TL;DR
A meta-analysis of 148 studies covering 308,849 people found that those with stronger social relationships had a 50% higher likelihood of survival — an effect the authors describe as comparable to well-established mortality risks like smoking and larger than physical inactivity or obesity.1 Loneliness also predicts later depression with an adjusted odds ratio of 3.87 in an 11,766-person longitudinal cohort.2 And yet the trend data offers “surprisingly no empirical support” for loneliness actually increasing over recent decades, in the US or in comparable rich countries.34 The condition is serious. The epidemic is a headline.
A stable one-in-three is not reassuring. It is only not new. Photo: Sandy Ravaloniaina on Unsplash.5
Two Claims That Travel Together and Shouldn’t
Almost every discussion of modern social life bundles two separate assertions into one sentence. The first: loneliness is bad for you, seriously and measurably. The second: loneliness is rising, sharply, because of something recent — phones, cities, individualism, the decline of community.
These need to be assessed separately, because the evidence for them is wildly asymmetric. One is among the more robust findings in health psychology. The other, on the actual longitudinal data, barely exists. Keeping them fused produces a conversation that is simultaneously too alarmed and not serious enough.
The Part That Is Very Well Established
Start with the strong claim, because it deserves its reputation.
Julianne Holt-Lunstad and colleagues pooled 148 studies covering 308,849 individuals followed for an average of 7.5 years, and found a 50% increased likelihood of survival for people with stronger social relationships (OR = 1.50, 95% CI 1.42–1.59).1 The authors’ own comparison is the line that made the paper famous: the magnitude is comparable with well-established mortality risk factors like smoking, and exceeds others such as physical inactivity and obesity.1
Two details are worth keeping, because they’re usually dropped. First, the effect varied a great deal by how social connection was measured: complex measures of social integration produced the strongest association (OR = 1.91), functional support measures were intermediate (OR = 1.46), and crude binary indicators like whether someone lives alone produced the weakest (OR = 1.19).1 That ordering is informative — living alone is a poor proxy for being disconnected. Second, the effect held consistently across age, sex, initial health status, cause of death and follow-up length, which is the kind of robustness that makes a finding hard to dismiss.1
And It Isn’t Only Mortality
The mental-health side has comparably serious numbers. A longitudinal analysis of Health and Retirement Study data from 2006–2016, covering 11,766 adults aged 50 and over, found that baseline loneliness predicted the later development of depression with an adjusted odds ratio of 3.87 (95% CI 3.55–4.21).2 The same analysis found loneliness predicting fatigue (OR = 1.88) and pain (OR = 1.54).2
The direction of causation is the sharp version of the question, and it’s worth stating what this design does and doesn’t settle. Because loneliness was measured at baseline and depression at follow-up, the temporal order is at least the right way round — this isn’t depressed people reporting that they feel isolated. What it can’t fully rule out is that early, unmeasured depressive symptoms drove both. The broader literature generally treats the relationship as running in both directions, each raising the risk of the other, which would make it a loop rather than a one-way street — and loops are exactly what people don’t exit unaided.
The Part With No Support
Now the claim that carries the public conversation, and where the evidence collapses.
Reviewing the actual trend data, Our World in Data’s assessment is blunt: “there is surprisingly no empirical support for the fact that loneliness is increasing, let alone spreading at epidemic rates.”3 The specifics are worth listing, because they’re consistent across independent efforts. Hawkley and colleagues found no cohort differences between Americans born 1920–1947 and 1948–1965. Trzesniewski and Donnellan, examining US high-school seniors from 1976 to 2006, found no evidence of cohort trends. Clark, Loxton and Tobin found a statistically significant decline in loneliness among US high-school students. Repeated surveys of Swedish adults aged 85–95 across a decade showed no increase. Reviews covering England, Finland and Germany point the same way.3
A separate account of the same literature reports rates of loneliness stable in England, the US, Finland, Sweden and Germany over recent decades, with the one clear time-series movement being the COVID period and its reversal: among US adults aged 50–80, reported loneliness ran 34% in 2018, rose to 42% during the pandemic, and had fallen back to 33% by 2024 — essentially returning to baseline.4 Sweden is the notable countertrend, where loneliness among older adults appears to be declining.4
The picture holds when you widen the lens. A study drawing on over 218,000 responses across nearly 150 countries found roughly one in five people reporting loneliness on a typical day — and looking at 2023–2024 data across 113 nations, no clear trend, with almost two-thirds of countries reporting lower rates in 2024 than 2023.6
That study also surfaces the variable the epidemic story usually omits entirely. Loneliness ran at 15.7% in high-income countries against 31.3% in low-income ones — roughly double — and over the period examined it rose in poorer regions while declining in wealthier ones.6 Which suggests loneliness tracks material and economic circumstance far more than it tracks whatever cultural villain is currently being blamed in rich countries.
There’s also a measurement confusion doing real work in the popular version: time spent alone and living arrangements are behavioural measures, while loneliness is a subjective state, and the two come apart. Living alone has risen in many countries without a matching rise in reported loneliness — which is the same lesson the Holt-Lunstad effect sizes taught from the other direction, where crude “lives alone” indicators were the weakest predictor in the set.1
Why the Distinction Actually Matters
This isn’t pedantry about a word. The two claims imply different responses.
If loneliness is a rising epidemic caused by something recent, the implied fix is to reverse the recent thing — ban the phones, revive the institutions, treat it as an emergency requiring crisis intervention. If loneliness is a stable, serious, chronic feature of how these societies are organised — which is what the data supports — then the implied response is sustained and structural rather than urgent and reactive, and you should be suspicious of anyone promising to solve it by removing one novel villain.
The epidemic framing also carries a specific cost that critics have named: it implies the problem is novel, growing, and out of control, none of which the data supports, and emergency framing tends to produce short-lived interventions rather than the slow work of building places where people actually see each other repeatedly.
Where I’d Hold This Loosely
Three caveats, and one of them cuts against my own framing.
Most of the strongest longitudinal evidence — the OR of 3.87, the mortality meta-analysis’s most robust strata — comes from older adults, which is a population where isolation is both more common and more medically consequential. Extrapolating those figures to a lonely twenty-five-year-old is not warranted, and the trend data on younger cohorts is precisely where the “no increase” finding is strongest.
Second, “no aggregate increase” is fully compatible with large increases in specific subgroups. A flat national average can conceal a rise among, say, young men or recent migrants, and the absence of a population trend is not evidence that nobody’s situation has worsened.
Third — and this is the caveat against myself — absence of a trend says nothing about the level. If a third of older US adults report loneliness and that figure has been stable for decades, “there is no epidemic” is true and also close to irrelevant to the person in that third. A stable one-in-three is not reassuring. It’s just not new.
Which is the whole shape of the thing: the alarm is misdirected, and the problem is real. Those are compatible, and treating them as if they weren’t is how a genuine public-health finding ends up filed alongside moral panics about screens.
Footnotes
-
https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1000316 ↩ ↩2 ↩3 ↩4 ↩5 ↩6
-
https://theconversation.com/there-is-no-loneliness-epidemic-so-why-do-we-keep-talking-as-if-there-is-259072 ↩ ↩2 ↩3
-
https://unsplash.com/photos/empty-park-bench-at-night-with-soft-lighting-55r-WEIQMhM ↩
-
https://theconversation.com/is-there-really-a-global-loneliness-epidemic-research-reveals-a-more-complex-picture-286543 ↩ ↩2
