An essay · 09 movements · ~30 min

published 21 May 2026

Trying to Grow Roots in Concrete

Why so many people feel alone, and what the evidence says about where that comes from.

Loneliness usually gets treated as a personal failing.

The evidence points somewhere else.

Scroll when ready.

The short versionThe whole argument in six lines. Each one links to the section where it’s shown and sourced.

About one person in six, worldwide, feels lonely right now, by the WHO’s count. Lonely people also carry about a quarter higher odds of an early death.
Where you live and what you earn change the odds a lot: 24% in the poorest countries, 11% in the richest.
Loneliness spreads: if someone you know is lonely, you are 52% more likely to be too.
This has been building for eighty years. The share of households that are just one person has nearly quadrupled since 1940, daily time with friends has almost halved since 2003, trust in strangers sits well below where it was in the 1970s, and men in particular have far fewer close friends than they used to.
Young people used to be the least lonely age group. By 2024 they were the loneliest.
Loneliness, the essay argues, works a lot like hunger: a signal doing its job in surroundings where the thing it asks for got harder to find. You can score yourself on the same three questions the big studies use. The fixes with evidence behind them are small: see one friend in person, become a regular somewhere, tell someone about the transition you’re in, join something local.

Movement I — Scale

One in six.

About one person in six worldwide feels lonely right now. The WHO puts the figure at 15.8%.

1 in 6

people worldwide currently experience lonelinessWHO 2025

871,000

deaths a year linked to social disconnectionWHO 2025

26%

higher risk of early death for the lonely vs. the well-connectedHolt-Lunstad et al. 2015

29%

higher risk of early death for the socially isolatedHolt-Lunstad et al. 2015

32%

higher risk of early death for those living alone (after controls)Holt-Lunstad et al. 2015

The WHO Commission on Social Connection released the first systematic global synthesis on 30 June 2025. Its headline figure — 15.8% globally, roughly 1 in 6 — pools 23 datasets covering 153 countries and territories. 95% uncertainty interval: 12.8–20.1%.

Forest plot · Holt-Lunstad et al. (2015)

What the meta-analysis says.

Holt-Lunstad et al. (2015) pooled 70 long-running studies covering about 3.4 million people. Each row below is one kind of disconnection and the extra risk of early death it carried. The key underneath explains every mark.

Lonely people carry about 26% higher odds of an early death; the socially isolated about 29%, and those living alone about 32%. The effects are modest, and all three point the same way.

How to read this

The dot: — the best single estimate of the risk. The further right it sits, the higher the chance of dying earlier.
The line through it: — how sure the researchers are. A short line is pinned down; a long one means the true number could sit anywhere along it.
The dashed 1.0 line: — the mark for no added risk at all. Every dot here sits to its right, so each kind of disconnection carried some.
k and N: — the weight of evidence behind the row. k is how many separate study estimates were combined; N is how many people they followed. The top row watched about 880,000 lives across 25 studies.
1.32 [1.14, 1.53]: — the dot and its line written as numbers: about 32% higher risk, with the two figures in brackets marking the two ends of the line.

Why it mattersA headline turns all of this into one number, like “26% higher.” The chart shows what that leaves out: the size of the effect, the doubt around it, and the number of people behind each row. The loneliness row rests on the fewest studies, so its line is among the longest.

Effects are stronger in samples with average age under 65; the risk falls hardest on lonely younger adults. The 1.00 line marks no excess mortality.Holt-Lunstad et al. 2015

Movement II — Geography

Where the loneliness lives.

Loneliness comes in very different doses depending on where you live: about 10% of people in Europe, 24% in Africa, and nearly the same spread again between rich countries and poor ones.

In the loneliest region, Africa, about 24% feel lonely right now; in the least lonely, Europe, about 10%. By income the gap is just as wide: 24% in low-income countries against 11% in high-income ones.

By WHO region · loneliness prevalence 2014–2023

African Region

24%

Eastern Mediterranean

21%

South-East Asia

18%

Region of the Americas

14%

Western Pacific

11%

European Region

lowest globally

10%

Source: WHO Commission on Social Connection, 2025.WHO 2025 Africa and Europe sit 14 percentage points apart on the same question.

By country income · same question, sorted by wealth

Low-income countries

24%

High-income countries

11%

People in low-income countries are more than twice as likely to say they are lonely as people in high-income countries, a gap the WHO report calls out directly.WHO 2025

Movement III — Contagion

It moves between people.

