«NETWORKING IN EVERYDAY LIFE by Bernard J. Hogan A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy ...»
7.6.1 Including the frequency of communication The ﬁnal model adds in the frequency of communication. These models alter the variance explained considerably. Firstly, as shown in the bottom of the ‘Model 3’ column of Table 7.2, adding the frequency of communication signiﬁcantly increases the overall explained variance. The R2 increases from 0.14 to 0.22. However, when examining the level-1 and level-2 R2 values, I ﬁnd that the level-2 variance explained (R2 ) has actually decreased, while the level-1 variance explained (R1 ) has increased sixfold from 7 percent to 42 percent. A decrease in variance explained at one level is possible in these sorts of models (Gelman and Hill, 2007; Rabe-Hesketh and Skrondal, 2008). What this suggests is a third or more of the variance in how many media are used can be attributed to the frequency of use for the various media. This demonstrates a variant on the media multiplexity hypothesis that Wellman has termed ‘the more, the more hypothesis’.9 By this he means that the more people will contact by any medium the more they will contact by all media. This “the more, the more” hypothesis appears to be far stronger than the media multiplexity hypothesis about tie strength.
From an accessibility perspective this makes sense. Individuals will give the most access to those they communicate with most frequently. Of course there are exceptions (such as not giving one’s boss one’s instant message address). These are to be expected, especially considering these variables did not account for all of the variation within networks. Nevertheless, frequency of communication is a powerful force moWellman, personal communication.
CHAPTER 7. MEDIA USE WITH SPECIFIC NETWORK MEMBERS 202tivating one’s decision to employ additional media with others. Also, “the more, the more” refers only to social activity, email, and instant messaging, each of which had positive coefﬁcients. By contrast, telephone had a negative coefﬁcient. After exploring this relationship further, it is clear that telephone contact is negative only when controlling for other frequency of contact variables, particularly socializing. That is to say, given a particular level of contact in person, the more one talks on the phone the less other media they are going to use together. This, perhaps, suggests a certain substitution effect between telephone and other media in speciﬁc cases. It can also be considered from an accessibility perspective. Very frequent telephone contact implies that it is easy to contact via telephone. With less telephone contact, ego might be either reluctant to “cold” call alter, or be just as interested in accessing them another way. I believe this particular ﬁnding calls for further inquiry.
An additional consequence of including the frequency of communication in these models is that it mutes the explanatory power of several of the previously signiﬁcant variables, most notably, tie strength. Because there is a stronger relationship between frequency and number of media than between tie strength and number of media, and frequency is correlated with tie strength, this ﬁnding is unsurprising. The third model also masks the explanatory power of distance, degree, and kin status. Like tie strength, distance was also correlated with frequency of contact, especially in person contact.
Also like tie strength, distance was not a particularly strong explanatory variable. The same explanation holds for kin status, as in person contact, socializing, and email contact were all lower for kin than non-kin.
Of all four variables that became non-signiﬁcant in model 3, only degree was not correlated with frequency of contact. However, degree’s standard error was still much lower than the other three. It is possible that under other conditions (such as a larger
7.7 Discussion These four models (the variance components model plus the three shown in Table 7.2) have revealed copious micro-ﬁndings, many of which are aligned with the original hypotheses. Collectively, they seem to have made the media multiplexity hypothesis more complex, rather than less. That was partially the goal. As seen in the original model, the stronger the tie, the more media one is going to use with that person. This ﬁnding is statistically signiﬁcant but not substantively signiﬁcant. It explains nearly no variation in the overall model even though the coefﬁcient has a p-value 0.05.
With the use of speciﬁc statistical controls that take into account some of the more obvious reasons why individuals would use more media-contexts, this relationship between tie strength and media use becomes more signiﬁcant, but it is still not a particularly strong explanatory variable for the logic of media use within personal networks.
Rather in this case, the theoretically constituent parts of tie strength—being spatially proximate, being in frequent contact, having many mutual relationships, and being a particular kind of tie—are more useful in predicting the number of media-contexts used. As mentioned above, tie strength is a multidimensional construct, and other measures are also relevant, such as length of time known, propensity to be reciprocal and extent of support given and received. Indeed, these are not taken into account.
However, of the various correlates to tie strength espoused in the literature, the one’s included above seem the most logical variables to include in this model. Moreover, neither supportive relations nor reciprocity signiﬁcantly contributed to the model.
Finally, the goal here is not to completely undermine the relevance of tie strength to media multiplexity. It is to show that there are numerous practical attributes of a relationship and a network that are as important as tie strength, if not more so, for
7.8 Summarizing the hypotheses Hypothesis 1 suggested there were role speciﬁc values for media multiplexity. There is no clear evidence for this to be the case outside of the broad distinction based on ascribed versus acquired ties, given the signiﬁcant negative effects of being kin. These ﬁndings might be extended into more nuanced understandings of role in a larger data set. However, they may not. It might be the case that there is simply too much variance in how people interpret speciﬁc roles beyond the clear distinction of kin-non-kin.
What a neighbour means for some is completely different than for others. Yet, there is a clear consistency in what it means to be kin (Wellman and Wortley, 1989; Fischer, 1982). As noted above, this consistency means that individuals will include kin in the network that they do not talk with particularly frequently or by many media.
