Abstract
As researchers took note of the emerging ubiquity of new media, they predicted how digital technologies would facilitate an increasingly fragmented audience. New media (i.e. technologies with online capabilities) were observed to possess previously unmatched levels of content options and audience control over consumption. Many researchers have since observed that the current audience landscape has not reached previously anticipated degrees of fragmentation, leading to questions about potentially mitigating factors. In this study, we utilize emerging network analytical procedures to examine the role of interpersonal relationships in both exacerbating and mitigating audience fragmentation. We find support for the notion that social ties can mitigate fragmentation with regard to particular types of media use, notably, those most narrowly defined. Implications of this cross-disciplinary study are discussed.
One of the most striking features of the new media environment is the ease with which people can select the precise content they wish to consume. Multi-channel television systems and the Internet have given audiences substantial power to tailor their media consumption (Chaffee and Metzger, 2001). Observing this emerging media landscape near the start of the 21st century, some scholars predicted that people would pursue narrow content domains, disregarding others (e.g. Sunstein, 2001). Instead of being collectively exposed to a uniform diet of messages (e.g. via a prime-time network lineup or limited set of stories in a daily newspaper), people would separate themselves into isolated pockets of consumption.
In the intervening years, many researchers have continued to predict the ongoing fragmentation of media audiences (e.g. Mancini, 2013). When faced with plenty, people choose the programming and information that most closely fit their personal needs and preferences (e.g. Taneja et al., 2012; Tewksbury, 2005). There is contemporary evidence to suggest that some demographic characteristics—those tied to personal interests (e.g. age and education)—are associated with selective media consumption patterns across varying media platforms (Kim, 2016). Recently, communication researchers have begun to question more actively the role of network boundaries, asking how far fragmentation can go (e.g. Webster and Ksiazek, 2012). Given that there are millions of content options on the Internet and hundreds of channels on television systems, is there a practical limit to the extent people will pursue idiosyncratic exposure patterns? Or is a dramatically individualized media experience the inevitable result of current trends?
In this study, we utilize emerging network analytical procedures to examine the role of interpersonal relationships in both exacerbating and mitigating audience fragmentation. Reardon and Rogers (1988) argued that social ties and media consumption are frequently, and erroneously, treated as a dichotomy of social influence. Lines of research on these two approaches occur predominantly in isolation from one another. They argued that researchers should focus more closely on the interaction of these forces as they influence each other as well as many of the beliefs, attitudes, and behaviors that define social life. Webster and colleagues have followed this route via research examining the manner in which language (Ksiazek and Webster, 2008) and geographic location (Taneja and Webster, 2016) influence patterns of media use. They find that these sociocultural attributes predict specific forms of consumption clustering. This type of research has demonstrated more macro-level social constraints of fragmentation. In this study, we explore how an individual’s micro-level, specific social ties may constrain fragmentation.
Thus far, the bulk of the research on fragmentation has tended to treat people as data points pulled from their social contexts (e.g. Tewksbury, 2005; cf. Wonneberger et al., 2009). Much of this research treats people in an atomized fashion, one in which individuals make choices purely on the basis of internal predispositions. Where limits to these tendencies are observed, they are frequently attributed to structural features pertaining to the individual (taken in isolation) or the media environment. Examples include variations in audience availability (Wonneberger et al., 2012) or disparities in content promotion (e.g. long tails; Elberse, 2008), respectively. The former is an audience-centered attribute referring to the general amount of time consumers allot for media use. The latter is a media-centered attribute facilitating content awareness. Both, however, are determinants of the individual’s unique potential to be present for specific content consumption.
This research is necessary, but it tells us little about how media selection operates in the context of concrete, day-to-day social life (Webster and Wakslag, 1983). If our social environment is unconnected with our media selection, highly individualized fragmentation is a plausible outcome of increasing proliferation of media options. After all, if the preferences and experiences of one’s friends, family, and neighbors play no role in one’s media selection, one’s individual preferences should exert all the more power over media choice. However, if people choose media content at least partly on the basis of how they interact with other people (Webster and Wakslag, 1983), accounts of highly atomistic fragmentation seem less plausible. We look at relatively tight social relationships within a bounded community to determine the extent to which an individual’s interpersonal relationships contextualize media selection.
