What is an Explanation?

Have you ever come across an explanation for a phenomenon, event, entity, occurrence, or fact and wondered how anyone could possibly find it satisfying? I increasingly find myself asking that question. It became especially noticeable while I was writing about the strategic use of bad arguments for propagandistic purposes. That post focused on argument, but I realized that much the same problem applies to explanation. Rather than simply cataloguing bad explanations and their rhetorical uses, I want to ask a more basic set of questions: What constitutes an explanation? How can we identify and evaluate one? Are there general criteria for judging explanatory quality? My aim is to identify what makes an explanation good so that we have better tools for evaluating the plethora of shit explanations we encounter in the wild.

Part I — What Makes Something an Explanation?

A Philosophical Perspective

Philosophers do not treat “explanation” as a single thing. Instead, they treat it as a family of practices with a common aim: making something intelligible by showing how or why it happens, or why it is the case. What counts as an explanation—and what makes one good—depends partly on the kind of “why?” question being asked (causal, mathematical, mechanistic, metaphysical, and so on) and partly on what the audience is trying to do with the answer (predict, intervene, understand, justify, or unify). Even so, several cross-cutting criteria recur across the literature.

The core structure of an explanation is the relation between the explanandum and the explanans. The explanandum is what is to be explained—a fact, event, pattern, lawlike regularity, or outcome. The explanans is what does the explaining: causes, laws, mechanisms, mathematical facts, grounding facts, and so on. An explanation succeeds when the explanans stands in the right explanatory relation to the explanandum. “Right” here means more than mere association: the relation must answer the relevant “why?” question.

How does an explanation stand in relation to what it explains? Different theories propose different “explanatory relations.” Here are the major ones.

Explanatory relation	Description
Covering-law / derivational relation (Hempel-style)	On the Deductive–Nomological (DN) picture, to explain an event is (roughly) to show it was to be expected from general laws plus initial conditions: a derivation of the explanandum from a lawlike set of premises. This is historically central, but also heavily criticized (because derivation can happen without genuine explanation, and because it struggles with asymmetries—e.g., flagpole/shadow cases). This relation type is logical/nomological entailment (or high-probability support in statistical variants).
Causal relation	A great deal of contemporary work argues that (many) explanations work by identifying causes or causal structure—not just regularities. Modern causal accounts focus on things like mechanisms, causal graphs, interventions, and counterfactual dependence. This type of relation is some form of causation (often understood via counterfactuals/interventions or mechanisms).
Unification relation	Another tradition says explanation unifies: you explain by showing how diverse phenomena follow from a small set of argument patterns or principles. (Think: “this reduces many facts to one framework.”) This aims to capture why very general theories can feel explanatory even when they aren’t obviously “naming causes.” The scientific-explanation literature treats unification as a serious competitor to purely causal accounts. This relation type is the integration of many facts under fewer explanatory patterns.
Counterfactual / interventionist relation	an explanation tells you what would change if you changed certain factors. The explanans matters if it supports the right counterfactuals (“if X were different, Y would be different”). This is closely tied to interventionist thinking about causal explanation. This type of relation is stable counterfactual dependence under interventions (or “difference-making”).
Non-causal explanations (especially mathematical)	Many philosophers think some real explanations are non-causal—especially in mathematics and sometimes in science when mathematics explains a phenomenon. The key idea is that understanding can come from proving, deriving, or showing structural constraints, even without citing causes. The types of relations here are mathematical/structural dependence, constraints, proofs, symmetry/optimization facts, etc.
Metaphysical / grounding explanations	Beyond science, philosophers talk about metaphysical explanation; explaining a fact in virtue of more fundamental facts (often discussed in terms of “grounding” or dependence). Example shape: “Why is this true? Because it holds in virtue of these more basic facts.” The relation type is metaphysical dependence (grounding, constitution, essence, etc.).
Pragmatic / question-relative relation	explanation is often contrastive and context-sensitive. We rarely ask “Why P?” in a vacuum; we ask “Why P rather than Q?” and we care about some causal factors rather than others. This helps explain why different answers can each be good explanations depending on what contrast class is salient and what the audience needs. (This theme runs through much of the explanation literature.) This type of relation is about providing information that appropriately answers the specific why-question asked (often implicitly contrastive).

What makes an explanation good? There isn’t one universally accepted checklist that applies identically to every domain (physics vs. history vs. metaphysics), but philosophers converge on a cluster of evaluative virtues. Think of them as dimensions along which explanations can be better or worse.

Evaluative virtue	What it requires
Truth or credibility of the explanans	A good explanation must be grounded in claims that are true, or at least sufficiently credible and approximately true in the relevant respects. If the explanans is simply false, then what we have is not really an explanation in the full sense, but perhaps a convenient narrative, heuristic, or fictional model. Philosophers usually allow that many explanations, especially in science, involve idealization or simplification. A model may omit friction, treat populations as homogeneous, or assume perfectly rational agents, yet still explain well because it captures the relevant structure of the phenomenon. What matters is not literal exactness in every detail, but whether the explanation is true enough, and true in the right way, to illuminate why the explanandum obtains. Still, outright falsehood remains a serious defect, because an explanation built on false premises cannot genuinely reveal what is responsible for the phenomenon.
Explanatory relevance, rather than mere correlation	A good explanation does more than point to something regularly associated with the phenomenon; it identifies factors that genuinely matter to its occurrence. This is the difference between explanatory relevance and mere correlation. Two things may track each other closely without one helping to explain the other. Philosophers therefore look for relations of dependence: causal, structural, mathematical, or metaphysical. In causal contexts, this often means identifying mechanisms, processes, or variables that make a difference to whether the explanandum occurs. In non-causal contexts, it may mean showing that the phenomenon follows from deeper principles, laws, or grounding relations. The basic point is that a good explanation cites what is responsible for the phenomenon, not merely what happens to accompany it.
Modal or counterfactual support	Many philosophers regard this as one of the central marks of a strong explanation. A good explanation does not only tell us why something happened in the actual case; it also helps us understand what would happen under different circumstances. It supports the right counterfactuals. It tells us what would change if certain factors were altered, removed, or intensified, and what would remain stable despite superficial variation. This is important because explanation is closely tied to understanding dependence. If an explanation shows that changing X would change Y, then it reveals something about how Y depends on X. This is why interventionist and difference-making accounts are influential: they treat explanation as illuminating the patterns of dependence that extend beyond the single observed case.
Depth	A good explanation has depth when it goes beyond surface description and reveals the underlying structure that makes the phenomenon intelligible. A shallow explanation may cite a regularity without showing why that regularity holds, or may stop at an immediate cause without tracing the deeper organization behind it. A deeper explanation, by contrast, identifies the mechanisms, structures, or principles that make the phenomenon robust and stable across a range of cases. In causal explanation, depth may come from uncovering layered mechanisms or chains of dependence. In mathematical explanation, it may come from showing that something holds by necessity given a structural feature. In metaphysical explanation, it may come from revealing grounding relations. Depth matters because it increases our sense that we are not just observing a pattern, but seeing why the pattern holds.
Scope and unification	An explanation is often better when it can account for a wider range of phenomena using the same conceptual resources. Rather than explaining one isolated fact in a piecemeal way, a strong explanation may show how many facts fit together within a single framework. This is the virtue of scope: it applies across cases, not just to one instance. It is also the virtue of unification: it reduces apparent fragmentation in our understanding by bringing diverse phenomena under common principles. Philosophers influenced by unificationist theories see this as a major explanatory good. Even outside those theories, there is broad agreement that it is an advantage when an explanation can illuminate not only this case, but many related ones, without requiring a separate ad hoc story for each.
Simplicity or parsimony, though not at any cost	Simplicity is widely treated as an explanatory virtue because explanations burdened with unnecessary assumptions, complications, or exceptions often seem less satisfying and less credible. A simple explanation typically avoids ad hoc additions and uses fewer independent assumptions to account for the same phenomenon. This can make it more elegant, more generalizable, and more epistemically manageable. However, philosophers are careful not to treat simplicity as an absolute criterion. An explanation should not be preferred merely because it is simpler if that simplicity comes at the expense of truth, adequacy, or relevance. A more complex explanation may be better if reality itself is complex. So parsimony is a genuine virtue, but it is a subordinate one: valuable when all else is roughly equal, not decisive in every case.
Precision and informativeness	A good explanation should provide enough detail to be genuinely informative. Vague or overly generic explanations can create the illusion of understanding without delivering it. Precision matters because it tells us exactly which factors are doing the explanatory work, under what conditions they operate, and how they relate to the phenomenon in question. In scientific contexts, this may involve specifying variables, mechanisms, boundary conditions, or functional relationships. In historical or philosophical contexts, it may involve clearly distinguishing background conditions from triggering factors, or necessary conditions from sufficient ones. Informativeness is closely connected to usefulness: an explanation is better when it not only sounds plausible, but gives us substantive insight into how and why the explanandum comes about.
Coherence with the rest of what we know	Other things being equal, explanations that fit well with established background knowledge are usually preferred. This includes coherence with well-supported theories, empirical findings, and broader frameworks of understanding. An explanation that clashes sharply with everything else we know carries an added burden of justification. Coherence matters because explanation does not occur in isolation; explanatory claims are assessed within larger networks of belief and theory. Still, this criterion is not absolute. Revolutionary episodes in science show that sometimes the best explanation initially appears to conflict with accepted views, only later reshaping the background framework itself. So coherence is a significant virtue, but not an inviolable constraint. It counts in favor of an explanation unless there are stronger reasons to depart from existing theory.
Non-circularity and explanatory asymmetry	A good explanation should not simply restate the explanandum in different words, nor should it presuppose what it is supposed to explain. Circularity undermines explanation because it fails to provide any independent basis for understanding the phenomenon. Relatedly, many philosophers stress that explanation is often asymmetric. Even if two facts are logically or mathematically connected, one may explain the other without the reverse being true. The classic example is that the height of a flagpole explains the length of its shadow under certain lighting conditions, but the shadow length does not explain the flagpole’s height. This asymmetry shows that explanation cannot be reduced to mere derivability or logical entailment. A good explanation must track the right direction of dependence.
Responsiveness to the question, or pragmatic fit	An explanation can be accurate and yet still fail as an explanation if it answers the wrong question. Explanatory adequacy depends partly on context: what is being asked, what contrast is in view, and what the audience is trying to understand. Someone asking why a patient developed a disease may want a proximate biological cause, a statistical risk factor, or a broader social explanation, depending on the context. Likewise, asking why an event occurred often implicitly means why it occurred rather than some alternative. This is sometimes called the contrastive dimension of explanation. Pragmatic fit matters because explanations are answers to questions, not merely free-floating statements of fact. Recognizing this helps explain why multiple different explanations of the same phenomenon can all be good, provided each addresses a distinct explanatory demand.

