Oscar Gerritsen 30-6-2016 10276580
University of Amsterdam
Predicting Music
Tastes through Social
Ties
1
1. Abstract
The goal of this research is to create a model that can predict variations in music taste based on the structure of friendship-ties in a network of Sociology students at the University of Amsterdam in 2015/2016 (N=70). The theoretical foundation that legitimize the pursuit of such a model are grounded on the principle of homophily. The idea that ‘birds of a feather flock together’ (McPherson & Smith-Lovin, 1987).
To estimate which birds of this network flock together, first their friendships on Facebook are gathered. The structure of these friendships form the basis for the relational network. Second, the tie-strength of the friendships in the network is estimated by measuring the overlapping friends two students have in common. Third, strong friendships are estimated as friendships with more than five friends in common. These three relational structures form the independent variables of this research.
This study contains two basic methods to establish the correlation between music tastes and social ties amongst the students. First, the similarity in music taste between students is measured as the ratio of similar interests for 10 music genres. Second, strong similarity between students is asserted by selecting students with similar interest for at least seven out of these 10 music genres. These two are the dependent variables of the predictive models from this research.
The results of this research show that there is no significant correlation between similarity in music taste and the structure of friendship-ties amongst the Sociology students at the University of Amsterdam in 2015/2016.
These results contradict earlier studies where the similarity between friendship-ties and music taste have been observed amongst students (Lewis, Gonzalez & Kaufman, 2012). This can be explained by the narrow definition of similarity in music taste and the small sample size of this research.
Furthermore, the mechanisms of social selection and social influence that drive homophily in social networks might not be based on music taste with the Sociology students at the University of Amsterdam in 2015/2016.
Consequently, this research calls for future research to use other forms of taste that can serve as effective markers for social differentiation, such as fashion, food or movies, in order to see whether these markers are strong enough to predict variations in taste based on the structure of social ties.
2
2. Table of contents
1. Abstract ... 1 2. Table of contents ... 2 3. Foreword ... 4 4. Introduction ... 6 4.1 Outline ... 6 4.2 Targeted advertising ... 7 4.3 Privacy on Facebook ... 9 4.4 My approach... 10 5. Theoretical Framework ... 11 5.1 Homophily ... 115.1.1 Social selection and social influence ... 12
5.2 Taste and social class ... 13
5.3 Music taste ... 14
5.3.1 Homophily in music taste ... 15
5.4 Tie-strength ... 16
5.4.1 Tie-strength on social networking sites... 17
5.5 Privacy on social networks ... 18
5.5.1 The taste for privacy ... 19
5.5.2 The taste for expression ... 19
6. Methodology ... 20 6.1 Facebook network ... 20 6.1.1 Friendship-matrix ... 22 6.1.2 Tie-strength-matrix... 23 6.1.3 Strong-tie-matrix ... 24 6.2 Questionnaire ... 25 6.2.1 Music genres ... 25 6.2.2 Demographics ... 29 6.2.3 Facebook behavior ... 35
3
6.3 Methods of Analysis ... 38
6.3.1 Quadratic assignment procedure regression ... 38
6.3.2 Models of analysis ... 39
7. Analysis ... 41
7.1 Similarity in music tastes and social ties ... 41
7.1.1 Similarity in music taste ... 41
7.1.2 Friendship-ties and similarity in music taste ... 42
7.1.3 Tie-strength and similarity in music taste ... 44
7.1.4 Strong-ties and similarity in music taste ... 45
7.1.5 Similarity in music taste and social ties ... 45
7.2 Strong similarity in music tastes and social ties ... 48
7.2.1 Strong similarity in music taste ... 48
7.2.2 Friendship-ties and strong similarity in music taste ... 50
7.2.3 Tie-strength and strong similarity in music taste ... 51
7.2.4 Strong-ties and strong similarity in music taste ... 52
7.2.5 Strong similarity in music taste and social ties ... 52
8. Conclusion ... 55
9. Discussion ... 56
9.1 Research design ... 56
9.2 Privacy on Facebook ... 57
9.3 Music tastes and social ties ... 57
10. Bibliography ... 59
4
3. Foreword
First and foremost, I want to thank dr. Thomas Leopold for agreeing to supervise this project. Thomas has provided excellent inspiration and an incredibly relaxed atmosphere to work in. This has undoubtedly been very beneficial for the fruits of this research.
As I recall, most of the meetings with my supervisor took the form of an inspiring 45-60 minute dialogue on the many different ways to explore the relation between tastes and ties, on the possible groundbreaking implications of this research, on how to trick my busy colleagues into filling out the questionnaire, and on discussing possible future career opportunities.
Some of these innervating meetings felt like those late nights where I had an intriguing discussion and felt quite certain that we had unraveled some of life’s greatest mysteries, only to find myself the next morning unable to reproduce most of the golden consensus achieved that night into any structured form. Let alone into a master thesis.
Thankfully, our meetings were on a more regular basis, and Thomas was always able to re-establish our train of thought. A rare gift, which soon brought me to the weird stage of finalizing the insights generated by this research before I had written a single word.
Our meetings were often concluded with the comforting words; “All looks well Oscar, nothing to worry about”. I remember these words having the ambiguous effect on me of either having a supervisor that did not care at all about the progress of my thesis, or that all was in fact looking well - which could also not have been the case, when I compared my scribblings and doings to the relentless work ethic of my fellow students.
Second, I would like to thank Kobé de Keere for all the support he has offered in constructing and finalizing this research. Although Kobé kept insisting that he is “not an expert on Bourdieu,
whatsoever”, I have never met anyone who could transfer Bourdieu’s theory on the forms of capital and the structure of taste in such a lively manner.
Next, I would like to thank my grandparents, Fem and Wim Oudshoorn, for giving me the opportunity to pursue my interests in Sociology after finishing my bachelors of Engineering,
notwithstanding the fact that a master degree in Sociology does not greatly enhance my prospects on the job market. In fact, having a master degree in Sociology probably decreases these prospects, seeing as long-term students are quite often frowned upon.
5 Furthermore, I would like to thank my mother Henriette Oudshoorn, and my sister Laura
Gerritsen, for supporting me on this journey. Whilst I am certain that being declined to graduate as an Engineer after completing the premaster Sociology at the Amsterdam University of Applied Sciences because; “it does not have anything to do with Engineering”, ascribes to the negative consequences of my ‘joie de vivre’ as a student, these two kept insisting of calling my premaster “a brilliant shortcut in the pursuit of two degrees”. This moderate form of cognitive behavioral therapy has had a positive effect on my life as a senior student.
Often we shared our differences on the either psychological or social causes of human behavior over an excellent prepared cod and a bottle of white wine at the dinner table. In the nature-nurture debate, I was always outnumbered two-to-one.