Loneliness spreads through social networks the way obesity and smoking do. In the Framingham data it stays measurable up to three people out from wherever it started, so some of what any one person feels began with someone else they know.Cacioppo et al. 2009

ONE DEGREE
52%
if a direct connection is lonely, you are 52% more likely to be too
TWO DEGREES
25%
25% more likely if a friend of a friend is lonely
THREE DEGREES
15%
15% more likely at three degrees out — the effect disappears at four

“Loneliness occurs in clusters, extends up to three degrees of separation, and is disproportionately represented at the periphery of social networks.”

— Cacioppo, Fowler & Christakis (2009)Cacioppo et al. 2009

Movement IV — Trade-off

Thick ties, thin ties.

Modern economies run on people who can pick up and move, for a job or for cheaper rent. Every move trades a few old, deep ties for a wider circle of newer, thinner ones. Eighty years of that shows up in the charts below.

Since the mid-1900s, one-person households have climbed from under 8% to 29%, daily time with friends has fallen from about an hour to roughly half that, and the share who say most people can be trusted has slipped from 46% to 34%. More people, thinner ties.

Thick · few · slowDrag the years forward.Thin · many · fast

194019402025

Three series · same scrubber · annual data

U.S. one-person households

1940–2024 · % of all households

7.7in 1940

1970 · No-fault divorce begins to spread (CA 1970 → nationwide)

Time Americans spend socializing in-person

2003–2024 · minutes/day

60in 2003(scrubber at 1940)

2007 · iPhone launches — joinpoint #1

Americans who say “most people can be trusted”

1972–2024 · % of adults

46in 1972(scrubber at 1940)

1988 · Onset of accelerating decline

Sources: U.S. Census HH-1/HH-4 (households, 1940–2024); BLS ATUS socializing with friends (2003–2024, own axis — ATUS did not exist before 2003); GSS variable TRUST (1972–2024). Intermediate ATUS years are reconstructed from published joinpoint segments; see data.json for the full series.

Four U.S. trend lines · 2005–2024

What moves with it.

Four American trend lines, 2005 to 2024, all starting from the same point so you can compare how far each moved. The thick line is time in person with friends. The other three went the opposite way over the same two decades.

Since 2005, in-person time with friends is down about 40%, while the share of adults without a partner is up 11%, teen sadness is up 39%, and the suicide rate is up 26%. As friend time fell, the others climbed.

Friend time, in personBLS American Time Use Survey, all U.S. adults. Annual mean minutes per day socializing in person with friends outside the household. The clearest continuous measure we have of physical disconnection.Kannan et al. 2023
Adults living unpartneredPew Research Center analysis of U.S. Census ACS data: share of prime-working-age adults (25–54) who are neither married nor living with a romantic partner. Cohabiting couples are counted as partnered — the cleanest available 'in a relationship' metric in published U.S. statistics.Fry et al. 2025
Teen persistent sadnessCDC Youth Risk Behavior Survey, U.S. high school students (grades 9–12). Share who report 'felt sad or hopeless almost every day for two weeks or more in a row, so that they stopped doing some usual activities' in the past year. Biennial since 1991.Centers 2024
Suicide rateCDC NCHS age-adjusted suicide rate per 100,000 U.S. standard population, all ages. Rose roughly 30% from 2002 to 2018 then plateaued. The 2023 Surgeon General's advisory explicitly named social disconnection as a contributor.Curtin et al. 2024

Each series has its own problems: the partner data undercounts cohabiting couples, the teen survey changed methods after 2019, and the pandemic shook all four.

Methods & vintage

All four series were reconciled against their primary published sources on 2026-07-10. Endpoints rendered as hollow dots with a (p) suffix indicate provisional or carry-forward values where the primary source has not yet released a final number — see the per-series notes below.

