refer to in this thesis as ‘unipartite’ networks and what are more commonly referred to as ‘one-mode’ networks in the literature, networks with nodes of a singular type. It covers commonly used measures for analysis of these networks and the social phenomena they typically try to explain. At the simplest level, this would mean analysis of what are often called ‘one-mode’ networks, with a single type of node. These networks in simple form are the population of interest that we wish to investigate, and the basic units of analysis are the ties between actors.

41

As I have discussed above, if we want to detect and assess network phenomena such as closure, brokerage, roles and positions, or, considering attributes, homophily, we need to use a ‘whole’ network approach where we look at a set or sets of nodes, connected by a set or sets of ties and conduct analyses of all nodes and ties in each network simultaneously. As such, further discussions of unipartite, bipartite and tripartite networks refer implicitly to ‘whole networks’.

3.4.1 Network Phenomena The purpose of this section is to give an overview of the theoretical literature on networks, explaining the ‘network phenomena’ that have been extensively written about by scholars in multiple different disciplines, as well as methods for the investigation of these phenomena and examples of substantive research questions that I address in the thesis in relation to these theories. From a ‘whole’ network point of view, scholars are interested in both the overall structure of the network and substructures within it; roles and positions within the network as well as the effect of exogenous covariates on the structure of the network. We wish to find out why certain ties exist, and others do not. A key tenet of network-based research is an emphasis on dependency, that our variables of interest, ties between individuals, depend on and can be influenced by each other. Below I outline key theories in network research that have attempted to explain tie formation mechanisms and the social processes that underpin them with reference to their relevance in this thesis.

Social networks are based on the idea of relational interdependence and many theories on social networks attempt to explain both the mechanisms behind the formation of ties as well as the flow of information or communication through ties and how this relates to social capital in networks, as theorised by Coleman (1988). The idea of social capital is that there are relational structures that are beneficial to an actor in some way. It may be that higher social capital affords one particular person strong social support, swift communication or easy access to information by virtue of their individual position in a network, in the context of every other person’s position in that network. As such, we may wish to measure a perceived tendency towards certain tie formation processes that are driven by actors’ want for increased social capital or to avoid loss of social capital in a given context.

Two seminal theories that identify with theories on social capital are Granovetter’s (1976) strength of weak ties theory and Burt’s (1992) structural holes theory, both of which share underlying principles in relation to how social structure is formed. Granovetter argues that strong ties are more likely to overlap with each other, positing that closure happens in a network often when two individuals A and B are friends and A is also friends with another individual C; it is likely that C and B will become friends by nature of their mutual friendship 42 with A. This tendency towards network closure and overlapping ties a key concept in social network analysis and networks are often characterised by their tendency towards closure and cohesive subgroupings of individuals. The key argument in Granovetter’s work though, is that if we wanted to use our friendship network to look for a new job, for instance, we are more likely to find a job through acquaintances as they are more likely to be sources of novel information. This is because our close friends are more likely to overlap with each other and perceived ‘weak’ ties connect us to new groups of individuals that are sources of new information and should actually be considered to be ‘strong’.

Burt’s (1992) theory of structural holes follows similar reasoning, arguing that homogeneity of attributes including knowledge, opinions and behaviour is more prevalent within groups than between groups in a network. Brokers are individuals that create bridging ties between two otherwise disconnected groups of people and in bridging these gaps develop social capital for themselves as they are privy to otherwise unforeseen information, ideas and opportunities.

In the underground punk network, I argue that although there is status to be gained by performing at certain venues or with certain bands, the world is very much one that subsists below the mainstream where national or global fame and status are neither achieved nor sought. Instead bands will seek out positive gig experiences where the gig runs smoothly, the sound of the room is adequate, and the audience reacts well to their music. Different bands will have different strategies for obtaining gigs and different venues that they wish to perform at based on varying information and experiences of both the band as a group and as individuals. As punk is a somewhat niche genre of music in the North West, in comparison with indie music, for example, the pool of musicians is relatively small. This small pool of musicians, we can hypothesise, may result in a high incidence of overlapping band membership ties. It is in these overlapping band membership ties where musicians that are members of multiple bands are likely to have social capital as they are privy to different gigs and have more information about audiences and reactions to certain music as well as what the venue sounds like along with individual experience of performances and facilities and equipment that the venues provide. This information can then be used to make an informed decision about whether to pursue gigs at that same venue with another band. I investigate this kind of closure empirically, assessing whether membership of multiple bands drives ties between bands and venues in Liverpool and Manchester in chapter 5.

A further key theory in social network analysis is that of homophily (McPherson, Smith-Lovin and Cook, 2001), the phenomena whereby ties are more likely to form between similar pairs of nodes in a network and often expressed using the adage “birds of a feather flock together”. Once again using the friendship network example, it has been long established that school

43 friendships are largely homophilous by gender. Homophily, however, is not restricted to demographic attributes, it can also encompass behaviour of individuals such as smoking, activities and hobbies in common or even political persuasion. It is essentially a similarity that we wish to measure in a well-defined way that we hypothesise may be related to ties being present in our network. The dynamic nature of the way in which homophilous ties are formed is explored in research on social selection and influence processes. Social selection is defined as tie formation that happens as the result of attributes or behaviours in common. Social influence asserts that attributes in common develop ex post facto in what we might describe as social contagion, via existing ties.

In analysing two culturally different cities in Liverpool and Manchester I hypothesise that geographical location will have some bearing on the formation of band ties between bands and venues. In addition to this, there may be homophilous relationships in the bipartite band - venue network. It may be the case that bands located in Liverpool prefer to perform in their home city and similarly for Manchester and this may have bearing on other tie configurations in the network. Selection and influence processes are in this case unidirectional. Selection processes can occur when a band chooses to perform at a venue located in the same city as them and influence processes may occur when bands choose to be located in a city close to venues they wish to perform at. Selection processes, it could be argued might occur in the opposite direction where venues choose bands based on their location. This, I argue is problematic and erroneously gives agency to venues. In this thesis I consider venues as the social space where bands perform as a separate entity to any people that are associated with venues or promotion of events at those venues. Social influence processes, on the other hand, are quite obviously impossible in the opposite direction, bands cannot influence a venue to change its geographical location.

In my view, selection is more likely to occur in the punk network than influence pulling bands to relocate to a given city. For a band to relocate to a city it would mean uprooting several individuals and quite possibly entire families from their homes to relocate because of the availability of venues. It may be the case that individuals relocate to cities because there are more opportunities to form bands and bands are more likely to form in cities because of the available pool of musicians as well as venues to perform at but this is not what is measured here. Instead bands may choose to perform at more venues in the city they are located in because of the ease of travel, familiarity with venues, ease of communication, ability to attract a local audience among a myriad of other factors. This dependency between geographical locations of bands and venues and the ties between them is empirically tested in chapter six.

44

3.4.2 Networks of Music Worlds In the extant literature on music networks, there has been a focus on unipartite networks with many studies focusing on the ties between individuals including musicians involved with popular styles of music and classical composers (Crossley, 2008, 2009, 2015a, 2015b; Hammou, 2015; McAndrew and Everett, 2015; O’Shea, 2015). One aspect of network analysis that is crucial when we want to analyse social structure is to specify and explain the types of ties we are considering between actors in the network. Well-defined ties are essential to draw meaningful conclusions about our data. One approach to tie definition, perhaps in order to simplify the analysis of these networks is to look at ‘relationships’ between individual actors. O’Shea (2015) surveyed women in a network of organisers and in this case the participants in the network specified their own relationships with the other actors from a roster list. Crossley’s (2008, 2009, 2015a, 2015b) research used a ‘strong ties’ method where key players in the music worlds of punk and post punk were identified as connected based on whether they had played in a band together; had another professional relationship; lived together; or were known to have had a strong friendship. The argument for this method was that these ties were easily identifiable in the archival material and therefore more likely to be robust and also the fact that strong ties are more likely to have influence on the actors conduct. The issue with this is that this network is in fact multiplex and conflates affiliations with dyadic ties. The ties where professional relationships exist represent an exchange of services whilst the ‘played in a band together’ ties represent membership of different bands and therefore these ties are a projection of a bipartite structure where co-affiliation is used as grounds that two actors are connected (for a summary of this method of analysis of affiliation networks see Borgatti & Halgin, 2011). This projection simplifies the social structure of the network and omits some key information. For instance, there may be a structure whereby two musicians played in the same band and also there was a longstanding friendship between them, or the alternative case where there are longstanding friendships that did not result in formation of bands and this might be interesting to explore but is limited by the projection method. Unipartite projections result in loss of important information and it is now well established that even random people-event networks will result in one-mode projections that have above random structures (Agneessens and Roose, 2008; Wang, et al., 2009; Wang2013; Wang et al., 2013).

3.4.3 Dual Projection As an alternative to focusing attention on one node-type of interest, Everett and Borgatti (2013) suggest a dual projection approach for the analysis of bipartite networks. They posit that for the Davis et al. (1941) data on Southern women, we can project both a women by women network with ties between women who attended the same events, and an event by 45 event network with ties between events that were coattended by at least two women. Everett and Borgatti argue that we are able to return to the original network if we wish, although it will only ever be an equivalent network so there is the potential that the ‘wrong’ woman would be connected to an event but that this is insignificant when analysing network structure as automorphically equivalent women or events would have the same structural properties and therefore any conclusions about the network would not be altered. The dual projection approach would be limited in this way, however, if we wish to include attributes of the women or events that may have an effect on network structure, such as social class of the women and categories of events. Everett and Borgatti (ibid.), however, do not propose the method as a catch-all solution, merely an additional tool with which to analyse networks with two distinct node sets. Further to limitations of attribute inclusion, using dual projection in a tripartite context of musicians, bands and venues, would entail creating four different projected networks, and this is further confused by the fact that there would be two band by band networks, one tied by musicians in common and another tied by venues in common.

Attributes are likely to be of interest in tripartite analysis of a music world. Composition of a band is important in the formation of new band ties, for instance, an additional guitarist could easily join a band that currently has one guitar player, but there is usually only space for one drummer. Even so, bands usually prefer to stick with one compositional structure, unless they adapt and change the types of music they play, so it is important to know that space opens up when one band member leaves. Similarly, the locations of bands and venues may be of interest when analysing structural patterns of ties, as well as the interaction between these two variables. For instance, it may be that bands located in Manchester may tend to favour performing in their home city over performing in Liverpool and vice versa with bands from Liverpool. It may also be the case that bands ‘piggyback’ onto gigs with other bands from their city and stick together out of town. These questions cannot be addressed adequately using projected networks, whereas with direct forms of analysis we are able to simultaneously assess structural patterns and attributes. In addition to the loss of covariate information, critically, projections of a tripartite network would prevent the analysis of tripartite network closure. In a music network we can investigate, for example, if musicians are members of multiple bands - a common aspect of small music worlds due to limited resources in terms of personnel and shared conventions - (see chapter 2), then it may be more likely that those two bands play at the same venue. Alternatively, if two bands have played at the same venue then it may be more likely that they share a band member. These phenomena cannot be tested by means of projected networks and therefore they are inappropriate for my analyses. Consequently, below I move to explain bipartite network analysis, that allows for networks with two different types of nodes. 46

3.5 Bipartite Network Analysis The alternative to projection of the underground punk music data is to analyse these networks in their two bipartite forms, one network of musicians as members of bands and a second network of bands performing at venues using what Borgatti and Halgin (2011) call the ‘direct’ method of analysis. Having noted above that unipartite network analysis would involve transformation of data and loss of information, this section is dedicated to explaining what I mean by bipartite network analysis, the progress that has been made in academic research in this area and how we can move forward with analysing tripartite structures.

Bipartite networks consist of two different types of nodes with ties exclusively between the different types. As noted by Borgatti and Everett (1997) data may be collected and represented in a bipartite form and network analysis methods applied to it, such as opinions that two people hold in common, but this not relational data per se, and it is not the goal to collect data in a format that is appropriate for network analytic methods but rather that the connections are meaningful. What we are interested in parsing out in our analyses as social network analysts are sociological mechanisms behind the formation of meaningful relational ties, why and how they are formed, what patterns of connections are generated and by whom. In a bipartite context this contributes to our understanding of the duality of persons and groups (Breiger, 1974). Relational ties that form bipartite networks are common in society and occur when, for example, directors sit on multiple boards (Mizruchi, 1996), academics co-publish research papers (Small, 1973), criminals operate in tandem (Frank and Carrington, 2007), political groups attend protests (Diani, 1999; 2015) or individuals attend social events together (Davis et. Al, 1941); connections that are often referred to in the literature as affiliations.

More recently direct bipartite network methods have been used to analyse interlocking directorates where directors of firms sit on multiple boards (Koskinen and Edling, 2012; Sankar, Asokan and Kumar, 2015) and organisational problem solving in software development (Conaldi and Lomi, 2013). In recent academic work on networks and music bipartite methods have been used when analysing music networks of folk music artists and events (Hield and Crossley, 2015) and Breiger’s (1974) concept of duality is briefly mentioned in research on Jazz Worlds (McAndrew, Widdop and Stevenson, 2015).

Breiger (1974) posits that there is duality in persons and groups, noting the importance of considering both node sets, women and events, in analyses of the classic Southern women dataset (Davis et al, 1941), showing that projection can be used to attain two separate adjacency matrices that form two networks. In the events network, event nodes are tied to each other through common attendees whilst in the network of Southern women, the nodes are tied through common attendance at events. 47

Whilst Breiger’s contribution to theoretical concerns around duality and multiple node sets is enlightening, he is constrained by the methods at his disposal for analysis. Fortunately, since Breiger’s seminal paper, there has been significant academic interest in the area of networks with multiple node sets that has resulted in methodological breakthroughs in the analysis of such networks. Borgatti and Everett (1992) specified regular blockmodels to assess roles and positions in bipartite networks and in a separate analysis, revisited the Southern women (Davis et al., 1941) data to demonstrate new approaches for obtaining traditionally unipartite measures of centrality, centralization and subgroups for bipartite networks without the need for transformation (Borgatti and Everett, 1997).

Thankfully the tools to analyse these types of networks have greatly improved such that we are now able to analyse this data in its original structure. Following considerable research into both methodologies and measures, ‘two-mode’ deterministic analysis options are available in the popular UCINET8 software program; statistical analysis of ‘bipartite’ networks using Stochastic Actor Oriented Models (SAOMs) in R9 package RSiena10 have been developed by Koskinen and Edling (2012) and bipartite Exponential Random Graph Models (ERGMs) are implemented in MPNET11 software, specified by Wang et al. (2013).

Despite methodological developments, Breiger’s theoretical concept of duality, however empirically constrained at the time, remains an important tenet of bipartite analysis and is of particular importance to our understanding of the way in which music worlds are created and sustained. The bands in the network, while comprised of individuals, need to also be considered as separate entities. As McAndrew, Widdop and Stevenson (2015, in Crossley, McAndrew and Widdop, 2015, p.227) put it:

“…a band or ensemble partly defines the musicians who belong to it, while each musician is individually defined by their network of band memberships”.

The musicians have individual attributes and identities and these are important factors in their choices to be members of certain bands. An easy argument to make then would be that the bands in the network are simply aggregations of musicians in a group but this is an over- simplification of the structure of the network. Bands have their own identities, songs and lyrics and music that is performed and produced by the band as a singular entity and once these songs are recorded they exist in perpetuity and are forever associated with the band. In addition to this, one property that is particular to bands is that membership can be fluid, some people move in and out of bands and that does not necessarily mean the band ceases to exist, bands carry on performing the same songs and carry with them the same identity as before, admittedly usually with one or more core members still intact. It is apparent then that this concept of duality fits the network of bands and musicians. 48 e3.5.1 Bipartite vs Multilevel Further developments in the analysis of complex network structures have led to new statistical methods for the analysis of what are termed ‘multilevel networks’. These networks should not be confused with multilevel statistical models (Snijders and Bosker, 1999) that are used more generally in quantitative social research to take account of variance within nested group structures, or their version that have been adapted for use with network data (Snjiders, Spreen and Swaagstra, 1995; Snijders and Kenny, 1999; Agneesens and Koskinen, 2016; Tranmer and Lazega, 2016). The use of the word ‘multilevel’ here refers explicitly to the structure of the network and not the method applied to analysis of the data.12

Fig.1 A representative multilevel network comprising two sets of nodes, A and B. It is useful to think of this network as combining three smaller networks including the micro and macro levels A and B, where nodes of the same type are connected to one another, as well as at the meso-level X that forms a bipartite network between the two node sets (Wang et al., 2013).

Multilevel networks13 have been of major interest in recent social networks research with development in both theory and methods for analysis, with a multitude empirical papers on subjects ranging from global fisheries governance (Holloway and Koskinen, 2016) to organisational learning (Zappa and Robins, 2016) and more recently international production and trade networks in high tech industries (Smith, Gorgoni and Cronin, 2019) as well as applying established network concepts such as social selection to multilevel structures (Wang et al., 2016)

The principle of multilevel networks is that networks exist with multiple node types, most commonly individuals and organisations and ties exist between these sets of nodes. In addition to this, a central tenet of a multilevel network is that ties between nodes of the same type are also permitted and perhaps of great interest. In my underground punk network, however, I wish to focus on musicians, bands and venues and the ties between different types of nodes. It would be very easy to apply well established methods for the analysis of multilevel networks

49 with two types of nodes to my data, but again this would mean projection of ties by collapsing either musician-band ties that result in a tie between bands based on common musicians; or collapsing band-venue ties, resulting in ties between bands based on common venues. Again this prevents me from investigating tripartite closure and results in the loss of important data and so would be inappropriate for analysis of the punk live performance network. Instead I propose that tripartite structures are the most appropriate and best representation of the network and outline how I conceptualise them below.

3.6 Tripartite Networks In order to arrive at my own conception of tripartite analysis, below I assess the limited academic literature related to these types of networks. I define tripartite networks as having three different types of nodes, with ties present only between nodes of different types. In this thesis I aim to analyse networks of musicians, bands and venues, three node types of critical importance in an underground music world.

The duality inherent in networks of individuals and organisations is recognised by Breiger (1974) in his seminal article on The Duality of Persons and Groups. The need to take multiple node types into account is recognised by Diani (1999, 2015) in his work on social movements and civil society (see also Diani & Kousis, 2014). A key proponent of applying and extending Breiger’s original idea to cultural networks of individuals, organisation and events, Diani stresses the importance of the dual nature of both individuals that are members of multiple organisations; often acting as bridges (see Granovetter, 1973; Burt, 1992) between these organisations that aid communication and cooperation; and organisations that attend multiple events, leading to interactions between organisations that result in cooperation and competition; stressing the importance of this duality and the role that it plays in the mobilisation of social movements. I agree strongly that we need to take this duality into consideration in analyses of networks that involve these three types of nodes - individuals, organisations and events. Whilst Diani explains his ideas about duality in principle, they are not fully realised in the methods he applies to his empirical analysis. Diani relies on projected networks that are analysed separately as; individuals connected when they are members of the same organisation; organisations connected if they have multiple members; organisations connected where they attend the same event; and finally, events connected where multiple organisations have attended them. There is cognitive dissonance between the theory of duality and the method of analysis that Diani uses. By projecting these ties and analysing networks as separate entities, information is lost and the full nature of the duality of individuals and organisations, as well as organisations and events is neglected. It is only by considering these sets of nodes simultaneously that we can account for this duality fully.

50

To give an example, let us consider the case where one person is a member of multiple organisations, if we collapse these ties into simply connecting organisations, we neglect the fact that this one person is a member of multiple organisations and the opportunity to ask why specific people are members of specific organisations as well as the overall structure of membership patterns, we lose the multiple membership ties and reduce multiple memberships to simply a value based on the number of members that an organisation shares. The same is true of events that organisations attend, we lose information on specific organisations and the specific events that they attended. If we extend this loss of information further to individuals, organisations and events in a tripartite network of individuals, organisations and events then we restrict ourselves from asking structural questions in a tripartite context. We lose the ability to assess which organisations are interacting at which particular events, the interaction loses its specific social context, the sites of interaction that are important foci in the network. As well as tripartite analysis of roles, positions and bridging ties, developing tripartite methods would also allow the investigation of tripartite closure effects (specified in detail in chapter 5), where we might want to analyse the effect of multiple organisational membership ties on event attendance for those organisations, or the roles of multiple organisations attending common events that have an effect on membership ties.