Audience selectivity and fragmentation
In an influential article about the “end of mass communication,” Chaffee and Metzger (2001) suggest that the fragmentation of shared media experiences will render existing media exposure and effects theories obsolete. Comparing the network era of television with content and exposure patterns in multichannel television systems and the Internet, they suggest that fragmentation was an almost inevitable result. However, it seems very likely that people will stop short of complete atomization of exposure. For instance, Salganik et al. (2006) observe that an awareness of the musical preferences of other consumers is associated with consumption of content in line with those perceived environmental preferences. Thus, considerations about joining an anonymous bandwagon can influence the degree to which an audience may dissipate. In an analysis of television and Internet exposure data, Webster and Ksiazek (2012) find that people are not randomly spread across media and outlets. Rather, they consume relatively distinct repertoires of content, selection patterns that may overlap with other consumers’ choices. Similarly, Yuan and Ksiazek (2015) find limited fragmentation in their study of television viewing behavior in the United States and China. While China is characterized by a system of consumption wherein people take some advantage of proliferation of channels outside of the mainstream, the United States exhibits limited internal diversification of consumption choices. Yuan and Ksiazek find that—even with a large number of content options—clusters of Americans gravitate toward similar selections. Taneja and Webster (2016) emphasize an important facilitator of this clustering. Using comScore Internet audience-measurement data, they observe that sociocultural factors can “… spawn ‘small but loyal’ audiences [and can be] an instrumental force in creating communities of ‘large and loyal’ audiences” (Taneja and Webster, 2016: 176).
When people who share beliefs and attitudes make media choices, they presumably might make choices in common. The notion that consumers may be prone to selective exposure tendencies is a theme shared by models of instrumental utility (e.g. Atkin, 1973) and uses and gratifications (e.g. Katz et al., 1973). Individuals become increasingly segmented in their media diets when they have many options with which their motivations and preferences can be articulated. Motivations people have for media selection can include activity that involves the establishment, maintenance, and enhancement of interpersonal relationships. A number of researchers have suggested that people engage in specific patterns of media use with some consideration of its social value. For example, Barton (2009) observed that desires for the social utility function of media were associated with heightened consumption of certain reality television programming. Other researchers have observed that people will seek specific program-types to satisfy needs related to interpersonal conversation and relationship maintenance (e.g. Bantz, 1982). The Internet, similarly, can satisfy certain social needs (e.g. LaRose and Eastin, 2004; Tewksbury et al., 2008). Media consumption can be an activity to enjoy with friends and a source of information to share with others in subsequent conversational bonding. In both cases, one’s social ties (or anticipated social ties) with others are associated with selection of media content.
Writing several years before use of the Internet became widespread, Webster and Wakslag (1983) proposed a model of television viewing that integrated personal factors (individual needs), contextual forces (co-viewing partners), and external structures (television program schedule) that affect how people select television programs. Most subsequent discussions of fragmentation have focused on the intersection of personal factors such as attitudinal predispositions (Garrett and Stroud, 2014) and external structures such as media affordances (Wonneberger et al., 2012) without giving adequate consideration to contextual forces such as social and professional relationships.
