Article — "The Confidence Gap May Be a Clock, Not a Competence Problem: Reflection Time and Measurement": Article — "The Confidence Gap May Be a Clock, Not a Competence Problem: Reflection Time and Measurement": Hand two underconfident people the exact same extra seconds to reflect before they rate themselves, and watch them split. One becomes better calibrated; the other sinks deeper into doubt.
Same label. Opposite response to the same manipulation. So which one has the confidence problem?
This is the strange result at the center of a 2026 study by Sucharit Katyal and Stephen Fleming, run on 1,447 online participants. They built a dynamic computational model to track how underconfidence unfolds over time, and found that "underconfidence" was never one thing.
Underconfidence tied to anxiety symptoms got worse the longer people took to make their metacognitive judgments. More time to introspect, more damage.
Gender-related underconfidence did the reverse. The more reflection time women were given, the more that underconfidence shrank.
Two people arrive at the same low number. In the study's model, greater metacognitive judgment time is associated with pushing anxiety-related and gender-related underconfidence in opposite directions.
That is not a fact about competence. It is a fact about the clock.
Two people arrive at the same low number. That is not a fact about competence. It is a fact about the clock.
Katyal and Fleming's model tells you why. Both anxiety and gender involved a shift in confidence criteria—the internal bar you set before you'll call yourself sure. But only anxiety involved something worse: a temporal accumulation of negatively biased evidence about one's own ability.
For the anxious person, per the study, thinking longer is associated with accumulating more negatively biased evidence about one's ability. For women in the sample, the extra time did the opposite—it let a considered judgment override a fast, prepotent bias.
So the famous "confidence gap" between women and men, popularized in outlets like The Atlantic as a story about women who "kept our heads down and played by the rules," may be, per this study of metacognitive judgment time, at least partly related to how long the instrument let people think.
Which means the number your test produces is not a clean readout of what someone can do. It is the output of a timed process, and the timing co-authors the result.
Consider what confidence actually is at the level of mechanism. It is not a lookup of stored ability. It is post-decisional—it forms after you've acted, as you keep processing what you just did.
This is why timing changes everything. In a 2024 memory study by Klopukh and Darby, confidence reportedly matched accuracy more tightly when it was probed after a three-second delay than when it was probed immediately.
Three seconds. Per the study, that is associated with a difference between a confidence rating that tracks performance more closely and one that doesn't.
If a three-second pause can improve calibration, then any test that probes confidence immediately is not measuring a stable trait. It is measuring a trait plus a stopwatch setting nobody chose deliberately.
And the direction of the timing effect isn't fixed either. The diminishing criterion model, developed to explain metacognition in complex tasks, shows that on hard problems people don't hold their standard steady as they invest time—they compromise it.
Per the model, the longer you work a difficult problem, the more you relax the bar for what counts as "confident enough." Which produces the counterintuitive result that more time can mean lower confidence—not because you learned you were wrong, but because the goalpost moved.
Put the pieces together and the picture is unsettling. Confidence ratings depend on when you ask, how long you let the person accumulate evidence, and whether that evidence stream is neutral or self-poisoning.
None of those are properties of the person's skill. All of them are properties of your measurement procedure.
So before you conclude that a group is underconfident and needs fixing, you have to rule out a duller explanation: your instrument handed them a different clock.
This is not a quirk of psychology labs. It is a general disease of measuring instruments that ask crowds to rate something, and the market has its own version.
Researchers at Clearer Thinking reportedly had 327 participants rate 185 items on two dimensions: hype—how cool, exciting, or status-conferring a thing is—and value, how much it actually helps people get what they truly want.
If hype were a rough proxy for value, the two would move together. They correlated at 0.23.
Statistically related, largely independent. The crowd's excitement about a thing tells you almost nothing reliable about whether the thing is good.
The examples are brutal. According to Clearer Thinking, Nikola Corporation staged a promotional video of its zero-emissions truck "driving majestically" through the Utah desert—except the truck wasn't driving, it was rolling down a hill.
The company briefly outvalued Ford, was worth over $30 billion, and produced roughly 600 trucks in ten years before filing for bankruptcy. Enormous hype, almost no value.
Run it in reverse and you get Ignaz Semmelweis, whose 1840s discovery that handwashing slashed maternal death rates took twenty years to be accepted. Enormous value, almost no hype, and women died in the interval.
Here is the connection to the confidence gap. When a crowd rates hype, it is doing the fast, prepotent thing—registering excitement, status, coolness—arguably the same way a snap confidence judgment registers a gut feeling before reflection kicks in.
Value is the considered judgment. It takes the delay, the accumulation, the deliberate second pass. And when the instrument only captures the excited response, it reads out hype and calls it worth—just as a confidence test that probes immediately reads out a gut bias and calls it ability.
In both cases the measurement procedure quietly substitutes the fast signal for the slow one, and the two only weakly coincide.
Which points to a discipline, not a slogan.
When a gap moves as you change reflection time or rating format, your first hypothesis should be a method effect, not a competence story. The Katyal and Fleming split is the proof of concept: the same "underconfidence" label fractured into two mechanisms the moment they touched the clock.
So redesign the instrument before you redesign the people. Standardize the reflection window so everyone gets the same number of seconds instead of whatever their anxiety or habit hands them.
Add a deliberate delay when you actually want calibration—the three-second pause is close to free. And separate the excited response from the considered one, the way hype and value have to be asked as two questions because they answer to two different processes.
The temptation, always, is to read a low number as a verdict on a person. She's underconfident. They didn't see the value. He was too sure of himself.
But a number that flips its meaning when you add a few seconds of thinking was never a verdict. It was a timestamp wearing a verdict's clothes.
Fix the clock first. Then find out who's actually in the room.