First efforts to formalise tripartite analysis were made by Fararo and Doreian (1984), extending the previous work to formalise bipartite matrices and graph structures by both Breiger (1974) and Wilson (1982). Further to this Borgatti and Everett (1992) specified three- mode matrices of criminals, crimes and victims, whilst Mische, Pattison and Robins (1999) theorised on the conceptualisation of ‘k-partite’ network structures. In addition to these methodological papers, Bohman (2012) attempts empirical analysis of tripartite data of ties between directors, boards and owners. He discusses one, two, and three-mode approaches to the analysis of a network of interlocking directorates in Swedish firms, arguing that every time a network is projected and collapsed into a simpler form, we lose data and perhaps that lost data will reveal new information that we should really be taking into account in conclusions that we draw about the roles of actors as well as the structural patterns in the network that might be of interest. Bohman fixes the level of owners and simply assigns a Bernoulli-type probability between two possible ties of directors to boards where there are two owners; three firms and three directors and compares random simulations of these ties to the ties observed in the real network. Whilst I agree with Bohman’s sentiment that the third level may yield new information, as well as his interest in the four-cycle patterns and tripartite dependencies, his method although novel, ignores the duality that plays out at the different levels and the hierarchy of dependence between ties (Wang, et al., 2013) that social networks should take into account and explore. 51

3.6.1 Bringing the Venues Back in I have highlighted above the attempts made by both Diani (1999, 2015) and Bohman (2012) to conceptualise and analyse networks with three types of nodes and asserted that whilst I agree with the sentiment for investigation of these networks, the methods applied to their analyses are inadequate as they are unable to take into account the nature of duality that plays out in these structures. Below I elaborate upon the importance of venues in an underground punk music world and the need for tripartite methods to take account of these sites of social interaction.

Whilst the functions of collective action to achieve an output might be similar to social movements, my point of departure for analysis of the underground punk network differs from Diani’s. Rather than focusing on events attended by organisations, or in this case, bands, my focus is on the social sites of interaction in my network, the venues at which the bands perform. Music differs from social movement networks in that the outputs of a band, the performances take place within defined spaces that have cultural meaning, rather than one-off events that could take place in a number of different locations. The social space of a venue is inherently important. For most of the bands in the underground punk music world the purpose of their existence is to perform live gigs, it is where the enjoyment comes from for the band and where the audiences congregate to see the bands in action as well as where social interaction takes place between bands and their audiences. Musicians form bands to play gigs, the recording of that music is often secondary. It is an opportunity for bands to gather new fans and sell their music and other merchandise to those fans. Bands that perform at these types of venues do not have big marketing or public relations budgets, so this is the way they expose themselves to wider audiences and how they grow their fan base, it is very organic in this way because there are very few opportunities for radio play and other more traditional means of exposure. It follows then, that because activity in the underground punk network is centred around live performances, that venues should be a central component in any analysis of the world. This links to the idea of foci (Feld, 1981, 1982) and that a music world is often focused around certain social spaces where interaction occurs. Crossley (2015a) notes this in his comparison of post-punk music worlds in three English Cities that were concentrated around particular venues; Electric Circus and The Factory in Manchester; Eric’s in Liverpool and The Limit in Sheffield. Further to this, the venues are also important because they are the sites where the bands live as separate articles to the individuals that comprise them.

They create and project their identities through the music that they perform at these venues and the audience perceives the band as one entity and consumes the music as a whole, they sing along with the band and praise or ridicule it as they see fit. In this way, venues are the

52 place where the duality of bands and musicians occurs in practice and the lines between individual musicians and the bands they belong to become somewhat blurred. Including the venues at which the bands play in the network means we now have three different types of nodes; the musicians, the bands they play in and the venues at which they perform; where ties only exist between different node types; which gives us a ‘tripartite’ network. Musicians are members of multiple bands and bands play at multiple venues and these ties to specific bands or venues can only be accounted for with tripartite analysis without reducing the multiple memberships to a tie between two musicians, bands or venues. The tripartite structure is inherently interesting, there are possible dependencies at play between musician-band ties and band venue ties, interactions occur within venues and new ties are formed in these interactions that foster the continuity of the underground punk world.

3.7 Discussion Networks are complex and can take many different forms. As researchers in the discipline of social network analysis we need to ensure that we match appropriate research methods to the research questions we ask. It is my contention in this thesis that development of tripartite thinking in a ‘music worlds’ context, as well as development of tripartite methods, is needed to properly answer the research questions that I pose. Previously, developments towards analysing networks with nodes of more than one type have been mostly limited to bipartite networks, which have been either bipartite affiliation type networks or more recently, multilevel networks where ties between nodes of the same type are also accounted for. I have argued that there is scope for the development of methods to analyse three-mode, tripartite networks, especially in the case of music worlds consisting of musicians, bands and venues, where inclusion of all three types of nodes is integral to any reasonable attempt to explain the activity in an underground music world setting, one that is based primarily around live performance. A tripartite approach allows use to simultaneously take account of the resources, conventions and collective activity involved as well as the critical importance of venues as foci for interactions and sites of duality where musicians and bands are both perceived and project themselves as dual entities and conventions dynamically evolve. I move forward in the following chapter to discuss methodological concerns, the collection of my dataset and how it was processed for analysis. I give a broad overview of the data and outline basic structure of the networks that I analyse in chapters five and six.

53

Chapter Four: Methodology 4.1 Introduction In the previous two chapters I have conducted two thorough literature reviews. Chapter two assessed a rich body of sociological literature, leading me to arrive at my own conception of music worlds and advocate for the importance of venues in an analysis of the live performance activity within a music world. Chapter three assessed the theory of networks, arguing that to answer research questions concerning three different types of nodes we should aim to use tripartite methods to avoid loss of important information. In both chapters I noted Breiger’s (1974) concept of the duality of persons and groups and the central role it plays in helping to conceptualise the network of the underground punk music world in Manchester and Liverpool as well as the foundation it provides for formal graph theoretical conceptions and statistical models of networks with multiple, distinct node sets. Breiger’s theory also contributes to our understanding of the role which venues play; providing the social space where bands’ dual identities, as both collections of individual musicians and cohesive organisations, are played out in interactions between each other and the audience both in the way they project themselves and the way they are perceived.

In this chapter I set out to give an overview of the methodological choices I have made in order to analyse this world. I begin by setting out how I propose to operationalise the concepts of music worlds and duality by investigating the network of live performance in Manchester and Liverpool in a tripartite context. I then move on to outline the research aims of the thesis and what I hope to achieve through the analysis of this network, followed by four key research questions. Next, I elaborate upon the data that was collected, how the dataset was constructed for analysis, as well as ethical considerations and limitations of the dataset. I conclude this chapter with a brief description of the dataset as a precursor to chapters five and six, which are dedicated to the empirical analysis of this data using statistical models that have been developed specifically for social networks.

4.2 Operationalising Music Worlds Becker (1982) posits that art is the product of collective activity, not isolated genius, and the concept of an art world and by extension, music world, encapsulates all those involved in this collective production, all of the activities that are encapsulated by Small’s (1998) term ‘musicking’. As I mentioned in the introductory chapter of this thesis, if we think about the collective production of music, there are perhaps three main creative outputs, compositions - the songs that are written, recorded music and live performances. By compositions, I mean the writing and arranging of original songs, the music and lyrics. In the context of underground punk, as well as most forms of popular music, this most likely does not involve 54 formally scoring the music to produce sheet music as is the case in the world of classical music, rather songs are written with the intention to be recorded or performed. The activity of composition, more commonly referred to as song writing, is perhaps the only output of these three that could possibly be considered as an individual pursuit. Although often done in collaboration with other musicians, it is possible to compose a song in isolation and this composition will exist irrespective of whether or not it is ever performed or recorded, or even learned by a group of musicians. Although composition is an important creative output, in the underground punk world, it is only important insofar as it allows musicians and bands to record those songs and perform them live, both of which are activities that rely heavily on collective activity, conventions and resources. Of these two outputs, it is my contention that live performance is not only the most important of the two but that in the minds of most musicians and bands actively involved in the underground punk world, it is principally the end goal and purpose of their involvement in music-making.

Live performances rely not only on collective activity, conventions and resources, but on venues and it is these venues that become foci for the underground punk music world, spaces that foster creativity and where conventions are established and formed in dynamic interplay between bands and their audiences. The role these venues play in a local music world differs greatly from the context of the major venues that perhaps most gig-goers will be familiar with. In the corporate, big-money, major music industry world, once a band is established and operates full-time, the general pattern of production of new music will start with the writing of new songs by one or more people in the band, perhaps with help from other established song writers or ‘hit-makers’ outside of the band; those songs are then critically appraised and the best ones selected for single releases, followed by an album which is recorded and released, followed by a series of tour dates in support of that album. The list of songs, or ‘setlist’ that the band then performs at these gigs will include new material from the recently released album as well as back-catalogue songs that the fans of the band will be familiar with. Venues in this world then serve to interact with audiences and play the songs that they expect to hear, it is entertainment that generates income for the band and various other stakeholders in a profit-making business. These venues are most likely booked as part of a larger tour of major cities and selected for capacity and the ability of the band to sell the amount of tickets required to fill the venue.

From experience as a participant observer in the underground punk world, it is evident that venues play an entirely different and more significant role at this lower level. A small-scale, local world of this nature operates somewhat differently to the higher level, major venue world of musicians playing punk music or other forms of popular music as their full-time

55 occupation. The musicians and bands that operate in the underground punk world are often not full-time musicians and likely at best will break-even from their musical endeavours. It is certainly not intended as a profit-making activity, rather it is a creative outlet. Some participants may operate a near to full-time touring schedule but often musicians pick up other jobs in between these tours. The bands in this world perform mostly at small venues of capacities fewer than 100 people; barring any major support slots they manage to get at major gigs; often to audiences who may not have seen them perform live before and may only have listened to recordings of the band’s music in the weeks or days leading up to the gig, if at all when they perform at gigs that are out of town. Some bands are fortunate enough to have a small following that constitute a loyal audience when they perform either in their hometown or in other cities. Audiences that attend the gigs are often in bands themselves or regularly attend local punk shows. Audiences build around certain venues, that simultaneously become the places bands want to play and the places audiences want to go to watch bands, reinforced by positive experiences of performing and watching live performances respectively. There are a large number of bands involved in the world and as such there are often bands performing that are unknown to the audience. These bands have to win over the new audiences that may be a mixture of people open to hearing new music or those that want to hear the types of music they have come to expect to hear at these venues. In this way the established conventions around style are inherently tied up with venues and they are reshaped in these spaces by new bands that perform there.

Live performance is critically important in shaping the sound of a band. It is in these live performances that bands gauge the reaction of their audience; they decide which songs ‘work’ and which do not. Venues at this underground level also offer a safe space for bands to try out new material and develop ideas about what works well outside of the rehearsal room, a critical appraisal process that informs choices about the music which they play. This dynamic interplay between bands and audiences that shape conventions in venues might be different for bands depending upon their location. For instance, bands that are from Liverpool and play at a hometown gig are likely to have greater knowledge of the audience and the venues than a band that travels to perform and play in Liverpool from out of town and similarly for bands and venues located in Manchester. It might be the case that hometown bands are willing to try different venues that are unknown in the hope that they can attract a local audience, in order to break conventions that are expected at a certain venue, whereas travelling bands may be likely to be more risk averse as it requires more effort to travel to and from the gig and they will want to make sure the journey is worthwhile. In order to assert that conventions play a key role in the choices of bands to perform at venues, we must also consider the role of musicians as members of the bands in the network. 56

The decision to perform at a given venue is likely to be a collective process with all band members needing to agree to a gig. Band membership, however, is not mutually exclusive and some musicians in the underground punk world perform in multiple bands. If experience informs venue choice, then it is logical to assume that if a musician has a positive experience at a venue then they will want the other bands that they are members of to perform at that venue. If band membership does not drive venue choice, then it may be that conventions are the driver of those choices. Band overlaps as drivers for venue choices are analysed in chapter five, whilst conventions and locations are investigated in chapter six. As I have argued throughout the previous chapters, the underground punk world in Manchester and Liverpool persists and subsists in a network of live performance. This network is comprised namely of three sets of key nodes in the network, musicians that are members of certain bands and bands that perform at certain venues located in each of the two cities. It is important to note that the network I capture in this thesis is not the genesis of a music world, a creative explosion of culture, an innovation of brand new music or the documenting of the formation of ties that create the ‘critical mass’ of participants needed to move towards the birth of a newly defined music world (Crossley, 2008, 2009, 2015a, 2015b). What I aim to capture is a snapshot in time, a period of three years for which the underground punk world was active and to show how this world is sustained through live performances at key venues.

4.3 Research Aims There are two main aims of this thesis. The first is to uncover the structure of the underground punk music world in two major cities in Northern England, Manchester and Liverpool, focusing on the structure of the activity in the network in terms of live performances at small venues, both in patterns that emerge around certain venues and also how the multiple membership of musicians in different bands might contribute to these patterns. Through this investigation into these networks two significant contributions to the current social network analysis literature will be made. The first will be a substantive contribution to the understanding of music worlds and the literature that surrounds this area of network analysis, highlighting the critical importance of venues in sustaining a world and the dynamic nature of conventions that plays out in these social spaces. As well as helping to aid the understanding of a specific music world of underground punk music in a specific geographical area, the North West of England, it is hoped that the conclusions reached in this thesis can be extended to other music worlds in other parts of the world, testing for both similarities and differences in the role of venues and the way in which the worlds operate. As well as adding to current understanding on music networks, I seek to make a contribution to knowledge for wider cultural networks research and to introduce these areas of investigation

57 to the use of statistical modelling as a viable approach to explain network phenomena in cultural contexts.

The second major research aim is to begin to develop a statistical methodology for testing the prevalence of tripartite network phenomena more widely, to enable investigation into whether or not tripartite closure patterns in networks with three different types of nodes are of importance in the network, an area where academic literature is severely lacking.

4.4 Research Questions This thesis addresses four key research questions:

1. Is there a tendency towards tripartite closure in the network of live performances in Manchester and Liverpool?

2. Do overlapping band-membership ties drive the formation of band-venue ties?

3. What role do venues play in the underground punk world of Manchester and Liverpool

4. How does the role of venues play out over a three-year time period from 2013-2015?

The empirical analysis that seeks to address these questions is separated into two subsequent chapters. Chapter 5 seeks to address the first two questions, whilst chapter 6 seeks to address the latter two. Prior to empirical analysis, I now seek to outline the data collection and dataset construction process below.

4.5 Data Collection In order to answer the research questions posed above I set out to capture the network of live performance activity in Manchester and Liverpool from 2013-2015 and construct the network for analysis. As the population of interest is unique, data was collected specifically for the purposes of this thesis. Network analysis differs from standard quantitative social research in that it does not seek to necessarily infer from a representative sample to a population of interest, nor does it aim to show how largely uncorrelated variables are related to some outcome variable of interest. Rather we are often interested in small populations of interest and consider a network as an outcome of specified social interactions, tie variables in the network that are dependent upon one another. Capturing the network of musicians, bands and venues presented a multitude of considerations and key decisions that I now discuss in turn.

The first concern that I address, which is a primary concern in the majority of network research, is boundary specification. There is a need to define exactly what the population is

58 that we are trying to capture. This process differs from standard statistical analysis where we aim to collect a representative sample of a population that we wish to infer to. The issue in social networks research is that if we take the view that our variables of interest, the network ties, are dependent on one another then there are perhaps ties outside our focused analysis that affect the ties in our network. Sometimes network boundaries may seem, on the surface, extremely clear. In the classic cases of friendships within a school classroom, or communication within an organization we might claim to have captured these entire populations of interest. Even in the apparently more concrete example of friendships within a classroom, it could be argued that we really should look at friendships in the whole school because students have peers in different classes and even different year groups and friendship ties in one classroom may be dependent on other friendship ties within and between other classes in the same school as well as constrained by them. Music worlds present a more complex challenge due to their sheer scale; indeed one could argue that there exists a global music world that is entirely interconnected; as well as the theoretical assumption, discussed in chapter 2, that a high number of complex interactions have to take place in order for music worlds to be created and to subsist, including a myriad of personnel and organizations.

I contend that these issues with boundary specification can be mitigated by ensuring two key details are given care and attention; a focus on the research questions at hand to generate reasonable hypotheses about who or what our network should include and a focus on clarity in any conclusions we draw about the network and who or what specific node sets we are drawing these conclusions about. For instance, the substantive conclusions in this thesis are hoped to provide a basis for further investigation into underground music worlds in the UK as well as internationally, but we cannot infer that we have captured the mechanisms for other localities or other periods of time without further investigation. I elaborate upon the boundaries of the underground punk network under investigation below.

First and foremost, the boundaries of the underground punk network in Manchester and Liverpool were derived from the research questions above in section 4.4. Temporal boundaries were set as the period from 2013-2015 inclusive so that the population of gigs to be considered took place from 1st January 2013 up to and including 31st December 2015. The data collection phase of this research began in early 2016 and as a result, the end of 2015 seemed to be a reasonable and clean cut off for the end of the collection period. This also ensured the data was as recent as possible. Working backwards from this endpoint, the start of the assigned period, 2013, was appropriate to ensure three complete years of data collection that would provide a sound basis for longitudinal statistical models with three observation periods in discrete years.

59

The second network boundary set by the research questions were the geographical boundaries. Manchester and Liverpool were selected as worlds to explore. Principally, this was based on my own interest in the punk worlds of these cities both a participant musician and active audience member at many gigs in both cities. Drawing on my own knowledge I also knew that active worlds existed in both cities and felt that there would be a significant level of activity for investigation and that there may be nuances and differences between the two cities that would be interesting to explore. As I drew attention to in the introduction, these two cities are also steeped in musical history and this underground punk world is a significant part of musical culture in Manchester and Liverpool that is currently being created. It was decided that geographical boundaries were strict, only venues that were within the cities were included in the networks; no venues in the larger metropolitan areas of Greater Manchester or Merseyside were included. This gives a clear boundary to the network in order to aid clarity in interpretation and comparison and in conclusions that can be drawn about the networks. One important aspect of the geographical boundary specification to note here is that geographical location is only limited to the venues. The aim of this research is to capture the music worlds in Manchester and Liverpool and therefore bands that travel from outside of these two cities to perform at venues within them are included in the network alongside bands that were not.

Following the time period and the geographical location, the next stage of the research design was to define the relational ties that I wished to capture. Often relational ties can be intuitive, for instance, friendship ties, often collected through primary face-to-face interviews, or can be constructed from theory. For instance, Crossley (2008, 2009, 2015a, 205b) defines a ‘strong- tie’ methodology, where evidence for friendship, collaboration or a romantic relationship is used as a tie that is considered relevant to forming a critical mass for the birth of punk and post punk. Affiliation ties are relatively simple to collect. Directors on boards, for instance, are concrete affiliations, as are co-publication of papers, they are factual in nature. The nuances of tie definition come into play however, when we consider the mechanisms behind the tie formations. What are the social processes behind the ties? Academics publish papers together through similar research interests, knowledge exchange, research groups, jointly awarded funding, PhD students and their supervisors and a multitude of other factors. Similarly, directors sit on multiple boards through a variety of mechanisms, they may be recruited through other directors, through boards or in tripartite structures via owners (Bohman, 2012). As has been mentioned previously my focus is on live performance activity involving three node sets – musicians, bands and venues.

Here I define two different types of tie between these node sets that constitute the tripartite structure of the underground punk network. The first is a tie from band to venue where a

60 band performs at a given venue. The key assumption here is that the band has agency in that choice, they make the decision to perform at a given venue. This is temporally constrained by the year that performance took place. The second type of tie is a band membership tie where musicians are members of a given band. It is assumed that musicians are performing members of the band and are not restricted to appearing on recorded music or contributing to song writing activities. In a tripartite structure there are no ties within node sets, that is, ties between musicians and other musicians, bands and other bands and venues and other venues. Also, in this case, there are no ties between musicians and venues. Now then, after defining boundaries of the punk network and establishing clear tie definitions, the data source will be discussed.

The first concern when trying to capture the network ties was what the most appropriate source of data would be. In general, there is no large scale social-survey data source for social networks, although some ego-network measures have been translated from large scale social surveys. The nature of networks means that concrete ties have to be recorded that are based upon some conception of affiliation or interaction and therefore exist only in the particular network which we are interested in and as a result, network data is usually gathered by the researcher(s) in a bespoke fashion. Network data can be collected face-to-face via interviews, from participant observation via anthropological studies (Mitchell, 1969), or inferred from secondary sources and archival data (Crossley, 2008, 2009, 2015a, 2015b).

The most straightforward method of data collection on live performance networks is via the events themselves and the sources of events data for live performances ostensibly originate from marketing materials for those events. Traditionally this may have included newspapers, magazines, flyers and posters advertising the gig. Flyers have been used previously to construct networks of live performance (Watson, 2015) to analyse co-performance in the hardcore punk world in the USA from 1979-1986. Physical flyers are still produced for some local punk events but not for every event and in addition there is no easy way to collate these flyers and no obvious archival source containing them. This is, in part, due to the fact that much of the promotion of gigs has moved to channels. Myspace was used ostensibly at first until first people, and soon after, businesses, transitioned to Facebook as a space for promotion, marketing and communication with stakeholders. The start of the data collection period, 2013, coincides with the more prominent use of Facebook events pages, providing a rich source of events data. Gigs in the underground punk world are primarily communicated online via events pages on Facebook that fans can interactively engage with and add to their own calendar of events they are attending.

