Strategic Intelligence
Strategic Intelligence
HomeArtificial ScientistIntelligence InfrastructureHacking EducationPolicy IntelligenceNext ComputingAGI ArchitecturesSocial PhysicsFuture EconomicsArchiveAbout
Sign in
Social Physics

The Blame Reflex — Why Trust Breaks in Complex Societies

When public problems outgrow what people can see, judge and influence, grievance looks for someone to blame.

Metamatics's avatar
Jakub Žegklitz-Bareš's avatar
Metamatics and Jakub Žegklitz-Bareš
Nov 22

The evidence says this happens to people who have lost status, voice and control, not to people who lack intelligence. The loop it sets off can be named, graded and broken.

ENSI Research · Built on a library of 46 primary documents · Compiled 2026-10-20


The argument, before the list

Most writing on the “trust crisis” begins with a number and ends with a lament. The numbers are real. Only 40% of people across the OECD have high or moderately high trust in their national government (OECD 2026, p. 21). Trust in news is at 37%, the lowest since the Reuters Institute began measuring in 2015 (Reuters Institute 2026, p. 18). Seventy per cent of respondents in 28 countries are hesitant or unwilling to trust someone different from them (Edelman 2026, p. 17). The lament is familiar too: people have become irrational and tribal, and social media has made them worse.

This report starts somewhere else. It treats the trust problem as a problem of how people handle harm they cannot explain. Inflation, energy transitions, migration, pandemics and automation are systemic. Each has many causes and no single author. The moral mind, however, runs on a template of “an intentional agent and a suffering moral patient” (Gray et al. 2012, p. 101). Blame, unlike anger, needs a target. Without an agent causally linked to the harm, “blame does not arise because there is not target for it” (Malle et al. 2014, p. 151). When suffering is real and no agent is visible, people supply one (Gray et al. 2012, pp. 111–112). The mismatch between agentless harm and an agent-seeking mind is what this report calls the blame reflex.

The tempting claim is that complexity, combined with loss of control and status, pushes people toward emotional and provocative explanations, toward claiming victimhood and toward scapegoating. Here that claim is a hypothesis, and it is tested. Part of it holds. The existential threat model (van Prooijen 2020), Need for Chaos (Petersen et al. 2023), scapegoating under deprivation (Bilewicz & Krzeminski 2010) and competitive victimhood (Noor et al. 2012) all trace a path from threat to agent-blaming explanation. Part of it fails. Six experiments that manipulated people’s sense of control found no effect on conspiracy belief (d = −0.03; Stojanov et al. 2020, pp. 11–12). On identity-laden questions, people with more numeracy polarise more (Kahan et al. 2013, p. 24). Everyday blame tracks evidence about intentions far more than it tracks outcomes or prejudice (Malle et al. 2014, pp. 160–162).

What survives is sharper than the original. Complexity does not produce scapegoating by itself. It creates latitude, room in which identity, grievance and status loss decide which explanation feels right. The people most prone to share hostile rumours are not the least able. They are those who have personally lost status, feel lonely and feel abandoned by elites (Petersen et al. 2023, pp. 12–15). The evidence leaves no room for the “dumb people” framing. It points to identity protection, marginalisation, status loss and powerlessness, which are conditions that societies create and can change.

That changes who must act. Rothstein and Stolle put it plainly: governments “cannot put the blame on their citizens for the lack of social capital”, because that lack “is caused by dysfunctional government institutions” (Rothstein & Stolle 2007, p. 30). Read this way, the blame reflex is a diagnostic. It shows where people bear harm without an accountable agent and without a say. The report builds toward a seven-move loop through which blame is generated, aimed, returned and locked into insular trust, and it grades every link.

Summary of the main points

  • Trust is a willingness to be vulnerable where one cannot monitor. It rests on ability, benevolence and integrity, and it is how societies reduce complexity (Mayer et al. 1995; Rousseau et al. 1998; Luhmann in Gambetta 1988).

  • Trust has forms that behave differently. Particularised, strategic and generalised (moralistic) trust follow different rules. Political trust fell in the Great Recession while generalised trust barely moved (Algan et al. 2017).

  • Institutions dominate the determinants. Impartiality and low corruption correlate with generalised trust at about r = .64–.68 (Delhey & Newton 2004; Rothstein & Stolle 2007). Economic loss matters. Ethnic diversity has a real but small effect, a partial correlation of −0.026 (Dinesen et al. 2020).

  • Voice is the steepest gradient. 69% of those who feel they have a say trust their government, against 22% of those who do not (OECD 2026, p. 50).

  • Trust is re-sorting rather than collapsing. It is draining from distant institutions and pooling in near circles (Edelman 2026, p. 15).

  • The blame hypothesis is half right. Threat, deprivation and status loss are linked to conspiracy, hostile rumours, scapegoating and competitive victimhood. But acute loss of control does not cause conspiracy belief, and skill amplifies identity bias.

  • “Dumb people” is the wrong diagnosis. Identity, marginalisation and powerlessness are the drivers.

  • Blame moves in seven steps: comprehension gap, loss of control, grievance, search for an agent, target selection, reciprocal victimhood, insular trust. The control link is the weakest.

  • The breakers are institutional: real voice, fair process, buffers against loss, and bridging-based ranking.

How this report is organised

Section 1 defines trust in seven propositions. Section 2 sets out nine factors behind social and institutional trust. Section 3 reads the 2025–26 OECD, Edelman and Reuters surveys. Section 4 states the hypothesis. Section 5 gives eight lines of support and Section 6 six counter-findings. Section 7 sets aside the “dumb people” story. Section 8 assembles and grades the seven moves of the blame loop. Section 9 draws six reframes. Online amplifiers (angles 06–08 of the library) and remedies (angles 09–10) appear only where they show which link can be cut.


1. What trust is: seven propositions

1. Trust is a willingness to be vulnerable. The field’s most-cited definition is “the willingness of a party to be vulnerable to the actions of another party based on the expectation that the other will perform a particular action important to the trustor, irrespective of the ability to monitor or control that other party” (Mayer et al. 1995, p. 5). A twelve-paper cross-disciplinary forum converged on “a psychological state comprising the intention to accept vulnerability based upon positive expectations of the intentions or behavior of another”. It needs two conditions, risk and interdependence. “Trust would not be needed if actions could be undertaken with complete certainty and no risk” (Rousseau et al. 1998, p. 4). Gambetta defines it as a subjective probability, centred on a mid-point of uncertainty, assessed before one can monitor (Gambetta 1988, pp. 230–231). Trust is needed exactly where people cannot check, so as issues become harder to monitor, the demand for trust rises.

