«NETWORKING IN EVERYDAY LIFE by Bernard J. Hogan A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy ...»
1990) was meant to make the task more straightforward.
The governing logic of this process was that individuals need only indicate which ties were present, rather than being asked about all potential ties. Also, by encircling sets of individuals, the respondent could cut down on both the number of edges drawn and the visual clutter. The steps for denoting edges was based on the idea that
Figure 4.3: Stylized version of the process of drawing edges according to the steps presented.
than vice versa. This is in keeping with Granovetter’s theory of triadic closure as well as prior research (Granovetter, 1973; Kalish and Robins, 2006).
Individuals generally considered this to be an unexpectedly fun and interesting task. This is very novel for social network research, as the previous matrix questions were considered monotonous and burdensome. Ethics reviews commonly ask what beneﬁts will come to the respondents from participating. A common ‘canned’ response is personal insight. I believe this technique makes good on that assertion.
The third stage of a name generator involves name interpreting questions. These are questions designed to elicit additional data about the alters as well as data about the relationship between ego and alter. In the Connected Lives interviews we opted for detailed description about speciﬁc alters to superﬁcial description about all alters, with two exceptions, gender, and role, which we captured for all alters in the network.
To select the speciﬁc alters we designed an algorithm for sampling alters. We asked interviewers to select the three closest alters, and then select the alter that was elicited ﬁrst from each ring. We could tell which alter was elicited ﬁrst because each name tag had a tiny number in the corner numbered from one (the ﬁrst alter elicited) to 33 (the last alter). The interviewer continued to select the alters with the lowest number until
was chosen based on prior research suggesting that given the opportunity, individuals are most likely to abandon the name interpreting process after 15 alters (Manfreda et al., 2004). Finally, if an alter was married to one of the alters that was already selected, they would be excluded. Additional details about this process are given in Hogan et. al., (2007).
For each of the selected alters, the interviewer asked a series of questions about the alter’s demographics as well as questions about media use with that alter. The speciﬁc instrument used (internally referred to as the ‘minisurvey’) is shown in Appendix B.
In practice, the minisurvey was so onerous that many interviewers reduced the number of individuals sampled in the minisurvey from 15 to 11 or 12. This was because there was almost 40 minutes of interview that had to take place after the minisurveys were completed, and interviewers were concerned about respondent withdrawal. Nevertheless, the sample of alters selected from the minisurvey is considered to be a faithful representation of the full network. As reported in Hogan et al. (2007), the sampled alters do not differ proportionately from the rest of the alters in terms of gender, ring or role, with one exception. There were more extended family members in the networks than in the sample. I assert that this is because individuals are especially prone to including the husbands and wives of extended family members, yet we could only include one of a couple in the minisurvey sample.
4.7.6 Interview Results The results of the name generator show networks that are slightly smaller than those given in the survey. As expected, there is still a positive skew for both very and somewhat close alters. This is reinforced by the fact that the networks had a mean 11.6 very close, 12.2 somewhat close and 23.8 total alters, even though the medians were 10,10, and 21, respectively.
Table 4.4: Name generator network size by ring, for very close, somewhat close and total alters As expected, many very close alters are also present on the second ring.
More curious is that 9 percent of alters initially labeled very close appear on the outer third and fourth rings. However, the 0.5 percent of very close alters placed on the outer ring come only from four respondents, two of whom have three very close alters on the fourth ring and two who have one. These alters are consistently among the lowest ranked very close alters (such as the 17th, 20th, and 21st of 21 very close alters), and they are not connected to the largest component.
Ties labeled ‘somewhat close’ are well distributed across rings two through four, with most on the third ring. There is some cognitive overlap between the weakest very close alters and the strongest somewhat close alters, a feature that is captured in the four ring schema. Like the small number of very close alters on the outer ring, the ﬁve percent of somewhat close alters on the inner ring were rarely connected to the largest component. But unlike the marginal very close alters, these few somewhat close alters who are on the inner ring were recalled early in the name generation process. As such, I suspect they are actually very close alters who were inadvertently omitted during the very close naming stage because they are not connected to other very close alters.
and later placed into more ﬁnely grained division by rings reveal an interesting difference in perceptions of socioemotional closeness. When the respondent is ﬁrst asked to name individuals as ‘very close’ or ‘somewhat close’, the respondent only considers her individual relationship to each alter. However, when the respondent has to arrange these names on one sheet, she must assess the closeness of alters in relation to each other. At this point, the respondent often promotes some alters to the inner rings and demotes others to the outer rings. Capturing the respondents behaviour thus shows a beneﬁt of participant-aided visualization: Arranging the alters in an overall structure induces the respondent to think about individuals in relation to each other. This is relevant to the eventual analysis, as one of the key research questions in Chapter 7 is whether media use varies by closeness of the individual. By having this second check on socioemotional closeness I believe I am in a better position to test this claim with validity.
There is much variation in the structures of the networks. This variation can also be described quantitatively. Here I focus on the number of components and the density of the overall graph.5. For both measures, ego and ties between ego and alters are excluded. The mean density of the 86 networks is 0.17, which increases to 0.30 when isolates are excluded.