61

These events pages are an ideal source of network affiliation information as they contain the date of the gig, the venue at which the event took place and the names of all the bands that are due to perform. The event pages are also archived by default, so it is possible to view past events and the details after they have taken place. Events pages are archived in chronological order within other Facebook pages belonging to venues, bands and promoters. I used my own knowledge as a participant in the music world to begin systematically searching through pages of promoters, bands and venues that I knew to be involved in the punk world around the time in question for events that they were a part of, and that took place in Manchester and Liverpool from 2013-2015. The time periods were strictly the calendar years from the 1st January to 31st December each year and each period contained 365 days (leap years took place in 2012 and 2016, either side of the period in question). These pages elicited event pages from which data could be recorded, but also other band, venue and promoters’ pages that I was able to snowball from to capture more events. Events are archived by default on Facebook for venues, providing the ability to work back through previous events systematically for venues, promoters and bands that met criteria for inclusion after initial searches.

I collected the events data from the pages manually, recording events in tabular format with one row allocated per gig, recording the date, the venue, any promoters involved and then the bands that performed. Bands were typically listed in the free text ‘details’ field at the bottom of the event page and were usually ‘tagged’, meaning the text on the page that said the name of the band was linked to a profile page for that band. There was also typically a digital poster attached to the page that listed the bands that performed. The discussion section at the bottom of the page was also checked to ensure that the bands were correct as these sections would sometimes mention last minute replacement bands or additional bands explicitly, but often stage times were posted on the day of the gig. If these stage times listed different bands to the ones advertised on the digital poster or in the details section, I deduced that the bands listed on the stage times, if posted on the day of the gig, were the bands that actually performed at that gig.

The process of manually searching for and collecting the data presented some additional network boundary considerations. As was mentioned in chapter one, in order to capture an underground punk network, we must define what punk is and what it is not. Defining punk is extremely difficult task that is extremely complex where clarity can seem an insurmountable task. To that end, here I focus only on punk as a musical genre, to maintain relevance to this thesis and not as a fashion or attitude in its own right. Genre can be defined by both music and lyrics and this again creates blurry boundaries, with some bands described as punk for lyrical content and others for musical style or simply a DIY ethos, which is an attitude bound

62 up with the punk world. In a concerted effort to remove myself as arbiter of what is punk and what is not, I outline criteria for inclusion in the network. The first decision made in this matter was to use self-definition by the bands, promoters, venues and events pages to describe themselves as punk on Facebook. Initially I drew on my own experience as an active participant in two different punk bands in North West England to begin searching for venues, promoters and band profile pages that I considered to be part of the underground punk world and therefore were likely provide text on the pages to self-identify as punk. This self- identification might be through the genre section on a band’s profile page or other sections of pages or event descriptions where the word punk was used or a known sub-genre of punk including hardcore, Oi! or Anti-Folk that do not necessarily include the word punk in their genre names.

One issue that arose using this method was that some bands used eccentric names for genres such as the prominent Liverpool band Pete Bentham and the Dinner Ladies that refer to their own genre as ‘kitchencore’. It also came to light during the snowballing process that there were several community ‘groups’ that existed where event pages and posters for gigs were posted that could not be found anywhere else and would also need to be searched through to gather as complete a dataset as possible. The most fruitful sources of event pages proved to be the profile pages of promoters and public ‘groups’, community pages dedicated to punk where events and digital posters were posted regularly. Unlike pages for promoters, bands and venues, where events are archived chronologically on a single page, the community groups were very difficult to search as they were not attached to specific events, instead they simply had a ‘news feed’ comprised of hundreds of posts which had to be scrolled through manually to check for links to event pages.

In order to ensure all possible posts were displayed an option had to be selected to display all posts. These were then manually searched starting with the most recent gigs in the timeframe, from 31st December 2015, back through to 1st January 2013 to make sure of a comprehensive record of all events. The groups also sporadically included posters with no event links attached. In these cases, the information on the poster was recorded and subsequently followed up through manually searching for an event page via the default Facebook search function. Not all events and posters posted to these community groups took place in Manchester and Liverpool and this was similarly the case when searching through bands’ lists of events that they performed at. Promoters tended to mostly put on gigs in either only Manchester or only Liverpool but there were some outside the geographical boundaries of the two cities that were disregarded, even where the promoter was primarily based in one of the cities in question and would usually put on gigs within the city boundaries. Using a multitude

63 of different page types did elicit repeat events where multiple event pages had been set up by each party involved, in some cases three separate events originating from the band, the venue and the promoter. Where this repetition occurred, the information was consolidated. In the majority of cases there tended to be one event page that contained the most detailed information, had the most recent activity on the page in terms of discussion posts, was posted in community groups and had the most confirmed attendees and therefore was taken to be the most accurate and reliable source of information in the case of any conflicting details between repeat event pages.

One limitation that may be levelled at this dataset is that the data is restricted to Facebook events and that this method may exclude events that were not promoted on Facebook. I contend that this is likely to be a small number of gigs and would consider them to be fringe events and not part of this core world, perhaps they capture something different altogether. For instance, some bands may not use Facebook because they are anarchists and do not want to promote their music events through a multinational corporation, some bands may be extremely right wing and take a more covert approach to promotion of their events, but again, this is not what I am trying to capture here. I believe I have managed to capture the core activity of the underground punk music world in Manchester and Liverpool from 2013-2015, musicians, bands and venues that use Facebook as a means to advertise, promote and communicate their gigs and contend that this serves as the best available record of live performance activity in these two cities over the time period in question.

4.6 Ethical Considerations All data collected for this thesis was done in compliance within The University of Manchester’s ethics framework. The research was deemed to fall under the School of Social Sciences pre-agreed framework, meaning further presentation to the ethics committee was not required. In addition to University approval, I address my own epistemological ethical considerations. Online sources of information are often subject to scrutiny both in terms of the accuracy and reliability of the information available, which I have addressed above, and in terms of the ethical implications of using data from online sources and in particular, social media, which I now address below.

One of the main tenets of ethical social research is informed consent of participants when collecting primary data. As primary data collection is time-consuming and costly, the traditional alternative to this is to use secondary data for which consent has already been granted. In the social statistics discipline this secondary data this is usually in the form of large- scale surveys conducted by external organisations to create data explicitly for research use, for example, Understanding Society (University of Essex, 2018). In social network analysis, 64 however, secondary data often takes the form of either bespoke network data collected that has been collated for the purposes of reproduction of analyses and methodological testing, such as the Southern women data (Davis et. al, 1941), originally collected as primary data, where informed consent was required in the first instance; or the network of Florentine families (Padgett, 1994) built from historical archived documents; among others that are available in the popular UCInet software (Borgatti, Everett and Freeman, 2002). Online data, and in particular, data from social media sources is problematic to position in this traditional framework. Much of the debate around ethical use of social media data is tied to defining what is private and what is public data. Data privacy is a growing societal concern with personal data collected online by companies being used from everything from marketing purposes to show users bespoke advertisements, to more sinister motives such as the Cambridge Analytica scandal, where the company was accused of collecting data with the intention of covertly influencing the USA 2016 presidential elections in favour of Donald Trump (Cadwalladr and Graham-Harrison, 2018). In the wake of the Cambridge Analytica scandal in addition to its controversial research program ‘Atlas’ that paid users to install an app that collected all their private user data (Constine, 2018), Facebook has come under increasing pressure to address data privacy concerns and CEO Mark Zuckerberg dedicated much of his address at the 2019 Facebook Developer Conference to privacy and stating that “…the future is private…” (TIME, 2019). European policy makers have attempted to mitigate their citizens concerns by implementing the new General Data Protection Regulation (GDPR) whereby personal data held by any company can be requested to be deleted by the people whom it concerns (EU, 2016), although how effective this regulation is or how it can be policed remains to be seen.

Varying levels of privacy are offered to users on different social media platforms. Twitter, for instance sets a profile to public by default. If a profile is set to public, limited information contained in a profile can be viewed by anyone who views the page URL, without needing to set up a Twitter account. Once an account has been set up, an additional tier of information can be accessed on public profile pages. Further to this, there is an option to set an account to ‘protected’ where users have to request to ‘follow’ a user and that user then decides whether or not to allow access to their profile data. This is similar to Instagram, where user accounts are by default set to public, but provide the option for a private account that requires authorisation by the user for access to their images. Facebook is slightly more complex with multiple tiers of privatisation available for individual users.

In a conscious and concerted effort not to violate user privacy, the Facebook data that I collect and analyse in this thesis, however, is not at individual level. It is collected from public pages that can be accessed without ever having to create a Facebook account; band pages,

65 events pages, venue pages and promoter pages all share this trait. It is not only a question of access to the data; however, I posit that the bands, promoters and venues all purposefully post event data on Facebook for the purposes of public consumption to advertise their gigs and attract and audience and create profile pages in the first place to be public facing and for people to search and find them; to promote and advertise themselves as a band, venue or promoter and consequently this data is firmly in the public sphere and not the private one and is therefore ethically sound and does not require informed consent in the same way as personal data and should be treated in line with secondary archival material.

4.7 Data Descriptions In this section I provide a brief overview of the resulting dataset that was collected. Following the procedure described above, the final raw dataset contained details of 357 gigs and 725 unique bands that performed at 41 unique venues over the three years. I detail how this dataset in its raw form is unsuitable for analysis of the underground punk world and show how I intend to create two separate datasets that will be subsequently used for empirical analysis in chapters five and six.

Table 1 Cross tabulation showing number of events by year and city Calendar year Location 2013 2014 2015 Total Manchester 63 37 53 153 Liverpool 72 71 59 202 Total 135 108 112 355

Table 2 Counts of bands and venues by geographical location in the total collected dataset Bands Venues Location Count Percentage Count Percentage Manchester 85 11.7% 25 61.0% Liverpool 66 9.1% 16 39.0% Other 574 79.2% - -

Total 725 100% 41 100%

66

As I have mentioned throughout the first three chapters, venues are central to this thesis and have been mostly overlooked in the literature on the networks of music worlds. Before I address their importance in tripartite closure in chapter 5 and their critical role in sustaining the underground punk worlds in Manchester and Liverpool in chapter 6, it is important to provide an overview of the venues themselves. Venues in the underground punk world are generally on a small scale, holding anywhere from fifty to one-hundred people and rarely more than this. The majority of these venues are multipurpose, usually pubs or bars with live music spaces ranging from a dedicated room for live music to a cramped corner that leaves bands fighting for space on a pub carpet and trying not to get in each other’s way.

Table 3 Two-mode degree centrality for venues in 2013

Venue Degree MelloMello 0.351

Kraak Gallery 0.079 Star & Garter 0.212 Sound Food & Drink 0.030 Fuel 0.011 Wahlbar 0.128 Big Western 0.019 Gullivers 0.130

Pilgrim 0.030 Eric’s Live 0.005 Picket 0.011 Tiger Lounge 0.035 Indigo 0.033 Lomax 0.011 Brink 0.008 Drop the Dumbulls 0.109

Bay Horse 0.033 Royal Oak 0.008 Sound Control 0.008 Maguire’s Pizza Bar 0.109 Kazimier 0.033 Retro Bar 0.008 Next to Nowhere 0.008 24 Kitchen Street 0.011

Ruby Lounge 0.008

67

Table 4 Two-mode degree centrality for venues in 2014 Venue Degree

Maguire’s Pizza Bar 0.424 MelloMello 0.173 Indigo 0.018 Wahlbar 0.055 Star & Garter 0.162 Gullivers 0.044 Retro Bar 0.052 Tiger Lounge 0.018

District 0.033 Next to Nowhere 0.033 Bay Horse 0.018 Sound Control 0.044 Joshua Brooks 0.018 Blade Factory 0.022 Sound Food & Drink 0.085 Kazimier 0.022 Odder Bar 0.018 Antwerp Mansion 0.018 Lomax 0.018

Dry Bar Live 0.018 Kraak Gallery 0.018 Fuel 0.018 Ducie 0.018

We can see from the sociograms in appendices 1, 2 and 3 that these networks are densely clustered around relatively few venues, particularly around MelloMello in Liverpool in 2013. We can see how the central venues shift from one year to the next by looking at their degree centrality in tables 3, 4 and 5 below. There is a dramatic drop off in the centrality of MelloMello and it is far less central in 2014 than 2013. This is compounded by it shutting down in late 2014. Surprisingly, although much has been spoken about recently on subject of the transient nature of venues, this is the only venue in the dataset that disappears from existence over the three year time period. The degree centrality tables as well as the sociograms in the appendices show that core venues appear to remain relatively stable. Table 1 shows events by year and city. 2013 and 2015 appear to be similar across Manchester and Liverpool with regard to the raw number of events taking place, but in 2014 there are almost twice as many events in Liverpool than Manchester, with an extreme drop off of events in Manchester.

68

Table 5 Two-mode degree centrality for venues in 2015 Venue Degree Sound Food & Drink 0.182 Maguire’s Pizza Bar 0.159 Ducie Bridge 0.090 Star & Garter 0.136

Thirsty Scholar 0.012 Gullivers 0.040 Sound Control 0.061 District 0.017 Joshua Brooks 0.009 Retro Bar 0.032 Gorilla 0.009 Next to Nowhere 0.026

Antwerp Mansion 0.009 Arts Club 0.006 Nordic Church 0.009 Tiger Lounge 0.009 Castle Hotel 0.009 Eagle Inn 0.009 Klondyke 0.009 Miners Arts & Music 0.006 Dulcimer Bar 0.006

Although these simple descriptive statistics start to give us an idea of the structure of venue choice, such as the most popular venues in Manchester and Liverpool, it is difficult to interpret with any sensibility. We can see from table 2 that the majority of the bands in the network or from outside of Manchester and Liverpool, representing a huge travelling contingent. Many of these bands perform only once in the entire time period in question. There are 725 unique bands in the dataset over three years and there are only 368, 314 and 346 active bands in each given year 2013, 2014 and 2015, and there is significant composition change between time periods.

In order to proceed with empirical analysis, I created two new datasets. For the tripartite analysis I took multiple samples of 16 bands that were active in 2015, eight bands that I knew to have members that performed in multiple bands and eight randomly sampled from bands that performed at least twice in 2015, in order to eliminate transient bands. Due to the large number of transient bands, however, repeating the sampling procedure yielded similar sets of bands each time and as such, I proceeded with the core set of bands that intersected every sample drawn. Additional data was then collected on performing members of those chosen bands from their public Facebook profile pages to connect to the existing data on the venues at which they performed. The resulting tripartite network of 44 musicians, 16 bands and 14 69 venues serves as a testing dataset for a novel tripartite methodology that I propose in chapter five. The second dataset is a longitudinal bipartite dataset with three observation periods 2013, 2014 and 2015, that comprises the eighty bands that were consistently active in all three observation years. The fact that only eighty bands out of 725 unique bands in the collected dataset were active in every year shows just how problematic the total collected dataset is in making any substantive conclusions. This dataset contains all forty-one of the venues in the dataset as I contend that bands make an active choice to perform at certain venues, and as I model the change in network ties over time probabilistically in chapter six, and these venues are available to perform at, they should be included in the model. Having reduced the collected dataset, I am now able to answer research questions three and four using the second dataset in chapter six, and now I move to answer research questions one and two in the following chapter, by testing a novel method for tripartite analysis using the small sample dataset.

70

Chapter Five: Analysis of a Tripartite Network 5.1 Introduction In chapter four I explained how I operationalise the theories on both art worlds (Becker, 1974, 1976, 1982) and the duality of persons and groups (Breiger, 1974) in order formulate tripartite research questions and collect network data on three distinct node sets; musicians, bands and venues with ties between musicians that are members of bands, and bands that perform at certain venues in Manchester and Liverpool. In this chapter, I endeavour to introduce, describe, explain and test a novel method for the investigation of tripartite network structures, and in doing so, answer research questions 1 and 2, uncovering the nature of dependency in tripartite four-cycle closure patterns.

I begin by reinforcing the motivation for tripartite network analysis and summarise key tenets of this analysis, the importance of venues and the duality of persons and groups (Breiger, 1974) and propose three key aims of the analysis. I then give a brief description of the data that will be used in the analysis, followed by a section introducing exponential random graph models (Robins, 2007; Lusher et al., 2013), the class of statistical model I have chosen to demonstrate analysis of a tripartite network and subsequently introduce its bipartite extension (Wang et al., 2009; Wang, 2013; Wang, Pattison and Robins, 2013). As formal methods for tripartite statistical modelling are not yet defined, I separate the tripartite network into two halves, a network of musicians that are members of certain bands and a network of bands that have performed at certain venues, estimating bipartite models and simulating goodness of fit parameters for each network. I subsequently move on to specify the tripartite network formally in an adjacency matrix (Fararo and Doreian, 1984) and explain how I conceptualise tripartite closure, leading to the introduction of statistical network parameters for testing the prevalence of this closure in a small sample network of 44 musicians, 16 bands and 14 venues, a subset of the total collected network.

5.2 Motivation The underground punk music world can be thought of as a performance-focused network of musicians, the bands they are members of, and the venues at which they perform. Conventions, common facets of the music world agreed upon by groups of participants, which facilitate and simultaneously constrain its operation, are dynamic. They can also form and reform in the social spaces where musicians interact with one another, but bands also perform and interact collectively with their audience. Venues, in studies of music worlds are often noted as significant but the dualities of interactions in social space by individual musicians and collective bands have so far been ignored. In this chapter I argue that it is 71 possible to investigate this duality to understand the drivers of conventions and that this can be investigated through tripartite network analysis. There have been few previous attempts to analyse tripartite networks. Notable papers analyse; a network of Swedish firms, owners and directors (Bohman, 2012); a social movement networks of Brazilian organizations, projects and events (Mische and Pattison, 2000) and criminals, crimes and victims (Borgatti and Everett, 1992). As the literature on these networks is sparse, methods of analysis for tripartite networks are underdeveloped and as a result I present a new method for the investigation of tripartite networks using statistical models.

The purpose of the analysis in this chapter is three-fold:

1. To develop a procedure that assesses the viability of tripartite investigation versus two separate bipartite analyses. 2. To develop a generally applicable method for analysis of four-cycle closure in tripartite networks. 3. To investigate substantively whether tripartite closure between bands and venues is driven by band membership.

In order to explore if there is merit in a tripartite approach to analysing this data, models for two separate bipartite networks are estimated, one for the musician-band network and another for the band-venue network. The joint predictive distributions (GOF distributions) for these two networks combine to form a predictive distribution for the tripartite network that can be used to ascertain what tripartite dependencies are not accounted for. In the predictive tripartite network, either of the two bipartite networks can be held fixed making it possible to investigate which bipartite network accounts for the tripartite dependencies of the observed network. The results may suggest what additional insights might be gained from fully modelling the tripartite network by explicitly modelling local, tripartite graph configurations.

5.3 Data Background The dataset that I will explore here is a cross-sectional, tripartite network of 44 punk musicians, 16 bands and the 14 venues from 2015. It is a subset of the larger underground punk music network analysed in chapter six. All data in this subset was collected using Facebook events to record the bands and venues, and band members were sourced from the relevant bands’ public Facebook pages. The venues are located in Manchester and Liverpool, two cities in the North West of England thirty miles apart. The reduced dataset was necessary to keep the network relatively simple in order to test this new method of analysis. The bands in the network are either unsigned or are on independent record labels and the gigs took place at small venues in both cities. The bands are considered part of the underground punk world

72 in that most of the gigs are put on and promoted by members of the bands themselves. The majority of these bands do not make a living from playing music and rely on other forms of income in the form of employment or welfare. As such, there is no monetary incentive for choosing to perform at certain venues over others necessarily. Instead, the motivation for many bands will be to perform in front of a captive audience that enjoys their music. From a social capital perspective, bands that have members who are also members of other bands may provide them with more information as well as contacts that afford them the opportunity to perform at certain venues as well as make more informed choices about certain venues. In light of this I use exponential random graph models to investigate whether multiple band membership drives venue choice in the network through tripartite closure structures.

5.4 Exponential Random Graph Models Social networks are by their nature interdependent. It is not appropriate to reduce units of analysis to individuals as is the case in most statistical modelling processes. Here we want to uncover social structure and we are recognising that the ties between musicians, bands and venues are inherently dependent. An evolution of ideas about statistical network dependence can be identified in the literature. The first and most simple idea is Bernoulli dependence, where ties between two network nodes; represented in an adjacency matrix where ties are binary variables with values of either 1, where a tie exists or 0 where it does not; are taken as independent from each other. This is the case in the Erdős-Renyi model (Erdős and Renyi, 1959) where all possible networks would be equally likely. This is an unrealistic assumption for networks. Frank and Strauss (1986) introduced the idea of Markov dependence, where every tie between two nodes may depend on any other tie involving either of those two nodes. This statistical idea is congruent with network theories on closure, including seminal work on the strength of weak ties (Granovetter, 1976) and structural holes (Burt, 1992). In the case of a one-mode punk music network where nodes are musicians and are assumed to be pair-wise tied if they have played together in the same band, if two musicians, A and B, are connected, this tie might be dependent on whether musicians B and C have played together in a band. It is this notion of conditional dependence that first took the idea of network modelling towards assuming dependence and away from traditional statistical analyses where independence is assumed. The Markov dependence assumption defines the Exponential Random Graph model (Robins, 2007; Lusher et al, 2013) for the distribution of ties in the network. In its basic form the set of nodes (actors) in the network is denoted 푉 = {1, … , 푛} and the tie variables between two actors 푖 and 푗 in V as 푋푖푗 which takes the value 1 if a tie is present and 0 otherwise. The set of tie variables 푋푖푗 : 푖, 푗 ∈ 푉 may be represented by its adjacency matrix 푋. The ERGM is a log-linear model that represents a network, a graph 푋 = 푥, through various

73 summary measures (called network statistics) 푧(푥) including the number of edges, centralization measures and mutual ties (Lusher et al., 2013).