2. Trustworthiness rests on ability, benevolence and integrity. Mayer and colleagues distilled 22 earlier lists of antecedents into three (Mayer et al. 1995, p. 11). Ability is domain competence. Benevolence is wanting to do good to the trustor “aside from an egocentric profit motive”. Integrity is adherence to acceptable principles (pp. 10–12). A lack of any one can undermine trust (p. 14). In public life these become competence on hard problems, care for “people like me” and even-handed rules. The OECD’s drivers (reliability, responsiveness, integrity, openness, fairness) translate them closely (OECD 2026, p. 12). Attribution matters too. When a bad act is traced to orders from above, the agent’s integrity survives, because the acts “are not deemed to be his or her fault” (Mayer et al. 1995, p. 20). Who is blamed decides whose trustworthiness falls.

3. Trust reduces complexity, and complexity raises its price. Luhmann separates confidence (alternatives not considered) from trust (chosen despite possible disappointment). The distinction predicts where blame goes: “In the case of confidence you will react to disappointment by external attribution. In the case of trust you will have to consider an internal attribution” (Luhmann in Gambetta 1988, pp. 110–111). Increasingly complex societies “may in fact generate systems which require more confidence as a prerequisite of participation and more trust as a condition of the best utilization of chances” (p. 112 [99]). Yet testing trust “require[s] a relatively concrete setting” and “a previous structural reduction of complexity” (p. 116 [103]). Where systems are too complex to test trust concretely, people fall back on confidence. When confidence fails, disappointment is attributed outward.

4. Trust comes in three forms. Delhey and Newton set out the standard typology (Delhey & Newton 2004, p. 5 [3]). Particularised (”thick”) trust lives in small face-to-face communities with close sanctions. Instrumental or calculating trust, also called strategic or tit-for-tat trust, is a cost-benefit judgment based on track record. Generalised (”thin”) trust in strangers marks large, mobile, heterogeneous societies. Uslaner’s moral reading, quoted by Algan and colleagues, is that “moralistic trust is a commandment to treat people as if they were trustworthy … [and] the belief that others share your fundamental moral values” (Algan et al. 2017, p. 66 [374]). The distinction explains a key result. Unemployment shocks damaged political trust far more than generalised trust, “consistent with the argument that general trust has a moral component inherited through education and socialization” (p. 61 [369]). Strategic trust breaks at the first violation (Rousseau et al. 1998, p. 9). Moralistic trust moves slowly.

5. Particularised trust is double-edged. In-group reciprocity sustains cooperation. It can also create in-groups “embedded in a matrix of hostile relationships with outsiders”, with feuds and discrimination “against anyone of a different color, religion, or ethnicity” (Ostrom 1998, p. 17). Generalised trust “may not extend easily to all others in general, as opposed to others who are like us” (Delhey & Newton 2004, p. 30). A society can therefore lose generalised trust while particularised trust rises. The result is not low trust but insular trust, the endpoint of the loop in Section 8.

6. Trust is self-reinforcing and asymmetric. Trust, reciprocity and reputation are “positively reinforcing”, so “a decrease in any one of these can lead to a downward spiral” (Ostrom 1998, p. 13). “Everyone honest” and “everyone cheats a little” are both stable, self-fulfilling states (Dasgupta in Gambetta 1988, pp. 70–72). Destruction is easier than repair. “While it is never that difficult to find evidence of untrustworthy behaviour, it is virtually impossible to prove its positive mirror image”. Trust rests “not on evidence but on the lack of contrary evidence”, which “makes it vulnerable to deliberate destruction” (Gambetta 1988, pp. 246–247). The data agree. In 20 countries a 10-point fall in police confidence went with a 6-point fall in generalised trust, while rises brought no significant gain (Rothstein & Stolle 2007, p. 23). Greek trust in government fell from above 50% in 2004 to 7% in 2012 and never exceeded 37% afterwards, even as services improved (OECD 2026, pp. 24–26). Fast decay and slow rebuild is a parameter every model of trust must carry.

7. Imposed control crowds trust out, and voice builds it. “Trust is not a control mechanism but a substitute for control … Some controls actually appear to signal the absence of trust” (Rousseau et al. 1998, p. 8). Too much monitoring can make people “feel they are not trusted and thus become less trustworthy” (Ostrom 1998, p. 17). Coercion breeds “power and resentment” (Gambetta 1988, p. 233). The mirror image is voice. Groups that met face to face and voted on their own sanctioning system reached 93% of optimal yield with only 4% defection (Ostrom 1998, p. 8). Political trust reverts to “mere confidence when the opinion spreads that you cannot really influence political behaviour through the ballot” (Luhmann in Gambetta 1988, p. 111). Together the seven propositions set the frame. Trust is needed where people cannot monitor, it is cheaper to destroy than to build, and it rests on a felt capacity to influence. Complex, distant governance strains all three.

2. Nine factors of social and institutional trust

Four studies in the library explain why trust is high in some places and low in others: a 60-country analysis (Delhey & Newton 2004), an institutional theory with macro and micro tests (Rothstein & Stolle 2007), a regional panel with an instrumental-variables design (Algan et al. 2017) and a meta-analysis of 1,001 estimates (Dinesen et al. 2020). The OECD’s 2026 survey adds driver estimates. Two outcomes must be kept apart. Generalised trust in “most people” is slow-moving and largely a property of the society. Political trust in parliaments, parties and governments is fast-moving and partisan. The factors run roughly from the strongest evidence to the weakest.

