Homophily and friendship dynamics

(1)

Sociologiska Institutionen

Masteruppsats i sociologi, 30 h.p.

Ht 2017

Handledare: Love Bohman

Homophily and friendship dynamics

An analysis of friendship formation with respect to homophily principle and distinctiveness theory

Author: Sepehr Saeidibonab

(2)

b

Abstract

People always find themselves interacting with others and forming ties with them; these ties shape an individual’s social network which help form the self-conception and identity of a person. In discussing the essence of social networks and how they are formed the concept of homophily is of high significance. Therefore, the purpose of this research is to show the association between homophily and the process of friendship formation. As the structure of any social network is important in tie formation, I have also intended to study homophilous tie formation from a distinctiveness theory perspective, suggesting that individuals with minority characteristics are more prone to form friendship ties with each other. The types of homophily studied in this research are gender, religion, nationality/ethnicity, and political views. The data is gathered from the cohort which started grade 10 in upper secondary education in a school in Stockholm in Autumn 2012.

The analyses were conducted using logistic regression. The results indicated the existence of gender homophily and national homophily. However, religious homophily did not appear to have a robust association; political homophily was only robust for individuals who were participating in political meetings. However, due to lack of sufficient data, the relations between network structure and homophilous relations could not be accurately tested. Since the data were not collected randomly and the school was chosen due to its specific characteristics, it is not possible to generalize the results of the research to all the adolescents living in Stockholm. However, this research sheds some light on the mechanisms at play in friendship formation among adolescents.

Keywords

Homophily, Social network, Network structure, Distinctiveness theory

(3)

c

1 Introduction

While moving around through social spaces, people establish various relations which have considerable effect on them; these relations or networks shape or even change their desires, beliefs and opinions. In other words, individuals, almost constantly, find themselves interacting with others and forming ties with them. These ties shape the individual’s self-concept and define his/her identity, desires, beliefs and opinions (see: Edling & Rydgren, 2014; Hedström, 2005). Hence, studying social networks can benefit us in better understanding the individuals’ behavior and future actions which, ultimately, would lead to a more comprehensive grasp in how society works. People get involved in various networks either due to individual characteristics, social contexts and/or social opportunities; these social networks can be both intentional and unintentional.

Essentially, one of the main elements to consider in social network formation is proximity. That is, people become friends with people who are geographically close to them. For instance, students become friends with class mates, co-workers become friends with other co-workers, etc. Hence, a focused social organization is important in network formation (see: Feld, 1981). However, in discussing the essences of social networks and how they are formed or structured, in addition to propinquity, the concept of homophily is of high significance; as is indicated in the old saying:

“birds of a feather flock together”. Homophily principle indicates that people tend to socialize with similar others. That is, others who share some characteristics with the individuals. These characteristics can include age, religion, ethnicity, nationality, opinion and values, etc.

(McPherson, Smith-Lovin, & Cook, 2001).

In recent years, societies have become more and more diverse and multicultural due to globalization and the dramatic increase in migration (Van Der Wildt, Van Avermaet, & Van Houtte, 2015). This event produces a huge opportunity for individuals to make acquaintance with people from other cultures and exchange information and ideas with people from all around the globe. However, this idea is not penetrated in most social networks and people still prefer homophilous relations (Smith, Van Tubergen, Maas, & McFarland, 2016).

Individuals, in general, prefer homophilous relations for various reasons. For instance, people usually prefer to be friends with others of the same religion. This can be due to the fact that having the same religious belief generates trust among two individuals and also they share the same

(5)

2

opinion regarding the concept of “good people” (Windzio & Wingens, 2014); further, with regards to national or ethnic homophily, it is argued that the issue of rational choice drives people to choose friends from the same national or ethnic background (Leszczensky & Pink, 2015).

Even though homophily is an important factor in tie formation, it can also lead to segregation.

Namely, when the formed ties are only based on similarities (ethnic, religious, etc.), ideas, knowledge and resources are unequally divided among groups and information is not passed completely through social networks; this event produces issues such as prejudice and conflict among individuals (Smith, Van Tubergen, Maas, & McFarland, 2016).

Homophily is a principle that should be considered when studying social networks and friendship dynamics. People have tendencies to become friends with someone who shares with them a specific quality, such as gender, age, race, nationality, etc. This principle, drives people to choose and form their friendships, their networks, and ultimately, their social world. This, so called, social world has strong significance for the individuals as it shapes or filters the information they receive, their attitudes and their very own experience of social interaction (McPherson, Smith-Lovin, &

Cook, 2001).

In order to study the element of homophily and friendship formation, most scholars turn into the most evident institution: schools. Since schools are “relatively closed institutions” in which the long process of socialization is conducted, it can easily be seen as a ready for analysis institution by scholars. Furthermore, schools contain a large capacity when it comes to the work of policy makers since they can treat a considerable number of social problems through schools (Smith, Van Tubergen, Maas, & McFarland, 2016; Coleman, 1994).

In this research, I intend to study the process of friendship formation at tenth grade (junior year of high school) among students in a school in Stockholm which mostly includes so called “second generation immigrants”. The data used is very well suited for the study of the homophily principle because the start of tenth grade is the start of a new school; therefore, the setting and the social context is completely new for the student. Further, in this particular school the students come from 85 different schools in different areas of the city; that is, a significant number of them do not know each other from previous years of education.