ERGMs take the general form:

1 푃푟 (푋 = 푥) = 푒푥푝 ∑ 휃 푧 (푥) 휅 푄 푄 푄

The probability of a given network 푋 = 푥 is given by a sum of network statistics 푧 that have been weighted by parameters 휃, where 휅(휃) is a normalizing constant. Network statistics are counts of configurations in the network 푋 = 푥 and can be thought of as sub graphs of the network. The probability of the network depends on how many of those configurations are present and the parameters inform us of the importance of each configuration (Lusher et al., 2013, p.9).

5.4.1 Bipartite Extension Bipartite networks, noted in chapters three and four, are networks with nodes of two different types, where ties occur only between the two different types of nodes. If we take for example, the network of bands and venues, then these two sets of nodes can be specified as 푉(푛)and 퐵(푚) where 푛 and 푚 are the number of nodes in each set respectively. The set of all possible bipartite networks is defined as 푋(푛, 푚).

5.5 Bipartite Analysis 5.5.1 Musician-Band Network In this bipartite network, there are 44 musicians and 16 bands. To model this data, I will use the bipartite ERGM function in MPnet (Wang, Robins and Pattison, 2009). The first two columns in table 6 below show the configurations for parameters that were specified in MPnet.

The first parameter is simply a dyad, which we would expect to fit because there are no isolates in the network, i.e. every musician is a member of a band. The second parameter represents the case where one musician plays in two different bands, one type of a bipartite two-star. The third parameter represents two musicians that play in the same band and the majority of bands have two or more members so this should definitely be included. The three- path parameter represents the case where one musician plays in two different bands and another musician plays in one of those bands but not the other.

74

5.5.2 Musician-Band Estimation Results

Table 6 ERGM model estimation results for the bipartite musician-band network. Significant parameters are more than twice their standard errors an indicated with an asterisk (*). Good convergence of parameters indicated by t-ratios < 2

Configurations Circle = musician Description Square =band Estimate (s.e) t-ratio Edge - musician a -3.375 (1.638) 0.063 * member of a band Two-star - musician a member of two bands -0.168 (1.378) 0.081

Two-star - band with two members -0.091 (0.477) 0.081 Three-star -band with three members 0.119 (0.068) 0.068

Three-path – one band with two members where one member is also 0.004 (0.140) 0.088

a member of a separate band

We know there are no isolates in the network because of the nature of these ties; every musician must belong to at least one band, therefore the ERGM was estimated conditional on the graphs having no isolates. Good convergence of these models is indicated by t-ratios of less than 0.1 (following the guidelines of Snijders, 2002), which is the case for all parameters in this model. Parameters are determined to be significant where the absolute values are larger than twice the standard error, indicated in the table by the asterisk (*). The only significant parameter in this model is the simple dyad parameter. This is what we would expect as ties from musicians to bands are nested, there are no musicians without bands and every band must contain at least one musician (solo artist), and usually at least two, but the reverse it not necessarily true. We are however, limited in what we can say from this model because there is little context in terms of how these ties play out in the context of band operations. The main purpose of a band is to perform and without including the venues, there is no way to link these band-sharing configurations to band activities. Tripartite analysis may help to assess whether these configurations are related to band activities, for instance, if a musician is in two different bands it might then be more likely that those two bands have performed at the same venue. 75

5.5.3 Musician-Band Goodness of Fit After the estimation of the model a goodness of fit was run (see table 7 below) indicating that the model was a good fit for the parameters chosen as well as for other possible parameters in the network with t-ratios of less than 2. Isolates read NaN (not a number) because the model was conditional on no isolates. The fact that the simple model manages to reproduce the observed network configurations, in particular the 4-cycles, seems to indicate that the collective production of music in terms of style does not stem from the musicians themselves. With the limitations of data in mind, we would still expect band members to cluster around bands if similarity of bands were driven by similarity of their constituent band members. This is not merely an artefact of few musicians being members of only a single band as there are multiple opportunities for 3-paths to be closed to create 4-cycles.

Table 7 Goodness of fit statistics for musician-band network

Parameters in bold included in estimated model

Parameter Count t-ratio XEdge 53 -0.524 Star2A 10 -0.521 XStar2B 83 -0.603 XStar3A 1 -0.452 XStar3B 87 -0.606 X3Path 63 -0.590 X4Cycle 1 -0.451 XECA 0 -0.477 XECB 9 -0.522 IsolatesX A 0 NaN

IsolatesX B 0 NaN XASA 9.5 -0.522 XASB 52.4 -0.569 XACA 9.5 -0.535 XACB 82.5 -0.603 XAECA 0 -0.491 XAECB 3.81 -0.390

5.5.4 Band-Venue Network In this bipartite network, there are 16 bands and 14 venues. As with the musician-band network I will analyse this data using the bipartite ERGM function in MPnet. Table 8 below shows the configurations for parameters that were specified in MPnet, where in this case the square represents a band and the triangle represents a venue.

As above the first parameter is simply a dyad and again there are no isolates in the network. In this network the second parameter represents one band connected to 2 different venues, meaning they have played at these different venues in the given time period. Bands tend to 76 play at more than one venue in a year, even in close proximity so this is likely to occur in the network frequently. This models variance in degree distribution among bands. Some may play at many different venues in the North West in a given year versus some that may play at a single venue. This is likely to reflect the fact that some bands tour in a wider geographical range than other bands and as a result may play fewer times in Manchester and Liverpool. The third parameter represents the case where 2 different bands have played at the same venue and if there are more of these 2-star configurations than can be accounted for by single ties, captures the centralization around venues by bands. The three-path parameter here represents the case where one band has played at two different venues and another band has played at only one of those venues. In this network it is useful to use three-paths as a control for the four-cycle parameter, where two different bands have played at the same two venues. This four-cycle closure may be a result of the number of gigs in the network and the distribution of these gigs among available venues. It could, however, be the result of a ‘cumulative advantage’ process (de Solla Price, 1976) where some venues become more popular than others through attracting many different bands.

5.5.5 Band-Venue Estimation Results

Table 8 ERGM model estimation results for the bipartite band-venue network. Significant parameters are more than twice their standard errors an indicated with an asterisk (*). Good convergence of parameters indicated by t-ratios < 2 Configurations

Square = band Description Triangle =venue Estimate (s.e) t-ratio

Edge – band perfromed at -2.631 (0.663) -0.035 * a given venue Two-star – two different bands performed at same 0.251 (0.251) -0.031 venue Two-star – one bnd performed at two 0.4694 (0.079) -0.020 *

different venues Three-path – for a pair of venues, one band has performed at both whilst 0.119 (0.068) -0.020 * the other has performed at only one of the pair. Four-cycle – two different bands perform at the -0.067 (0.033) -0.016 * same pair of venues

77

Again, there are no isolates in the network because of the nature of these ties; every band has played in at least one venue (the network data collected was based on performances), therefore the model was estimated respecting this restriction. There are 4 significant parameters in this model (table 8). The dyad parameter is significant which is expected with no isolates because every band in the network has performed, at the very least, at one venue. The second parameter, where two bands play at one venue is not significant. This means that there is no significant centralisation in the degree distribution of venues. In other words, there is no evidence of some venues being more popular than others. The third parameter where a band has played at 2 different venues is significant, indicating that some bands play at a lot more venues than other bands. This could be a sign of an unequal distribution of popularity of bands – some bands get all of the fame whereas others have to make do with a small number of occasional venues. It also could be the result of different touring patterns, where the wider reach of some bands results in fewer performances in specific areas.

The effect capturing bipartite reach, the three-path, is positive and statistically significant, indicating that the band-venue network is more ‘fractured’ than would be expected from an equal assignment of bands to venues. The four-cycle parameter, however, is negative and statistically significant.. It is important to note the difference here in parameter values. The three-path value is a much larger absolute value than the four-cycle parameter and this is reflected in the number of X3path (384) and X4Cycle (29) configurations observed in the goodness-of-fit-simulations below in table 9, indicating that the majority of three-paths do not resolve as four-cycle closure patterns – pairs of bands do not tend to perform at pairs of venues together. Possible underlying mechanisms here may be that bands seek to perform on different bills in the same year, exploring different venues and different gigs with differing band line-ups. It may also be the case that there are clusters of bands that choose to perform together at the same venue. This clustering of pairs of bands around singular venues may reflect underlying similarities in style and audience. On its own, this is evidence for a similarity in style and sound arising in the interaction with the audience, bands respond to positive feedback from audiences; conventions are shaped by this interaction so that audiences expect to hear certain styles of music or certain performance styles at certain venues and similarly bands know what audiences will react well to, choosing to perform together at venues that have provided positive experiences. Another explanation may be that the bands in the sample tend to perform at these venues disjointly, at separate events. This would mean that interaction between bands in the sample does not take place within the venue and that other mechanisms may play a part in venue choice, such as multiple membership of bands, that is not explained by the current bipartite models. Tripartite investigation may help to parse out mechanisms behind formation of three-path configurations in the band-venue bipartite 78 network, investigating whether a pair of bands that share members are more likely to perform at the same venue than a pair of bands that do not share any common members.

This model is slightly more interesting on its own than the musician-band network. It is in this portion of the network that the activity of the bands takes place and the music world of underground punk is based; on live performances. However, there is little context in terms of how these ties play out in the context of the musicians who are the constituent members of these bands and if their tie patterns have any bearing on how we can interpret the network. Tripartite analysis will investigate whether the structure of ties at both levels gives us additional information about the network structure.

5.5.6 Band-Venue Goodness of Fit Again after the estimation of the model a goodness of fit was run (table 9) and indicated that the model was a good fit for the parameters chosen as well as for other possible parameters in the network with t-ratios of less than 2 (isolates are again irrelevant for the GOF as none are present).

Table 9 Goodness of fit statistics for band-venue network Parameters in bold included in estimated model Parameter Count t-ratio XEdge 41 -0.081 Star2A 44 -0.077 XStar2B 102 -0.073 XStar3A 26 -0.220 XStar3B 260 -0.148 X3Path 384 -0.076 X4Cycle 29 -0.069

XECA 73 -0.112 XECB 384 -0.162 IsolatesX A 0 NaN IsolatesX B 0 NaN XASA 32.88 -0.033 XASB 42.13 -0.054 XACA 35.15 -0.176 XACB 88.25 -0.073

XAECA 61.5 -0.045 XAECB 110.07 -0.042

79

5.6 Tripartite Analysis In order to analyse these networks in a tripartite context, we first need to bring the two separate network datasets together into an adjacency matrix (this concept is explored formally by Farraro & Doreian (1984)). In fig.2 and fig.3 below, example matrices are shown to illustrate the differences between bipartite matrices and a tripartite matrix for a simple example dataset with 4 musicians, 3 bands and 2 venues.

Bipartite Matrices Bands Venues B1 B2 B3 V1 V2

M1 B1 1 0 0 1 0

M2 B2

1 1 0 Bands 1 1

Musicians M3 1 1 1 B3 1 0

M4 1 1 1

Fig. 2 Two separate bipartite matrices for a network with 4 musicians, 3 bands and 2 venues.

Tripartite Matrix Musicians Bands Venues M1 M2 M3 M4 B1 B2 B3 V1 V2

M1 1 0 0 X X X X X X M2 X X X X 1 1 0 X X M3 Musicians X X X X 1 1 1 X X M4 X X X X 1 1 1 X X B1 1 1 1 1 X X X 1 0 B2 0 1 1 1 1 1 Bands X X X B3 0 0 1 1 X X X 1 0 V1 X X X X 1 1 1 X X

Venues V2 X X X X 0 1 0 X X

Fig. 3. Tripartite matrix for the same network of 4 musicians, 3 bands and two venues. The black crosses represent where the matrix is null because there is zero probability of ties existing in this case.

80

Fig. 4 – Tripartite network of 44 musicians, 16 bands and 14 venues Manchester and Liverpool for 2015. The main purpose of the tripartite analysis in this thesis is to analyse band membership as a driver of closure in the network. As such, I now give a brief explanation of transitive closure before moving to specify what I mean by tripartite closure.

5.6.1 Transitive Closure When network scholars talk about network closure, it is often in a unipartite context and usually involves looking at transitivity in a network between three nodes.

Fig 5 - An open triad (left) and a closed triad (right) in a network of three friends A, B and C

Granovetter (1973) gives the example for transitivity using three people A, B and C (fig.5), where B is friends with both A and C, but A and C are not friends. This represents an open triad. However, when A and C are also connected this is called a closed triad. If B is friends

81 with both A and C then it is likely that eventually A and C will interact (a necessary condition for friendship) and over time therefore, there is a greater than zero probability that A and C will become friends. The tendency towards these tie patterns is usually measured through transitivity, which is the number of observed transitive triads, over the number of observed transitive plus intransitive triads (treated as the possible triads that could have been formed).

Using fig. 5 above, we can think of A, B and C as three bands, connected if they have played at the same venue together (a one-mode projection of a two mode network) On the left we have the situation where band B has played at the same venue as band A, band B has also played at a the same venue as band C. Bands A and C have never performed at the same venue together. In the network of the underground punk world this would be interesting because there are a small number of venues and a comparatively large number of bands so there must be a reason why two of these bands have not performed at the same venue. It may be down to attributes of the bands, or the type of venues they play at, or perhaps it is something that is present at the level of individual musicians. From this one-mode projection all this information on musicians and venues is conflated and the detail is lost so there is a need for a more appropriate method for the investigation this type of closure, to ascertain the reasons for it in the context of the whole tripartite network.

5.6.2 Closure in a Tripartite Context Tripartite closure, shown below as a four-cycle graph configuration, is likely to occur where one musician is a member of two bands. In this case, a tendency towards network closure would indicate that, by extension of the four-cycle principle, if a musician is in multiple bands then those bands are likely to perform at the same venue. It seems logical that, following a positive experience of performance at a particular venue, that musicians that have played at one venue could suggest to another band, which they are a member of, that they should make best efforts to secure a gig there. Another example of where a four-cycle would be formed may also occur simultaneously where one musician is in two bands that perform at the same gig and therefore the same venue at the same time. This analysis aims to investigate whether shared band membership drives tripartite closure between bands and venues.

82

Fig 6. Tripartite network configurations: three-path (left) represents a lack of closure, whilst the four-cycle (right) represents closure in a tripartite network. 5.6.3 Recombining Bipartite Networks in GOF Distribution Under the assumption that there is no dependency across levels, the distribution of the combined musician-band networks and the band-venue networks serves as a null-distribution for ascertaining the prevalence of tripartite motifs (fig.6). To define tripartite network configurations, we introduce the following notation.

Let

푥푖푗 = {1 푖푓 푝푒푟푠표푛 푖 푖푠 푖푛 푏푎푛푑 푗 0 표푡ℎ푒푟푤푖푠푒 , 푖 = 1, … 푛, 푗 = 1, … , 푚 and

푦푠푡 = {1 푖푓 푏푎푛푑 푠 푝푙푎푦푠 푎푡 푣푒푛푢푒 푡 0 표푡ℎ푒푟푤푖푠푒 , 푠 = 1, … 푚, 푡 = 1, … , 푣

This now lets us define three configurations relevant for cross-level interactions. Cross-level assortativity The statistic

퐴(푥, 푦) = ∑ 푥+푗푦푗+ 푗=1 sums the cross-products of the band and venue degrees and is large if there is a high positive correlation between the degree distributions. This would mean that bands with many members tend to play at many venues. Given the low heterogeneity in the number of band members this may not be interesting in its own right, but the count would have a substantive interpretation if bands had fluent memberships and large bands could substitute their members and thus play at many venues with different set-ups.

83

Tripartite three-paths The count of tripartite three-paths

푛 푚−1 푚

푃3(푥, 푦) = ∑ ∑ ∑ 푥푖푗푥푖푘(푦푗+ + 푦푘+) 푖=1 푗=1 푘=푗+1 is a count of instances where venues are indirectly connected to a band through a band member. Such a three-path creates the opportunity for cross-level closure, something which could manifest itself in a band having a band-member that has played at a venue even if that band has played there themselves. If the collective production of a style or sound is driven by band members, we would expect that there were few instances of these three-paths as the shared band-membership would induce a similarity of bands that would then seek to perform to similar audiences, leading to the closure of these three-paths, forming four-cycles.

Tripartite four-cycles Of focal interest for investigating the interplay of style and membership and audience is the four-cycle configurations

푛 푚−1 푚 푣

퐶4(푥, 푦) = ∑ ∑ ∑ 푥푖푗푥푖푘 ∑ 푦푗푡푦푘푡 푖=1 푗=1 푘=푗+1 푡=1 that close the three-paths. If band-membership were driven by the venues (i.e. audiences that bands perform to) or if venue selection were driven by individual members, then we would expect to see more of these four-cycles than the effects in the individual bipartite networks explain. An incidence of these four-cycles over and above the bipartite models would indicate that the choice of audience was really driven by the style induced by individual members rather than the clustering of bands around venues being driven by the style that they adapt to or is created in interacting with their audience. If there is not over-incidence of these four- cycles it means that the clustering around venues is not contingent on individual band members, suggesting that styles are not dependent on band-composition only.

84

5.6.4 Tripartite Simulation results

Fig. 7 distributions of simulated networks where the red mark indicates observed network. Networks in centre of distribution indicate they can be explained by bipartite analysis. Networks in the tails of the distribution would indicate tripartite analysis is likely to provide better models. 5.7 Tripartite Results We can see from figure 7 that the incidence of tripartite closure in the observed networks, illustrated by the counts of three-path and four-cycle local configurations, is close to the centre the distributions of tripartite counts in the simulated networks. As the GOF procedure is constrained to have the same number of bands and venues, as well as the same number of ties between them, as the observed networks, we can conclude that the prevalence of tripartite closure is not significantly different from random networks that would occur by chance. Put simply it appears there is no deliberate effort made by bands that share members with other bands to perform at the same venues. These results indicate that bands choose to perform at venues independently of band membership ties, showing that the clustering around venues (and therefore performance to particular audiences) in the bipartite models is not driven by individual preferences but preferences that form in band interaction with audiences at venues which warrants further investigation. One additional key result is that we now know we do not need to model the tripartite networks explicitly to account for unobserved tripartite mechanisms that were unexplained by the bipartite networks. If we return to the results from 85 the earlier bipartite ERGMs in section 5.4, we can see that significant parameters lie in the band-venue part of the network, with significant clustering around venues; with lower incidences of the four-cycle parameter; indicating where bands will play two venues in common with each other; than the unresolved three-paths. This shows that any two given bands will be likely to perform at most at one venue in common, suggesting that bands choose specific venues to perform at, rather than a scattergun approach to venue selection. This warrants further investigation into the mechanisms behind this clustering which I pursue in chapter 6, to highlight the importance of conventions, duality and interaction within the social spaces of venues as drivers for tie formation.

5.7.1 Further Possibilities The data that has been used in the tripartite analysis above is a sample of the total collected network. The collection of further, more detailed data on individual band membership as well as data on different music networks would allow us to discover if the overlapping band memberships drive tripartite closure in other instances or not. Additional tripartite network datasets would also assess whether the procedure is robust and whether it is affected by networks of a much larger size and increased complexity.

5.7.2 Attribute Data The tripartite network dataset could also include attribute data. These data would allow us to see whether the tripartite four-cycles are conditional on attributes at different levels; venues for example are located in Manchester or Liverpool. Both Manchester and Liverpool have rich but very different musical histories and the variation in network substructures might be conditional on venue location; bands are also located in different cities and the musicians in those bands may have different demographic attributes that may affect both their level and type of activity in the network.

5.7.3 Venues vs Events It would be worthwhile using the same method with a dataset where the venues were replaced with the gigs (events) that took place. This differs slightly because the bands that play at the same venues may not necessarily play at these venues at the same point in time and the ties to venues and events tell us different things about the way that the network operates. Here it is important to separate closure that occurs in the manner discussed above (i.e. where bands perform at the same venue, but not necessarily on the same occasions) or closure that occurs where bands perform together at the same event, at the same time. In this context the bands are performing together. The frequency of these patterns could depend multiple factors that could be explored using node attributes. Venues/events are located in certain geographic areas; bands are located geographically, have certain sub-genres and some sing political songs; 86 musicians are also located geographically, are different ages and genders and have different employment statuses that my impact on their ability to function in the network. For example, it might be the case that more drummers share bands and if they do, then it is likely that they will play at the same event because then the equipment can be shared; also, the converse might be true where a drummer might not want to play with two different bands in one night because they will be exhausted.

