012 min
Where the narrative fallacy shows up
Illustrative scenario. A product launch misses its signup target. An engineering manager runs a retrospective with the team. Within an hour the group agrees on a cause. The onboarding flow had one step too many, the step was added late, and the team ignored an early warning from a support lead. Each of these facts is true. The group lists them in time order, each one leads to the next, and the explanation feels complete.
Now change one input. The launch beat its target by 20 percent. The same facts are on the table: the extra step, the late addition, the early warning. The retrospective now finds a different cause. The team made a bold late change, trusted its own judgement, and the support lead's warning turned out to be too cautious. The facts did not change. Only the result changed, and the explanation changed with it.
That is the tell. When the same facts support opposite stories depending on the outcome, the story was built from the outcome and not from the facts. This is related to hindsight bias, the belief that an outcome was predictable once it is known, but it is a separate error. Hindsight bias is about how predictable the result looks. The narrative fallacy is about the chain of causes the group builds to explain it.
The pattern is general. After any event, people choose the facts that lead to it, put them in order, and read the order as cause.
022 min
Why the narrative fallacy happens
The narrative fallacy comes from three ordinary steps in how people explain events. Each step is useful on its own. The error appears when the three run together and nothing checks the result.
Step 1: selection
A past event has more surrounding facts than anyone can hold in mind, so people keep a few. Daniel Kahneman, who adopted the term in his 2011 book Thinking, Fast and Slow, describes which few they keep. He writes that the explanatory stories that people find compelling are simple; are concrete rather than abstract; assign a larger role to talent, stupidity, and intentions than to luck; and focus on a few striking events that happened rather than on the countless events that failed to happen.
The last part matters most. A story can only contain events that happened. The events that did not happen, but could have changed the result, never appear in it.
Step 2: linking
Once a few facts are chosen, the mind connects them. Taleb says an explanation binds facts together and makes them easier to remember. A list of five unrelated events is hard to hold. A sequence in which each event causes the next is easy to hold. The link between two events is added by the person explaining. It is not observed.
Step 3: confidence
The finished story feels like understanding. Kahneman calls the underlying rule WYSIATI, short for "What You See Is All There Is": people treat the information they have as if it were all there is to know. He writes that you build the best possible story from the information available to you, and if it is a good story, you believe it. He adds that it is easier to build a coherent story when you know little, because fewer pieces have to fit.
This is why more detail does not always cure the problem. A person with little information produces a smooth story. A person with a lot of information has to leave facts out to keep the story smooth.
Where the mechanism is uncertain
The reason the mind builds such stories is not settled. One proposal comes from split-brain research. Michael Gazzaniga, who studied patients whose two brain hemispheres had been surgically separated, writes of a left-hemisphere "interpreter" that constructs theories about the relationship between perceived events, actions and feelings. His paper says the phenomena appear to be related to this interpreter. That wording is a proposal, not a measured fact.
032 min
Where the narrative fallacy comes from
Nassim Nicholas Taleb introduced the term in his 2007 book The Black Swan. It is a claim from a book for general readers, not the result of an experiment of his own. Taleb defines it as "our limited ability to look at sequences of facts without weaving an explanation into them, or, equivalently, forcing a logical link, an arrow of relationship upon them."
Kahneman then made the term familiar to psychologists. In chapter 19 of Thinking, Fast and Slow, titled "The Illusion of Understanding", he credits Taleb for the idea. Taleb introduced the notion of a narrative fallacy to describe how flawed stories of the past shape our views of the world and our expectations for the future.
Older experiments show the tendency the term describes. In 1944 Fritz Heider and Marianne Simmel at Smith College showed subjects a film of about two and a half minutes. It contained a large triangle, a small triangle, a disc and a rectangle with a section that opened and closed like a door. The shapes only moved. In the first experiment subjects were told only to write down what happened in the picture. Only one subject described the film almost entirely in geometrical terms. All the other subjects interpreted the movements as actions of animate beings, in most cases persons, and in two cases birds. Nineteen of them reported a connected story.
Gazzaniga's split-brain work, summarised in his 2000 paper in the journal Brain, added a possible mechanism. A well-known case from his lab involved a patient whose right hemisphere saw a frightening video. She felt the emotional response to the video her right hemisphere had seen but had no idea what caused it. Her left hemisphere then supplied an explanation that placed the cause in the experimenter.
These three sources make different kinds of claims. Taleb's is a conceptual label. Heider and Simmel's is a description of what subjects wrote. Gazzaniga's is a proposed brain mechanism. None of them is a test of the narrative fallacy as Taleb defined it.
