Analyzing Social Media Data and Web Networks

M. Cantijoch, R. Gibson, S. Ward, M. Cantijoch, R. Gibson, S. Ward

• Publisher: Palgrave Macmillan
• Year: 2014
• ISBN:
  9781137276773
• Topic: Ciencias sociales
• Subtopic: Estudios de medios

As governments, citizens and organizations have moved online there is an increasing need for academic enquiry to adapt to this new context for communication and political action. This adaptation is crucially dependent on researchers being equipped with the necessary methodological tools to extract, analyze and visualize patterns of web activity. This volume profiles the latest techniques being employed by social scientists to collect and interpret data from some of the most popular social media applications, the political parties' own online activist spaces, and the wider system of hyperlinks that structure the inter-connections between these sites. Including contributions from a range of academic disciplines including Political Science, Media and Communication Studies, Economics, and Computer Science, this study showcases a new methodological approach that has been expressly designed to capture and analyze web data in the process of investigating substantive questions.

Part I

Structure and Influence

1

Political Homophily on the Web

Robert Ackland and Jamsheed Shorish

Homophily is a central concept within sociological research and describes the preference of actors in social networks to form ties on the basis of shared attributes, such as gender and race, as well as subjective characteristics such as political affiliations and desires for certain consumer goods. The study of homophily can provide important insights into the diffusion of information and behaviours within a society and has been particularly useful in understanding online community formation given the self-selected nature of the information consumed.

In this chapter, we introduce the concept of homophily and show that in order to accurately measure homophily, one needs to control for factors such as group size and the existence endogenous network ties. We then provide a discussion of how Web data can be used to advance research into political homophily, which is the phenomenon whereby people seek out others who share their political affiliation. We contend that the Web provides several unique opportunities for political homophily research, but there are associated challenges that must be taken into account.

Any research involving Web data for understanding social and political behaviour should first establish that the observed online behaviour is a valid or meaningful representation of its offline counterpart – this has been referred to as the requirement that there be a ‘mapping’ between the online and offline world (Williams 2010), or that the online data have ‘construct validity’ (Burt 2011). Our chapter discusses the construct validity of Web data for political homophily and offers three tests of such validity.

One of the tests is that the Web data display differential homophily, where communities exhibit idiosyncratic tie preferences within their community, rather than a uniform tendency of flocking for the population at large. In this context, we revisit the well-known ‘Divided They Blog’ 2004 weblog network data of Adamic and Glance (2005) and show how a particular statistical social network analysis technique (exponential random graph modelling or ERGM) can be used to quantitatively characterise political (uniform and/or differential) homophily in the blogosphere. We use the VOSON (Virtual Observatory for the Study of Online Networks) hyperlink network research tool to construct a network of the US political blogosphere in 2011 and assess how political homophily has changed since 2004. Complementing the traditional insights gained from qualitative network visualisation techniques, we show that differential homophily has become more characteristic of the political landscape exhibited by weblogs from 2004 to 2011.

Assortative mixing and homophily

Assortative mixing in social networks refers to a positive correlation in the personal attributes (age, race, ethnicity, education, religion, socioeconomic status, physical appearance, etc.) of people who are socially connected to one another. There is strong evidence that people assortatively mix when it comes to forming friendships, marriages and sexual partnerships – this is, the ‘birds of a feather flock together’ phenomenon (see, for example, McPherson et al. 2001 for a review).

With regard to marriage, research reviewed in the aforementioned McPherson et al. (2001) shows that Americans exhibit a preference for ‘same-race alters’ far in excess of preference for similarity based on other characteristics such as age and education. There is also evidence that people assortatively mix on the basis of political preferences, and recent work by Alford et al. (2011) shows that the correlation between spouses’ political attitudes is larger than for other personality and/or physical traits.

A tendency towards politically homogeneous social interactions affects the degree of exposure to different political perspectives, and this can have an impact on, for example, the operation and effectiveness of municipal councils and civic associations (‘crosstalk’; see, for example, Weare et al. 2009). In addition, the concept of assortative mixing assists in the classification of political networks and the factions they represent, as discussed in, for example, Kydros et al. (2012). And in the wake of the mass shooting tragedy in Newtown, Connecticut, in December 2012, network visualisations have permeated mass-media outlets to such an extent that the gun control debate in the United States is also best understood – and presented – from an assortative mixing perspective (cf. Stray 2013).

