All Chapters

The Missing Chapter

The Conjunction Fallacy

Why a good story is the enemy of good probability

An extension of Jordan Ellenberg's "How Not to Be Wrong"

Chapter 75

Meet Linda

Linda is thirty-one years old, single, outspoken, and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.

Which is more probable?

A. Linda is a bank teller.
B. Linda is a bank teller and is active in the feminist movement.

If you picked B, you're in excellent company. About 85% of people do — including, as we'll see, a disturbing number of people who should know better. You're also wrong.

This is the conjunction fallacy, and it's one of the most important results in the history of cognitive science. It was introduced in 1983 by Amos Tversky and Daniel Kahneman in a paper with the admirably boring title "Extensional Versus Intuitive Reasoning," and it demonstrates something profound about how human minds work.1 We are not probability calculators. We are story processors. And when a good story and good math walk into a room together, the story wins almost every time.

The conjunction rule is perhaps the simplest law of probability: the probability of two things both being true can never exceed the probability of either one alone. P(A ∧ B) ≤ P(A). It's not a deep theorem. It's barely even a theorem. It's more like the mathematical equivalent of "a thing can't be bigger than the thing that contains it."

Think about it with sets. The collection of all bank tellers in the world is some group of people. The collection of all feminist bank tellers is a subset of that group — it lives entirely inside it, the way Rhode Island lives inside the United States. You cannot drive from Boston to Providence without entering Rhode Island, but you can easily drive from Boston to Providence without leaving the United States. The smaller thing is inside the bigger thing. Always.

Bank Tellers (all of them) Feminist Bank Tellers This can never be bigger than that ↙
Every feminist bank teller is a bank teller. The reverse is not true. The subset cannot exceed the whole.

So why does almost everyone get this wrong?

Chapter 75

The Representativeness Heuristic

Tversky and Kahneman's answer was the representativeness heuristic: when asked "how probable is it that Linda is X?", people don't actually compute a probability. Instead, they ask themselves a different question — "how much does Linda resemble X?" — and report that answer as if it were a probability.2

And Linda, with her philosophy degree and her anti-nuclear demonstrations, resembles a feminist bank teller much more than she resembles a mere bank teller. "Bank teller" feels random, arbitrary, disconnected from the story we've just been told. "Feminist bank teller" fits. It completes the narrative. It makes the character make sense.

This is a deep feature of human cognition, not a shallow mistake. We are, at our core, narrative reasoners. We don't evaluate claims by computing their extensional probability — the share of possible worlds in which they obtain. We evaluate them by their coherence, by how well they fit into a story. And more detail, more specificity, more conjunction, almost always makes a story more coherent even as it makes the underlying event less probable.

"The more detailed a story is, the less likely it is to be true — and the more likely it is to be believed."

Try it yourself. I'll give you the Linda description and some options to rank. Be honest — go with your gut.

🧪 The Linda Experiment

Read the description, then rank these statements from most probable to least probable by clicking them in order. Go with your gut feeling.

Linda is thirty-one years old, single, outspoken, and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.

Click statements in order from most to least probable:

Chapter 75

It's Not Just Linda

You might think the Linda problem is a party trick — a clever bit of misdirection that only works because the description is specifically designed to mislead. But Tversky and Kahneman showed it's far more general than that. The conjunction fallacy appears whenever a conjunction is more "representative" than one of its constituents.

Consider this scenario they tested in 1983: subjects were told that a hypothetical participant in Wimbledon had lost the first set. They were then asked to rank the probability of various outcomes. Most people rated "will lose the first set but win the match" as more likely than "will win the match."3 The comeback narrative was simply too compelling.

Or consider their health survey scenario: subjects judged it more probable that a 55-year-old man would have a "heart attack and survive" than that he would simply have a "heart attack." Because "heart attack and survive" is a story — a little drama with a happy ending. "Heart attack" is just a clinical event, floating in the void without resolution.

Here's what makes the conjunction fallacy truly alarming: expertise doesn't cure it. Tversky and Kahneman tested doctoral students in the decision science program at Stanford — people who had taken multiple courses in probability and statistics. The conjunction fallacy rate among these trained statisticians? Still above 80%.4

Knowing the rule isn't enough. You have to feel it, and the representativeness heuristic is older and louder than any statistics class.

