Whose trust? Social capital across economics and sociology
Economists measure trust with a survey question and regress it against growth. Sociologists meant something else entirely — and the gap between the two changes what each can see.
Sociology had it first
“Networks of civic engagement … foster sturdy norms of generalized reciprocity and encourage the emergence of social trust. Such networks facilitate coordination and communication, amplify reputations, and thus allow dilemmas of collective action to be resolved.”
— Robert Putnam, “Bowling Alone: America’s Declining Social Capital,” Journal of Democracy, 1995, p. 67
Putnam was not reporting a survey. He was reporting a structural fact: Americans were joining fewer bowling leagues, fewer PTAs, fewer Rotary clubs. The thinning of those networks was, to him, a different and prior kind of fact than the number you get when you ask people whether they trust strangers. Hold onto that distinction. It is the whole argument.
Before economics borrowed the phrase, it already had a default way of talking about the soft glue between strangers in a market. That default is institutions: the formal and informal rules that constrain how transactions happen. In Douglass North’s framing, institutions are the rules of the game — property rights, contract enforcement, the courts, and the unwritten conventions that make a handshake mean something. Trust, on this account, is what you get when the rules are good enough that you don’t have to verify everything yourself.
That is the baseline the sociological apparatus is about to be set against. It is not wrong — it is a real and load-bearing piece of economics. But notice what it does to trust: it treats trust as a residual, the thing left over when institutions work, rather than as an object with its own internal structure. Sociology starts from the opposite end.
The apparatus economics borrowed from
By the time economists reached for the term in the late 1990s, sociology had spent three decades building a precise structural machinery around it. Four figures carry most of the weight. Read them in their own words; the point is not that they are evocative but that they are analytically specific in a way the survey measure is not.
“Social capital is the aggregate of the actual or potential resources which are linked to possession of a durable network of more or less institutionalized relationships of mutual acquaintance and recognition.”
— Pierre Bourdieu, “The Forms of Capital,” 1986
Bourdieu’s move is conversion. He puts three currencies on the table — economic, cultural, and social capital — and argues they trade into one another inside field-specific games. A useful introduction, a reference, an inherited position: these convert a network into money or status by specific mechanisms. Social capital here is not a feeling people have about strangers. It is a property of where you sit in a web of relationships, and of what that position lets you draw down. You cannot read it off a one-item survey because it lives in the structure, not the head.
“Social capital is defined by its function. It is not a single entity but a variety of different entities, with two elements in common: they all consist of some aspect of social structures, and they facilitate certain actions of actors … within the structure.”
— James Coleman, “Social Capital in the Creation of Human Capital,” American Journal of Sociology, 1988
Coleman sharpens it into a mechanism. Three forms do the work: obligations and expectations (the favour-bank, where a debt held is a resource), information channels (who tells you what, and when), and norms backed by effective sanctions (which only bite when the network can enforce them). His key claim is about closure: dense, closed networks — where everyone knows everyone — generate enforceable trust precisely because deviation gets punished. Trust here is not an average over a population. It is a structural property that switches on when the network is closed and off when it is open.
“Some regions of Italy … have many active community organizations. … In these communities citizens are engaged by public issues, not by patronage. … By contrast, life in the less civic regions … is structured by hierarchy.”
— Robert Putnam, Making Democracy Work: Civic Traditions in Modern Italy, 1993
Putnam turned the apparatus into a falsifiable empirical claim and won the bet. Tracking the performance of Italy’s twenty regional governments after a 1970 reform handed them identical formal powers, he found their effectiveness diverged sharply — and the divergence was predicted not by wealth or by the new institutions but by the density of horizontal civic association reaching back centuries. Northern regions with old traditions of choral societies and mutual-aid clubs ran better governments than southern regions organized around vertical patronage. A structural feature of the network, measured historically, forecast institutional performance decades later. No aggregate-trust survey could have produced that result, because the result is about structure across time.
“Those to whom we are weakly tied are more likely to move in circles different from our own and will thus have access to information different from that which we receive.”
