Does Community Substitute for Connection?

Since Bowling Alone, a half-century of social-capital writing has carried a generous implication: that the texture of the place you live — its churches and leagues, its PTAs and volunteer fire companies, its habit of returning lost wallets — is a public good. Robert Putnam measured league memberships; John Helliwell measured the well-being returns; Eric Klinenberg argued that libraries and other "social infrastructure" could stand in for the private ties Americans were losing. The premise underneath all of it is that community fabric, like clean air, is non-excludable. Everyone breathes it.

But whom, exactly, does it lift? Consider the person social-capital policy is implicitly for: someone who lives alone, has no spouse or partner, and belongs to no congregation. For that person, two readings of the literature make precisely opposite predictions — and the literature, read carefully, does not know which is true.

Two hypotheses, opposite predictions

Compensation says community substitutes for connection. If your county is thick with neighbors who help each other, with low crime and functioning institutions, then the absence of a partner or a congregation matters less; the public fabric catches what private life drops. Klinenberg's palaces for the people, Cacioppo's loneliness research read optimistically — the prediction is that the well-being gap between isolated and connected people should narrow as county social capital rises.

Accumulation says the opposite: social capital is captured by the people positioned to use it. The clubs, congregations, and carpools that constitute a high-capital county are accessed through exactly the ties the isolated lack. On this reading, civic fabric is less like clean air and more like a standing invitation that requires someone to hand it to you. The prediction: the gap should widen — the connected pull further away in the very places celebrated for community.

These are not strawmen; both have respectable pedigrees, and both cannot be true. What has been missing is a dataset large enough to measure a within-county contrast — isolated versus connected people — across the full range of American community life. The Gallup US Daily poll, 3,530,270 interviews conducted from 2008 through 2017 and joinable to the Joint Economic Committee's county social-capital index, may be the best descriptive testbed anyone has. This article runs the test and reports what comes out, in plain units, with the caveats attached where they belong.

Measuring the lonely, measuring the fabric

Isolation here is an additive score built from three blunt markers, each verified against the file's codebook: living in a one-person household, being unpartnered (never married, separated, divorced, or widowed), and rarely or never attending religious services. Among the 989,355 interviews with all three markers, a Cantril-ladder response, and a matched county, 32.6 percent score zero — we will call them connected — 30.8 percent carry one marker, 23.3 percent two, and 13.3 percent all three. The score behaves exactly as the loneliness literature says it should: average life evaluation falls in a clean staircase, from 7.341 rungs (on the 0–10 ladder) at score zero to 7.037, 6.862, and 6.562 at scores one, two, and three. Carrying all three markers is associated with a deficit of 0.78 rungs — roughly the gap between the United States and Poland in world league tables.

The fabric side comes from the JEC Social Capital Project, which scored every US county on family unity, community health, institutional health, and collective efficacy, rolled into one index. The 989,355 interviews span 2,984 counties; the swarm below shows the 1,389 counties contributing at least 300 interviews each — 2.34 million interviews — arranged along that index. The geography is familiar to anyone who has read the JEC report: dense urban cores and the rural South sit low; the upper Midwest sits high. Note how compressed the vertical axis is — county average life evaluation varies far less than the index does.

Before the main event, one number for scale. Moving from the bottom decile of county social capital to the top is associated with a rise in average life evaluation — for everyone pooled — from 7.030 to 7.122. Less than a tenth of a rung across the entire civic range of the country. Whatever social capital does for well-being, it is not, in this unweighted cross-section, a large lever. The question is whether even that small lever tilts toward the isolated or away from them.

The verdict: a ribbon that refuses to move

Here is the test in its purest form. Sort the 989,355 interviews into ten equal-size bins by their county's social-capital index. Within each decile, take the mean ladder for the connected (score 0) and for the isolated (score 2 or higher). The vertical distance between the two lines — the gold ribbon — is the well-being cost of isolation. Compensation predicts the ribbon tapers to the right. Accumulation predicts it flares.

The ribbon barely moves. Under the central definition (two or more markers), the gap is 0.527 rungs in the least-civic decile of America and 0.575 rungs in the most civic — and it bulges to 0.617 in decile 7. Fit a line through the ten gaps and it drifts upward by 0.0071 rungs per decile: the wrong direction for compensation, but glacially. The isolated themselves gain almost nothing from civic context — 6.802 in decile 1, 6.822 in decile 10 — while the connected gain slightly more, 7.329 to 7.397.

Now move the slider. Define isolation loosely (any one marker) and the gap is 0.432 rungs at one end, 0.431 at the other — flat to the third decimal. Define it strictly (all three markers) and it runs 0.772 to 0.750, peaking at 0.859 in decile 4 — if anything a hair narrower at the top, but with no trend a fair reader would bet on. Across three definitions and ten deciles, the ribbon's message is the same: the penalty for isolation is portable. It arrives at close to full strength in Door County, Wisconsin, and in the densest urban cores alike.

Which fabric? The sub-indices disagree

"Social capital" is a composite, and the composite hides a small but genuine disagreement among its parts. The JEC index averages four sub-indices; the county file also carries a direct measure of emotional support (the share of residents who say they rarely get it, axis flipped below so that right always means more support). Re-running the decile test on each:

Family unity, community health, and collective efficacy all show the gentle flare — gaps that open by 0.04 to 0.06 rungs from bottom decile to top. Emotional support shows the largest drift in the raw deciles (0.501 to 0.612), though as the regressions below show, that one dissolves entirely once you account for who lives where. The interesting deviant is institutional health — confidence-in-institutions, voting, census response — where the gap is flat to slightly narrowing (0.591 to 0.579). It is the one strand of social capital that is not accessed through personal ties: you do not need a spouse to benefit from a functioning county government. That is exactly where the accumulation logic predicts compensation could survive, and it is the only place a trace of it appears.