A concise way to put the overall point is this: good explanations are usually assessed along three broad dimensions. First, they must track genuine relations of dependence or relevance, whether causal, mathematical, or metaphysical. Second, they should exhibit epistemic virtues such as scope, simplicity, coherence, and precision. Third, they must fit the explanatory question being asked. A strong explanation is therefore not just true, but illuminating, informative, and appropriately targeted.

Explanations can also be indeterminate in the sense that more than one explanation may be equally good. There are several reasons this can happen, and it is not always a problem. One is contextual plurality: two explanations can both be good because they answer different contrastive questions. For example:

“Why did the window break?”
- Explanation 1 (causal/mechanistic): because a rock hit it at speed.
- Explanation 2 (structural): because the glass had a microfracture that made it fragile.

Both can be good depending on whether you’re asking about the triggering event or the predisposing condition. This is “pluralism” driven by the fact that explanations select a subset of the causal/structural story as relevant. There are also different explanatory types for the same explanandum. Sometimes you can explain the same phenomenon causally and mathematically/structurally, and it’s not obvious one dominates. The literature on non-causal explanation and mathematical explanation is partly about when this happens. A more problematic version is when there is underdetermination by evidence (a deeper indeterminacy). A stronger sense of “indeterminacy” is when the available evidence doesn’t decide between competing hypotheses/explanations that fit equally well. This is closely connected to underdetermination in philosophy of science. Here, two explanations might be Empirically equivalent (fit all the same observed data), equally coherent, simple, and powerful (at least by our current measures), yet mutually incompatible about what the world is like.

This possibility of a tie often arises in “Inference to the Best Explanation”; roughly, we infer the hypothesis that best explains the evidence. Sometimes there is no uniquely best explanation, like in cases where the explanatory virtues trade off. For example:

Explanation A is simpler but narrower.
Explanation B is messier but has greater scope.
Explanation C is elegant but relies on speculative entities.

Philosophers disagree about whether there are principled, universal ways to weight these virtues, or whether judgment and context are unavoidable. When are two explanations “equally good”? It depends what “equally good” means:

Same explanandum, same question-context, same virtue profile (a genuine tie).
Same explanandum, different context/contrast (both good but not competitors).
Same explanandum, different explanatory aims (predictive control vs. understanding vs. unification).
Same evidence fit, different metaphysical commitments (underdetermination).

Philosophers often think (2) and (3) are common; (1) is possible; (4) is a major philosophical issue in theory choice and realism debates.

So what is an explanation “according to philosophers,” in one integrated picture? Explanations aim to produce understanding by connecting the explanandum to an explanans in the right way. The “right way” is not one relation. Causal dependence is central in many domains, but there are also unificatory, mathematical/structural, and metaphysical (grounding) explanatory relations. Quality is multi-dimensional. Explanations are evaluated by relevance/difference-making, truth/credibility, counterfactual support, depth, scope/unification, simplicity, coherence, precision, and question-fit. Pluralism and indeterminacy are real possibilities. Sometimes multiple explanations coexist because of context, levels of description, or different explanatory projects; sometimes evidence and virtues fail to pick a unique winner (underdetermination).

So if you’re handed an “explanation,” try asking:

What exactly is the explanandum? (What is being explained—event, law, regularity, fact?)
What kind of why-question is it? (Causal? mathematical? metaphysical? contrastive?)
What dependence claim is being made? (Cause, mechanism, lawlike derivation, structural necessity, grounding?)
Does it support the right counterfactuals? (What would change if key factors changed?)
Is it relevant, not just associated?
Is it deep and general without being hand-wavy?
Does it trade off virtues sensibly? (Simplicity vs. scope vs. precision)
Are there rivals that fit equally well? If so, what extra considerations could break the tie?

Part II — Why Explanations Feel Satisfying

Psychological Aspects of Explanation

Explanations can feel correct, plausible, and satisfying even when they are badly mistaken. We appear to crave explanations, and the mere availability of one can bring a sense of relief. That reaction is closely connected to uncertainty reduction, cognitive closure, preference for order, and intolerance of ambiguity. These factors affect not only how we evaluate explanations, but also how long we search for alternatives and how willing we are to reconsider an answer once we have accepted it.

The philosophical discussion treated explanation primarily as an epistemic tool: a way of tracking real dependence through causes, mechanisms, mathematical constraints, grounding relations, and similar structures. The psychological question is different. Here the issue is explanation as a psychological product: how an account can produce felt understanding, relieve tension, and restore a sense of order. Much of the craving for explanation arises because felt understanding is partly driven by motivational and affective systems that are only loosely coupled to truth-tracking.

When something happens that violates expectation (“Why did that happen?”), you don’t just lack information—you often get a state of aversive arousal: confusion, anxiety, loss of control, or “meaning threat.” A major framework here is the Meaning Maintenance Model (MMM): humans need the world to “hang together” as a web of expected relations; when that coherence is threatened, people compensate by restoring meaning—sometimes in the same domain, sometimes elsewhere (“fluid compensation”). Another closely related line is compensatory control: when personal control is threatened, people seek order/structure, sometimes by detecting patterns—even illusory ones—and by endorsing systems that promise order. The "relief" feeling is real and predictable; an explanation can function like psychological closure—restoring coherence, reducing “I don’t know,” and returning a sense of predictability/control—even if the explanation is inaccurate.

Several mechanisms can make an explanation feel plausible even when it is badly off base. Coherence and fluency can feel like truth. A story that ties its elements together smoothly can acquire high subjective plausibility despite weak evidence; this is one reason conspiracy narratives can be so sticky, since they convert apparent randomness into order. The Illusion of Explanatory Depth (IOED) creates a related problem: the feeling of correctness can substitute for actual explanatory quality. People often think they understand complex causal systems much better than they do, and the gap becomes visible only when they are required to generate the missing details. Explanatory satisfaction and explanatory accuracy can therefore diverge sharply.

Explanation as uncertainty reduction

There is a deep relationship between explanation and uncertainty reduction. There are three kinds of “uncertainty” explanations can reduce:

Epistemic uncertainty: uncertainty about what is true (which hypothesis is correct).
Predictive/pragmatic uncertainty: uncertainty about what will happen next and what to do (action guidance).
Existential/meaning uncertainty: uncertainty that feels like incoherence, randomness, or loss of control.

Explanations reduce uncertainty by compressing many observations into a small set of organizing relations (“this happened because X, under conditions Y”). In Bayesian terms, they can reduce your “hypothesis space” and sharpen expectations (lower entropy), which is inherently stabilizing. (The Bayesian literature generally treats cognition as managing uncertainty by integrating evidence with prior beliefs.) But, uncertainty reduction isn’t automatically “good epistemology”, because people often reduce uncertainty by narrowing too fast (prematurely collapsing onto one hypothesis), freezing too early (resisting updates), and choosing explanations that satisfy motivational needs (order, control, group identity) over accuracy. So explanation is a powerful uncertainty-reducer, but it can be either truth-conducive (reducing uncertainty by actually tracking structure), or comfort-conducive (reducing uncertainty by feeling coherent).

