What the CDC's Loneliness Numbers Actually Sample

The CDC’s Morbidity and Mortality Weekly Report on loneliness, published in June 2024, is frequently cited as a federal, authoritative estimate of how lonely Americans are. It is federal. Whether it is the estimate people think it is depends on what the Behavioral Risk Factor Surveillance System, the instrument behind it, was built to do.

BRFSS is a telephone survey administered by state health departments, coordinated by the CDC, that has run continuously since 1984. It samples adults through landline and cell phone random-digit dialing, weights the results to match each state’s known demographic distribution, and pools state data into national estimates. It was designed to track chronic disease risk factors — smoking, seatbelt use, diabetes screening — not subjective psychological states. Loneliness items were added as an optional module, which several states chose not to field. The MMWR report’s national estimate is therefore not built from a single uniform national sample; it is built from a patchwork of states that opted in, weighted afterward to look national.

That matters for three reasons that rarely make it into summaries of the finding.

The 2022 data problem

The report was published in June 2024, but the underlying survey wave was collected in 2022. Two years is not a small gap in a fast-moving area. The period immediately following the acute phase of the pandemic, when much of the loneliness literature — including the Harvard Graduate School of Education’s Making Caring Common survey, fielded in late 2020, and the Surgeon General’s 2023 advisory, which drew on data through 2022 — was still describing conditions shaped by remote work, school closures, and reduced in-person contact. A 2022 telephone survey published in 2024 sits between those two moments: past the worst of pandemic isolation, but not necessarily representative of conditions in 2024 or 2025. Anyone citing the CDC figure as a current snapshot of American loneliness is citing something two calendar years and one significant social transition removed from its own publication date.

Who answers a landline or cell-phone survey

BRFSS’s mode is not incidental. Telephone surveys, even ones that call cell phones, systematically undersample certain groups relative to their true population share: people who screen calls from unknown numbers, people without stable phone service, people in institutional settings, and — persistently across survey research — younger adults, who are less likely to answer an unsolicited call at all. This is not a hypothetical concern specific to BRFSS. It is a known property of telephone survey methodology generally, and it cuts in a particular direction for loneliness research: if younger adults are the group most likely to report loneliness, as the Harvard survey found for adults 18 to 25 and as the AARP Foundation’s 2018 survey found by a different route for older cohorts, and if younger adults are also the group least likely to be reached by telephone sampling, then a telephone-based prevalence estimate has a plausible structural reason to run low, independent of anything respondents actually feel.

The AARP survey sidesteps this particular problem by using a different mode and sampling frame entirely — a national survey of 3,020 adults 45 and older using the 20-item UCLA Loneliness Scale, a validated psychometric instrument rather than a single self-report item. The CDC module, by contrast, typically asks a brief single-item or short-form question about frequency of loneliness, embedded among dozens of other health items in a phone interview that respondents did not sign up for expecting to discuss their emotional state. Single-item measures and multi-item validated scales do not produce numbers that mean the same thing, even when both are called “loneliness prevalence.” This is a recurring problem across the loneliness literature generally: the BMC Public Health review of the state of the field, published in 2023, flags inconsistent measurement as one of the central barriers to comparing findings across studies, and the CDC-versus-AARP contrast is a clean illustration of exactly that barrier rather than an exception to it.

What “national” means when states opt in

The deeper sampling issue is coverage, not mode. BRFSS is administered state by state, and the loneliness module was optional. States decide independently whether to field it, which means the pool of respondents answering the loneliness question is not a random sample of the United States — it is a random sample within each state that chose to ask, aggregated afterward. If the states that opted in differ systematically from the states that did not — in age structure, urbanicity, health infrastructure, or the political salience of loneliness as a policy issue — the aggregate figure inherits that selection, no matter how carefully each individual state’s sample was weighted internally. The MMWR report does not claim uniform 50-state coverage, and a careful reading of the methods section is required to see which states actually asked the question. A topline national percentage, quoted without that caveat, implies a completeness the underlying data collection did not have.

None of this is a claim that the CDC estimate is wrong. State-level weighting within BRFSS is a mature, well-documented process, and the survey has decades of validation behind it for its original purpose of chronic disease surveillance. The concern is narrower: that a surveillance system engineered to track smoking rates and diabetes screening was adapted, via an optional module, to answer a question — how lonely is the country — that it was not built to answer with the same precision, and that the resulting number circulates with a confidence its sampling frame does not fully support.

Comparing across instruments, carefully

This is why the CDC figure, the AARP figure, the Harvard figure, and the Surgeon General’s roughly-half-of-adults figure from the 2023 advisory do not converge on a single number, and should not be expected to. They differ in mode (phone versus online panel versus other), in instrument (single item versus UCLA scale versus other short forms), in population (45-plus versus 18-25 versus general adult), and in which year’s conditions they actually captured. Treating them as four measurements of the same underlying quantity, and then puzzling over why they disagree, misreads what each one is. Treating them as four measurements of related but distinct quantities, taken with different tools pointed at different slices of the population, is closer to correct — and considerably less satisfying if what is wanted is a single clean number for a headline.

What would settle this

A stronger national estimate would field the same validated multi-item instrument — the UCLA scale, for preference, given its use in the AARP survey and the wider academic literature — across a probability sample drawn to guarantee full state coverage rather than opt-in modules, repeated on a fixed annual cadence so that trend, not just level, becomes interpretable. It would also report mode effects directly: field a subsample by web panel and by phone in the same wave, so the size of the telephone-coverage gap for younger respondents could be estimated rather than argued about. Until something like that exists, the honest way to use the CDC’s 2022 loneliness figures is as a lower-bound, state-selected, telephone-mode estimate of a particular kind of loneliness question — not as the loneliness rate for the United States.