In the period roughly extending 1980-2010, the rapidly growing field of social networks saw an acceleration in the number of studies explicitly concerned with incorporating social network ideas into data collection projects.23 A common feature among many of these projects was their focus on a single type of relationship at a time—or relying on one “name generator” per study. A study’s name generator(s) identifies what type of tie(s) you will elicit for each respondent. The use of any name generator comes with a set of coinciding considerations that determine how, and how many, of those ties you will gather.
Among the most commonly discussed name generators is the one used by the General Social Survey (GSS). The GSS is a nationally representative, population-based survey of US residents that is repeated on regular intervals. In addition to its core set of questions, the GSS also allows for panels of questions to be added by researchers. In 1985, they added a component that aimed to enumerate and briefly describe the composition of personal networks (Marsden 1987; Burt 1984). The GSS question asks “From time to time, most people discuss important matters with other people. Looking back over the last six months—who are the people with whom you discussed matters important to you?” (Marsden 1987: 123, footnote 1). Just like this “important matters” question in the GSS, any individual-level dataset can readily incorporate a name generator to capture a variety of relationship types from survey respondents—some common examples ask who one seeks advice from (or provides advice to), shares resources with, participates in shared activities with, etc.24
An important consideration for each name generator is how specific or general the type of tie is that the researcher aims to capture, and how the collection strategy would shape the ability to gather data that aligns with those aims.25 In survey contexts, the respondent is referred to as ego, while those partners they name in response to a name generator, are generally referred to as their alters. More general name generators make for more inclusive alter sets than more precise ones. For specific research questions, studies will often employ more precise name generators, focusing more directly on alters specific to the focal research topic (e.g., social support, friendship, or domain-specific interactions). For example, the Malawi Longitudinal Study of Families and Health (MLSFH) had various stages of the project with name generators specifically asking about those with whom respondents discussed matters related to HIV/AIDS (Behrman, Kohler, and Watkins 2007), religion (adams and Trinitapoli 2009), and sexual behavior (Clark 2010). While such topic-specific name generators can produce alter sets that are more directly salient to the researchers’ questions (Burt et al. 2012), they effectively provide no clear indicator of each ego’s general relational context from which such specific relationships are drawn.26
Any social network research project can include one or more name generators. Focusing on one per study can provide researcher focus, and reduce respondent burden (Marin and Hampton 2007). A burgeoning literature suggests that even in a research study with relatively focused aims, a single name generator may not sufficiently capture the study’s aims, given different interpretations between researchers and research participants, or among research participants (Klofstad, McClurg, and Rolfe 2009; Marin and Hampton 2007). Having information about multiple types of ties that actors may share is called multiplexity. Ultimately, determining the proper number of name generators must be answered in the context of a research project’s aims. However, that decision should not be made solely based on which name generator(s) are of theoretical import to the research, but also in conjunction with a variety of considerations about how name generators are posed, and that each name generator used often requires a number of “name interpreter” questions—I discuss a number of these in the following section. While the importance of name generators may be clear for survey research, studies employing other primary research designs must also carefully evaluate how many different tie types researchers will aim to capture.
Once a researcher has determined how many tie types to examine for each focal node (ego), they must also consider a number of factors regarding how they will elicit alters for each name generator. For example, which potential pool of alters will ego’s ties be pulled from? In a survey context, responses can either be provided via free listing (also known as open recall), or fixed choice (also referred to as roster-based) response strategies. In free listing approaches, respondents are left to produce from memory the list of alters corresponding to the specified name generator, whereas fixed choice approaches ask respondents to select their alters from a pre-populated list. In many research designs, a full enumeration of the population simply isn’t feasible for a variety of reasons (e.g., membership may be partially unknown), leaving free listing as the only possibility. Open recall studies have been shown to produce more limited alter-nomination lists, and elicit ties more concentrated within clustered domains of one’s personal network.27
Researchers can also choose to cap (or not) the number of alters elicited for each ego on each tie type gathered. For, example, in the MLSFH study on AIDS-discussion partners, interviewers were instructed to prompt the respondents for any additional partners if they nominated fewer than three alters. Many survey studies focus on a relatively small number of elicited partners (three and five are among the most commonly used caps), because these were thought to be sufficient to capture ties that are especially close (Fischer 1982), and/or to reduce the respondent burden that comes from the additional questions each nomination would require (Paik and Sanchagrin 2013).
If researchers do decide to cap alter nominations, they also must consider whether the identified cap will be implicitly or explicitly incorporated into the research design. Implicit nomination caps are those known only to the researcher eliciting the nominations. For example, in the Colorado Springs “Project 90,” which gathered information on social, sexual and needle sharing partnerships among commercial sex workers, intravenous drug users, and their partners (Rothenberg et al. 1995), respondents were limited to reporting on no more than 16 alters across the three name generators used (Potterat et al. 2004); this cap was encoded into the sheets interviewers used for recording data, but was not made known to the respondents. Alternatively, explicit caps are those that are made known to the respondents, sometimes directly in the wording of the question. For example, in the National Longitudinal Study of Adolescent Health (“Add Health”; ), the network module asked middle and high schoolers to name their five best male and female friends. Such caps—whether implicit or explicit—can shape the number of nominations provided in unanticipated ways. An explicit cap could introduce a “floor effect” wherein the question wording may encourage respondents to inflate the number of nominations they provide, in an effort to match the number stated in the question prompt. For example, given the wording of the Add Health question, some have speculated that gender-based homophily (Kandel 1978; McPherson, Smith-Lovin, and Cook 2001) may be under-represented in Add Health data,28 since respondents may have nominated more opposite-gender friends than they would have without such an explicit numeric anchor in the question prompt.29 Alternatively, capping the number of nominations elicited for each name generator could also produce ceiling effects on responses provided, artificially truncating the resulting distribution of reported ties.30