— Mark Granovetter, “The Strength of Weak Ties,” American Journal of Sociology, 1973
Granovetter, and later Ronald Burt, located the value in the gaps. Most people find jobs not through close friends but through acquaintances — weak ties — because weak ties bridge otherwise-disconnected clusters and carry novel information across them. Burt formalized the flip side in Structural Holes (1992): the broker who spans a gap between two clusters captures value purely from position, not from any personal trait. Information, opportunity, and mobility flow along the contours of the network. A survey item that asks “can most people be trusted?” cannot see any of this, because the answer to that question is the same for the well-connected broker and the isolate.
What the relational apparatus buys you
Put the four together and you have a relational picture: social capital as a property of positions, ties, and the resources moving through them. This is not a fuzzy, pre-scientific stand-in for something economics would later make rigorous. It is its own rigorous apparatus, and it catches a class of facts the aggregate cannot — which relationships in which networks produce which effects, and what happens to trust-as-a-resource when the structure reorganizes. The Italian-regions divergence, the weak-tie job find, the broker who profits from a hole in the network: these are findings, not metaphors. When economics arrives at the word, it inherits a real challenge, not a vague invitation.
So sociology had a precise structural apparatus with thirty years of empirical traction by the mid-1990s. Then economists started using the term. What did they import — and what did they leave on the dock?
Economics imports the term
“Trust and civic cooperation are associated with stronger economic performance. … A ten-percentage-point rise in trust is associated with an increase in growth of four-fifths of a percentage point.”
— Stephen Knack & Philip Keefer, “Does Social Capital Have an Economic Payoff?” Quarterly Journal of Economics, 1997
This is the paper that made trust respectable in a growth regression. Knack and Keefer took the survey item, aggregated it to the country level, put it on the right-hand side, and found a robust, sizeable coefficient. It is a real empirical result. It is also a complete reconception of what social capital is — and that is the move worth watching closely.
Move one: trust becomes an aggregate variable. The World Values Survey asks a single question — “Generally speaking, would you say that most people can be trusted, or that you can’t be too careful in dealing with people?” The share of a country’s respondents who pick “most people” becomes a national trust score. Norway sits near 70 percent; some low-trust societies sit in single digits. That one number, per country, is what economics regresses against growth, investment, and contract enforcement.
Move two: the trust-and-growth lineage. Knack and Keefer (1997) established the correlation. Zak and Knack (2001) added a transaction-cost mechanism: where trust is low, people spend resources verifying and protecting against each other, and that overhead drags on growth. Algan and Cahuc (AER, 2010) closed the obvious objection — that growth might cause trust rather than the reverse — by instrumenting current trust with the inherited trust of the descendants of immigrants to the United States, whose forebears’ home-country trust levels plausibly affect nothing today except through transmitted attitudes. The causal-flavoured estimate survived. The literature is large, replicated, and policy-active.
Move three: trust becomes an individually measurable game parameter. Berg, Dickhaut, and McCabe (1995) invented the trust game. An investor is handed an endowment and may send any fraction to an anonymous trustee; the experimenter triples whatever is sent; the trustee then decides how much, if any, to return. The amount sent measures trust; the amount returned measures trustworthiness. Because the game is anonymous and one-shot, narrow self-interest predicts sending nothing — yet investors reliably send, and trustees reliably return. The protocol spread to thousands of follow-on studies across cultures, ages, and conditions, and individual send-rates correlate with the same person’s answer to the survey item.
The investor sending fraction $s$ of an endowment $E$ chooses $s$ to maximize expected payoff given a belief $r$ about the fraction the trustee will return on the tripled stake:
$$\max_{s \in [0,1]} \; (1-s)E + r \cdot 3sE$$A purely self-interested trustee returns $r = 0$, so a backward-inducting investor sends $s = 0$. Observed behaviour — positive $s$, positive $r$ — is the measured trust the parameter is meant to capture.
I send you \$5. The experimenter turns it into \$15 in your hands. Now it is up to you: keep all \$15, or send some back. If I expect you to share, I send; if I expect you to keep it, I hold my \$5. How much I send is a clean, repeatable number for how much I trust a stranger — and most people, in most places, send more than zero.