Closure, ambiguity, and premature stopping

The Need for Cognitive Closure (NFC) often favors comfort-conducive explanations and changes what counts, psychologically, as a “good” explanation. NFC is the motivation to reach a definite answer rather than remain uncertain. Kruglanski’s theory emphasizes two tendencies: urgency (obtain closure as quickly as possible) and permanence (preserve that closure for as long as possible). Together they produce the characteristic pattern: seize on an explanation early, then stick with the resulting judgment. NFC is also measured as a disposition involving preferences for order and structure, discomfort with ambiguity, decisiveness, and close-mindedness. Under higher NFC, people tend to overweight speed, definiteness, simplicity, and coherence—traits that yield closure—and underweight calibration, depth, evidential sensitivity, and exploration of alternatives, which keep uncertainty alive longer. A “good explanation” can then become one that ends the question rather than one that best survives adversarial testing. Kruglanski’s model predicts that heightened closure motivation can reduce information processing and hypothesis generation while increasing confidence. The pathway looks like earlier stopping, stronger primacy effects, less critical probing, and confidence inflation. This is one reason plausible-but-wrong explanations can dominate: NFC can privilege decisiveness over diagnosticity.

Intolerance of ambiguity and preference for order are other drivers of this phenomena. Budner’s definition: intolerance of ambiguity is the tendency to perceive ambiguous situations as sources of threat, whereas tolerance treats them as desirable/interesting. This matters because if ambiguity is appraised as threat, then explanatory evaluation becomes emotionally loaded. Competing explanations are experienced as “unsettledness,” not as healthy epistemic pluralism. “I don’t know yet” becomes aversive. The mind becomes motivated to end ambiguity, not to map it. Preference for order can improve functioning (you get decisions made), but it can also distort comparison between explanations. It rewards explanations that partition the world cleanly (binary categories, single causes, clear villains) and penalizes explanations that are interaction-heavy, probabilistic, multi-level, or that keep uncertainty explicit. This becomes especially visible with complex phenomena (economies, ecosystems, politics, personal relationships) where the highest-quality explanation often is partial conditional, multi-factor, and honest about uncertainty. When order/ambiguity intolerance is high, those hallmarks of epistemic maturity can feel like weakness.

How accepted explanations become sticky

“Seizing and freezing,” permanence, and resistance to reconsideration are often consequences of fluency effects, IOED, NFC, intolerance of ambiguity, and preference for order. Kruglanski’s model basically says this is high under NFC. Once an explanation is adopted, several well-known cognitive dynamics reinforce it:

Selective exposure / confirmation bias: you preferentially seek or notice evidence consistent with the chosen explanation.
Interpretation bias: ambiguous evidence gets read as supportive.
Commitment and identity: explanations become socially and self-reputationally costly to abandon (“I was wrong”).
Cognitive economy: reopening the question reintroduces uncertainty and effort.

The result can look like “permanence” even if the person thinks they’re being evidence-responsive. And once you’ve frozen, the explanation often becomes cognitively fluent (“obvious”), which increases subjective plausibility and reduces felt ambiguity. That can create a self-sealing loop: confidence rises because the explanation is familiar, not because it’s stronger. This connects to IOED too: people can feel they understand because they have a label/story, not because they can generate the mechanism.

How motivation narrows the search space

These traits shape the search for explanations themselves (not just evaluation). Your motivational stance changes the space of explanations you ever consider. Under high NFC / high ambiguity intolerance search becomes exploitative rather than exploratory. You generate fewer competing hypotheses and do less counterfactual probing (e.g., “What evidence would make me abandon this?”). You may shift toward strategies that reduce cognitive load (simpler rules, reliance on early cues, reliance on consensus). There’s also a broader pattern: threats to control/uncertainty can increase pattern perception and acceptance of order-imposing narratives. So the “need for explanation” isn’t just preference over finished products—it’s a force that shapes which products ever make it to the shelf.

Explanatory strength vs. explanatory satisfaction

Truth-tracking explanations emphasize the virtues philosophers foreground: relevance and difference-making, depth, counterfactual support, evidential fit, and robustness. Psychologically satisfying explanations are often evaluated by a different set of cues: coherence, simplicity, narrative completeness, reduced ambiguity, restored control or meaning, and a metacognitive feeling of “rightness.” Under low stress, high curiosity, and strong accountability norms, the first set of criteria can dominate. Under threat, time pressure, fatigue, identity stakes, or high dispositional NFC, the second can dominate without the evaluator noticing the shift. People who tolerate ambiguity may therefore be better positioned for comparative evaluation because they can sustain partial explanations, probability distributions, model uncertainty, and iterative updating. If the evidence supports several models, they can keep several alive; NFC and ambiguity intolerance make that continued openness psychologically costly.

Practical safeguards

If you want tools that directly counter seizing/freezing and ambiguity aversion, the best ones are those that force delayed commitment and make alternatives cheap:

Explicitly write the contrast question: “Why P rather than Q?” (prevents generic closure answers).
Generate 2–4 competing explanations first (even if you like one).
Pre-commit to disconfirmers: “What observation would change my mind?”
Separate plausibility from support: Plausible story ≠ strong evidence.
Mechanism test: “Could I explain the causal chain step-by-step?” (IOED antidote).
Accountability/accuracy prompts reduce premature closure effects in the NFC framework (they lower the benefits of quick closure).

Illusion of Explanatory Depth in More Detail

The illusion of explanatory depth is the reliable finding that people think they understand complex mechanisms far better than they actually do—and they only realize the gap when they’re forced to produce a detailed explanation rather than merely recognize one. Rozenblit & Keil’s classic experiments had people rate how well they understood everyday devices/phenomena, then explain them step-by-step, then re-rate; confidence typically drops sharply after explaining. IOED is strongest for explanatory / causal-mechanistic knowledge (how systems work) and is weaker for other kinds of knowledge like simple facts or procedures.

I want to linger on this mechanism because it is somewhat more differentiated than cognitive closure, preference for order, or fluency effects. IOED concerns what we accept as sufficient to fill an explanatory gap; in that sense, it bears directly on what we treat as a good explanation. It appears to bias the application of criteria such as simplicity, coherence, scope, mechanistic detail, and counterfactual power in predictable directions.

IOED functions as a psychological engine that lets shallow explanations feel like deep understanding—so it’s tightly linked to uncertainty reduction and need for closure. Explanations reduce uncertainty psychologically before they reduce it epistemically. A quick, coherent “because-story” often reduces the uncomfortable feeling of not knowing. IOED makes that easy: you get the feeling of understanding from a rough sketch, and you don’t notice the missing steps until you’re challenged to cash it out. So IOED acts like a fast, low-effort closure mechanism. This pairs naturally with seizing and freezing: grabbing the first satisfying story (seizing), then sticking with it because it already delivered relief (freezing). The NFC framework describes exactly these urgency/permanence tendencies.

One reason IOED happens is that people evaluate their understanding at the wrong “resolution.” Alter, Oppenheimer, & Zemla argue IOED occurs because people use an inappropriately abstract construal when judging understanding—focusing on the gist/purpose (“what it does”) instead of the mechanism (“how it does it”). Abstract construals feel simpler, cleaner, more coherent, and less ambiguous. So if you have a strong preference for order, you’re even more likely to accept high-level stories as “good enough,” which amplifies IOED.

IOED involves a systematic underuse of explanatory standards—the kinds of checks philosophers and scientists treat as depth-producing, such as:

specifying mechanistic steps (not just labels),
identifying difference-makers (what changes the outcome),
stating boundary conditions (when it stops working),
supporting counterfactuals (“what if X were different?”),
being able to answer follow-up ‘how exactly?’ questions without collapsing.

In everyday cognition, people often don’t spontaneously apply those standards. IOED is what you get when “I can tell a coherent story / recognize a story” is mistaken for “I can meet the deeper standards.” Rozenblit & Keil describe several contributors to the initial feeling of knowing, including confusing labels with mechanisms and confusing environmental support (having the object around, or being able to look things up) with internal understanding. So IOED is a predictable mismatch between the standards people use to judge their understanding and the standards that actually track explanatory depth.

How do people come to overestimate their understanding of complex mechanisms? There are a few big mechanisms.

