Wrong Direction (Reverse Causation Fallacy)¶

Catch the error of assigning a causal arrow A→B when it actually runs B→A on the strength of a symmetric association alone, by refusing any oriented claim until temporal precedence or interventional response fixes the direction.

Core Idea¶

The wrong-direction fallacy — also called reverse causation or causal reversal, catalogued in informal logic as one reading of non causa pro causa — is the reasoning error of assigning a causal arrow from A to B when the actual causal direction runs from B to A, on the basis of an observed association alone. The error is a direction error, not an existence error: there is genuine causation in the system; the reasoner has simply oriented the arrow backwards. This distinguishes it from the other two classic interpretations of a correlation (common cause, where a third variable drives both; and chance, where there is no causal structure at all): wrong-direction concedes causation and concedes the correlation, but gets the which-way wrong. The structural mechanism of the error is that correlation is symmetric — A correlates with B exactly as strongly as B correlates with A — so an observed association carries no inherent signal about which variable is upstream without additional evidence. The additional evidence that can fix direction is of two kinds: temporal precedence (if B reliably precedes A in time, B is the more plausible upstream cause, though even this is defeasible) and interventional response (if intervening on A — holding it fixed or randomizing it — does not move B, while intervening on B does move A, the causal arrow runs B→A). Absent such evidence, the practitioner who concludes A→B from a positive correlation has committed the wrong-direction fallacy, and any policy built on A causes B will be ineffective or actively counterproductive. Canonical examples in argumentation and epidemiology include: the inference that diet soda causes obesity because obese people consume more diet soda (the obesity drives the soda choice, not the reverse); the inference that police presence causes crime because crime-heavy neighborhoods have more police (deployment follows crime, not the reverse); the inference that hospitalization causes death because death rates are higher in hospitals. In each case, the correlational evidence is real and the causation is real but the direction of inference is inverted.

Structural Signature¶

Sig role-phrases:

the observed association — a symmetric correlation between A and B, which carries no inherent which-way signal (A correlates with B exactly as strongly as B with A)
the inferred arrow — the asserted causal claim A→B drawn from that association alone
the true arrow — the actual direction B→A (or bidirectional): a direction error, not an existence error — causation and correlation are both conceded
the forced disjunction — the exhaustive menu the association cannot adjudicate among (A→B, B→A, common cause, chance), with wrong-direction staking one cell as the standing trap
the temporal-precedence test — if B reliably precedes A in time, B is the more plausible upstream cause (defeasible)
the interventional-response test — moving the alleged cause should move the alleged effect while moving the effect leaves the cause unmoved; which intervention bites fixes the arrow
the suspended verdict — where neither asymmetry test has run, the disciplined reading is "direction unknown," not a confident orientation
the asymmetric-stakes warning — a policy built on A→B when truly B→A is not merely inert but counterproductive, so the unexamined-direction case is owed the probes most urgently

What It Is Not¶

Not the bare "correlation is not causation" warning. That cruder lesson cautions against inferring any causal link from association, and so leaves untouched the reasoner who concedes a real causal relationship and still orients the arrow backwards. Wrong-direction picks out exactly that residual error — it grants the correlation and grants the causation, faulting only the which-way.
Not an existence error. The reasoner has not hallucinated structure into noise; there is genuine causation in the system. The fault is one of orientation, not of asserting a link where none exists. This is what separates it from the chance reading of a correlation, where there is no causal structure at all.
Not confounding by a common cause. A third variable driving both A and B is a different escape-hatch entirely: confounding leaves neither an arrow between A and B, whereas wrong-direction is a genuine A–B causal arrow pointing the wrong way. The remedy differs too — adjust for the third variable versus fix the orientation.
Not the post hoc fallacy. Inferring "A caused B because A came first" reads causation out of temporal succession alone; wrong-direction reads the wrong arrow out of a symmetric association whose temporal order may be unclear or itself reversed. Where post hoc over-trusts sequence, wrong-direction has typically ignored sequence (and intervention) altogether.
Not as harmless as admitting ignorance. Confidently asserting the reversed arrow is worse than conceding "direction unknown," not merely as bad: a policy built on "A causes B" when in truth B causes A is not inert but counterproductive, acting on the cause to move an effect that is actually upstream of it.
Not settled by temporal precedence alone. That B reliably precedes A makes B the more plausible upstream cause, but the test is defeasible — anticipation, reverse-running mechanisms, and measurement lag can all break it. Temporal order shifts the arrow's probability; only interventional response, where available, more firmly fixes it.

