Theory & Research Mar 24, 2026 17 min read

The Causation Illusion: How We Manufacture Causes from Coincidence

#blog.tag.d1 #blog.tag.logical-fallacies #blog.tag.causation #blog.tag.correlation #blog.tag.reasoning #blog.tag.bok #blog.tag.encyclopedia

Humans are causal animals. From the earliest moments of cognitive development, we compulsively search for why things happen. This instinct served our ancestors well — those who correctly identified the cause of a rustling bush (predator, not wind) lived to reproduce. But this same instinct, operating in the complex modern world, generates a distinctive class of reasoning errors: causal fallacies. TellDear's Dimension 1 (Logical Fallacies) catalogues over 60 formal and informal fallacies. This article examines the cluster that corrupts our understanding of why — the fallacies that manufacture causation from coincidence, sequence, and pattern.

I. The Causal Hunger: Why We Can't Stop Finding Causes

Before examining individual fallacies, it's worth understanding why causal errors are so persistent. The answer lies in what cognitive scientists call our "causal model of the world" — the mental architecture that automatically generates causal explanations for observed events.

When you see a billiard ball strike another and the second ball moves, you don't merely observe a sequence. You perceive causation. David Hume famously argued that we never actually observe causation — only temporal sequence and spatial contiguity — but our brains can't help themselves. We see causes everywhere, even in animated geometric shapes (as Fritz Heider and Marianne Simmel demonstrated in their famous 1944 experiment, where subjects attributed intentions and causal agency to moving triangles).

This causal hunger is a feature, not a bug. But it has a critical failure mode: it generates causal narratives faster than our critical faculties can evaluate them. By the time we've consciously considered whether A actually caused B, our intuitive system has already constructed a convincing story about how and why it did. The causal fallacies described below exploit this gap between automatic causal inference and deliberate causal evaluation.

II. The Six Core Causal Fallacies

1. Post Hoc Ergo Propter Hoc — The Sequence Trap

Post hoc ergo propter hoc ("after this, therefore because of this") is the foundational causal fallacy. Its structure is deceptively simple: Event A occurred before Event B, therefore A caused B. The rooster crows before sunrise, therefore the rooster causes the sunrise.

The fallacy feels absurd when stated with roosters and sunrises, but it becomes remarkably persuasive in less transparent contexts:

Medicine: "I took this supplement and my cold went away in three days." (Most colds resolve in three days regardless.)
Policy: "Crime decreased after we passed this law." (Crime was already declining before the law, or a dozen other variables changed simultaneously.)
Business: "We hired a new CEO and profits rose." (Market conditions improved, or restructuring initiated by the previous CEO finally took effect.)
Superstition: "I wore my lucky shirt and we won." (The team's performance is independent of your wardrobe.)

What makes post hoc reasoning so persistent is that it is sometimes correct. Temporal sequence is a necessary condition for causation (causes must precede effects). The error lies in treating it as a sufficient condition. Real causal inference requires much more: controlled comparison, elimination of confounders, a plausible mechanism, and replication. Post hoc reasoning skips all of these and goes straight from "before" to "because."

The fallacy is particularly dangerous in domains where controlled experiments are difficult or impossible — macroeconomics, social policy, historical analysis. In these fields, temporal sequence is often the most salient evidence available, which makes post hoc reasoning feel like the only game in town. As the statistician George Box noted, the fact that we cannot run the counterfactual — we cannot observe what would have happened without the intervention — is precisely what makes causal inference in observational data so treacherous.

2. False Cause (Non Causa Pro Causa) — The Misattribution Engine

False cause is the broader category under which post hoc falls, but it deserves separate examination because it captures causal misattributions that go beyond mere temporal sequence. In false cause reasoning, a causal relationship is asserted between two events or phenomena based on inadequate evidence — not necessarily because one preceded the other, but because of superficial correlation, thematic similarity, or motivated reasoning.

Consider the perennial confusion between correlation and causation. The number of films Nicolas Cage appears in per year correlates strongly with the number of people who drown in swimming pools. Ice cream sales correlate with murder rates. Per capita cheese consumption correlates with the number of people who die tangled in their bedsheets. These are real statistical correlations, and they are entirely meaningless causally.

Tyler Vigen's "Spurious Correlations" project has documented hundreds of such cases, and they are funny precisely because the causal absurdity is transparent. But when the correlation involves variables that could plausibly be related — education spending and test scores, immigration and crime, screen time and depression — the same logical error becomes much harder to detect. The correlation provides a scaffold on which our causal hunger eagerly constructs a narrative.