During these last two years at the University of Amsterdam, I was lucky enough to have
encountered love. I sincerely want to apologize to Rafne ten Oever for all the nights that I had to gather and analyze data, write mediocre essays or study for tests. I hope that these nights have not been given away in vain, as they should result in many more exciting and much more prosperous evenings together.
Last, I want to thank my father Jan-Willem Gerritsen, who paved the way for all my curiosity and expertise in Sociology, with his academic career and life’s work on ‘The control of fuddle and flash: a sociological history of the regulation of alcohol and opiates’. Although he has long been gone, some of his interests in the arts of Sociology have undoubtedly been transferred on to me, which might prove that my mother and sister were right all along.
6
4. Introduction
4.1 Outline
This research examines the correlation between music tastes and social ties amongst students at the University of Amsterdam in 2015/2016. The goal of this research is to create a working model that can predict music tastes for students, based on their friendship-ties within a Facebook network (N=70).
There are two basic methods applied in this thesis. First, similarity in music taste for each pair of students is measured as a ratio of similarity between them. Second, strong similarity in music taste is assigned to students that are at least 70% similar in their music taste. Then, the structure of friendships is used to estimate the ratio of similarity for each pair of students.
It is expected that friendly students will have more similar music tastes compared to non-friendly students. Moreover, it is contemplated that the differences between students’ taste in music are
observable, and that these observations are the result of structural differences in their friendship network. In short, this thesis is grounded on the sociological theory of homophily.
The principle of homophily states that similarity breeds connection. This principle structures social networks of any kind, including friendship-networks. Friends often have similar sociodemographic, behavioral and other characteristics (McPherson, Smith-Lovin & Cook, 2001; Kandel, 1978; McPherson & Smith-Lovin, 1987; Laumann, 1973; Hallinan & Williams, 1989; Marsden, 1988).
The main argument for similarity in friendships lies in the conducts of their relationships. Relationships in which people often interact in a friendly manner, must involve some level of mutual understanding. This mutual understanding is often expressed by showing similar interests, tastes or believes (Granovetter, 1973; Krackhardt, 1992; De Klepper, Sleebos, Van de Bunt & Agneessens, 2010).
This research examined whether the principle of homophily is strong enough to be implemented as a method to predict someone’s music taste, based on the structure of his or her social network.
7
4.2 Targeted advertising
The first reason to do so, is that this method might be beneficial to improve the practices of the so called ‘data brokers’. Data brokers are organizations aimed at collecting, interpreting and selling the digital footprints of citizens. This information is then often used to improve marketing services
(Hoofnagel, 2004).
This research is not original in trying to utilize the theory of homophily on a predictive model on consumer taste. Indeed, there are many examples of data brokers built around this principle found in contemporary society. One prominent and controversial player in this business is Facebook.
In March 2015 Facebook has over 936 million daily active users, of which 798 million users are active on mobile devices. Facebook provides friendship ties, a message service and the opportunity for uploading media content such as photos, videos and texts. Almost 94% of Facebooks $3.54 billion revenue of 2015 is derived from advertising. From this $3.32 billion advertising revenue, 73% was generated through mobile advertising (Facebook, 2015).
Contrary to popular opinion, this revenue is not created by directly selling the personal information of their users. Instead, Facebook offers companies a specific audience for targeted advertising, such as: ‘Dutch male citizens between 30-50 years old, who like cars’ or ‘Spanish students that like jazz’ (NRC, 2014; Fuchs, 2012; Weintraub, 2011). Targeted advertising for 936 million active users is what generates the majority (94%) of Facebook’s revenue.
The commercial value of targeted advertising can simply be expressed as the creation of an accurate match between customer and content. Advertising content that responds to the monitored activity of individuals has a better chance to resonate with those individuals. Therefore, targeted advertising has a better chance of selling products and services, compared to non-targeted ads. It has been reported that targeted advertising is more than twice as valuable, and more than twice as effective as non-targeted advertising (Knowlson, 2001; Beales, 2010).
The problem with using data provided by Facebook is that the information is voluntarily produced by its users. Therefore, the content always depends on the activity of the user. This results in large discrepancies within the commercial exploitability of user-generated data.
8 When a Facebook user provides much personal information to the social networking site, the ad can respond better. For instance; Jane, a twenty-one years old female, who lives in Amsterdam, and likes to listen to ‘Madonna’, watch ‘the Gilmore Girls’ and dine at ‘Restaurant Panini’, has all these
characteristics listed on her Facebook profile. This offers the advertising industry plenty of leads to work with. Now, compare this to a Facebook profile of Jane who only lists that she is twenty-one years old. In the latter Facebook profile, Jane could also be a male, or a senior. Therefore, the potential yield of a targeted advertising campaign on tampons aimed at Jane the male, is drastically reduced.
This oversimplified example is used to show how private profiles can devalue the targeted advertising platform offered by Facebook. In order to revitalize the commercial prospects of online social networking sites, this research examined how accurate estimations on music tastes can be made based on the structure of friendship networks.
By applying the theory of homophily in practice therefore, this research can offer valuable insights for data brokers. If similarity breeds connection, connections should be able to predict similarity. When these estimations are accurate, the model provided by this research can be extrapolated towards other domains of taste such as; food, fashion, movies, books or cars.
9
4.3 Privacy on Facebook
The rapid growth of online social network sites has coincided with an increasing concern over personal privacy. Facebook responds to this concern by allowing its users to control the privacy level of their profile, thus limiting outside access to this information. Although this restricts most outsiders and third-parties from gathering personal information, their protection is considerably less effective to the collection and use of data by Facebook itself. Facebook can use almost all personal information provided by the user on the social networking site (Lewis, Kaufman & Christakis, 2008; Alsenoy et. al, 2015).
Still, privacy poses a problem for the targeted advertising business. How can they reach users that do not reveal most of their interests on the online platform? What can they say about their taste and behavioral patterns? What accurate offer can they make those that do not share their interests online?
Here, the principal of homophily in social networks might offer valuable insights for the targeted advertising industry. If friendships are revealing someone’s interests, believes and other characteristics, then the match between customer and content can still be made.
Through the embeddedness of a Facebook user within their personal social network, Facebook can estimate the taste of these individuals, by exploring the taste of their friends (Alsenoy et. al, 2015; Granovetter, 1985). In short, Facebook can use social network analysis to estimate the taste of private profiles.
This method acknowledges the privacy settings from Facebook profiles, and still generates personal information. Therefore, social network analysis might offer a way to bypass the wall of privacy offered by Facebook. This would leave the private Facebook user in the illusion of being safeguarded from targeted advertising, while still become the target. Therefore, if this method is successful, one might be tempted to conclude that there is no such thing as privacy on Facebook.
10
4.4 My approach
In this research I explore the potency of predicting music taste of students through their user-generated content on Facebook. In particular, the accuracy of predictions on music taste for Facebook-users that do not reveal much of their personal taste online, are explored through their embeddedness within a social network.