Assortative mixing is thus an important and fundamental aspect of social networks and consequently has received much research attention. However, assortative mixing is simply an empirical measure that describes the structure or composition of a social network (that is, which types of nodes have a higher probability of being connected) – it says nothing about the exact processes that have led to the formation of a particular social network. While it is reasonably easy to measure the level or extent of assortative mixing in a social network, it is much more difficult to discover why people are mixing on the basis of shared characteristics. We outline three main reasons why a given social network might exhibit assortative mixing. ¹

First, there might be homophily – a term first coined by Lazarsfeld and Merton (1954) which refers to people forming a social tie because they prefer to be connected to someone who is similar to themselves. Homophily can, in principle, operate with respect to any attribute – physical characteristics such as race and gender, ‘cultural preferences’ over books and music, and political attitudes. However, when the person has choice over the attribute then it is harder to distinguish whether ‘birds of the feather are flocking together’ (attributes are influencing friendship formation) or whether someone is becoming more like their friends (friendships are influencing attitudes and preferences).

Second, there are opportunity structures that influence social tie formation. In particular, group size is important: the smaller a particular group (for example, racial category) the more likely (all other things considered) that its group members will form social ties outside of the group (Blau 1977). If group size is not controlled for, then there can be erroneous conclusions about the ‘homophilous’ behaviour of different groups. Independent of group size, the propinquity mechanism can also influence whether two people form a social tie – these shared ‘foci effects’ might relate to spatial proximity (for example, living in the same neighbourhood) or shared institutional environments (for example, working in the same organisation) – see, for example, Feld (1981) and Mouw and Entwisle (2006).

Finally, there are endogenous network effects, which are mechanisms that are not directly related to the attributes of individuals, but exert influence on social tie formation. First, there is the process of sociality: two people might become friends simply because they are both social people and like to form lots of social ties. Second, social networks tend to exhibit two properties: (1) reciprocity – if A extends the hand of friendship to B, there is good chance that B will reciprocate the friendship; and (2) transitivity – the tendency for friends-of-friends to become friends (this is referred to as triadic closure). It has been argued by the proponents of balance theory (see, for example, Davis 1963) that the social norms reciprocity and transitivity reduce the social and psychological strain that arises from unreciprocated ties and being in a situation where one’s friends are not themselves friends.

Reciprocity and transitivity can also impact on the measurement of homophily: if a particular group does have a genuine preference for forming in-group ties, then this preference will be amplified by the processes of reciprocity and transitivity. Furthermore, if there are differences between the extent of reciprocity and transitivity across different social groups (for example, one race has a cultural tendency to reciprocate friendships or introduce friends to each other), then this may obscure the cross-group comparison of homophily.

The problem for researchers studying homophily is that both opportunity structures and endogenous network effects can ‘mask’ the true level of homophily in a social network. Currarini et al. (2009) demonstrate one approach for constructing measures of homophily where differences in group size are controlled for. Below, we demonstrate a statistical technique that provides estimates of homophily in a social network, where both group size and balance mechanisms are controlled for.

The Web and political homophily

It was mentioned in the previous section that there is evidence people are more likely to be socially connected to other people who share a political affiliation. This section considers how research using Web data is providing new insights into political homophily. First, we discuss how Web data from social network sites (such as Facebook), blogs and microblogs (such as Twitter) provide several key opportunities – and sometimes challenges – for studying political homophily. Next, we examine evidence for whether political homophily exists on the Web and if so, whether it has characteristics that are similar to political homophily in the offline world. Finally, we consider whether Web data may provide insights into how political attitudes are formed, and we also ask the question: might the Web itself contribute to political homophily?

Opportunities for studying political homophily

Web data provide several opportunities and challenges for social networks research. This section provides a summary, with particular focus on research into political homophily. First, Web data are created in a naturalistic environment, and so there may be less problem of recall error and respondent burden with regard to the collection of social tie data. However, there is the additional problem that all relevant social network ties may not be observable to the researcher. If the research aims to, for example, understand the role of social networks in political preference formation using Facebook data, it may be that significant offline social ties are not represented in the data (for example, friends and family who are not on Facebook).