False cause reasoning also operates through thematic association. If two phenomena seem conceptually related, we tend to assume they're causally linked. "Video games are violent, and some gamers commit violence, therefore video games cause violence." The thematic connection between fictional violence and real violence makes the causal claim feel intuitive, even though decades of research have failed to establish a clear causal link. Our minds conflate "related to" with "caused by."

The antidote to false cause reasoning is the discipline of asking: What would it take to actually establish this causal claim? Usually, the answer involves controlled experiments, natural experiments, or at minimum a careful analysis of confounding variables — none of which a mere correlation provides. For a deeper exploration of how statistical patterns mislead, see How Numbers Lie.

3. The Texas Sharpshooter Fallacy — Drawing Targets Around Bullet Holes

The Texas Sharpshooter fallacy gets its name from a joke about a Texan who fires randomly at the side of a barn, then paints a bullseye around the tightest cluster of bullet holes. It describes the error of identifying a pattern in random data and then treating that pattern as if it were predicted in advance.

This fallacy is everywhere in scientific and pseudo-scientific reasoning:

Cancer clusters: A community notices an unusually high rate of cancer and attributes it to a local factory. But in any large population, some clusters will appear by chance alone. The question is whether the cluster is statistically meaningful relative to how many clusters you'd expect by chance — a question that requires careful epidemiological analysis, not pattern-matching.
Financial markets: After a market crash, analysts identify the "warning signs" that were "clearly visible" beforehand. But they're selecting these signs from thousands of indicators, most of which pointed in different directions. The "pattern" is constructed retrospectively from noise.
Bible codes: The claim that hidden messages can be found in the Torah by selecting every nth letter. Given any sufficiently long text, you can find almost any message by varying the skip interval — you're painting bullseyes around random clusters of letters.
Dietary studies: Researchers measure 50 health outcomes and find that eating chocolate correlates with two of them. They report these two as "significant findings." But with 50 measurements, you'd expect about 2-3 "significant" results by chance at the p < 0.05 level. This is the multiple comparisons problem, and it's the statistical formalization of the Texas Sharpshooter.

The cognitive root of this fallacy is our inability to intuitively appreciate the vastness of the space of possible patterns. In any complex dataset, there are astronomically many possible patterns to find. Some of them will look striking by pure chance. The Texas Sharpshooter mistake is treating a pattern discovered in data as if it were a hypothesis formulated before looking at the data. In statistics, this distinction — between exploratory and confirmatory analysis — is fundamental. In everyday reasoning, it's almost universally ignored.

4. The Gambler's Fallacy — When Randomness Feels Purposeful

The Gambler's Fallacy is the belief that independent random events somehow influence each other — that a roulette wheel that has landed on red six times in a row is now "due" for black. It's a causal fallacy because it implicitly attributes causal agency to the sequence itself, as if the wheel remembers its history and adjusts accordingly.

The mathematics are unambiguous: each spin of a fair roulette wheel is independent. The probability of black on the next spin is exactly the same whether the previous six spins were all red, all black, or alternating. The wheel has no memory. And yet, the feeling that "it's due" is almost irresistible. Why?

The answer lies in our expectations about randomness. Psychologists Amos Tversky and Daniel Kahneman showed that people have a deeply flawed model of what random sequences look like. We expect random sequences to be more balanced and less streaky than they actually are. When we see a streak (six reds in a row), we perceive it as a deviation from randomness that needs to be "corrected" — as if the universe has a quota system for outcomes.

The Gambler's Fallacy has consequences far beyond casinos:

Judicial decisions: Research by Daniel Chen, Tobias Moskowitz, and Kelly Shue found that asylum judges, loan officers, and baseball umpires all show Gambler's Fallacy patterns — after several decisions in one direction, they become more likely to decide in the other direction, regardless of the merits of the case.
Investing: "This stock has fallen for five consecutive days — it must bounce back." (It might, or it might continue falling. Past movement doesn't cause future movement in efficient markets.)
Everyday reasoning: "We've had three girls — the next one is bound to be a boy." (Each conception is essentially an independent event with approximately 50/50 odds.)

The flipside of the Gambler's Fallacy is the hot hand fallacy — the belief that a streak will continue. Interestingly, recent research suggests that the hot hand in basketball may be partially real (though smaller than people perceive). This asymmetry illustrates why causal reasoning about sequences is so difficult: sometimes streaks are meaningful, and sometimes they're not, and our intuitions are poor guides to which is which.