Factor 1: Impartial, effective institutions. This is the strongest and most consistent correlate. Generalised trust correlates with rule of law at .683, government effectiveness at .641 and corruption at −.665 across countries (Delhey & Newton 2004, p. 20). The figures are .6457 for impartiality and .6455 for efficiency in Rothstein and Stolle (2007, p. 35). Their thesis is that “trust thrives most in societies with effective, impartial and fair street-level bureaucracies” (p. 2), through four mechanisms: safety; inference from officials to everyone else (”if those in position of responsibility cannot be trusted, then ‘most other people’ can surely not be trusted”); watching fellow citizens; and direct experience of fair or biased treatment (p. 18). In Canada, each point on a 0–100 court rating raises the odds of generalised trust by about 10% (p. 28). The authors admit causality is not identified and call for quasi-experiments around institutional reform (pp. 29–30).

Factor 2: Which institutions. Citizens separate political institutions (parliament, parties, government), order institutions (army, police, courts) and power-checking institutions (press, TV). The split appears in 56 countries and 64,997 respondents (Rothstein & Stolle 2007, p. 32). Political institutions are trusted on partisan grounds, while order institutions are trusted for “even-handedness and/or impartiality” (p. 13). Hence the 2025 OECD ranking: armed forces 66%, police 63%, courts 54%, local government 46%, news media 40%, parliament 37% and parties 26% (OECD 2026, pp. 35–36). The OECD calls the ordering a possible “sign of ‘skeptical trust’” (p. 34). For blame, the split means that a shock can drain political trust without touching the institutions people meet every day.

Factor 3: Economic shocks, measured as change. Algan and colleagues give the best-identified evidence in the library. EU unemployment rose from 7% in 2007 to 11% in 2013 (Algan et al. 2017, p. 2). Instrumenting with the pre-crisis share of construction, each extra point of unemployment raised the anti-establishment vote by 2 to 4.4 points (p. 37). Five extra points of unemployment cut trust in the national parliament by 3.65 points in 2SLS (p. 55). “It is the change in unemployment—rather than its level—that correlates with voting for nonmainstream parties” (p. 4). In the Brexit vote, a one-standard-deviation larger rise in unemployment raised Leave by 4 to 5.5 points, against about 1 point for the level (p. 41). Loss, not deprivation as such, moves people. The authors concede that their instrument “is not fully exogenous” (p. 4).

Factor 4: Shocks hit political trust, not generalised trust. Interpersonal trust did not fall after the crisis (”if anything … increased somewhat”; Algan et al. 2017, p. 18). The standardised effect of unemployment on political trust was about twice its effect on interpersonal trust, and on satisfaction with democracy almost five times (pp. 45, 49–50). The effects were stronger among people without a degree and among older cohorts (p. 5). Distrust rose toward national and EU politicians but not toward the police or the UN. The authors read this as citizens having “assigned the blame for the rise in unemployment to the inefficient national and European institutions” (p. 66). This is inferred from patterns, not measured. It is still the clearest sign in the library that blame for systemic harm is routed to whoever seems responsible.

Factor 5: Inequality and insecurity. Inequality correlates with generalised trust at −.466, with a multivariate β of −.25 (Delhey & Newton 2004, pp. 20, 23). Rich, egalitarian countries average 38% trusting, against 20% for poor, unequal ones, though the middle categories carry large errors (p. 21). Within countries, people with financial concerns trust their government 18 points less (OECD 2026, p. 48). Edelman’s gap between high- and low-income trust grew from 6 points in 2012 to 15–16 in later waves, and reaches 29 in the US (Edelman 2026, p. 7). Only 32% expect the next generation to be better off (p. 10).

Factor 6: Ethnic diversity, at its true size. Precision matters most for this factor. A meta-analysis of 1,001 estimates from 87 studies finds “a statistically significant negative relationship”, a partial correlation of −0.0256, called “rather modest” (Dinesen et al. 2020, p. 10). A best-practice subset gives −0.0283 (p. 18). The effect is strongest for trust in neighbours and at neighbourhood level, almost three times the country-level effect, which is not significantly different from zero (pp. 11–12). Because in-group trust falls too, the pattern fits “hunkering down” better than group threat (pp. 11–12). The authors conclude that “apocalyptic claims regarding the severe threat of ethnic diversity for social trust … are exaggerated” (p. 17). Diversity is a weak cause of distrust but a convenient explanation for it. That gap between effect and salience is the fingerprint of scapegoating.

Factor 7: Voice, the steepest gradient. In the OECD survey, 69% of people who feel “people like them have a say” trust the national government, against 22% of those who do not. The gap is 47 points, ranging from 60 in Australia to 36 in Chile (OECD 2026, p. 50). Only 31% feel they have a say, unchanged since 2021 (pp. 48, 51). The barrier is attitudinal. Forty per cent think officials “do not care about people like me”, and “voting is not sufficient anymore for citizens to feel heard” (p. 131). This is Luhmann’s warning in survey form, and it makes voice the natural lever against the second move of the blame loop.

Factor 8: Competence on complex problems. The OECD’s strongest driver of trust in government is whether decisions use “the best available evidence”. It is “by far most strongly associated”, followed “at a distance” by balancing current and future generations (OECD 2026, pp. 30–32). Only 42% think evidence is used (p. 29). Where trust moved by at least 4 points, governance perceptions explain “at least forty percent” of the change, and responsiveness and reliability on complex issues outweigh fairness or integrity (pp. 32–34). Service satisfaction is high, at 67%, but it mostly drives trust in local government and the civil service (pp. 13, 15). This is the most direct support in the library for the premise of this report. National trust is won or lost on the problems people find hardest to judge for themselves.

Factor 9: The interpersonal baseline and partisanship. High interpersonal trusters are about twice as likely to trust the police (75% vs 38%) and four times as likely to trust the government (54% vs 14%) (OECD 2026, p. 59). Supporters of governing parties trust the government 26 points more, a gap that has widened. Gender and age gaps are only 7 and 5 points (p. 48). One expected factor does not make the list. Voluntary membership fails once other variables are controlled (β = .05; Delhey & Newton 2004, pp. 23, 29), and joiners’ higher trust “seems to be an effect of self-selection”. “Many associations are in fact established to create distrust” (Rothstein & Stolle 2007, pp. 6–7). Belonging to groups, online or offline, does not by itself build bridging trust.

The pattern across the nine is clear. The large effects are relational and institutional: impartiality, voice, competence, partisanship and the interpersonal baseline. Economic effects are real and work through loss. Compositional effects are small. An account of distrust that leads with the small effects deserves suspicion, and so does any blame narrative built on them.


