Illusion of explanatory depth

The zipper

Two friends are packing for a trip when one jacket’s zipper jams. “Zippers are simple,” one says. “I know exactly how they work.” Asked how the slider actually locks the teeth together, they start to answer, stop, and say: “Well, the teeth sort of... hook into each other when you pull it.”

The confidence came from years of using zippers and seeing them up close. But using something and seeing it aren’t the same as having a model of how it works, and the attempt to explain shows the model was never there. This is the core case: a device so familiar it feels understood.

The running toilet

A homeowner hears the toilet running all night and decides not to call a plumber: “It’s just a tank and a flush lever. I understand how it works.” With the lid off, they can’t tell whether the problem is the flapper, the float or the fill valve, or how any of them decide when the water stops.

Here the illusion has a practical cost: it shaped a decision. The homeowner’s sense of understanding was built on the visible, simple-looking parts. How those parts interact to control the water level is the mechanism, and it’s hidden inside the tank. Seeing the parts of a system is easily mistaken for knowing how it works.

Knowing the names of the parts

A keen cyclist can name every component on a road bike: derailleur, cassette, chainring, cable stop. A friend asks how shifting actually moves the chain to a bigger gear, and the cyclist says: “The derailleur does it.” Pressed on how the derailleur does it, they can’t say.

This is the less obvious form, and it can affect people with real expertise in something nearby. The cyclist’s knowledge is genuine, but it’s knowledge of labels and of how to use the bike. Knowing what a part is called and what it’s for feels like knowing how it works, one level down.

• When you claim to know how to use something, not how it works. Being confident you can operate a device, follow a recipe or recount a film’s plot is a different kind of knowledge. People judge those far more accurately.
• When you’ve actually explained it. Confidence in an understanding you have recently put into words, taught or tested against an expert account is earned.
• When you know you’re relying on someone else’s understanding. Using a thing without understanding it, and trusting that engineers or plumbers do, is how most knowledge works. The error is believing the understanding is in your own head.

The test: could I write out, step by step, how it works, including the parts I can’t see?

Knowing how to use it

Someone who has used a pressure cooker for years tells a guest: “I know how this works. Lock the lid, bring it up to pressure, then release it through the valve before opening it.”

This sounds like a confident claim to understand a device. But what they describe is a procedure, and they can in fact carry it out, step by step. They aren’t claiming to know the physics of why pressure cooks food faster. That’s the use, not mechanism condition above, and studies of the illusion find that people judge this kind of knowledge well.

The explanation that holds up

A bike mechanic says she understands how a derailleur works, and when asked, explains how the cable pulls the cage sideways against a spring, how the limit screws stop it, and why the chain climbs to the next cog.

The confidence is the same as the cyclist’s in the example above, but here it survives the test of explaining. She has the mechanism, not just the vocabulary. That’s the you’ve actually explained it condition.

Leonid Rozenblit and Frank Keil, who named the illusion, proposed several reasons it’s stronger for explanations than for other kinds of knowledge:

• The object does the remembering. When a device is in front of you, its visible parts support a feeling that you’ve grasped it, as if seeing were storing. In their studies, the more of a device’s parts were visible rather than hidden, and the more of its parts people could name, the more confident people were at the outset.
• Understanding one level feels like understanding all of them. Explanations are layered: the brakes stop the car, the pads grip the disc, and so on down. Grasping the top level, or knowing a part’s name, can pass for understanding the levels below.
• There’s no clear finish line. You can check whether you know a capital city or how to bake a cake by producing the answer. An explanation has no obvious end point, so it’s hard to test yourself without actually trying.
• People rarely explain things. With little practice, there’s little feedback about how well you’d do.

Later work supports parts of this. Jeffrey Zemla and Daniel Corral found that people rated their understanding of devices lower when shown the internal parts rather than the whole object, or when quizzed on mechanism rather than on use, consistent with people blending knowing how to use something with knowing how it works.

The illusion is related to, but distinct from, general Overconfidence. Rozenblit and Keil found that the same people were well calibrated about procedures and film plots and only modestly overconfident about facts, so it isn’t simply a tendency to overrate everything one knows. It also differs from the Dunning–Kruger effect, which concerns people with low skill misjudging their performance: the illusion of explanatory depth appeared in graduate students at an elite university. When the illusion meets an explanation that doesn’t fit one’s shallow model, it can surface as incredulity: the explanation seems impossible because the gap it fills was never noticed. A person who does understand a mechanism faces the opposite difficulty, the Curse of knowledge: imagining how little someone else understands.

• Try to explain it, step by step, before relying on your understanding. This is the tested remedy: writing out a mechanistic explanation lowered people’s ratings of their own understanding across many studies. One set of experiments found that explaining any mechanism lowered ratings of unrelated ones too, so a single attempt may recalibrate more broadly.
• Ask about the hidden parts. If you can describe only what you can see, you may understand less than it feels.
• Don’t expect the realization to change much else. In three preregistered replications, asking people to explain how complex public policies would work made them rate their understanding lower, but did not make their opinions more moderate, as an earlier study had reported.

Status: replicates robustly. The drop in people’s ratings of their own understanding after they try to explain something has been found again in several independent sets of preregistered studies. What that drop measures, and why it happens, are less settled.

• Rozenblit and Keil (2002) ran 12 studies. In the core design, graduate and undergraduate students rated how well they understood 48 items, then wrote step-by-step explanations of devices such as a speedometer, a zipper, a flush toilet and a cylinder lock, answered a diagnostic question (for example, how you would pick the lock), and read an expert explanation, rating their understanding after each step. Ratings fell. The pattern held at a less selective state university and with other devices. Independent judges who rated the explanations scored them about where the participants’ lowered ratings ended up. Warning people in advance that they’d be tested reduced the drop but didn’t remove it. Knowledge of facts (capital cities) showed a small drop; knowledge of procedures and of film plots showed none; explanations of natural phenomena showed a large one.
• Lawson (2006) tested understanding directly rather than through self-ratings: asked about basic bicycle design, people made frequent, serious mistakes, such as putting the chain around the front wheel as well as the back one. Errors were fewer, but still present, among bicycle experts.
• Fernbach and colleagues (2013) found the same drop in understanding ratings for complex public policies and reported that explaining also moderated people’s opinions. Crawford and Ruscio (2021) ran three preregistered close replications (306, 405 and 343 participants). People again rated their understanding lower after explaining, but the effects on opinions did not replicate.
• Meyers and colleagues (2023), in three preregistered studies, found the drop again, but also that explaining one thing (say, how a zipper works) lowered ratings of unrelated things (how snow forms) just as much. In their second and third studies, a control group that explained nothing at all also lowered its ratings, though only slightly (about d = 0.10) and by less than those who explained. The authors note that simply being asked twice may account for some of the effects published so far, and that the usual drop doesn’t come only from discovering gaps in the specific thing explained. (Here d measures the size of a difference; 0.2 is conventionally “small”.)
• Körner, Schütz and Petersen (2024), in three preregistered studies with 607 participants, found strong evidence for the illusion with devices, including when outside observers rated how good the explanations were. They found only weak evidential value for an earlier proposal that thinking in abstract terms is what drives it.

What remains uncertain is the mechanism, and how much of the standard measure reflects the illusion rather than people’s general tendency to revise ratings downward when asked again. Meyers and colleagues’ findings suggest the act of explaining produces a broad sense of “I know less than I thought”, not only a discovery about one object. That people’s sense of understanding outruns what they can explain, at least for devices and natural processes, is well supported.