This research provides a double-edged relevance by contributing to the discussion of privacy on Facebook, as well as to the body of knowledge on targeted advertising. These two fields are in constant motion, contesting for space in the private sphere (Fuchs, 2012; ECHR, 1950; Alsenoy et. al, 2015 ).
If this research is accurate in making predictions on music taste, then even Facebook users who only show little or none of their taste can still be described through their friends. This will diminish the concept of privacy and user-protection offered by Facebook. At the same time, the accuracy of this description offers fruitful insights into the commercial potential of online social networks. A chance to target consumer groups that are, at first glance, invisible.
The Facebook network at hand consists of master students Sociology from the University of Amsterdam enrolled between 2015-2016. The relational structure of friendships is mapped through Facebook friendships within the group ‘UvA Sociology MA tracks 2015/2016’.
This one-mode network is analyzed on the dyad level. The research focuses on similarity in music taste and the structure of friendship-ties between two actors. On the dyad level, predictions of ego’s music taste are made based on alter’s music taste and the strength of their friendship-tie. This tie-strength is estimated by the degree of overlapping friends between two actors.
The research builds on similarity in friendship-networks using the principle of homophily. The goal is to see if the homophily principle can provide accurate predictions on preferences in music for the actors.
Readers should note that only a small portion of information can be gathered from Facebook directly, due to their privacy settings. For instance, this research is not able to incorporate personal posts, chats, search queries, likes between persons or groups, and many other information that either describe a person’s music taste or their relation towards others within the network (Alsenoy et al., 2015).
In order to slightly fill this gap, additional personal information is obtained through a
11
5. Theoretical Framework
5.1 Homophily
The theory of homophily is grounded on the micro-sociological tradition of symbolic
interactionism. From this perspective, humans act towards their social world based on the subjective
meaning they ascribe to the things they have encountered in their life. They give meaning through their interactions with society, because most human behavior is embedded within networks of interpersonal relations. As the result of many interactions where individuals give similar meanings towards social entities, common symbols arise (Blumer, 1969; Granovetter, 1985; Goffman, 2005).
An example of such a symbol is the wave of a hand, which is generally understood as a way to greet someone. Through the lens of symbolic interactionism this general understanding is merely a temporary social consensus, established through many social interactions over the course of time. Therefore, the meaning of symbols can change over time and space (Blumer, 1969).
Although the general meaning of a symbol is socially constructed, its personal meaning still remains subject to the individual’s interpretation. People that differ on demographics such as age, education or nationality often attribute different meanings towards symbols. This is due to the structural differences of the social networks in which they are embedded. Therefore, the relational structure of social networks is the main focus of micro-sociology, because that is where the subjective interpretation of social reality is constructed (Blumer, 1969; Goffman, 2005; Mark, 2003).
Symbolic interactionism is at the core of the homophily principle, because it establishes that most human behavior is guided through the subjective interpretations of social interactions. People who have a relation tend to be much more similar than chance alone would predict. They are similar in relation to others in their social network. Therefore, the key of explaining homophily in social groups lies within the many individual interactions that structure the relationships of that social network. To explore these relations is to explain where social consensus arises and structural similarity begins (Granovetter, 1985).
12
5.1.1 Social selection and social influence
Although it is widely accepted that similarity breeds connection, there is still no consensus on what causes this phenomenon. This research will theorize the relation between a student’s music taste and their social relations in two ways; by social selection (1) and through social influence (2).
The first postulates that friends develop a relationship because they already have similar characteristics. According to this explanation, people select their friends based on their subjective interpretation of the characteristics they have in common with the other. Therefore, friendships are forged from the basis of mutual understanding. From the perspective of social selection, similarity breeds connection.
Although social selection has been described by some as the most important causal mechanism for the appearance of homophily in social networks, both causes are acknowledged in the scientific community (Lewis et al., 2012; Cohen, 1977; Ennett & Bauman, 1994).
The theory of social influence postulates that similarity in social networks comes from the social relations itself. Friends become increasingly similar through their friendships. Their social relations gradually shape and redefine their perspectives on life. From the perspective of social influence, connection breeds similarity.
Although most scientist argue that social influence is the weaker component of homophily in social networks, some argue that social influence is the dominant mechanism in social settings where organizational constraints are high. For instance, when the possibilities of friendships are limited and people are required to work together to fulfill a job or a project (De Klepper et al., 2010).
Since high organizational constraints are also part of the student network at hand, it remains unclear whether similarity between students comes from their friendship or vice versa.
13
5.2 Taste and social class
Another, more structural explanation for homogeneity amongst friends is offered by Pierre Bourdieu (1979). He argues that people acquire their tastes throughout their upbringing within a social milieu. Humans incorporate the social structures of their social milieu (i.e. the tastes, dispositions, sensibilities, appropriate behavior), and embody these structures as their own. From this perspective, expressing a music taste can be understood as expressing a set of aesthetic dispositions which are acquired from the social milieu one has experienced throughout his or her life. Expressing music taste is therefore an act of social positioning (Bourdieu, 1979).
On first glance, Bourdieu (1979) argues more in favor of social influence as the causal engine of homophily in society. However, social selection is by no means neglected in his work. Because the expression of taste can be seen as a form of social positioning; humans actively reproduce the symbols of their internalized social milieu when they express their taste. In doing so, they select others that are part of a similar social milieu and can relate to their taste. Therefore, social selection and social influence are both actively and simultaneously contributing towards homophily in social networks (Bourdieu, 1979).
Whether homophily exists in social networks due to social selection or due to social influence is not the main topic of this research. In this research, I focus on wielding the homophily principle into a predictive model for estimating similarity in music taste. Therefore, the principle that taste relates to the social structure of relations is what is most important for guiding this research.
14
5.3 Music taste
This research focuses on music taste, because music is often recognized as an expression of one’s social identity. Music can divide people into many different subcultures, each providing an exclusive social identity (Bennet, 2000; Tekman & Hortaçsu, 2002).
Some examples of subcultures built around music genres are; the ‘House-culture’ in London, the ‘Hardcore-scene’ in Rotterdam, the ‘Jazz'-culture’ from New-Orleans, the Jamaican ‘Reggae-culture’ or the ‘Hiphop-scene’ from New York. All these music genres have different origins, ideologies, and aesthetics, and therefore these music genres also have different social identities (Eyerman & Jamison, 1998; Hunt, Joe-Laidler, Moloney, Poel, & Mheen, 2011; Hersch, 2008; Kubrin, 2005; King, Bays & Foster, 2002).
These different social identities are less likely to establish mutual understanding in social interactions, which will probably result in different friendships in the network. Consequently, I expect that variations in music genres are likely to result in variations in friendships.