(6)

3

I intend to analyze the existence of homophily principle, with regards to network structure, in gender, religion, nationality and political opinion among the adolescents in this school. In other words, the research question is: How is structure in a new social context associated with gender, national and political homophily among junior high school students?

2 Theory and previous research

The social networks which people establish during adolescent years, have significant effect on their social and cognitive development. Entrance to high school is, generally, a primary level for entering a social context outside of family; a place in which students interact, make friends and form peer groups.

The theoretical discussion in this research is divided into two major parts; theories regarding gender, political, religious and ethnic/national homophily, and theories regrading social network structure. In this section, first, I will have a quick browse on friendship dynamics among students;

second, the specified aspects of homophily will be debated; finally, the network structural theories will be discussed.

2.1 Students and friendship dynamics

A long line of research, during the last decade, has shown the adolescents’ tendency to choose their friends based on similar characteristics. For instance, Hartup (1993) claimed that adolescents choose their friends according to similar age, race, sex and social class; he also mentions the importance of similarity in two general areas of attitude: school related and “contemporary culture”

related. In his opinion homophily in attitude is also a significant factor in adolescent friendship formation. His work on adolescent friendship formation is of great importance as it showed the substantial contribution of adolescents’ social network on their development. However, his claims were quite limited to the notion of similarity as the origin in friendship formation. This view, while partly true, has been modified when studying the process of adolescent friendship formation in a social network context.

Roman (2016) conducted a comprehensive research on homophily with regards to individual background, status of immigration and religious preferences in a school in Stockholm. What is interesting about this research is that she also takes interest in the outcome of school structure on

(7)

4

friendship formation. She also considers religious homophily and students’ background, in addition to ethnic homophily.

Mayer and Puller (2008) also studied the friendship dynamics among students on university campuses. Their aim was to analyze the tie formation process according to similarities in race, socioeconomic background, and ability. Their findings suggested that the fraction of friends with similar ability or parental education is not significantly different from a randomly assigned network; meaning, students were integrated regardless of the mentioned factors. However, the same students were racially segmented, indicating a high leniency towards racial homophily.

Mayer and Puller’s research on friendship formation is of substantial significance since they account for the element of network structure and, most importantly, they emphasize on the subject of network segregation. This notion is of great importance when it comes to homophily as it can lead individuals towards prejudice and stereotypical thinking. That is why researchers have worked on network segregation to be able to come up with a means of measuring and also implementing policies for reducing it (See: Moody, 2001; Bojanowski & Corten, 2014).

2.2 Types of homophily

Merton and Lazarsfeld (1954) distinguished between two major types of homophily: status homophily in which friendship is based on similarity in group membership or individuals’ position in a group and value homophily which is a tendency towards the correspondence of values among friends. McPherson and Smith-Lovin (1987) introduced a network based mechanism for homophily. In their theoretical model, they introduced two mechanisms in which homophilous relations are made: induced homophily and choice homophily. Induced homophily is when a network is formed around a focus and the individuals in the network are homogenous, any random network formed around this social focus would be people with similar attitudes or beliefs.

Therefore, this type of homophilous relationship is merely induced by the systematic features of the network structure. On the other hand, choice homophily in a social network, with heterogeneous individuals, indicated that people choose others similar to them and this similarity is not imposed by the characteristic of the network so therefore people choose their social ties (Ibid). While this theoretical mechanism takes network effects into account, it is rather difficult to distinguish either mechanism when it comes to empirical analysis. For instance, considering a social context such as school, both structural constraints and individual choices can be seen when

(8)

5

it comes to making friends (Kossinets & Watts, 2009); therefore, in this research four types of homophily which can be empirically analyzed are taken into account. These four types are gender, religious, ethic/national and political homophily. Afterwards, I will move on to presenting some arguments regarding the relations of social structure and homophilous relations.

2.2.1 Gender homophily

There is a relatively salient and obvious individual/social identity feature that can breed homophilous relationships and that is gender. As stated by Shrum et al. (1988) one of the most consistent findings in studies regarding the peer relations, is the fact that boys and girls are

“relatively isolated”. Even though, they argue that this isolation varies in different ages, the respective gender specific homophily is present in most age groups.

The fact that children tend to select friends of the same sex is called to attention by numerous researchers (Morimoto & Yang, 2013; McPherson, Smith-Lovin, & Cook, 2001; Mehra, Kilduff,

& Brass, 1998). Gender homophily is strongly present in childhood; however, as children grow and reach adolescence, gender homophily starts to decrease (Shrum, Cheek, & Hunter, 1988); this can be due to the fact that adolescents usually get involved in romantic relationships; therefore, they, generally, prefer interactions with the opposite sex.

On the other hand, gender homophily can also be seen through structural viewpoints. For instance, previous research regarding gender homophily in workplace indicated that men tend to have larger and more gender homophilous networks than women (see: McPherson, Smith-Lovin, & Cook, 2001; Morimoto & Yang, 2013). However, Morimoto and Yang (2013), in their research, discovered the contrary; namely, women’s tendency for building gender homophilous friendships was higher than men’s; therefore, when considering gender homophily, one must take into account the context in which it is being studied. For instance, in the former case, the study was conducted in a workplace in 1992 and 1997, whereas the latter was done in a graduate school in 2013. As the working environment was mainly composed by males, it seems rather simplistic to conclude that men are more likely to form gender homophilous informal networks (see: Morimoto & Yang, 2013).