3. Where trust stands: five readings of the 2025–26 surveys

The three largest current surveys in the library are OECD (about 2,000 per country in 38 countries, fieldwork September–November 2025; OECD 2026, p. 17), Edelman (33,938 respondents in 28 countries, fieldwork October–November 2025; Edelman 2026, p. 2) and Reuters (97,520 respondents in 48 markets, fieldwork January–February 2026; Reuters Institute 2026, p. 5). All three are online, quota-weighted, non-probability panels. Levels cannot be compared across instruments: “trust in media” is 54 in Edelman, 40% in the OECD and 37% in Reuters. Directions and gradients can be compared.

Reading 1: stable at a low level. Across 29 countries in both waves, OECD trust in national government went from 40.0% to 40.7% (OECD 2026, p. 22). The Edelman Trust Index rose from 56 to 57 (Edelman 2026, p. 5). The “collapse” story fits parties (26%), parliaments (37%) and news (37%), not institutions in general.

Reading 2: news is still falling, partly by composition. Trust in news fell 3 points, dropping in 29 of 48 markets (Reuters Institute 2026, pp. 10, 18). News met on social media is trusted by 22% and AI chatbot answers by 20%, so “trust overall is prone to go down” as consumption moves there (pp. 10, 20). Social and video networks are now the leading gateway to online news, at 54% (p. 9).

Reading 3: trust is retreating inward. Net trust shifts after societal events were −16 for government leaders and −11 for major news organisations, against +11 for neighbours, family and friends and +11 for coworkers (Edelman 2026, p. 15). Local government is trusted 6 points above national (OECD 2026, p. 36). Weekly exposure to sources of a different political leaning fell 6 points to 39% (Edelman 2026, p. 14).

Reading 4: insularity and grievance travel together. Seventy per cent are insular: 30% unwilling and 40% hesitant to trust someone who differs (Edelman 2026, p. 17). Majorities are insular in 25 of 28 markets and across left, centre and right (pp. 48–49). Among the insular, 61% report moderate or high grievance, against 45% of the open (p. 20). The insular would trust an institution led by “someone different” 28–30 points less (p. 21). These are proprietary, cross-sectional constructs whose appendix is not in the library.

Reading 5: majorities want less blame, and the engaged minority rewards more. Eighty per cent rate a government that “avoid[s] rhetoric that blames or vilifies particular groups” as effective. Eighty-one per cent want “accurate headlines instead of exaggerated or fear-inducing ones” (Edelman 2026, p. 30). Encouraging cooperation “without taking a side” earns trust from 35%, and backing the respondent’s own position from 13% (p. 26). Yet the minority that prefers like-minded news shares, comments and pays more, rewarding “the production of partisan news that an important and highly engaged minority may want” (Reuters Institute 2026, pp. 61, 63). Angles 06 and 08 of the library show that the feed usually follows the minority.

The measured landscape is less a collapse than a re-sorting. Trust is draining from distant, representative and mediated institutions and pooling in near circles. Grievance concentrates among the unheard and the economically exposed. This is the terrain on which the blame hypothesis must be tested.

4. The hypothesis, stated so it can fail

The working hypothesis has four links, each to be checked separately:

  1. Complexity. As public issues grow more systemic, technical and distant, more people face harm they cannot trace to a cause.

  2. Control and status. That comprehension gap, especially with loss of control and status, is felt as threat.

  3. Explanation style. Threat pushes people toward emotional, provocative, agent-centred explanations such as conspiracy theories, hostile rumours and moralised outrage.

  4. Blame and victimhood. Those explanations select a culprit (institutions or an out-group), cast one’s own group as victim, and lower trust across groups and in institutions.

One premise must be flagged first. No document in the library measures whether public issues have actually become more complex over time (digest 05). Link 1 is an assumption about the world. What can be tested is what happens when people face complexity, threat and loss.


5. The case for: eight lines of support

Support 1: the existential threat model. Existential threat, “feelings of anxiety or uncertainty, often because of distressing events that call one’s values, one’s way of life, or even one’s existence into question”, triggers sense-making through pattern perception and agency detection. That process yields conspiracy theories “only when antagonistic outgroups are salient” (van Prooijen 2020, pp. 2–3). Uncertainty favours “rigid and radical leaders, who offer simple (and therefore understandable) solutions for complex problems” (p. 8). The evidence is experimental. Belief rises when a president’s assassination leads to war and when an opposition leader dies rather than survives. Induced uncertainty and status-quo threat raise it too (p. 10). Van Prooijen himself names the out-group condition “scapegoating” (p. 5).

Support 2: Need for Chaos. Eight US surveys (combined N = 10,921) identify “a desire for a new beginning through the destruction of order and established structures” (Petersen et al. 2023, pp. 4–5). One standard deviation of it raises intentions to share hostile political rumours by about 10 points, with a range of 4–19 (p. 8). The motive is mobilisation “against disliked groups” and amusing friends. Truth is negatively related to sharing (p. 11). Need for Chaos grows from status-driven risk-taking combined with loneliness (p. 13). At its peak, partisans of both sides share rumours “to hurt both parties” out of “a feeling of being abandoned” (pp. 11–12). This finding covers links 2–4 at once.

Support 3: dyadic morality. Moral cognition completes its own template. Suffering demands an agent. Intentional shocks of equal voltage feel more painful, and unfair splits make people “perceive the hand of an intentional agent”. “When suffering cannot be attributed to human agents, people often blame nonhuman agents” (Gray et al. 2012, pp. 111–112). This explains why systemic harm does not stay systemic in people’s minds.

Support 4: the Path Model of Blame. Blame runs in ordered steps: a norm violation, a causal agent, intentionality, then reasons or preventability (Malle et al. 2014, p. 151). Two features open it to distortion. Values can be “preset” by group membership (”the rival always intentionally harms us”; p. 158). And the usual check, that “the community can fact-check the blamer’s warrant” (p. 149), weakens among strangers. Third-person blame then “becomes akin to road rage … or hateful anonymous comments on the internet”, often “a first step toward socially excluding the offender” (p. 173). Scapegoats “can in fact increase the coherence of a group” (p. 174).