While discussing the anticipated proliferation of audience fragmentation in the contemporary media environment, Sunstein (2001) argued that the public could expect a deterioration of their societal-level “social glue.” This notion of “social glue” refers to the ability of individuals to effectively communicate with one another via the utilization of common experiences and shared sources of information. Expecting social glue to deteriorate appears sound if people expose themselves to different facts or perspectives on an issue (or different issues altogether). Potentially underestimated in this account is the value that certain individuals ascribe to maintain this “social glue,” especially relative to the value presumed to be associated with the ability of individuals to fragment toward individual interests. If maintaining some sense of social connectedness is prioritized by consumers, their considerations about their social environment would be a conceivably critical force shaping how audience fragmentation will become manifest. For example, with regard to social viewing, Wonneberger et al. (2009) observe that watching television “… together with friends or family means that the viewer’s preferences affect the group opinion on program choice and that those preferences are, in turn, affected by the group” (p. 238), stressing a critical link between media selection and relationships. Taneja et al. (2012) suggest that “… the structural features of social life still leave their mark on how people organize their use of media into various repertoires” (p. 953). Wonneberger et al. (2011) test an interactive model of internal and external contributors to news selection. They find an interaction of interest in news and the presence of co-viewers such that interest is less predictive of viewing when people are in the presence of others. The co-viewers appear to exert a stronger influence on viewing duration. Thus, situational factors such as the presence of others can exert a substantial influence on news exposure patterns.
One way to both theoretically and empirically examine the role of a given social structure is through social network analysis. Although the study of social ties within networks is diverse, there is a generally agreed-upon assumption that many of the phenomena (e.g. attitudes, behavior, and norms) that attract social scientists may be influenced by social ties. As Borgatti et al. (2013) put it, networks matter because “… an actor’s position in a network determines in part the constraints and opportunities that he or she will encounter, and therefore identifying that position is important for predicting actor outcomes” (p. 1). Indeed, a general assumption in network theory suggests that taking individuals out of their networks is undesirable because it strips away influential social ties and assumes actors exist in a static vacuum (Wellman, 1988). With regard to this study, social networks could conceivably influence certain goal-directed choices (e.g. media consumption for social utility) people make about the content to which they expose themselves. The social glue utility of shared media experiences once thought to be a casualty of audience fragmentation could, in fact, constitute an important factor limiting it.
We expect that social ties and media consumption in an option-rich environment will be associated such that media-use clusters should form around friendship networks. Friemel (2012) undertook a related investigation among teenagers in Switzerland. He examined whether television use could predict (i.e. social selection), or be predicted by (i.e. social influence), participants’ choice of conversation partners. He observed an association between social ties and shared media use, most notably characterized by social selection processes. Other researchers have similarly identified social selection to be the primary force behind the association of social ties with shared media use, though evidence persists denoting a role of social influence (e.g. Aral et al., 2009). Friemel’s (2012) study was informative, though a number of areas for additional inquiry persist regarding broader generalizability, theoretical implications, types of media-use behaviors, and analytical technique. This study employs participants who are dissimilar from Friemel (2012) in terms of geographic location, culture, and age in order to explicitly discuss observations specifically with regard to implications for audience fragmentation. In this study, we take the next step of analyzing fragmentation pertaining to Internet-use behaviors, in addition to television use. Both of these media hold some of the greatest capacity for fragmentation and are the ones most widely consumed by American adults and youths (Emarketer, 2013; Herrick et al., 2014). Furthermore, as we will later discuss, we rigorously analyze the phenomenon with a leading analytical procedure recently developed by network analysts. Therefore, our analyses are better equipped for comprehensively gauging the role of participants’ network positioning in a high-bandwidth media landscape.
Media use has the capacity to enhance social bonding. Mutual consumption of specific television programs offers individuals an opportunity to bond over commonality of potential personality attributes articulated by some of their most specific taste preferences. As such, we predict the following:
H1. Stronger friendship ties will be associated with reports of shared favorite television programs.
Given the narrowness of specific program consumption, it is possible that commonality of personal characteristics might be best articulated with a related, though broader, measure of similar taste preferences. As a result, we expect:
H2. Stronger friendship ties will be associated with shared television genre consumption.
Sport is a popular genre of media content with mutually exclusive subgroups. This affords a potent capacity for particular individuals to be die-hard fanatics of certain sports while overwhelmingly ignoring others. Those who are friends may expose themselves to similar sports content as a type of relationship maintenance or enhancement. Therefore, we predict:
H3. Stronger friendship ties will be associated with shared sports consumption patterns.
The types of news that individuals consume could be related to the formation of preferred habits of information sharing or debate, as well as relationships that cater to those preferences. Hence, we expect:
H4. Stronger friendship ties will be associated with shared news genre preferences.
