Beyond Correct vs. Incorrect: Investigating Graded Feedback Processing in a More Naturalistic Skill Learning Paradigm

According to reinforcement learning theory, humans make decisions and behavior adaptations based on reward- prediction errors, the difference between actual and anticipated outcomes. In human research, rewardprediction errors have been studied through the proxy measure of the ERP component reward positivity (RewP). Studies investigating the relationship between RewP and behavior have mostly relied on paradigms in which outcome feedback is dichotomous (e.g., correct vs. incorrect). However, in naturalist skill learning settings, outcome feedback is presented in a more graded manner (e.g., “you missed the target by about 40 cm”). Thus, it is important to examine the RewP in such naturalistic settings for a more ecologically- valid assessment of learning processes. Accordingly, in the present study we recorded EEG from 64 participants while they practiced a nondominant arm beanbag tossing task and received graded feedback. We quantified the RewP elicited by the graded feedback on a trial- by- trial basis and used a linear mixed- effects model to regress it on the accuracy of each trial while also testing whether this relationship changed over time or was dependent on participants’ skill level. In line with previous findings, RewP was more positive after more accurate as opposed to less accurate trials. Therefore, results of this study support the use of the RewP as a measure of reward- prediction errors in more naturalist skill learning paradigms and shed light on the dynamic changes in reward- prediction errors during practice.