«NETWORKING IN EVERYDAY LIFE by Bernard J. Hogan A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy ...»
In each case, the data from the survey and interview were ﬁltered to match the Census data. For example, the Canadian census community proﬁle only publicly reports on the education levels of those 20-64. This is why the N for the survey and interview sample is signiﬁcantly lower than the expected number of cases (350 and 86, respectively). The only case where data were substantially underreported was income.
4.5.1 Personal characteristics The survey sample is somewhat biased in terms of age, gender, and nationality. These biases were in the expected direction for a self-administered questionnaire. Such surSee http://www12.statcan.ca/english/Proﬁl01/CP01/Index.cfm?Lang=E
CHAPTER 4. RESEARCH METHODS AND MEASURES 62veys are more likely to be done by older individuals, as well as females. Immigrants are underrepresented as several of them found the survey (written only in English) too difﬁcult to complete. Nevertheless, none of the biases are greater than 4 percent.
Age shows a substantial regression towards middle age. The median age of the sample is not far off of the median age of the population, however, there are fewer elderly individuals in the sample as well as fewer younger individuals. This is the result of two forces. Biasing the sample against elderly individuals is the fact that this survey was framed in terms of technology use. As such, a number of older individuals felt it was not relevant. Other older individuals were in care, others still had died. This last point is because the sampling frame was approximately a year old by the time we deployed the surveys. This also helps to explain the relative lack of younger individuals. Younger people are more likely to be transient, to rent, to share a place with other young people, and to not have a landline telephone. Thus, a landline telephone number-based sampling frame will underrepresent these individuals. (Wellman et al., 1996) The consequences of this data is that a number of networking styles (discussed in Chapter 5, will be underrepresented. In that chapter I partition the sample into discrete categories based on a ‘style’ of media use and social activity. It is highly probable that the clustering algorithm would pick up on the same styles if this analysis was more representative. The difference being that the relative proportion of styles shown in the population probably varies from a true population estimate.
The interview sample shows the same sorts of biases, although these biases are even stronger. Whereas 53 percent of the population is female, 59 percent of the interview sample is female. The interview sample also shows a positive skew on age, as well as a regression towards middle age. This is due primarily to access. Older individuals, particularly middle aged and those who are retired were more willing
Table 4.2: Household characteristics of East York, the survey sample and the interview sample interview still covers individuals ranging from early twenties to individuals in their eighties, and the age distribution is still normal.
4.5.2 Household characteristics A number of household and marital types were sampled with clear precision. Namely, those who are separated, divorced, widowed, and childless couples comprise nearly the same amounts of both the population and the sample. However, the sample does overrepresent married couples, particularly married couples with children. Correspondingly, the sample underrepresents one-person households as well as living arrangements other than a married couple. Similar to the reasons for underrepresenting younger individuals, I believe this is due to these individuals being transient and thus particularly difﬁcult to access through the sampling frame.
of the population than the survey. However, the interviews were even less likely to capture single-person households. I believe that this is again exacerbated by the transience of these individuals. About 15 people had moved between the survey and the interview and thus could not be contacted. Single people made up the largest share of those who could no longer be contacted.
4.5.3 Socioeconomic characteristics Both the survey and the interview faithfully represent the labor force participation of the population. Labor force participation is deﬁned as individuals who are working (either part or full time) either for an employer or self-employed as well as individuals who are on paid long term leave. This represents 66 percent of the population, as well as 66.7 percent of the survey sample and 65.1 percent of the interview sample.
The Connected Lives survey follows the standard practice of allowing individuals to select an income range rather than give a speciﬁc number. This is why the median income is reported as 50k to 75k, rather than as a speciﬁc value. The median income given by statistics Canada for 2001 was 46.5k. Adjusting for four years of inﬂation, the income of East York should be 50.7k.3 Therefore I believe that income biases in the East York sample are slight at best and generally due to the underrepresentation of singleperson households as well as younger individuals. Also, income has the most missing data of any demographic variable. The missing data are biased toward those of less education. While 15 percent of those with an undergraduate or advanced degree did not report income, 24 percent of those with high school or less did not report income.
Thus controlling for other factors, it appears that there are few biases in the survey values of income.
education while underrepresenting those with less education. This is probably a consequence of the survey deployment. Those who have been to university are more sympathetic to a university-run survey than those who have not.
4.6 Interpretations of the biases of the samples The biases inherent in the Connected Lives survey should not interfere with the conclusions drawn from the analysis. The purpose of the analysis, in general, is to articulate a logic of networking in everyday life. The assertion of the prevalence of speciﬁc logics of networking in the population is a secondary concern to merely spelling out these logics in the ﬁrst place. So, if 21 percent of the sample versus 24 percent of the population are active users of both cell phones and email, then this difference will not interfere substantially with our understanding of why and how this group uses cell phones and the Internet.
That being said, a number of the biases, such as age and gender, are also worth considering as explanatory and control variables. Where appropriate these variables will be kept in models.
Finally, on a positive note, while not perfect, this study does do a good job of representing the diversity of demographic indicators in East York. Apart from education and family composition no indicator is more than six percent off of the population value, and most are within three percent or less. The indicators of income and labour force participation were particularly accurate. Also, given the potential language barrier of new immigrants, this sample also clearly does an effective job of capturing the share of new Canadians in East York. As such, it is unlikely that any conclusions from
4.7 Key concepts and selected instruments This study was focused on an analysis of media use and social networks, primarily.