The naturalistic nature of Web data also poses both challenges and opportunities for collecting data on the key attribute of interest: political affiliation. Focusing once again on the example of Facebook, data on the political preferences of an individual will only be available to the researcher if the Facebook user has decided to fill out the appropriate profile fields, and there may be something different about such individuals that make them less representative of the population under study (they may be more politically motivated or active than the average person in the population). Also there is a potential issue of the accuracy of the political preference data: with certain populations of study (for example, university students), there may be social pressure to display a particular political affiliation in the Facebook profile that doesn’t reflect the person’s true political preferences.

With this caveat in mind, a second major advantage of Web data is that it is often possible to collect complete network data (where links between all actors in the network are recorded). This allows the computation of both node-level metrics (such as degree, betweenness and closeness centrality) and network-level metrics (such as density and centralisation) that may be important to understand the phenomenon being researched. Whole network data are necessary for being able to model ‘supra-dyadic’ phenomena, that is, where it is not just the direct ties between a person (‘ego’) and his or her social contacts (‘alters’) that are important in understanding that person’s behaviour or outcomes, but also the connections between alters themselves (and, indeed, connections between people more two or more degrees of separation from ego).

However, while Facebook might provide an opportunity to collect complete network data for a particular population, for example US college students, it needs to be recognised that this population cannot be representative of the general population. Hence, conclusions that are drawn about the extent of political homophily among college students may not be able to be generalised to a wider population.

It also needs to be recognised that for some Web data sources, it may not be feasible to identify a bounded population from which to collect complete network data. For example, in research on the blogosphere, investigators often need to use ‘snowball’ sampling in order to build their network data because there is no sampling frame from which to randomly sample observations. Non-probability sampling techniques such as snowball sampling typically cannot be used to make inferences about population statistics – it may not be valid to make strong conclusions about the extent of political homophily in the political blogosphere, for example, when snowball sampling has been used. Further, the fact that snowball sampling may be required to construct the complete network may also make it difficult to assess the population share of, for example, conservative and liberal US political bloggers, and this can have implications for the measurement of political homophily (see the following paragraphs).

A third and final major advantage of Web data for studying political homophily is the fact that many Web datasets are longitudinal: research subjects’ political and other attributes are recorded over time, as are their social network data. This opens up the possibility for studying how political preferences and social networks co-evolve over time, allowing potential insights into the social processes underlying political preference formation.

However, there are associated challenges involved with the use of time stamped Web data in the context of research into political homophily. First, it has been noted that while Web environments such as Facebook provide useful data for social tie formation, they are less useful as sources of data on tie dissolution: people do not tend to ‘unfriend’ in Facebook, because the costs of maintaining a Facebook friendship are minimal. Noel and Nyhan (2011) have shown that homophily in Facebook friendship retention can confound causal estimates of social influence. The implication is that one needs to be cautious when using longitudinal Web data (for example, from Facebook) for researching how social ties impact on political preferences, since homophily in friendship retention (people with shared political preferences are less likely to unfriend one another) can exert upward bias on estimates of the extent to which political preferences are transmitted through social networks.

Another potential problem with time stamped social network data (both online and offline) is that people can drop out of the sample over time, and if the rate of attrition is related to the political behaviour of interest then differential rates of attrition can impact on research findings. While differential rates of attrition may not be a concern when one is studying mainstream political behaviour, it may be more of a problem if the focus of study is on radical or extreme behaviour. That is, a Facebook user who has recently started engaging in radical political behaviour might be more inclined to change his or her profile to private and thus become invisible to researchers, or stop using Facebook entirely, and this will impact research into social influence and pol...

Table of contents

• Cover
• Title Page
• Copyright
• Contents
• List of Figures and Tables
• Notes on Contributors
• Introduction: The Importance of Method in the Study of the ‘Political Internet’
• Part I: Structure and Influence
• Part II: Contents and Interactions
• Part III: Mixed Methods and Approaches for the Analysis of Web Campaign
• Index