Support 5: competitive victimhood and the victimhood trait. “Framing yourself as a victim is a more effective way of escaping blame than framing yourself as a hero” (Gray et al. 2012, p. 113). Groups in conflict compete for the victim role because it brings cohesion, licence and freedom from responsibility, and recruits third parties (Noor et al. 2012, pp. 358–359). In Northern Ireland and Chile, competitive victimhood “was a unique negative predictor of inter-group forgiveness”, mediated by trust in the out-group’s intentions (p. 362). Individually, the Tendency for Interpersonal Victimhood is stable (three-week retest r = 0.77). It predicts entitlement to immoral behaviour (β = 0.27), lower dyadic trust (r = −0.18) and desire for revenge (r = 0.68) (Gabay et al. 2020, pp. 5–7). It also lowers belief in one’s ability to cope (p. 9). Victimhood and loss of agency feed each other.

Support 6: collective narcissism. Conspiracy theories “valorize the self and the in-group by allowing blame for negative outcomes to be attributed to others”, recruited “to relieve the self or in-group from a sense of culpability for their disadvantaged position”. They are predicted by collective narcissism, “a belief in the in-group’s greatness paired with a belief that other people do not appreciate it enough” (Douglas et al. 2017, p. 4). Seeing one’s group as superior makes salient out-groups look antagonistic. Collective narcissism predicts conspiracy theories about “different nations, minority groups, and competing political parties” (van Prooijen 2020, pp. 15–16). The primary studies are cited by these reviews but not held in the library, so no effect sizes are given.

Support 7: scapegoating under deprivation. In a Polish national survey (n = 1,098), deprivation predicted discriminatory intent toward Jews (β = .24) and belief in Jewish control (β = .21). Belief predicted discrimination (β = .34), giving an indirect effect of β = .07 (95% CI .05–.10) (Bilewicz & Krzeminski 2010, p. 239). Deprivation also predicted discrimination directly in Ukraine (β = .22; p. 241). This is link 4 measured in a population. Experienced loss raises hostility to a target group, partly through an ideology naming it as the hidden agent.

Support 8: the comprehension gap. The “illusion of explanatory depth”, the finding that people overrate their grasp of complex systems, comes from Sloman, Fernbach and colleagues. That work is not held in this library and is not used as evidence here. The library has proxies. On a standard covariance task, 59% of a national sample answered wrongly, and only 48% of the more numerate half solved the neutral version (Kahan et al. 2013, pp. 17–18). Citizens manage by being “experts at something else: identifying who knows what about what” (p. 32). The OECD’s top driver, perceived evidence-based decision-making (OECD 2026, pp. 30–32), is the institutional face of the same fact. Where people cannot judge for themselves, trust is their only access to the truth. A break in trust leaves an explanatory vacuum.

The online environment adds a premium. Each out-group word raises the odds of sharing by 67% (Rathje et al. 2021, via digest 06). Engagement ranking amplified political anger by 0.75 SD and out-group animosity by 0.24 SD beyond users’ own choices (Milli et al. 2025, via digest 08).


6. The case against: six counter-findings

Counter 1: manipulating control does not cause conspiracy belief. This is the most direct test of link 2, and it fails. Across six online experiments, Stojanov and colleagues “observed no effect of control manipulations on conspiracy theory beliefs, while replicating previously reported correlational evidence” (Stojanov et al. 2020, p. 1). In Study 1 the manipulation moved felt mastery, yet low-control participants scored descriptively lower on conspiracy ideation (pp. 4–5). The pooled effect was d = −0.03 (95% CI −0.20 to 0.13), and Bayesian posteriors favoured the null at 0.90–0.93 (pp. 11–12). Earlier positive studies had power of only .17–.22 (pp. 12–13). The correlation is real (r = −0.237 and −0.217) but may be confounded or reversed. Conspiracy exposure increased powerlessness in another study (p. 13). The manipulations were weak, acute MTurk primes. The authors’ own field study found that tornado victims endorsed weather-related conspiracies more for about three months (pp. 13–14). Acute loss of control does not cause conspiracy belief. Chronic, situated powerlessness remains untested.

Counter 2: the more skilled reason more motivatedly. Kahan and colleagues test a “Science Comprehension Thesis”, which blames conflict on “defects in the public’s knowledge and reasoning capacities”, against identity-protective cognition (Kahan et al. 2013, p. 2). Identity wins. When the problem was framed as gun control, high-numeracy partisans whose outlook the answer affirmed were 45 points more likely to solve it than those it threatened. Among low-numeracy partisans the gap was 25 points (p. 24). Numeracy raised accuracy by 21–32 points for congenial answers and by only 3–11 for threatening ones (p. 25). The cause is social. A wrong belief costs an individual almost nothing, but disagreeing with one’s group risks “the loss of trust among peers, stigmatization within his community; and even the loss of economic opportunities” (p. 6). Such reasoning is “perfectly rational, from an individual-welfare perspective” (p. 30). Complexity gives everyone room to settle on the congenial answer, and the able use that room best.

Counter 3: complexity without identity produces no polarisation. In the equally hard skin-rash version, partisan differences were “not meaningful” (Kahan et al. 2013, p. 25). “Very few facts … ever become symbols of group identity”. Without that, people “of all levels of science comprehension” track the evidence (p. 5). There is a related claim, from Sloman and colleagues (2022), that asking people to explain how a policy works moderates views on consequence-based issues but not on issues tied to sacred values. That study is not held in this library and is noted only as a pointer. Kahan’s contrast is the in-library evidence for the same boundary. Explanation works where an issue concerns consequences and fails where the issue has become a badge.

Counter 4: everyday blame tracks evidence. Outcome severity correlates with blame at only r = .17 (Robbennolt’s meta-analysis). Mental-state information “explained 6 times more variance” than outcomes, and character effects vanished within subjects (Malle et al. 2014, pp. 160–163). There is “little support for blatant motivated reasoning in blame judgments” (p. 162). Even Nazi propaganda was “a systematic ‘argument’”, not “irrational lashing out” (p. 174). The distortion sits upstream, in which agents are made salient and which intentions are preset. The fault lies in the inputs people receive, not in the faculty they apply to them.