042 min
A real case: how Kahneman retells Google's rise
Kahneman uses the history of Google to show the narrative fallacy at work, and he tells it in the way a business magazine might. Two creative graduate students in the computer science department at Stanford come up with a better way to search the internet. They obtain funding and make a series of decisions that work out well. Within a few years the company is among the most valuable in America.
The account also contains one stroke of luck. In Kahneman's telling, a year after founding Google, its founders were willing to sell the company for less than $1 million, and the buyer said the price was too high. A reader notices that incident and counts it as luck. Kahneman's point is that naming one lucky event makes it easier to underestimate the many other ways in which luck affected the outcome.
He then lists what the story leaves out:
- The decisions the founders made that could have gone the other way. Because every critical decision turned out well, the record suggests almost flawless prescience.
- The actions of the competing firms that Google defeated, who in the story look blind, slow and inadequate.
- The events that did not happen. In his words, the human mind does not deal well with nonevents.
The halo effect, where a good result in one area makes a person or company look good in every other area, adds to the story. It makes the founders look invincible.
Kahneman then states the test he uses. The ultimate test of an explanation is whether it would have made the event predictable in advance. No story of Google's unlikely success will meet that test, because no story can include the myriad of events that would have caused a different outcome.
This is a worked example of the three steps above. The story selects a few facts, links them in time order, and ends in a conclusion about what makes businesses succeed. A reader who accepts that conclusion has learned a lesson that the evidence does not carry.
052 min
A second case: the word "knew"
The first case is a story told about a company. The second is a single word used in conversation. Kahneman writes that he has heard of too many people who "knew well before it happened that the 2008 financial crisis was inevitable."
He objects to the verb. Some people thought well in advance that there would be a crisis, but they did not know it. In everyday language, "know" applies only when what was known is true and can be shown to be true. The people who expected the crisis could not show it at the time. Many intelligent and well-informed people were keenly interested in the future of the economy and did not believe a catastrophe was imminent.
What is wrong with the verb, in Kahneman's view, is not that some individuals get credit for foresight they do not deserve. It is that the language implies that the world is more knowable than it is.
The variable that differs from the Google case is the unit of the story. Google's story is a full narrative with characters and a plot. The "I knew it" sentence is one short clause with no plot at all. It still carries the same claim: the outcome was visible in advance to anyone who looked. A short sentence does the same job as a long story, and it is easier to repeat.
Together the two cases show the range of the problem. It lives in business histories, and it lives in the vocabulary a team uses in a meeting.
062 min
How the narrative fallacy shows up in product, design and data work
All scenarios in this section are illustrative.
A launch retrospective. The opening scenario of this article is the standard case. The team writes a timeline, and the timeline becomes the explanation. The decision at risk is what the team changes next time. If the cause was invented, the team fixes a step that did not matter.
A founder's origin story. A founder tells investors and customers a history of the company in which each early decision leads to the next. The decision at risk is whether another team copies the decisions. Kahneman's Google analysis says the lesson drawn from such a history is largely illusory. This is a case where the story is often told on purpose and often believed by the person telling it. Suffering from the fallacy and exploiting it can both be true at once.
A research synthesis. A designer reviews interviews with eight participants and writes a single user journey: the person hears about the product, tries it, hits a problem, gives up. The journey is coherent. The decision at risk is the roadmap built on it. Each interview may have contained only part of that sequence, and the designer supplied the links between parts.
A dashboard explanation. A data analyst sees a drop in weekly active users and finds a release that shipped in the same week. The explanation arrives quickly and fits. The decision at risk is rolling back a release that did not cause the drop. Two events in the same week are a sequence, not a cause. Correlation and causation covers the test that separates them.
In each case the signal is the same. The explanation arrived fast, used few facts, and no one tried to find a version of events in which it would be false.
072 min
How to guard against the narrative fallacy
These techniques change the process of explaining, so they do not depend on the person noticing the error in themselves. No technique removes the habit. Each one makes the story easier to check.
Write the prediction before the result.
Before a launch, write down the expected outcome, the reasons, and a probability, with a date. After the result, the team compares the story it tells now with what it wrote then. This works because it fixes the facts known at the time, before the outcome selects among them.
Run the reversal test.
Take the retrospective and ask what it would say if the result had been the opposite. If the same facts support an equally good story, the story does not explain the result. The opening scenario shows this test being applied by accident.
List the events that did not happen.
Ask what had to go right, which choices could have gone another way, and which similar teams made the same choices and failed. Survivorship bias, where only the cases that succeeded are visible, hides most of those teams. Looking for them recovers the missing half of the evidence.
Apply Kahneman's test.