People may bond over shared Internet-use activities. If one friend enjoys online gaming, shopping, or music exploration, it is feasible that he/she may speak of the benefits to close others or even encourage them to follow suit. Therefore, we predict:
H5. Stronger friendship ties will be associated with shared Internet-use activities.
The Internet has many ways one can pursue one’s news interests (Tewksbury and Riles, 2015). People who frequently go to a particular outlet could do so, as a result of a preference for its style of presenting information. Concurrent exposure among individuals with social ties to one another provides an avenue for discussion about each individual’s perspective on topics, news frames, or writers. Hence, we predict:
H6. Stronger friendship ties will be associated with shared patterns of exposure to online news outlets.
Method
Sample
Study respondents consisted of undergraduate college students at a large Midwestern university. Because network boundaries are very permeable and sometimes difficult to observe, they are often the most vexing factors to determine in empirical analysis. We investigated populations that had closed boundaries on two factors: (1) membership in the same social organization and (2) inhabitation of a single location. As such, we used two social Greek letter organizations for our network populations.
One fraternity and one sorority were selected to enhance generalizability. Furthermore, three organizational criteria were utilized: (1) there could be no prominently emphasized religious, honorary, or professional affiliation with the organization (i.e. we were interested in social Greek organizations); (2) the number of members living inside the chapter house had to be at least 50; and (3) an evaluation of a statement of interest by the organization was required to ensure adequate member participation.
To determine network boundaries, only members who lived in the organization’s house were included. Additionally, in order to encourage individual participation, subjects were offered US$10 to complete the survey. In all, our final sample included 42 (74% response rate) fraternity and 42 (71% response rate) sorority members. There is no general rule of thumb for minimum response rates in network analyses, but several studies have suggested the most common network statistics (e.g. average degree, reciprocity, and clustering) are robust enough for the current participation levels (e.g. Kossinets, 2006).
Procedure
Researchers visited the houses of each organization to administer the questionnaire. Participants worked alone and were told that responses and any comments were to be kept to themselves. Participants not available at the time of administration were allowed to complete the study online within 72 hours of the initial data collection. One fraternity member and eight sorority members completed the study online. 1
Measures
Friendship ties
We collected data on the strength of friendship ties and, thus, were able to extract a directed and valued network. Directed networks are useful to capture relationships that might be asymmetrical, common in friendship networks. For instance, hypothetical person A might indicate that she is friends with person B, but B might indicate that she is not friends with A (i.e. lack of reciprocity). This is important because phenomena such as reciprocity fluctuate. Valued networks differ from binary networks by putting a value or weight (i.e. 1–10) between nodes to characterize the strength of a relationship. They more accurately represent friendship relationships because levels of friendship are rarely binary and usually differ from one person to another (Granovetter, 1973).
To assess the strength of friendship between each person, respondents were given a roster of the other in-house members—the preferred method of network data collection when the full population is known (Borgatti et al., 2013)—and asked to indicate the strength of their friendships on a scale of 1–10. The scale was based on an extensive friendship scenario, asking respondents to rate their friendship based on different prompts (see Hendrickson et al., 2011).
Media-use variables
We created several semantic networks to compare with the network of friendship ties. This approach reduces measurement bias because all of the hypotheses are measured similarly, and it allows for a more comprehensible interpretation of model terms and comparison across independent variables (i.e. is one relationship stronger than the other?). To do so, we created two-mode networks and then converted them to one-mode networks. Two-mode networks, in this case, contain two elements that constitute the network: house members and media use. A two-mode representation links each house member with his/her media consumption. The two-mode matrices were converted to one-mode networks of house members only. For instance, if persons A and B consumed three of the same types of media content, they would have a value of three between them.
Favorite television program
The television program network relied on subjects’ free responses. Respondents were asked to list the top five favorite television shows they had watched in the last year.
Television genre preference
A list of popular television genres was presented to participants and their agreement with the statements, “I often watch _____ shows,” was assessed with a 7-point scale of agreement (1 = strongly disagree; 7 = strongly agree). The genres were comedy, reality television, sports, news, talk shows, and documentaries.