This leads to one of those uncomfortable conclusions that mathematics occasionally forces upon us: the human mind was not designed for probability. It was designed for narrative. Evolution built us to track agents, intentions, and plot lines — who did what to whom and why — because that's what kept us alive on the savanna. Whether the lion was "probably" behind that bush was less important than whether the story of a lion behind that bush made sense given the evidence.

Amount of Detail Added Level Story Coherence Actual Probability The Danger Zone →
As you add detail, stories feel more plausible but become less probable. The gap between the curves is where the conjunction fallacy lives.
Chapter 75

The Frequency Fix

There is, however, a partial cure. In the 1990s, the evolutionary psychologist Gerd Gigerenzer proposed that the conjunction fallacy largely disappears when problems are framed in terms of frequencies rather than probabilities.5

Instead of asking "What is the probability that Linda is a bank teller who is active in the feminist movement?", try this:

There are 100 women who fit Linda's description. How many of them are:

When the question is phrased this way, people can suddenly see the set relationships. If you write down "5 are bank tellers" and then try to write down a number for "bank tellers who are also feminists," it's obvious that number can't be bigger than 5. The frequency frame makes the nesting visible in a way the probability frame doesn't.

Gigerenzer's interpretation was radical: maybe the human mind isn't broken. Maybe it's just adapted for frequencies — the format in which information actually arrived in the ancestral environment — rather than for single-event probabilities, which are a relatively modern mathematical abstraction. You never encountered "the probability of a lion" on the savanna. You encountered "three out of the last ten times I walked through that tall grass, something tried to eat me."

The debate between Kahneman and Gigerenzer — are we fundamentally irrational, or are we rational in the wrong format? — is one of the great intellectual feuds of twentieth-century psychology.6 But for our purposes, the practical takeaway is clear: when you need to think about conjunctions, think in frequencies.

📐 Conjunction Calculator

Set the probabilities of two independent events and see why their conjunction must be smaller than either one alone.

P(A) — Probability of Event A 50%
P(B) — Probability of Event B 50%
P(A) = 50%
50%
P(B) = 50%
50%
P(A ∧ B) = 25%
25%
Conjunction Probability
25.0%
Always ≤ min(50%, 50%) = 50%
Out of 1,000 people
250
Have Event A
500
Have Event B
500
Chapter 75

Conjunctions in the Wild

The conjunction fallacy isn't just a laboratory curiosity. It shapes how we think about some of the most consequential domains of human life.

The Planning Fallacy

Consider any complex project — building a house, launching a startup, writing a PhD thesis. The plan consists of many steps, each of which must succeed: the permits come through and the contractor is available and the materials arrive on time and the weather cooperates and the inspector approves the foundation and...

Each individual step might have a high probability of going smoothly — say 90%. But string ten such steps together and your overall probability drops to 0.910 ≈ 35%. Twenty steps? About 12%. The plan that seemed rock-solid when you considered each piece individually is actually more likely to fail than to succeed.7

The Planning Fallacy
P(plan works) = P(step₁) × P(step₂) × ⋯ × P(stepₙ)
Even with P = 0.9 per step, 20 steps gives only ~12% total probability

This is the conjunction fallacy writ large. We evaluate each step's plausibility independently — "sure, that'll probably work" — and our brains quietly treat the whole conjunction as if it has the same plausibility as each piece. But multiplication is brutal. It only takes a few "probably fine" steps to produce an overall plan that's "probably doomed."

Conspiracy Theories

Conspiracy theories thrive on the conjunction fallacy, though they exploit it in reverse. A conspiracy theory adds detail — the secret meetings, the hidden documents, the carefully timed cover-ups — and each detail makes the story feel more coherent, more alive, more plausible. The representativeness heuristic kicks in: this detailed scenario resembles something that could really happen.

But every added detail is another conjunction, another multiplicative penalty on the overall probability. The conspiracy theory that feels most convincing — the one with the richest, most specific account of how it all went down — is, by the cold logic of probability, the least likely to be true.

The Detail Paradox

In stories, detail creates verisimilitude. In probability, detail creates impossibility. The human mind was built for stories.