15
5.3.1 Homophily in music taste
In a longitudinal study amongst 1640 students from different universities in the United States, the relation between a student’s music taste and their friendships on Facebook has been recognized. Students that had a similar taste in music were more likely to be friends on Facebook. Furthermore, this research concluded that most similarity in music taste could be attributed to the mechanism of social selection instead of through the process of social influence. Only for classical- and jazz-music did the student’s taste seem to be developed through the process of social influence (Lewis, et al., 2012).
This could be established because the researchers used relational information from Facebook over a period of four years. Therefore, this longitudinal study made it possible to see whether students had similarity in taste before or after they had become friends.
These findings acknowledge that both the mechanisms of social selection and social influence are valid explanations for the way in which homophily in social networks arises.
Lewis et al. (2012) provided another important finding for the purposes of this research, namely that expressing a taste can have a different social impacts depending on its medium. Movies and music had a stronger social influence on friendship-ties compared to books.
Furthermore, expressing a taste can also have a different social impact depending on its social utility. Tastes that were widely shared amongst the student group did not serve as effective markers of social differentiation. Therefore, these tastes were no longer part of a student’s social identity (Lewis et al., 2012).
In general, music seems to be a medium which can have a strong social impact on friendship-ties amongst students. Based on these considerations, I hypothesize that music tastes are likely to be more similar amongst friendly students, compared to non-friendly students.
16
5.4 Tie-strength
Not all friendships amongst the students can be considered to be equal. Some friends are more close than others. In order to account for the different gradations of friendships, researchers often assign weights to relations. The weight of relations is called their tie-strength. One way to measure the tie-strength in friendships is to take the degree of overlapping friends within a network (Granovetter, 1973).
Although the research done by Lewis et al. (2012) provided excellent evidence for the relation between music tastes and social ties, it treated all Facebook-friendships in the network as equal, ‘weak’-ties. Therefore this research did not take into account the effect of stronger friendships on similarity in music taste amongst the students. Considering both mechanisms that drive homophily in social networks, this can be perceived as a shortcoming for understanding the relation between music tastes and social ties.
If students are similar in music taste due to social influence, they might be more influenced by strong friendships compared to weak friendships. On the other hand, if social selection breeds similarity in music taste, then it is also interesting to look at the strength of friendship-ties, because strong friends might be more similar in their music taste. Therefore, this research incorporates the tie-strength of friendship-ties in the student network.
17
5.4.1 Tie-strength on social networking sites
Adamic, Buyukkokten and Adar (2003) found that the strength of friendship-ties also increases through similar friendships on social networking sites. This result might indicate that the strength of an online friendship-tie can be estimated by calculating the overlapping friendships between two actors.
In this study, I will utilize this measure in order to give more depth to the relational structure of the student network.
In ‘Tunes that bind’, Baym and Ledbetter (2011) predicted the strength of friendship-ties based on similarity in music taste on the music-based online social network called ‘Last.fm’. They found that although shared tastes in music may foster weak-tie friendships, they do not often converge into strong friendship-ties. Strong friendships were uniquely dependent on the communication behavior of the users. The more ways partners on music-based social networking sites communicated, the stronger their relationship was likely to be (Baym & Ledbetter, 2011).
The results of this research are interesting, because they implicate that stronger friendship-ties will not necessarily result in stronger similarity in music tastes. Even though this might be true for music-based social networking sites, this might not be the case for the student network at hand. Therefore, it remains interesting to incorporate the strength of friendship-ties into the analysis.
Based on these findings, I hypothesize that stronger tie-strength will increase the similarity in music taste between two students from the network.
18
5.5 Privacy on social networks
Another important finding for the purposes of this research is that Facebook profiles often reflect actual personalities due to their social embeddedness. ‘Creating idealized identities is hard to accomplish because profiles include information about one’s reputation that is difficult to control (e.g., wall posts) and friends provide accountability and subtle feedback on one’s profile’ (Back et al., 2010).
This finding suggests that information derived from Facebook profiles often describe actual personalities. Therefore, information from Facebook profiles can be molded into information on potential customers for the targeted advertising industry.
Furthermore, the suggestion from Back et al. (2010) allows this research to apply offline network theories to the structures of online social networks. If Facebook profiles often reflect actual
personalities, then similar processes of social selection or influence might also be observable. Indeed, the principle of homophily has also been recognized in the structure of online social networks (Bisgin, Agarwal & Xu, 2010).
Predictions of a user’s music taste based on the similarity with his or her social connections can be seen as a legitimate way to bypass the privacy settings of an individual’s Facebook profile. Since the personal information is generated through statistical estimations, and not acquired through the breaching of one’s privacy settings, this information can be legitimately used for targeted advertising.
By incorporating the relational structure of Facebook friendships into this research, a glimpse of the commercial potential, and therefore also the practical limitations of privacy on Facebook, will be revealed.
19
5.5.1 The taste for privacy
It is interesting to find out what motivates users to hide their profile information on online networking sites such as Facebook. In ‘The taste for privacy’, Lewis et al. (2008) analyzed more than 1500 students’ Facebook profiles, of which 33% are private. They found that female students were more likely to have a private profile than men, and students having friends or roommates with private profiles were also more likely to have a private profile (Lewis, et al., 2008).
Furthermore, this research showed that students with private Facebook profiles have a unique set of cultural preferences, which Lewis et al. (2008) named their ‘taste for privacy’. For instance,
students that liked books of ‘Dan Brown’ or ‘Jane Austin’ were far more likely to have a private profile. In contrast, students that listened to ‘the Beatles’ almost always had a public profile (Lewis et al., 2008).
This ‘taste for privacy’ is informing because it might imply that there are common expressions of music taste amongst the private profiles of this research. Moreover, private profiles might form separate components of friendships in the network, making it harder to estimate the music taste of these users through the music taste of their friends. This might indicate that privacy on Facebook remains possible if the friends in the network also have a private profile. Therefore, it is quite interesting to explore users’ motivation behind expressing their taste on social networking sites.
5.5.2 The taste for expression
Liu (2007) examined statements of taste from over 125.000 profiles on MySpace, in order to discover the motives of users expressing their tastes online. He concluded that ‘prestige and
differentiation’ were the primary motives for users to list their taste in music, movies, television and books online.
Prestige can be understood as wanting to gain status by affiliating with other profiles to become a prominent member of the group. Differentiation can be understood as distinguishing oneself from other profiles to establish uniqueness (Liu, 2007).
This social utility for the expression of taste on social networking sites has led me to believe that there are enough users who will express their music taste on the Facebook network, in order to predict the music taste of their friends.
20
6. Methodology
6.1 Facebook network
To measure the relationship between music tastes and social ties, this research focuses on master students of Sociology at the University of Amsterdam. These students are all part of the Facebook network ‘UvA Sociology MA tracks 2015/2016’ (N=70).
