Skill Ratings Raise Risk-Taking 22% in 600 Puzzle Runs

What happens to a person's willingness to take a risk when they are told, explicitly, that they are good at something? The question sounds almost trivial until you notice how many systems now hand out competence labels — leaderboards, percentile rankings, badges, tiered skill levels — before the person has done anything risky at all. A recent line of work on puzzle-solving behavior offers a useful test case: when participants were given a skill rating before a series of 600 timed puzzle runs, their rate of choosing the harder, higher-variance option rose by roughly 22 percent compared with a control group given no rating. The interesting part is not the number. It is what the number implies about how feedback loops, self-concept, and uncertainty interact.

The Design Behind the 22 Percent

The setup matters more than the headline. Participants solved short logic puzzles under time pressure. Before each run, they chose between two versions: a "safe" version with a modest, predictable point value, and a "stretch" version worth substantially more but solvable only about half the time. Both versions drew on the same underlying skill. The only difference was variance.

Half the participants received a skill rating after a calibration block — a percentile score presented as a measure of their puzzle ability. The other half received neutral feedback about their completion speed with no comparative ranking. Across 600 runs, the rated group selected the stretch option 22 percent more often. They also persisted longer after failures, and their post-failure accuracy dipped less than the control group's.

This is a small effect in absolute terms and a large one in behavioral terms. A 22 percent shift in risk selection, sustained across hundreds of trials, is the kind of change that compounds. It is also the kind of change that organizations, educators, and product designers routinely induce without intending to.

Why a Rating Changes What Feels Risky

Variable-ratio reinforcement and the shape of confidence

The psychology here is not mysterious. B.F. Skinner's work on variable-ratio reinforcement established that unpredictable reward schedules produce the most persistent behavior — the same principle that makes intermittent feedback so sticky. A skill rating is not itself a reward, but it functions as one: it is a summary judgment delivered at an arbitrary moment, and it reframes every subsequent choice as a chance to confirm or disconfirm it.

Once a person holds a competence label, the stretch option stops being a gamble on ability and becomes a test of a claim. That shift in framing is doing most of the work.

Loss aversion, inverted

Kahneman and Tversky's loss aversion predicts that people overweight potential losses relative to equivalent gains. A skill rating appears to loosen that asymmetry — not by making people reckless, but by changing what counts as a loss. If you believe you are in the 80th percentile, declining the harder option feels like a loss of identity rather than a preservation of points. The reference point moves from "points accumulated" to "self-image maintained."

This is why the effect shows up in persistence as well as selection. Participants were not just choosing harder puzzles; they were protecting a story about themselves.

The competence-confidence gap

There is a well-documented gap between measured competence and self-assessed competence, and it runs in both directions. The puzzle data suggests that external ratings can close that gap artificially — and quickly. A single percentile score, delivered once, was enough to shift 600 subsequent decisions. That is a remarkably short feedback loop for a remarkably durable effect.

What the Puzzle Data Actually Shows — and What It Doesn't

The 600-run figure is worth pausing on. It is large enough to smooth out warm-up effects and individual variance, but it is also a laboratory number. Puzzle tasks have clean success criteria, immediate feedback, and no social cost to failure. Real-world risk-taking — a career change, a public commitment, a capital allocation — has none of those properties. The 22 percent figure should be read as a directional signal, not a transferable constant.

A useful comparison comes from the literature on "expertise-induced" overconfidence. Studies of physicians, financial analysts, and air traffic controllers have repeatedly found that domain expertise correlates with faster decisions and, in some conditions, reduced sensitivity to disconfirming evidence. The puzzle study isolates one input — a competence label — and shows it can produce a version of that effect in under an hour. That is the finding worth carrying forward.

Where This Intersects With Everyday Reading and Decision-Making

If you read widely in behavioral economics and decision science, you have probably encountered this pattern before under different names: the "hot hand" debates in sports, the "escalation of commitment" literature in management, the "nudge" literature in public policy. All of them describe the same underlying mechanism — a signal about the self changes the calculus of the next choice.

What makes the puzzle study useful as a bridge is its economy. It strips the mechanism down to one variable and one outcome. If you want to understand why a leaderboard changes user behavior, why a promotion changes negotiation posture, or why a public commitment makes people double down, the 600-run experiment is a clean reference point.

A practical implication for anyone who designs feedback

The forward-looking question is not whether skill ratings raise risk-taking. It is whether the risk-taking they raise is well-calibrated. A 22 percent increase in stretch-option selection is beneficial when the stretch option is genuinely winnable and the feedback is honest. It is harmful when the rating is inflated, the task is not actually skill-based, or failure carries real cost.

The design lesson is to treat ratings as interventions, not decorations. If you attach a competence label to a person, you are changing their subsequent choices. That is a lever, and levers deserve to be pulled deliberately.

For readers, the corresponding lesson is to notice when a rating has been applied to you — and to ask whether the harder option you are now drawn to is harder because it is better, or harder because someone told you that you could handle it.