In conclusion, it can be indicated that even though it varies by age, gender homophily is present in social networks and it is associated with the structure of any specific social context. Therefore, in

(9)

6

this research, in addition to analyzing gender homophily with regards to sex, I will try to see the affiliation of minority/majority on gender homophily (see: section 2.2.5).

2.2.2 Religious homophily

Even though religion is usually considered as an individual matter, it is in fact a social phenomenon (Cheadle & Schwadel, 2012). Religion is produced and reinforced by social interaction and cannot be considered, solely, as an individual aspect as it plays a substantial role in organizing social relations; this fact has been long discussed in sociological texts. For example, Simmel argued that

“the faith which has come to be regarded as the essential, the substance, of religion, is first a relation between individuals” (Simmel, 1905, p. 366). Furthermore, Durkheim (1965 [1915]) also discussed the social significance of religion; also, according to Edling, Rydgren and Bohman (2014) religion is a source of “mutual identification” and social solidarity, in a way that it generates a group of shared symbols and shared history. Therefore, not only religion helps the individual to define and give meaning to his/her surroundings (individual level) (Ibid), it also helps shape and organize the social relations in a specific context.

Needless to say that religion is one of the important factors of making friends in various social contexts. Numerous researches have shown that religious homophily in fact exists in social networks. For instance, Cheadle and Schwadel (2012) studied the friendship dynamics of religion and they concluded that “the youth in this study prefer friendships to those who are religiously similar” (p.1209). Furthermore, Scheitle & Adamczyk (2009) strongly argue that religion is an important factor in relation formations. In any case, there are numerous researches with regards to proving the existence of religious homophily and/or basing the research on religious homophily (see: Wuthnow, 2007; McPherson, Smith-Lovin, & Cook, 2001; Scheitle & Adamczyk, 2009).

Many scholars have analyzed the reasons for religious homophily; some researchers argue around the issue of trust; that is, since religion supplies its followers with routine rituals and moral responsibilities, it gives them a definition for “good people”; therefore, two people of the same religion have the notion that they both have the same behavioral guidelines which leads to mutual trust between them (Windzio & Wingens, 2014; Wuthnow, 2007). Further, some claim that religious homophily is related to rational choice; being a member of any religion requires individuals to sacrifice time and energy for participating in religious rituals and activities (e.g.

going to church, praying, etc.); hence it would be more beneficial for people to interact with

(10)

7

someone of their own religion to preserve time and energy (Stark and Finke, 2000). Moreover, having friends of the same religion is also beneficial in cumulating social capital; numerous researches have shown that establishing friendship with people of the same religion has a positive effect on the social capital of the individual (see: Diehl & Ruckdeschel, 2009; Windzio & Wingens, 2014). As a result, having friends of the same religion is beneficial for conserving and/or cumulating social capital.

The significance of religious homophily has been discussed thoroughly by many scholars.

However, the purpose of the present research does not allow for further discussion around this subject. It is sufficing to say that religious homophily is a salient feature in social networks.

Therefore, in this research the existence of homophily among the adolescents is analyzed and also the effect of the network structure on this kind of homophily is studied.

2.2.3 Political homophily

Research regarding the existence and dimensions of political homophily is quite limited (Huber &

Malhotra, 2013), but there seems to be a considerable tendency among adults to associate with people of the same political opinion (McPherson, Smith-Lovin, & Cook, 2001).

Political homophily can be theoretically distinguished into three types: identity homophily, issue homophily and engagement homophily; identity homophily is the process of ego’s self-definition through political notion. For instance, identification with a political party or with an ideological disposition; according to Huber & Malhorta (2013), political identity is usually formed at early years in life and continues throughout time. That is why it is believed that political identity is a group issue and not related to political opinions.

Issue homophily happens when people get sorted in networks through agreement on a political issue; for instance, economic or foreign policy. Engagement homophily is defined as the tendency to get in touch with people who are similarly committed to their political or civic rights; for instance, their engagement in political activities, number of issues that they are willing to vote for and the amount of interest in joining demonstrations (Huber & Malhotra, 2013).

Political homophily can be considered as a branch of value homophily. As argued by McPherson et al. (2001) value homophily is a tendency to get in contact with people of the same values and opinions. However, there are some issues regarding the concept of political, or in general, value

(11)

8

homophily. Huber and Malhorta (2013) mention induced homophily and convergence as the sources of difficulty in analyzing political homophily. Induced homophily, indicates the contextual limitations of making friends in a network. This notion is widely studied by Scott Feld (see: Feld, 1982). The second dilemma can be explaned via the fact that there are other forms of similarity which are quite stronger in getting people together. For instance, as argued by McPherson et al (2001) race, sex and religion are the most substantial factors when it comes to homophily.

Therefore, it can be easily argued that similarities in political opinions can be formed among friends. In other words, according to Huber and Malhorta, “political views and orientations are malleable over time” (Huber & Malhotra, 2013, p. 8). Therefore, it is rather challenging when trying to determine political homophily, for in fact it could be a mere convergence.

2.2.4 National/ethnic homophily

Generally, when one begins to discuss homophily, the idea of racial or ethnic homophily comes to mind; racial similarity creates the strongest grouping and division in society. The main reason for this argument is salience. That is, when a personal or social characteristic of two individuals are quite noticeable, it is easier for them to form a homophilic relationship (Mehra, Kilduff, & Brass, 1998; McPherson, Smith-Lovin, & Cook, 2001).