Scope of Application¶

Because the wrong-direction fallacy is a named diagnostic — a standing trap plus the two asymmetry tests (temporal precedence, interventional response) that escape it — rather than a causal mechanism, it applies wherever its minimal precondition holds: an oriented causal claim drawn from a symmetric association. That precondition is met across the whole empirical-science substrate, so the fields below are genuine literal uses of the identical diagnostic, not metaphor; what travels is the named trap, while the more general insight it crystallizes belongs to the parent primes correlation and causality.

Informal logic and argumentation theory — the catalog home, where it is the stock fallacy filed under non causa pro causa / "wrong direction" (Walton, Govier) and taught as one of the three escape-hatches a correlation cannot rule out.
Epidemiology — the bedrock motivation for prospective cohort designs and Mendelian randomization, instruments chosen precisely to fix the arrow that cross-sectional association leaves underdetermined.
Economics and policy analysis — a recurring misdiagnosis (does loose money cause inflation or inflation loose money? does welfare cause unemployment or unemployment welfare uptake?), where a policy built on the reversed arrow is counterproductive, not merely inert.
Clinical reasoning — the standing error of confusing a symptom-of-disease with a cause-of-disease when the two merely co-occur.
Quantitative social science — psychology, sociology, and educational research, where the same residual error and the same two remedies apply unchanged to observational findings.
Network analysis and ML interpretability — where attributing a directional influence from an observed co-variation invites the identical reversal, and the asymmetry probes are the same escape.

Clarity¶

Naming the wrong-direction fallacy makes legible that an observed association underdetermines causal direction even after causation has been granted. The cruder lesson — "correlation is not causation" — warns only against inferring any causal link at all, and so leaves untouched the reasoner who concedes a real causal relationship and still confidently orients the arrow. Wrong-direction picks out exactly that residual error: it concedes the correlation and concedes the causation, faulting only the which-way. The reason it can do so is the symmetry of correlation itself — A correlates with B precisely as strongly as B correlates with A — so the association, however strong, carries no internal signal about which variable is upstream. The sharp question the label forces before any causal conclusion is drawn is therefore not "is there a relationship?" but "which way does the arrow run, and what besides the correlation tells me so?"

The fallacy also sharpens the taxonomy of what a bare correlation cannot rule out, by separating three claims an undisciplined inference fuses. Once an association is seen, the candidate explanations are forced apart: A→B, B→A, a common cause driving both, or chance — and wrong-direction names the specific failure of asserting the first when the truth is the second, distinct from the existence error of reading structure into noise and from the third-variable error of confounding. Holding these apart matters because they demand different remedies and carry different stakes: confidently inferring the wrong direction is worse than admitting the direction is unknown, because policy built on "A causes B" when in fact B causes A is not merely inert but can be counterproductive. The label simultaneously tells the practitioner what evidence would settle direction — temporal precedence (does the supposed cause reliably precede the supposed effect?) and interventional response (does moving the alleged cause move the alleged effect, while moving the alleged effect does not move the cause?) — converting a flat "X is linked to Y" into a set of asymmetry tests whose answers fix the arrow.