5. Regression Fallacy — Mistaking Statistical Noise for Causation

The Regression Fallacy occurs when we attribute a causal explanation to what is actually regression to the mean — the statistical tendency of extreme observations to be followed by less extreme ones. It's arguably the most underappreciated causal fallacy, responsible for vast amounts of wasted money, bad policy, and false beliefs.

The mechanism is simple. Any measurement that combines a true signal with random noise will show regression to the mean. If a student scores exceptionally well on an exam (because their true ability plus lucky noise aligned), their next score will likely be closer to their true ability — i.e., lower. If a city has an unusually high crime year (true crime rate plus random fluctuation), the next year will likely be closer to the baseline — i.e., lower.

Now add a causal intervention between the two measurements. A teacher scolds a student after an exceptional score, and the student performs worse next time. A mayor implements a new crime policy after a spike year, and crime drops the following year. In both cases, the regression to the mean would have happened regardless of the intervention. But the intervention is given causal credit.

Daniel Kahneman describes this as one of the most important — and most commonly misunderstood — phenomena in statistics. He tells the story of Israeli flight instructors who praised trainees after exceptionally good landings and criticized them after poor ones. They observed that praise seemed to be followed by worse performance, while criticism was followed by improvement. Their conclusion: criticism works, praise doesn't. The actual explanation: extreme performances (both good and bad) naturally regress toward average, regardless of the instructor's response.

The regression fallacy undermines evidence-based evaluation in virtually every domain:

Medicine: Patients tend to seek treatment when symptoms are at their worst. Symptoms naturally fluctuate, so many patients improve after treatment regardless of its efficacy. This is why controlled trials with placebo groups are essential — and why anecdotal evidence of "cures" is so unreliable.
Sports: The "Sports Illustrated curse" — athletes who appear on the cover after an exceptional season tend to perform worse the next year. Not because of a curse, but because exceptional seasons are partly luck that regresses.
Education: Schools identified as failing (bottom 5%) are given special interventions. They improve. Schools identified as excellent (top 5%) are held up as models. They decline. Both are largely regression to the mean, but the narrative credits (or blames) the interventions.

6. The Single Cause Fallacy — Narrative Simplification

The Single Cause Fallacy (also called causal oversimplification or reductive fallacy) occurs when we attribute an effect to a single cause when it actually results from multiple interacting factors. It's less a failure of logic than a failure of narrative — our stories prefer one protagonist, one villain, one explanation.

Complex events almost always have multiple causes operating at different levels:

World War I: "caused by" the assassination of Archduke Franz Ferdinand — except it was also caused by the alliance system, arms races, imperial competition, nationalist movements, railroad mobilization schedules, and decades of diplomatic failures.
Obesity: "caused by" eating too much — except it involves genetics, gut microbiome, food environment, stress, sleep, socioeconomic factors, hormonal regulation, and dozens of other variables.
The 2008 financial crisis: "caused by" subprime mortgages — except it also required deregulation, credit default swaps, rating agency failures, leverage ratios, monetary policy, and collective delusions about housing prices.

The Single Cause Fallacy is not merely an intellectual error — it has profound practical consequences. If you believe obesity is caused by personal choices alone, you design interventions around individual willpower (which largely fail). If you recognize it as a multi-causal system, you might also address food policy, urban planning, economic inequality, and advertising regulation. Single-cause thinking produces single-lever solutions, and single-lever solutions rarely work for multi-causal problems.

This fallacy connects deeply to what TellDear's Dimension 3 captures as status quo bias — when a complex system produces bad outcomes, the single-cause narrative typically identifies a cause that conveniently avoids structural change. For a broader exploration of how cognitive biases shape our decisions, see The Architecture of Bad Choices.

III. How Causal Fallacies Interact

In practice, causal fallacies rarely appear in isolation. They form reinforcing clusters that make the resulting false beliefs remarkably resilient:

The Alternative Medicine Cascade: A person feels ill (extreme state) → seeks alternative treatment → improves (regression to the mean) → attributes improvement to treatment (post hoc) → tells others about their "cure" (anecdotal evidence + hasty generalization) → seeks patterns of who else improved (Texas Sharpshooter) → constructs a single-cause theory of disease (single cause fallacy). Each step is individually flawed, but together they construct an edifice that feels overwhelmingly convincing.

The Policy Feedback Loop: Crime spikes (partly noise) → new tough-on-crime policy → crime decreases (regression) → policy declared successful (post hoc) → policy expanded → next spike attributed to insufficient enforcement (single cause) → even tougher policy → another regression → further confirmation. The policy becomes self-validating regardless of its actual effectiveness.