Both of these domains present speciﬁc challenges to the researcher. This section will highlight the speciﬁc instruments used and indicate how these instruments were employed to meet these challenges. This section will cover social network instruments ﬁrst, followed by media use instruments.
4.7.1 Measuring the personal network A personal network is a set of individuals who share a meaningful relationship with a given person. One of the most challenging parts of operationalizing this deﬁnition is deciding how meaningful is meaningful enough, especially given the need to ensure accuracy in self-administered questionnaires and to minimize respondent burden in an interview setting.
A note on wording: Because both the respondent and the respondent’s network members can be called individuals, this can lead to confusion about who one is discussing. To get around this potential confusion, personal network scholars use the convention of calling the respondent ‘ego’ and any member of the respondent’s network ‘alter’. It is also convention in social network analysis to name the link between any two individuals as an ‘edge’ if it is undirected and an ‘arc’ if it is directed. In this study all relationships are considered symmetric and thus I will be primarily referring to edges. Affective symmetry is a signiﬁcant assumption and one that is made in the interests of simplicity. If we were to interview the alters and get their perspective we would be in a position to understand whether this tie is symmetric and whether the strength is similar. The word ‘tie’ will also be used frequently. In general, a tie means the same thing as an edge. Yet there are special instances where there is a conceptual distinction. One such case is when considering individuals who have multiple
tween ego and alter, although the two share a multiplex/multistranded relationship.
Also, ties are infrequently used in a metonymic way; that is, the word ‘tie’ actually refers to an ‘alter’. This is typically when alter is modiﬁed by an adjective. So ego has an alter, but that alter is a strong tie, or perhaps a work tie.
It is conventional to use socioemotional closeness as a measure for who to include in a personal network. This has been the case since the 1970s, and the ﬁrst personal network studies. The idea that some people are ‘close’ to ego while others are merely acquaintances is also easy to explain to respondents and is well understood culturally.
One of the criticisms of using closeness as a yardstick for who to include in the network is that it may ignore people who have an inﬂuence over ego even if they are not especially close. There are two prominent cases of this. The ﬁrst is when alter is in frequent but non-voluntary contact with ego. For example, individuals may not be close to people from work, but yet see them everyday. A boss may have direct inﬂuence over ego’s social calendar by asking ego to work late, even if ego does not like the boss. This is a valid concern. However, it is seen as a necessary compromise.
We could either use closeness as an organizing principle for the network, or frequency of interaction. Given the history of the use of closeness in personal network studies (Wellman, 1979; Fischer, 1982; Marsden and Campbell, 1984; Boase et al., 2006) and the problems inherent in recalling frequency data (Bernard et al., 1979; Freeman et al., 1987), measuring closeness ﬁrst and then recalling some frequency data later seemed to be an appropriate track.
The second concern is when there is a transitive tie that is relevant for social activity. For example, ego may frequently attend parties with alter. These parties are hosted by alter’s friend, and ego is not close to alter’s friend. Thus, even though the friend is important in deﬁning the opportunity structure for social engagement, they do not appear in the network. This particular case is simply outside of the scope of
I believe these concerns are study-speciﬁc. This is a study about how individuals organize their interaction and communication with others. Those individuals that a person wishes to have voluntary contact with are the most likely individuals to be considered close. Also, any individual who maintains a consistently important role in a person’s social life will tend to become a close individual.
4.7.2 Capturing the network In an interview setting and especially a self-administered questionnaire, it is impossible to expect the respondent to elicit all the individuals to whom they are tied. Past estimates suggest that an individual has anywhere between 250 and 2000 alters depending on the criteria used. The most extensive strategies involve diaries (Boissevain, 1974; Fu, 2007) or speciﬁc comparative sampling measures, such as how many people one knows in prison (McCarty et al., 1997; Zheng et al., 2006). As such, the researcher should select a strategy that ﬁlters out ephemeral and insigniﬁcant ties while highlighting the alters with whom ego shares a relevant tie. For this reason, network researchers commonly refer to the strength of the tie, and seek to capture the strongest relations.
Tie strength was clearly formulated by Granovetter in “The Strength of Weak Ties”.
Here he suggests that tie strength is a “combination of the amount of time, the emotional intensity, the intimacy (mutual conﬁding), and the reciprocal services which characterize that tie” (1973, 1361). In the wake of Granovetter’s research, scholars have sought to uncover a single parsimonious concept that encapsulates most, if not all of these dimensions of tie strength.
Marsden and Campbell (1984) tested the relevance of three of the above dimensions of tie strength, namely the frequency of interaction, the emotional intensity and the intimacy of the relationship. They contend that the emotional intensity of the rela
the idea of “closeness”, and for this reason, the network instruments on the Connected Lives project employ the idea of ties as being either “very close” or “somewhat close”.
As can be seen from the survey instruments, the language of “somewhat close” and “very close” ties permeates the network instruments. However, the survey design team also provided an elaboration of closeness that maps closely on to Granovetter’s
original concept. The survey states:
Please think about the people in your life who do not live with you. We would like you to consider those who you are VERY close to and those who you are SOMEWHAT close to.
•Those that you regularly discuss important matters with,
•Those that you regularly keep in touch with, or
•Those that are there for you when you need help.
•More than just ‘casual acquaintances’, but not ‘very close’.