Three moves, one importation. Each is real economics, not a token gesture. But look at what they share: trust is a number — aggregate or individual — never a network property. The relational structure that did all the work in Stage 1 has been compressed out.
The bar the importation has to clear
Set the importation against what Stage 1 established, item by item. Bourdieu’s point was that social capital depends on relational position — the aggregate score erases position by averaging over it. Coleman’s point was that enforceable trust switches on under network closure — the score cannot tell a closed network from an open one with the same mean. Putnam’s Italian result turned on the historical density of horizontal association predicting later institutional performance — the score is a snapshot with no structure and no history. Granovetter’s weak ties carried information precisely because of where they sat in the graph — the score is the same whether you are a broker or an isolate.
This is not yet a verdict against economics. It is the standard the importation now has to be judged against: four specific things sociology’s apparatus captured that a single national number, or a single game parameter, gives up.
Genuine work, partly flattening
Start with what the importation gets right, because it is a lot. The trust-and-growth literature catches real cross-country variance that survives the credibility revolution’s tougher standards — the Algan-Cahuc instrument is not a trick, and the correlation does not evaporate when you take endogeneity seriously. The trust game produces replicated cross-cultural patterns and links individual behaviour to the same survey measure, which means both instruments are tracking something real. And the institutional-economics edge of the field — North’s rules, Acemoglu and Robinson’s work on state capacity, Ostrom on the self-governance of commons — takes structure far more seriously than the headline regressions and sits much closer to the sociological apparatus than a fair critic should ignore.
That lineage — North through Coase, Ostrom, and Acemoglu, all of whom treat trust and civic norms as institutional substrate rather than as a survey residual — is mapped in History of Economic Thought, the institutionalist tradition. It is the part of economics least vulnerable to the flattening charge.
But the dominant trust-and-growth importation is partly flattening, in a specific and nameable way: it compresses a relational apparatus into aggregate-and-individual measures. That compression is the right call for the cross-country policy question — you cannot run a network analysis at the scale of nations, and the survey average is a defensible operationalization when the country is your unit. It is the wrong call for the within-country question, where the network structure is exactly the thing that changes. Which raises the obvious test: find a case where giving up the structure makes you misread the world.
So economics’s apparatus is right for one class of question and throws away structural information for another. Where does throwing it away actually mislead? There is a live, contested case sitting in the news.
Where the framings diverge
“The story of the American twentieth century was an upswing from ‘I’ to ‘we’ and then a downswing back to ‘I.’ … But the downswing was not a simple, uniform fraying. It was bound up with deepening polarization — a sorting of Americans into camps that trust their own and distrust the other.”
— Robert Putnam & Shaylyn Romney Garrett, The Upswing, 2020 (paraphrasing the book’s thesis)
By 2020 the headline numbers were stark. Gallup and the Edelman Trust Barometer showed trust in American institutions — government, media, business, even science — at multi-decade lows. The aggregate reading writes itself: Americans no longer trust their institutions. But Putnam, who launched the modern decline narrative, had by 2020 revised the mechanism. The fall, he argued, was not uniform. It was a reorganization — and that is precisely the kind of fact the aggregate cannot see.
Set up the institutional object first, then run the relational apparatus on it. The thing whose decline everyone is measuring — trust in institutions — is, in economics’s own terms, trust in the rules of the game and in the people who administer them. That object lives in the institutional apparatus from Stage 1, and the granular complement is the experimental measure from Stage 2: survey aggregates and game-measured trust are the two instruments the home discipline brings to the table.
Now the relational machinery. Here is the mechanism the aggregate misses. When network ties reorganize so that within-cluster density rises and the bridges between clusters thin, the mean trust per relationship need not fall at all — people may trust their own side more than ever — yet the national aggregate drops, because cross-cutting trust between the camps has collapsed. The level did not change uniformly; the distribution reorganized. Computational-sociology work on exactly this — DellaPosta, Shi, and Macy on the “Why Do Liberals Drink Lattes?” clustering dynamic, Chris Bail on social-media echo chambers, Damon Centola on how behaviour and belief spread through network structure — gives the mechanism empirical teeth.