This sample is primarily selected because of its accessibility. Since I am part of the master’s program of Sociology at the University of Amsterdam, many students that are part of this network are acquaintances of mine. This increases the possibility of data collection from Facebook, because Facebook does not provide much profile information to strangers. In order to get the information on the structure of friendship-ties, the researcher must be part of the group.
Second, this sample provides a clearly defined boundary to the network analysis. When looking at music tastes and friendships on Facebook, a problem arises with the diversity of social settings for each actor. Most students have many friendships on their Facebook profile from outside of the University of Amsterdam. These friendships could also have shaped their music taste. Therefore, if all friendships are included in the analysis it becomes unmanageable. While most methods of social network analysis are based on the assumption of a closed network, in reality the actors of a group are almost always also active outside of the network.
When strong similarity in music taste between friends is observed, this might be due to a different social setting where the actors interacted, for instance at a concert or a festival. Therefore, it is important to find a common denominator within the group in order to keep the scope of the research manageable. Furthermore, the boundaries of this network are put in place to make the research process transparent and replicable. Consequently, the results of this study can be compared to other research on ‘closed groups’ such as companies, classes, clubs or other venues.
The common denominator of the actors within this group is that they are all part of the same program, at the same university and within the same Facebook network. This generates a solid basis for the establishing of friendship-ties on Facebook.
21 In the first step of this research the friendship network of each student is extracted from each of the student’s Facebook profiles. Since the relational data on Facebook is voluntarily and naturally provided by its users, this method avoids having large roster questions in the survey.
Furthermore, network analysis based on survey research often uses interaction events to measure the relationships between actors, such as; ‘gossiping’, ‘working together’ or ‘spending leisure time with’. In a survey, these interaction events are susceptible to the interpretation of the respondent, which can result in measurement errors. Since each respondent has different ideas on what ‘gossiping’, ‘working with’ or ‘leisure time’ means, the outcome of the survey is often not coherent. In this research, a relation is objectively measured as the mutually confirmed ‘friendship’ between two students on Facebook.
In order to obtain the relational data, I have become Facebook friends with 70 students from the Facebook network ‘UvA Sociology MA tracks 2015/2016’. This allows me to look at their Facebook profiles and see which students are befriended. Therefore, the assumption that all friendship-ties in this network are revealed is met.
The boundaries of this research provide a closed network, because all friendships are visible and each student is consciously part of the group. These students have voluntarily created a social entity, which main goal is to share knowledge, ideas, jokes and services in matters of Sociology. This research will discover whether this social network is also structured through the homophily principle.
22
6.1.1 Friendship-matrix
First, the friendship-ties obtained from Facebook are anonymized in order to protect the privacy of the students. Then, this relational information is used to create a 70x70 friendship-matrix. The anonymized students are represented on each axis, in the same order.
Since each friendship-tie on Facebook can mean a different type of friendship in real life, all friendship-ties are initially coded as ‘weak’-ties. Therefore, the weight of a friendship-tie in the
friendship-matrix equals one. All students that are not befriended in the Facebook network are coded as zero.
For instance, if students A and B are befriended with everyone in the network, but students C and D are not friends, their 4x4 friendship-matrix would look as follows:
𝑆𝑡𝑢𝑑𝑒𝑛𝑡 𝐴 𝐵 𝐶 𝐷
𝐴 − 1 1 1
𝐵 1 − 1 1
𝐶 1 1 − 0
𝐷 1 1 0 −
This quantitative measure of friendship-ties amongst the students is convenient, because every friendship-tie can be approached as a similar relation. Every friendship-tie means that two students have mutually agreed on a ‘friendship’ on Facebook. Therefore, this research avoids the subjective component of what a friendship means for each individual.
Next, the 70x70 friendship-matrix is plotted into a one-mode network, which is analyzed on the dyad level. First, this research will select the students with the most and the least friendships, and look at their similarity in music taste with other actors in the network.
Then, the friendship-matrix is used as the independent variable in a linear regression model that predicts ego’s music taste, based on similarity with alter’s music taste, and their friendships on the network. This model shows whether having a friendship-tie increases the chance of sharing similar music tastes. The analysis is controlled for by sociodemographic- and other behavioral characteristics that could also account for the similarity in music taste between two actors in the network.
23
6.1.2 Tie-strength-matrix
In the second analysis, the friendship-matrix is used as the basis to establish the tie-strength of actors in the network. Scholars have argued that the strength of friendship-ties on social networking sites increases through overlapping friendships (Adamic et al., 2003; Granovetter, 1972). In this study, I utilize this measure in order to create a tie-strength-matrix.
The tie-strength-matrix is built by calculating the amount of friendships two students have in common. For each friend two students have in common, their tie-strength increases with a weight of one, provided that these two students have a friendship on Facebook. If the students do not have a friendship on Facebook, their tie-strength remains zero.
For example, If we take the 4x4 friendship-matrix from the previous paragraph, where students A and B are befriended with everyone and students C and D do not have a friendship, then the 4x4 tie-strength-matrix looks as follows:
𝑆𝑡𝑢𝑑𝑒𝑛𝑡 𝐴 𝐵 𝐶 𝐷
𝐴 − 3 2 2
𝐵 3 − 2 2
𝐶 2 2 − 0
𝐷 2 2 0 −
Students A and B have a friendship with two friends in common, therefore their tie-strength equals three. Students A and C have a friendship, but have only one friend in common (student B), therefore their tie-strength equals two. Although students C and D have two friends in common (students A and B), they do not have a friendship, therefore their tie-strength equals zero.
This measure for tie-strength is used for each pair of actors in the student network. All relational data obtained from Facebook is recoded into a 70x70 tie-strength-matrix. The tie-strength-matrix is used as the second independent variable of a linear regression model.
Predictions of ego’s music taste are made based on alter’s music taste and their tie-strength. The linear regression model used for this prediction shows the effect of each increase in similar friendships for their similarity in music taste. Are students with more friends in common more similar in their music taste?
24
6.1.3 Strong-tie-matrix
The tie-strength-matrix is then used to create a 70x70 strong-tie-matrix for all of the students. This matrix represents the students with a tie-strength higher than five. For each pair of students with more than five friends in common on the network, their relation in the strong-tie-matrix equals one. Students that have a tie-strength of five or lower receive a code of zero.
The goal of this matrix is to make a binary distinction between strong and weak friendships, to explore its effect on similarity in music taste. Students that have more than five friends in common are isolated in order to analyze the differences between the groups. The strong-tie-matrix is used as the third independent variable of a linear regression model, where stronger friendships are compared to weaker friendships.
25
6.2 Questionnaire
After establishing the relational structure of the student network, the characteristics of the students are measured through a survey. This questionnaire contains one broad question on music genres and several questions on variables that can influence either the relational structure of friendships in the network, or the similarity in music taste.
Most of these questions are taken into account in order to provide some sociodemographic information on the student, other questions relate to the Facebook behavior of the student.