Counter 5: threat can raise trust. After 9/11, conspiracy theories surged, yet Bush’s approval was historically high. Control threat raises both conspiracy belief about government and support for it (van Prooijen 2020, p. 4). Conspiracy theories can even protect system satisfaction by blaming “a few bad apples” (Douglas et al. 2017, p. 5). The outcome depends on whether government is construed as in-group or as “an antagonistic outgroup within society” (van Prooijen 2020, p. 15). The prior state of trust decides which way threat breaks.

Counter 6: neither target nor trigger is fixed. Classical scapegoat theory could not predict whom people blame, and induced frustration did not change attitudes to minorities in experiments (Bilewicz & Krzeminski 2010, p. 236). In Ukraine, deprivation barely predicted belief in Jewish control (β = .08; indirect β = .02), and other groups may be blamed instead (pp. 241–242). Personal status loss raises Need for Chaos (β = 0.05), while group status loss slightly lowers it (β = −0.02) (Petersen et al. 2023, pp. 14–15). Unemployment raised the view that immigrants harm the economy but not views of their cultural impact, and average attitudes did not worsen after the crisis (Algan et al. 2017, pp. 20, 51). History and elite supply select the target, not frustration alone.


7. Not the “dumb people” story

A common framing must be set aside: complex issues overwhelm the less intelligent, who then reach for simple villains. The library contradicts each part of it.

  • Skill does not protect. High-numeracy partisans polarise more (45 vs 25 points; Kahan et al. 2013, p. 24).

  • The drivers are positional. Need for Chaos rises with personal status loss, loneliness and feeling “abandoned” (Petersen et al. 2023, pp. 12–15). The low-education link to conspiracy belief runs partly “by feelings of powerlessness”, and minorities’ belief reflects “a general belief that the societal system is rigged” (van Prooijen 2020, p. 11).

  • Blame-processing is lawful. It follows evidence about intent (Malle et al. 2014, pp. 160–162). What varies is what people are shown.

  • Victimhood follows experience. It is predicted by anxious attachment (β = 0.46; Gabay et al. 2020, p. 6) and transmitted by “chosen traumas” and “us-them” journalism (Noor et al. 2012, p. 355).

  • Loss, not level, predicts revolt (Algan et al. 2017, p. 4).

  • Contempt makes it worse. “Silencing, ridicule, and other exclusionary reactions will only exacerbate the feelings of marginalization” (Petersen et al. 2023, p. 18).

The defensible statement is this. Identity protection, marginalisation, status loss and powerlessness, not low intelligence, drive the turn to blame. Complexity widens the space in which those forces operate. Responsibility moves from individual minds to the conditions of voice, fairness and security that societies provide.

8. How blame shifts and gets projected: the seven moves of the blame loop

Once each link is graded on its own evidence, the supports and counter-findings fit one dynamic account, a loop of seven moves. Grades are strong (experimental, quasi-experimental or meta-analytic), moderate (consistent correlational or mixed experimental evidence), contested (direct tests disagree) and assumed (untested in the library). Each move ends with the point where it can be cut.

Move 1: the comprehension gap opens. Issues become systemic and distant. People can no longer test trust in “a relatively concrete setting” and fall back on confidence (Luhmann in Gambetta 1988, p. 116). Knowledge is held socially, through “who knows what about what” (Kahan et al. 2013, p. 32). Only 42% think government uses the best available evidence (OECD 2026, p. 29). Grade: assumed for the trend, moderate for the mechanism. Cut: legibility, meaning a visible and contestable use of evidence, the OECD’s strongest driver.

Move 2: the gap is felt as loss of control. Felt mastery and conspiracy ideation correlate at about r = −.22 to −.24 (Stojanov et al. 2020, pp. 4–6), and feeling unheard separates trusters by 47 points (OECD 2026, p. 50). Grade: contested. Acute manipulation does not move belief (d = −0.03; Stojanov et al. 2020, pp. 11–12), and causation may run from conspiracy exposure to powerlessness (Douglas et al. 2017, p. 3). The defensible version concerns chronic, positional powerlessness. Cut: real ownership. Self-designed rules reach 90–93% of optimal cooperation (Ostrom 1998, p. 8; Ostrom 2009, p. 18).

Move 3: loss becomes grievance. Grievance builds from change: rising unemployment (Algan et al. 2017, p. 4), personal status loss (Petersen et al. 2023, pp. 14–15) and feeling “abandoned” (p. 12). It clusters with insularity, at 61% among the insular against 45% among the open (Edelman 2026, p. 20). Grade: strong for economic loss leading to political distrust (instrumented panel), moderate for status and grievance. Cut: buffers for losers. The anti-incumbent effect of trade-related job loss was smaller where workers got training and income support (Margalit 2011, cited by De Vries in Algan et al. 2017, p. 77).

Move 4: grievance searches for an agent. Suffering without an author presses to fill the agent slot (Gray et al. 2012, pp. 111–112), and blame cannot form without one (Malle et al. 2014, p. 151). Under threat, explanations that recast systemic harm as an evil group’s intention gain appeal (van Prooijen 2020, pp. 8, 10). Hostile rumours are shared to mobilise, not because they are true (Petersen et al. 2023, p. 11). Grade: strong for the cognitive template. Moderate for threat leading to conspiracy, since the samples are not disadvantaged and the mediation evidence is “somewhat indirect” (van Prooijen 2020, p. 13). Cut: accountable agents. Harm with a named owner who explains and can be challenged fills the slot with warrant instead of fantasy.

Move 5: a target is selected, either institutions or an out-group. Along the institutional route, blame follows perceived responsibility. Trust fell in national and EU politicians but not in the police or the UN (Algan et al. 2017, p. 66), and government becomes a target once construed as “an antagonistic outgroup within society” (van Prooijen 2020, p. 15). Along the out-group route, blame lands where intent is preset by stereotype (Malle et al. 2014, p. 158). In Poland it ran partly through belief in Jewish control (indirect β = .07; Bilewicz & Krzeminski 2010, p. 239). Elites switch routes. Incumbents “shift their attention to values or identity-related matters when economic conditions deteriorate” to “dodge the responsibility” (De Vries in Algan et al. 2017, pp. 77–79). Online, out-group words are the strongest predictor of sharing (+67% per word; Rathje et al. 2021). Grade: moderate. Target choice is the least predictable step (Bilewicz & Krzeminski 2010, p. 242). Cut: elite restraint, endorsed by 80% (Edelman 2026, p. 30), and feed design. Reranking hostile posts in real feeds shifted out-party warmth by about 2–2.5 degrees in a week (Piccardi et al. 2024, via digest 10).