Ethnic homophily, or ethnic segregation is an ever present feature of friendship networks; scholars claim that it may be the most salient factor when it comes to friendship formation (McPherson, Smith-Lovin, & Cook, 2001; Smith, Maas, & van Tubergen, 2014; Kao & Joyner, 2004;

Leszczensky & Pink, 2015).

Ethnic or national homophily is considered a highly significant issue, especially among young adolescents (Smith, Maas, & van Tubergen, 2014); that is, while it is an evident matter in social networks, ethnic homophily would eventually lead to ethnic segregation and consequently ethnic or national prejudice which can cause unwanted results (Stark & Flache, 2012). Furthermore, when it comes to the subject of immigration and so called “second generation immigrants”, a lack of native friends would result in low language proficiency in the host country, low market skills and consequently, low chances of finding a suitable career in the future (Ari, 2012; Leszczensky &

Pink, 2015).

Baerveldt et al. (2004) in their work on ethnic boundaries, studied the ethnic boundaries and the preference for intra-ethnic friendship formation in Dutch schools. In their work, in addition to

(12)

9

individual inclinations, they acknowledged the effect of social properties on inter-ethnic and intra- ethnic friendships; they argue that, according to opportunity hypothesis, the ethnic majority, in a social network, has less preference towards inter-ethnic relations than the ethnic minority. Their study, however, concluded otherwise, since the Turkish minority preferred intra-ethnic friendship to a larger extent than the Dutch majority (Baerveldt, Van Duijn, Vermeij, & Van Hemert, 2004).

The point of importance for this research is the consideration of social elements in studying ethnic homophily. Two hypotheses were presented in this paper which were related to social circumstances: opportunity hypothesis and social identity hypothesis. According to the latter hypothesis, people tend to identify themselves with similar groups and differentiate themselves from “others”. This process can, eventually, lead to prejudice and segmentation (Stets & Burke, 2000).

Sanders (2002) discussed the reasons of boundary formation among immigrants in plural societies;

he discussed the importance of race, parents’ education and their influence on the so called “second generation” as the basis for establishing homogenous or heterogeneous relationships. Although, his work seems quite important in studying the effects of social networks in plural societies, it is in a macro scale and without considering other forms of homogeneity/heterogeneity, such as gender, attitude or beliefs.

Another example of research on this subject is the work of Shrum and Hunter (1988). In their research, they analyzed gender and racial homophily among students in school. Using a longitudinal data, they determined that gender segregation started declining from junior high school, but racial homophily started to increase through time. Even though their research is a middle-range study, Shrum and Hunter fail to control for various confounders such as family background and religion; furthermore, their categories of different races seemed quite limited.

However, they presented interesting results which was the foundation for later studies on gender and racial homophily in schools and universities.

Even though ethnic homophily has its negative effects, most adolescent social networks are based on this phenomenon. There have been various policies implemented for reducing ethnic homophily and segregation (see: Stark & Flache, 2012), but this element has been one of the most powerful in the process of friendship formation (McPherson, Smith-Lovin, & Cook, 2001).

(13)

10

There are two arguments regarding ethnic homophily that will be discussed here. First, Smith &

van Tubergen (2014) who base their argument upon cultural similarities. They are of the opinion that “adolescents are likely to prefer friends who have similar leisure activities as it helps them in the development of their identity and self-esteem” (p. 35); cultural similarity makes it easier to find a mutual leisure activity among friends, which can be enjoyed by all of them. Therefore, with respect to cultural similarity, adolescents prefer making friends with someone of the same ethnic background. The second argument is, once again, extracted from rational choice theory. While starting a friendship tie and maintaining it is a rather costly activity, having to form the tie with someone of the same ethnic background usually decreases these costs and is more rewarding.

(Lazarsfeld & Merton, 1954). When ego tries establishing a tie with someone of another ethnic background, he/she would spend the same time and energy and probably not get the same reward as in an ethnic homophilous relationship (Leszczensky & Pink, 2015; Windzio & Bicer, 2013).

In the present research, national/ethnic homophily is analyzed using the respondents’ and their parents’ birthplace. Also, the relations between network structure and national/ethnic homophily will be studied.

2.3 Homophily in social network

In the previous sections, various features of homophily were discussed. In this section, the aim is to present arguments regarding the effects of network structures on homophilous relations.

The theoretical arguments mentioned above mainly concentrate on the existence of homophily in social networks. However, they usually have overlooked the affiliation of the structure of a social network on homophily. Namely, considering a social context such as school, both structural constraints and individual choices can be seen when it comes to making friends and it is rather complicated when these concepts are studied in real life social networks (Kossinets & Watts, 2009).

With regards to the critics to the aforementioned theories, in this research the theory of distinctiveness will be discussed. Distinctiveness theory suggests that “People in a social context tend to identify with others with whom they share characteristics that are relatively rare in that context” (Mehra, Kilduff, & Brass, 1998, p. 442). For instance, two black people are more likely to get in contact in a crowd of white people than in a crowd of black people. According to this theory, the notion of rarity of a characteristic in a social network, makes that characteristic a lot

(14)

11

more salient for the ego and therefore elevates the chances of forming a tie with an alter of the same characteristic.