Friend time, in person· vintage 2026-05-21 · 4 provisional points2003–2020 values are from Kannan & Veazie 2023 (SSM-Population Health), which applied joinpoint regression to BLS ATUS microdata and identified statistically significant slope changes in 2007, 2013, and 2019. 2021–2024 values are smoothed estimates of the post-COVID partial recovery, consistent with BLS ATUS annual summary releases; the underlying microdata for 2024 was published by BLS in June 2025. Excludes household members.
Adults living unpartnered· vintage 2026-07-10 · 1 provisional pointSource-published anchor points from Pew Research Center: 29% in 1990, 30% in 2000 (decennial census), 35% in 2010 (ACS), 38% in 2019 (ACS, peak). The 2023 value (36%) is a population-weighted blend of the age splits Pew published in its January 2025 update (42% for ages 25–39, 29% for 40–54) — Pew did not publish a 25–54 series in that update. The 2005 value is a linear interpolation between Census 2000 and ACS 2010 (no 2005 Pew analysis published). 2024 is a carry-forward of the 2023 ACS reading. The 2020 ACS sample was not released by Census due to COVID-19 data quality issues, so 2020 here is interpolated. Cohabiting partnerships not involving the household head (~2% of unpartnered) are under-counted by ACS — a known overestimate of the unpartnered share, applied consistently across years.
Teen persistent sadness· vintage 2026-05-21 · all points finalWeighted national prevalence from the CDC YRBS Data Summary & Trends Report (2013–2023) and earlier YRBS DSTR tables for 2005–2011. The 2021 wave was administered Sept–Dec 2021 in still-disrupted school contexts; the 2023 wave used a mixed mode design. Both methodology shifts likely affect comparability across the 2019–2023 segment, though the 2007–2019 trend is monotonic and unaffected by these issues. Independently confirmed by Mojtabai 2025 (BAPC 3.0% [95% CI 2.6–3.4%], n=119,654).
Suicide rate· vintage 2026-07-10 · 1 provisional point2009–2023 values match the CDC NCHS Data Brief 541 (December 2024) final NVSS mortality file exactly. 2005–2008 values are from earlier CDC NVSR Final Data reports. The 2024 value (13.7) is provisional from KFF’s analysis of CDC WONDER 2024; CDC NCHS has not yet released a final 2024 data brief. Rates are age-adjusted to the 2000 U.S. standard population using ICD-10 codes U03, X60–X84, and Y87.0.

Every value above (plus the underlying citations, source URLs, and indexing method) is exposed as machine-readable JSON at /essays/the-social-body/data.json for replication.

Movement V — The close ones

The friendship recession.

The close ties gave way too, and men took the worst of it. In 1990 most American men said they had six or more close friends. Only about a quarter say so now, and the share with none at all went from almost nobody to one in seven.

In 1990, 55% of men had six or more close friends; by 2021 only 27% did, and the share of men with no close friend at all rose from 3% to 15%. Close friendship thinned for everyone, and men lost the most.

1990 → 2021

Across all U.S. adults, the share reporting no close friends went from 3% to 12% over the same window. Source: Survey Center on American Life, “The State of American Friendship” (3% → 12% for all adults with no close friends).Cox 2021

Two anchor years, three decades apart. Each line is one group’s share; the steep one falls while the shallow ones climb. The surveys ask about close friends, the kind you'd call from a hospital parking lot or at midnight when the car dies. That is the tie these lines are tracking.

Movement VI — Generational inversion

The curve flipped.

For decades, well-being in age was a U: high in youth, dip in middle age, rebound in old age. Blanchflower, Bryson & Xu (2025) showed that across 44 countries the U has collapsed: distress now starts high in youth and eases with age. Loneliness is now highest in the young and lowest late in life.

Around 2010, loneliness was lowest among young adults, near 12%, and peaked in middle age; by 2024 it’s highest among teenagers, around 21%, and lowest in old age. The loneliest group flipped from old to young.

c. 2010 · classic U-shape2024 · declining with age

Sources: Blanchflower, Bryson & Xu (2025), PLOS ONE — disappearance of the unhappiness hump across 44 countries.Blanchflower et al. 2025 WHO age bands (2025).WHO 2025 Cigna generational index (2025).Cigna 2025

The tempting story is that phones did this, starting around 2010. But the pullback shows up earlier in the record. Across seven national surveys of American teenagers, the share of high-school seniors who had ever been on a date fell from 87% in 1980 to 63% by 2016, and the slope was bending down years before the first smartphone shipped.Twenge et al. 2019 Phones sped up something that was already moving.

Something about growing up changed fast enough to flip that curve in under fifteen years.

Movement VII — The Mirror

Three questions.

Each scores 1 to 3. Most people finish in under a minute, and your answers never leave this browser.

The three-question scale runs from 3 to 9; in the big aging studies about a third score the lowest, not lonely, while roughly 28% land at 6 or higher, the ‘lonely’ range. Answer and you’ll see where you fall among them.

Q.1 of 3 · UCLA-3

How often do you feel you lack companionship?

Q.2 of 3 · UCLA-3

How often do you feel left out?

Q.3 of 3 · UCLA-3

How often do you feel isolated from others?