Manages Complexity¶

When a practitioner meets a fresh "X is linked to Y" finding, the interpretive space is wide and ill-defined: an association can be spurious, real-but-confounded, real-and-direct, or real-and-reversed, and within "real" the arrow can run either way or both — so the question "what does this correlation mean?" invites an open-ended hunt that, undisciplined, tends to collapse straight to the most readable story (the cause that comes first to mind becomes the cause). Naming the wrong-direction fallacy is one cut that tames that space. It does not try to settle the whole interpretation; it forces the otherwise-fused readings of a correlation apart into a fixed, exhaustive menu — A→B, B→A, common cause, chance — and stakes out one specific cell of that menu as a standing trap. The boundless question "is my causal story right?" is thereby compressed into a finite disjunction the data must adjudicate among, and the reasoner who has not consciously placed a finding into one of those four cells knows exactly what work remains. The sprawl of possible misreadings collapses to a small typology, and the wrong-direction label marks the one cell where causation and correlation are both conceded and only the orientation is wrong — the residual error that the cruder "correlation is not causation" leaves entirely untouched.

The deeper compression is that direction itself, which feels like it should be visible in a strong enough association, is shown to be invisible there in principle — and that one structural fact replaces case-by-case agonizing with a uniform procedure. Because correlation is symmetric, A correlating with B exactly as strongly as B with A, no amount of correlational evidence carries a which-way signal; the practitioner therefore never has to ask, for each new finding, whether this correlation happens to reveal direction, because the answer is always no. What the analyst tracks instead reduces to two asymmetry probes that the symmetric association cannot supply on its own: temporal precedence (does the supposed cause reliably precede the supposed effect?) and interventional response (does moving the alleged cause move the alleged effect, while moving the alleged effect leaves the alleged cause unmoved?). The arrow reads off the result of those probes — and where neither has been run, the disciplined verdict is "direction unknown," not a confident orientation. The branch structure carries the stakes: confidently asserting the wrong arrow is not merely as bad as admitting ignorance but worse, because a policy built on "A causes B" when in truth B causes A is not inert but counterproductive — so the typology does not merely sort errors, it ranks them, telling the practitioner that the unexamined-direction case is the one whose downside is asymmetric and therefore the one the probes are most urgently owed.

Abstract Reasoning¶

The wrong-direction fallacy licenses a set of moves in causal inference from observational data, all anchored to the fact that a symmetric correlation carries no which-way signal even after causation is granted. The signature boundary-drawing move is to refuse any oriented causal claim from association alone: before concluding "A causes B," the analyst asks "which way does the arrow run, and what besides the correlation tells me so?" — and where neither asymmetry test has been run, the disciplined verdict is direction unknown, not a confident orientation. This is the regime edge the concept enforces: correlation entitles the reasoner to "there is a relationship," and to nothing about its direction, so the analyst reasons from the absence of asymmetry evidence to a suspension of the directional claim rather than defaulting to the reading that comes first to mind.

The two interventional/predictive tests are the moves that actually fix the arrow, and each generates a checkable prediction. The interventional test reasons by manipulation asymmetry: if A→B, then holding A fixed or randomizing it should move B, while intervening on B should leave A unmoved; the reversed pattern — intervening on A does nothing to B, intervening on B moves A — diagnoses B→A. So the analyst predicts the result of an experiment before running it and reads the direction off which intervention bites: a randomized trial assigning diet soda versus sugar soda to non-obese subjects that shows no obesity effect refutes the soda→obesity arrow and locates the cause upstream in the obesity. The temporal test reasons by precedence: if B reliably precedes A in time, B is the more plausible upstream cause, so prospective and temporally-ordered data can discriminate direction where cross-sectional data cannot — and the analyst predicts that a cohort design will settle what a snapshot leaves underdetermined. Both predictions are defeasible (temporal precedence especially), so the analyst treats them as evidence that shifts the arrow's probability, not as proof, and prefers the interventional signal where the two are available.