5.7.4 Further Configurations A further step to investigating tripartite analysis and what it might tell us would be to look at further configurations that might represent cumulative advantage in tripartite networks (fig. 8) where a musician that is in two bands might play at multiple different venues or events with both of those bands. There also might be a threshold on the number of shared members in two bands and whether that makes a difference in the number of venues or events that those two bands play at.

Fig. 8 Cumulative advantage in tripartite closure

Further investigation into tripartite analysis needs to be done in order to encompass more types of network data including attributes and directed ties as well as further specifications for different configurations of interest.

5.8 Discussion This chapter has explored tripartite structures of musicians, bands and venues in the underground punk world of Liverpool and Manchester. It specified tripartite closure patterns in music networks and proposed a novel method to test for drivers of ties between bands and venues. A method for the testing of the prevalence of unexplained tripartite closure structures using GOF distributions from bipartite exponential random graph models (ERGMs) to assess if they have accounted for mechanisms of tie formation in small networks has been suggested. The method was presented in general terms and specified statistically to ensure that it would be transferrable to other work wishing to investigate dependencies in tripartite network structures, particularly in other cultural contexts and especially in work on music worlds. It has 87 contributed an important first step in the development of explicit tripartite methods to analyse structures that appear in music networks, social movements and beyond. Further to this analysis and the innovative methodological development that it contains, it would also be appropriate to conduct analysis with alternative datasets, although tripartite datasets might be difficult to source as there are limited examples in the literature of scholars attempting this analysis (Borgatti and Everett, 1992; Mische and Pattison, 2000; Bohman, 2012). Statistical modelling techniques would also need to be developed for situations where the observed network lies in the tails of the distributions of the induced null-distribution of tripartite networks, and therefore cannot be explained adequately through separate bipartite analyses.

In this small punk live performance network, the observed counts of tripartite three-path and four-cycle structures sit close to the centre of the simulated networks from the GOF distribution. From this result above, it is also evident that the two separate bipartite analyses are sufficient for the explanation of these tripartite structures (three-path and four-cycle). If we look back at the bipartite analyses conducted earlier in this chapter then we see that significant parameters occur in the band-venue network. This testing procedure then, provides evidence that most of the variability in this network is focused around venues and which bands perform at those venues. This justifies further investigation into the larger bipartite network of bands and venues that has been collected from Facebook events to see what role venues play in the larger collected network of bands and venues. Aside from the introduction of an innovative methodology for the detection of tripartite closure patterns, I have shown that overlapping band memberships do not drive closure in the network. This gives a strong indication that performance ties between bands and venues, where bands make a deliberate choice to perform at those venues, have other mechanisms at play in their formation. I hypothesise that bands make choices based on their interactions within venues as dual entities; both as a collective group and as individuals and these interactions and experiences shape their choices. In lieu of this I investigate the larger bipartite network in the following chapter.

88

Chapter Six: Bipartite Dynamic Network Analysis of Bands and

Venues

6.1 Introduction Following on from the tripartite analysis of the underground punk network in Manchester and Liverpool, the aim of this chapter is to conduct subsequent empirical analyses on the underground punk music network over the three-year time period from 2013-2015. After introducing and testing a novel form of tripartite investigation in the previous chapter, there was strong evidence that network ties are formed between bands and venues somewhat independently of the band membership ties in the musician-band half of the network. The hypothesis that was made preceding the tripartite investigation, that there would be a tendency towards tripartite network closure based on membership of multiple bands, did not hold for the network. It was also concluded that the two separate bipartite analyses were sufficient for explaining the prevalence of tripartite phenomena in this case and so the bipartite model estimation results could be interpreted in the knowledge that information was not lost due to any lack of accounting for dependencies between musician-band and band-venue ties. Parameter effects that were statistically significant in the model were found in the band-venue part of the network and this part of the network is what will take precedence in this chapter, expanding the sample network from chapter five, using the entire collected bipartite network of 725 bands and 41 venues as a starting point for analysis.

First, I discuss the motivation behind the analysis of the bipartite network with a focus on venues and outline what I hope to discover through comparison and interactions between the geographical locations of both bands and venues. I then introduce the concept of network dynamics with reference to substantive areas of interest in the literature as well as methodological solutions for the analysis of longitudinal data, again making the distinction between standard statistical analysis of longitudinal data and longitudinal statistical models for network analysis. I reason that models using continuous time, in particular the Stochastic Actor-Oriented Models (Snijders, 1996, 2001, 2017; Snijders, van de Bunt and Steglich, 2010), are the best solution for analysis over discrete time models, such as Temporal Exponential Random Graphs Models (TERGMs) that are also specified in the literature (Robins and Pattison, 2001; Hanneke, Fu and Xing, 2010), but have been acknowledged to be problematic (Block et al., 2018). I introduce these models theoretically and statistically using the specification in Snijders, van de Bunt and Steglich (2010) and expand this with Koskinen and Edling’s (2012) bipartite extension, which I use to analyse the punk live performance network, elaborating on the theoretical and substantive implications of, and reasons for, modelling a network of bands and venues in this way. Next, I give a brief overview of the data I have 89 chosen to model, followed by the parameters that were chosen for inclusion in the models. Finally, models are estimated, and the results are discussed in the context of bands and venues in the underground punk music world.

6.2 Motivation I have established in the analysis of the tripartite network in the previous chapter that there is significant clustering around venues by pairs of bands in the underground punk music network in Manchester and Liverpool. This clustering happens around venues that are successful in gathering a regular audience, that become the ‘places to play’ in each city. It is my contention that this is the result of the interaction between bands and audiences in those spaces and so the venues provide a platform to form and reform conventions around the music which the bands choose to play and the audiences expect to hear and positively engage with over time.

The purpose of analysis in this chapter is three-fold:

1. To investigate the change in structure over time in the underground punk music world 2. To investigate the role of geographical location in venue choice over a three-year period 3. To uncover the role that venues play in sustaining a music world.

6.3 Network Dynamics Network dynamics concern the change in network ties between actors over time. This is a natural phenomenon and relational ties are formed and broken every day. For some networks, if we take, for example, a secondary school class, the network of friendships in that classroom will change from when those pupils start secondary school at age eleven to when they leave school at age sixteen. Some of those friendships may endure for the entire five years whilst others may break and possibly reform at a later date. The formation and dissolution of those friendships may be dependent on multiple and complex endogenous structural mechanisms or as the result of exogenous actor covariates. Changes in ties may be the result of the current network and the structural positions of the actors within it, for instance in triadic closure, where friendships are formed through friends of friends, exogenous factors such as age, sex or other relevant covariates of our node sets, attributes among pairs of nodes, known as dyadic covariates, as well as residual unexplained processes.

In the underground punk world bands choose to perform at certain venues and these choices are subject to change over time, bands may choose to play at the same venue year after year or may choose to play at different venues or a variety of venues. These choices of venues to play at in one year are affected and influenced by the network structure; the choices made in the 90 previous year and the experience of interaction between band and audience within the venue; as well as location attributes of bands and the interaction between band locations and venue locations. These choices and the reasons for them are analysed below using statistical models that have been developed specifically for dynamic networks. As with the exponential random graph model in the previous chapter, these models differ from standard statistical models for longitudinal data, in that they do not assume independence of variables assigned to individuals, rather they account for dependencies between the tie variables that connect nodes in a network.

We can take account of network dynamics through modelling these tie changes probabilistically using statistical models. In general, network scholars have conceptualised statistical models of network dynamics in two different ways:

1. Networks modelled in discrete time 2. Networks modelled in continuous time

The first type of model, where networks are considered to change in discrete time, is implemented via the use of temporal exponential random graph models (TERGMs), a longitudinal extension of the exponential random graph model for social networks. TERGMs have unresolved issues with consistent interpretation, noted by Block et al. (2018). The second type of model, where networks are modelled in continuous time, is largely and most notably implemented via the use of Stochastic Actor-Oriented Models (Snijders, 1996, 2001, 2017; Snijders, van de Bunt and Steglich, 2010), which Block et al. (2018) note have no such issues with interpretation and compare favourably against TERGMs when assumptions of actor agency in the model are satisfied. Consequently, I favour Stochastic Actor-Oriented Models for the analysis of the underground punk networks in Manchester and Liverpool and adopt the assumption that bands in the network make distinct choices to perform at specific venues in Manchester and Liverpool over a time period of three years from 2013-2015.

6.3.1 Network Growth vs Network Activity It is pertinent here to distinguish between what I posit are two methods of measuring network change over time. The first of these is network growth, where ties are measured cumulatively. This process of tie accumulation makes sense for some types of network, for instance, academic co-publication networks. Ties exist in a bipartite network of academics and publications and these ties can never be broken, once an academic paper is published it exists in perpetuity and therefore change in the network will only happen in the form of new tie creation to new papers and the network will grow in size or remain static. Network growth has also been used conceptually in a music context by Crossley (2008, 2009, 2015a, 2015b) where he asserts that strong ties, created through collaboration, are accumulated to reach the ‘critical 91 mass’ of participants needed to birth a new music world, as was the case with punk in London and post punk in Manchester, Liverpool and Sheffield.

However, in this thesis, what I aim to capture is the collective action that sustains a music world long after the genesis of a new one and so this focus on network growth and accumulated ties is inappropriate. Instead, what I focus on in this analysis is network activity, in particular the activity of live performance. Bands make choices to perform at certain venues and these choices differ over time. These choices are influenced by a range of factors but most importantly, previous experience of performing at the venue. This focus on activity allows us to analyse how a music world is sustained, whether certain venues emerge as key foci in the network; attracting a higher number of bands than other venues, as well as repeatedly chosen by bands over the three years as their places to play. It also aims to capture the transient nature of some venues in juxtaposition to the more permanent nature of others. In order to parse out these choices around key venues, the data is first elaborated upon below.

6.4 Network Data for Analysis As noted in chapter four, the network data that I collected, in its raw form was extracted from Facebook event pages. The total dataset consists of 725 bands that performed at forty-one venues over three years from 2013-2015. Of those 725 bands, only eighty are consistent in the network over the entire three-year period, performing at thirty-nine of those venues. This in itself speaks to the nature of the underground punk world and in particular the nature of the venues within it. It appears that the bands and venues that sustain a world within a city are a small core number of bands that consistently perform at venues that become the foci of that world.

Before analysis can be conducted, the dataset must be reduced for two reasons. The first is that we are restricted by the modelling process in that to model the network longitudinally using stochastic actor-oriented models we need the node sets for bands and venues to be consistent so that we are able to draw meaningful conclusions about bands’ patterns of venue choice over time. It is possible to specify joiners and leavers in the network to account for bands that are missing in certain years, but this is problematic in a network of this size as the number of joiners and leavers as a proportion of the total number of bands is extremely high. The second reason pertains to the research questions and the way in which network ties are defined in the model. In saying a tie exists between a band and a venue, we are saying that the band has chosen to perform at that venue. If bands are consistently performing at venues in Liverpool and Manchester over three years, then these ties are aligned more strongly with this assumption and therefore our conclusions about the patterns of venue choice and the importance of venues will carry greater weight. 92

The venues node set, however, is reduced only by two nodes in the dataset for analysis. The two nodes that are not included are never chosen by the eighty consistent bands in the network and so are removed from the dataset to avoid biasing parameter estimates in the model. Otherwise, we are assuming that venues in the dataset remain open and available for bands to perform at, it is simply the case that none of the bands choose to perform at these venues during certain time periods. I choose to model the data in this way as I believe it is closest to how the underground punk world operates, venues are available to perform at, they have the required resources for live performances to take place but the eighty key bands may not choose to perform at them for a variety of reasons, discussed in the context of the model results that follow later in this chapter.

Location attributes for both bands and venues are also included in the dataset in order to assess the role of location of both bands and venues in the choices that bands make when performing at a venue. Data on venue location was sourced from Facebook profile pages of both bands and venues. We can see from table 10 below that of the thirty-nine venues included in the analysis, twenty-four are located in Manchester, with fifteen located in Liverpool and of the eighty bands in the network, fourteen are located in Manchester, eighteen in Liverpool and the remaining forty-eight bands outside of either city.

Table 10 Counts of bands and venues by geographical location Bands Venues Location Count Percentage Count Percentage Manchester 14 17.5% 25 64.1% Liverpool 18 22.5% 14 35.9% Other 48 60.0% - - Total 80 100% 39 100% In the first instance, the three categories for band location are separated into two dummy variables:

Band Liverpool was constructed to take a value of 1 when a band is located in Liverpool and a value of 0 when it is not located in Liverpool.

Band Manchester similarly was constructed to take the value of 1 when a band is located in Manchester and 0 when it is not located in Manchester.

As with standard statistical modelling techniques, only two dummy variables are required for band location as when values are 0 for both variables, this indicates a band that is not located in either city and this ‘other’ category, seen above in table 10, serves as the reference category to compare against when interpreting a network model.

93

Venues in the dataset are located in either Liverpool or Manchester and so a binary variable to account for this was constructed:

Venue Location Manchester is a binary variable that takes the value of 1 when a venue is located in Manchester and 0 when it is not located in Manchester and de facto is located in Liverpool.

Network data was constructed in three separate affiliation matrices for each discrete year, 2013, 2014 and 2015. The rows of the matrices pertain to specific bands and columns pertain to specific venues. Ties between bands and venues take the value 1 where a band performed at a given venue in that year and 0 where there was no performance at a given venue in the particular year. One venue in the dataset, MelloMello in Liverpool closed at the end of 2014 thereby removing the opportunity to choose to perform at that venue. As such I code the matrix with ‘10’ to implement ‘structural zeros’ (Ripley et al., 2019) to ensure ties are deemed impossible in the modelling process.

It is important to note that our node attributes, in this instance, remain constant through all three years whilst the network ties are dynamic and subject to change. The combination of the three network affiliation matrices and three node attribute variables gives us our dataset for statistical modelling using stochastic actor-oriented models, which I now move to explain.

6.5 Stochastic Actor-Oriented Models Statistical models for network dynamics were introduced briefly in section 6.3 above. Here I aim to define in more detail the stochastic actor-oriented model (SAOM) (Snijders, 1996, 2001, 2017; Snijders, van de Bunt and Steglich, 2010). The SAOM, also often referred to as a ‘Siena’ model, after the software package it was originally implemented in, is a statistical model for the analysis of longitudinal network data and network dynamics.

One important aspect of the model to note is that is it assumed actors in the network change their ties one at a time, a constraint first proposed by Holland and Leindhardt (1977) that has been adopted in SAOMs. This ‘method of moments’ assumes that actors know the current state of the network and use this knowledge to make informed decisions about tie formation in a series of ‘mini-steps’. At each mini-step actors have an opportunity to form a tie or break a tie in the network and there can be an infinite number of mini-steps that take place for the network to change from its current state at the first time of observation, in the case of underground punk the network at the end of 2013, to the next observation time (see Snijders, 2001 for the detailed algorithm specification). This opportunity for change is represented by rate parameters in the model. Additional network configurations and behaviours can be specified, as well as exogenous actor attributes. These models allow us to investigate the propensity for actors to form certain tie patterns over the time periods in question, in this

94 case, I am able to investigate the different strategies for choosing venues that bands will implement.

6.6 Modelling the Network Stochastic actor-oriented models (SAOMs) are currently implemented in the RSiena package (Ripley et al., 2019; Snijders, 2017) available for the open source R statistical software. I draw on the comprehensive manual available (Ripley et al., 2019) and key research articles (Snijders, van de Bunt and Steglich, 2010; Koskinen and Edling, 2012) to ensure best practice in model building. Although it is possible to specify SAOMs for non-directional data (Ripley et. al, 2019) the models assume some level of agency of the actors in determining ties, and so directionality in the tie data is preferred.

As I aim to model a bipartite network of bands and venues, I use the extended specification first proposed by Koskinen and Edling (2012). In their analysis of interlocking directorates of Swedish directors and companies, ties are assumed to be directed from corporate boards to directors whereby the boards choose which directors they recruit.

In order to specify the directionality of the ties in the underground punk network, careful consideration was needed. In reality, there are multiple communications between multiple actors that take place before the band performs at a given venue. Principally there are perhaps three roles involved, the band as a collective group, a venue and optionally a promoter that organises the event. Prior to a band becoming established in the world they will have to make the first approach to a venue or promoter with the hope that they can book themselves a gig. Once a band is established, the first communications may come from the venues or promoters who ask the band to play at an event they are organising. After this first communication there will be multiple communications back and forth between the three parties, as well as some discussion among band members as to availability.

It is my contention, in this analysis, that the final decision is made by the band collectively to perform at a given venue. In addition to this, the sample I use to assess the importance of venues uses eighty core bands that have all performed over three years with only a small percentage of these bands performing in their home city. Established bands and in particular, established bands that travel to other cities to play gigs, from my own experience as a punk participant, will generally choose to play gigs at venues where they have had a positive experience in their interactions with other bands and the audience in the venue. This active role of the band in venue choice lends itself to the use of stochastic actor-oriented models for analysis of the network as the agency assumption is satisfied and we can conclude that ties are

95 formed with bands identified as ‘senders’ of a live performance and venues as ‘receivers’ of ties, where ties are possible in this direction only.

In Koskinen and Edling’s (ibid.) formulation, they specify a pairwise method to account for the high level of composition change in their network of boards and directors. In their example, directors and boards are included in the network if and only if they are actively involved in the network. In the case of interlocking directorates, it makes sense to conceptualise the network in this way, for a board to exist it must have sitting directors and vice versa, if an individual no longer sits on any company boards they de facto rescind the title of director.

My analysis of the underground punk world, however, differs from this conceptualisation because venues that are isolates are an important part of the network. The set of venues I include in the longitudinal models remains constant. This set of venues is obtained as the set of all thirty-nine venues performed at by at least one of the eighty core bands that had performed in at least one venue in every discrete calendar year from 2013-2015. If we know that core punk bands have performed at these venues then we know these venues must be available to perform in. Searching systematically on social media and for venue-specific websites online revealed, to the best of my knowledge that all but one of the venues included in the network were open and had facilities for live music over all three years in question and entertained punk bands at least once over that time period. The one exception to this is MelloMello in Liverpool as the venue closed at the end of 2014 and therefore it would be impossible to have a tie from any bands to that venue in 2015. To account for this, I use ‘structural zeros’ in the data to indicate that a tie is impossible. For all other venues I make the assumption here then, that these venues are available and so should be included in every observation of the bipartite network, whether performances took place in a given year or not. This allows us to investigate the structure of ties and whether venues are consistently used or are transient in nature and unpack the reasons for this. In order to do this, I focus upon network activity rather than growth of accumulated ties and so opportunity for both tie creation, where bands perform at new venues and dissolution, where they do not perform at the venue in the given year are permitted.

In standard statistical modelling procedures, we have several explanatory variables and one outcome variable. Assuming relative independence of the explanatory variables, we are then able to conduct exploratory analysis to assess distributions of variables across our sample and compare our explanatory variables to our outcome variable using correlations and crosstabulations. In social network models, the landscape differs as our parameters fall into two categories. Parameters can be specified as both exogenous parameters; referred to often

96 as node attributes, in this case the band and venue locations, akin to more traditional variables used in statistical models; and endogenous parameters that concern the structure of the network. In SAOMs the structure of the network dynamically changes over time, so measures are generally thought of as tendencies towards certain patterns of tie formation over the time period in question. A multitude of parameters are available for inclusion in the SAOM and are listed in the manual for RSiena (Ripley et al., 2019). Suffice to say it is not appropriate to list all possible parameters and reasons for their omission in the models that follow, instead I outline only the chosen modelling parameters below.

6.6.1 Parameter Specification In order to model the network, we must first specify the parameters for inclusion in the model. The parameters included in any statistical model should be informed by theory, exploratory analysis and, most importantly, the research questions we are trying to answer. The aim of the models below is to analyse the role of venues in the underground punk music world. I also wish to investigate the effect of the geographical location of the venues; is there a difference in structure and tie formation with venues in Manchester compared to Liverpool? I hypothesise that as there are a larger number of venues in Manchester and therefore a larger number of opportunities to play. Parameters are separated into two functions, a rate function, that measures the estimated number of opportunities for actors to change their ties and an evaluation function that includes the endogenous, structural parameters and exogenous attributes that serve as covariate parameters in the network. Four models are specified below, the rate function element of both models is the same and so this is explained in the first instance, followed by the evaluation function parameters that are additionally included in each model in turn. 6.6.2 Model 1 Rate: 2013 – 2014

Rate: 2014 -2015

As there are three observations of the network, there will be, by the nature of the model, two rate parameters, that capture the estimated opportunity for change in ties between each observation moment, 2013-2014 and 2014-2015. The rate parameters estimate the opportunity for tie formation and dissolution in the network for each actor, the opportunity for network change. This will always be larger than the observed number of changes for each actor as it is assumed that opportunities will be not always be realised as actual change in network ties.

97

Outdegree is included by default in every model to account for the density of the network. Most social networks by their nature are sparse with low densities (density measured as the fraction of all possible ties between nodes that are actually observed). Sparsity is true of the underground punk network so we expect that this parameter will take a negative value that represents a low density, literally, a tendency away from forming ties with every possible node.