Movement VIII — The Sensor

The feeling is accurate.

And that changes where the fix belongs.

Maybe the best word for it is sensor. Hunger is the body asking for food, and pain is the body asking you to stop doing whatever you're doing. Loneliness looks like the same kind of signal, except the thing it asks for is other people. If you only quiet the signal, whatever tripped it is still there.

Loneliness evolved to push you back toward the group, and for most of human history the group was easy to reach: you walked over to the fire. The signal still goes off today. But now the fire is across six lanes of traffic, or it costs $7 at a café where you rent your seat one coffee at a time. Sometimes the walk back just isn't there anymore.

A signal that never shuts off starts to wear on the body, the way constant hunger would: more inflammation, a weaker immune response, and over the years a higher chance of dying early. Holt-Lunstad's meta-analyses put the mortality risk of weak social ties in the same range as smoking.Holt-Lunstad et al. 2010 Holt-Lunstad et al. 2015 The 2023 U.S. Surgeon General's advisory turned that into the comparison most people have heard: about fifteen cigarettes a day.Surgeon General 2023

The usual response is advice. Put down your phone. Join a club. Some of it helps. But nobody tells a person living in a food desert to just eat better. Plenty of places are short on chances to meet people in the same way.

Some of this is recent, and some of it was built on purpose. Dating moved onto apps that work like real-estate listings, and past a certain point an endless scroll of faces stops helping: people begin filtering for flaws, and everyone being swiped through, themselves included, starts to feel replaceable.Schwartz 2004 Young men and women grow up in separate feeds now, with less of the ordinary, low-stakes contact that attraction used to grow out of.Evans 2025 Derek Thompson calls the era the anti-social century: Americans spend more time alone than at any point on record, and the retreat was underway well before the pandemic.Thompson 2025

“What prepares men for totalitarian domination in the non-totalitarian world is the fact that loneliness, once a borderline experience usually suffered in certain marginal social conditions like old age, has become an everyday experience of the ever-growing masses of our century.”

— Hannah Arendt, The Origins of Totalitarianism, 1951.Arendt 1951

Arendt published that sentence in 1951, six years after the war ended. In 2009, Cacioppo & Hawkley summarized two decades of lab work showing that lonely brains lean toward threat: they spot angry faces faster, they read ambiguous expressions as hostile, and social slights stick with them longer than compliments do.Cacioppo et al. 2009 Set that next to Arendt's sentence and her worry gets specific: people primed to see threat are easier to win over with an enemy and a promise of belonging. Oddly, the two fields barely cite each other.

Which may be why sensor is the right word for more than the feeling. We built a way of life that asks people to stay light on their feet and easy to replace, then handed it to a species that runs on attachment. Feeling lonely inside that arrangement is a reasonable response to it. And when the same alarm keeps going off in house after house, at some point you stop checking alarms and start checking the wiring.

Four experiments

What seems to help.

All four are small enough to try this week, and each one links to the research behind it.

Experiment 01
See one friend this week.
Pick one person and meet them, or at least call. Texting doesn't count for this one. The mortality studies keep finding that feeling isolated matters more than how many people you know, so one reliable friendship goes a long way.Holt-Lunstad et al. 2015
Experiment 02
Find a third place.
Klinenberg calls libraries, parks, public pools, and barbershops 'social infrastructure'. Pick one you can walk to and keep showing up.Klinenberg 2018
Experiment 03
Name the transition.
Loneliness spikes when life changes shape: a new city, a divorce, a death, retirement. If you're in one of those windows, say so to somebody. In the WHO's data these are the loneliest stretches, and also the ones where help lands best.WHO 2025
Experiment 04
Join something local.
In person, not online. The ties that cross class, age, and religion are the ones Putnam found keep a democracy healthy, and they happen to be the kind you can't make from a couch.Putnam 2000

Roots take time, and they take other people.

A note from the author

Why I made this.

I work in project management. Over a decade, mostly creative operations. In that line of work, when something ships late or breaks, the cause usually sits in the conditions around whoever gets blamed: the brief nobody clarified, or the meeting that should have happened in week two. After a while you stop looking for someone to blame and start looking at the setup.

I built this because I wanted to understand something that touches all of us, directly or sideways. I'd just come back from a month in Europe, somewhere that arranges daily life differently. How people eat, where they sit, what time things close, who you end up talking to without planning it. That made me notice how much rides on how connected we feel: what we call happiness, how we treat each other and the planet, even our politics. I'm still working that out, which is part of why I wanted to read the research itself instead of the takes about it.