Sports preference
A list of major American sports was presented to participants along with the item stem, “With regard to sports, I often watch _____ games.” Options included baseball, basketball, football, hockey, and soccer.
News genre preference
Participants were presented with a list of news genres and asked, “Please indicate how closely you follow each type of news in newspapers, on television, or on the internet.” News genre options included politics, sports, international, science and technology, health, medical, and entertainment.
Internet activity
A wide list of Internet activities was compiled (e.g. go on Facebook or Twitter, listen to music, and shop), and participants were asked, on a 7-point scale, to indicate how often they had engaged in each activity. The endpoints were “never” (1) and “always” (7).
Internet news outlet
In total, 14 online news websites (e.g. Fox News, Buzzfeed, Facebook, and The New York Times) were presented to participants who were asked to indicate how often they received news from each. Responses were given on a 7-point scale with endpoints of “never” (1) and “always” (7).
The television program preference variable was composed of a two-mode network consisting of members linked with the programs they reported. It was converted to a one-mode network in which a valued tie indicated the amount of common programs shared between each pair of individuals. All other variables were measured on a scale, rather than a nominal response. The scale had to be dichotomized to designate whether a respondent consumed that type of media regularly. The cutoff value in both cases was the midpoint, which determined general agreement with the statement. The midpoint value of 4 indicates “agree” for television and sports preference and “about half the time” for news genre, Internet activity, and Internet use. As such, stronger ties between individuals indicated that respondents had similar patterns of media consumption preferences across each variable.
Control variables
When modeling a network, it is important to include effects not hypothesized, in order to account for some of the most basic factors that have been observed to be related to network composition in the past. As statistical controls, we included three types of theoretical mechanisms for the emergence of social networks: (1) structural effects, (2) homophily effects, and (3) attribute effects. Structural effects simply refer to the “formation of ties due to the presence of other ties” (Robins, 2013: 5). Following Robins (2013), we controlled for some of the most common types of tie dependencies in the network science literature, including reciprocity, transitivity, cyclicity, in-degree popularity, and out-degree activity. Similarly, theories of homophily follow the tenet that birds of a feather flock together (McPherson et al., 2001) and is well suited to a network perspective because similar node attributes can serve as motivations to select ties. Therefore, we control for common attributes that may influence friendship strength: year in the university, major of study, and race.
Finally, we control attribute popularity and activity effects. Controlling for these factors is typical when including homophily effects. Not including attribute effects with homophily terms is akin to including an interaction effect without simple main effects in regression. As such, we control for in-degree popularity and out-degree activity for (1) year in the university, (2) race, and (3) media consumption. We do not include major of study because it had too many unique qualitative values (27 in the sorority and 24 in the fraternity) and too few cases per category to reasonably model. For media consumption, we create a continuous variable representing how much people agreed to all the questions. It represents, for example, whether a respondent generally consumes a lot of Internet news based on how many times they agreed with the questions (e.g. they answered at least “half of the time” for watching MSNBC). We do not include all unique television programs because most respondents (97.5%) answered five television programs, and thus, there would be little variance in this measure.
Data analysis
To assess the relationships between the strength of friendship ties and various patterns of media consumption, we modeled the two friendship networks using valued exponential random graph modeling (VERGM), a recent extension of binary exponential random graph modeling (ERGM). ERGMs are used to statistically evaluate the various factors that influence the probability of network ties (Robins et al., 2007). They work by constantly comparing an observed network to thousands of similar looking simulated networks. Using a Monte Carlo Markov Chain (MCMC) algorithm to continually compare the observed network to the simulated ones, ERGMs specify maximum likelihood estimates (MLEs) for each parameter when they begin to look very similar (i.e. converge). Thus, a significant and positive estimate indicates that the odds of a tie are more likely to appear than by chance alone, whereas a negative value indicates the opposite (i.e. similar to a regression coefficient).