Distinctiveness theory is quite useful when it comes to situations where individuals have just entered a social context and seek new friends; for instance, the first year of high school. It is during high school that adolescents develop the idea that they are social beings, thus they try to identify themselves with others (Cheadle & Schwadel, 2012). With respect to this notion, one or more of personal characteristics of the adolescents will be rare and consequently more salient. The argument here is that adolescents will identify themselves with that specific feature of their identity and hence form friendship based on rare similarities.

In conclusion, I hypothesize that gender, religious, political and national/ethnic homophily is promoted when these phenomena are rare in a social context (i.e. in the studied school).

In the present research, however, I intend to study friendship ties in relation to gender, religion, national and political homophily. In addition, it is in this study discussed if homophily, with respect to the mentioned aspects, also is influenced by the number of individuals sharing the specific characteristics. In other words, accompanying the theories supporting gender, political and national homophily, I will also test the plausibility of distinctiveness theory and Blau’s theorem on social structure. Blau discusses the quantitative characteristics of social structure in analyzing inter-group association among majority and minority groups in a social network. In his opinion, Firstly, the intergroup involvement of the minority exceeds the majority; secondly, the probability of intergroup relations decreases with increasing group size and thirdly, the more a majority discriminates against a minority, the smaller the difference between the majority and the minority intergroup relations (Blau, 1977).

Furthermore, this research argues that the study of social networks among adolescents is quite important. Cheadle et al (2012), for instance are of the opinion that while adolescence is the period where individuals are highly resistant to the fact that they are social beings, but at the same time, they are entering a social space which is separate from home. Therefore, in this research, it is argued that the mechanism adolescents use to find and make friends, will continue throughout their social life.

(15)

12

Overall, the importance of this research lies in two facts; first, the argument that social structure of the school is associated with the homophilic relations at school; second, the importance of the friendship dynamics in teenage years.

3 Research question

In this study, the aim is to the study the probability of existence of friendship ties with accordance to the homophily principle. The types of homophily analyzed are gender, religion, nationality/ethnicity and political opinions. In addition to this subject, this research intends to see the association between homophilous friendship ties and the prevalence of the variables that the homophily is based on. Therefore, the results will be analyzed with the distinctiveness theory in mind, to see if individuals with minority characteristics are more prone to form ties with each other.

4 Data

The data which is going to be used in this research is assembled by Sara Roman and is gathered from the cohort that started grade 10 in upper secondary education in a school in Stockholm at three time points (October 2012, May 2012 and January 2013) during the academic year 2012/2013. It includes a total of 115 individuals and contains information about the social network in the cohort as well as individual characteristics of the students. The data was collected by Roman by means of questionnaires that were distributed to the students in their classrooms during class time. Each classroom was visited at least twice. To reduce the ethical issues of the survey, the students were thoroughly informed about the detail of the research, beforehand; further, the identity of the school and the respondents were protected; also, participation in the research was voluntary and students could withdraw from filling out the questionnaire anytime they wanted.

The data collection was approved by the regional committee for ethical vetting (see: Roman, 2016).

The school was not chosen randomly; Roman had chosen this specific school since it contained a lot of students who were born in another country or were, so called, “second generation”

immigrants. Therefore, the data and the analyses, here, are not necessarily representative of all of the adolescents in Stockholm.

(16)

13

Individual characteristics include information about demographic characteristics (such as, age, gender, etc.), ethnic and religious background, political orientations, status of mental and physical health and spare time activities.

Network data are gathered by asking the students to name three of their closest friends in school and discovering the quality of their friendships by asking whether they spend time with their friends outside of school and if they can talk to these friends about their personal problems. The cases are then coded into numbers and they are so called “edge lists”, basically a list of pairs that have a relation. Every row represents a pair that was reported as a response to the question "with which other students in the cohort have you spent time after school during the last three weeks?"

(Roman, 2016).

Political preference measures are questions regarding the respondent’s tendency towards left or right wing parties and also the political parties they would vote for if there was an election.

Religious affiliation was measured with questions regarding the importance of religion for the respondent and his/her parents, and also the respondent’s own religion. National affiliation, as contentious as it is, was measured by the respondent’s parents’ birth country.

In consideration of the fact that this study focuses on the formation of immediate ties, longitudinal data is not necessary for analyzing the research question. Therefore, only the first wave of the dataset is going to be used for analysis. Furthermore, some of the questions that were necessary for the analysis, were not asked from the respondents in the second and/or third wave of data gathering; hence, those waves were not considered in the analysis.

4.1 Dependent & Independent variables

The dependent variable is whether two students know each other or not. Hence, the subject under study is the possible link which can exist between each pair of individuals in the data (the extraction of the dependent variable is further discussed in section 5). The independent variables are being the same gender, same religious affiliation, having the same religion, mutual significance of religion for the respondents’ parents, same birth country for at least one of the parents.

For analyzing homophilous relations based on gender, a binary variable was developed to show whether two persons in the two by two network data are of the same gender or not. Second, two variables were created: 1. At least one in the pair is male 2. At least one in the pair is female to see

(17)

14

the difference in friendship formation between boys and girls; and third, two binary variables were made for both male and female friendships.

Regarding religious homophily, in the questionnaire participants were asked to indicate their religion. The answers included no religious affiliation, Christian, Jewish, Muslim, and other. The

“other” option was used as the reference variable in the analysis. In the network data a binary variable was created to see whether two students were of the same religion or not.