Moods, and a placebo that behaves

Life evaluation is one construct; daily experience is another, and the file keeps them separate. The same decile test on yesterday's moods — the share reporting sadness, worry, and, as a deliberate placebo, physical pain — shows the same constancy. Isolated Americans report sadness at rates about 8 points higher than the connected in the least-civic decile, and about 8 points higher in the most civic. Worry: a steady 4-to-5-point excess. Pain — which no theory of social capital should touch — shows a 7-to-8-point excess that likewise goes nowhere.

Formalizing it — with the attenuation in plain sight

Deciles are honest but coarse. The formal version interacts the isolation score with the (standardized) county index in a least-squares regression, with standard errors clustered on the 2,984 counties. The interaction coefficient answers the question directly: for each additional isolation marker, how much does one standard deviation of county social capital change the penalty? Negative means the penalty deepens — accumulation. Positive means it shrinks — compensation.

Run raw, with nothing else in the model, the interaction is −0.0092 rungs (clustered SE 0.0031). Add the controls — age, education, income, race, county median income, and the rural–urban continuum — on the identical 767,036 complete-case interviews, and it attenuates to −0.0053 (SE 0.0027, p = 0.049). The attenuation is itself informative: roughly two-fifths of the raw interaction was composition, not context — isolated people in civic counties are demographically different from isolated people elsewhere — and there is no reason to believe the controls caught all of it. We show both estimates side by side rather than burying the raw one.

What should you make of −0.005? Be honest about the magnitude. A person carrying all three markers, moving across one standard deviation of county social capital, sees the gap widen by three times that — about 0.016 rungs on the controlled estimate. The full bottom-decile-to-top-decile journey spans about 3.1 standard deviations of the index, so the gap widens by roughly 0.05 rungs across the entire civic range of the United States — against a baseline gap of half a rung. The interaction is statistically detectable because three-quarters of a million observations can detect almost anything. Substantively, it is a rounding error on the isolation penalty itself, which stands at −0.259 rungs per marker raw and −0.172 with controls.

The secondary outcomes sharpen rather than soften this. For Gallup's "thriving" classification the interaction is −0.19 percentage points per marker per SD with controls (p = 0.001) — accumulation again, again small. For sadness and worry the interactions are statistical zeros, raw or controlled. And the placebo behaves exactly as a placebo should: physical pain shows a raw "interaction" of +0.22 percentage points (p = 0.003) — the pain gap appears to widen with civic fabric, an "effect" no theory of social capital predicts — and it collapses to +0.06 and insignificance once the controls absorb the demographic sorting of who is isolated where. A reader who wants a one-line summary of why the raw estimates cannot be read causally can point at the pain panel: composition alone manufactures a raw interaction there nearly as large as the one we found for the ladder.

Across the sub-indices, the controlled interactions diverge just as the deciles hinted: collective efficacy −0.0120 (p < 0.001), family unity −0.0056 (p = 0.031), community health −0.0040 (p = 0.11), emotional support a flat zero after controls — and institutional health +0.0080 (p = 0.004), the lone compensation-signed estimate in the table. The strands of social capital that run through private ties tilt toward accumulation; the strand that runs through public institutions tilts, faintly, the other way.

What this can and cannot say

Composition versus context is the whole battle, and a cross-section cannot win it — only referee it. People choose where to live; the unmarried non-attender in a high-capital Minnesota county got there, or stayed there, by some process that also touches her well-being. The controls attenuate that selection; they cannot eliminate it. Every "effect" above is an association. The county index is measured once, around 2018, and joined to interviews from 2008–2017 as time-invariant context. The Gallup file carries no survey weights, so every figure is an unweighted sample estimate: contrasts and slopes are the reliable objects here, not population levels. And the isolation score is behavioral, not felt — living alone is not loneliness, and religious attendance is one club among many (though it is the one civic membership the file measures directly, and the one Putnam called the single most important).

What the data can say, they say loudly, with the precision of 989,355 interviews. There is no decile of American community, no sub-index of civic fabric, and no definition of isolation under which the well-being gap between connected and isolated people meaningfully closes. The compensation hypothesis — the generous reading, the policy hope that building community fabric is a remedy for private disconnection — finds no support at county scale. The accumulation hypothesis wins the sign test: the interaction is negative, raw and controlled, for life evaluation and for thriving. But it wins on points, not by knockout; its magnitude is a few hundredths of a rung where its strongest advocates might have predicted tenths.

The fairest verdict is the one neither camp was rooting for: the isolation penalty is approximately a constant of American life. It travels intact from the civic deserts to the civic gardens, shaved or fattened only at the margins. Community, at least as counties embody it, neither substitutes for connection nor hoards its benefits at meaningful scale — it mostly operates on a different channel altogether. For the policy conversation that runs from Putnam to Klinenberg, that is a genuinely uncomfortable finding, and the discomfort is the point: if we want to lift the well-being of people without partners, households, or congregations, the lever marked "build more community" is — on the best descriptive evidence available — far too small for the job. The ties themselves are the active ingredient. There is, as yet, no civic substitute.