Because the main network in this study is valued, not binary, we relied on Krivitsky’s (2012) extension of ERGM to valued networks. The most significant change in VERGMs is the focus on modeling the strength of ties rather than merely their presence or absence. This requires a modification of what is known as the reference distribution. For binary networks, this is often a Bernoulli distribution of a 0.5 probability of a tie existing. However, in VERGMs, the reference distribution must be defined for all of the possible values in which relationships can occur (i.e. range of values between any two nodes) because the models are being simulated over values of ties between nodes, not just the presence or absence of ties. Because the distribution of friendship ties is limited to a value between 1 and 10, we used a binomial distribution across each value as the reference measure. In other words, because there is an unknown probability between choosing any of the 10 values, we would expect a binomial distribution of each of the values if there were no model terms biasing the model.
Once models were fitted, we followed procedures to identify model degeneracy as advised by Goodreau et al. (2008), including t-tests on observed values compared to estimated ones, Geweke statistics, and time series MCMC estimate plots. We considered models converged when (1) sample statistics did not differ significantly from the observed statistics, (2) neither Geweke statistic was statistically significant, and (3) when each model statistic varied stochastically around the mean through each iteration. We used the Statnet extension ergm version 1.13 (Goodreau et al., 2008) and ergm.count version 3.2.2 (Krivitsky, 2012) in R for the analysis. With this data analysis approach, significant relationships indicate that particular groups of friends are clustering around a media-use activity in which those who hold weaker friendship attachments do not engage.
Results
Table 1 contains descriptive statistics on the networks used to test the hypotheses. Correlations were performed using quadratic assignment procedure (QAP; Krackhardt, 1987). QAP is a permutation-based test that corrects for interdependence between dyads in order to determine the correlation between two networks of the same nodes. They can be interpreted very similar to Pearson correlations, a negative value indicates a negative correlation, and positive value indicates the opposite.
Descriptive statistics and QAP correlations.
QAP: quadratic assignment procedure; SD: standard deviation; DF: distance-weighted fragmentation.
Descriptive statistics refer to in-degree centrality. t-tests are calculated for difference in means for each house.
A cutoff value of 5 was used to determine ties for fragmentation.
A cutoff value of 4 was used to determine ties for fragmentation. Cutoffs were used because both networks were single components.
p < .01, *p < .05.
Table 1 also contains measures of fragmentation. Fragmentation is notoriously difficult to measure in social networks, and there have been several proposals on how to measure it (see Borgatti, 2006). We adopt Borgatti and Everett’s (2006) measure of distance-weighted fragmentation (DF) because it is more sensitive to structural changes in networks that have the same number of components, the more traditional way of measuring fragmentation (i.e. the more distinct the components, the more fragmented). DF is a global measure of the average reciprocal distance between every pair of nodes. The higher the value from 0 to 1, the more fragmented the network. 2 Although there is no associated threshold to indicate that a particular type of media use qualifies as fragmented, it is worth noting that, in general, there were slightly higher levels of fragmentation in the sorority than in the fraternity across both friendship and media use.
The final ERGM containing the tests of our hypotheses is specified in Table 2. To get an overall sense of how similar and different the fraternity and sorority were from one another, we included a column that displayed the difference between each MLE. The differences were flagged if the corresponding z-test was significant. 3 Generally, there should be significant caution when comparing MLEs from different ERGMs because things such as sample size and density can considerably affect the values of estimates (Krivitsky et al., 2011). However, because the networks are identical in size and followed the same procedures in obtaining friendship ties, there is some value to this comparison.
VERGMs with MLEs and SEs (control variables).
VERGM: valued exponential random graph modeling; MLE: maximum likelihood estimate; SE: standard error.
Cell entries are MLEs; SEs are in parentheses. “Popularity” designation indicates association of variable with others saying they were tied with respondent. “Activity” designation indicates association of variable with respondents saying they were tied to others. The reference category for Class is Freshman and White for Race. In the “Difference” column, the difference is significant if the resulting z-test is >1.96 in absolute value.
p < .01, **p < .001.