Move 6: the target becomes a victim and blames back. The blamed group claims the victim role, which brings cohesion, licence and freedom from responsibility (Noor et al. 2012, pp. 358–359). The claim lowers trust in the other side’s intentions and so lowers forgiveness (p. 362). The victimhood trait predicts revenge (r = 0.68) and costly punishment of mild offenders (Gabay et al. 2020, pp. 7–8). Without warrant checks, the ordinary cycle of complaint and countercomplaint has no repair step. Even among couples, 95% of complaints are contested (Malle et al. 2014, pp. 173–174). Political elites amplify the cycle: out-group posts by members of Congress gained more than those by media accounts (Rathje et al. 2021, via digest 06). Grade: moderate, and strongest for violent conflicts. Cut: reciprocal exchange of empowerment for victims and acceptance for perpetrators (Noor et al. 2012, pp. 360–361). Two warnings apply. Recognition alone can entrench victimhood claims (Gabay et al. 2020, p. 9), and “a climate of trust should be established” first (Noor et al. 2012, p. 365).

Move 7: trust contracts into the in-group. The end state is trust re-sorted, not trust at zero. Outrage raises the expresser’s trustworthiness within the group (Jordan et al. 2016, via digest 06) while trust falls across groups. Net trust moves toward neighbours (+11) and away from government leaders (−16) (Edelman 2026, p. 15). The insular would trust an institution led by “someone different” 28–30 points less (p. 21), so every change of government drains trust from half the population. US out-party warmth fell from 48° in the 1970s to 20° (Finkel et al. 2020, via digest 07). Luhmann foresaw “retreat into smaller worlds of purely local importance” and “fundamentalist attitudes” (Luhmann in Gambetta 1988, pp. 116–117). Grade: moderate for the shift, assumed for its link to earlier moves. Cut: bridging. A deliberative weekend warmed partisans toward each other by 13–14 thermometer points, and the most extreme warmed most (Fishkin et al. 2021, pp. 14–15).

The loop closes. Insular trust widens the comprehension gap, because fewer sources are trusted across lines. Conspiracy exposure lowers trust even in institutions unrelated to the theory (Douglas et al. 2017, p. 4). “The single best predictor of belief in one conspiracy theory is belief in a different conspiracy theory” (van Prooijen 2020, pp. 17–18). Distrust “may become the source of its own evidence” (Gambetta 1988, p. 247).

The links at a glance

  • Strong: suffering → search for an agent · economic loss → political distrust and anti-establishment vote · out-group content → online reach · conspiracy exposure → lower institutional trust.

  • Moderate: threat → conspiracy when an out-group is salient · deprivation → scapegoating · competitive victimhood → lower out-group trust · trust retreating to near circles.

  • Contested: loss of control → conspiracy (correlated, not shown causal) · target selection · recognition as cure or fuel.

  • Assumed: that public issues are becoming more complex.

The Blame Field

The loop is implemented as an interactive simulation, “The Blame Field” (visualization/blame-field.html). Two hundred simulated people in four neighbourhoods are tied by threads whose opacity shows trust. Three controls match the cut points above: issue complexity (local and legible to global and technical), feed ranking (engagement to bridging) and ownership and voice. A crisis trigger adds a temporary spike in complexity. Grievance builds in proportion to the comprehension gap times the lack of control. When the gap is high, blame lands on the institutions. Otherwise it lands on the least-trusted group, which blames back. Engagement ranking gives blame extra reach, and people with a real say send proposals instead. The simulation states its limits. It treats control and suspicion as correlated more than causal (Stojanov et al. 2020), and it lets the feed amplify grievance but not create it (Guess et al. 2023; Boxell et al. 2020). It is a tool for reasoning about leverage, not a forecast.


9. What the model implies: six reframes

1. Read blame as a map of missing accountability. Scapegoating concentrates where harm is real, systemic and has no answerable owner. The first response is to supply that owner. Correcting the blamers comes second.

2. Put voice before information. Information campaigns left court trust unchanged over decades (Tyler 2007, Rottman essay, p. 31), and information alone can backfire in partisan settings (Fishkin et al. 2021, p. 3). Voice is worth 47 points (OECD 2026, p. 50), but only if it is real. “Postdecision voice” still raises perceived fairness, which makes voice theatre both effective and a future betrayal (Tyler 2007, White Paper, p. 11).

3. Cushion losses, not just levels. Grievance is driven by change. Compensation for those hit by shocks reduces the political fallout (Algan et al. 2017, p. 77).

4. Keep facts from becoming badges. Once a fact signals membership, skill amplifies the split (Kahan et al. 2013, pp. 24, 32). Do not frame technical questions as loyalty tests, and watch for others doing it.

5. Never answer grievance with contempt. Ridicule feeds marginalisation (Petersen et al. 2023, p. 18). Visible paternalism such as content warnings backfired with conservatives (Jia et al. 2023, via digest 10). Fair process raises trust even after a loss (Tyler 2007, p. 27).

6. Rank for bridging, not for blame. Feeds that pay for out-group hostility speed up Moves 5 and 6. Bridging systems that reward content endorsed across divides slow them (Ovadya & Thorburn 2023; Wojcik et al. 2022). The evidence that feeds create polarisation is weak (Guess et al. 2023; Boxell et al. 2020). The evidence that they amplify the loop is substantial.


Close: what to do first

The blame reflex is not a defect of the public. It is what an agent-seeking moral mind does under harm it cannot trace, when it has no say and something to lose. Complexity supplies the latitude. Status loss and threatened identity supply the motive. Engagement-ranked media supply the reward. All three are shaped by institutions, so the first moves fall to governments and platform governors:

  • make the use of evidence in complex decisions visible and contestable;

  • give people a real say in the rules behind the harms they bear;

  • cushion losses before they harden into grievance;

  • refuse the politics of the convenient scapegoat;

  • measure and rank for what divided groups endorse together.