The second variable is the importance of religion for the participant. Participants were asked about the significance of religion for them. The answers ranged from “very important” (4), “fairly important” (3), “not very important” (2) and “not at all important” (1). These answers are recoded to important (choices 3 and 4) and not important (choices 1 and 2). The same importance for religion is created to see whether mutual significance of religion is important in friendship formation.

With respect to parents’ religiousness, a merged variable was made that indicated the significance of religion for at least one parent from both sides of the network data. It is worth mentioning that since the frequency of the data is not quite high, this merged variable was made to make the models more sensible.

With regards to the effect of network structure, the “same religion” variable was divided into three categories: nonreligious friendships, Christian friendships and Muslim friendships. These categories indicate the possibility of friendship if both students are of same religion. However, the variable for Jewish friendship was omitted since there were no such cases in the data that both students were Jewish and friends.

For measuring national/ethnic homophily, the countries in which the students or the parents of the students were born, were categorized into five regions; these regions include Europe, Middle East and North Africa, Sub-Saharan Africa, Sweden and other regions. Other regions include students (or their parents) born in Bangladesh, Bolivia, Colombia, India, Pakistan, Peru, Sri Lanka, Thailand, Turkey, and Uzbekistan; the reason for grouping these countries as “other” was mainly due to the fact that the frequency for students from these countries were low. These categorizations were based on Sara Roman’s categorization of the birthplace variable, with the exception of

“European” categorization (see: Roman, 2016). Afterwards, three sets of binary variables for

(18)

15

mutual birthplace were created; one for if two students shared a birthplace, another to see if at least one parent of two students shared a birthplace and a third, to see if both parents of two students had the same birthplace.

Moreover, for seeing the effect of the structure of the existing social network on different types of national/ethnic homophily there were nine variables created. The first four was to analyze the probability of friendship among those born in the same region; the other are five variables, also bianry, to see the effect of students from different regions on the possiblity of them making friends.

Therefore, five variables were created for four regions and one “other” regions. These variables are to see if being born in a region affects the probablity of making friends. The results for national or ethnic homophily and the probability of being friends are shown in table seven.

For identity homophily, the operationalized predictors were being interested in politics, voting for a political party if there were a hypothetical election and political orientation (right and left wing).

The focus in this research is whether or not students are similar in these indicators which would imply analogous political identity. Nonetheless, from all these variables only voting for a similar political party was used in the model, due to lack of sufficient cases and a lot of missing values.

Engagement homophily is divided into three political activities: engaging in political meetings, participating in demonstrations and being a member of a political organization. These variables were, at first, merged into a single index of political engagement. However, since there were a high number of missing values in the data, the engagement index was dropped and only participating in political meetings were used in the regression models (see table eight)¹.

Control variable is a binary variable indicating if two students attended the same school in ninth grade. This variable was controlled to see if possible previous friendship is also a factor in friendship network formation. Furthermore, same gender friendship was also used as a control variable in all regressions (except the one where gender was the main interest, obviously). Also,

1 Within the subject of issue homophily, I concentrated on the views of students regarding the issue of migration.

With respect to this issue, an index was created regarding their opinion about government’s policy when it comes migrating to Sweden and accepting refugees. These policies included the money spent on immigrants, the quantity of immigrants and refugees moving to Sweden, whether immigration is a beneficial phenomenon in Sweden and the responsibility of rich countries for accepting refugees. This index was also dropped due to lack of data.

(19)

16

considering the effect of propinquity on tie formation, I have also controlled for being in the same classroom.

5 Method of research

There were two sets of data at my disposal. One containing network data, which is a list of pairs who have established friendships, and one containing all respondents’ answers to the questionnaire. For the purposes of the study, the two datasets were merged in the form that every possible relationship between each student was put together, then a binary variable was defined to show which pair had a relationship in real life. Afterwards, all the other variables were merged into this dataset for both the first column respondents and the second column respondents to be able to compare their answers and find out the similarities and the differences. Conclusively, I had two datasets, one containing information regarding the real descriptive characteristics of the research group, and one consisting of network information and the comparison of their answers.

In order for the reader to better understand the method of research, two tables are presented below.

The table on the left is a group of hypothetical students in a school and the table on the right is a matrix of all possible ties which can be formed. The value zero means that there are no ties between the two individuals and value one means that a dyad (tie) exists. The black cells in diagonal of the matrix indicates that an individual, per definition, cannot befriend him/herself (i.e. there are no entries in the diagonal). Further, since the matrix is symmetric (e.g. if there is a relation between a and e, there must be a relation between e and a, etc.) the part of the matrix which is above and to the right of the diagonal holds no information, since this information should be redundant. As seen, there are 45 possible ties in this network of 10 individuals and 13 of them are realized as friendship ties. This is how the dependent variable was generated in the research.

(20)

17

Table 1. How the network data (dyads) were produced

Respondents a b c d e f g h i j

a a

b b 0

c c 0 0

d d 0 0 0

e e 1 0 1 0

f f 0 1 0 0 0

g g 0 0 0 0 0 1

h h 0 1 0 0 0 1 1

i i 1 1 0 0 1 0 0 0

j j 1 0 0 1 0 0 0 1 0

Hence, the entity under study is if a friendship link exists between each pair of individuals. Table two shows a small excerpt of how the analyzed data generated from the matrix looks like. The subject under study (the dependent variable in the subsequent analysis) is the variable called friends, in the table below. Now, variables such as if the two students are of the same gender or not can be added.

