Sarah B. Malykke¹, Audrey Siqi-Liu¹, Kelvin S. Oie³, Dwight J. Kravitz^1,2, Stephen R. Mitroff¹; ¹The George Washington University, ²US National Science Foundation (SBE/BCS), ³U.S. DEVCOM Army Research Laboratory

Errors are inevitable in human behavior, and thus it is critical to understand their causes and effects, especially in applied settings. Prior research has shown that individuals often speed up before an error, perhaps due to attentional lapses or diminished cognitive control. Conversely, after an error, individuals typically slow down, which may reflect disrupted processing or an adaptive increase in caution. While much research has examined pre-error speeding or post-error slowing independently, here we examined both phenomena simultaneously to better understand how errors emerge and disrupt performance, with a specific goal of exploring how pre- and post-error behaviors might relate. We examined error performance in a very large dataset (n=17,784) from a mobile game (Airport Scanner), where individuals completed an object-sorting task which involved categorizing items by touching the item on the screen and swiping it up or down into a bin. A linear mixed-effects model revealed consistent speeding leading up to an error (as early as 12 trials before), followed by significant and long-lasting post-error slowing (up to 15 trials after). Correlation matrices focused on trial-to-trial performance revealed strong relationships between adjacent trials leading up to an error, reflecting systematic behavior across participants as they approached the error. However, upon the error, correlations between adjacent trials were disrupted, suggesting a breakdown in the trial-by-trial behavioral consistency. Post-error, the trial-by-trial correlations gradually recovered demonstrating the reestablishment of response time carryover effects. These relationships were absent in a group of matched participants who did not make an error. These findings highlight the importance of studying pre-error speeding and post-error slowing together to better understand how errors influence task performance over time—and how pre-error performance may be indicative of both error occurrence and post-error behavior.

Acknowledgements: W911NF-23-2-0210, W911NF-24-2-0188