How someone's friends get counted

The standard instrument asks a person to name the people with whom they have a certain kind of relationship, then asks questions about each of them: age, occupation, length of the tie, frequency of contact. It produces a list, not a global figure reported off the cuff.

That method avoids rough estimation, but it imposes lengthy work on the respondent, which in practice limits the number of names collected. Many surveys also cap the list, so that the number obtained sometimes says more about the questionnaire than about the network.

The most used generator asks with whom the person discusses important matters. Others ask whom they would turn to in difficulty, whom they saw in the past month, or whom they consider a friend. Claude Fischer's survey in Northern California, published in 1982, combined several of them for that reason.

Those questions do not select the same people and produce networks whose size varies several-fold. Comparing two surveys that do not ask the same question therefore makes no sense, even when the titles of the publications use the same word.

The important matters question, for example, picks out confidants but not necessarily friends: a spouse or a colleague often appears, while a long-standing friend seen for other reasons may be missing. It measures a function, not a category of relationship.

An American publication in 2006 reported a sharp fall in the number of reported confidants since 1985, a result very widely taken up as proof of social disintegration.

Later reanalyses, notably by Claude Fischer, questioned collection artefacts, notably a position effect in the questionnaire and an unusual number of respondents naming nobody. The reported decline has been considerably qualified, though its first version has never stopped circulating.

The number of names given depends on the effort made, itself sensitive to questionnaire length, fatigue and the position of the question. Respondents learn that a name given brings additional questions.

That mechanism produces a downward bias, the stronger the longer the questionnaire, and it is one of the explanations retained in the 2006 episode. The interviewer matters too: several studies have found systematic differences in network size depending on who administered the questionnaire.

Position generators, developed notably by Nan Lin, ask whether the person knows anyone in a given occupation, which measures the reach of the network without requiring names. They tell us about access to social milieus rather than about intimacy.

Scale-up methods, which estimate network size from the number of known people bearing a rare first name, give far higher orders of magnitude, because they measure acquaintance rather than relationship. None of these instruments is the right one: each answers a different question.