Confidence, Correcting Metacognitive Errors, and Applications in Higher Education
- Silva, Paulina N.
- Advisor(s): Sarnecka, Barbara W.
Abstract
It is useful to know ourselves - to be able to reflect accurately on the state of our own knowledge, skills, and cognitive processes; in other words, to have good metacognition. But it turns out that metacognitive errors (e.g., being under- or overconfident) also provide us with opportunities to learn about what information we do and do not know. In this dissertation, colleagues and I ask, what kind of metacognitive errors are beneficial, and in what way? Past research on response errors (whether you answer a question correctly or not) has shown that how likely a response error is to be corrected depends on how confident the learner was when they made the response error. Specifically, the more confident you are when you make a response error, the more likely you are to give the correct response on a later re-test. This is known as the hypercorrection effect. In this dissertation, I explore whether metacognitive errors are subject to an analogous effect; in other words, I explore whether the chances of correcting a metacognitive error depend on how confident the learner was when they made the metacognitive error. Using a trial-and-error learning task with adult participants, I show that metacognitive errors are subject to the inverse of this effect: under-confident responses are more likely to lead to accurate predictions later on than overconfident responses. I find further evidence of this effect in two secondary analyses using undergraduate student data from an online learning platform, first using the same analysis that was used on the experimental data, and again using a Markov process approach that considered all possible responses students could make, not just their metacognitive errors. In another experiment, I explore how uncertain outcomes affect this inverse hypercorrection and discuss how uncertain outcomes may make metacognitive judgments particularly difficult, despite the fact that participants showed evidence of learning the probabilities associated with the outcomes. The secondary analyses and an additional survey study of statistics instructors also demonstrate how basic research can be applied to enhance education practices and aid in the evaluation and development of educational materials.