The Superstition Machine: Random good outcome → seek explanation (causal hunger) → identify coincidental factor (false cause) → repeat the behavior → sometimes it "works" again (intermittent reinforcement + Texas Sharpshooter) → belief solidifies → disconfirming instances forgotten (cherry picking) → ritual becomes entrenched.

These cascades show why individual fallacy identification, while necessary, is insufficient. You need to see the system of causal errors — how one fallacy creates the conditions for the next, and how the compound effect creates beliefs that resist any single point of correction.

IV. Causal Fallacies in Science: The Replication Crisis

The scientific replication crisis — the finding that many published results fail to replicate — is, in significant part, a story about causal fallacies operating within the structures of science itself.

Texas Sharpshooter at Scale: When researchers test multiple hypotheses (or multiple variations of a hypothesis) and report only the "significant" results, they're painting bullseyes around random clusters. This practice, called p-hacking or data dredging, has been documented across psychology, medicine, economics, and other fields.

Post Hoc Narratives: After finding an unexpected significant result, researchers construct a post hoc theoretical justification — "we predicted this because..." — transforming an exploratory finding into an apparently confirmatory one. This is known as HARKing (Hypothesizing After Results are Known).

Regression Artifacts: Studies that select extreme groups for intervention (the sickest patients, the worst-performing schools, the most depressed individuals) will almost always show improvement — not because the intervention works, but because extreme groups regress to the mean. Without careful control groups, regression artifacts masquerade as treatment effects.

The methodological reforms sweeping science — pre-registration of hypotheses, registered reports, larger sample sizes, adversarial collaborations — are essentially structural defenses against causal fallacies. They force researchers to commit to their bullseyes before shooting, to predict before observing, and to compare against what would have happened anyway.

V. Defending Against Causal Fallacies

Recognizing causal fallacies in the wild requires cultivating a specific kind of cognitive discipline — what we might call causal skepticism. Not the nihilistic kind ("we can never know anything about causation") but the productive kind ("causation is real but hard to establish, and our intuitions about it are unreliable").

Ask the counterfactual: "What would have happened without this supposed cause?" If you can't answer this question — if you don't have a control group, a baseline, or a natural experiment — your causal claim is standing on very shaky ground. This is the fundamental question of causal inference, and it has no shortcut.

Check for regression: "Was the starting point extreme?" If you're evaluating an intervention that was applied after an extreme observation (a crisis, an exceptional performance, a dramatic failure), ask whether regression to the mean alone could explain the subsequent change. Often it can.

Count the causes: "Is there really only one explanation?" For any complex event, force yourself to list at least three plausible contributing causes before settling on one. This combats the narrative simplification of the single cause fallacy and opens up more productive avenues for understanding and intervention.

Demand mechanism: "How, specifically, would A cause B?" A correlation or temporal sequence without a plausible mechanism is much less convincing than one with a clear causal pathway. This doesn't mean mechanism is required (we used aspirin for decades before understanding its mechanism), but its absence should increase our skepticism.

Respect base rates: "How often does this happen anyway?" If something happens 30% of the time without intervention, the fact that it happened after your intervention tells you almost nothing. You need to know whether the rate changed, not merely whether it occurred. This connects to the hasty generalization fallacy — drawing broad conclusions from too few observations.

VI. The Deeper Problem: Narrative Causation

Underlying all causal fallacies is a deeper cognitive tendency that may be impossible to fully overcome: our preference for causal narratives over statistical descriptions. We understand the world through stories, and stories require causes. "A happened, then B happened, because of C" is a story. "There is a 0.3 correlation between A and B after controlling for confounders D through K" is not a story. It's a description. And descriptions, no matter how accurate, don't satisfy our causal hunger.

This is not merely a cognitive limitation — it's built into our language, our media, our institutions. Newspapers report that the stock market "fell because of" inflation fears. Politicians claim that their policies "caused" economic growth. Doctors tell patients that their disease was "caused by" a specific factor. In each case, the causal language is typically an oversimplification of something much more complex, uncertain, and multi-factorial.

The philosopher Nassim Nicholas Taleb calls this the "narrative fallacy" — our tendency to construct coherent causal stories from random events. The psychologist Daniel Kahneman describes it as the operation of "System 1" — our fast, automatic, narrative-generating cognitive system — overwhelming "System 2" — our slow, deliberate, analytical system. Both are describing the same fundamental tension: we are wired for stories, but reality is wired for statistics.

The goal of studying causal fallacies is not to stop telling causal stories — that would be neither possible nor desirable. It's to develop the metacognitive awareness to ask, at the right moments: "Is this a story I'm telling, or something I actually know?" That question, asked consistently and honestly, is the beginning of genuine critical thinking about causation.