“Gist-to-depth” confusion (abstract summary feels like a full model): People often possess a functional schema (“toilets flush waste; the lever triggers water”) and mistake that for a causal model (“here’s how the valve, siphon, and tank dynamics produce the flush”). Until you’re forced to generate steps, your mind treats the gist as evidence of depth. This is exactly what the construal-level account predicts: abstract representations inflate perceived understanding.
“Visibility” and environmental scaffolding (the world lends you understanding you don’t actually have): For many artifacts, mechanisms look perceptually obvious (“working parts are right there”), which tempts you to think you’ve internalized the causal structure. Rozenblit & Keil explicitly discuss confusing environmental support (what the environment provides on demand) with what you actually know. Modern version: Google/YouTube/AI can make this even stronger—because access feels like possession.
Division of cognitive labor (you know that someone knows, so it feels like you know): A lot of what we call “my understanding” is socially distributed: I know who could explain it (a mechanic, an engineer) and I can use the artifact successfully. That practical familiarity can be misread as explanatory mastery.
Fluency and familiarity get misread as depth: If a topic is familiar, it feels easy to think about; that ease often gets interpreted as “I understand it.” Mills & Keil show people can miscalibrate explanatory understanding and become more accurate after trying to explain—suggesting the act of explanation reveals what fluency was hiding.
Closure motivation reduces “stress-testing”: When people want closure, they’re less likely to do the things that would puncture IOED: generate alternative explanations, run counterfactual checks, seek disconfirming evidence, and demand mechanistic detail. NFC doesn’t just make you accept weaker evidence; it makes you stop asking the questions that would expose shallowness.

IOED isn’t just a cute “you don’t know how a zipper works” fact. It can affect confidence calibration (high certainty with shallow support), and polarization/extremity (especially when people feel they understand complex policy mechanisms). A famous example: Fernbach et al. found that political extremism can be supported by an illusion of understanding, and that prompting people to explain mechanisms (rather than merely justify positions) can reduce perceived understanding and sometimes moderation follows. This is basically IOED and closure dynamics in the wild: mechanistic explanation prompts force a higher explanatory standard, breaking the “feels right” lock-in.

If you want to know whether you really understand something:

Rate your understanding.
Explain it step-by-step (mechanism, not just purpose).
Specify two failure cases (when it wouldn’t work).
Answer one counterfactual (“what if X were removed/changed?”).
Re-rate your understanding.

That procedure forces the standards people otherwise underuse—and it’s exactly why IOED reliably collapses when explanation is required.

Additional Examples from Cognitive Science

Cognitive science is full of examples where we accept explanation-shaped cognition (coherent, fluent, closure-giving) in place of explanatorily virtuous cognition (mechanistic depth, counterfactual support, discriminating evidence). Here are some of the clearest, well-studied patterns.

Inventing causes we do not actually know

Confabulation and the “introspection illusion”

People often don’t have direct access to the real causes of their judgments and actions, yet they can generate confident, plausible-sounding explanations anyway. Nisbett & Wilson (1977) review evidence that verbal reports about higher-order mental processes often reflect post hoc theorizing rather than introspective access. This is “succumbing to weak explanation” because the mind is highly skilled at producing coherence even when the causal basis is unknown or unavailable. It’s a built-in narrative generator that reduces uncertainty (“I know why I did that”) and supports identity protection (“I’m the sort of person who…”), even if the mechanism is missing.

Choice blindness: explaining choices you didn’t make

In choice blindness, people sometimes fail to notice when their choice has been covertly swapped and then provide reasons for the “chosen” option—reasons that can be detailed and confident. Johansson et al. develop this as evidence that people can rationalize outcomes and construct justifying explanations with minimal anchoring to their actual decision process. This shows how easily we accept an explanation because it fits a self-image and provides closure, not because it accurately tracks causes.

Mistaking persuasive cues for explanatory strength

The seductive allure of neuroscience: “brain words” inflate perceived explanation quality

A classic demonstration of persuasiveness outrunning explanatory virtue: Weisberg et al. (2008) found that poor explanations of psychological phenomena were judged more satisfying when they included irrelevant neuroscience information. People substitute surface cues of “scientificness” (technical detail, brain references) for deeper standards (mechanism, relevance, counterfactual support). This is a very direct instance of systematic underuse of explanatory standards.

Narrative transportation: stories lower scrutiny and increase belief

When people become absorbed in a narrative, they tend to accept story-consistent claims more readily and detect fewer problems. Green & Brock’s work on transportation shows that highly transported readers find fewer “false notes” and adopt more story-consistent beliefs. This is weak-explanation vulnerability by design: narrative coherence and emotional engagement can suppress the impulse to test alternatives or ask “what would falsify this?”

Availability-based reasoning: vivid exemplars substitute for base rates and mechanisms

We often judge what’s likely or common by how easily examples come to mind. Tversky & Kahneman’s classic work on availability shows systematic biases when ease-of-recall is used as a proxy for probability/frequency. This produces weak explanations. Vivid anecdotes make certain causal stories feel “obvious,” leading us to prefer narratively available explanations over statistically grounded ones.

Miscalibration/overconfidence (e.g., Dunning–Kruger) as an “explanation confidence” amplifier

When people lack skill in a domain, they may also lack the skill to accurately judge their own competence, leading to inflated confidence. Kruger & Dunning (1999) is the classic demonstration of miscalibration among low performers. If you don’t know what a good explanation requires, you can’t easily detect when yours is shallow—so weak explanations feel sufficient.

Imposing order on uncertain or noisy evidence

Hindsight bias: once we know the outcome, we invent inevitability

Hindsight bias (“I knew it all along”) is partly powered by coherence-building explanations after the fact. Fischhoff’s classic work demonstrates distorted remembered probabilities once outcomes are known. Roese & Vohs’ review emphasizes that a coherent explanation that resolves surprise fuels the feeling of inevitability—explanatory coherence becomes a signal that the outcome “had to happen.” Hindsight explanations feel deep because they reduce uncertainty maximally (the past becomes “ordered”), but they can be shallow or misleading because they’re constructed with privileged outcome knowledge.

Illusions of causality: mistaking correlation/contingency for real causation

People can acquire a strong belief that “X causes Y” even when X and Y are unrelated—especially under common learning conditions (high base rate of outcomes, attention to coincidences, etc.). Matute and colleagues review causal illusions and how they can support superstition and pseudoscience. This is a weak explanation trap because causal language provides instant explanatory closure (“that’s why”), so we lock onto it even when the dependence relation is missing.

Pattern perception under threat/lack of control (including conspiratorial explanations)

When people feel a loss of control, they are more likely to perceive meaningful patterns in randomness. Whitson & Galinsky (2008) show that lacking control increases illusory pattern perception, including conspiracies and superstitious beliefs. This has a direct tie to uncertainty reduction: when uncertainty/control threat is high, the mind prioritizes order-restoring explanations (even if they’re not truth-tracking).

Teleological bias: purpose-based “because” as a default explanation

Humans (children strongly, adults under some conditions) often accept “X is for Y” explanations even in domains where they’re inappropriate (e.g., natural phenomena). Kelemen’s work shows broad teleological preferences in childhood and evidence that adults can default to teleological explanations under time pressure. This is weak-explanation-friendly: teleology is cognitively cheap—it compresses complexity into intention/purpose, which is extremely satisfying and identity-compatible (“the world makes sense”), even when mechanistic explanation would be more accurate.

Across these literatures, the recurring pattern is:

Satisfaction cues	Strength cues
Coherence, simplicity, vividness, technical gloss, purpose/agency, closure, identity fit.	Discriminating evidence, mechanism detail, counterfactual support, boundary conditions, robustness to alternatives.

The concepts we discussed earlier—need for closure, seizing/freezing, intolerance of ambiguity—tilt people toward the first set of cues, which is exactly why weak explanations can be psychologically “sticky.”

Part III — Narrative, Identity, and Explanation

Explanation vs. Story and Narrative

Explanation and story/narrative overlap, but they’re not the same. Explanation is primarily about a dependence relation that answers a why/how-question (causal, mechanistic, mathematical, grounding, etc.). Narrative is primarily about sense-making via temporal/agentic structure—organizing events into a coherent sequence with roles, motives, turning points, and a “point” (what it means). Sometimes a narrative is an explanation (especially in history, action explanation, biography). But a narrative can also be explanatorily shallow while still feeling deeply satisfying.

What makes something a narrative (vs. an explanation)? We already know what an explanation is, so let’s turn to narrative. A narrative typically includes a Temporal sequence (this happened, then that happened), Agents and intentions (who did what and why), Coherence/plot (events “hang together”), and Significance (why it matters; what it means). Jerome Bruner’s view describes two contrasting modes of thought. One is Paradigmatic / logico-scientific, aimed at truth via argument, evidence, general principles. The second is Narrative, aimed at intelligibility through story form (humanly meaningful coherence). So narratives are often meaning-first; they make events feel intelligible by fitting them into an understandable plot.