6.2.1 Music genres
Question 7a-w from the survey contains questions on 23 music genres. These music genres are based on the likes from the students’ Facebook-profiles. The questionnaire only contains music genres that are liked at least once by one of the students on their Facebook profile. Missing genres such as ‘salsa’ or ‘hardcore’, are discarded from the survey. This is done in order to keep the link between music genres that can serve as effective markers of social differentiation.
For the question on music genres a seven-point Likert-scale is used. Students can answer with; ‘Love it’, ‘Like it very much’, ‘Like it’, ‘Neither like nor dislike it’, ‘Dislike it’, ‘Dislike it very much’, ‘Hate it’ and ‘Don’t know’. This gives more depth to the research for comparing music tastes, because the students can vary in how much they like or dislike a specific music genre. It is useful to measure a student’s music preference on a seven-point scale, to see whether strong preferences are similar amongst friends. If a binary scale was used for this question, the measurement of music taste would have become too inclusive.
Furthermore, by using a seven-point Likert-scale, this research can separate music genres with strong preferences to more neutrally liked genres amongst the students. As research by Lewis et al. (2012) showed, the expression of taste can have different social impacts depending on its social utility. Music tastes that are widely shared are less likely to serve as effective markers of social differentiation, and are therefore not part of a student’s social identity (Lewis et al., 2012).
The seven-point Likert-scale grants this research a way to filter out music genres that are not effective markers of social differentiation. This is achieved through a selection of music genres, where at least 30% of the respondents show a strong preference towards it (‘Love it’, ‘Like it very much’, ‘Dislike it very much’, ‘Hate it’).
26 The rationale behind this selection is that strong preferences in music tastes are related to strong social identities. This is not fully in line with the explanation that has been offered by Lewis et al. (2012), where tastes that were widely shared (including tastes with strong preferences) may have lost their power as effective markers of social differentiation, because they were shared by almost all members of the group.
The results of the survey show that there are 10 genres where more than 30% of the students show a strong like or dislike. A strong like is measured when students either filled in ‘Love it’ or ‘Like it very much’ on question 7a-w from the questionnaire. A strong dislike is measured when students either responded with ‘Hate it’ or ‘Dislike it very much’ on that same question.
This resulted in the following 10 specified music genres; ‘Rock’, ‘Soul’, ‘R&B’, ‘Hiphop’, ‘Reggea’, ‘Classical’, ‘Techno’, ‘Jazz’, ‘Funk’ and ‘Metal’. Metal is the only category where more than 30% of the students responded in a strongly negative way. The other nine music genres are all perceived as strongly positive by at least 30% of the students. Similarity in these 10 music genres is the dependent variable of this research.
Table 1.
The 10 specified music genres with strong preferences
Music Genre Percentage of students
with strong preference
1. Rock 45.7% 2. Soul 35.7% 3. R&B 32.9% 4. Hiphop 37.1% 5. Reggea 32.9% 6. Classical 40% 7. Techno 31.4% 8. Jazz 47.1% 9. Funk 38.6% 10. Metal 30%
27 Next, these 10 music genres are used to create a taste-matrix where the similarity of music taste between students is represented. The similarity in music taste is measured by recoding the answer categories of the specified music genres into a five-point Likert scale. This is accomplished in order to create more overlap in music taste between the students. ‘Love it’ and ‘Like it very much’ are recoded as
strong positive preferences; ‘Hate it’ and ‘Dislike it very much’ are recoded as strong negative
preferences. ‘Like it’ is recoded as an average positive preference; ‘Dislike it’ is recoded as an average
negative preference, and ‘Neither like or dislike it’ and ‘Don’t know’ are recoded as neutral preferences.
6.2.1.1 Taste-matrix
The taste-matrix uses the five-point Likert scale to assign similarity between each pair of
students. Similarity in music taste is measured as the ratio of overlap between each pair of students. The taste-matrix is built by taking the sum of the respondents overlapping preferences and dividing this number by the total amount of music genres.
There are 10 specified music genres in the taste-matrix. Therefore, each similar interest or disinterest increases the relation between two students in the taste-matrix with 10 percentage points. For instance, two students that have strong positive preference towards Rock (‘Love it’ or ‘Like it very much’), strong negative preference towards Jazz (‘Hate it’ or ‘Dislike it very much’), and have neutral preference towards Hiphop (‘Neither like nor dislike’ or ‘Don’t know’), share similarity in 3 music categories. Their taste overlap in the taste-matrix equals 0.3 or 30%.
Similar disinterests in music genres are also part of a student’s taste profile because it can be a part of their social identity. Negative or non-affiliation with a music genre can also be an effective marker of social differentiation. Student’s that both genuinely hate Metal might find solace and mutual
understanding in their hatred of the genre, which can be a cornerstone of their friendship.
Readers should note that when a student ‘Likes’ a music genre, and another student ‘Loves’ that same music genre, they do not fall into the same taste preference category because their levels of preference differs too much. In this research, similarity in taste profiles has not been made too inclusive, because the 10 music genres that are selected for the taste-matrix are already based on strong
28 The students’ similarity in music taste is represented as a ratio in the taste-matrix. At a
minimum, students can share 0.0 or 0% in their preferences for the 10 music genres. The maximum amount of similarity occurs when a pair of students have similar preferences in all of the 10 music genres. The ratio for this pair of students then equals 1 or 100%.
For example, students A and B have four music tastes in common, students A and C have one music taste in common, and students B and C have two music tastes in common. Then, their 3x3 taste-matrix looks as follows:
𝑆𝑡𝑢𝑑𝑒𝑛𝑡 𝐴 𝐵 𝐶
𝐴 − 0.4 0.1
𝐵 0.4 − 0.2
𝐶 0.1 0.2 −
This is measured for all 70 students in the network. The resulting 70x70 taste-matrix is used as the first dependent variable in the linear regression models 1a, b and c, with respectively friendship-ties (a), tie-strength (b) and strong-tie-strength (c) as the independent variables.
6.2.1.2 Strong-taste-matrix
The taste-matrix shows an overview of the ratio of similarity in music tastes for each pair of students. This matrix is used as the basis to build the strong-taste-matrix. The strong-taste-matrix is a binary variable that selects students with strong similarity in music taste and compares them to students with lower similarity in music taste. In the strong-taste-matrix only students that have similar
preferences in at least seven out of the 10 music genres are represented as one. All students that share lower similarity in music taste are coded into zero.
The strong-taste-matrix is used as the second dependent variable in linear regression models 2a, b and c, with respectively friendship-ties (a), tie-strength (b) and strong-tie-strength (c) as the
29
6.2.2 Demographics
Similarity in friendship-ties and music tastes cannot be explored without measuring similarity in other variables that can account for either constructs. Only once the influence of these other variables is known, can the researcher be anywhere near optimistic that the relationship he or she observes is accurate. These are the control variables of the network analysis.