A diagnostic move supplies a cheaper, prior check: read the candidate mechanism in each direction and ask which is more plausible. Confronting "police presence causes crime," the analyst reasons that deployment plausibly follows crime (resources sent where crime is high), which flips the proposed arrow on mechanistic grounds before any data are gathered; confronting "hospitalization causes death," that the sick are sent to hospitals, so illness is upstream of both. This generation of the reverse story — narrating how B could produce A — is the move that surfaces wrong-direction as a live possibility in the first place, converting an association into a competition between two oriented mechanisms. Finally, an interventionist move attends to the asymmetric stakes the fallacy carries: because a policy built on "A causes B" when in truth B causes A is not merely inert but counterproductive, the analyst reasons that the unexamined-direction case is the one whose downside is worst, and so treats the asymmetry probes as most urgently owed exactly where a costly intervention is about to be staked on the unverified arrow — directing scarce verification effort to the inferences whose reversal would do active harm.

Knowledge Transfer¶

The wrong-direction fallacy is a named diagnostic — a standing trap plus the two asymmetry tests that escape it — rather than a causal mechanism of its own, and its precondition is minimal: any oriented causal claim drawn from a symmetric association. That precondition is met across the entire empirical-science substrate, so the diagnostic transfers literally and as the same thing wherever correlational data are interpreted, not by analogy. In its catalog home, informal logic and argumentation theory (Walton, Govier), it is the stock fallacy filed under non causa pro causa / "wrong direction." But the place it actually bites is the observational sciences, and there the construct carries intact: in epidemiology it is the bedrock motivation for prospective cohort designs and Mendelian randomization (instruments chosen precisely to fix the arrow); in economics and policy it is the recurring misdiagnosis (does loose money cause inflation or inflation loose money? does welfare cause unemployment or unemployment welfare uptake?); in clinical reasoning it is confusing symptom-of-disease with cause-of-disease; in psychology, sociology, educational research, network analysis, and ML interpretability the same residual error and the same two remedies — temporal precedence and interventional response — apply unchanged. Across all of these the diagnostic question ("which way does the arrow run, and what besides the correlation tells me so?"), the suspended verdict where neither asymmetry test has run ("direction unknown"), and the asymmetric-stakes warning (a policy built on A→B when truly B→A is not inert but counterproductive) transfer without translation, because the structural fact underneath — correlation is symmetric and so carries no which-way signal — is substrate-independent.

This wide reach is exactly why the entry is not a prime, and the honest accounting is case (B): the thing that genuinely recurs across all those sciences is not "the wrong-direction fallacy" as a named rhetorical category but the more general structural insight that a correlation underdetermines causal direction, which only temporal or interventional evidence can settle — and that insight is already carried by the parent primes correlation (whose treatment explicitly lists reverse causation as one of correlation's escape-hatches, alongside common-cause and chance) and causality. The general pattern travels; the home-bound cargo is the fallacy framing — the label, its place in the three-escape-hatch taxonomy of informal logic, the non causa pro causa lineage, the rhetorical packaging as a "named pattern of bad reasoning." Those belong to fallacy taxonomy and do not themselves do work outside it. So the cross-domain lesson should be stated as the parent prime (correlation does not fix direction; demand precedence or intervention), with "wrong direction" understood as the argumentation-domain crystallization of that insight — the name a rhetorician gives the trap, useful for cataloguing and teaching, but adding no structural content beyond what correlation and causality already encode. The boundary to mark is therefore not instrument-versus-metaphor (the diagnostic is no mere metaphor anywhere its precondition holds) but instrument-naming versus mechanism-ownership: the cross-domain force is owned by the parent, the rhetoric-specific label is what is local. See Structural Core vs. Domain Accent.

