«NETWORKING IN EVERYDAY LIFE by Bernard J. Hogan A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy ...»
4. Network composition: The last section suggested that individuals of different roles
CHAPTER 6. WITHIN-NETWORK VARIATIONS AND NETWORKED INDIVIDUALISM 164occupy different places in the network. Family are usually collected as one large group whereas friends are distributed throughout the network and voluntary association members are usually set apart. So just as network structure, in the abstract sense of closeness, can foreseeably play a part in particularity, so can either the prevalence of certain roles or the heterogeneity of roles in the network. Here I use the number of alters belonging to each of the eight roles discussed above. (To note, other models using the percent of alters instead showed similar results.) Since the particularity measure is normal and the cases are considered independently, it is worthwhile to seek to explain the measure through linear regression.4 This data set includes only 80 viable cases for analysis. This is because of missing data issues with four of the other possible cases. Thus, I am reluctant to perform a simple nested linear regression on these cases (i.e., simply pile on additional variables regardless of their contribution to the model). The inclusion of additional variables cuts into precious degrees of freedom and increases the standard error in the model.
This means that even if additional variables increase the raw R2 value, they can still decrease the adjusted R-squared value and thus interfere with the model as well as render as non-signiﬁcant variables that would otherwise be signiﬁcant. Therefore, I employ a nested forward-selection regression rather than a simple nested regression.
Just like a nested OLS regression, variables are added into the model in batches. However, only the signiﬁcant variables are kept in the model from one nesting to the next.
I use a P-value of ≥ 0.1 as criteria for exclusion.
Table 6.4 presents ﬁve models predicting particularity.
The ﬁrst four represent the
Table 6.4: Nested OLS regression models predicting particularity
WITHIN-NETWORK VARIATIONS AND NETWORKED INDIVIDUALISM 166planatory variables removed to highlight the increased strength of association. Through the models one can see a number of noteworthy ﬁndings, although the overall picture is consistent with many of the claims made up until now. What is perhaps most striking is the effect of being an immigrant in this sample. This variable is signiﬁcant throughout all ﬁve models and has the highest strength of association with the dependent variable. In Chapter 5 I noted that immigrants were disproportionately located in the “daily online” cluster, and that they were prominent users of online chatting and instant messaging. This is shown again in this model whereby immigrants show a great deal of particularity among their alters. This particularity is most likely due to their interest in maintaining long-distance ties, and their interest in using whatever means possible to that end. Moreover, recent immigrants are probably the quintessential networked individuals as their networks almost by deﬁnition are far ﬂung and consequently loosely connected, spanning friends and family from the home country as well as others from their host country (Hiller and Franz, 2004).
Apart from immigrant status, a couple of social location / demographic variables were also signiﬁcant in the ﬁrst model, although this signiﬁcance washed out once I included communication frequency. In the case of age, this is unsurprising as older individuals are less prone to using ICTs, so before ICT use was included, age was a signiﬁcant variable. Relationship status was not particularly signiﬁcant to begin with, although given that the coefﬁcient was negative, it suggests that individuals who are coupled tend towards consistency with networks, perhaps by differentiating their use of media between the couple.
The second model included selected measures of ICT status. While cellphones were not employed in the calculation of the particularity index there is reason to believe, based on the results of the prior chapter, that cell phone users would be more particular in their media use, especially since cell phone use is frequently coupled with email use. This was not borne out in the data. However, Internet use was sigCHAPTER 6. WITHIN-NETWORK VARIATIONS AND NETWORKED INDIVIDUALISM 167 niﬁcant, and remained signiﬁcant throughout the remaining models. In this analysis, Internet use was split into three categories: non-user, light user (an hour or less a day) and heavy user (more than an hour a day). These categories were based on Statistics Canada’s analysis of Internet users (Veenhof, 2006). Non-users were the reference category, and thus not included in the model. Interestingly, simply using the Internet did not explain variations in particularity. Only heavy Internet use was a signiﬁcant explanatory variable. This will become increasingly relevant as more individuals move on to the Internet and use it more frequently for different media such as social software and microblogging. It also reinforces the idea that Internet use affords but does not determine fragmentation of media use, as light Internet use was not signiﬁcant.
Finally, the strongest predictor in this model, by far, was the total instances of planning. This measure is based on the survey questions introduced in Chapter 5. There I examined clusters of media use. In this analysis, I simply summed all of the variables into a single planning score. Since planning is highly related to the use of different media, it is unsurprising that it is a signiﬁcant predictor, however, it does reinforce the validity of the particularity index as the planning variable was taken from survey measurements of planning by media across all alters, whereas the particularity measure was taken from interview measurements for each individual alter.
Perhaps the most interesting aspect of this analysis is the paucity of signiﬁcant structural variables. Many of these variables are signiﬁcant in other per-alter models, including the multilevel model in the next chapter and work by other researchers on this data set (Carrasco, Hogan, Wellman, and Miller, Forthcoming). This is to imply that it is not a problem with the measures. Rather, it seems to be the case that network structure is not related to the particularity of media use. To note, one variable was signiﬁcant in model three (the number of isolates) but the signiﬁcance of this variable was washed out by the number of online only network members. As we may recall from part I, online-only alters are the most likely role to be isolated, so it is underCHAPTER 6. WITHIN-NETWORK VARIATIONS AND NETWORKED INDIVIDUALISM 168 standable that it masks the effect of this variable (or rather, that the only structural effect is really an effect of having online-only alters).