Narratives and explanations overlap, but they perform several functions differently:

Narrative functions	Explanation functions
Meaning and coherence: Narratives reduce the “this is random / senseless” feeling by imposing structure. Bruner explicitly treats narrative as a way minds organize reality.	Truth-tracking dependence: Good explanations (especially causal/mechanistic ones) support the right counterfactuals: what would change if we intervened.
Memory, communication, and coordination: Stories are compressible, memorable, shareable. They transmit social norms, identity, and lessons.	Prediction and control: Explanations are often valuable because they enable forecasting and intervention, not merely coherence.
Motivation and emotion regulation (closure): Narratives can give relief: they turn uncertainty into a settled storyline (“here’s what happened and why”), which is psychologically stabilizing.	Comparative evaluation: Explanations are (in principle) rankable by epistemic virtues: fit to evidence, relevance, depth, scope, etc.

Narratives are optimized for human intelligibility and social transmission; explanations are optimized for dependence-tracking and epistemic control (though the best accounts often achieve both). You can have explanations lacking narrative form, and narratives lacking any feature of good explanation.

Narratives are sometimes a type of explanation. J. David Velleman argues that narrative can function as a genre of explanation because it renders events intelligible—not just listing what happened but organizing it so we “get it.” But this is exactly where the danger lies. Narrative intelligibility can be achieved by coherence alone, even when the underlying explanatory relation is weak. So narrative can be explanatory, but it can also be merely explanation-shaped. This means it can be persuasive, signalling explanatory adequacy without having explanatory depth or virtue.

A story can be persuasive because it has narrative virtues; coherence, vividness, identifiable protagonists/antagonists, emotional resonance and satisfying closure. But explanatory depth is about explanatory virtues: relevance/difference-making, mechanism/detail where appropriate, robustness and counterfactual support, honesty about uncertainty, non-ad-hocness and evidential grounding. These two sets of virtues can correlate—but they often diverge. Persuasive stories often reduce critical evaluation. Psychology has a well-studied mechanism: narrative transportation—being absorbed into a story can change beliefs and reduce scrutiny. Green & Brock’s work shows that “transportation” into narratives is associated with increased story-consistent beliefs, and that transported readers detect fewer “false notes.” So persuasion can undermine depth because it alters the evaluator, not just the content: it lowers analytic checking and raises acceptance.

Kahneman describes how intuitive thought (System 1) measures success by the coherence of the story it constructs, largely ignoring missing data—“WYSIATI” (what you see is all there is). This is a direct bridge to what we were getting at earlier. A persuasive narrative can deliver uncertainty reduction and closure while masking low evidential support or missing mechanisms

Narratives are one of the easiest ways to generate IOED-like “I understand it” feelings, because they supply causal-looking structure (A led to B led to C), they supply intentional explanations (someone wanted X), and they supply closure (the story ends). That can produce felt understanding without the ability to answer “What’s the mechanism?”, “What would falsify this?”, “What if we intervene on X—does Y change?” or “What boundary conditions make this fail?” So persuasive storytelling can become a substitute signal for explanatory competence.

If you want to compare two accounts of the same phenomenon, ask:

Intervention test: does it tell you what would happen if you changed a key factor? (explanation-strength)
Mechanism test: can it specify intermediate steps rather than jumping from cause to outcome?
Constraint test: what would have to be true for this to work, and what would break it?
Alternative test: does it actively rule out plausible competitors, or just feel coherent?
Uncertainty honesty: does it quantify/acknowledge unknowns, or convert them into plot?

A story that wins mainly on vividness/coherence but fails these tests is likely persuasive without being deep. The bottom line is Narratives are sense-making machines: they make events coherent, memorable, emotionally navigable, and socially transmissible. Explanations (in the epistemic/philosophical sense) are dependence-tracking machines: they aim at the right “because” relation and support counterfactuals/interventions. Persuasiveness can absolutely mask low explanatory depth, because transportation and coherence cues can reduce scrutiny and inflate confidence.

Narrative and Identity-Protective Cognition

Narrative explanation is a powerful way to make events intelligible.
Identity protection is a powerful motive in which “intelligibility” we accept.
Anecdotes are the delivery system that makes narratives vivid, memorable, and socially credible—sometimes at the expense of truth-tracking.

How can we distinguish identity-protective anecdote from legitimate explanatory case?

Dan Kahan and colleagues use identity-protective cognition to describe how people selectively credit or dismiss information in ways that protect the beliefs dominant in their group (or that protect social standing within it). The core idea is: when a factual claim becomes entangled with “who we are,” reasoning is recruited to defend identity rather than update belief. This matters for narrative because narratives are not just “accounts of what happened.” They typically come bundled with heroes/villains, blame/innocence, moral lessons, and implied group alignments (“people like us” vs “people like them”). So narrative explanation often functions as identity signaling: adopting it broadcasts membership, loyalty, and shared values.

Political psychology shows that when issues are identity-relevant, people engage in motivated skepticism: they more aggressively critique counter-attitudinal evidence and more readily accept congenial evidence—even when the objective quality is similar. That translates directly into explanation-evaluation. Evidence that threatens identity gets treated as “flawed,” “biased,” or “missing context.” Explanations that protect identity feel “obvious,” “realistic,” and “what’s really going on.” Narrative and explanation become social: closure isn’t just comfort; it’s belonging.

Narratives don’t just persuade; they can change how we scrutinize. A key mechanism is narrative transportation (Green & Brock): when people are absorbed into a story, they tend to adopt more story-consistent beliefs and detect fewer “false notes.” In other words, immersion often reduces critical monitoring. That matters for identity protection because transportation makes it easier for a narrative to do two things at once: produce felt understanding (“now it makes sense”), and immunize itself from counterevidence (“criticism feels like nitpicking” or “missing the point”). Narrative explanation is structurally well-suited to identity defense because it’s inherently meaning-and-value laden, and because its persuasive mode can suppress the very checking behaviors that expose shallow explanation.

Anecdotes are “exemplars”: they act like evidence even when they shouldn’t. Communication research (Zillmann & Brosius’ exemplification work) argues that vivid single cases (a “case report” in the media sense) can strongly shape people’s perceptions of how common or serious an issue is, because they’re concrete, emotional, and attention-grabbing. And classic judgment research explains why: we overweight what comes easily to mind (availability). A vivid anecdote becomes mentally “available,” so it feels representative even when it’s not. Anecdotes are uniquely useful for identity-protective purposes because they feel like first-hand truth (“I saw it,” “my friend lived it”), are hard to rebut without seeming callous (“you’re denying someone’s lived experience”), can be chosen selectively to fit the group narrative, and compress moral meaning into a memorable form (a single face can stand for a whole worldview). So in identity-charged domains, anecdotes often function less like neutral evidence and more like identity-laden tokens: they anchor the group’s “what’s really going on” story.

Does that mean anecdotes can’t legitimately explain? No—but “anecdote” has two different meanings.

Anecdote as exemplar	Anecdote as case
Anecdote as an exemplar (weak for truth, strong for persuasion/identity): This is the “I know a guy” story used to infer general patterns. It’s prone to selection bias, base-rate neglect, and motivated cherry-picking. It can be meaningful, but it’s typically low-grade evidence for general claims.	Anecdote as a case (can be genuinely explanatory under the right methodology): A single case can sometimes carry real explanatory weight if it’s treated as a case study rather than a rhetorical exemplar. In qualitative social science, process tracing uses within-case evidence to identify the intervening causal steps that produced an outcome (the mechanism), not just that outcome. This is explicitly presented as a way to make causal inferences by examining diagnostic steps inside a case.

Anecdote as exemplar

Anecdote as case

Anecdote as an exemplar (weak for truth, strong for persuasion/identity): This is the “I know a guy” story used to infer general patterns. It’s prone to selection bias, base-rate neglect, and motivated cherry-picking. It can be meaningful, but it’s typically low-grade evidence for general claims.

Anecdote as a case (can be genuinely explanatory under the right methodology): A single case can sometimes carry real explanatory weight if it’s treated as a case study rather than a rhetorical exemplar. In qualitative social science, process tracing uses within-case evidence to identify the intervening causal steps that produced an outcome (the mechanism), not just that outcome. This is explicitly presented as a way to make causal inferences by examining diagnostic steps inside a case.

An anecdote can be explanatory when it’s not just a story, but a mechanism-revealing trace. Anecdotes can be explanatory under particular conditions we mentioned earlier; mechanistic detail, diagnosticity (would this evidence look different if the explanation were false? Not just emotionally driven details), boundary conditions and counterfactual sensitivity (what would change if key factors changed, what alternative pathways were possible), representativeness / typicality (if the anecdote is generalizing), and triangulation across other sources such as base rates, broad datasets, and corroborating sources.