The survey contains several questions on demographic variables that control for the similarity in music taste and social ties. In the next section these variables are explored and a rationale is provided for integrating them into the survey. Readers should note that there might be other variables that can also contribute to the establishing of friendship-ties or the development of similar music tastes, which are not included in this research for pragmatic reasons.
Binary similarity-matrixes are constructed for each of the demographic variables included in this research, in order to compare the relational structures of the student network. Each binary similarity-matrix describes if actors are similar on the demographic variable or not.
For instance, if student A and B are both female but student C is male, then their 3x3 similarity-matrix for gender looks as follows:
𝑆𝑡𝑢𝑑𝑒𝑛𝑡 𝐴 𝐵 𝐶
𝐴 − 1 0
𝐵 1 − 1
30 6.2.2.1. Gender
Gender is often part of homophily in social networks. Similarity in gender likely increases the establishing of friendship-ties amongst the students. Friendships are often forged between two people who can share their experiences and emotions, because they understand each other. Therefore, gender can structure the friendships in this network (McPherson, Smith-Lovin & Cook, 2001; Elkins & Peterson, 1993).
Furthermore, gender can influence a student’s music taste, when conforming to dominant gender roles in contemporary society. Gender roles dictate what is masculine or feminine music in culture. According to Christenson & Peterson (1988), ‘gender is central to the ways in which popular music is used and tastes are organized’. Other research on the subject found that men were more often into heavier contemporary music and women more often into chart pop (Colley, 2008).
In the sample 54 students are female and 16 students are male. On first glance, gender does not segregate the friendship-relations of the student network into separate groups. Friendships based on similarity in gender are not strongly represented in the student network. Both the student with the most friendships (26) as with the least friendships (0) is female.
6.2.2.2. Nationality
Similarity in nationality also positively contributes towards homophily in social networks. Cultural differences can become barriers for understanding each other, sharing experiences, establish trust and become friends. Language, as one example of a cultural barrier, is a crucial aspect of
communication which, in turn, is an integral part of establishing friendships (Sias, et al., 2008). In a research done by Furnham & Alibhai (1985) on the friendship networks of foreign college students, they discovered a strong preference for foreign students to establish friendships with co-national students first, other co-national students second and local students last. This might indicate a clustering of friendships in the network amongst students born in the Netherlands and those born in other countries. In the survey the division of local students born in the Netherlands and students with other nationalities is questioned, in order to establish binary similarity matrixes.
31 Furthermore, music taste is often influenced and guided by cultural standards (Schuessler, 1948). Therefore, both the key constructs of this research, music tastes and social ties, are likely to be
influenced by a student’s nationality.
In the network, 46 students are born in the Netherlands and 24 are from other countries. Homophily based on nationality seems to be present in the student network. Foreign and non-foreign students tend to have more friendships with students from their own group.
6.2.2.3. Age
Differences in age are also often noted as a barrier for establishing friendships. Age defines a stage in life, where a person’s interests as well as their physique differs. Feelings related to
companionship, satisfaction, intimacy, nurturance and reliable alliance are reported to be significantly greater in friendships of similar age compared to friendships of different age (Holladay & Kerns, 1999).
Similarity in age is measured as students with a maximum range of 3 years between each other. When the age of students differs more than 3 years, they are not considered to be similar for the purposes of this research.
On average, students range between 22 and 28 years old. The youngest student is 21 and the oldest student is 45 years old. Students do not seem to have most friendships amongst students of their own age in the network. Therefore, similarity in age does not structure the social relations in the student network.
32 6.2.2.4. Track
One of the control variables used in this research is based on the student’s academic career. If students follow the same track of Sociology at the University of Amsterdam, then they are more likely to have developed friendship-ties, compared to students of different tracks. Through friendship-ties music tastes can be transferred and adopted, which might lead to stronger similarity in music taste for these students. Therefore, the track is questioned in the survey.
There are eight master tracks in Sociology at the University of Amsterdam. Each of these tracks is represented on the Facebook group ‘UvA Sociology MA tracks 2015/2016’ by at least one student. Most students are in the track ‘Social Policies and Social Problems’ (16), the least students are in the ‘Research Master’ (1) and in ‘Urban Sociology’ (1).
Similarity in master track seems to segregate the structure of the friendships in the student network strongly. Students that follow the same track are far more likely to have established a friendship-tie amongst each other, compared to students of different tracks.
6.2.2.5. Cultural capital
This research includes insights from Bourdieu’s (1986) extensive work on the structural relation between tastes and social class. According to his theory, what constitutes ‘taste’ in society is defined by those who possess some form of cultural capital, in order to distinguish themselves from others.
There are many forms of cultural capital. Possessing knowledge or skills in language, art, architecture, fashion, photography, music or cinema, all contribute to one’s cultural capital. Gaining cultural capital presupposes that the individual has the time and the means to act upon cultural
activities, such as learning to play an instrument or paint. For the working class, the time and means are often not available to be converted into cultural capital, because they are needed for their survival. Therefore, the amount of cultural capital is related to one’s social class (Bourdieu, 1986).
According to Bourdieu (1986), most cultural capital is transferred from the parents to their children. Parents teach their children most of the tastes and appropriate conducts of their social milieu. In order to account for the different social milieus amongst the students, the social position of the parents is measured. The educational background of the parents is used in order to describe possible variations in a student’s cultural capital (Sullivan, 2001; Choy, 2001).
33 The educational background of a student’s parents is considered high when at least one of the parents has a master’s degree or advanced graduate work (ph. D). If one of the student’s parents has completed education below or at the level of a bachelor’s degree, then the student is considered to have low cultural capital for the purposes of this research.
In general, most students at the university are not first-generation students, they have parents that also went to college (Choy, 2001). Therefore, it is interesting to see whether there are structural differences in music taste and social ties between the first-generation students and the second-generation students.
In the student network, 42 students are first-generation students and 28 are second-generation students. Differences in a student’s cultural capital do not seem to structure the friendship-ties in the network. First- and second-generation students often have friendship-ties between them.
In table 2 the demographics for the master students of Sociology at the University of Amsterdam, 2015/2016 are represented.
34 Table 2.
Descriptive Statistics for Master Students of Sociology at the University of Amsterdam, 2015/2016
Student at the UvA 0-1 year 1-2 years > 2 years
Gender Male 4 7 5 Female 23 16 15 Nationality Dutch 4 23 19 Other 23 0 1 Age 21 1 0 0 22 4 2 1 23 6 1 5 24 2 5 6 25 4 7 4 26 3 0 2 27 4 3 1 28 1 2 1 30 0 1 0 31 1 1 0 32 1 0 0 45 0 1 0 Track 1. Cultural Sociology 5 1 5 2. Gender, Sexuality and Society 10 2 3 3. Urban Sociology 1 0 0 4. Comparative Organization
and Labor Studies 2 5 3 5. Migration and Ethnic Studies 2 1 1 6. Social Problems and Social Policies 6 5 5 7. Research Master 0 0 1 8. General Sociology 1 9 2 Cultural capital Low 11 19 12 High 16 4 8
35
6.2.3 Facebook behavior
After the music tastes and demographics, the survey continues with several questions on the Facebook behavior of the student. By quantifying the friendships amongst students, this research assumes that each friendship is similar in its nature. The behavioral variations that lead to the establishing of a Facebook friendship are problematic for this assumption. Therefore, this research controls the establishing of Facebook friendships by measuring the Facebook activity and the Facebook friendship-policy of the students.