Loneliness gets written about in two ways that both bother me: confident claims with no numbers, and numbers presented like the people inside them don't exist. I wanted to see the source material myself.

So I read it. The WHO published its first global synthesis recently. Holt-Lunstad's meta-analysis has been sitting there since 2015 with mortality data on three million people, and back in 2009 Cacioppo mapped how loneliness spreads through a social network using Framingham data. The generational flip is newer; Blanchflower documented it a few months ago. Young adults are now lonelier than the elderly in much of the rich world, which reverses what we thought we knew. None of this is hidden. The names are below.

What I tried to do was put it all on one desk. Arendt's 1951 paragraph next to the 2009 amygdala data she never got to see. Three long-running U.S. trends on one slider you can drag through the decades. Your own UCLA-3score plotted against the HRS population, scored in a browser that never sends your answers anywhere. The pieces were all out there. As far as I can tell, nobody had put them side by side.

The deeper question for me is why systems keep producing outcomes nobody inside them chose. Loneliness turned out to be a clean example. We keep calling it personal, and meanwhile there's a decade of structural data sitting underneath.

Here's what I haven't figured out: is the collapse in social trust in the GSS series causing young-adult loneliness, or is loneliness causing the trust collapse? The data is consistent with both.

I'm not an academic. I read carefully and I want to keep learning, and projects like this one are how I do it: trying to assemble something and finding out where my understanding breaks.

If any of this is useful, take it. If a number is wrong, tell me. Methods drawer is below.

Causal methods

How to read these numbers.

If you skipped straight to this section, you probably work with data and want to know whether to trust the piece. This is the model underneath it.

The assumed causal model.

Loneliness is the exposure. Mortality is the outcome. Age, socioeconomic status, and baseline health are confounders — they cause both. Inflammation, sleep quality, and health behaviors are mediators — loneliness operates through them. Frailty creates a back-door (reverse) path that the literature has not fully resolved. Estimates from observational studies that adjust for confounders but not mediators identify a total causal effect; estimates that adjust for mediators (as some studies do) identify only the residual direct effect and will be biased toward null.

Loneliness sits on the left as the suspected cause and early death on the right as the result; age, income, and health can push on both, sleep and inflammation carry the effect through the middle, and a dashed loop marks illness feeding back into loneliness. It’s the whole chain of what might cause what.

Hover (or tab through) any node to read its role in the model — Enter or Space pins it.

EXPOSUREOUTCOMECONFOUNDERMEDIATORSELECTION / REVERSE

Dashed arrow: reverse causality (declining health reduces contact, which raises measured loneliness).

Conventions follow Greenland, Pearl & Robins (1999).Greenland et al. 1999 Adjusting for confounders (top row) is required; adjusting for mediators (middle) attenuates the total causal effect and is generally undesirable when the estimand is “does loneliness cause death?” The dashed back-door arrow encodes the reverse-causality problem that randomized trials — ethically impossible here — would close.

E-value sensitivity analysis.

VanderWeele & Ding (2017): the E-value is the minimum strength of association (on the risk-ratio scale) that an unmeasured confounder would need to have with both exposure and outcome to fully explain away an observed effect. For the headline mortality OR point estimates, E-values sit between 1.83 and 1.97; the loneliness lower bound (OR 1.04) yields E ≈ 1.24 — more fragile. Obesity sits near E ≈ 2.37; smoking near E ≈ 3.41. A confounder strong enough to nullify the loneliness–mortality association would need to be one we should already know about.

To fake the link between loneliness and dying earlier, a hidden cause would need an E-value near 1.9, an association about as strong as a well-known health risk. Obesity sits near 2.4 and smoking near 3.4. The bar is high, though the weakest edge of the estimate, about 1.24, is more fragile.

Living alone OR 1.32 · E 1.97 (lower bound 1.54)
Robust — requires a confounder roughly as strong as obesity to explain away.
Social isolation OR 1.29 · E 1.9 (lower bound 1.31)
Moderately robust — confounder weaker than smoking but stronger than sedentary lifestyle.
Loneliness OR 1.26 · E 1.83 (lower bound 1.24)
Sensitive — a moderate unmeasured confounder could plausibly nullify the effect.
Overall (pooled) OR 1.30 · E 1.92 (lower bound 1.59)
Robust — requires a confounder roughly as strong as obesity to explain away.