Hypothesis 2 suggested that more highly connected individuals will share more media with ego. This ﬁnding is relevant both at the individual level (where degree is signiﬁcant) and at the network level (where density is signiﬁcant). This is a particularly strong and useful ﬁnding as it can be effectively applied to future media systems, such as who to include on Facebook, who to suggest for addresses in one’s email list, or who to list on one’s instant messenger chat list. It does not require complex network analysis, and has a face validity as a logic that can be employed by an individual. It also lends itself to a theory of social accessibility. Those who share the most ties with ego may also serve as gatekeepers for ego. If one is to use a number of media, it would be with these individuals. They may organize affairs, with each one having differential access to the rest of the network, but mutually having a great deal of coverage.
However, just like tie strength, the signiﬁcance of degree is washed out by the frequency of contact. Yet, even this reinforces the logic that individuals jointly organize and communicate. Those with whom they are in more frequent communication are
Hypothesis 3 suggested that more spatially proximate individuals will share greater numbers of media with ego. This is a relatively commonsense claim when considering interaction patterns. Only the hype of the early days of the Internet worked against the idea that individuals would communicate in more ways with those who are distant. In early work and punditry one may build a claim that people would need to use an array of media to compensate for a lack of in-person contact among those who are more distant. Yet, by considering the Internet as embedded in everyday life, it starts to support the architecture of everyday actions. The vast majority of these actions are grounded in the coordination and mutual interaction of physically proximate individuals.
Hypothesis 4 suggested that those with more frequent in-person contact will share more media with ego. The evidence slightly supports this claim. The p-value is ever so slightly above the traditional cut-off of 0.05. Yet of all the frequency of interaction variables, in-person interaction was the least signiﬁcant. This is probably because of all the means of interaction, in person contact (rather than socializing) is the form of contact least regulated by ego. As Boase (2006) notes, one’s choice of interaction partners in person is highly variable. By contrast, one’s choice of interaction partners by media can be more carefully controlled.
The ﬁnding helps to reinforce Haythornthwaite and Wellman’s second claim of media use—that there is a Guttman scale of media use. A secondary analysis found some evidence for a partial Guttman scale. Using the LoevH algorithm for media use on the dichotomous variables use/did not use media, I found that evidence for a Guttman scale starting from in person contact, followed by socializing, email, and cell phone use. Landline telephone and instant message use could not be included in this scale, thus making it a partial rather than full Guttman scale of media use. The fact that telephone was not included could reasonably be expected from the earlier
that in many ways it stands apart from other media.10 Nevertheless, the point is that people start with in-person interaction ﬁrst. After this point, individuals are more selective about those with whom they will interact.
Hypothesis 5 suggested that when controlling for these other effects, that the residual effects of socioemotional tie strength will not explain how many media-contexts individuals use with others. This hypothesis is true when including the frequency of communication. That is to say, the media multiplexity hypothesis is either spurious or under-speciﬁed. Since tie strength was signiﬁcant in model 2 (such that someone on the ﬁst ring will use, on average, 0.4 more media than someone on the outer ring), I cannot deny the explanatory power of tie strength outright. However, I do believe that in the case of everyday life communication, there are far better ways to explain the number of media used than socioemotional tie strength. As a consequence, I recommend researchers focus on the relevant and constituent aspects of tie strength, rather than merely taking tie strength for granted as a motivator of increased media use.
Of all the aspects of networking in everyday life, one might consider the number of media used to be a relatively marginal one. I would argue differently, especially in contemporary urban societies. Social access is one of the key ways in which individuals sustain their ties with each other. To say, “I am available whenever you need me” is a powerful statement. To merely have someone on your instant messenger list means they are available when you want to chat. To email someone is to denote that The LoevH algorithm calculates Lovinger’s H, a measure of expected versus measured errors in a ranking of dichotomous variables (Loevinger, 1948; Hardouin, 2004). If there are signiﬁcantly less errors than expected, the value is kept in the model. In the aforementioned scale, the scale Loevinger’s H coefﬁcient (rather than Loevinger’s H for speciﬁc items) was 0.2799, which gives a very signiﬁcant p-value of 0.001. However, this scale should not be taken as the last word as it is a scale of the ordering of media used between dyads. Unlike the HLM models, this scale does not take into account how dyads are nested in networks.
CHAPTER 7. MEDIA USE WITH SPECIFIC NETWORK MEMBERS 207you can discreetly plan future events, as well as trade all sorts of documents, from inspirational chain letters to family photos to party invitations. As actor Peter Ustinov mused over thirty years ago, “Contrary to general belief, I do not believe that friends are necessarily the people you like best, they are merely the people who get there ﬁrst” (Ustinov, 1977).
While sociologists are frequently less than great at predicting the future, it is safe to assume that there are new media on the horizon. Since embarking on this dissertation, email and instant messaging have been partially usurped by social media (such as Facebook and MySpace) and “twittering” (the practice of broadcasting very short life updates). If we do not sort out the logic behind media multiplexity (be it role, communication frequency, degree or proximity), we will be faced with the seemingly perennial task of migrating tie our relationships from one medium to another. While the network-as-cognitive-object will likely remain intact, the actual means for contact these individuals will be partially strewn across ever more media.
This is the modern equivalent of Simmel’s web of group afﬁliations. In Simmel’s day, alters were generally associated with speciﬁc places and speciﬁc times. Some individuals were known through church, others through a fraternal organization and others still from work. Individuals, mused Simmel, expressed their identity through their speciﬁc combination of these groups. Yet, they were always existing at any given group at any given time.
Now, accessing individuals is simultaneously far more simple and far more complex. Simpler insofar as one need not travel, nor even be in contact at the same time.