When Choosing the Harder Task Pays—and When Confidence Overshoots

The attraction of a harder task

Choosing a demanding assignment is not necessarily reckless. If success brings a larger reward, the harder option can be the better bet for someone equipped to complete it. The difficulty is judging when that is true.

In Choosing Complexity: Cognitive Ability, Overconfidence, and Task Choice, Purdue University economists David Gill and Zhongheng Qiao investigate this decision using a theoretical model and a preregistered laboratory experiment. Their November 2026 arXiv preprint reports that both higher measured cognitive ability and greater overconfidence predict choosing more complex tasks. These are associations, not evidence that changing either trait would cause someone to choose differently.

Turning complexity into a choice

The model describes a simple tradeoff: complexity raises the reward for success but also increases the probability of a mistake. Higher cognitive ability lowers mistake probabilities. Under the model’s assumptions, this makes greater complexity relatively more attractive as ability increases.

The experiment involved 307 students at Purdue in November 2025. Participants first completed an 11-question visual-pattern reasoning test, then estimated how many answers they had correct. The researchers defined overconfidence as the estimated score minus the actual score. Negative values therefore represented underconfidence, rather than a separate personality category.

After five practice tasks, each participant made two paid choices. Each choice offered a less complex and a more complex rule-following task. Participants typed letters according to rules responding to computer-generated letters, with the rules remaining visible. Success required all 20 responses to be correct within 60 seconds.

In one pair, success paid $5 for the less complex task or $9 for the more complex one. In the other, the corresponding payments were $6.50 and $11. Failure always paid $1. The order of the two pairs was randomized.

Ability and confidence both predicted ambition

Participants with higher reasoning-test scores were more likely to select the more complex task in both pairs. Higher scores also predicted higher realized task earnings.

Confidence mattered beyond measured ability. Comparing people with the same reasoning score, those who estimated their performance more generously were more likely to choose the harder option. Across the overconfidence range containing most participants—from underestimating their score by two answers to overestimating it by two—the pooled regression associated that difference with a roughly 21-percentage-point increase in choosing greater complexity.

That finding does not establish that confidence itself caused the choice. Ability and confidence were measured, not experimentally assigned, and the reasoning test need not capture every skill relevant to the tasks.

Was the harder choice actually better?

Evaluating choices required more than counting successes. Each participant attempted only the selected task, so the researchers could not directly observe how that person would have performed on the alternative. A failure on a sensible gamble also does not make the original decision wrong.

To address this selection problem, the authors jointly modeled task choices and success, including an unobserved person-level factor that could influence both. They then predicted each participant’s success probability on each option and calculated expected earnings: the success payment weighted by its probability, plus the failure payment weighted by its probability.

When evaluating one pair, they excluded that participant’s outcome in that pair from the individual-factor prediction, while retaining the estimated model parameters. This reduced circularity from judging a decision using its own outcome. They also carried estimation uncertainty through the analysis using 500 bootstrap replications.

Under this model-based benchmark, about 75% of decisions selected the task with higher expected earnings. All decisions classified as suboptimal favored the more complex task; none favored excessive caution.

What the findings do—and do not—support

Evidence that higher ability improved decision quality was much weaker than evidence linking it to harder choices. The ability–efficiency association reached only the 10% significance threshold in the less complex pair. Ability did not significantly predict the simpler yes-or-no measure of choosing optimally.

The optimality findings also depend on modeling assumptions. The unrestricted statistical model did not converge, so the authors shared coefficients across choice equations and across success equations, allowing separate intercepts. Moreover, maximizing expected money is the study’s benchmark, not a complete account of participants’ preferences.

These short laboratory tasks with students do not establish how employees choose long-term projects. The work is a preprint, and its proposed workplace remedies—clear success-rate information, trial assignments and feedback—were not tested here. Its practical message is therefore a proposal: preserve ambitious options, while helping people assess whether the larger prize justifies their probability of failure.