Method: E(RR) = RR + √(RR · (RR − 1)).VanderWeele et al. 2017 Solid markers = point estimate; open markers = lower 95% CI bound. Reference lines = the published all-cause-mortality risk ratios of smoking, obesity, and sedentary lifestyle. A confounder strong enough to nullify any of these effects would itself be a major public-health story we should already have heard of.

Sample sizes — the evidence stack.

Every figure in the essay is anchored to a dataset of known size. The mortality meta-analysis pools 3.4 million participants. The HRS distribution rests on roughly 20,000 adults per wave. The contagion finding draws on 5,124 Framingham subjects, still the largest study of its kind.

The mortality finding pools about 3.4 million people; the contagion finding rests on 5,124. The surveys in between run from a few thousand per wave to tens of thousands.

Holt-Lunstad 2015 meta-analysisHolt-Lunstad et al. 2015

Across 70 prospective cohorts; provides the OR=1.26–1.32 mortality estimates.

N = 3.4M

participants pooled

WHO Commission 2025 synthesisWHO 2025

Pools 23 datasets including Meta-Gallup, Eurobarometer, and national surveys.

N = 153

countries & territories

U.S. Census CPS (annual)Census 2024

Source for the one-person-household time series (HH-1, HH-4).

N = 60k

households / year

BLS ATUS (annual)Kannan et al. 2023

Single-day time diaries; 2024 release sample size 7,700.

N = 9.7k

respondents / year

General Social SurveyGSS 2024

72,390 person-waves cumulative 1972–2022; source for the TRUST series.

N = 3.5k

respondents per wave

U.S. Health & Retirement StudyCrowe et al. 2021

Source for the UCLA-3 distribution; biennial since 1992, ongoing.

N = 20k

adults 50+ per wave

Cacioppo, Fowler & Christakis 2009Cacioppo et al. 2009

Source for the three-degrees-of-separation contagion finding.

N = 5.1k

Framingham Heart Study participants

10²10³10⁴10⁵10⁶10⁷

Full reproducibility manifest.

Every series, citation, forest-plot row, DAG node, and E-value is also exposed as machine-readable JSON at /essays/the-social-body/data.json. Each citation carries its canonical source URL; the co-movement series additionally carry a vintage (last-verified date), a direct source link, and per-series methodology notes. The endpoint is versioned and CORS-open; pull it into a notebook and re-run any of the analyses.

GET /essays/the-social-body/data.json ↗

$ curl -sS "/essays/the-social-body/data.json" \
    | jq '.forestPlot.rows[] | {label, or, ciLow, ciHigh, n}'Host resolves on load — relative path works in-browser.

What the data won’t tell you

Three places this page leans.

Places where I’m relying on inference, or on choices someone else made when the data was collected. Read them before you take a number from here and reuse it.

Caveat 01
“Loneliness spreads up to three degrees of separation.”
The original network-contagion methodology cannot fully separate genuine social influence from people forming friendships with already-similar people (homophily) or being exposed to the same neighborhood / shock (shared environment). Cohen-Cole & Fletcher (2008) showed the same model produces statistically significant “network effects” for acne, height, and headaches — phenomena that cannot plausibly be contagious. VanderWeele’s (2011) sensitivity analysis concluded the loneliness and happiness findings are less robust to these alternative explanations than the obesity and smoking findings from the same lab.Cohen-Cole et al. 2008 VanderWeele 2011
Caveat 02
“1 in 6 people are lonely globally.”
WHO 2025 is the first systematic global synthesis — which means there is no comparable earlier baseline. We cannot say from this number whether loneliness has risen, fallen, or stayed flat. WHO itself states: “Previous data are too limited to determine whether the rates of social isolation and loneliness have risen or fallen.” Country-level rates vary by a factor of more than two (Europe 10%, Africa 24%), driven partly by real environmental differences and partly by survey-instrument heterogeneity.WHO 2025
Caveat 03
“Lonely people have 26% higher mortality.”
The 1.26 odds ratio is the fully-adjusted random-effects pooled estimate, 95% CI [1.04, 1.53]. The lower bound just barely clears 1.00 — the effect is statistically real but the CI is wide, meaning the true effect could be anywhere from a 4% increase to a 53% increase. The unadjusted overall meta-analytic estimate is 1.53, considerably larger; the adjustment matters. Most importantly, this is a population-level association, not a destiny: many lonely individuals live long lives, and the pathway is mediated by behavior (sleep, diet, healthcare-seeking) and physiology (inflammation, immune dysregulation) that interventions can address.Holt-Lunstad et al. 2015

Colophon · primary sources

Every figure in this piece traces back here.