Indegree popularity measures the popularity of venues. If all venues are equally popular then this should tend towards zero. Because this popularity measure takes time into account, a positive parameter indicates that venues gain popularity due to already existing popularity, a phenomenon widely found in social networks. In layman’s terms we might call this the ‘rich get richer’ phenomenon. It has been discussed by several scholars as ‘the Matthew Effect’ (Merton, 1968), ‘cumulative advantage’ (de Solla Price, 1976) and more recently, ‘preferential attachment’ (Barabasi and Albert, 1999). It is important to include this effect in the model to test the hypothesis that venues become ‘foci’ of the underground music world through positive experiences at venues that may reinforce the desire to revisit those venues the following year.

Anti in-isolates measures the tendency of bands to want to connect to venues that would otherwise have been isolates with no bands performing at these venues, whilst also wishing to maintain ties to venues that would otherwise be isolates if the tie was dissolved.

Indegree at least 2 measures the tendency for at least two bands to perform at a given venue, with a positive parameter indicating a preference for at least two bands to perform at the majority of the venues. In the general case of the underground punk network, we expect that this would be a strongly significant parameter as every gig usually involves two bands at the very least, more often than not three or more bands, meaning that venues should have far more than two bands connecting to them over a period of three years. However, as we are analysing the choices made by a much smaller number of key bands, this may not be the case.

6.6.3 Model 2 In addition to the parameters specified above the second model additionally contains:

Venue Manchester is a binary attribute of a venue in the network that takes the value 1 if a venue is located in Manchester and 0 when it is located in Liverpool. This is included to test for a rudimentary preference for performing in Manchester. It is assumed that this may be the case as there are a larger number of venues to choose from in Manchester (24) than in Liverpool (15).

98

6.6.4 Model 3 In addition to the parameters in the second model, the third model also additionally includes:

Band Liverpool is a binary attribute of a band in the network that takes the value of 1 if a band is located in Liverpool and 0 when it is not. On its own this serves as a simple measure of whether bands located in Liverpool tend towards forming ties with a greater number of venues than those that are not located in Liverpool.

Band Manchester is a binary attribute of a band in the network that takes the value of 1 if a band is located in Liverpool and 0 when it is not. On its own this serves as a simple measure of whether bands located in Manchester tend towards forming ties with a greater number of venues than those that are not located in Manchester.

Taken together, as they are in the model, Band Manchester and Band Liverpool serve as dummy variables for a categorical node attribute with three values, where bands are located either in Liverpool, Manchester or elsewhere. As the ‘elsewhere’ i.e. not located in either Liverpool or Manchester does not have its own assigned variable, it serves as the reference category. Positive values for either variable then can be interpreted as bands in either city forming ties with more venues than bands located outside of the two cities.

6.6.4 Model 4 As I wish to investigate the tendency towards homophilous tie formation by geographical location, where bands from Liverpool may be more likely to choose to perform at venues in Liverpool and similarly bands from Manchester may prefer to perform at the Manchester venues, I construct two dyadic covariates:

In general: If

퐴 = [푎푖푗] 푖푠 푎푛 푚 × 푛 푚푎푡푟푖푥 and

퐵 = [ bij] 푖푠 푎 푝 × 푛 푚푎푡푟푖푥 the product

푇 퐴퐵 = [ 푐푖푗] 푖푠 푎푛 푚 × 푝 푚푎푡푟푖푥

This means that if we treat the attribute variable Band Manchester as an 80 × 1 matrix 퐴 and treat the attribute variable Venue Manchester as a 39 × 1 matrix 퐵 and take the product 퐴퐵푇

99 then we can construct an 80 × 39 matrix that serves as a dyadic covariate that takes the value 1 when both band and venue are located in Manchester and 0 otherwise.

Similarly if we take the attribute variable Band Liverpool as an 80 × 1 matrix 퐷 and we compute 1 − 퐵 where 1 is the scalar value and 퐵 is as above then we generate a new matrix 퐸 that takes the value 1 when a venue is located in Liverpool and 0 otherwise. Subsequently, by the same process above, the product 퐷퐸푇 gives us a dyadic covariate that takes the value 1 when both band and venue are located in Liverpool and 0 otherwise.

In the model:

Band Venue Lpool is a dyadic covariate that takes the value of 1 where both band and venue are located in Liverpool. On its own this serves as an indicator for the preference of bands based in Liverpool to choose to perform at venues that are also located in Liverpool.

Band Venue Manc is a dyadic covariate that takes the value of 1 where both band and venue are located in Manchester. On its own this serves as an indicator for the preference of bands based in Manchester to choose to perform at venues that are also located in Manchester.

Taken together these two covariates serve as an indicator of the preference of bands to perform at venues in the city of their location, which we can hypothesise is likely to be the case as we can assume home locations will be more familiar and bands will be able to draw a crowd in their hometown.

6.7 Descriptive Statistics Here I highlight descriptive statistics for each of the networks in discrete years, 2013, 2014 and 2015 for which the data was collected. The simplest measure we can calculate for a network is the density which is calculated as the proportion of all possible ties between nodes in a network that are realised in the observed network. In the underground punk networks, ties are captured as live performances at a given venue in a given year. For bipartite networks, ties are only permitted between the two different node sets and, as SAOMs assume agency in the model, are directed from bands to venues as choices to perform at those venues. In a network of 80 bands and 39 venues then, the maximum number of ties is achieved when every band in the network has performed at every venue in the network in a given year. Node sets of 80 bands and 39 venues are consistent across all three years so 80 × 39 ties to give a total of 3120 possible ties in each network for each year. The observed networks are relatively sparse with 147, 144 and 135 ties in each year 2013-2015 giving densities of 0.047, 0.046 and 0.043 respectively, meaning roughly 4-5% of possible ties in the network are realised in each year. The number of ties and therefore the density remains stably sparse across all three years.

100

The fact that the network is sparse is expected for the band-venue network and may be the result of a number of factors.

The global network structure is dependent on strategies of bands in choosing the venues at which they perform, this choice is not straight forward and may depend on the amount of information at a band’s disposal, which in turn is contingent upon the dual nature of bands including both their collective experience in interaction with an audience and experience and information they retain as individuals. As the bands included in the network all have ties to at least one venue each year, we can take they are established somewhat, at least by the second year and certainly the third year. This means the information at their disposal will differ greatly from brand new bands. New bands may perform at many different venues as they find their feet in the world of underground punk and will most likely be eager to play a large number of gigs but perhaps have fewer resources to play elsewhere in other cities. Established bands, on the other hand are likely to deliberately choose venues that they know will have a high probability of a positive experience.

The next measure of note is the average degree in each network. The degree of an actor in the network, in this case outdegree, as ties are directed from bands to venues only, is simply the number of different venues that a band performs at in each given year. The average degree is the mean of these degree values for every actor in the network. This is also consistent across all three years, with average degrees of 1.837, 1.8 and 1.688 meaning that on average, bands form ties with fewer than two venues in each year. All bands are active across all three years meaning they form ties with at least one venue. The average degree values suggest that some bands form ties with more than one venue whilst others simply choose to perform at only one venue in either Manchester or Liverpool in each year. This information gives us the hypothesis that there is a preference for bands that sustain the underground punk music world in both cities to play at a single venue in each year.

Two measures are possible in the RSiena package to test for the suitability of stochastic actor- oriented models for the analysis of the observed network data, Hamming Distance and Jaccard indices, calculated for the two periods of change, from the 2013-2014 network and from the 2014 -2015 network. Hamming distance is a count of number of tie variable changes in the network between the two periods. Hamming distances were calculated as 172 and 199 for the two periods showing a large number of tie changes. This is somewhat expected as some venues are transient and used only in one year, whilst others host gigs in all three observation periods.

Snijders, van de Bunt and Steglich (2010) note that Jaccard indices are a measure of stability in the network. They note that values less than 0.2 may be problematic. In the live performance 101 networks, the Jaccard indices are 0.254 and 0.167 respectively for periods 2013-2014 and 2014-2015. Although the value for the second period is low, this is due to the sparsity of the network in general with many isolate venues in each observed year. Ripley et al. (2019) note that lower Jaccard indices may not be problematic in very sparse networks where the majority of nodes have degrees lower than 2, which is the case for the punk network. On this basis, if models converge well, then we can deem them unproblematic for interpretation as this sparsity makes sense in the substantive context of underground punk venue choices.

6.7.1 Model Building Procedure Models were estimated using a stepwise procedure where each subsequent model builds on the previously estimated one. Model 1 was estimated first and subsequently, parameter estimates were used as starting values for model 2. Model 3 was estimated using starting parameter values of model 2 and finally model 4 was estimated using starting parameter values of model 3. This decision was made in lieu of the descriptive statistics in order to provide stability as the networks were found to be sparse with low Jaccard scores. Model 1 is attempting to solely explain tie structure and with overall maximum convergence of 0.065 converged extremely well against the recommended figure of less than 0.25. Model 2 introduces an attribute for the location of venues. I wanted to ensure that the model was stable with solely the venue attribute as the aim was for further attributes and dyadic covariates to be interpreted in the context of venue choice. Again, the model converged well with an overall maximum convergence ratio of 0.032. Model 3 introduced the two location attributes for bands and again the model converged well with an overall maximum convergence ratio of 0.055. At this stage, the model had proved stable with the network parameters and nodal attributes, which served as the basis for the calculation of the dyadic covariates in section 6.4 above. This meant that these dyadic covariates could be introduced in model 4 to estimate the final model for interpretation.

Standard errors were computed using the score function method (Schweinberger and Snijders, 2007) and 4000 phase 3 iterations were specified as recommended for publication (Ripley et. al, 2019). Similar to standard statistical models, parameters may be interpreted as significant when the estimates are more than twice the computed standard error values.

102

Table 11 Parameter estimates (standard errors) for a longitudinal model fitted for three observations of the bipartite live performance network of bands and venues in Manchester and Liverpool from 2013 – 2015. Models were estimated using the method of moments (Snijders, 2001, 2007) with 4000 iterations in phase 3. Standard errors were computed using the score function method (Schweinberger and Snijders, 2007). Additional variables were introduced stepwise, where estimates for model 1 were subsequently used as starting values for model 2 and similarly from model 2 to model 3 and model 3 to model 4. Model 1 Model 2 Model 3 Model 4 Rate 2013 - 2014 4.431 (0.553) 4.475 (0.556) 4.507 (0.568) 4.972 (0.661) variables 2014 - 2015 4.728 (0.595) 4.725 (0.596) 4.762 (0.613) 5.376 (0.771) Network outdegree (density) -1.821 (0.092) -1.835 (0.096) -1.849 (0.096) -2.164 (0.121) variables four cycles 0.005 (0.028) 0.005 (0.031) -0.001 (0.033) -0.015 (0.035) indegree popularity 0.049 (0.007) 0.050 (0.007) 0.052 (0.007) 0.062 (0.007) anti in-isolates -3.263 (0.777) -3.275 (0.825) -3.294 (0.861) -3.228 (0.810) indegree at least 2 -0.002 (0.462) 0.037 (0.456) 0.021 (0.494) 0.044 (0.491) Node Venue Manchester 0.068 (0.108) 0.097 (0.110) 0.698 (0.163) attributes Band Liverpool 0.329 (0.140) -1.610 (0.442) Band Manchester 0.311 (0.152) 0.191 (0.270) Dyadic Band Venue Lpool 3.003 (0.470) Covariates Band Venue Manc 0.153 (0.303) Models estimated using RSiena package in open source R statistical software. Overall maximum convergence ratios 0.030, 0.063, 0.070 and 0.069 respectively for models 1, 2, 3 and 4 indicating good convergence of all models. Individual parameter t-ratios were below 0.1 in every case.

6.8 Bipartite SAOM Results From the final results above for model 4, we can see that overall convergence is good, with the maximum overall convergence ratio of 0.058 being well below 0.25 and convergence of estimated parameters was also good, with all t-ratios below 0.1, threshold values recommended for good model convergence by Ripley et. al (ibid.)

6.8.1 Rate Parameters The rate parameters are separated into two periods, the first between 2013 and 2014 and the second between 2014 and 2015. For the final model, model 4, these rate parameters show that there was an average of approximately just under five opportunities for tie changes per band in the network during the first period and an average of over five opportunities for tie changes per band during the second period. It is important again to note that these rate parameters are representations of the approximate opportunity for change and not the observed changes in ties, in fact, the rate parameter will always be larger than the observed number of tie changes. This is because an opportunity for a change in ties does not always result in a change – the band may have the opportunity to perform at a venue but decide not to take the opportunity and form a tie. As such we can only interpret that there were more opportunities for change during the second period from 2014-2015 than during the first period. This is what was

103 expected from difference in scores for Jaccard similarity index, which was lower for the second period, discussed in the earlier section on descriptive statistics, meaning that the network was less stable between 2014-2015 resulting in a higher volume of tie change that was also indicated by the slightly larger hamming distance.

Although standard errors are reported for the rate parameters in the model, these should not be interpreted as indicators of statistical significance as we are measuring the change in networks and a value that is not significantly different from zero would indicate no change in network ties, which is certain not to be the case, therefore an interpretation of this kind is redundant. However, parameters in the second part of the model, the evaluation function, can be interpreted as statistically significant when the parameter values are more than twice the value of the standard error.

6.8.2 Network Parameters As ties are only formed in a single direction, from bands to venues, the negative outdegree parameter simply indicates there is a tendency by bands away from forming a high number of ties to venues. The second parameter for the tendency towards the formation of four-cycle structures is similar across the first three models. In models 1 and 2 the value is positive and in model 3 the value is negative, but in all three cases the value is extremely low, showing close to zero effect as well as no statistical significance, with the standard error being more than twice the parameter estimate. In model 4 however, when nodal and dyadic attributes for locations of bands and venues are controlled for, the value is slightly larger and negative, which in general would indicate a tendency against forming these closure patterns, but again is not statistically significant. Consequently, we can deduce that there does not appear to be evidence of this type of closure over time where pairs of bands make pairs of venue choices in common.

The third network parameter, indegree popularity, is both positive and significant, with an estimate more than eight times the computed standard error. In spite of this, the coefficient value is relatively small, indicating that a very small number of venues become more popular than others over time. It seems as though the venues that are popular at the beginning of the observation data retain that popularity over time and suggests that there is likely to be an element of partisanship in venue choices and that bands may elect to perform at the same venues year after year.

The fourth parameter, anti-in isolates, is large, negative and statistically significant with a standard error approximately four times the parameter estimate. This indicates a tendency away from wanting to establish ties that prevent isolates, as well as not wanting to dissolve ties that would create isolates if they were dissolved and shows that the eighty core bands in the 104 network make deliberate venue choices. The fifth parameter, indegree at least 2, is positive but is not significant in explaining the observed change in network ties. This is likely due to the fact that the average degree statistic, seen in section 6.7 above is just below 2, with some bands forming more than two ties and others choosing to perform at a single venue each year.

6.8.3 Nodal Attributes The binary covariate for venue, which is coded either 1 or 0 depending on whether the venue was located in Manchester or Liverpool respectively is positive and statistically significant with a parameter estimate of more than four times its standard error. This shows a preference in general for bands to form more ties with venues in Manchester than in Liverpool over the three years.

The first and second nodal attributes represent two dummy indicators for band location, with the reference category referring to bands that are not located in either Manchester or Liverpool. The first of these, band Liverpool, has a relatively large negative value and is highly significant with a parameter estimate more than four times its standard error. This shows that in general that bands based in Liverpool have a strong adverse preference for new tie formation in general compared to bands from neither Liverpool nor Manchester in the network. The second dummy indicator, band Manchester, is much smaller in magnitude and positive, but is not significant with a standard error larger than its parameter estimate. This shows that bands in Manchester are slightly more prone to forming new ties than bands not located in either city in the network, but not significantly more prone to this behaviour. Taking these two dummy variables together then, because we are controlling for all three categories of band location, we can also see that Liverpool bands behave much differently to Manchester bands. Whereas Manchester bands show a slight preference for forming new ties, Liverpool bands are somewhat against it. The mechanisms behind these preferences may be explained by the two dyadic covariate parameters, which I discuss below.

6.8.4 Dyadic Covariates The two dyadic covariate parameters serve as an indication of homophily (McPherson et al, 2001) of choice by bands to perform in the city in which they are located. The first of these parameters, band venue L’pool, has a large positive value and is highly statistically significant with a parameter estimate that is more than six times its standard error. This indicates that bands based in Liverpool overwhelmingly prefer to form ties with Liverpool venues over Manchester venues. The second dyadic covariate, band venue Manc, is positive but not statistically significant with a standard error larger than its parameter estimate. It does indicate a slightly larger number of ties that Manchester bands formed with Manchester venues than

105

Liverpool venues but as it is not statistically significant, we cannot infer a preference in general.

If we consider the model as a whole we can say that the network is centred around a relatively small number of venues that bands repeatedly choose to perform at. These venues gain slightly more bands over time, indicated by a small positive indegree popularity parameter but generally venues that are foci in 2013 are foci in 2015, with the exception of MelloMello that operated as a focus for two years before closing its doors at the end of 2014. Controlling for geographical location of bands, as well as the preference for bands to form ties with homophilous venues, we can see that new ties tend to be formed with venues in Manchester more so than Liverpool venues. Controlling for band and venue location in Manchester, we can see that bands in Liverpool are perhaps more partisan to their city of origin, choosing to perform at venues that are based in Liverpool whilst at the same time forming few new ties overall, showing a renewed commitment each year to a single venue.

6.9 Discussion Overall the networks are relatively sparse with little overlapping between bands and a high degree of clustering around particular venues over a three-year period. Model results show that venue choice is deliberate, and venues are chosen by key bands in the network above a simply random choice. We can see from the small but significant indegree popularity effect that a small number of venues become foci over the three years in question. These deliberate venue choices are made on the basis of the interaction between bands and their audiences in the social space of the venues. Some venues are ‘tried out’ and then simply forgotten, whilst others are chosen repeatedly over three years by a key group of consistent bands.

Performances have to take place in a given year for a tie to exist, they are not simply accumulated over time and so key venues are shown to be reaffirmed over three years. From a social capital point of view, bands look to draw the largest number of fans possible to their gigs, a bigger crowd usually makes for better gig, provided there is a receptive audience. In order to draw bigger crowds, bands may make the decision to perform at fewer events in order to make their appearances rarer and therefore sought after for audiences. I contend that tie formation is also shaped by the dual interactions of bands as collective groups and bands as individuals with their audience. When bands as collectives and as individuals have a positive experience performing in a venue to a captive audience they will wish to perform again in that same environment and after multiple visits become familiar with it. They know what time to arrive, what equipment they need to bring, how the band will sound and what the audience expect. They are also privy to the reactions of audiences to other bands and the interplay between bands and audiences in shaping conventions and pushing the boundaries of what 106 audiences expect to hear. All of these factors motivate bands to perform at their best and therefore be more likely to please audiences and attract more fans by word of mouth and in doing so reinforce the cycle of increasing their social capital.

This becomes more difficult to achieve when different venues are tried out, they may be tested for one gig or a small number and then fall by the wayside. This may be due to a change of management at a previous venue or simply that touring bands want to fit a gig into their touring schedule, requiring a new venue to be sourced when key venues are unavailable. This situation breaks established conventions for bands that are used to the way an event is run at the key venues that are used regularly. Everything from room hire, equipment provided or required, sound tech, room sound, curfew, age limits, capacity, accessibility, availability of parking, proximity to public transport may be unknown or uncomfortable. Something as simple as drinks prices that are too high may put the audience off from attending. All of these factors lead to an unfamiliar environment where everyone is more unsure how to act and interact in the social space, which can lead to a frosty reception for the bands in the awkwardness of the environment. Venues that are familiar with established conventions, lead to a dynamic interplay between bands and their audiences and this in turn affords bands a safe space to experiment with their sound and break and reform conventions around what audiences expect to hear.

We have to consider that the bands in this network are consistently performing in the network over three years and that 2013, as the initial observation year is not the genesis of a new musical movement, rather I capture a snapshot in time and as such, I assume that these eighty core bands are somewhat established. For new bands, the same result may not hold true as they seek out as many opportunities as possible to perform as many times as possible in whichever venue they can find. As bands become established, however, they are much more likely to choose venues where they have established themselves and benefitted from positive experiences that have allowed them to prosper, however they might define that, which could be in terms of a large audience that responds positively, the venue that provides the resources the band needs, they are paid well, they manage to sell merchandise or that the venue sonically sounded pleasing to them. I contend that as established bands they are unlikely to perform more than once in a venue if they have a negative experience as they will not want to waste time and effort on what they will predict to be a bad gig. This contrast between new and established bands has the potential to be explored in further work but more data would need to be collected for this purpose.

