The Illusion of Explanatory Depth
A reader's summary of the finding that people overestimate how well they understand how things work — covering the 2002 device study, the contrast with facts and procedures, the 2013 political experiments, the limits of the result, and why asking someone to explain a mechanism can change a conversation that asking for reasons does not.
The term at a glance
The illusion of explanatory depth is a metacognitive error: a mistaken judgement about one's own knowledge. It is usually described as a confidence that collapses on contact with a request to explain. It does not say that people are foolish or know nothing. It says that the felt sense of understanding is a poor measure of how much explanation one could actually produce.
Origin
The phrase comes from Leonid Rozenblit and Frank Keil's 2002 paper “The misunderstood limits of folk science: an illusion of explanatory depth” in the journal Cognitive Science. Keil, a cognitive psychologist then at Yale, had been studying how lay people reason about causes and categories, and the paper grew out of that work on the shallowness of everyday theories. It is distinct from the better-known Dunning–Kruger effect, which concerns people who are weak at a skill misjudging their own ability, whereas this effect appears across the range of people and concerns a specific kind of knowledge.
History and context
The result belongs to a long line of findings that people are poor judges of their own minds, including Richard Nisbett and Timothy Wilson's 1977 paper on the limits of introspection. It gained a wider audience when Fernbach and colleagues applied it to politics in 2013, and again through Sloman and Fernbach's 2017 book. It sits close to several ideas covered elsewhere on this site, such as epistemic learned helplessness, which describes giving up on evaluating arguments one cannot check, and confirmation bias, which keeps the shallow view intact once held.
Main ideas
Feeling you understand is not understanding
The illusion of explanatory depth is the gap between how well people think they understand a mechanism and how well they can actually explain it. Most people feel they know how a flush toilet or a zipper works. The feeling comes from familiarity: they have seen the thing work many times, and they can recognise the parts, which is easily mistaken for being able to describe the causal chain.
The study: Rozenblit and Keil, 2002
Leonid Rozenblit and Frank Keil's paper in Cognitive Science asked Yale students to rate, on a seven-point scale, how well they understood everyday devices such as a zipper, a flush toilet, a cylinder lock, a helicopter and a quartz watch. The students then wrote out a step-by-step explanation and rated their understanding again. The second rating fell, often sharply, because the attempt exposed how little of the mechanism they could actually state.
It is specific to explanations
The same researchers found the drop did not appear for knowledge of facts (such as the capital of a country), of procedures (such as how to bake chocolate chip cookies) or of narratives (such as the plot of a film). Those can be checked by simply doing or recounting them. Causal explanations are different: they chain many steps, and a vague sense of the whole can hide that most of the steps are missing.
Why the illusion forms
Rozenblit and Keil proposed several contributors. People confuse knowing what a thing looks like and does with knowing how it does it, they rarely test their explanations, and the world is full of visible mechanisms whose insides are never shown. Steven Sloman and Philip Fernbach's 2017 book The Knowledge Illusion adds the idea of a community of knowledge: because we constantly lean on other people and on tools, we lose track of where our own knowledge stops and theirs begins.
The political version: Fernbach and colleagues, 2013
In a Psychological Science paper, Philip Fernbach, Todd Rogers, Craig Fox and Steven Sloman asked Americans to rate their understanding of policies such as a cap-and-trade system for carbon emissions, a flat tax and merit pay for teachers. Participants who were asked to explain in detail how the policy would work rated their understanding lower afterwards and became less extreme in their stated positions. Participants who were only asked to list reasons for their position did not.
Not just an everyday-objects result
A related and earlier demonstration is about drawing a bicycle from memory. In studies by Rebecca Lawson and others, many adults who ride bicycles could not place the frame, chain and pedals correctly in a sketch, even though they had seen thousands of bicycles. Recognising an object is cheap and generating its structure is not, which is the same asymmetry seen at the level of ideas.
Why a feed is a good place to catch it
Online arguments about policy, technology and economics are conducted at a distance from any mechanism. A confident post about how a tariff, a vaccine or a language model works usually rewards a strong conclusion, and nothing in the format asks for the intermediate steps. A reply that says “walk me through how that works” is often the first time the poster is made to produce them. The pattern is a close relative of the overconfidence of the half-informed and of bikeshedding, where the easy-to-grasp question draws the most opinions.
Explain how, not why
The practical lesson of the 2013 studies is the difference between two questions. “Why do you support this?” invites a list of reasons and tends to harden the position. “How exactly would it work?” invites a causal chain and tends to expose the gaps. The second question is a gentle form of what steelmanning asks of both sides, and it is hard to read as an attack because it is the same question I can ask of myself.
Critique
- The political moderation effect has not replicated cleanly. In 2021 Jarret Crawford and John Ruscio reported three preregistered attempts to repeat the 2013 experiments and did not find that explaining a policy made people more moderate, although the drop in self-rated understanding was more robust. The core device finding is better supported than its political extension.
- Understanding is not all-or-nothing. Nobody needs to explain a zipper from first principles to use one well. Relying on experts and tools is rational, and the community-of-knowledge view treats it as a feature of human cognition, not a defect. The error lies in overrating one's own part of the load, not in sharing it.
- Rating scales are fuzzy. A person who rates their understanding a five before explaining and a three afterwards may have been using the scale differently the second time. Some of the drop could reflect a recalibration of what counts as understanding, not a loss of confidence in a fixed belief.
Impact
The useful habit is to test an opinion by trying to explain the mechanism behind it, out loud or in writing, before posting it. If I cannot get from cause to effect in a few plain steps, I probably hold a feeling about the topic more than an explanation of it, and I can say so or go and read. The same test is a courteous one to offer others, framed as curiosity about how something works, which usually keeps a thread friendlier than a demand for evidence would. It is also a reason to treat the confidence of strangers online, including mine, as a weak signal about what is actually understood.