Cocky kids: Sticky overconfidence among 4-to-6-year-olds

The tendency to be excessively confident in one's abilities has been observed across professions and cultures. However, the overwhelming majority of studies on overconfidence rely on adult samples. Yet there is no reason to believe that we develop such an omnipresent bias only once we come of age.

Like Comment
Read the paper

It may be fair to say that designing and conducting a solid behavioral experiment is a nontrivial task. This is especially true whenever participants are asked to physically execute a given prompt. Now, consider the same setting only that your participants have a significantly shorter attention span, need to be intrinsically motivated throughout the experiment, unambiguously vocalize their displeasure, and have high expectations when it comes to the fun factor of your experimental task.

I knew that conducting a randomized controlled trial in judgment and decision-making with children would be challenging. It took months to review the literature and countless consultations with experts from different fields to find a suitable experimental task (shout out to Heather Kirkorian). Prior research largely relied on memory games that lack the central elements of risk and ambiguity which characterize decision-making under uncertainty. I decided to adapt the Children's Gambling Task, a card game that was originally designed by Kerr and Zelazo (2004) to study the development of hot executive functions in 3-to-4-year-olds. The findings seem to support the targeted age range of this experiment. I added a number of age-appropriate confidence indicators in order to be able to compare perceived performance to a child's actual outcome.

However, the purpose of this study was not only to test overconfidence among children in a more applicable experimental setting but also to causally examine whether a sublet video intervention could reduce excessive confidence in 4-to-6-year-olds. Designing such a clip was equally demanding. The aim was to create a sequence that may challenge a child's suboptimal card game strategy without threatening their ego. The latter is necessary to avoid framing effects. Furthermore, balancing both experimental arms was given high priority in order to minimize statistical noise. For instance, deep neural networks were used to generate the voices of all narrators and protagonists. Furthermore, the displayed characters were gender-matched with the participants. Lastly, both, the patterns on the back of the cards as well as the display order of the decks, were fully randomized. The video had to mirror all of these aspects to provide a realistic card game situation that children in the intervention group could relate to (shout out to Jenny P.). In retrospect, the video may have been too sublet and guided reflection tasks might offer a promising avenue for future research.

Figure 1. Intervention video for female participants with triangles to the left and waves to the right. Only girls with the exact same card and deck randomization would see this version of the video.

Despite all these preparations, there was still a mountain to climb. A pretest revealed that the experimental setting was appropriate. However, the logistics of the data collection process was not. Specifically, it became clear that a single researcher could not handle the card game as well as the input survey that would record the responses. Consequently, two trained researchers, who needed to be unaware of the underlying hypothesis, were required for every single case. In essence, the study would not have been possible without substantial institutional support (shout out to Dee Warmath) and pronounced community acceptance (shout out to Amy Wagner and the UW Child Development Lab, Eagle's Wing Child Care, Bernie's Place, and the Madison Children's Museum). I was incredibly fortunate to work with a team of highly motivated and professional researchers who were able to create a scientifically sound yet welcoming experience for all participants (shout out to Katie S., Breanna B., Whitney W., and Megan L.).

The results are, as so often in the behavioral and social sciences, an illustrious bag of novel insights. For instance, girls outperformed boys as they adopted a more risk averse and, ultimately, more successful strategy. However, by the end of the experiment, girls were notably more overconfident in their future performance compared to boys. How is that possible? On average, female participants seemed to closely align their estimates with their current performance. In other words, they heavily discounted information from the more distant past. The results suggest that girls overestimated their abilities if they had a winning streak and underestimated themselves whenever they lost a few times in a row. A gender effect in early childhood has never been consistently shown in prior studies on overconfidence. It would be interesting to see whether the findings continue to hold with an adult sample and a complex adaptation of the Gambling Task.

You can read the full paper here.


Kerr, A., & Zelazo, P. D. (2004). Development of “hot” executive function: The children’s gambling task. Brain and cognition, 55(1), 148-157.

Dominik M Piehlmaier

Lecturer (Assistant Professor), University of Sussex Business School