In addition, through this reaffirmation of key venues over three years from 2013-2015, I have shown that the structure of live performance worlds of Manchester and Liverpool are very

107 different in relation to where bands are geographically based. Liverpool bands tend to strongly prefer to play in Liverpool over Manchester, this may be an effect of the desire to perform to a familiar and receptive audience, or it may be due to past experience performing in venues in the city that have been deemed to be unsuccessful. This is coupled with the fact that Liverpool bands also are strongly predisposed not to form new ties with venues in general. From my own experience as a participant in this world, I can say with some authority that there is an almost intangible difference if atmosphere and conventions of behaviour and expectation of audiences in Manchester and Liverpool. Manchester, it seems, is much more structured with venues that tend to host certain pre-defined genres of music and promoters that will not entertain certain bands if their style does not fit the show they are putting on and as a result the feeling tends to be that these gigs are almost curated, in much the same way that Finnegan (1989) tends to refer to venues as having established conventions of style and genre that folk musicians have to match up with. Liverpool, on the other hand, often boasts much more, at face value, eclectic lineups for gigs and audiences tend to be open to a myriad of different styles of music, judging bands on quality rather a predisposition for certain types of music.

This again is interesting from a social capital point of view, bands located in Manchester and Liverpool appear to employ different strategies when choosing to perform at venues in either city; if we make the assumption that these established bands are attempting to maximise their chance of a successful gig. To unpick this a bit more, we can look at the nodal attributes in the model. Controlling for everything else, in general, there is a tendency towards forming more ties with Manchester venues, shown by the positive and statistically significant values for the indicator for venues located in Manchester. The final parameter in the model, however, band venue Manc, is not statistically significant, explaining that Manchester bands will not necessarily choose to perform at venues in Manchester. This shows us that in general that Manchester venues must be chosen more often than not, over Liverpool, by touring bands and the Manchester core bands that sustain the Manchester world are a mixture of Manchester based and touring bands. We might say that this provides the Manchester bands with social capital as they are able to interact with bands located further afield and subsequently acquire valuable information and contacts from venues in other cities in the UK that are in turn the best places to play in those cities, allowing less trial and error to have to take place when travelling further afield. Liverpool bands that sustain the underground world, however, seem to be content with performing in their home city. It could be argued that this would put these bands at a disadvantage in terms of their social capital. However, if the purpose of venue choice is to perform at a successful and enjoyable gig to an appreciative audience then perhaps this is not a disadvantage at all.

108

From what I have discussed above the fact that touring bands seem to choose Manchester more often than Liverpool may be related to the idea of social capital and mutual contact and gig exchange between the Manchester and Touring bands. It may be that if Liverpool bands do tend to stay in Liverpool, they are less likely to meet people and spread the word about venues in Liverpool. This type of reciprocal exchange is difficult to pin down here without further research as there are a number of other factors at play including transport links; Manchester is more easily accessible from most places in the UK than Liverpool as it lies close to several major motorways. It may be that Liverpool bands have not faired well in Manchester and this information has been relayed to other core Liverpool bands or they may have had a similarly negative experience. The boundaries of the network are also relatively small as many bands seek gigs further afield travelling South towards Birmingham and London or North towards Leeds and beyond.

In terms of conventions, I argued at the end of chapters two and five that bands had the opportunity to choose venues and that these venues allow the bands to form and reform conventions around the way their music sounds in their interactions with an audience. This is undoubtedly true when we consider the interaction between established and brand new bands as it is likely to be the brand new bands that shape new sounds and are the ‘sources of novel information’ that Granovetter (1973) and Burt (1992) speak lengthily about in their work. This network of core bands, however, may operate differently. Bands appear to be repeatedly choosing to perform at the venues in which they feel comfortable, where they are used to their surroundings and if they are consistently performing at the same venues, we can imagine that they perform to regular audiences that are fans of theirs. When all the mundane conventions are taken care of, this might afford the bands opportunities to push at the boundaries of the established conventional sounds that their audiences expect to hear as they are in a safe environment, which perhaps affords them the opportunity to make mistakes. It may be that Manchester bands are more constrained in this way than Liverpool bands, in that if they are relying on a reciprocal exchange mechanism in order to book gigs in other cities then they may wish to ‘play it safe’ and impress touring bands by playing songs they know they are able to perform to a high standard and that audiences will respond to. Liverpool bands, on the other hand, may be more willing to take risks as they are performing more often than not to a ‘home crowd’ with less at stake than the Manchester bands, which affords them the opportunity to be more innovative, which may link back to what I mentioned earlier about Liverpool audiences and line-ups of bands at gigs being more eclectic and less curated than in Manchester.

109

Overall the stochastic actor-oriented model has highlighted the inner workings of the music worlds of Manchester and Liverpool, showing that they are sustained by a group of core bands and venues that are the foci of the worlds. From the brief discussion above, I have brought to light several facets of this world, strategies of bands surrounding social capital, reciprocal gig exchange and the dynamics of conventions across geographical locations of both bands and venues that could be followed up in future research. In the following chapter I look to draw conclusions on the thesis as a whole.

110

Chapter Seven: Conclusion 7.1 Introduction As a discipline, social network analysis has a rich body of both methodological and substantive literature on a wide range of subject areas from school friendships to disease transmission, academic co-publication to interlocking directorates, social movements to music worlds among a whole host of others. This thesis is positioned at the intersection of growing substantive interest in the social networks of music worlds (Crossley, 2008,2009, 2015a, 2015b) and a thriving community of academics pushing for theoretical and methodological advancements in the analysis of network data, in particular networks with multiple types of nodes and statistical modelling of social networks. The case of the underground punk music world of Manchester and Liverpool has provided a unique context for the investigation of tripartite network structures, an area which has largely been overlooked in the literature and that certainly lacks empirically tested research with real-world datasets. It has also provided the opportunity to introduce complex statistical modelling in tandem with theory on music worlds to enhance the future body of work influenced by Becker’s (1974, 1976, 1982) Art Worlds theory and subsequently Crossley’s (ibid.) work on social networks of music worlds.

The aim of this thesis was to uncover the structure of live performance in the underground punk music world in Manchester and Liverpool, two major cities in the North West of England over a period of three years from 2013-2015. It has succeeded in bringing together key theories on music worlds and duality in networks as a foundation for the exploration of the importance of venues in a tripartite network of musicians, bands and the venues at which those bands perform and has shown that the development of new tripartite methods allows us to ask complex questions around dependencies among different node sets in a network, with particular emphasis on tripartite closure. It has further investigated the critical importance of venues in a dynamic network and the nature of venue choice in an underground music world where conventions are dynamically formed and reformed in social space. Below I recap how each of the chapters of the thesis has achieved this aim.

7.2 Summary of Chapters Chapter One introduced the research problem, outlining the reasons for investigating the underground punk worlds of Manchester and Liverpool, introducing the key concepts of punk, music worlds and social network analysis as well as a brief overview of my own experience of punk as a participant. It also identified the lack of attention in the social networks literature given to tripartite networks and more generally, networks with nodes of more than two types, in conjunction with the lack of attention given in the substantive literature on music worlds to the importance of venues as foci that sustain music worlds and 111 provide the opportunity for bands, in their dual identities as individuals and groups, to obtain social capital and form and reform conventions in social space. It posited four original contributions to knowledge to stimulate debate among both network methodologists and substantive researchers interested in both music worlds as well as other cultural sociological domains.

Chapter Two reviewed literature on punk music, highlighting how much focus has been given to the idea of subculture, cultural consumption and aesthetics with punk often positioned as a form of youthful resistance against political and social restraint that emerged from relative obscurity. It then moved on to discuss Crossley’s (2008; 2009; 2015) conceptions of the emergence of punk and post-punk as a critical mass of participants operating in a network of relational ties with a focus on the production of music and subsequently defined four key facets of music worlds; collective activity, conventions and resources, highlighted by Becker (1974, 1976, 1982); as well as venues, often neglected in academic research on music worlds; in the context of the underground punk music world. Finally, it highlighted the principally qualitative body of literature on music worlds as well as more recent academic research using network-focused approaches to advocate for further development of the network approach through introducing the use of statistical models to the music worlds approach.

Chapter Three gave a comprehensive overview of the large body of literature on network theory, beginning with ego-networks and why they were inappropriate for the exploration of the underground punk music world and that whole network analysis should be used instead to fully capture the true network structure. It then moved to systematically highlight and explain network topologies, beginning with unipartite networks and explaining that whilst many affiliation networks are often projected to form these types of network, we should aim to respect the original data structure where possible to avoid loss of information. In light of this, bipartite networks were explained in tandem with Breiger’s (1974) concept of duality with reference to recent literature as well as highlighting methods to analyse these types of network directly. Subsequently, the sparse body of literature on tripartite methods was discussed, noting efforts by Fararo and Doreian (1984), Borgatti and Everett (1992) and Bohman (2012) to formalise the concept of analysis of network with nodes of more than two types, going on to explain the critical importance of venues in networks of music worlds and advocating the use of tripartite methods to adequately capture closure mechanisms between musicians, bands and venues.

Chapter Four introduced the methodological concerns and the research design of the thesis. It explained the operationalisation of the music worlds concept developed in chapter two alongside Breiger’s (1974) notion of duality in the tripartite context set out at the end of

112 chapter three, positing that the focus on live performance in this underground punk setting should include careful consideration of venues, arguing that venues are particularly significant in the shaping of a band’s sound at this relatively small level of operation. Further to this, possible mechanisms behind venue choice were discussed, hypothesising that band membership may drive decisions by bands to perform at certain venues, analysed in chapter five, as well as hypothesising positive interaction between a band’s location and the location of the venue in either Liverpool or Manchester, analysed in chapter six. Subsequently, research aims were defined to highlight key contributions of the thesis to both the substantive understanding of music worlds and the methodological analysis of these worlds as well as tripartite networks more generally, leading to four key research questions considering tripartite closure and the role of venues. After defining the research questions, data collection was discussed elaborating on the systematic collection of network data from Facebook event pages to construct a raw dataset of 725 unique bands performing across 357 gigs at forty-one unique venues over three years. The boundaries of the network were discussed as well as possible imitations of the dataset before a brief summary of the data including descriptive statistics and sociograms was presented.

Chapter Five was the first of two key chapters of empirical analysis using statistical network models. It introduced exponential random graph models (ERGMs) recently developed for social network analysis (Robins, 2007; Lusher et.al, 2013) and subsequently specified its bipartite extension (Wang et al, 2009). Two bipartite ERGM models were then fitted to analyse two separate parts of the larger tripartite network, the musician-band network and the band-venue network. A novel method was then proposed to test for the prevalence of tripartite closure in the network and to assess whether multiple membership of bands was a driver of closure patterns. The hypothesis that multiple band membership was driving closure did not hold and the earlier bipartite analysis, in the case of this network, was sufficient for the explanation of tripartite patterns, in particular the band-venue network, where statistically significant parameters were found. It was concluded that subsequent attention should be directed towards the larger band-venue network to investigate further the critical role of venues.

Chapter Six was the second key empirical chapter and focused attention on the critical role of venues in the underground punk music worlds of Manchester and Liverpool from 2013-2015 and the choices that a relatively small number of key bands make to play at a small number of key venues, creating a core network of live performance activity that sustains this underground music world. It brought attention to the importance of network dynamics in music worlds and how choices by bands to perform at certain venues can be captured using longitudinal models

113 of network activity. Discrete-time temporal exponential random graph models (TERGMs) were acknowledged to have issues with interpretation (Block et. al, 2018) and consequently, stochastic actor-oriented models (Snijders, 2010) for the analysis of dynamic social networks were introduced, followed by their bipartite extension proposed by Koskinen and Edling (2012). A model was estimated that showed the music worlds of Manchester and Liverpool were sustained by a small number of key venues, whose status as foci (Feld, 1981) in the network is reaffirmed each year as bands repeatedly choose to perform at them. It was shown that bands located in Manchester and Liverpool vary in their strategies when choosing venues, with bands from Liverpool found to have a strong preference for performing at venues in Liverpool and bands from Manchester more agnostic about venue location, hypothesising that social capital may be gained through different strategies of venue choice for different bands located in different cities.

7.3 Discussion

I set out in this thesis to answer four key research questions:

1. Is there a tendency towards tripartite closure in the network of live performances in Manchester and Liverpool?

2. Do overlapping band-membership ties drive the formation of band-venue ties in tripartite closure?

3. What role do venues play in the underground punk world of Manchester and Liverpool?

4. How does the role of venues play out over a three-year time period from 2013-2015?

The first and second research questions were investigated empirically in chapter five, analysing a sample network of forty-four musicians, sixteen bands and fourteen venues. I specified three-path and four-cycle configurations and hypothesised that tripartite closure would occur when pairs of bands with one member in common perform at the same venue in a given year. After modelling the two constituent sub-networks comprised of musician-band ties and band- venue ties respectively, I combined the GOF simulated distributions to produce joint predictive distributions for three-path and four-cycle tripartite configurations, finding that observed networks lay in the centre of these distributions, indicating that there was no apparent tendency towards tripartite closure, as the number of observed configurations was not significantly different from randomly simulated networks. From this result we can conclude that there is no tendency towards tripartite closure between musicians, bands and venues in this sample of core bands in the year 2015 but it would be erroneous to conclude

114 that this is the case across the core network for all three years or for any larger networks with differing nodes and structures. We can say the sample indicates that this is likely to be the case for a larger network perhaps but extending this to different music worlds in other cities or other time periods would be a false claim, further investigation and new data would be needed. Further conclusions may be drawn around dynamic closure patterns if more data could be collected on band membership. This would require new data to be collected as Facebook is a static data source that provides details of current band members. Band membership can be transient and membership or band composition, the instrumental make- up of the band, may shift between time periods so this would have to be collected either via primary data collection methods for historic data, or on an ongoing basis over time if new data on future gigs was recorded from Facebook. Dynamic data would allow more robust testing of this tendency towards tripartite closure.

The second research question has proved more difficult to answer. Empirically, for the 2015 network, overlapping band ties do not drive tripartite closure in the network. However, in general, there remains a rich body of theory to support this idea that overlapping band membership would drive tripartite closure through resource sharing and social capital mechanisms. Again, dynamic investigation of a longitudinal network would provide powerful test of this hypothesis. It may also be the case that band members with certain attributes, such as the instrument they play, the amount of free time they have or even if they have access to a car or not, may dictate when tripartite closure occurs. As ever, inclusion of further data types or configurations would lead to more complex models, but the basis for analysis I have provided is sound and could in principal be extended to incorporate these more complex configurations.

With regards to research questions three and four, I feel I have managed to define the role of venues in the underground punk network in a robust manner theoretically, adopting a music worlds approach to explain that venues provide social spaces that become foci of a world. They are spaces where bands perform as dual entities and simultaneously project themselves and are perceived as a whole group and as separate individuals. Conventions are dynamically shaped in venues and subject to evolving change, which in turn will impact dual experiences of band members and bands in the world, perhaps leading to new band memberships and choices to perform at certain venues. These choices that may be made in the venue itself or may be made afterwards; interactions in a venue inform future tie creation and dissolution. Venues sustain the world by providing bands with a place to perform live, the principal purpose of being in a punk band and for audiences to observe and listen in ways that satisfy expected norms embodied by conventions or disrupt these norms and provide a space for

115 innovation in these interactions between bands and audiences. In addition to developing this idea theoretically, I have also shown empirically in chapter six that choices of venues by a core group of bands are reaffirmed over time, with a small number of venues established as the foci of this underground punk world as bands choose to perform there year after year. I managed to collect a comprehensive dataset from Facebook event pages, which captures the world I set out to analyse and explain. This dataset was reduced in order to assess patterns of venue choice by a cohort of core bands that were active in all three years 2013, 2014 and 2015. Using robust stochastic actor-oriented models I showed that a small group of core venues are foci for the core bands. This empirical analysis gave rise to interpretations of differing strategies for different bands located in different cities. These interpretations generate further hypotheses that can be empirically tested if more primary, qualitative data is collected in an effort to establish a more nuanced understanding of patterns of venue choice by certain groups of bands. This empirical analysis has provided a robust platform for further investigation in this area.

In addition to answering the four research questions above, this thesis has expanded current thinking on music worlds to argue for the importance of venues as a key factor in sustaining music worlds alongside collective activity, conventions and resources. I have shown that venues that are chosen by core bands in a network, are reaffirmed in their subsequent choices each year and posit that the mechanisms behind these choices are a function of collective activity, conventions, resources, the dual nature of bands and the social spaces of venues, bringing together key theories of Art Worlds (Becker, 1974; 1976; 1982) and duality in networks (Breiger, 1974) in the process. Substantively, I shown that the structure of the worlds in Manchester and Liverpool appear to be very different and both have the potential to afford opportunities and impose constraints on bands in terms of social capital. Differing choices by bands in Liverpool, Manchester and from outside these cities reveal that there are differing strategies that lead to different social capital in different ways for different bands. Whilst Liverpool bands repeatedly choose venues in their hometown, Manchester based bands are more agnostic about where they perform. The core of the Liverpool world is sustained by Liverpool-based bands whereas the core of the Manchester world is sustained mainly by touring bands. Liverpool bands may benefit from a kinder audience that allows them to experiment more with ‘unconventional sounds’ by the same token they may expect to hear other bands to push the boundaries of what is currently popular. On the other hand, Manchester bands are more likely to perform with touring bands and so this may restrict their ability to reform conventions around how their music sounds as they may want to give the best account of themselves in order to gain social capital with the touring bands and partake in reciprocal exchange of contacts and gig information to enable them to perform in other cities. 116

In developing my own conception of music worlds, I have extended the work done by Crossley (2008, 2009, 2015a, 2015b) and others (O’Shea, 2015; McAndrew and Everett, 2015a, 2015b; Emms and Crossley, 2018) by advocating for the pivotal role of venues as well as that conventions are dynamically formed and reformed in social space. In addition to the substantive contribution, I have pushed the boundaries of music worlds research methodologically through the use of bipartite and tripartite networks as well as exponential random graph and stochastic actor-oriented models.

In focussing on musicians, bands and venues, I have developed a novel method for analysis of tripartite networks, allowing the investigation of tripartite closure structures and the drivers behind them. Although it was revealed that band-membership did not drive tripartite closure in my sample network, it serves as a basis for further work and provides an important step towards the modelling of complex networks with multiple node types, an area of great interest currently within social network analysis, expanding greatly on earlier work by Fararo and Doreian (1984) and Borgatti and Everett (1992), expanding work by Diani (2015) in highlighting the importance of duality and complex structures in cultural networks and Bohman (2012) in tripartite structures and lost data by providing a robust statistical method to test for the prevalence of tripartite structures and drivers of four-cycle closure.

7.4 Further Implications

It has been shown that venues are of critical importance to the underground punk music world in Manchester and Liverpool and that although the number of venues remains relatively stable, the music world is sustained by a comparatively small percentage of all available venues. These are the venues that become the foci of this local punk world; that allow punk musicians to foster creativity and to expand and develop their sounds, to gauge reactions from a regular and attentive crowd and to establish and change conventions in both the way the music sounds and the way it is performed. In my own work I hope to further test the conclusions in this thesis through collection of data for subsequent time periods as well as for other major cities in the UK. It is hoped that this research will also spark debate about the roles of venues in the formation, continuation and reformation of conventions as well as their importance to sustaining an underground music world. The differing strategies of bands in both Manchester and Liverpool may afford them each social capital in different ways and this is another potential area for further investigation. As I mentioned in the introductory chapter to this thesis, bands rarely make any money from performing in the underground punk world and consequently, many factors will be at play in how social capital is viewed by different bands and how they navigate the choices of performances at certain venues as dual entities comprised of persons and groups (Breiger, 1974). 117

It is hoped that the contribution made by this thesis will encourage other researchers of social networks to push the boundaries of methods for the analysis of networks with nodes of more than two types and advocate that where these structures exist, we should make best attempts to analyse them in their original form. In this regard, I hope to continue working on methods and further investigation around tripartite closure patterns, for instance, cumulative advantage in tripartite closure that was mentioned at the end of chapter five. It is also imperative that more network data is collected that is suitable for tripartite analysis in order to provide datasets for empirically testable models to be developed.

Finally, it is hoped that scholars of music worlds will begin to incorporate statistical models into their research and that by doing so they can both ask more complex questions of their data, but also to retain substantive sensibility in the interpretation of such models.

.