Identity-protective cognition predicts a very specific pattern: Identity-congruent anecdotes get treated as “real evidence" and Identity-incongruent anecdotes get dismissed as “exceptions,” “edge cases,” or “anecdotal.” That asymmetry is exactly what identity-protective cognition is about: selective acceptance/discounting to maintain-group-aligned belief. And narrative transportation makes this worse by reducing detection of “false notes” and increasing story-consistent beliefs. Anecdotes often serve identity-protective functions by default in contested domains, unless strong norms (methods, accountability, adversarial testing) force them into a truth-tracking role.

Just-So Stories and the Barnum Effect

“Just-so stories” and the Barnum (Forer) effect are very much in the same family as the other “weak explanation” traps we’ve been discussing. They’re two different failure modes of explanation, but they share a common engine: we often treat coherence, plausibility, and self-relevance as if they were evidential support and explanatory depth.

A “just-so story” arises when plausibility substitutes for explanatory warrant. In cognitive-science-adjacent debates (especially around evolutionary explanation), “just-so story” is a critique of explanations that are narratively plausible and coherence-producing, but weakly constrained by evidence and often post hoc (made to fit what’s already known) rather than generating risky, discriminating predictions. Gould & Lewontin’s famous critique of “the adaptationist programme” describes a pattern: atomize an organism into traits and propose an “adaptive story” for each; then treat “sounds plausible” as enough—while giving too little attention to constraints, alternatives, or non-adaptive processes. They explicitly fault the approach for (among other things) “reliance upon plausibility alone as a criterion for accepting speculative tales” and for unwillingness to seriously consider alternatives. Just-so stories work because they match several default heuristics:

Narrative completion: we hate gaps; we’ll fill missing links with a story-schema.
Teleology/agency defaults: “it evolved for X” is cognitively easy because it packages function like intention (even when the real history could be messy or plural). (This connects with teleological bias in explanation endorsement.)
Closure & seizing/freezing: once a story makes the phenomenon feel “understood,” we stop exploring alternatives unless norms force us to. (Your earlier points.)
Low falsifiability pressure: if the story doesn’t make clear, testable discriminators, it can survive indefinitely as “plausible.”

A just-so story can sound like explanation while failing the “explanatory virtue” tests:

Does it generate novel, risky predictions?
Does it identify constraints (what couldn’t have happened)?
Does it get more plausible by surviving comparisons with rival stories, not just by sounding good?

If the answer is mostly “no,” you’re in just-so territory.

The Barnum (Forer) effect occurs when vagueness + self-relevance substitutes for specific accuracy. It is the tendency to rate vague, general personality statements as highly accurate and personally tailored.

Forer’s original classroom demonstration is especially clear. He gave 39 students a “personal” typed sketch with their name on it, but all of the sketches were identical and were designed for “more nearly universal validity.” The class nevertheless endorsed them strongly. The statements were broad and hedge-friendly (“at times…”), mixed flattering observations with mild negatives, and invited self-projection (“I can see myself in that”). Students consequently gave the sketches high validation ratings and accepted many items as true.

Meehl later coined the term “Barnum effect” as a normative warning for clinical assessment. He wanted to stigmatize interpretive practices in which test-based personality descriptions appear to fit mainly because of triviality and base rates, rather than genuine discriminating validity. Barnum statements are not always framed as explanations, but they can function like explanations by answering questions such as “What kind of person am I?” or “Why do I feel this way?” Think of ridiculous personality tests or horoscopes: they can produce felt understanding with very little informational content—explanatory satisfaction without explanatory strength.

Because these descriptions are self-relevant, they can also be unusually sticky: they satisfy identity needs (“I’m seen; I make sense”), which can short-circuit scrutiny. In this respect, both just-so stories and the Barnum effect exploit psychological anchors for acceptance while evading the standards that track genuine explanatory quality. They are explanation-shaped but weakly anchored to specificity, mechanism or detail, counterfactual support, and discriminating evidence. Instead, they rely heavily on fluency, closure, self-identity, and low-cost plausibility.

Both are high-powered uncertainty reducers. Just-so stories turn “why?” into a coherent causal/function narrative. Barnum statements turn ambiguity about self into a stable-seeming self-model. This is comforting—and that comfort is part of why they bypass quality control.

Anchored Narratives

Anchored narratives provide one of the clearest meeting points between legal psychology, argumentation theory, and the themes developed above. We need stories to make sense of evidence, but stories can also seduce us into premature closure unless they are disciplined by explicit evidential and argumentative constraints. The anchored-narratives tradition asks how legal stories can retain their cognitive usefulness while being forced to answer to evidence, warrants, alternatives, and defeaters.

In real trials, fact-finders don’t just add up isolated bits of evidence. They tend to construct a narrative scenario of “what happened” that links actions, intentions, causes, and timing. This is central in the Story Model of juror decision making (Pennington & Hastie): jurors build stories to organize evidence and then evaluate which story best fits verdict categories. A key psychological implication (very aligned with uncertainty reduction): when evidence can be organized into a coherent story, decision-making often feels easier and more settled; when it can’t, uncertainty and dispute persist.

The Anchored Narratives Theory (ANT) associated with Wagenaar, van Koppen, and Crombag is often presented as both descriptive (this is how legal fact-finding actually proceeds), and normative (this is how it should proceed to avoid error): legal decisions should be based on stories that are anchored in commonsense generalizations and evidence. The anchor idea is meant to prevent “just-so” stories: a story shouldn’t merely be coherent; its crucial links should be supported by warrants/generalizations and by evidence. Bernard Jackson’s paper “‘Anchored narratives’ and the interface of law, psychology and semiotics” sits right at this interface and compares narrative approaches in legal fact-finding (including the anchored narrative tradition) while stressing that courtroom narrative is also a communicative/semiotic performance, not just a logical structure.

Argumentation theorists (especially in AI & Law) agree with the psychologists that stories are indispensable—but they worry that stories are “necessary but dangerous” unless we make the support relations explicit and contestable (so that counterevidence, exceptions, and alternative stories are handled rationally rather than rhetorically). Floris Bex, Henry Prakken, and Bart Verheij develop anchored narratives into a hybrid theory that combines a story-based layer (a causal/temporal scenario of events), and an argumentative layer that makes explicit how pieces of evidence support (or attack) elements of the story, typically via generalizations (warrants) that link evidence to story claims. Bex et al. explicitly describe ANT as requiring that legal decisions be based on stories “anchored” in common-sense generalisations; their contribution is to formalize the interaction between story construction, evidential support, and defeasible reasoning (exceptions/conflicts). A later, accessible overview is Bex’s “hybrid theory of stories and arguments applied to a case” (TOPiCS), which presents the approach as a structured way to reason about evidence by combining scenarios with argumentative support/attack relations.

Argumentation theorists like the hybrid approach because it directly targets the failure mode we have been circling. Narrative coherence produces psychological satisfaction and closure. Anchoring via explicit arguments forces epistemic accountability: What exactly supports each story element? What generalization is being relied on? What exceptions/defeaters exist? Which competing story explains the evidence at least as well? The result is a built-in antidote to “persuasiveness without explanatory depth.”

Story schemas / story schemes are how legal stories get evaluated (beyond “it sounds right”). In cognitive psychology and AI, schemas/scripts (e.g., Schank & Abelson’s classic “restaurant script”) are generalized patterns that help people interpret events and fill in missing steps. Bex explicitly places his “story schemes” in this tradition, citing work on story grammars and scripts/explanation patterns. Bex uses story schemes as generalized templates for criminal event-types (e.g., “revenge murder,” “robbery gone wrong,” etc.). These can help in two ways. They can help construct stories, suggesting plausible missing links (a double-edged sword psychologically). And they can help evaluate stories, checking whether a proposed story has the right structure (episodes, causal links, motives) and whether it instantiates a known plausible pattern. Crucially, Bex treats “good story” as not just “supported by evidence” but also “well structured and plausible”—and he tries to make that evaluative notion explicit rather than intuitive.

Walton’s work is foundational for argumentation-theoretic tools that pair well with this:

Argumentation schemes (common patterns of defeasible reasoning) plus critical questions (how to test them).
Legal argumentation and evidence: analyzing how reasoning operates in trials and how specific types of arguments can be evaluated.
Abductive reasoning (often aligned with “inference to the best explanation”): proposing a hypothesis/story that would explain the evidence, while keeping it defeasible and contestable
Dialogue models of explanation: explanation as something that happens in an interaction with rules/obligations (requests, challenges, burdens, successful transfer of understanding).

Walton also explicitly links legal reasoning to scripts/stories/anchored narratives in the context of abductive reasoning and explanation in dialogue. Where Bex gives you a “story + anchors” architecture, Walton gives you a “scheme + critical questions + dialogue” architecture for stress-testing the anchors (and sometimes the story itself).

This framework directly connects narrative satisfaction to the earlier distinction between psychological closure and epistemic quality.