After controlling for other mechanisms (Table 2, structural, homophily, and attribute effects), our hypotheses (Table 3) concerned the relationship between different patterns of media consumption and social networks or, in other words, network multiplexity. The most prominent finding stems from H1. Both organizations exhibited clustering around preferred television programs within, though not between, friendship clusters. This provides full support for H1. Though no other media-use variable was associated with social network in both organizations, two media-use behaviors were associated with friendship-based clustering in the sorority. Organization-wide, respondents’ heightened friendship levels were associated with similar patterns of sports consumption and preferred Internet-use activities, providing partial support for H3 and H5, respectively. No other set of hypotheses received support. All of the results are discussed below.
VERGMs with MLEs and SEs (main variables).
VERGM: valued exponential random graph modeling; MLE: maximum likelihood estimate; SE: standard error.
Cell entries are MLEs; SEs are in parentheses. “Popularity” designation indicates association of variable with others saying they were tied with respondent. “Activity” designation indicates association of variable with respondents saying they were tied to others. In the “Difference” column, the difference is significant if the resulting z-test is >1.96 in absolute value.
p < .01, **p < .001.
Discussion
The results of this study provide some support for the expectation that the structure of social ties is related to media-use patterns. We observed evidence consistent with theorizing about how social ties could mitigate widespread fragmentation in the contemporary content-rich environment (depending on the type of media content) and lead to clusters of media-use activities. Direct support for the hypothesized relationships appeared in both organizations with regard to self-reported favorite television programs, the most narrowly defined media-use behavior. Of our six media-use variables, the three which received no support were related to genre preferences (i.e. television genre), news preferences (i.e. preferred Internet news outlets), or both (i.e. preferred news genre). Thus, our findings provide support for our expectations primarily with regard to the more specific accounts of media use.
Stronger friendship ties were associated with reported favorite television programs (indicating a clustering of this media-use behavior within social ties but not between), whereas television genre preferences were not. Both of these variables pertain to consumption of television programming, but they operated dissimilarly. One potential explanation may lie with the level of abstraction at which television genre operates compared with specific television program. This might have been an issue because television genre preference is a media predilection that might be too broad for assessing whether individuals are actually engaging in similar media consumption. For example, though two respondents might individually agree that they enjoy the comedic genre, they might disagree on the type of programs that should be included in this genre or on which programs most closely exhibit their preferred type of humor. Even with mutual genre liking, a person who enjoys South Park and another who prefers The Big Bang Theory would not have much to discuss or bond over with regard to this media preference. An absence of a relationship between sociocultural similarity and shared genre preference is, however, supported in previous research (Taneja and Webster, 2016).
Strength and structure of friendship ties were associated with preferred sports consumption for the sorority. Regular consumption of particular sporting events could reasonably be considered analogous to regular consumption of a particular television program. Someone might be expected to tune in to each “episode” to stay up-to-date with what is happening in a sport. As members of the sorority indicated stronger friendship ties, they also shared a common pattern of interest in specific sports. A potential reason why we were not able to observe social ties mitigating fragmentation of sports consumption in the fraternity may have been that there was no fragmentation of this type of media use in this organization to begin with. Another explanation may be that the uniformity of strong friendships in this organization was associated with a uniformity of media-use behaviors. Such an explanation, however speculative, aligns with our primary theorizing.
Internet-use activity clusters were associated with friendship ties in the sorority but not the fraternity. One could conceive of Internet-based choices (i.e. using Facebook or Twitter, listening to music, playing games, etc.) as the Internet equivalent of choosing a particular program on television. Both involve making a choice among the offerings available in a medium and, therefore, may operate similarly. Internet activities, however, are clearly more multifaceted. For example, measuring activities such as listening to music, playing games, or watching YouTube videos do not assess the specific types of content people seek. Therefore, although there is some specificity to this media-use behavior, it is not nearly as narrowly defined as is preferred television program. It is possible that due to the high reported affiliation in the fraternity, we were not able to discern how more isolated social clusters mapped onto the formation of media-use clusters for this variable, as was the case with our other social network.