Examples¶

Canonical¶

A standing example in criminology is the observation that neighborhoods with more police have more crime. Read naively, the correlation invites the arrow police→crime: perhaps a heavy police presence provokes or over-criminalizes residents. But the correlation is symmetric and settles nothing; the mechanism runs the other way. Police are deployed to high-crime areas — crime drives the deployment, so the true arrow is crime→police. The diagnostic tell is mechanistic plausibility: staffing decisions obviously respond to crime rates. And the interventional test confirms it and flips the policy conclusion: randomized hot-spot policing experiments, which add officers to randomly chosen high-crime locations rather than letting deployment follow crime, generally find that added presence reduces crime there — the reverse of what the raw correlation suggested. A policy of withdrawing police to lower crime would have been actively counterproductive.

Mapped back: The police–crime correlation is the observed association, symmetric and directionless; police→crime is the inferred arrow and crime→police the true arrow. Randomizing added patrols is the interventional-response test that fixes the arrow, and "withdrawing police would backfire" is the asymmetric-stakes warning made concrete.

Applied / In Practice¶

Epidemiology built an entire method — Mendelian randomization — around escaping this fallacy. Observational studies repeatedly find biomarkers correlated with disease and are tempted to read the biomarker as the cause. C-reactive protein (CRP), a marker of inflammation, is strongly associated with coronary heart disease, which suggested that lowering CRP might prevent heart attacks. But the arrow could run the other way — early disease raising CRP — or both could share a cause. Mendelian randomization fixes direction by using genetic variants that raise lifelong CRP as a natural randomizer, since genes are fixed at conception and precede disease. Large collaborative studies found that people genetically predisposed to higher CRP had no higher heart-disease risk, so CRP is a downstream marker, not a cause. The correlation was real; the arrow did not run from CRP to disease.

Mapped back: The CRP–heart-disease correlation is the observed association; CRP→disease is the inferred arrow. Genes being fixed at conception supply the temporal-precedence test, and using genetic variants as a randomizer is the interventional-response test — together fixing the arrow and averting acting on a CRP-lowering drug that the asymmetric-stakes warning flags as futile.

Structural Tensions¶

T1: Direction error versus existence error (the residual the cruder lesson misses). "Correlation is not causation" warns against inferring any causal link, and so leaves entirely untouched the reasoner who concedes a real causal relationship and still orients the arrow backwards. The wrong-direction fallacy picks out exactly that residual: it grants the correlation and grants the causation, faulting only the which-way. The tension is that the popular slogan feels like the complete lesson while covering only one of a correlation's four readings — its blanket skepticism about causation cannot catch an oriented error made after causation is correctly granted. Guarding against existence errors and guarding against direction errors are different disciplines, and mastering the first gives no protection against the second. Diagnostic: Is the doubt about whether there is a causal link at all (existence) or about which way a conceded causal link runs (direction)?

T2: Symmetric correlation versus the readable-story pull (the association carries no which-way signal). The structural fact is that correlation is symmetric — A correlates with B exactly as strongly as B with A — so an association, however strong, contains no internal signal about direction. The tension is that human inference abhors this vacuum and fills it with whichever mechanism comes first to mind, so the reasoner routinely feels the arrow's direction is given by a strong enough correlation when it is in principle invisible there. The stronger and cleaner the association, the more confident (and unwarranted) the felt orientation. The concept's discipline is to insist the correlation entitles the reasoner to "there is a relationship" and to nothing about its direction, against a cognitive default that supplies a direction for free. Diagnostic: Is the asserted arrow supported by evidence beyond the correlation, or is a symmetric association being read as if it pointed one way?

T3: Temporal precedence versus interventional response (two tests of unequal strength). The two escapes from the fallacy differ in reliability. Temporal precedence — the supposed cause reliably precedes the effect — is a cheap and often available check, but it is defeasible: anticipation, reverse-running mechanisms, and measurement lag can all break it. Interventional response — move the alleged cause and watch the effect, move the effect and watch the cause stay put — more firmly fixes the arrow but is often expensive or impossible to run. The tension is that the accessible test is the weaker one and the decisive test is the costly one, so a practitioner frequently must choose between a defeasible signal they can get and a firm signal they cannot. Treating precedence as proof over-trusts the cheap test; demanding intervention everywhere is often infeasible. Diagnostic: Does the direction rest on temporal precedence (defeasible — check for anticipation and lag) or on an intervention that actually moved one variable and not the other (firmer)?