Arendt, H. (1951). The Origins of Totalitarianism.
Harcourt, Brace & Co.. Open source ↗
Blanchflower, D. G., Bryson, A. & Xu, X. (2025). The declining mental health of the young and the global disappearance of the unhappiness hump-shape in age.
PLOS ONE, 20. Open source ↗
Bureau of Labor Statistics (2024). American Time Use Survey — Time Spent in Social Activities.
Synthesized with Sarah Cowan & Daniel A. Cox analyses (Georgetown / Survey Center on American Life). Open source ↗
Bureau of Labor Statistics (2024). American Time Use Survey — 2024 Results (Table 1; socializing & communicating).
BLS news release USDL-25-1162. Open source ↗
Cacioppo, J. T. & Hawkley, L. C. (2009). Perceived Social Isolation and Cognition.
Trends in Cognitive Sciences, 13(10), 447–454. Open source ↗
Cacioppo, J. T., Fowler, J. H. & Christakis, N. A. (2009). Alone in the Crowd: The Structure and Spread of Loneliness in a Large Social Network.
Journal of Personality and Social Psychology, 97(6), 977–991. Open source ↗
Centers for Disease Control and Prevention (2024). Youth Risk Behavior Survey — High school students reporting persistent feelings of sadness or hopelessness in the past year, 2005–2023.
CDC YRBS Data Summary & Trends Report (biennial). Open source ↗
Cohen-Cole, E. & Fletcher, J. M. (2008). Detecting implausible social network effects in acne, height, and headaches: longitudinal analysis.
BMJ, 337, a2533. Open source ↗
Cox, D. A. (Survey Center on American Life) (2021). The State of American Friendship: Change, Challenges, and Loss.
Survey Center on American Life — May 2021 American Perspectives Survey. Open source ↗
Crowe, C. L., Domingue, B. W., Graf, G. H., et al. (2021). Associations of Loneliness and Social Isolation with Health Span and Life Span in the U.S. Health and Retirement Study.
Journals of Gerontology Series A, 76(11), 1997–2006. Open source ↗
Curtin, S. C., Garnett, M. F. & Hedegaard, H. (2024). Changes in Suicide Rates in the United States, 2002–2023 — Age-Adjusted Death Rates per 100,000.
CDC NCHS Data Briefs No. 509 (2024) and No. 541 (2025). Open source ↗
Evans, A. (Stanford University / King's College London) (2025). The Great Gender Divergence — young men and women moving into separate social worlds.
The Great Gender Divergence project; cf. Burn-Murdoch, Financial Times (26 Jan 2024). Open source ↗
Fry, R. & Parker, K. (2025). Rising Share of U.S. Adults Are Living Without a Spouse or Partner (2021) and Share of U.S. Adults Living Without a Romantic Partner Has Ticked Down in Recent Years (2025).
Pew Research Center Social & Demographic Trends — analyses of U.S. Census Bureau decennial census and American Community Survey 1-year estimates. Open source ↗
Greenland, S., Pearl, J. & Robins, J. M. (1999). Causal Diagrams for Epidemiologic Research.
Epidemiology, 10(1), 37–48. Open source ↗
Holt-Lunstad, J., Smith, T. B. & Layton, J. B. (2010). Social Relationships and Mortality Risk: A Meta-analytic Review.
PLOS Medicine, 7(7), e1000316. Open source ↗
Holt-Lunstad, J., Smith, T. B., Baker, M., Harris, T. & Stephenson, D. (2015). Loneliness and Social Isolation as Risk Factors for Mortality: A Meta-Analytic Review.
Perspectives on Psychological Science, 10(2), 227–237. Open source ↗
Hughes, M. E., Waite, L. J., Hawkley, L. C. & Cacioppo, J. T. (2004). A Short Scale for Measuring Loneliness in Large Surveys.
Research on Aging, 26(6), 655–672 — UCLA 3-Item. Open source ↗
Hunt, M. G., Marx, R., Lipson, C. & Young, J. (2018). No More FOMO: Limiting Social Media Decreases Loneliness and Depression.
Journal of Social and Clinical Psychology, 37(10), 751–768. Open source ↗
Kannan, V. D. & Veazie, P. J. (2023). US trends in social isolation, social engagement, and companionship — nationally and by age, sex, race/ethnicity, family income, and work hours, 2003–2020.
SSM — Population Health, 21 (PMC9811250). Open source ↗
Klinenberg, E. (2018). Palaces for the People: How Social Infrastructure Can Help Fight Inequality, Polarization, and the Decline of Civic Life.
Crown. Open source ↗
Lyons, R. (2011). The Spread of Evidence-Poor Medicine via Flawed Social-Network Analysis.
Statistics, Politics & Policy, 2(1), Article 2. Open source ↗
National Opinion Research Center (NORC) (2024). 1972–2024 General Social Survey Cumulative File (variable: TRUST).
GSS Data Explorer / SDA Berkeley archive. Open source ↗
Putnam, R. D. (2000). Bowling Alone: The Collapse and Revival of American Community.
Simon & Schuster. Open source ↗
Schwartz, B. (2004). The Paradox of Choice: Why More Is Less.
Ecco / HarperCollins — choice overload and satisfaction. Open source ↗
Silver, L., Keeter, S., Kramer, S., Lippert, J., et al. (Pew Research Center) (2025). Americans' Trust in One Another (and Why It's Declining): The share answering 'most people can be trusted' fell from 46% in 1972 to 34% in 2018.
Pew Research Center, 8 May 2025. Open source ↗
Steptoe, A., Shankar, A., Demakakos, P. & Wardle, J. (2013). Social isolation, loneliness, and all-cause mortality in older men and women.
PNAS, 110(15), 5797–5801. Open source ↗
The Cigna Group / Morning Consult (2025). The 2025 Loneliness in America Report — generational breakdown showing the youngest adults most lonely and Boomers least.
Cigna Group, Loneliness in America Report (2025 release). Open source ↗
Thompson, D. (2025). The Anti-Social Century.
The Atlantic — February 2025 cover story. Open source ↗
Twenge, J. M. & Park, H. (2019). The Decline in Adult Activities Among U.S. Adolescents, 1976–2016.
Child Development, 90(2), 638–654 — Monitoring the Future, N ≈ 8.4 million. Open source ↗
U.S. Census Bureau (2024). America's Families and Living Arrangements: 2024.
U.S. Census Bureau Current Population Survey. Open source ↗
U.S. Census Bureau (2024). Historical Households Tables — HH-1 (Households by Type) & HH-4 (Households by Size), 1940–2024.
U.S. Census Bureau, Current Population Survey. Open source ↗
U.S. Surgeon General (Vivek Murthy) (2023). Our Epidemic of Loneliness and Isolation.
U.S. Department of Health & Human Services advisory. Open source ↗
VanderWeele, T. J. (2011). Sensitivity Analysis for Contagion Effects in Social Networks.
Sociological Methods & Research, 40(2), 240–255. Open source ↗
VanderWeele, T. J. & Ding, P. (2017). Sensitivity Analysis in Observational Research: Introducing the E-Value.
Annals of Internal Medicine, 167(4), 268–274. Open source ↗
World Health Organization (2025). From Loneliness to Social Connection: Charting a Path to Healthier Societies.
WHO Commission on Social Connection — flagship report. Open source ↗
World Health Organization (2025). Social connection linked to improved health and reduced risk of early death.
WHO press release, 30 June 2025. Open source ↗