Table 2. An excerpt of how the data looks like

Individual 1 Individual 1 Friends Same sex

a b 0 1

a c 0 0

a d 0 0

a e 1 1

a f 0 0

a g 0 1

a h 0 0

a i 1 1

a j 1 0

b c 0 0

b d 0 0

b e 0 1

As pointed out, the units of analysis in the research are the possible dyads, the ties which are or are not realized between two students in the school. The analysis is seeing how the odds that a tie actually exists is associated with e.g. whether the pair of individuals are of the same sex or not.

(21)

18

Since the dependent variable is binary, logistic regression model was used to test the hypotheses.

Analysis were conducted using statistical software Stata (Version 13.0, StataCorp, College Station, Texas).

The chosen school for the study has special characteristics. It is not a representative of other schools in the Stockholm region. Further, the data is from all the students who were present in the school; that is, the analysis is conducted with the research population and not a research sample.

Therefore, significance of the odds ratio is not relevant in this research. That is why instead of significance, robustness of the coefficients is considered in the analysis. A robust coefficient, in this research, is defined as a coefficient with a relatively low standard error. Therefore, in the analysis, the focus is on the relative value of the standard error to the odds ratio to see the robustness of the coefficient.

6 Results

A total number of 111 students were analyzed in this research. As seen, this number is lower than the original amount, which was 115. This is because in the first wave, there were four students missing (they were absent on the day or were not yet registered at school). Table three gives an overview of different characteristics of the students in this school.

Table four shows the distribution of the dependent and independent variables. These variables are mostly binaries to measure if two students are of the same characteristics. It is worth mentioning that the high number of n in this table is due to the fact that the frequency for the independent variables were extracted from the crossed dataset. That is the dataset in which the data for the first wave were merged and crossed with the network data.

(22)

19

Table 3. Frequency distribution of gender, religion, birth place (mother, father and the student) and political preference and engagement.

Frequency Percent Cum. Perc Gender

Male 43 38.74 38.74

Female 63 56.76 95.5

Missing 5 4.5 100

Religion

Non-religious 17 15.32 15.32

Christianity 39 35.14 50.45

Judaism 1 0.90 51.35

Islam 41 36.94 88.29

Other 8 7.21 95.50

Missing 5 4.50 100.00

Father's birthplace

Europe 9 8.11 8.11

Middle East and North Africa 22 19.82 27.93

Sub-Saharan Africa 33 29.73 57.66

Sweden 25 22.52 80.18

Other 13 11.71 91.89

Missing 9 8.11 100.00

Mother's birthplace

Europe 10 9.01 9.01

Sweden 29 26.13 82.88

Other 11 9.91 92.79

Missing 8 7.21 100.00

Student's birthplace

Europe 3 2.70 2.70

Sweden 68 61.26 85.59

Other 8 7.21 92.79

Missing 8 7.21 100.00

Political party²

Moderaterna 16 14.41 14.41

Vänsterpartiet 5 4.50 18.92

Socialdemokraterna 32 28.83 47.75

Miljöpartiet 3 2.70 50.45

Blank vote 2 1.80 52.25

Would not vote 5 4.50 56.76

Missing 48 43.24 100.00

Engagement in political meeting

Yes 21 18.92 18.92

No 71 63.96 82.88

Missing 19 17.12 100.00

n 111 100.00

2 Name of the political parties are in Swedish. No other political party was mentioned.

(23)

20

Table 4. frequency distribution of similarity between all possible pairs of students and observed friendship formation.

Frequency Percent Gender homophily

Same gender

Yes 2856 45.39

No 2709 43.05

Missing 540 8.85

Total 6105 100

Religious homophily Same religion

Yes 1725 27.42

No 3840 61.03

Missing 540 8.85

Total 6105 100

Same importance

(religion)

Yes 2773 44.07

No 2583 41.05

Missing 749 12.27

Total 6105 100

National/ethnic homophily

Same birthplace (student)

Yes 2299 36.54

No 2954 46.95

Missing 852 13.96

Total 6105 100

Same birthplace (at least one parent from each side)

Yes 629 10

No 4624 73.49

Missing 852 13.96

Total 6105 100

Political homophily

Vote for the same party

Yes 640 10.17

No 1313 20.87

Missing 4152 68.01

Total 6105 100

Engage in political meetings

Yes 210 3.34

No 3976 63.19

Missing 1919 31.43

Total 6105 100

Observed friendship formation Friends with

each other

Yes 300 4.91

No 5805 95.09

Total 6105 100

6.1 Gender homophily

As seen in table five, the same sex friendship has an odd above one; that is, holding everything else constant, the odds of two students knowing each other is higher when they are of the same gender, compared to two students of different gender. Compared to the odds ratio, the relatively low standard error of this odds ratio indicates that the coefficient is robust.

However, while the odds ratio of Model 1 is 2.51, the odds ratio of Model 2 is 1.31. This could be because in the first model the probability of males making friends is calculated, whereas in the second model the probability for females making friends is estimated. The above one ratio of “one

(24)

21

in the pair is male” and the below one ratio of “one in the pair is female” is an indicator that a group of friends is more likely to form if it includes boy students than girl students. This is in line with Shrum’s theory that boys tend to form large networks and girls have leniency towards forming smaller networks (Shrum, Cheek, & Hunter, 1988). Moreover, distinctiveness theory is supported when comparing male and female friendship coefficients in models 3 and 4 (see: Mehra, Kilduff,

& Brass, 1998). As seen in model 3 and 4 in table five, individuals are more prone to form same- sex friendship ties than mixed-sex friendship ties. This goes for both men and women. However, the odds for same-sex friendship (compared to mixed-sex friendship) is lower for women than men. Further, a Chi-squared test indicate that this difference is highly robust. This number also indicates the effect of distinctiveness theory on friendship formation when it comes to sex.