The lack of other clear structural indicators reinforces the network pie charts shown earlier in this chapter. Despite being at two ends of the particularity measure, these individuals have nearly structurally identical networks. Even as a null ﬁnding, it is a rather important one for future analyses of networked individualism. It suggests that the individualization of media use and the fragmentation of networks are not especially related. Granted, this is a restricted data set (being from one community in one city at one time). Nevertheless, it is (to my knowledge) the ﬁrst attempt to link these two conceptual keystones of networked individualism. I will elaborate on the consequences of this model in the summary below.
The last model tested included network composition measures. These measures do not examine the heterogeneity of alters, but the presence of speciﬁc alters in the network. Herein, only one role was signiﬁcantly associated with increased particularity— online-only alters. This should not come as a surprise, since these alters, by default, are only accessed online. To note, the values of these variables refer to the raw number of alters in the network by role. Alternate models using the percentage of alters in the network showed nearly identical coefﬁcients. The interesting null ﬁndings in this model indicate that having large numbers of family did not tend individuals towards less particular strategies, nor did having large numbers of friends predict correspondingly particular strategies.
The ﬁnal preferred model shows a modest adjusted R2 of 0.36. To note, the raw R2 value was lower in the preferred model than in model four, even though the adjusted R2 was higher. This highlights the value of being especially prudent with variable selection in these models.
CHAPTER 6. WITHIN-NETWORK VARIATIONS AND NETWORKED INDIVIDUALISM 1696.4.4 Summarizing the results from Part II The focus of this section was on intra-network variation in media use. That is, under what conditions are people going to use a variety of strategies with their alters, and under what conditions will they use a consistent strategy. If networking is about accessibility, these are two ends of a scale with one implying “I have a stable strategy, know it and you can easily access me” and the other implying “I will use whatever is necessary to get in contact with you, and to make myself accessible”. One pole is about consistency and the other is about accommodation. But unlike the prior chapter where I partitioned individuals into mutually exclusive categories, here I merely plotted these strategies on a spectrum of particularity. The beneﬁt of having a single scale of particularity is that I can then model particularity in a relatively straightforward fashion. I employed this model to test ﬁve speciﬁc hypotheses about variation in networking. The ﬁrst three come from Wellman’s theory of networked individualism.
To these I added overall planning propensity and sociodemographic characteristics, most speciﬁcally, immigrant status.
Hypothesis 1 (on media use) was a non-linear relationship between Internet use and particularity. That is, since the Internet affords different perceptions of one’s network, heavy Internet users will employ a more particular strategy, although mere use of the Internet will not be associated with more particular networking. This hypothesis was validated. Heavy Internet use persisted as a signiﬁcant variable throughout all models, even when controlling for overall planning propensity. Given that the coefﬁcient as well as the level of signiﬁcance remained stable (at approximately b = 0.4, p 0.03 & p 0.01) across these models, I consider this to be a strong ﬁnding. Heavy Internet use is not simply a part of frequent planning, but a factor in its own right.
Hypothesis 2 (on group composition) was that the presence of speciﬁc groups, namely roles with substantially more homophilous ties than heterophilous ones, would be associated with a more general strategy. That is, the more people one knows from
CHAPTER 6. WITHIN-NETWORK VARIATIONS AND NETWORKED INDIVIDUALISM 170a speciﬁc group, the easier it is to coordinate with that group and to be in contact with group members in consistent ways. I cannot reject the null hypothesis in this case.
The only group that contributed signiﬁcantly to the model was the presence of online alters. Almost by deﬁnition these alters should make a difference since they are online only. All other roles did not contribute signiﬁcantly.
While I suspect there are ever more complex ways to parse the relationship between role and variation in media use, I am hesitant to try and squeeze a signiﬁcant ﬁnding from this situation. At least in this context, it appears that there is simply too much variation per role for these roles to be considered as singular forces acting on ego. However, the optimistic interpretation of this situation is that it reveals a clear distinction in how individuals think about their ties to alters and how they act on them. They think in terms of roles, but employ speciﬁc media independent of the structure of these roles.
Hypothesis 3 (on network structure) was that more fragmented networks would be associated with more particular behaviours and a less general approach to maintaining ties. Again, for the most part I cannot reject the null hypothesis in this case. Conventional structural variables were not signiﬁcant. That is, particularity did not vary signiﬁcantly with network size, density, number of components or centralization. In particular, number of components is a great example of network fragmentation. Each component that is not connected to other alters can be considered a separate sphere of life that has to be maintained by ego. It is highly plausible that one would have to be especially sensitive to different strategies in order to maintain more separate spheres, but this does not seem to be the case. This may have been presaged by the network examples given at the beginning of the previous section: “Cathy” was able to manage a very large number of dyads and isolates who were not close with each other despite not using email or the telephone. For her it was about keeping busy and ﬁnding alter-appropriate activities that were easily scheduled (such as a movie or stage