METHODOLOGY · All citation URLs re-verified 21 May 2026. Series sources: Census HH-1 (decennial 1940–2020) merged with CPS HH-4 (annual 2010–2024). BLS ATUS “social engagement with friends” anchored to Kannan & Veazie 2023 published values (60 min/day in 2003, 34 in 2019, 20 in 2020) with intermediate years reconstructed from their joinpoint segments (APC −4.38, +1.58 n.s., −6.89); 2021–2024 partial-recovery estimates from BLS news release. GSS variable TRUST shows actual wave years (waves occur ~biennially after 1994). Forest plot reproduces Holt-Lunstad 2015 Table 3 fully-adjusted random-effects pooled odds ratios with 95% CIs. UCLA-3 placement uses the published HRS distribution shape (Steptoe 2013, Crowe 2021); answers stay in the browser. Generational inversion combines Blanchflower, Bryson & Xu 2025 with WHO 2025 age bands. WHO regional dot lattices round prevalence to integers; the underlying point estimates and 95% uncertainty intervals (Africa 24.3% [20.4–29.0], EMR 21.0% [15.9–27.1], SEAR 18.3% [11.2–29.3], AMR 13.6% [10.2–18.6], WPR 11.0% [6.1–21.7], Europe 10.1% [8.2–12.5]) live in the WHO PDF.

← Back to carlosvasquez.me