T4: Asymmetric stakes versus the suspended verdict (confident reversal is worse than "unknown," but decisions press). The fallacy carries an asymmetry the concept insists on: confidently asserting the reversed arrow is not merely as bad as admitting ignorance but worse, because a policy built on "A causes B" when truly B causes A is not inert but counterproductive. The disciplined response where neither asymmetry test has run is therefore "direction unknown," not a confident orientation. The tension is that decision-makers are under pressure to act, and "direction unknown" offers no lever, so the discipline that correctly refuses a false arrow also withholds the answer a policy needs — pushing the urgency onto getting the asymmetry evidence exactly where a costly intervention is about to be staked on an unverified arrow. Diagnostic: Before acting on the arrow, has an asymmetry test fixed its direction — and if not, is the intervention one whose reversal would be actively counterproductive (making the suspended verdict the right answer)?

T5: Autonomy versus reduction (a named fallacy or the correlation/causality parents). The wrong-direction fallacy is a named diagnostic — a standing trap plus two asymmetry tests — that applies literally wherever an oriented causal claim is drawn from a symmetric association, which is the whole empirical-science substrate; so within epidemiology, economics, clinical reasoning, and quantitative social science it transfers as the same thing, not by analogy. But what genuinely recurs is not the rhetorical category; it is the more general insight that a correlation underdetermines causal direction, which only temporal or interventional evidence can settle — already carried by the parents correlation (which lists reverse causation among its escape-hatches) and causality. The home-bound cargo is the fallacy framing: the label, its place in the three-escape-hatch taxonomy, the non causa pro causa lineage. The tension is between a useful teaching label and the recognition that its cross-domain force is owned by the parents. Diagnostic: Resolve toward correlation/causality (a correlation cannot fix direction; demand precedence or intervention) when stating the cross-domain lesson; toward the wrong-direction fallacy when cataloguing or teaching the trap as a named pattern of bad reasoning.

Structural–Framed Character¶

The wrong-direction fallacy sits at the framed-leaning position on the structural–framed spectrum — clearly on the framed side, as a normatively charged, taxonomy-born fallacy label like ad hominem, but pulled well back from the framed pole by a genuinely substrate-independent structural core and by a diagnostic that transfers as literal recognition, not analogy, across the whole empirical-science substrate. The framed marks are real. On evaluative_weight it scores high: to call an inference "the wrong-direction fallacy" is to convict it — the label is a normative verdict that a piece of causal reasoning erred, not a neutral description of a mechanism, exactly the register of a catalogued informal fallacy. On institutional_origin it is likewise framed: the entry is fallacy-taxonomy furniture — the non causa pro causa lineage, its slot among the three escape-hatches a correlation cannot rule out, its packaging as a "named pattern of bad reasoning" in Walton and Govier — all distinctions drawn inside informal logic and argumentation theory. And on the human_practice_bound axis the fallacy framing is practice-relative: a "fallacy" is a defect in the human act of inference, so the label has nothing to grip where no reasoner is drawing an oriented causal claim.

But three features pull it firmly off the framed pole, and this is what separates it from a pole case like ad hominem. First, its portable core is not practice-constituted at all: correlation is symmetric and therefore carries no which-way signal is a mathematical fact, true observer-free, not an artifact of any reasoning tradition — so the structure the fallacy crystallizes is as substrate-independent as they come. Second, on import_vs_recognize the diagnostic transfers as recognition, not import: because its one precondition (an oriented causal claim from a symmetric association) is met across all empirical science, the identical diagnostic and its two asymmetry tests (temporal precedence, interventional response) are deployed literally in epidemiology, economics, clinical reasoning, and quantitative social science — not by analogy, which is why the entry insists the boundary here is instrument-naming versus mechanism-ownership, not instrument-versus-metaphor. Third, on vocab_travels it is split: the fallacy vocabulary (non causa pro causa, "named pattern of bad reasoning") is home-bound to rhetoric, but the operative diagnostic vocabulary (symmetric correlation, temporal precedence, interventional response, suspended verdict) travels intact wherever causal claims are read off data.

