3.1.2 Survey Methods

Figure 3.1: The General Social Survey name Generator.
Figure 3.1: The General Social Survey name Generator.

Probably the most prominent examples of gathering social network data entail either an ego-centric or socio-centric (i.e., complete) survey design (Marsden 2011). Furthermore, likely the most common of these is the GSS “Important Matters” name generator discussed in Chapter 2 and represented in Figure 3.1. As can be seen, even this relatively simple instrument requires both the actual text of the question to be read to the respondent, and meta-instructions for the interviewer on how to actually implement the item. Because this name generator will be followed by a number of name interpreter questions, it not only enumerates how many important matter discussion partners each respondent has, but also asks for an identifier that will be sufficiently meaningful to the respondent to know who is being referred to in those follow up questions (Burt 1984). However, since it is an ego network design (i.e., the information provided from any one survey’s responses will not be linked up to any other data), the identifiers provided do not need to be meaningful to the interviewer, or those who will later use the data. In fact, these are generally removed from the actual stored data.

Survey-based approaches have a long history in social networks for ego (e.g., Katz and Lazarsfeld 1955),56 partial (e.g., Lee 1969), and complete network approaches (Coleman, Katz, and Menzel 1957).

Among the most commonly known complete social network surveys is the Add Health friendship network data.57 The question format looks similar to the GSS, with a prompt (“List your five closest male/female friends.”) and provides five blanks for providing those names. However, the in-school version of this questionnaire presented here was self-administered, so the instructions are all directed to the respondent rather than an interviewer. Since this is a complete network design, and the researchers planned to link these nominations up to other members of the study population, the scantron responses identified each friend from a provided pre-populated roster of the population.58 This facilitates the process of matching each nomination to other members of the population.

The Colorado Springs “Project 90” also aimed to link named partners to other members of the study population. But, given that there was no population roster available, they relied on an open-recall approach to eliciting nominations across each of the ties about which they gathered information (sexual partners, drug sharing, social contacts). Within their name interpreter questions, Project 90 gathered substantial node attribute information about each nominated alter, both to provide that information about alters not subsequently recruited into the study, and to facilitate with the matching process for identifying which nodes (among the respondents and named alters) represented the same individual. This highly difficult node matching process consumed a large proportion of the research staff’s resources (Potterat et al. 2004), highlighting the benefits of a roster when possible for constructing representations of full networks if that is part of the study’s aims. Particularly within partial network designs such as this, researchers have increasingly relied on methods like tokens with unique recruiting partner identifiers on them to assist in this record linking process.59