What people do not admit in a survey about relationships

Social desirability denotes the tendency to answer according to what is valued rather than what is experienced. It is measurable, notably through dedicated scales, the best known of which was proposed by Douglas Crowne and David Marlowe in 1960.

Questions about relationships are particularly exposed to it, because they concern a highly valued dimension and because the answer cannot be verified. Nobody checks the number of friends one reports, and everyone knows what is expected.

The mode of administration alters answers: self-administered questionnaires obtain higher reports of isolation than face-to-face or telephone interviews. Roger Tourangeau and Ting Yan synthesised in 2007 the work on sensitive questions, which finds this mode effect on many subjects.

That gap is one of the most solid methodological arguments for the existence of a bias, because it does not depend on an assumption about what people feel. It is also one of the simplest to check, since it is enough to compare two modes of administration on comparable samples.

The reference loneliness scale, developed at the University of California, Los Angeles by Daniel Russell and colleagues from 1978 onwards, avoids the word for precisely that reason. Other instruments use indirect techniques, asking for instance to estimate the frequency of a behaviour in the population rather than in oneself.

So-called randomised response techniques, proposed by Stanley Warner in 1965, which guarantee anonymity by construction, have been used on other sensitive subjects and remain little used in this field.

The social cost of an admission varies by country, generation and background. A comparison between two populations therefore partly measures the difference in that cost, and not only the difference in situation. Reporting few friends does not carry the same weight everywhere, and that weight itself changes.

That is one reason international comparisons of loneliness are considered the most fragile of all relational indicators.

The opposite objection deserves to be set out. Some people under-report their relationships out of modesty, or because they reserve the word friend for very few. The bias therefore has no single direction, and correcting it requires knowing each group's norm.

If the legitimacy of reporting loneliness rises, a measured increase may reflect a fall in the cost of admission rather than a rise in the phenomenon. A rising series may then describe a change in norms as much as a change in lives.

That hypothesis is plausible given the place the subject has taken in public debate, and it is rarely tested, for want of an instrument measuring the response norm separately.