Ask whether the explanation would have made the event predictable in advance. If a person could not have used it to predict, it is a description of what happened and not an explanation of why.
Compare to a group.
Instead of asking why one project succeeded, ask how often projects like it succeed. A base rate for the whole group sets a limit on how much any single story can claim.
None of these needs special data. They need time before the outcome, and the habit of treating a fast explanation as a first draft.
081 min
Common misunderstandings about the narrative fallacy
"It means stories are always false." It does not. The fallacy is about confidence, not about whether an explanation is true. Taleb himself says explanations bind facts together, make them easier to remember, and help them make more sense. The problem he names appears where this habit increases our impression of understanding beyond what the facts support.
"It is the same as lying or spin." It is not. A person who builds an unsupported story usually believes it. Spin is a story told on purpose to persuade, which is closer to narrative transport, the state of being absorbed in a story.
"It only applies to the past." The Kahneman passage above says the opposite. Flawed stories of the past shape our views of the world and our expectations for the future. A team that explains a launch wrongly will plan the next launch wrongly.
"Every after-the-fact explanation is a narrative fallacy." Some explanations can be tested. A team that reads the logs of an outage and finds the exact configuration change that caused it has evidence for the cause, not a story. The fallacy applies when the chain of causes cannot be tested, or when no one tried to test it.
092 min
The narrative fallacy vs. nearby concepts
Several concepts describe errors that happen around the same time as the narrative fallacy. Each is separated from it by what it is about.
| Concept | What it is about | What separates it from the narrative fallacy |
|---|---|---|
| PsychologyHindsight bias | ||
| PsychologyConfirmation bias | ||
| PsychologyHalo effect | ||
| ResearchSurvivorship bias | ||
| ResearchCorrelation and causation | ||
| CommunicationNarrative transport |
The closest neighbours in practice are confirmation bias and hindsight bias. They often act together with the narrative fallacy and reinforce it. A team that has a belief about why launches fail will keep the facts that fit it, which is confirmation bias. The team then links those facts into a sequence, which is the narrative fallacy.
102 min
Where the evidence for the narrative fallacy is contested
The narrative fallacy is a descriptive label, not a measured effect. This is the main limit on how strongly it can be claimed.
It has no experiment of its own. Taleb proposed it in a book for general readers. The experiments that bear on it carry other names. Heider and Simmel's 1944 study is about how people describe moving shapes. Gazzaniga's interpreter is a proposed brain mechanism, and his own paper says the phenomena appear to be related to it. Neither study measures how often explanations are wrong or how much confidence is added. A reader cannot cite an effect size for the narrative fallacy.
The source book has critics. Mathematics professor David Aldous argued that Taleb is sensible in his discussion of financial markets and in some of his general philosophical thought, but tends toward irrelevance or ridiculous exaggeration otherwise. That objection is aimed at the book as a whole and not at this term in particular. The fair reading is that the label is useful and the surrounding argument should not be taken on trust.
The strongest objection is that some stories are right. If explanation after the fact were always unreliable, no one could learn from an incident or a case study. That is not what Taleb or Kahneman claim. They argue that the sense of understanding runs ahead of the evidence. Kahneman's own test, whether the explanation would have made the event predictable in advance, is a way to separate the two.
What a practitioner does while this is unsettled: use the term as a prompt to check an explanation, not as proof that one is wrong. The techniques above work whether or not the label itself is ever measured.
?7 questions
Questions people ask
What is the narrative fallacy?
What is an example of the narrative fallacy at work?
What is the difference between the narrative fallacy and hindsight bias?
How do you avoid the narrative fallacy in a retrospective?
Is the narrative fallacy a cognitive bias?
Does the narrative fallacy mean every explanation is wrong?
Why do stories feel more convincing than lists of facts?
§6 sources
Sources on the narrative fallacy
LessWrong, "Narrative Fallacy" (tag page quoting Nassim Nicholas Taleb's definition from The Black Swan, 2007).
Wikipedia, "The Black Swan" (book by Nassim Nicholas Taleb, 2007), including David Aldous's critical review. (Taleb_book)
Kahneman, D. (2011). "The Illusion of Understanding", chapter 19 of Thinking, Fast and Slow, excerpt reproduced by Hedge Fund Alpha.
Heider, F., & Simmel, M. (1944). An experimental study of apparent behavior. American Journal of Psychology, 57(2), 243–259.
Show all 6 sourcesShow fewer sources
Gazzaniga, M. S. (2000). Cerebral specialization and interhemispheric communication: does the corpus callosum enable the human condition? Brain, 123(7), 1293–1326.
Wikipedia, "Michael Gazzaniga" (account of the patient who felt fear without knowing the cause).
