National Web-Based Survey and Network Generator
A telecommunication company, SK telecom, sponsored this study as a part of a media user project through the Korean Society for Journalism and Communication Studies. The data for the present study have not been analyzed or reported on before. During the web survey development, 10 communication researchers in Korea peer-reviewed the questions, adding, subtracting, and revising questions. A survey company then took over the survey questionnaires, conducted a brief pilot survey, revised some questions, and implemented the survey. Although this survey was conducted via the web, it was designed to represent the Korean adult population as broadly as possible. The commercial company maintains a large master survey pool developed and maintained over time, providing small gifts for people who respond to their advertisements for participation, for registration to the master pool, and for participation in each survey.
Registrants provide demographic information and information about their media ownership and use. The company uses the demographic information to create a proportional stratified sampling frame by gender and age categories, appropriate for the specific survey topic (such as email and mobile phone users). The entire master pool is solicited by email, and the survey stays open until each category is filled, closing down each category when it reaches its criterion sample size. Typically, this takes one week to 10 days. The survey program also evaluates responses and rejects surveys with skipped questions and random or invariant responses. A total of 1,507 people responded to the present survey. As the survey collection ended when the stratified categories were filled, there is no “response rate” to report. We collected information on the respondents’employment category, their media use, up to five people in up to 15 social roles they communicate with through each of those media, and the closeness of those relationships, as a limited set of social contexts influencing media use.
The web survey asked respondents to indicate their gender and age (for descriptive purposes only) and to check their employment category from a list of 11 occupations: Administrator/management, IT technician or professional, salaried, sales/service, simple technician or laborer, agricultural/fishery, housework, middle or high school student, college student, no occupation, and other. (The survey considered a person as “salaried” if (a) s/he gets paid regularly (by month) and (b) if his/her company size is over 100 employees and his/her position’s rank is lower than a managerial position, or (c) if his/her company size is under 100 employees and his/her position’s rank is lower than a mid-level managerial position. It does not include government office workers; religion, art, athletic field workers; employees in political and non-governmental political organizations; self-employed, free-lance workers; doctors or nurses; lawyers or accountants; salespersons (at stores or door-to-door); paid-by-day workers; or others such as farmers, fisherman, students, professors, teachers, or military personnel.) Because of very low frequencies for some of the categories, they were grouped into six categories: salaried, homemaker, middle-/high-school student, college student, IT professional, and others. The “other” category, with 468 respondents, was not included in the analyses, to maintain a consistent sample across the analyses that did and did not include employment categories. Thus we used the sample with the five general employment categories throughout.
This sample of 1,039 respondents consisted of 44% males and 56% females, distributed across the following age ranges: 13–19 (29.8%), 20–29 (27.9%), 30–39 (27.7%), and 40–49 (14.5%), and distributed across the following employment categories: salaried (27.3%), home worker (23.5%), middle/high school student (14.6%), college student (22.9%), and technical/professional (11.6%).
On a separate set of web pages for each of five different media (face-to-face—FtF; email—EM; instant messenger—IM; mobile phone—MP; and short messaging service—SMS), each respondent identified a maximum of five communication partners with whom s/he communicates most frequently. The respondent listed each communication partner by number (i.e., “person 3”) and identified that partner distinctively across the media. That is, if a partner was identified in one medium but also appeared as a partner in the list for another medium, the respondent would mark the partner as the same person (“person 3”). This approach is midway between an ego network (respondents list the others with whom they have contacts, with no attempt or ability to assess links among those contacts) and a system network (respondents indicate contacts on a roster of a bounded system of actors).
Here, we used what Marsden (1990) refers to as a role relation name generator. For each person indicated, respondents checked the social role (spouse, children, parent, sibling, other relative, elementary/middle/high school friend, college friend, girlfriend/boyfriend/lover, other types of friend, work colleague, work boss/manager, work subordinate, other work related, teacher/professor, or online only). These social roles are fairly similar across General Social Survey (GSS)-type surveys (see Hashimoto, Ishii, Nakamura, Korenaga, Tsuji, & Mori, 2000; Van der Gaag, 2005). We avoided some possible problems of name-generator approaches (a brief review is available from the authors) by using the general referent of “who you most frequently communicate with” (through each medium) and separately asking for an indication of the closeness of each communication relationship (from 1 = not close to 5 = extremely close), which Marsden and Campbell (1984) concluded was the best single item indicator of relationship strength.
