What '1990 to Today' Actually Compares in Friendship Surveys

The 2021 Survey Center on American Life report on friendship contains one of the most quoted statistics in recent writing about young adults: 12% of Americans report having no close friends at all, up from 3% in 1990. Among men, the share with no close friends rose fivefold over the same period, while the share with six or more close friends fell from 55% to 27%. These numbers travel well because they are stark and because they appear to come from the same measuring instrument applied thirty years apart. That appearance is worth examining before the numbers are treated as a trend line.

What “the same survey” requires

A genuine time series needs three things held roughly constant: the wording of the question, the mode of administration, and the sampling method. A friendship question asked by telephone to a random-digit-dial sample in 1990 is not measuring the same thing as one asked online to an opt-in panel in 2021, even if the words on the page are identical. Response patterns to sensitive personal questions shift with mode: people tend to report more socially undesirable answers, including friendlessness, when a survey is self-administered online than when a live interviewer is on the line. That alone could account for some portion of a rise that has nothing to do with actual changes in social life.

The AEI report’s 1990 figure is presented as a baseline comparison rather than as an earlier wave of the same ongoing program — the Survey Center on American Life is a relatively recent research initiative, and the report does not specify that the 1990 data came from an identical instrument run by the same organization under the same conditions. That does not make the comparison worthless. It makes it a comparison between two different snapshots that happen to ask a similar question, which is a weaker claim than “this is what happened over three decades,” even though it is often written and read as the latter.

A closer case: Cigna’s own tracker

Cigna’s loneliness index offers a cleaner test case, because it is explicitly designed as a repeated measure run by the same organization. The 2020 report puts the share of U.S. adults who say they sometimes or always feel lonely at 61%, a rise of seven points from the prior year, with 73% of workers aged 18 to 22 reporting loneliness. Here the comparison is at least nominally within one series, tracked by one sponsor using one instrument. But 2020 was also the first year Cigna’s fieldwork ran through the early months of the pandemic, a period when survey administration for many organizations shifted in ways not always documented in the summary report available. A seven-point jump in a single year is large enough that mode or timing effects deserve scrutiny before it is read purely as a change in how lonely American workers actually became.

Retrospective self-report is a different measurement problem

Harvard’s Making Caring Common project took a third approach entirely: rather than comparing two independent waves, its 2021 survey asked young adults directly whether their loneliness had increased since the pandemic began. Forty-three percent of respondents aged 18 to 25 said it had. This is not a repeated measurement at all — it is a single measurement of memory. Asking someone to estimate the direction and rough size of a change in their own emotional state over roughly a year is a different task from measuring their current state twice and subtracting. Retrospective self-report is vulnerable to the respondent’s current mood coloring their recollection of the past, a well-documented bias in survey research generally. The 61% figure for serious loneliness among 18-to-25-year-olds in the same survey is a more solid number, because it describes a present state rather than a remembered change.

Where comparability holds up better

Not every source in this literature has the same problem. The AARP Foundation’s 2018 survey of adults 45 and older used the 20-item UCLA Loneliness Scale, a validated instrument used across decades of academic research rather than a bespoke question written for one report. That does not create a time series either — AARP’s survey was a single wave — but it does mean the finding that one in three older adults are lonely can be set alongside other studies using the same scale with more confidence than a comparison built on differently worded, differently administered questions ever could.

Robert Putnam’s account in Bowling Alone of declining civic and social participation across the twentieth century drew on dozens of distinct data sources — membership rolls, time-use diaries, survey archives — precisely because no single continuous survey covered the whole period. Putnam’s argument is convincing partly because the same directional pattern shows up across many imperfect, non-comparable sources, not because any one of them constitutes a clean series.

That is the more defensible way to read the friendship and loneliness literature generally. The direction — fewer close friendships, more reported loneliness among young adults, over recent decades — recurs across the AEI, Cigna, and Harvard data despite their incompatible instruments. The precise magnitude of change implied by any single “then versus now” statistic is weaker evidence than the number of citations it receives would suggest. A study that actually settled the question would track the same individuals, or at minimum the same probability sample design and question wording, at multiple points using a validated scale — something closer to what AARP did in a single wave, repeated.