The Courtroom

Prosecutors tell stories. That's literally their job. "The defendant drove to the victim's house at 11 PM, entered through the unlocked back door, took the knife from the kitchen drawer, and..." Each detail makes the story more vivid, more convincing to a jury. A narrative with timestamps and specific actions feels more true than a vague accusation.

Defense attorneys, interestingly, often do best when they keep things vague — "reasonable doubt" doesn't require a counter-narrative, just the absence of certainty. This asymmetry maps perfectly onto the conjunction fallacy: the prosecution's detailed story triggers our representativeness heuristic, while the defense's appeal to uncertainty is fighting our deepest cognitive instincts.8

Step 1 90% × Step 2 90% × Step 3 90% Step 10 90% = 35% total Each step feels fine. The chain is what kills you.
Ten 90%-likely steps multiply to just 35%. Twenty steps: 12%. This is why projects run late.
Chapter 75

What Linda Teaches Us

The conjunction fallacy is, at bottom, a lesson about the difference between two kinds of thinking — or, if you prefer Kahneman's later terminology, between System 1 and System 2. System 1, fast and intuitive, evaluates claims by narrative coherence. System 2, slow and deliberate, can in principle apply the conjunction rule. But System 2 is lazy, and System 1 is persuasive, and most of the time System 1 wins without System 2 even knowing there was a fight.

The practical implications are everywhere:

Think in frequencies. Instead of "What's the probability that...?", ask "Out of 100 cases like this, how many would...?" The set relationships become visible.

Watch for detail. When a scenario feels more plausible because it's more specific, that's the conjunction fallacy waving a red flag. More detail should make you more skeptical, not less.

Decompose plans. Before declaring a project plan "realistic," list every step that must go right, assign each a probability, and multiply. The result will depress you — and it should.

Beware compelling stories. In courtrooms, in political arguments, in sales pitches: the more vivid and detailed the story, the more you should slow down and think about what's actually been proven.

Linda doesn't care about your statistics education. She is thirty-one years old, single, outspoken, and very bright. And every time you hear her description, some ancient part of your brain will whisper: feminist bank teller, obviously.

The trick isn't to silence that whisper. It's to hear it, recognize it for what it is — the voice of a narrative engine doing exactly what it was evolved to do — and then, gently but firmly, do the math anyway.

Notes & References

  1. Tversky, A., & Kahneman, D. (1983). "Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment." Psychological Review, 90(4), 293–315. The foundational paper that introduced the Linda problem and systematically documented the conjunction fallacy across multiple experimental paradigms.
  2. Kahneman, D., & Tversky, A. (1972). "Subjective probability: A judgment of representativeness." Cognitive Psychology, 3(3), 430–454. The earlier paper that introduced the representativeness heuristic as a general framework for understanding probability judgment.
  3. The Wimbledon example and the health survey example both appear in Tversky & Kahneman (1983), demonstrating that the conjunction fallacy extends well beyond the Linda problem to sports predictions, medical judgments, and many other domains.
  4. Tversky & Kahneman (1983), Study 7. Graduate students in the decision science program at Stanford Business School showed conjunction fallacy rates of 85%, comparable to untrained subjects. Statistical sophistication provided essentially no protection.
  5. Gigerenzer, G. (1991). "How to make cognitive illusions disappear: Beyond 'heuristics and biases.'" European Review of Social Psychology, 2(1), 83–115. Gigerenzer's influential argument that frequency formats align with evolved cognitive mechanisms and dramatically reduce conjunction errors.
  6. For a balanced overview of the Kahneman-Gigerenzer debate, see Mellers, B., Hertwig, R., & Kahneman, D. (2001). "Do frequency representations eliminate conjunction effects? An exercise in adversarial collaboration." Psychological Science, 12(4), 269–275.
  7. Kahneman, D., & Tversky, A. (1979). "Intuitive prediction: Biases and corrective procedures." TIMS Studies in Management Science, 12, 313–327. Also developed extensively in Kahneman's Thinking, Fast and Slow (2011), Chapter 23.
  8. Pennington, N., & Hastie, R. (1992). "Explaining the evidence: Tests of the Story Model for juror decision making." Journal of Personality and Social Psychology, 62(2), 189–206. Demonstrates that jurors construct narratives from evidence and that verdict choices are driven by story coherence.