Weighing a Tie: Assessing Inaccuracies in the 2004 General Social Survey Network Items

Download data and study materials from OSF

Principal investigator:

Matthew E. Brashears

Cornell University

Email: meb299@cornell.edu

Homepage: http://www.soc.cornell.edu/faculty/brashears.html


Sample size: 2061

Field period: 10/9/2009-5/11/2010

Abstract

What is the size of the typical American discussion network? Recent efforts to use the General Social Survey (GSS) to answer this question have garnered substantial interest and led to questions about potential methodological defects. This project evaluates the sensitivity of estimates of network size derived from the GSS "important matters" name generator to respondent fatigue and training effects, and generates a new, nationally representative estimate of mean network size. Secondarily, it tests a method for improving responsiveness to the "important matters" item and assesses the amount and type of social support that is typical of discussion network ties. The results indicate that modern discussion networks have decreased in size, but that social isolation has not become more prevalent. It also disconfirms several explanations for shrinking networks, and supports the usefulness of the new proposed method of enhancing response to name generator items.

Hypotheses

Training Hypothesis: Respondents who believe that providing fewer names will result in more questions will provide fewer names in response to a name generator item.
Fatigue Hypothesis: The number of names respondents provide in response to a name generator is negatively related to the number of questions that precede it.
Priming Hypothesis: Respondents who are asked general questions about their relationships prior to the name generator will report more discussions of important matters than those who are not asked such questions.

Experimental Manipulations

The experiment utilized a two-by-two design crossing manipulations meant to test the training hypothesis (i.e. the training manipulation) and the priming hypothesis (i.e. the priming manipulation). These manipulations, described in greater detail below, should influence the number of names given in response to a network name generator. The fatigue hypothesis was not directly tested using an experimental manipulation, but did receive an indirect test via comparison of results from this study to results from other non-experimental but otherwise similar studies.

Testing the training hypothesis required a contrast between a case where respondents expect that providing more names would result in a larger number of subsequent questions (training manipulation present) and a case where respondents did not expect such a link (training manipulation absent). This expectation was induced by presenting the following text prior to the name generator:

"In a moment you will be asked to give a series of first names or the initials of persons with whom you have had a certain kind of contact. For each person you name you will be asked additional questions about that person (for example, their sex, education, age, etc.), questions about the type of relationship you have with them (for example, how long you have known them, how often you have seen them, etc.), and about the type
of relationship each named person has with every other person you name."

When the training manipulation was present, this text was presented to subjects prior to the name generator and when the training manipulation was absent, this text was not presented to subjects at all.

In order to test the priming hypotheses respondent attention had to be directed to their network ties prior to the name generator. This was accomplished using several items asking respondents to indicate approximately how many others they know who met certain criteria. These global estimator items ask the respondent to give a numerical estimate of the number of persons who meet certain criteria without the need to specifically name these individuals. In contrast, a name generator asks the respondent to name a specific set of persons who meet a specific set of criteria, hypothetically producing an exhaustive list, from which a count can then be generated. Answering a global estimator question should cause the respondents to think about the range of persons they know who meet the criteria, thereby improving response to the name generator itself. Respondents answered four global estimator items, indicating how many others they know by name and face, how many others there are with whom they could discuss important matters, how many others there are from whom they could borrow a large sum of money, and how many others there are with whom they could engage in social activities. In all four items there was no requirement that the respondent have actually engaged in the specified activity with a specific person during a particular period of time. Thus, while the important matters name generator asks about those persons with whom the respondent has actually discussed important matters within the preceding six months, the equivalent global estimator asks how many persons there are with whom the respondent could discuss important matters if they wished to, regardless of whether or not such a conversation has taken place. The global estimate is thus the number of persons in the set from which active discussants have been drawn, and the important matters global estimator item and important matters name generator item have a set–subset relationship. When the priming manipulation was present respondents were asked for global estimates before being exposed to the name generator and when the priming manipulation was absent respondents were asked for global estimates following the name generator and its corresponding name interpreters.

Outcomes

The primary outcome variable is a1_count, which gives a count of the number of names listed by respondents. Experimental condition is given by the xtess023 variable. In this variable "Block A" refers to the "discusses important matters" name generator and associated name interpreters. "Block B" refers to the priming manipulation (i.e. the global network estimator items) and "Block C" refers to the training manipulation (i.e. the block of text). Variables a4, a5 and a6 capture the willingness of one, randomly chosen alter to provide non-monetary material aid, companionate support, and monetary aid, respectively. Variable a8 captures the reason why certain individuals had no discussions of important matters to report. Variables b1-b4 contain the results of the global network estimator items.

Summary of Results

Results support the priming hypothesis and fail to support the fatigue and training hypotheses. Subjects exposed to the priming manipulation provided more names than those not exposed, as expected. Subjects who were trained (i.e. exposed to the training manipulation) provided more names than those who were not trained, counter to expectations. Finally, comparison of these results to other, longer survey results suggest that fatigue did not impact the number of names given. In general this suggests that discussion network size has decreased since 1985. The results also suggest that current approaches to measuring social isolation are unreliable. See Brashears 2011 for more details.

References

Brashears, Matthew E. 2011. "Small networks and high isolation? A reexamination of American discussion networks." Social Networks 33: 331-341.

Brashears, Matthew E. 2014. "'Trivial' Topics and Rich Ties: The Relationship Between Discussion Topic, Alter Role, and Resource Availability Using the 'Important Matters' Name Generator." Sociological Science 1: 493-511.