Explanation Quality vs. Narrative Satisfaction

Narrative virtues (coherence, vividness, closure, moral intelligibility) can create “this makes sense” feelings.
Explanatory virtues (relevance, mechanism/detail where appropriate, counterfactual support, robustness, evidential fit, sensitivity to defeaters) are what make the account actually strong.

Anchored narratives frameworks are basically an attempt to keep the cognitive and communicative power of narrative while forcing the story to “pay rent” in evidential currency.

Uncertainty reduction and “seizing/freezing”

The story model suggests decision makers often settle by selecting a story that organizes the evidence and fits a verdict category. "Seizing and freezing” worry is exactly what critics worry about: once a story provides closure, we may stop searching, discount anomalies, and interpret new evidence through the settled plot. The anchored-narratives/hybrid approach tries to prevent freezing by:
- making anchors explicit (so they can be attacked),
- requiring confrontation with exceptions/conflicts,
- and structurally supporting comparison among competing stories.
Illusion of explanatory depth (IOED) in legal stories

Story coherence can mimic depth. A story with a familiar schema (“that’s how these crimes usually go”) can feel mechanistically rich even if the crucial causal links are handwaved. Story schemes are powerful here in two opposite ways:
- Risk: they can encourage “schema completion” (filling gaps with defaults), inflating confidence.
- Benefit: used critically, they provide a checklist of missing elements and weak transitions that should be anchored rather than assumed.
Identity-protective narrative

Legal stories often carry implicit values (blameworthiness, victimhood, motive, character). Bex explicitly explores values as part of why stories persuade. So the “identity protection” dynamic you mentioned fits: story choice can be shaped by moral and group-aligned interpretations, not just evidence. The hybrid/scheme approach helps by relocating debate from “whose story feels right” to “which story’s anchors survive critical scrutiny.”

This can be operationalized as a practical synthesis of epistemic and psychological evaluation:

Story quality (narrative plausibility)	Anchor quality (evidential / argumentative support)
Internal coherence (no contradictions) Completeness (key episodes accounted for) Temporal/causal intelligibility (why transitions happen) Fit to an appropriate story scheme (if applicable)	For each key story step: what evidence supports it? What generalization/warrant links evidence → story claim? What exceptions/defeaters apply? What alternative story explains the same evidence (abduction competition)?

Part IV — Explanation in a Social Environment

Most of what we “understand” is socially mediated. We inherit explanations from teachers, institutions, media ecosystems, disciplines, and communities. Those social structures can improve inquiry through division of cognitive labor, error-correction, and adversarial testing, or deform it through echo chambers, propaganda, identity-protective cognition, and the strategic production of “clarity.” Social epistemology studies how agents pursue truth with the help of—and sometimes in the face of—other people, including inquiry by groups and institutions. Explanation is therefore not merely a relation between a theory and a phenomenon. It is also a product of social processes, a tool for coordinating groups, and an object of trust. This shifts attention toward a second-order set of questions about who explains, who is trusted, and which social arrangements make explanation reliable.

Question	Primary concern
Who gets to explain? Who is heard as an explainer?	Power and testimony
Which social structures make explanation reliable?	Institutions and norms
Which structures make “fake understanding” feel irresistible?	Clarity, propaganda, and echo chambers
Can groups understand? Can groups explain?	Collective epistemology

Engineered clarity and stopping rules

In "Seductions of Clarity", Nguyen describes explanation as a felt stopping rule. He treats clarity as an epistemic feeling that we use as a thought-terminating heuristic—a signal that inquiry is “done.” In his abstract, he explicitly says the feeling of clarity is associated with understanding and can be exploited: manipulators can “imbue a belief system with an exaggerated sense of clarity” to make us stop checking it. The feeling of clarity is not merely private; it can be engineered socially. Conspiracy theories offer a sweeping, compressive explanation that “connects everything” and gives users the sense they can generate explanations easily (high cognitive facility). Nguyen uses conspiracy theories as a main case study of seductively clear systems. Bureaucratic/quantified value systems produce crisp metrics that feel like understanding even when they omit morally or causally important dimensions. (This is also in Nguyen’s abstract as a case study.) The phenomenology of understanding can be decoupled from the epistemic virtues of explanation, and whole communities can be stabilized around that decoupling.

Groups, echo chambers, and epistemic dependence

Boyd argues that epistemic bubbles and echo chambers should be treated as epistemically pernicious groups, and that a core mechanism is what he calls groupstrapping: members treat the group itself as a testimonial source; agreement raises members’ confidence; that confidence then feeds back into attributing more credibility to the group than is warranted. In his paper’s opening, Boyd highlights two core harms of these groups; unwarranted confidence, and stagnation of inquiry (fewer prompts to explore or revise). Groupstrapping distorts explanations because it encourages what you might call closed-loop explanation; the community’s preferred explanation is treated as confirmed because the community repeats it with confidence, while outside sources are discounted, so the explanation never faces serious comparative tests. This is essentially confirmatory selection and overreliance on the in-group as an epistemic diagnosis.

Nguyen’s “Echo Chambers and Epistemic Bubbles” provides a key distinction that social epistemologists use constantly now. Epistemic bubbles is a situation where relevant voices are left out (often accidentally). Echo chambers are when outside voices are actively discredited, typically via trust-manipulation. Mere exposure to evidence can shatter an epistemic bubble, but may actually reinforce an echo chamber. Echo chambers don’t just block facts; they pre-structure what counts as an acceptable explanation and who counts as an acceptable explainer. This makes explanations “sticky” in the seizing/freezing sense: alternative explanations aren’t merely missing; they’re framed as corrupt, naive, or enemy propaganda.

Testimony, expertise, and epistemic dependence show how explanation is often “outsourced.” A lot of social epistemology begins with a sober fact: you can’t personally verify most things you believe. Modern knowledge is built on epistemic dependence (especially on experts and institutions). Hardwig’s classic “Epistemic Dependence” is one of the canonical statements of that problem: individuals rationally rely on expert communities because no individual can possess all the supporting evidence. Goldman’s Knowledge in a Social World develops “veritistic” social epistemology: evaluate social practices/institutions by how well they promote knowledge/truth (media norms, science, law, democracy). This creates a deep vulnerability. Because explanation is often received via testimony, the quality of explanation is entangled with the quality of social trust networks. “Good explanation” can’t just mean internal features (coherence, simplicity); it must include social-epistemic features: source reliability, diversity of critical scrutiny, independence of evidence, institutional incentives. Nguyen’s echo chambers paper explicitly links the danger to “manipulation of trust” and emphasizes epistemic interdependence as unavoidable—echo chambers “prey on our epistemic interdependence.”

Epistemic injustice and who gets to explain

Epistemic injustice is also relevant; it describes who gets to explain, who gets believed, and who gets understanding-resources. Fricker’s Epistemic Injustice identifies Testimonial injustice (credibility is unfairly deflated due to prejudice) and Hermeneutical injustice (gaps in shared interpretive resources unfairly disadvantage people in making sense of their experiences). This matters for explanation because explanations aren’t only evaluated for truth; they’re also gateways to social standing (“knowers,” “experts,” “credible witnesses”). Epistemic injustice distorts explanation in at least three ways:

Which explanations enter the public space (some people aren’t heard).
Which explanations are treated as plausible (credibility economy is biased).
Which phenomena can even be explained (lack of shared concepts makes certain experiences unintelligible or easily misexplained).

Communities can maintain identity by maintaining interpretive dominance—their favored explanatory frames crowd out rival hermeneutical resources.

Network effects and local rationality

Network epistemology adds another piece: it helps explain how false explanations spread even among reasonable agents. O’Connor & Weatherall’s The Misinformation Age argues that false beliefs can spread and persist through social network dynamics; what you believe depends heavily on who you interact with, and network structures can stabilize misinformation even without everyone being individually irrational. This supports a central social-epistemology theme: explanations are often “ecologically rational” within a local network (they fit what your trusted peers say and what you’re exposed to), yet globally unreliable. That meshes with Boyd’s stagnation-of-inquiry concern and Nguyen’s warning: if the system gives you easy explanatory facility, you feel done—whether or not the system is truth-tracking.

Traditional criteria (fit with evidence, causal/mechanistic adequacy, simplicity, unification, counterfactual robustness) are still relevant. Social epistemology adds higher-order and social criteria—especially important when explanation is socially transmitted.