Trust “is depleted through not being used” (Gambetta 1988, p. 247). The loop described here shows how it falls out of use. Reports 2 and 3 take up how to put it back to work.


Sources from the library cited in this report

  • Algan, Guriev, Papaioannou & Passari 2017 — The European Trust Crisis and the Rise of Populism — 02 - Determinants of Social Trust

  • Bilewicz & Krzeminski 2010 — Anti-Semitism in Poland and Ukraine: The Belief in Jewish Control as a Mechanism of Scapegoating — 04 - Blame, Scapegoating and Victimhood

  • Boxell, Gentzkow & Shapiro 2020 — Cross-Country Trends in Affective Polarization — 07 - Social Media, Polarisation and Declining Trust

  • Delhey & Newton 2004 — Social Trust: Global Pattern or Nordic Exceptionalism? — 02 - Determinants of Social Trust

  • Dinesen, Schaeffer & Sønderskov 2020 — Ethnic Diversity and Social Trust: A Narrative and Meta-Analytical Review — 02 - Determinants of Social Trust

  • Douglas, Sutton & Cichocka 2017 — The Psychology of Conspiracy Theories — 05 - Complexity, Uncertainty and Conspiracy

  • Edelman Trust Institute 2026 — 2026 Edelman Trust Barometer: Trust Amid Insularity — 03 - Measuring Trust

  • Finkel et al. 2020 — Political Sectarianism in America — 07 - Social Media, Polarisation and Declining Trust

  • Fishkin, Siu, Diamond & Bradburn 2021 — Is Deliberation an Antidote to Extreme Partisan Polarization? — 09 - Rebuilding Trust - Ownership, Agency and Participation

  • Gabay, Hameiri, Rubel-Lifschitz & Nadler 2020 — The Tendency for Interpersonal Victimhood — 04 - Blame, Scapegoating and Victimhood

  • Gambetta (ed.) 1988 — Trust: Making and Breaking Cooperative Relations (incl. Luhmann and Dasgupta chapters) — 01 - The Nature of Trust

  • Gray, Young & Waytz 2012 — Mind Perception Is the Essence of Morality — 04 - Blame, Scapegoating and Victimhood

  • Guess et al. 2023 — How Do Social Media Feed Algorithms Affect Attitudes and Behavior in an Election Campaign? — 07 - Social Media, Polarisation and Declining Trust

  • Jia, Lam, Mai, Hancock & Bernstein 2023 — Embedding Democratic Values into Social Media AIs via Societal Objective Functions — 10 - Prosocial Platform Design and Bridging Systems

  • Kahan, Peters, Dawson & Slovic 2013 — Motivated Numeracy and Enlightened Self-Government — 05 - Complexity, Uncertainty and Conspiracy

  • Malle, Guglielmo & Monroe 2014 — A Theory of Blame — 04 - Blame, Scapegoating and Victimhood

  • Mayer, Davis & Schoorman 1995 — An Integrative Model of Organizational Trust — 01 - The Nature of Trust

  • Milli et al. 2025 — Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media — 08 - Misinformation and Algorithmic Amplification

  • Noor, Shnabel, Halabi & Nadler 2012 — When Suffering Begets Suffering: The Psychology of Competitive Victimhood — 04 - Blame, Scapegoating and Victimhood

  • OECD 2026 — Drivers of Trust in Public Institutions 2026 Results — 03 - Measuring Trust

  • Ostrom 1998 — A Behavioral Approach to the Rational Choice Theory of Collective Action — 01 - The Nature of Trust

  • Ostrom 2009 — Beyond Markets and States: Polycentric Governance of Complex Economic Systems — 09 - Rebuilding Trust - Ownership, Agency and Participation

  • Ovadya & Thorburn 2023 — Bridging Systems — 10 - Prosocial Platform Design and Bridging Systems

  • Petersen, Osmundsen & Arceneaux 2023 — The “Need for Chaos” and Motivations to Share Hostile Political Rumors — 05 - Complexity, Uncertainty and Conspiracy

  • Piccardi et al. 2024 — Reranking Partisan Animosity in Algorithmic Social Media Feeds Alters Affective Polarization — 10 - Prosocial Platform Design and Bridging Systems

  • Rathje, Van Bavel & van der Linden 2021 — Out-group Animosity Drives Engagement on Social Media — 06 - Moral Outrage and Emotional Contagion Online

  • Reuters Institute 2026 — Digital News Report 2026 — 03 - Measuring Trust

  • Rothstein & Stolle 2007 — The Quality of Government and Social Capital — 02 - Determinants of Social Trust

  • Rousseau, Sitkin, Burt & Camerer 1998 — Not So Different After All: A Cross-Discipline View of Trust — 01 - The Nature of Trust

  • Stojanov, Bering & Halberstadt 2020 — Does Perceived Lack of Control Lead to Conspiracy Theory Beliefs? — 09 - Rebuilding Trust - Ownership, Agency and Participation

  • Tyler (lead) 2007 — Procedural Justice and the Courts (Court Review special issue) — 09 - Rebuilding Trust - Ownership, Agency and Participation

  • van Prooijen 2020 — An Existential Threat Model of Conspiracy Theories — 05 - Complexity, Uncertainty and Conspiracy

  • Wojcik et al. 2022 — Birdwatch: Crowd Wisdom and Bridging Algorithms — 10 - Prosocial Platform Design and Bridging Systems

Referred to but not held in the library, with no figures used: the work of Sloman, Fernbach and colleagues on the illusion of explanatory depth, including the 2022 study contrasting consequence-based and sacred-value issues, and the primary studies on collective narcissism by Golec de Zavala, Cichocka and colleagues, cited here only through Douglas et al. 2017 and van Prooijen 2020.

Discussion about this post

User's avatar

Keep reading

Strategic Intelligence

Articles by European Nexus for Strategic Intelligence. Our think tank focuses on advising the entrepreneurial and governance ecosystem how to advance critical intelligence infrastructure to advance our civilization to an era of safe abundance

  • About
  • Archive
  • Sitemap
© 2026 Intelligence Strategy Institute · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.