Frequency of using specific online news outlets was an Internet media-use variable that was not associated with social ties. The null finding can likely be explained by the notion that just because individuals share a use of an outlet, they might not be interested in the same stories or topics in that outlet. For example, among people who prefer to read the The New York Times online, some may enjoy its arts coverage and others enjoy the national or international news fare. The same idea can apply to preferred news genre, which was also not significant in both organizations. Preferred news genres, similar to news outlets, may not be illustrative of the various frames and perspectives to which consumers will expose themselves. Thus, news use effects related to social outcomes (e.g. Riles et al., 2015) may not be readily apparent. This is to say, different people may similarly engage in certain media-use behaviors, though in wholly dissimilar capacities. Our analyses did not incorporate a proxy measure of the general popularity of each media-use behavior within the wider population. Such a measure would have been all but impossible to produce with regard to the larger population for all of our media-use behaviors but potentially would have allowed us to parse out this de facto mutual usage and better detect how usage was related to ties. Hence, the statistical significance of key variables was not bolstered by the exclusion of this measure. Indeed, this could be one potential explanation for failing to observe additional hypothesized support and suggests that observed support may indicate particularly pronounced effects.
This study had a few limitations worth noting. First, we utilized Greek letter organizations as our participants. Although the use of these organizations might hinder generalizability, we needed to find closed-boundary organizations that were not defined by a confounding characteristic. Although a random sample of Americans, for example, would provide higher external validity, we would have no way of operationalizing a social network because we could not analyze ties on that scale. However, future research might consider a broader environment, like a city, that can take advantage of network snowball or respondent-driven techniques (Borgatti et al., 2013). It seems very plausible that similar patterns could emerge in the workplace, another context in which people spend many hours together each day. Indeed, much has been said about the reduction in the power of “watercooler” conversations in the workplace (e.g. Jones, 2013). The enhanced diversity along dimensions such as age would allow researchers to ask about the social implications of other media-use behaviors (e.g. use of social media) that might have much more variation than in a Greek letter organization.
Another limitation pertains to the use of measures of perceived friendship ties. There is evidence that self-perceived ties are often different than externally observed ones (Corman and Scott, 1994), much as self-reported media consumption can, at times, be a less than optimal measure of actual consumption (Prior, 2009). Self-report data (e.g. to measure media consumption or friendship ties) always carry the risk of being influenced by social desirability or memory errors. Ethnographic techniques of observing our participants might have yielded more externally valid accounts of these measures. Future research should consider this methodology. Future research might also benefit from a focus on frequency of television program exposure rather than on preferences. Such an approach could complement the work of this study. Finally, our exploratory study is cross-sectional in nature. It is possible that a specific causal direction between quality of social ties and media use might have manifested itself if the variables were examined longitudinally. Future research using longitudinal actor-oriented modeling (e.g. Friemel, 2012), for example, can differentiate the points at which social ties predict fragmented clusters of media use or vice versa.
Concluding remarks
If there are limits to technology-enabled audience fragmentation of our most specific media-use behaviors, social ties might be one source. Controlling for select indicators of homophily, we found that social network ties are consistently associated with television program preferences. Indeed, if media producers seek a loyal or growing audience, one route to get there is to take advantage of existing social networks. If a particular television program or Internet outlet seeks to grow its base, it should provide opportunities for audiences to bond with others over its content. Our results suggest that where fragmentation does occur, there is a potentially strong limit to fragmentation, and it is tied to the most basic element of social organization: personal relationships.
This examination offers a novel approach to overcoming sub-disciplinary divisions between researchers in mass, interpersonal, and organizational communication research. Network analytical procedures were utilized to examine the influence of interpersonal relationships on a media environment long perceived to facilitate audience fragmentation. Reardon and Rogers (1988) argue that it is “time for the sub-disciplinary boundaries to become much more permeable … If not, we may be missing a world of research opportunities” (p. 300). Here, we make strides toward exploring these research opportunities in the context of media audience fragmentation.
Footnotes
Acknowledgements
The authors wish to thank Colleen Couture and Eric Wiemer for their data collection and coding assistance.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.