Burt’s structural-holes apparatus from Stage 1 explains why the reorganization is self-reinforcing: as bridges between clusters thin, the few remaining brokers gain outsized influence over what crosses the gap, and they have every incentive to widen it. The same structural lens names a second blind spot — clustered low-trust pockets. A country with a high national trust average can contain dense neighbourhoods, professions, or online communities where trust has cratered, with effects on collective action and institutional performance far larger than their share of the population. The aggregate cannot see the pocket; the relational measure is built to.
Two readings of the same collapse
Take the aggregate reading at its strongest first, because it is not wrong. Institutional trust has fallen, the fall is robust across independent data sources, and it has measurable consequences: lower compliance, weaker state capacity, harder collective action. The trust-and-growth framework predicts that an economy losing institutional trust loses a productivity-relevant input, and the prediction holds. For a finance minister asking “is declining trust a macro risk for this country?” the aggregate measure is the right tool and the answer is yes. This is the case at full force, not a flimsy version set up to fall over.
It is, however, incomplete — and the incompleteness is the whole policy problem. The same data, read through network structure, shows not uniform decline but reorganization. Trust in your-side institutions stays high; trust in the other side’s collapses; the network of trust re-sorts into denser within-cluster ties and thinner cross-cluster bridges. Polarization on this reading is not a moral failing of individuals to be lectured out of them. It is a structural feature of how the network rewired. A policy that targets the aggregate — a national trust-in-institutions campaign — without targeting the structure that produced the polarization will spend its budget and miss. The structural-position framing here rhymes with how power, too, concentrates at network positions; the parallel is taken up in a sibling cross-topic walkthrough on power across the disciplines.
“Americans have simply stopped trusting each other and their institutions. We are becoming a low-trust society, and the data prove it.”
— A common framing of the institutional-trust collapse in op-ed and barometer commentary, 2020–present
Are we becoming a “low-trust society”?
The aggregate numbers say trust is falling across the board. But “the average fell” and “everyone trusts less” are different claims — and the relational apparatus says only the first one is true.
Which reading is reading the world?
The aggregate reading is the home-discipline reading and it is not wrong: institutional trust has fallen and the fall matters. But for the specific question — what is happening to American institutional trust, and what would reverse it? — it is incomplete in a way that changes the policy. The relational apparatus catches which institutions are losing whose trust, through which mechanism, propagating along which thinning bridges. This is the case where the by-question-scale calibration shows its teeth. The within-country, dynamic question is exactly the territory where sociology’s apparatus keeps its edge, and pretending the national average settles it is how you design an intervention that fails.
If each apparatus owns a scale of question where it wins, what does it look like to use both at once? For over a decade, a quiet body of hybrid work — network economics, computational sociology fed by experimental microdata, agent-based models — has been doing exactly that. Where has it landed?
The emerging synthesis
“Economic and social outcomes — from how diseases spread, to which products we buy, to how much education we pursue, to whether or not we can find a job — depend in important ways on the structure of the networks in which we are embedded.”
— Matthew O. Jackson, Social and Economic Networks, 2008
Jackson is an economist, and his program is a deliberate answer to the flattening. It takes network structure as a first-class object — modelled with the field’s own game-theoretic and equilibrium apparatus — rather than averaging it away. Outcomes that look like the product of an aggregate variable, he argues, are often the product of network structure in disguise. That is the sociological insight, re-derived inside economics.
Thread one: network economics. Jackson’s program models network formation as a strategic game, derives the network structures that emerge in equilibrium, and studies how shocks propagate through them. Acemoglu and Ozdaglar’s work on cascades and systemic risk shows how the topology of a network — not just the average connection — governs whether a local shock stays local or sweeps the system. Bramoullé and co-authors formalize peer effects so that an individual’s outcome depends on their position in the graph. The mathematics is unmistakably economic; the object it operates on is the structure sociology insisted mattered.
Thread two: computational sociology with experimental microdata. Centola runs controlled network experiments on online platforms, varying the structure of who can see whom and measuring how that changes what spreads. Macy and collaborators build agent-based models of polarization calibrated against survey and behavioural data. Here survey trust, game-measured trust, and network structure feed a single empirical model — both disciplines’ instruments treated as measurements of real and distinct things, not as rivals.