There is a clear negative correlation between density and the number of alters(r = −0.38, p 0.001). This is because density is simply the number of ties divided by the number of possible ties. As the number of alters increases linearly, the number of possible ties increases geometrically (since density is a function of the number of alters squared). So it becomes increasingly less likely that the number of ties will stay proportionate to the number of alters. There is also a strong positive relationship between A component is a sub-graph that has no connections to the rest of the network. Strictly speaking there is only one component in a personal network, since ego is connected to everyone. By removing ego, it is possible to get a better sense of the personal network that affects ego, rather than ego’s effect on the network (see discussion in McCarty and Wutich, 2005) An isolate is an individual who is unconnected to the rest of the graph. It is also the smallest possible component.
CHAPTER 4. RESEARCH METHODS AND MEASURES 84the number of alters and the number of components in a network (r = 0.71, p 0.001).
The relationship between network size and the number of components persists when isolates are removed (r = 0.72, p 0.001). This means that larger networks do not necessarily have more isolates that skew the number of components. Instead, larger networks have a greater number of separate groups.
4.7.7 Comparing network size across the two methods There is a strong correlation between the network size produced by the summation method and that produced by the name generator method (r = 0.67, p 0.001). That is to say, people who say they have few alters on the survey, mention only a few during the interview; those who say they have many on the survey, mention many during the interview (Figure 4.4). The strength of association is higher for very close alters (r = 0.74, p 0.001) than for somewhat close alters (r = 0.49, p 0.001). This means that the responses given for very close alters vary less between the survey and the interview than responses for somewhat close alters. Respondents seem to have surer grounds for deciding who their very close alters are than their somewhat close alters.
Although the number of alters produced by the name generator and the summation method are strongly correlated, respondents routinely mention a greater number of network members when they use the summation method. To estimate the difference in magnitude, I use bivariate linear regressions with no intercepts. Using this measure, the coefﬁcient for the independent variable indicates how far the dependent variable deviates from the diagonal (1:1 relationship) conditioned on that variable, and the R2 measure indicates the variability of this deviation. Respondents name 1.25 ‘very close’ alters on the survey for every ‘very close’ alter on the interview and 1.64 ‘somewhat close’ alters on the survey for every ‘somewhat close’ alter on the inter
Figure 4.4: Predicted values for number of names recalled using the summation and name generator methods name generator (R2 = 0.
78). It is not surprising that respondents disproportionately name more somewhat close alters on the survey, and have more variation in the number of somewhat close alters named. While very close alters are deﬁned by speciﬁc criteria, somewhat close alters are deﬁned in the survey as simply “more than just casual acquaintances, but not very close”. By contrast, interview respondents have to actually name their alters instead of giving an approximate count. As a result of this procedural difference, respondents are choosier in the interviews about which alters are somewhat close. Moreover, as mentioned above, some survey respondents round off large counts on the survey.
4.7.8 Media use—measures and interpretations Much like the measurement of social networks, there are a multitude of ways for measuring media use. These measures are complicated by the almost necessary act of considering media use temporally. As Zerubavel (1982) notes, there are four dimen
clock time) and recurrence. One can measure media use according to each of these dimensions, both in terms of aggregate media use, and media use with any individual.
Someone might regularly call their parents, but only do so at a speciﬁc time of the day for a speciﬁc duration.
One of the most accurate means for representing activity is to use a time-diary (Michelson, 2006). This technique, however, is very involved and is meant to capture an intensive picture of an individual’s day. This study, on the other hand, needs to assess the extensive use of many media with a number of alters. Since one interacts with only a fraction of one’s personal network on a given day or in a given week, it is necessary to use more crude categories. Both the survey and the interview use ordinal scales of frequency, generally in some variant of “daily”,“weekly”,“monthly” and “yearly”.6 To reduce the complexity of models using these ordinal responses, I have converted these to either days or times per month. This enables me to use regular parametric models, rather than more complex and often less accurate nonparametric equivalents.
Also, it is important to distinguish the kinds of media use for a given individual.
The time spent per day on the Internet is not an accurate measure of the time spent on Internet-based social activities such as email and instant messaging. Some individuals may spend most of their time online playing solitary games while others may only go online once a week, but do so explicitly to check their email. The same can be said of mobile phones. Some individuals may spend a substantial amount of time having expressive conversations with a single alter, while others would be able to have many short instrumental conversations using the same number of mobile phone minutes.
For this reason, the media use measures in this study will be ﬁt to the research question where possible. In Chapter 5, I examine the relationship between social activity I say “some variant” because several survey instruments also included mid-points, such as “more than weekly” and “more than daily”. The speciﬁc responses vary and are explained in-text alongside the measures where relevant.
CHAPTER 4. RESEARCH METHODS AND MEASURES 87and the media used to organize this activity. Therein, the measures is based on the number of times per month an individual uses a particular medium to plan (i.e., to use a medium instrumentally). By contrast, Chapter 7 focuses on the overall number of media that individuals use to access their alters. There, a simple dichotomous measure of use/non-use is sufﬁcient. Additional details about the media use measures are explained in the respective relevant chapters.
Chapter 5 Network proﬁles: Coupling media use and social activity
partment store in Manhattan for a brief public spectacle. In the back of the store, over a hundred people huddled around a very expensive rug. Besides showing up, their only instructions were to claim that they were part of a rural collective and all decisions especially ones on such an expensive goods were to be made as a group.
For fear of blowing their cover or aggravating the store owners, the group quickly dissipated back to their everyday lives (Rheingold, 2003).