118

Notes

1. Green Day Platinum reference: https://www.riaa.com/gold-platinum/certification- criteria/ 2. For a brief overview of fanzine culture see Martin, 2016. For an example of a modern online punk fanzine see: https://shout-louder.com/ 3. For a more in-depth explanation of the ‘fields’ versus ‘worlds’ debate, see Bottero and Crossley, 2011. 4. For a comprehensive overview of recent thinking on relational sociology and its position in the discipline see Dépelteau, ed., 2018. 5. In order to illustrate an extreme case of absence of conventions, Becker refers to composer Harry Partch, 1949 (cited in Becker, 1982, p.32-33) who would recruit university students at the start of the academic year to build instruments he had invented and learn unique notation he had devised, culminating in a live performance at the end of the year, a significantly longer time than it would take musicians that shared the conventions of Western classical music to be able to perform a new piece together. 6. More detail on the mechanisms behind tie formation is given in chapter 3. 7. For a general introduction to formal SNA concepts see Wasserman and Faust, 1994; Borgatti, Everett and Johnson, 2013; Scott and Carrington, 2011. 8. Borgatti, S.P., Everett, M.G. and Freeman, L.C. 2002. Ucinet 6 for Windows: Software for Social Network Analysis. Harvard, MA: Analytic Technologies. 9. R Core Team, 2019. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/. 10. A library of commands for R software that implements the statistical network models for network dynamics specified in Snijders, 2017. 11. Wang, P., Robins, G., Pattison, P., 2009. MPNet: program for the simulation and estimation of exponential random graph models. Melbourne School of Psychological Sciences, The University of Melbourne. 12. For a comprehensive overview of multilevel analysis for social networks see Lazega, E. and Snijders T.A.B., 2016. Multilevel Network Analysis for the Social Sciences. London: Springer 13. For an introduction to multilevel networks see Lomi, A., Robins, G. and Tranmer M., 2016. Introduction to multilevel networks. Social Networks, 44, 266-268.

119

References ABRSM, 2014. Making Music: Teaching, learning & playing in the UK. [online] Available at: [Accessed 20th October 2017]. Agneessens, F. and Roose, H., 2008. Local Sructural Properties and Attribute Characteristics in 2-mode Networks: p* Models to Map Choices of Theater Events. The Journal of Mathematical Sociology, 32, pp.204-237. Barabasi, A. and Albert, R., 1999. Emergence of Scaling in Random Networks. Science, 286(5439), pp.509-512. BBC, 2018. Pubs closing at rate of 18 a week as people stay at home. [online] Available at: Becker, H.S., 1974. Art As Collective Action. American Sociological Review, 39(6), pp.767-776. Becker, H.S., 1976. Art Worlds and Social Types. American Behavioural Scientist, 19(6), pp.703- 718. Becker, H.S., 1982. Art Worlds. Berkeley, CA: University of California Press. Becker, H.S., 2004. Jazz Places. In: Bennett, A. and Peterson, R.A. eds., 2004. Music Scenes: Local, Translocal and Virtual. Nashville, TN: Vanderbilt University Press. Beer, D., 2014. Punk Sociology. London: Palgrave Pivot. Bennett, A., 2000. Popular Music and Youth Culture. London: Macmillan. Bennett, A. and Peterson, R.A. eds., 2004. Music Scenes: Local, Translocal and Virtual. Nashville, TN: Vanderbilt University Press. Bennett, H. S., 1980. On Becoming a Rock Musician, Amherst, MA: University of Massachusetts Press. Blank Generation: Punk Rock, 2007. Seven Ages of Rock. [TV programme]. BBC, BBC2, 2nd June 2007. Block, P., Koskinen, J., Hollway, J., Steglich, C. and Stadtfeld, C., 2018. Change we can believe in: Comparing longitudinal network models on consistency, interpretability and predictive power. Social Networks, 52, pp.180-191. Blush, S., 2001. American Hardcore: A Tribal History. , Feral House. Blush, S., 2006. American Hardcore [DVD]. Culver City: Sony Pictures Entertainment Bohman, 2012. Bringing the owners back in: An analysis of a 3-mode interlock network. Social Networks, 34(2), pp.275-287. Bottero, W. and Crossley, N., 2011. Worlds, Fields and Networks: Becker, Bordieu and the Structures of Social Relations. Cultural Sociology, 5(1), pp.99-199. Borgatti, S.P. and Everett, M.G., 1992. Regular blockmodels of multiway, multimode matrices. Social Networks, 14, pp.91-120.

120

Borgatti, S.P. and Everett, M.G., 1997. Network analysis of 2-mode data. Social Networks, 19(3), pp.243-269. Borgatti, S.P., Everett, M.G. and Freeman, L.C., 2002. Ucinet for Windows: Software for Social Network Analysis. Harvard, MA: Analytic Technologies. Borgatti, S.P., Everett, M.G., and Johnson, J.C., 2013. Analysing Social Networks. London: Sage. Borgatti, S.P. and Halgin, D. 2011. Analyzing Affiliation Networks. In: Carrington, P. and Scott, J. eds., 2011. The Sage Handbook of Social Network Analysis. Sage. Bourdieu, P., 1984. Distinction: a social critique of the judgement of taste. London: Routledge. Bourdieu, P., 1993. The Field of Cultural Production: Essays on Art and Literature. Cambridge: Polity. Breiger, R.L., 1974. The Duality of Persons and Groups. Social Forces, 53(2), pp.181-190. Burt., R.S., 1976. Position in networks. Social Forces, 55, pp. 93-122. Burt, R.S., 1980. Innovation as a structural interest: Rethinking the impact of network position on innovation adoption. Social Networks, 2, pp.327-355. Burt. R.S., 1995. Structural Holes: The Social Structure of Competition. Cambridge, MA: Harvard University Press. Cadwalladr, C. and Graham-Harrison, E., 2018. Revealed: 50 million Facebook profiles harvested for Cambridge Analytica in major data breach. Guardian [online] 17 March. Available at: Chapple, J., 2018. Live Music Revenues to Top $30bn for First Time. [online] Available at: Chick, S., 2009. Spray Paint the Walls: The Story of Black Flag. London: Omnibus Press. Clark, J., Hall, S., Jefferson, T. and Roberts, B., 1976. Subcultures, Culture and Class: A theoretical overview. In: Hall, S. and Jefferson, T. eds., 1976. Resistance Through Rituals: Youth subcultures in post-war Britain. London: Hutchinson & Co. Cohen, S., 1991. Rock Culture in Liverpool: Popular Music in the Making. Oxford: Clarendon Press. Coleman, J.S., 1988. Social Capital in the Creation of Human Capital. American Journal of Sociology, 94, pp.S95-S120. Constine, J., 2019. Facebook pays teens to install VPN that spies on them. TechCrunch [online] 29 January. Available at: . Crossley, N., 2008. Pretty Connected: The Social Network of the Early UK Punk Movement. Theory, Culture & Society. 25(6), pp.89-116. Crossley, N, 2009. The man whose web expanded: Network dynamics in Manchester's post/punk music scene 1976–1980. Poetics, 37(1), pp.24-29. Crossley, N., 2011. Towards Relational Sociology. London: Routledge.

121

Crossley, N., 2015a. Networks of Sound, Style and Subversion: The Punk and Post-punk Worlds of Manchester, London, Liverpool and Sheffield, 1975-80. Manchester: Manchester University Press. Crossley, N., 2015b. Totally Wired: The Network of Structure of Post-Punk Worlds of Liverpool, Manchester and Sheffield, 1976-1980. In: Crossley, N., McAndrew, S. and Widdop, P. eds. 2015. Social Networks and Music Worlds London: Routledge. Crossley, N., Bellotti, E., Edwards, G., Everett, M.G., Koskinen, J. and Tranmer, M., 2015. Social Network Analysis for Ego-nets. London: Sage. Crossley, N. and Bottero, W., 2015. Social Spaces of Music: Introduction. Cultural Sociology, 9(1), pp. 3-19. Crossley, N., McAndrew, S. and Widdop, P., 2015. Social Networks and Music Worlds. London: Routledge. Davis, A., Gardner, B., and Gardner, R., 1941. Deep South. Chicago: University of Chicago Press. De Solla Price, 1976. A general theory of bibliometric and other cumulative advantage processes. Journal of the Association for Information Science and Technology, 27(5), pp.292-306. Della Porta, D. & Diani, M., 1999. Social Movements: An Introduction Della Porta, D. & Diani, M., 2015. The Oxford Handbook of Social Movements. Oxford: Oxford University Press Dépelteau, ed., 2018. The Palgrave Handbook of Relational Sociology. London: Palgrave Macmillan. Diani, M., 2015. The Cement of Civil Society: Studying Networks in Localities. Cambridge: Cambridge University Press. Diani, M. and Kousis, M., 2014. The Duality of Claims and Events: The Greek Campaign Against the Troika’s Memoranda and Austerity, 2010-2012. Mobilization: An International Quarterly, 19(4), pp.387-404. Diani, M. and Mische, A., 2015. Network Approaches and Social Movements. In: Della Porta, D. and Diani, M., 2015. The Oxford Handbook of Social Movements. Oxford: Oxford University Press. Edensor, T., Hepburn, P. and Richards, N., 2015. The enabling qualities of Manchester’s open mic network. In: Crossley, N., McAndrew, S. and Widdop, P., 2015. Social Networks and Music Worlds. London: Routledge. Emms, R. and Crossley, N., 2018. Translocality, Network Structure and Music Worlds: Underground Metal in the United Kingdom. Canadian Review of Sociology, 55(1), pp. 111-135. Engels, F., 2009. The condition of the working class in England (New ed. / introduction by Tristram Hunt. ed., Penguin classics). London: Penguin. Erdős, P, Rényi, A., 1959. On Random Graphs. Publicationes Mathematicae Debrecen, 6, pp.290- 297. Everett, M.G. and Borgatti, S. P., 2013. The dual-projection approach for two-mode networks. Social Networks, 35(2), pp.204-210.

122

Fararo, T. J., &Doreian, P. (1984). Tripartite structural analysis: Generalizing the Breiger- Wilson formalism. Social Networks, 6(2), 141-175. Feld, S.L., 1981. The Focused Organization of Social Ties. American Journal of Sociology, 86(5), pp.1015-1035. Feld, S. 1982. Social Structural Determinants of Similarity Among Associates. American Sociological Review, 47(6), pp.797-801. Finnegan, R., 1989. Hidden Musicians: Music-Making in an English Town. Cambridge: Cambridge University Press. Frank, O. and Carrington, P.J., 2007. Estimation of Offending and Co-offending Using Available Data with Model Support. Journal of Mathematical Sociology, 31, pp.1-46. Frank, O. and Strauss, D., 1986. Markov Graphs. Journal of the American Statistical Association, 81(395), pp.832-842. Gelder, K. and Thornton, S., 1997. The Subcultures Reader. London: Routledge. Gilmore, S., 1987. Coordination and Convention. Symbolic Interaction, 10(2), pp.209-227. Glasper, I., 2010. Armed with Anger: How UK punk survived the nineties. London: Cherry Red Books. Glasper, I., 2012. Trapped in a Scene: UK hardcore 1985-1989. London: Cherry Red Books. Glasper, I., 2014a. The Day the Country Died: A history of arnarcho punk 1980-1984. Oakland, CA: PM Press. Glasper, I., 2014b. Burning Britain: The history of UK punk 1980-1984. Oakland, CA: PM Press Gordon, D. and Pantazis, C., 1997. Breadline Britain in the 1990s. Bristol: Summerleaze House Books. Granovetter, M.S., 1973. The Strength of Weak Ties. American Journal of Sociology, 78(6), pp.1360-1380. Hall, S. and Jefferson, T. eds., 1976. Resistance Through Rituals: Youth subcultures in post-war Britain. London: Hutchinson & Co. Hammou, K., 2015. Between social worlds and local scenes: patterns of collaboration in francophone rap music. In: Crossley, N., McAndrew, S. and Widdop, P., 2015. Social Networks and Music Worlds. London: Routledge. Hanneke, S., Fu., W and Xing, E.P., 2010. Discrete temporal models of social networks. Electronic Journal of Statistics, 4, pp.585-605. Haslam, D., 2007. Young Hearts Run Free: The Real Story of the 1970s. London: Harper Perennial. Hebdige. D., 1979. Subculture: The Meaning of Style. London: Routledge. Hield, F., 2010. English Folk Singing and the Construction of Community. PhD thesis, University of Sheffield. Hield, F., and Crossley, N., 2015. Tastes, ties and social space: exploring Sheffield’s folk singing world. In: Crossley, N., McAndrew, S. and Widdop, P., 2015. Social Networks and Music Worlds. London: Routledge.

123 van de Hoeven, A. and Hitters, E., 2019. The social and cultural values of live music: Sustaining urban live music ecologies. Cities, 90, pp.263-271. Holland, P. and Leinhardt, S., 1977. A dynamic model for social networks. Journal of Mathematical Sociology, 5(1), pp.5-20. Holloway, J and Koskinen, J., 2016. Multilevel embeddedness: The case of the global fisheries governance complex. Social Networks, 44, pp.281-294. Kornienko, O., Agadjanian, V., Menjivar, C. and Zotova, N., 2018. Financial and emotional support in close personal ties among Central Asian migrant women in Russia. Social Networks, 53, pp.125-135. Laing, D., 1985. One Chord Wonders: Power and Meaning in Punk Rock. Milton Keynes: Open University Press. Lansley, S. and Mack, J., 2015. Breadline Britain: The Rise of Mass Poverty. London: Oneworld. Leblanc, L., 1999. Pretty in Punk: Girls’ Gender Resistance in a Boys’ Subculture. New Brunswick: Rutgers University Press. Levine, H.G. and Stumpf, S.H., 1983. Statements of Fear Through Cultural Symbols: Punk Rock as a Reflective Subculture. Youth and Society, 14(4), pp.417-435. Lusher, D., Koskinen, J. and Robins, G., 2013. Exponential Random Graph Models for Social Networks. Cambridge: Cambridge University Press. Mack, J. and Lansley, S., 1985. Poor Britain. London: Allen & Unwin. Martin, G., 2016. Fanzine Culture. [online] Available at: < https://www.bl.uk/20th-century- literature/articles/fanzine-culture>. Martin, P.J., 1995. Sounds and Society: Themes in the sociology of music. Manchester: Manchester University Press. McAndrew, S. and Everett, M., 2015a. Music as Collective Invention: A Social Network Analysis of Composers. Cultural Sociology, 9(1), pp.56-80. McAndrew, S. and Everett, M., 2015b. Symbolic versus commercial success among British female composers. In: Crossley, N., McAndrew, S. and Widdop, P., 2015. Social Networks and Music Worlds. London: Routledge. McPherson, M., Smith-Lovin, L., Cook, J.M., 2001. Birds of a Feather: Homophily in Social Networks. Annual Review of Sociology, 27, pp.415-444. Mizruchi, M.S., 1996. What Do Interlocks Do? An Analysis, Critique and Assessment of Research on Interlocking Directorates. Merton, R.K., 1968. The Matthew Effect in Science. Science, 159(3810), pp.56-63. Miller, S. ed., 2013. Touch and Go: The Complete Hardcore Punk Zine ’79-’83. Brooklyn: Bazillion Points Books. Millward, P., Widdop, P., Halpin, M., 2017. A ‘Different Class’? Homophily and Heterophily in the Social Class Networks of Britpop. Cultural Sociology, 11(3), pp. 318-336. Mische and Pattison. 2000. Composing a civic arena: Publics, projects, and social settings. Poetics, 27(2), pp. 163-194.

124

Novak, D., 2013. Japanoise: music at the edge of circulation. London: Duke University Press. Ogg, A., 2014. Dead Kennedys: Fresh Fruit for Rotting Vegetables, The Early Years. Oakland: PM Press. O’Shea, 2015. Embracing difference in feminist music worlds: a Ladyfest case study. In: Crossley, N., McAndrew, S. and Widdop, P., 2015. Social Networks and Music Worlds. London: Routledge. Pantazis, G., Gordon, D. and Levitas, R. eds., 2006. Poverty and Social Exclusion in Britain: The millennium survey. Bristol: Policy Press. PRS 2018. Live performances. Available at: Ripley, R.M., Snijders, T.A.B., Boda, Z., Vörös, A. and Preciado, P., 2019. Manual For RSiena [online] Available at: https://www.stats.ox.ac.uk/~snijders/siena/RSiena_Manual.pdf Robins, G. et al., 2007. An introduction to exponential random graph (p*) models for social networks. Social Networks, 29(2), pp.173-191. Robins, G. and Pattison, P., 2001. Random graph models for temporal processes in social networks. Journal of Mathematical Sociology, 25(1), pp.5-41. Savage, J., 1991. England’s Dreaming: The Sex Pistols and Punk Rock. London: Faber & Faber. Schweinberger, M. and Snijders, T. A. B., 2007. Markov models for digraph panel data: Monte Carlo-based derivative estimation. Computational Statistics and Data Analysis, 51(9), pp.4465-4483. Scott, J. and Carrington, P.J., 2011. The SAGE Handbook of Social Network Analysis. London: SAGE. Seven Ages of Rock, Episode 3: Blank Generation. [TV Programme] BBC Two, 2nd Jun 2007 21.00. Shank, B., 1994. Dissonant Identities: The Rock ‘n’ Roll Scene in Austin, Texas. Middletown, CT: Wesleyan University Press. Simmel, G., 1908. Soziologie Untersuchungenüber die Formen der Vergesellschaftung. Translated from German by Wolff, K.H., 1950 as The Sociology of Georg Simmel. Glencoe, IL: The Free Press. Small, C., 1998.Musicking: The Meaning of Performing and Listening. Middletown, CT: Wesleyan University Press. Smith, M., Gorgoni, S. and Cronin, B., 2019. International production and trade in a high- tech industry: A multilevel network analysis. Social Networks, 59, pp.50-60. Snapes, L., 2018. UK's first live music census finds small venues struggling. Guardian [online] Available at: Snijders, T.A.B., 1996. Stochastic actor-oriented dynamic network analysis. Journal of Mathematical Sociology, 21, pp.149-172. Snidjers, T.A.B. and Bosker, R.J., 1999. Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modelling. London: Sage.

125

Snijders, T.A.B., 2001. The Statistical Evaluation of Social Network Dynamics. Sociological Methodology, 31(1), pp.361-395. Snijders, T.A.B, 2002. Markov Chain Monte Carlo Estimation of Exponential Random Graph Models. Journal of Social Structure, 3(2). Snijders, T.A.B., van de Bunt, G.G. and Steglich, C.E.G., 2010. Introduction to stochastic actor-based models for network dynamics. Social Networks, 32(1), pp.44-60. Snijders, T.A.B., 2017. Stochastic Actor Oriented-Models for Network Dynamics. Annual Review of Statistics and its Application, 4, pp.343-363. Subcultures Network, 2014. Fight back: Punk, politics and resistance. Manchester: Manchester University Press. Thornton, S., 1997. General Introduction. In: Gelder, K. and Thornton, S., 1997. The Subcultures Reader. London: Routledge. Tilly, C., 2008. Explaining Social Processes. Boulder, CO: Paradigm TIME, 2019. Mark Zuckerberg Gives The Keynote Speech At Facebook Developer Conference | TIME. [video online]. Available at: . Townsend, P., 1979. Poverty in the United Kingdom: A Survey of Household Resources and Standards of Living. Harmondsworth: Penguin. University of Essex, 2018. Institute for Social and Economic Research, NatCen Social Research, Kantar Public. Understanding Society: Waves 1-8, 2009-2017 and Harmonised BHPS: Waves 1-18, 1991-2009. [data collection]. 11th Edition. UK Data Service. SN: 6614, http://doi.org/10.5255/UKDA-SN-6614-12 Wang, P. 2013, Exponential random graph model extensions: models for multiple networks and bipartite networks. In: Lusher D., Koskinen, J. and Robins, G. (eds.) 2013. Exponential Random Graph Models for Social Networks: Theory, Methods and Applications. London: Sage Wang, P., Robins, G., Pattison, P., 2009. PNet: program for the simulation and estimation of exponential random graph models. Melbourne School of Psychological Sciences, The University of Melbourne. Wang, P., Robins, G., Pattison, P and Lazega, E., 2013. Exponential random graph models for multilevel networks. Social Networks, 35(1), pp.96-115. Wang, P., Robins, G., Pattison, P. and Lazega, E., 2016. Social selection models for multilevel networks. Social Networks, 44, pp.346-362. Wang., P., Sharpe, K., Robins, G.L. and Pattison, P.E., 2009. Exponential random graph (p*) models for affiliation networks. Social Networks, 31(1), pp.12-25. Wang, P., Pattison, P.E. and Robins, G.L., 2013. Exponential random graph model specification for bipartite networks: a dependence hierarchy. Social Networks, 35(2), pp.211- 222. Wasserman, S., 1980. A Stochastic Model for Directed Graphs with Transition Rates Determined by Reciprocity. Sociological Methodology, 11, pp.392-412.

126

Wasserman, S. and Faust, K. 1994. Social Network Analysis: Methods and Applications. Cambridge, UK: Cambridge University Press. Watson, J., 2015a. A statistical analysis of the hardcore punk music network in the USA 1979–1986. MSc Dissertation, Social Statistics, University of Manchester, Manchester Watson, J., 2016. California Über Alles: The evolution of hardcore punk in the USA 1979– 1986. Presented at XXXVI Sunbelt Conference of the International Network for Social Network Analysis, Newport Beach, California, 5–10 April. Zappa, P. and Robins, G., 2016. Organizational learning across multi-level networks. Social Networks, 44, pp.295-306.

127

Appendix 1 Network sociogram for whole collected network data 2013 showing 368 active bands performing at 25 venues

128

Appendix 2 Network sociogram for whole collected network data 2014 showing 314 active bands performing at 23 venues

129

Appendix 3 Network sociogram for whole collected network data 2014 showing 346 active bands performing at 22 venues

130