Thread three: economics’s own institutional edge. The part of economics that was always closest to sociology has kept advancing — Acemoglu and Robinson on state capacity, North’s late work on cognition and institutions, Ostrom on how real communities self-govern commons without either market or state. Algan and Cahuc sit on the boundary: cross-country regression methodology paired with a careful mechanism story about inherited, transmitted trust, which is a relational claim wearing a regression’s clothes. This network-economics and complexity turn is part of the methodological expansion charted in History of Economic Thought, modern pluralism — the chapter where economics stops being a single method and becomes a toolkit.
Three threads, one movement — and it is a real one, not a hoped-for one. The integration is also genuinely partial. Network economics has not absorbed Bourdieu’s convertibility-of-capitals; computational sociology has not produced policy-actionable cross-country results; the institutional edge has not displaced the headline trust-and-growth regressions in the development debate. But the gap is narrower than it was in 1997, and the trajectory points one way.
Both measurements, one model
The engagement has already been done at full strength — sociology’s apparatus in Stage 1, economics’s in Stage 2, each in its own voice. Here the two stop competing. Bourdieu’s relational position and Knack and Keefer’s national average are both real measurements of real things at different scales; the hybrid work uses both as inputs rather than forcing a choice between them. The synthesis is integrative, not a victory for either side — which is the right way to read it.
Calibrate by question, not by discipline
The honest verdict is not “sociology was right” or “economics was right.” It is that two genuine analytic tools live at two scales of analysis, and the answer is to match the tool to the question:
- Cross-country, policy-level questions — does higher national trust track growth, investment, contract enforcement? Should aid build civic norms? Economics’s apparatus is the right tool. The survey aggregate is a defensible operationalization when the country is the unit; Knack-Keefer, Algan-Cahuc, and the state-capacity literature give the right evidence at that scale.
- Within-country, dynamic questions — why is trust declining, how does network position gate opportunity, what propagates polarization, where do low-trust pockets form? Sociology’s apparatus keeps its edge. The aggregate hides the distribution; relational measures catch what changes when trust changes.
- Hybrid questions, where the two scales interact, are where the integration is actually happening. Partial, but real — and the right expectation is that the hybrid work keeps narrowing the gap.
One last thing worth naming: this is a recurring pattern, not a one-off. Whenever one discipline imports a concept from another — trust from sociology, and, in a parallel case, “capitalism” from history and comparative politics — the same drama plays out: real intellectual work, real flattening, and a slow hybrid reconciliation. The trust importation and the capitalism importation are two worked examples of the same methodological challenge.
Where this leaves us
We started with a survey question standing in for a concept that sociology had spent three decades making structural. Putnam, Bourdieu, Coleman, and Granovetter built a relational apparatus — social capital as a property of positions, ties, and the resources flowing through them — with replicated findings the aggregate could never reach. Economics then imported the word and rebuilt it as a number: a national trust score in a growth regression, a send-rate in a trust game. That importation was real intellectual work, and it was partly flattening. The post-2020 institutional-trust collapse showed exactly where the flattening bites: read as a falling average it looks like depletion; read through network structure it is reorganization, and the two imply opposite remedies.
The verdict is to calibrate by question, not by discipline:
- Cross-country policy scale — economics’s aggregate apparatus is the right tool, and the survey measure is a defensible operationalization at the level of nations.
- Within-country dynamic scale — sociology’s relational apparatus keeps its edge, because the structure is what changes when trust changes.
- The interaction of the two scales — the live frontier, where the hybrid work happens.
That hybrid work — Jackson’s network economics, Centola and Macy’s computational sociology fed by experimental microdata, the institutional edge from Acemoglu and Robinson through Ostrom, and Algan-Cahuc straddling the boundary — is the live frontier in 2026. The integration is partial. But the gap between the two apparatus is narrower than it was when economics first borrowed the word, and the honest answer is no longer to pick a side. It is to ask which question you are answering, and reach for the tool built to that scale.