Since Robert Putnam's Bowling Alone, the strongest claim of the social-capital literature has been about places: counties thick with congregations, clubs, volunteers and trusted neighbors are supposed to be better to live in. But better for whom? Take a person with no spouse or partner, an empty apartment, and no congregation. One hypothesis — call it compensation, the spirit of Eric Klinenberg's "social infrastructure" — says community fabric substitutes for missing personal ties, so the well-being gap between the isolated and the connected should narrow where social capital is high. The rival — accumulation — notes that the connected are the ones actually inside the clubs and churches, so they capture the benefits and the gap should widen. The two stories make opposite predictions about one measurable object: the width of a ribbon.
We measured that ribbon in nearly a million Gallup interviews joined to the Joint Economic Committee's county social-capital index. People with none of the three isolation markers below rate their lives at 7.32 on the 0–10 Cantril ladder in the lowest-capital tenth of counties and 7.39 in the highest. People with all three markers go from 6.56 to 6.64. The gap: 0.76 rungs at one end, 0.75 at the other. Across the entire American gradient of civic fabric — from St. Louis city at one extreme to Milwaukee's church-dense suburbs at the other — the penalty attached to being unattached is, to two decimal places, a constant.
Lives alone
Household of one — 33 percent of the sample.hhsize3 = "1"Unpartnered
Never married, or separated, divorced or widowed; "partnered" includes domestic partners.marital3 ≠ "Married"Rarely attends
Never or seldom attends religious services — the one regular civic institution Gallup asks everyone about.attend3 = "Rarely"Score the three markers 0–3. Connected means none of them (33 percent of the sample); isolated, in the strictest reading, means all three (13 percent). The toggle on the chart relaxes that definition — any one marker, or two of three — and lets you watch whether the ribbon's behavior is an artifact of where we drew the line. It isn't.
The ribbon that refused to taper
Mean Cantril ladder (0–10, evaluative) by decile of county social capital — connected vs. isolated, the gap shaded
A tide that lifts both boats by about an inch
Read the chart the way the two hypotheses ask you to. Compensation predicts the amber band tapers to the right; accumulation predicts it flares. What it does is neither — it runs nearly parallel, wobbling within a tenth of a rung around three-quarters of a rung wide. Both lines drift up slightly in high-capital counties, by less than a tenth of a rung from decile 1 to decile 10. Where you live on America's civic gradient is associated with far less of your life evaluation than whether you are partnered, accompanied and affiliated: the place gradient is about a tenth of a rung; the isolation gap is three-quarters of a rung, everywhere.
The toggle is the robustness check. Define isolated as any one marker and the gap runs 0.42 to 0.43, end to end. Define it as two of three and it runs 0.51 to 0.57 — the only definition under which anything moves, and the direction is widening: the formal interaction is −0.022 rungs per standard deviation of the index (p = 0.0005). That is the accumulation direction, but mind the size: roughly six-hundredths of a rung across the whole national gradient, against a gap of half a rung. A whisper, not a verdict — and it disappears entirely under the strict definition.
The regression formalizes what the picture shows. We fit OLS on the individual interviews — ladder on isolated, the standardized index, and their product, with standard errors clustered by county — first raw, then adding age group, sex, race, education, income bracket, interview year, county urbanicity (RUCC) and county median income. Raw, the all-three isolation penalty is 0.77 rungs; with controls it attenuates to 0.54, because the isolated are also older, poorer and less educated than the connected — composition, openly shown. The interaction, the number this article exists to estimate, is −0.013 raw (p = 0.25) and −0.0002 with controls (p = 0.98). With 328,706 interviews in the controlled model, that is not an underpowered shrug; it is a precisely estimated zero.
The places behind the axis
Counties along the JEC social-capital index — circle area = interviews in the analysis sample; hover for names
The beeswarm is a reminder of what the axis actually ranks. The high end is suburban and exurban upper-Midwest and mountain-West counties — Ozaukee and Waukesha outside Milwaukee (index 2.1 and 1.9), Carver outside Minneapolis (2.1), Douglas between Denver and Colorado Springs (1.7). The low end is urban cores and the rural South: St. Louis city (−3.7), Baltimore city (−3.5), New Orleans (−3.1), Memphis's Shelby County (−3.0). Civic density in America co-travels with marriage, money, whiteness and suburbia — which is exactly why the model controls for urbanicity and county income, and why even the controlled comparison stays descriptive.
The feelings agree with the judgment — and so does the placebo
Life evaluation is a judgment; feelings are a different construct. So we repeated the design on Gallup's experiential "yesterday" items — did you feel sadness, worry, physical pain a lot of the day yesterday — all inside the same 2008–2012 window (these items change survey regime in 2013 and never co-occur with the isolation markers afterward). The isolated report more of everything: sadness 24 percent versus 14 in the lowest-capital decile, with a gap of about 10 points that is still about 10 points in the highest decile. Worry behaves the same with a gap near 7 points.
Physical pain is the falsification panel. Pain is real, but no theory of bowling leagues says community fabric should specifically close the pain gap between the isolated and the connected; if our design were picking up genuine social moderation, the social feelings should move more than pain does. They don't, because nothing moves: with controls, the interaction is −0.002 for sadness, −0.002 for worry, and +0.002 for pain — identical in size, all indistinguishable from zero. The one nominally significant affect interaction anywhere is raw pain (+0.006, p = 0.008), and it dies the moment controls enter — a tidy demonstration that what little movement exists in these ribbons is composition, not community.
Even the sub-indices can't agree
The JEC index is a composite, and the brief version of the compensation story really concerns its civic ingredients. Re-running the controlled model on each sub-index splits the signs: institutional health (civic participation and confidence) leans toward narrowing the gap (+0.024 per SD, p = 0.016); collective efficacy (low violent crime) leans toward widening it (−0.026, p = 0.002); family unity and community health are nulls. Two "significant" interactions with opposite signs, each worth a hundredth or two of a rung per standard deviation, are better read as fragility than as dueling mechanisms — with 328,706 interviews in the model, tiny accidents of composition clear p < 0.05 easily. The honest summary across all four ingredients is the same as the composite: no robust moderation in either direction.
What this does — and does not — show
It does not show that community is worthless to the isolated. These are cross-sections of people, not experiments on them: isolated people who live in Ozaukee County differ from isolated people in St. Louis in ways no regression fully removes, and our controls demonstrably matter (they cut the raw penalty by nearly a third). Composition versus context is the whole battle here, and observational data can referee it only partially. The county index is also a single ~2018-vintage snapshot joined to 2008–2012 interviews through approximate ZIP-to-county mapping, the sample is unweighted, and "rarely attends" folds the contentedly secular in with the disconnected.
What it does show is descriptive, but unusually well-powered: across nearly a million interviews, the association between social isolation and lower well-being is flat in county social capital — every definition, every outcome, both constructs, and the placebo behaves exactly like the treatment outcomes. That is evidence against strong versions of both canonical stories. Helliwell and Putnam showed that community context and personal ties each predict well-being; this test asks whether the first substitutes for the second, and the data answer: they simply add. If loneliness is, as Cacioppo argued, an individual-level risk, the resolution to it that shows up in this file is individual too. A county can be a palace for the people — and the ladder still cares most about who is standing next to you on it.