The portable structural skeleton is a correlation underdetermines causal direction, which only temporal precedence or interventional response can settle — and, as the entry establishes, that skeleton is precisely what the fallacy instantiates from its parent primes correlation (which lists reverse causation among correlation's escape-hatches) and causality, not what makes "wrong-direction fallacy" itself travel. The cross-domain force is owned by those parents; the home-bound cargo is the fallacy framing — the label, the taxonomy slot, the non causa pro causa lineage — which adds no structural content beyond what correlation and causality already encode. Its character: a normatively charged, taxonomy-born fallacy label whose distinctive cargo is argumentation furniture, but framed-leaning rather than framed-pole because the substrate-independent reverse-causation insight it crystallizes is carried by correlation/causality and travels as literal cross-science recognition, not analogy.

Structural Core vs. Domain Accent¶

This section decides why the wrong-direction fallacy is a domain-specific abstraction and not a prime, and it carries the case for its domain-specificity — there is no separate section for that.

What is skeletal (could lift toward a cross-domain prime). Strip the fallacy framing and a thin relational structure survives: a correlation is symmetric and so carries no which-way signal, so an oriented causal claim underdetermined by association can be fixed only by temporal precedence or interventional response. The portable pieces are abstract — a symmetric association, an oriented arrow the association cannot adjudicate, and two asymmetry tests (which variable moves first in time, which intervention bites) that break the symmetry. That skeleton is genuinely substrate-portable — indeed its core (correlation's symmetry) is a mathematical fact true observer-free — which is why the entry attributes it to the catalog primes the fallacy instantiates: correlation (whose treatment explicitly lists reverse causation among correlation's escape-hatches, alongside common-cause and chance) and causality. That underdetermination-plus-asymmetry-test core is what the wrong-direction fallacy shares with every causal inference from observational data — not what makes it the wrong-direction fallacy.

What is domain-bound. The distinctive content is fallacy-taxonomy furniture, and it is unusually thin: the label itself as a named pattern of bad reasoning; the non causa pro causa lineage; the slot among the three escape-hatches a correlation cannot rule out; the packaging in Walton and Govier as a catalogued informal fallacy; and the normative verdict the name delivers (to call an inference "the wrong-direction fallacy" is to convict it). These are argumentation-theory distinctions, drawn inside informal logic for cataloguing and teaching. The decisive test: remove the reasoner drawing an oriented causal claim — the human act of inference the "fallacy" indicts — and there is nothing for the label to grip, even though the underlying underdetermination fact still holds; conversely, strip the fallacy packaging and keep the structural insight, and what remains simply is correlation and causality doing the work. The fallacy framing is the home-bound accent; it adds no structural content beyond what the parents already encode.

Why this does not clear the prime bar. A prime is a relational structure whose vocabulary travels and whose cross-domain transfer is recognition of the same mechanism, not analogy — and here, unusually, the diagnostic transfers as literal recognition across the whole empirical-science substrate, yet the entry still sits below the prime bar. The reason is that what travels is not the named category but the parent insight. Within its precondition (any oriented causal claim from a symmetric association) the identical diagnostic and its two asymmetry tests are deployed literally in informal logic, epidemiology (cohort designs, Mendelian randomization), economics and policy, clinical reasoning, quantitative social science, and ML interpretability — not by analogy, so the boundary here is instrument-naming versus mechanism-ownership, not instrument-versus-metaphor. But that very wideness is the tell: the thing recurring across all those sciences is a correlation underdetermines causal direction, which only temporal or interventional evidence can settle, and that is already carried, in more general form, by correlation and causality. The wrong-direction fallacy is the argumentation-domain crystallization of that insight — the name a rhetorician gives the trap, useful for cataloguing and teaching — while the cross-domain force is owned by the parents. So the cross-domain reach belongs to correlation/causality; the fallacy label is the local, practice-relative accent that adds a normative verdict and a taxonomy slot but no portable structure. It clears the domain-specific bar as a genuine diagnostic but sits below the prime bar, because its only substrate-spanning content is already held by the primes it instantiates.

