america is not getting lonelier

A key empirical question in social network analysis is whether Americans have more or less friends over time. Famously, Robert Putnam argued that indeed, we were “bowling alone.” In contrast, critics contend that these are misinterpreted results. Some types of networks disappear, while other appear.

On the social network listserv, Claude Fischer provides the latest round in the debate. Fischer uses 2010 GSS data to claim that the decline in strong personal relationships reported by McPhereson et al. (2006 in the ASR) is due to survey question construction. I’ll quote Fischer’s entire announcement:

Results of the 2010 GSS Network experiment indicate that the increase from 8 to 25 percent of respondents with no confidants between 1985 GSS and 2004 GSS (McPherson et al., ASR 2006) was a result of procedural differences between the surveys.

In 2006, McPherson, Smith-Lovin, and Brashears (ASR) reported that in the 2004 General Social Survey about one-fourth of respondents said they had no one with whom they discussed personal matters, a dramatic increase from the roughly eight percent who said that in the 1985 GSS. This report garnered widespread attention. In 2009, Fischer (ASR) argued that the results were anomalous and likely an artifact. McPherson and colleagues reported slightly adjusted statistics (due to 41 miscoded cases), but strongly defended the basic finding of a sizeable, if perhaps overstated, surge in the percent who reported no confidants.

In 2010, the GSS ran a survey experiment, designed by Peter Marsden, in third wave of its 2006 panel to test whether procedural differences accounted for the 1985/2004 differences. Three random subsamples of the panel were asked the classic “discuss important matters” name-eliciting question, each roughly following the procedures used in the 1985, 1987, or 2004 instances of the networks module: (1) The 1985 procedure asked the question early in the survey, following only the “core” questions of the GSS. (2) The 1987 procedure had the important matters question come later in the interview, following a battery of questions asking respondents about membership in voluntary associations. (3) The 2004 procedure asked the same question near the end of the interview, following the battery of questions about voluntary association membership and questions asking respondents to provide detailed information about one particular association, including the name of a leader and contact information for him or her.

The top-line results of the experiment are:

Percent reporting that they discussed personal matters with no one

1985 procedure (ballot 1/A in 2010)

1985 – 8%
2010 – 5%

1987 procedure (ballot 3/C in 2010):

1987 –  4%
2010 – 13%

2004 procedure (ballot 2/B in 2010):

2004 – 23%
2010 – 21%

Therefore, procedural changes probably account for the year differences, especially for the 2004 v. 1985 contrast.

A recent paper by Paik and Sanchagrin, available on-line (, conducts a deeper analysis of these (and other) data and concludes that the best explanation is an interviewer effect – more interviewers in 2004 than in 1985 avoided or abbreviated the network module.

Personally, I have always been with the critics. Roughly speaking, my Bayesian prior on the “death of society” thesis was 30%. Maybe we might be more isolated, but my casual observation is that people invent new ways to be social. The McPhereson et al. paper revised my prior to about 60%. Now I’m back to 30%

Adverts: From Black Power/Grad Skool Rulz

About these ads

Written by fabiorojas

September 3, 2012 at 2:55 am

5 Responses

Subscribe to comments with RSS.

  1. If a question is this susceptible to variation based on microfluctuation in respondents’ moods, doesn’t that mean it’s a bad question?

    Philip N. Cohen

    September 3, 2012 at 5:44 pm

  2. Phil: According to the Paik & Sanchagrin paper, the problem isn’t with the respondents as much as with (a few) interviewers. As I understand their argument, when the network question is at the end of the survey, some interviewers ask it in a way that would minimize the number of confidants, possibly because each additional confidant triggered a series of follow-up questions.

    I once had a summer job as an interviewer for a random-dial telephone survey. (A job, incidentally, that all survey researchers should hold at least once.) The biggest problem, not surprisingly, was getting the right household member to talk to you, since we were also randomizing within households.

    The second biggest problem, though, was getting the respondent to *stop* talking. Nine times out of ten, people were so excited about the opportunity to tell someone all about themselves and their opinions that they wouldn’t want to get off the phone. The interaction was much less novel for the interviewers. And, yes, there were times when we just wanted to finish up an interview and take our break.


    September 3, 2012 at 7:47 pm

  3. Okay, that makes sense. I still have a preference for hanging major social change conclusions on questions with a little bit more objective grounding, such as, “Do you have a job?”, “Are you alive?”, “Have you graduated from college?, etc.

    (I agree it is a great experience to participate in, or at least witness, real survey interviewing. I had a great summer internship at NCHS where I got to listen in on some of the interviews for the Longitudinal Study of Aging. Quite an ear-opener.)

    Philip N. Cohen

    September 4, 2012 at 2:05 pm

  4. Due to spam, comments are closed for this post. Please email me if you would like post a comment.


    September 17, 2012 at 6:12 pm

  5. [...] lacking a confidant now looks dubious: a new analysis by the sociologist Claude Fischer concluded that the finding arose because of a change in how the questions were asked. There’s little other evidence to suggest that more people are feeling lonelier. Which makes it [...]

Comments are closed.


Get every new post delivered to your Inbox.

Join 971 other followers

%d bloggers like this: