Young children often believe they can do more than they realistically can. That confidence can be useful: Children who expect to succeed may be more willing to approach unfamiliar tasks, persist through frustration, and explore their surroundings. But inflated beliefs can also encourage risky choices, particularly when children misjudge their physical abilities. Now, new research suggests that one of the most appealing claims about childhood overconfidence—that it improves learning—may be based on a statistical illusion rather than a genuine psychological advantage.
In a study published in Psychological Science, researchers led by Mengqi Hu of Beijing Normal University revisited the idea that overconfident children show greater gains in memory performance. Their findings indicate that children who appear to improve more over time may simply be those who started with unusually low scores. When the researchers experimentally increased or decreased children’s confidence, the manipulation changed how capable the children believed themselves to be, but it did not alter how their memory performance developed during the task.
The question has its roots in a widely cited 2007 study of children’s memory. In that experiment, children predicted how many words they would remember before studying and recalling lists of words. Most children overestimated their performance, but those classified as highly overconfident appeared to show more favorable changes in recall than children who were less overconfident. The result was interpreted as evidence that confidence might be adaptive: Children who believed they could succeed might engage more deeply with the task, persist longer, and ultimately learn more.
Hu and her colleagues first attempted to replicate that finding with 30 children between 6 and 8 years old. At first glance, their results appeared consistent with the earlier report. The children categorized as highly overconfident showed a smaller decline in recall than those categorized as less overconfident. A closer analysis, however, revealed an important difference between the groups at the beginning of the task. The highly overconfident children had lower initial recall scores, while the groups did not differ in their later recall performance. The apparent improvement therefore seemed to reflect where children started, rather than how much they learned.
To investigate this possibility, the researchers used computer simulations involving data from 1,000 hypothetical children under several different conditions. In one model, overconfidence had no effect whatsoever on learning. Even under that assumption, applying the analysis used in the 2007 study produced an apparent advantage for the highly overconfident group. The simulation showed that the original statistical procedure could generate the same pattern researchers had interpreted as evidence of improved learning, even when confidence played no causal role.
The mechanism behind the result is a familiar statistical phenomenon called regression to the mean. Hu’s team calculated overconfidence by subtracting each child’s actual recall score from the number of words the child predicted remembering. Because the measure included actual recall, children with unusually low scores were more likely to receive large positive overconfidence scores and be placed in the highly overconfident group. Children with unusually high initial scores were more likely to be classified as less overconfident.
The researchers then assessed change in performance by comparing later recall with those same initial scores. But extreme measurements often move closer to the average when repeated, even without any real change caused by learning or treatment. A child who performs unusually poorly on one trial may perform closer to average on the next simply because of normal fluctuations in attention, guessing, memory retrieval, or measurement error. When low-scoring children are disproportionately placed in the overconfident group, this natural movement toward average performance can make that group appear to improve more.
The team next conducted an experiment designed to separate confidence from the statistical features of the original grouping method. Sixty-four children were randomly assigned to receive feedback intended to make them feel either highly capable or poorly capable at remembering information. The children were shown what they were told was an EEG-based memory assessment. Before each task, some were falsely informed that the device indicated excellent memory, while others were told that it indicated poor memory. The feedback successfully shifted the children’s confidence, creating a direct experimental test of whether confidence itself changed memory performance.
Despite the difference in reported confidence, the two groups showed similar patterns of recall across the task. Children told they had excellent memories did not improve more than those told their memories were poor. Yet when the researchers analyzed the data using the earlier study’s method—sorting children according to the difference between predicted and actual performance—the apparent benefit of overconfidence reappeared. This contrast is central to the researchers’ conclusion: The effect could be produced by the analysis even when experimentally changing confidence had no measurable effect on learning.
The findings do not mean that encouragement, optimism, or positive expectations are irrelevant to children’s development. Confidence can influence emotions, willingness to attempt difficult activities, and reactions to failure. But the study cautions against deliberately exaggerating children’s abilities on the assumption that inaccurate beliefs will automatically improve learning. Feedback may be more effective when it recognizes genuine effort, useful strategies, and measurable progress while also identifying areas that need improvement. For parents and educators, the message is not to discourage children, but to distinguish constructive support from claims of competence that are disconnected from performance. In this case, a celebrated learning advantage may have been created not by children’s beliefs, but by the mathematics used to classify them.
Subject of Research: The relationship between children’s overconfidence and learning performance, with a focus on memory recall, experimental confidence manipulation, and regression to the mean.
Article Title: Does Overconfidence Really Confer Adaptive Benefits to Children’s Learning?
Web References: https://journals.sagepub.com/doi/abs/10.1177/09567976261455283; https://www.sciencedirect.com/science/article/abs/pii/S0885201406000979
References: Hu, M., Zhao, W., Cao, M., Shanks, D. R., Hu, X., Luo, L., & Yang, C. (2026). “Does Overconfidence Really Confer Adaptive Benefits to Children’s Learning?” Psychological Science. DOI: 10.1177/09567976261455283
Keywords: Children, overconfidence, learning, memory, confidence, regression to the mean, statistical illusion, cognitive development, education, psychology