Relationships to Other Abstractions¶

Current abstraction Wrong Direction (Reverse Causation Fallacy) Domain-specific

Parents (3) — more general patterns this builds on

Wrong Direction (Reverse Causation Fallacy) is a kind of Informal Fallacy Prime

Wrong Direction is the informal-fallacy species that assigns a causal arrow from symmetric association while the genuine arrow runs in reverse.
Wrong Direction (Reverse Causation Fallacy) is part of Causality Prime

Wrong Direction contains causality because its identity requires both an asserted causal arrow and a genuine causal relation whose orientation the assertion reverses.
Wrong Direction (Reverse Causation Fallacy) presupposes Correlation Prime

Wrong Direction presupposes correlation because the fallacy begins with a symmetric observed association that carries no information about which variable is upstream.

Hierarchy paths (3) — routes to 3 parentless roots

Wrong Direction (Reverse Causation Fallacy) → Informal Fallacy

Show alternative paths (2)

Not to Be Confused With¶

"Correlation is not causation" (the bare warning). The cruder lesson that cautions against inferring any causal link from association. The wrong-direction fallacy is the residual error it leaves untouched: it concedes both the correlation and the causation and faults only the orientation of the arrow. Tell: is the doubt about whether any causal link exists (the bare warning, an existence question), or about which way a conceded causal link runs (wrong direction, a direction question)? Mastering the first gives no protection against the second.
Common-cause confounding. A third variable drives both A and B, so there is no causal arrow between A and B at all — a different escape-hatch entirely. Wrong direction is a genuine A–B causal arrow pointing the wrong way. The remedies differ: adjust for the confounder versus fix the orientation. Tell: does a lurking third factor produce the association with no direct A–B link (confounding), or is there a real A–B cause whose arrow is simply inverted (wrong direction)?
Post hoc ergo propter hoc. Infers causation from temporal succession — "A came first, so A caused B" — over-trusting sequence. Wrong direction reads the wrong arrow from a symmetric association whose temporal order is often unclear or itself reversed, typically having ignored sequence (and intervention) altogether. Tell: is the error over-reading temporal order as cause (post hoc), or orienting an arrow from a directionless correlation with no temporal evidence consulted (wrong direction)?
Chance / spurious correlation. The reading in which there is no causal structure at all — the association is coincidence or noise. Wrong direction concedes real causation and real correlation; its fault is orientation, not existence. Tell: is the claim that the correlation reflects no genuine causal relationship (chance), or that a genuine relationship's direction has been inverted (wrong direction)? (These are two of the four cells — A→B, B→A, common cause, chance — a bare correlation cannot adjudicate.)
The parent primes correlation and causality (and the non causa pro causa umbrella). The substrate-neutral core — a correlation is symmetric and so underdetermines causal direction, which only temporal precedence or interventional response can settle — is owned by correlation (which lists reverse causation among correlation's escape-hatches) and causality; the wrong-direction fallacy is the informal-logic crystallization of that insight and one reading of the broader non causa pro causa fallacy. Tell: strip the fallacy label and taxonomy slot and the portable content simply is correlation/causality (treated more fully elsewhere); "wrong direction" is the rhetoric-domain name for the trap, adding a normative verdict but no new structure.

Neighborhood in Abstraction Space¶

Wrong Direction (Reverse Causation Fallacy) sits in a sparse region of the domain-specific corpus (93^rd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (309 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12