When a Trend Line Isn't One: Measuring Loneliness Over Time

Cigna’s 2020 loneliness report states that 61% of U.S. adults report sometimes or always feeling lonely, up seven points from the year before. That single number, “up 7 points,” does more work in press coverage than any other figure in the report. It implies a trend: loneliness rising, measured consistently, on a fixed scale, from one year to the next. Whether that implication holds depends on a question that gets skipped almost every time the figure is cited: was the 2019 wave the same survey, fielded the same way, to a comparably drawn sample, using the same questions?

This matters more than it sounds like it should. A single cross-sectional survey tells us the state of things at one moment. A genuine series tells us about change. Those are different kinds of evidence, and confusing them is one of the most common errors in how loneliness statistics circulate.

What comparability actually requires

For two survey waves to support a claim about change, three things need to hold. The instrument has to be identical or validated as equivalent — same questions, same response options, same recall period. The sampling frame has to be comparable — same population definition, similar recruitment method, similar weighting. And the mode should not shift in ways that change how people answer — a phone survey and an online panel do not reliably produce the same loneliness rate even when the underlying question is word-for-word identical.

Miss any of the three and a reported change could reflect the instrument rather than the world. This is not a minor technical quibble. It is the difference between “loneliness rose” and “we changed how we asked about loneliness.”

The case that looks like a series but isn’t

The American Enterprise Institute’s 2021 survey on American friendship is one of the most frequently cited sources for the claim that American social life has thinned out over three decades. It reports that 12% of Americans have no close friends, up from 3% in 1990, and that the share of men with six or more close friends fell from 55% in 1990 to 27% in 2021, with men reporting no close friends increasing fivefold over the same period.

Those are dramatic numbers, and they may well point to something real. But it is worth being precise about what this comparison is. It is not a repeated survey series in the sense described above. The 2021 figure comes from a survey fielded in 2021; the 1990 figure comes from a different survey conducted around that time, and the two were not designed as a matched pair with identical wording, sampling, or mode three decades apart. Comparing them assumes a level of instrument stability across a 31-year gap that is difficult to verify and, given how survey methodology and even the meaning of “close friend” have shifted over that period, difficult to fully credit. The comparison is suggestive. It is not the same category of evidence as a panel study that reinterviews the same design at fixed intervals.

A genuine series: JACSIS in Japan

The JACSIS study offers a useful contrast. Its 2023 paper on social isolation and loneliness during the COVID-19 pandemic in Japan is explicitly longitudinal, tracking the same cohort across 2020 and 2021 rather than drawing two independent cross-sections and setting them side by side. Because the design holds the instrument and the sample constant and follows the same respondents through time, a reported change between the two waves is much more likely to reflect an actual shift in isolation or loneliness rather than a shift in who was asked, or how.

This is the standard that trend claims about loneliness should be held to, and the standard almost none of the widely quoted figures meet. Repeated cross-sectional surveys — the same questions fielded to a freshly drawn but comparably specified sample each year — are the second-best option and are still informative, provided the sampling and wording are actually held fixed. Cigna’s index and similar annual “loneliness index” products fall into this category. Its own report is transparent enough to flag the year-over-year point increase as such, but it does not include the methodological detail — sample composition, exact wording, mode — needed to independently verify that the 2019 and 2020 waves are actually comparable on the dimensions that matter.

Instrument stability versus instrument comparability

There is a separate, related issue worth naming: an instrument can be stable within a single wave and still be incomparable to other sources. AARP’s 2018 survey of adults 45 and older used the 20-item UCLA Loneliness Scale, a standardized instrument with a long history in the academic literature. That choice makes the 33% loneliness figure from that survey unusually well anchored — it can be read against decades of psychological research using the same scale, rather than against a bespoke question written for a single corporate or government report. That is a genuine strength, but it applies to comparing across different studies that share an instrument, not to tracking change within a single series over time, since AARP’s report is a single wave.

Gallup’s 2023 work illustrates the opposite failure mode within a single organization’s output. Its global survey with Meta reports 72% of people worldwide feeling very or fairly socially connected, drawn from interviews across 142 countries between June 2022 and February 2023. A separate Gallup analysis published weeks later reports 24% of people worldwide feeling very or fairly lonely, and a follow-up notes that connectedness and loneliness are not simple inverses of each other. These are not two waves of the same series measuring change; they are two different questions, administered as part of the same broader project, measuring two related but distinct constructs. Reporting them together as though they describe a single trend line — connection down, loneliness up, or any combination — would be an instrument-mismatch error even though both figures came from the same Gallup and Meta partnership in the same year.

What would settle it

A 2023 review in BMC Public Health surveying the state of loneliness and isolation research flags inconsistent measurement as one of the central barriers to comparing findings across studies, and the problem sits exactly here: at the point where a report wants to say something moved, and the reader has no way to check whether the movement is in the world or in the survey design.

A study capable of supporting a real trend claim about loneliness would document, at minimum, identical question wording across waves, a sampling frame specified with enough detail to assess comparability, and a fixed mode of administration. Panel designs that reinterview the same respondents, as JACSIS does, go further and remove sampling variation as a competing explanation entirely. Until a report specifies these things, a claim that loneliness has risen or fallen by some number of percentage points is better read as two snapshots than as one line.