CAN COMPLETE INFORMATION ON PAST COHORT PERFORMANCE REDUCE STUDENTS’ OVERCONFIDENCE? AN EXPERIMENTAL APPROACH USING A REAL-EFFORT TASK

This work investigates the influence of information about the performance of previous cohorts on the accuracy of predictions. Overconfidence is a cognitive bias that can lead students to underestimate the effort required to obtain the desired grades, affecting their academic success. To address this problem, I carried out a lab experiment using a between-subjects design, dividing the participants into four treatments. When predicting, the control groups (T1 and T2) had no information, while the treatment groups (T3 and T4) had information that included visual data on the distribution of performance from previous cohorts. The experiment followed a methodology derived from Abeler et al. (2011), using a real-effort task involving counting zeros in tables to assess performance accuracy. Participants’ mindsets and self-esteem were also measured through specific questions, allowing categorisation into deliberative or implemental mindsets and levels of self-esteem through the Rosenberg Self-Esteem Scale. The main conclusion of the study was that providing information immediately reduces prediction errors, with the treatment group showing a median prediction error of 0. This result emphasises the potential benefits of such interventions in educational contexts to improve students' grade predictions and enhance their academic performance. The final model, with 16 predictors, explained 86.6% of the variability in under/overconfidence, revealing the importance of actual performance and mindset. A Bayesian network model also indicated that actual performance on the task was crucial, although residual patterns suggested areas for further investigation.