The rather intriguing part of this table are odds ratios for being in the same class and being in the same school at ninth grade. As seen, the odd ratio for same ninth grade, has higher standard error when the “same class” control variable is entered in the model. The standard error of the “same class” is relatively small compared to the predicted odds ratio, which shows the robustness of the variable.

Table 5. Maximum likelihood logistic regression of gender homophily. Dependent variable is the odds of two students being friends. Control variables are a binary if they were in the same school and a binary if they were in the same classroom. Results

are presented in odds ratio. Standard errors in parentheses.

Model 1 Model 2 Model 3 Model 4 Independent variables

Same sex friendship 2.516 1.310

(0.348) (0.165) One in the pair is male (at least) 1.921

(0.272)

One in the pair is female (at least) 0.520 (0.074)

Female friendship 1.268 1.310

(0.153) (0.165)

Male friendship 2.377 2.516

(0.31) (0.348) Control variables

Same ninth grade 2.297 2.297 3.737 2.297

(1.161) (1.161) (1.754) (1.161)

Same class 8.278 8.278 8.278

(0.935) (0.935) (0.935)

Log Maximum likelihood -1240.69 -1240.69 -1432.09 -1240.69

n 5562 5562 5562 5562

(25)

22

6.2 Religious homophily

In this type of homophily, like national or ethnic homophily, having the same religion is not the only important factor in analysis; since the data deals with junior high school students, the element of religion among the parents of the students is also considered in the regression models.

Table 6. Maximum likelihood logistic regression of religious homophily. Dependent variable is the odds of two students being friends. Control variables are a binary if they were in the same school and a binary if they were in the same classroom. Results

are presented in odds ratio. Standard errors in parentheses.

Model 1 Model 2 Model 3 Model 4 Model 5

Independent variables

Same religion 1.445 1.432 3.518 3.588 2.435

(0.163) (0.164) (1.973) (2.012) (1.455)

Same importance (student) 1.284 1.254 1.285 1.302 1.334

(0.144) (0.152) (0.146) (0.148) (0.158)

At least one parent from both sides (merged) 1.075

(0.153)

Same religion friendship^*

Nonreligious 0.358 0.356 0.479

(0.228) (0.226) (0.323)

Christian 0.371 0.362 0.552

(0.213) (0.207) (0.337)

Muslim 0.443 0.431 0.666

(0.253) (0.246) (0.405)

Control variables

Same ninth grade 6.000 3.917

(2.446) (1.724)

Same class 7.718

(0.904)

Log Maximum likelihood -1328.525 -1328.397 -1326.89 -1319.563 -1154.603

n 5148 5148 5148 5148 5148

*Reference variable is "other religions"

As seen in table six, having the same religion is an important factor in making friends. In almost all the models represented in the above table same religion variable is has an odd above one and relatively low standard errors. However, this is not the case with the odds ratios indicating how the association between realizing a friendship link and religious similarity vary between different religions. In the last model, as the control variable, “same class”, is entered in the model, the odds ratio for having the same religion decreases and the relatively high standard error suggests a fragile coefficient. Therefore, while being in the same religion at first seems to be an important factor in forming friendship ties, being in the same classroom, as a social focus (see: Feld S. , 1981) is more

(26)

23

substantial (at least for all other religions than Islam, see further interpretations of the results below)³.

In the second model, I have controlled for the importance of religion among parents; more precisely, the similarity of significance (or non-significance) of religion in between at least one parent from both sides. The odds ratio for this variable indicates that importance of religion in the family is not an important factor in tie formation among the students. However, the odds ratio for importance of religion for the students themselves in the other models shows that the importance of religion for the individual is a factor for friendship formation

The network structure effect, as mentioned, has been studied through dividing same religion friendships into three categories. The net effect for each religion (each category) is the odds ratio of same religion variable plus the odds ratio for that religion. For obtaining the net effect and consequently seeing which religion increases the odds of being friends the linear combinations of estimates was calculated.

The results for three linear combinations suggest that since only the odds ratio for Muslim students was high and robust (comparatively lower standard error). The ratio was 1.53; that is, the odds that a tie exists between two random Muslim students is 1.53 times higher than the odds that a tie exist between two random students of different religions. Christianity and non-religiousness does not give a higher probability of intergroup friendship formation compared to friendship between different religious groups.

6.3 National/ethnic homophily

There are several factors to be considered in this type of homophily. First, the existence of national or ethnic homophily among students in the school. Second, the effect of parents’ birthplace on this form of homophily and Third, the effect of the actual social structure on the existing homophily.

3 For seeing the association of social focus and probability of friendship, an interaction between same classroom and same religion is needed; however, due to data limitation, this analysis could not be performed.

Homophily and friendship dynamics

Homophily and friendship dynamics

An analysis of friendship formation with respect to homophily principle and distinctiveness theory

Author: Sepehr Saeidibonab

Abstract

Keywords

Table of Contents

1 Introduction

2 Theory and previous research

3 Research question

4 Data

5 Method of research

6 Results