Criterion family	Questions to ask
Source-structure criteria	Independence: are the supporting sources genuinely independent or “copying” each other (closed loops)? Diversity: is the explanation tested against varied perspectives and counterevidence? Incentives: do institutions reward correction or reward clarity/theater/loyalty?
Trust-health criteria	Are trust relations calibrated, or manipulated (echo chamber dynamics)? Does the community treat disagreement as evidence of “enemy status” (Nguyen’s echo chamber mechanism)?
Phenomenology-control criteria	Does the system generate clarity too cheaply? Does it function as a “thought-terminating heuristic” (Nguyen’s diagnosis)?
Justice and inclusion criteria	Are some explainers systematically discredited (testimonial injustice)? Are there gaps in interpretive resources that make some experiences “unexplainable” or easily misexplained (hermeneutical injustice)?

Nguyen and Boyd bring into focus that explanation is not only a cognitive achievement but also a social technology. Embedded in communities with strong critical norms, it can be epistemically virtuous: it can produce real understanding and support reliable prediction or intervention. Embedded in echo chambers or groupstrapping dynamics, the same basic technology can be epistemically weaponized, producing intense felt clarity and closure while insulating itself from correction.

Environmental Conditions and Epistemic Reliability

Social epistemologists tend to grade explanations based on two criteria. The explanation as a piece of content (does it actually explain—evidence, mechanism, counterfactual support, etc.); and The explanation as a socially produced artifact (what kind of epistemic environment and institutional pipeline generated, selected, and stabilized it). The second criterion is the distinctive contribution of social epistemology: a persuasive, coherent explanation can be epistemically bad if it is the output of a distorted social process—especially in what Nguyen calls hostile epistemic environments designed to exploit our vulnerabilities. Below is a framework that captures how many social epistemologists would rank explanations. It shows how it connects to hostile epistemology, groupstrapping, and engineered clarity.

Evaluation level	What to examine
Internal explanatory virtues (content-level)	These are closer to philosophy of science / traditional epistemology, and social epistemologists usually keep them: Fit to evidence (and not just cherry-picked evidence) Relevance / difference-making (it tracks real dependence) Mechanistic depth (where appropriate), not just labels Counterfactual support (what would change if X changed) Scope and robustness (works across cases; stable under new data) Non-ad-hocness (doesn’t rely on handwavy “just so” patches) Social epistemology doesn’t reject these; it asks: why are people so often satisfied without them?
Social-epistemic virtues (environment-level)	These are “higher-order” features: the reliability of the social process producing the explanation. This is the big shift: you don’t just ask “Is the story coherent?” You ask “Was it produced and stabilized in a way that tends to generate true, well-supported explanations?” This is exactly Goldman’s “veritistic” social epistemology: evaluate social practices by how truth-conducive they are (how they promote true belief / knowledge).

Evaluation level

What to examine

Internal explanatory virtues (content-level)

These are closer to philosophy of science / traditional epistemology, and social epistemologists usually keep them:

Fit to evidence (and not just cherry-picked evidence)
Relevance / difference-making (it tracks real dependence)
Mechanistic depth (where appropriate), not just labels
Counterfactual support (what would change if X changed)
Scope and robustness (works across cases; stable under new data)
Non-ad-hocness (doesn’t rely on handwavy “just so” patches)

Social epistemology doesn’t reject these; it asks: why are people so often satisfied without them?

Social-epistemic virtues (environment-level)

These are “higher-order” features: the reliability of the social process producing the explanation. This is the big shift: you don’t just ask “Is the story coherent?” You ask “Was it produced and stabilized in a way that tends to generate true, well-supported explanations?” This is exactly Goldman’s “veritistic” social epistemology: evaluate social practices by how truth-conducive they are (how they promote true belief / knowledge).

Nguyen defines hostile epistemology as the study of how environmental features exploit our cognitive vulnerabilities, especially vulnerabilities that come from being finite, rushed, and dependent on shortcuts and trust. That directly changes explanation-evaluation at the narrow propositional level. Nguyen’s Seductions of Clarity argues that the feeling of clarity is a powerful “stop rule” (it tempts us to end inquiry), and that manipulators can fake clarity by designing systems that generate high cognitive facility and easy answer-production. So a social epistemologist will often discount an explanation’s apparent clarity if it arises from environments optimized for virality, outrage, identity signaling, “one weird trick” simplification, or conspiracy-style totalizing coherence. In a hostile environment, the same internally coherent explanation gets a lower grade unless it comes with unusually strong anchoring and transparency, because the environment is known to manufacture “felt understanding.”

Boyd’s “epistemically pernicious groups” idea (bubbles/echo chambers as group-level pathologies) diagnoses groupstrapping as group agreement boosts members’ confidence, which then boosts the group’s perceived authority, which further increases reliance on the group, producing a self-reinforcing loop. Explanations that are primarily supported by in-group repetition, circular citation networks, narrow “approved” sources, and systematic discrediting of outsiders get downgraded even if they “explain everything,” because the support structure is epistemically non-independent and self-sealing.

Longino’s influence in social epistemology of science is huge: objectivity is not just an individual virtue; it’s a property of communities with the right norms for criticism. A widely cited summary gives four conditions for “transformative criticism”: venues for criticism, uptake/responsiveness, public standards, and tempered equality of intellectual authority. In this frame, an explanation earns a higher grade if it has been produced in (or exposed to) communities that allow genuine critique, actually respond by revising beliefs/models, make standards explicit, and don’t let authority be monopolized in ways that shield a view from challenge. This directly counters the Nguyen/Boyd pathologies: it’s the anti-hostile and anti-groupstrapping epistemic ecology.

Sperber’s “epistemic vigilance” program argues that humans have cognitive mechanisms for evaluating testimony and guarding against misinformation because reliance on other people is unavoidable. Nguyen’s hostile-epistemology framework highlights the complementary danger: those vigilance mechanisms can be overloaded or hacked by the environment. Social epistemologists therefore increasingly treat a “good explanation” as one that does not demand heroic vigilance from ordinary agents. They value explanations that come with traceable sources, accessible checks, clear uncertainty markers, and institutional or accountability scaffolding.

Part V — A Practical Evaluation Framework

General Summary

A practical evaluation framework has three layers: the quality of the explanatory content, the psychology of the evaluator, and the social process that produced and stabilized the explanation.

Layer	Evaluation
1. Explanatory content	A: Mechanistic/difference-making, counterfactuals clear, boundaries stated, rivals addressed, evidence strong. C: Coherent story, some evidence, but thin mechanism, few discriminators, ignores key alternatives. F: “Explains everything,” unfalsifiable, mostly vibes/labels, no real dependence relation.
2. Evaluator psychology	Check whether coherence, fluency, closure, narrative completion, identity fit, intolerance of ambiguity, or an illusion of explanatory depth is substituting for evidential support. The key distinction is between an explanation that reduces uncertainty by tracking real structure and one that reduces uncertainty mainly by producing a feeling of understanding.
3. Social production	Key questions (with typical social-epistemology inspiration in parentheses): Source independence: are the “multiple sources” actually independent, or a citation echo? (Boyd-style red flag) Adversarial exposure: has it faced informed criticism from outside its home coalition? (Longino) Uptake: does the community revise in response to critique, or does it immunize? (Longino / Nguyen echo-chamber logic) Transparency: can you see the chain from evidence → claim, or is it “trust me / trust us”? (Goldman veritism) Incentives: are there incentives for being right (correction, replication, reputational cost for error) or for being compelling/loyal/outrage-inducing? (veritistic/institutional analysis) Hostility check: does the environment systematically exploit clarity, outrage, identity, or information overload? (Nguyen)

Layer

Evaluation

1. Explanatory content

A: Mechanistic/difference-making, counterfactuals clear, boundaries stated, rivals addressed, evidence strong. C: Coherent story, some evidence, but thin mechanism, few discriminators, ignores key alternatives. F: “Explains everything,” unfalsifiable, mostly vibes/labels, no real dependence relation.

2. Evaluator psychology

Check whether coherence, fluency, closure, narrative completion, identity fit, intolerance of ambiguity, or an illusion of explanatory depth is substituting for evidential support. The key distinction is between an explanation that reduces uncertainty by tracking real structure and one that reduces uncertainty mainly by producing a feeling of understanding.

3. Social production

Key questions (with typical social-epistemology inspiration in parentheses):

Source independence: are the “multiple sources” actually independent, or a citation echo? (Boyd-style red flag)
Adversarial exposure: has it faced informed criticism from outside its home coalition? (Longino)
Uptake: does the community revise in response to critique, or does it immunize? (Longino / Nguyen echo-chamber logic)
Transparency: can you see the chain from evidence → claim, or is it “trust me / trust us”? (Goldman veritism)
Incentives: are there incentives for being right (correction, replication, reputational cost for error) or for being compelling/loyal/outrage-inducing? (veritistic/institutional analysis)
Hostility check: does the environment systematically exploit clarity, outrage, identity, or information overload? (Nguyen)

Overall grade is a combination: a content-A can be socially fragile (e.g., correct claim arrived at by conspiracy-style methods), and a content-C can be socially valuable if it’s part of a healthy error-correcting pipeline.