For each of these variables binary similarity-matrixes are constructed in order to compare students with the same Facebook behavior to students with different Facebook behavior.
6.2.3.1 Facebook user activity
Variations in friendship-ties can be established by differences in Facebook user behavior. Students that are very active on Facebook, are more likely to have established Facebook friendships amongst the group. If a student does not pay much attention to Facebook, he or she might not value an online friendship-tie as much as someone who is very active on the social networking platform.
Since this network consists of all master students at the University of Amsterdam, it is likely that some of these students were already acquainted before this research. If they have a Facebook profile, they are more likely to have established a friendship-tie with another student. If a student just recently started with a Facebook profile, then their Facebook network is less likely to have as many fellow students as someone who has had their profile for a long time. Therefore, friendship-ties amongst students might be better explained by a student’s Facebook activity than their similarity in music taste.
A student’s Facebook activity is measured with two components; how often the student is on Facebook (a) and for how long the student has had their Facebook profile (b). On the first component, activity is considered high when students are on Facebook at least once a day. On the second component activity is considered high when students have had their Facebook profile for over two years.
In the student network 62 students are on Facebook every day and 69 students have had their Facebook account for over two years. Therefore, most student’s Facebook activity is considered high. In order to integrate Facebook activity in the network analysis, a similarity-matrix is created.
36 The similarity-matrix is constructed by taking the product of each tie between the students. This results in a matrix where only students with high Facebook activity are represented as a positive tie of one. When one or both of the students are low active users, their tie remains zero. Therefore, this measure only accounts for similarity in high Facebook behavior, because it is expected that this influences the structure of friendships in the network.
For instance, students A and B are highly active on Facebook, but student C and D are not. In this case, their 4x4 similarity-matrix on Facebook behavior looks as follows:
𝑆𝑡𝑢𝑑𝑒𝑛𝑡 𝐴 𝐵 𝐶 𝐷
𝐴 − 1 0 0
𝐵 1 − 0 0
𝐶 0 0 − 0
𝐷 0 0 0 0
Notice that student C and D do not share a positive similarity-tie, even though they are similar in their low activity on Facebook.
6.2.3.2 Facebook friendship policy
Some users are more eager than others to establish a friendship on Facebook. This discrepancy is troubling for the intentions of this research. If friendship invitations are send based on low or no contact amongst the students, then it is unlikely that these students are similar in their music taste, either through social selection or social influence.
In order to account for variances in the Facebook friendship-policy of the user, the survey questions what percentage of a student’s friendships on Facebook are acquaintances (not good friends), and whether a student would accept a friendship invitation from a stranger. Together, these two
questions describe the Facebook friendship-policy of a student. If both of these variables are high, then the student is considered to be a strong networker on Facebook. If at least one of these variables is low, then the student is considered to be a weak networker on Facebook. For this variable, the same positive similarity-matrix is constructed as in the previous paragraph.
In the network, 60 students have more than 50% acquaintances on their Facebook profile and 46 of the students would accept a total stranger on Facebook.
37 6.2.3.3 Facebook privacy
Students can differ in their reluctance to provide personal information on their Facebook profile. While some students express their music taste in a vigorous manner, others do not reveal any of their music taste on Facebook. Since this research explores if the music tastes of both groups can be
estimated through their friendships, it is interesting to find out which students of the network show their music taste and which do not.
Therefore, in addition to the survey, the music tastes from the students are obtained through their Facebook profiles. Students that have a maximum of five likes in music artists on their Facebook profile are considered to have private profiles. These students do not share much of their personal music taste on the social networking site. Therefore, they are less likely to be targeted by advertising
campaigns based related to music. This ‘taste for privacy’ might structure their friendships and their music taste (Lewis et al., 2008).
Students that have more than five likes in music artists on their Facebook profile are not
considered to have a private profile. These students are willing to share their music taste online. For this variable, similarity amongst the students is also measured using a positive similarity-matrix. Only
students that both have a private profile receive a positive tie in the similarity-matrix.
Note. Data are from the master students of Sociology at the University of Amsterdam in 2015/2016
Table 3.
Facebook behavior of the Master Students of Sociology at the University of Amsterdam, 2015/2016
Private profile Yes No
Membership > 2 years 25 44 ≤ 2 years 0 1 Activity Daily 20 42 Non-daily 5 3 Friends or acquaintances
≥ 50% of friends are acquaintances 21 39 < 50% of friends are acquaintances 4 6
Accept strangers
Yes 16 30
38
6.3 Methods of Analysis
The analysis begins after all the information from Facebook and the survey is collected. First, the network structures of the independent and dependent variables are visualized. Second, the key actors with the strongest and weakest similarity in their music taste are located. These actors are followed in the different friendship-networks, in order to explore the relation between friendships and music taste.
Then, the network is analyzed at the dyad-level. In this research there are six regression models that analyze the similarity in a student’s music taste and their friendship-network. Each regression analysis incorporates the data from the student network in a different manner, thus offering different insights into the relation between friendship-ties and music tastes. The regression models all use the
quadratic assignment procedure to estimate this relation.
6.3.1 Quadratic assignment procedure regression
In quantitative data analysis most hypotheses are tested based on the assumption that the units of the sample are independent. This assumption is problematic for network analysis, because the structure of the network is made up of actors that are not independent but are interdependent. The actors have relations with one another. As is the case in this research, the actors are related based on their friendship-ties.
In order to test the hypotheses on the relation between characteristics of actors and the relational structure of the network, a quadratic assignment procedure regression (QAP-regression) is used. The quadratic assignment procedure uses the relational data from matrixes and randomly shuffles this data many times, in order to create a vast amount of new matrixes. With each new matrix a new order of the relation between the dependent and the independent variable is simulated.
The null hypothesis of a QAP-regression states that there is no association between the dependent and the independent variable. By shuffling each value for the two matrixes an ‘infinite’ amount of times, creating an ‘infinite’ amount of matrixes, a null distribution of correlations can be created. This null distribution resembles what the sampling distribution looks as if there is no correlation between the dependent and the independent variable. In practice, the quadratic assignment procedure generates does not create an infinite, but a large amount of matrixes based on the size of the matrixes.