Analyses involved both the traditional respondent-by-variable dataset (individual level) and the social role matrices for each medium (network level).
For individual-level analyses, we used repeated-measures ANOVA to test for differences in amount, closeness, and overlap in social roles across employment categories (salaried, homemaker, middle/high school student, college student, and information-related technicians) and media. We performed 5x5 repeated-measures ANOVA with mean value of closeness for each medium as the within-groups factor and employment category as the between-groups factor.
For network-level analyses, we used all respondents’ answers about their communication relationships to create a matrix for each medium. (For these and other network analyses we used Ucinet 6.0; Borgatti, Everett & Freeman, 2002.) This matrix can be called G(k), where k represents the specific medium. Each of the 1,039 rows (i) is a different respondent, and each of the 15 columns indicates a different social role (j). The value in cell (i,j) indicates the number of mentions of communication with that social role (j) the respondent (i) reported for that specific medium. For example, survey participant 232’s communication partners through FtF include 1 spouse (we would hope no more than 1!), 2 work colleagues, and 2 work bosses/managers (for the maximum of five relationships).
Each of those five media matrices was then converted into a matrix Ak. Matrix Ak = Gk*Gk’. Gk is the original 1,039 respondent × 15 social roles matrix for medium k, and Gk’ is Gk transposed. Gk is then matrix-multiplied by Gk’. The resulting matrix Ak is a 15 × 15 social role by social role matrix, aggregated across the 1,039 respondents, for medium k. Note that these values are not the frequency of communication between those social roles; rather, they indicate the frequency with which each pair of these social roles exists in the overall configuration of social roles for each medium, across the entire sample.
As an example, consider the Ak matrix where k is FtF. A value on the diagonal (j,j) indicates the number of times the respondents overall identified that social role j as a communication partner (ranging from a low of 10 teacher/professor communication partners, to 263 spouse communication partners, to a high of 420 middle/high school friend communication partners) for the FtF medium. An off-diagonal (i,j) value indicates the number of times any two social roles (i and j) were both mentioned by any respondent. For example, 263 respondents reported communicating FtF with a spouse (diagonal value for the spouse social role), whereas 65 reported communicating with a spouse and with children, 28 with a spouse and with a college friend, and only two with a spouse and with someone only online (off-diagonal values).
We then used Ucinet 6.0 to transform those Ak frequency matrices into correlation matrices. The correlation between any two social roles in a particular medium matrix indicates the extent to which those two social roles have similar patterns of frequencies to all other social roles in the Ak matrix. Just as two variables are correlated in traditional analysis, here two columns are correlated. The correlations between each two columns are placed in the respective cell of a new matrix, and that new correlation matrix is used as input for subsequent analysis. Netdraw (a module of Ucinet 6.0) was used to display the results of multi-dimensional scaling of the correlation matrices visually, showing how the social roles are more or less “close” to each other in each medium. The configuration of these relationships within media can thus be described and compared both visually and statistically.
To compute the correlation (extent of similarity) between pairs of the matrices and the statistical significance of those correlations, we used the Quadratic Assignment Procedure (QAP) in Ucinet 6.0. This permutes (by default, 2,500 times) the rows and columns of one of the pairs of matrices and computes the correlation between the two matrices for each permutation. This is done by converting all the non-diagonal values from one matrix into a single column of values, and the same for the other matrix, and then the two columns are correlated as two variables are typically correlated. The program then creates a distribution of all those 2,500 correlations and determines where along that distribution of possible correlations the empirical correlation between the actual two matrices lies. This nonparametric approach to assessing statistical significance is required because the rows and columns of network matrices are not independent, unlike the traditional assumption underlying survey data and parametric statistics. In order to assess how the social configurations in the four new media were uniquely associated with the FtF social configuration, the multiple regression quadratic assignment procedure (MRQAP) with semi-partialing (Krackhardt, 1988) was applied to compute the overall R2 and each new media matrix’s partial beta coefficients. This uses the same approach as QAP but controls for interdependence among the explanatory matrices.