WASHINGTON, DC — A widely publicized analysis of social network size, which reported dramatically increasing social isolation when it was published in 2006, has sparked an academic debate in the August issue of the American Sociological Review (ASR), the flagship journal of the American Sociological Association.
The 2006 report by sociologists Miller McPherson, Lynn Smith-Lovin and Matthew E. Brashears found a near tripling in reports of Americans' social isolation—the percentage who said they discussed important matters with no one—between 1985 and 2004. The increase in social isolation was reduced markedly by sophisticated modeling of the data, yet a very significant decrease in social connection to close friends and family remained. Data underlying the findings came from the 1985 and 2004 General Social Surveys (GSS), collected by the National Opinion Research Center (NORC) and funded by the National Science Foundation. The GSS has been fielded since 1972.
But sociologist Claude S. Fischer of the University of California, Berkeley, calls the research team's findings highly implausible based on the immense scale of the reported change, anomalies in the GSS data and contrary results in data on other types of network ties.
"Results that seem to be too good, too strong or too stark to be true probably are, as seems to be the case in this instance," said Fischer. "The survey question used in 2004 to measure social network size yielded results that were so inconsistent with other data and so internally anomalous and implausible that they are almost surely the product of an artifact."
Fischer highlights peculiar statistics such as the soaring percentage (from 1 to 16 percent between 1985 and 2004) of respondents with post-graduate degrees who named no confidants. Fischer also describes several other available surveys which show that there were no meaningful changes in Americans' social connections over the same years.
The study in question, "Social Isolation in America: Changes in Core Discussion Networks over Two Decades," appeared in the June 2006 issue of ASR and was updated in a December 2008 ASR erratum after NORC announced a coding error affecting 41 responses in the 2004 GSS. The erratum corrected the 2006 findings and provided updated tables and figures, but did not substantially change the study's conclusions.
Responding to Fischer's comment, the research team of McPherson, Smith-Lovin and Brashears argues that because the survey questions are identical in the two studies, the burden of proof lies with Fischer to show the presence of an artifact. Their statistical analysis, they argue, strongly suggests that Fischer's proposed artifact is not credible.
McPherson and his coauthors point out that Fischer's so-called anomalies often come from very small sub-samples in the data. For example, the increased percentage of highly educated respondents with no confidants mentioned above results from a shift of just 22 cases in a study of nearly 3,000 respondents
In their reply to Fischer, the authors alert scholars to the risks of attempting to oversimplify complex research. According to McPherson, attention to the simple percentages rather than the researchers' statistical models produced an inaccurate picture of American's most intimate social circles in many media reports.
"We are very pleased that Professor Fischer's reanalysis supports our original contention that the 2004 data overstate the prevalence of social isolation, a caution that was included in the abstract of our original 2006 report," said McPherson. "However, we disagree that the data show no change in social isolation 1985-2004."
"Our statistical analysis and those of independent observers continue to find significant change in social isolation as measured in the publicly available GSS data," McPherson said. "Interested readers can easily access those data on several public Web sites (for example, www.sda.berkeley.edu) and do their own analyses. We eagerly await new network data from the 2010 General Social Survey."