Error-related Processes and Adaptive Behaviour Change
The ability to monitor ongoing performance is a function critical to adapting behaviour. Detecting an error may signal that the task has increased in difficulty and adapting our approach to the task is necessary to improve performance, or avoid making the same mistake again. There is an ongoing need to examine how we recognise and adapt our behaviour following an error in performance. Further, the question of whether post-error slowing is non-functional requires greater exploration.
Research Questions / Hypotheses
RQ: Does post-error slowing differ as a function of the uncertainty of errors? Hypotheses: We expect greater post-error slowing following errors in a condition where lure trials are presented randomly, as opposed to a condition where lures are presented in a predictable manner. We expect greater post-error slowing following aware errors than unaware errors.
80 participants completed the study. No exclusions have been applied.
A serial choice reaction time task was developed with modifications based on the error awareness task (EAT; Hester et al., 2005). The task was designed and presented using jsPsych and responses were recorded via the computer keyboard. Colour words were presented in a serial stream in either a congruent or incongruent colour. Participants were required to respond to the presentation of each stimulus by pressing the letter ‘A’, unless the stimulus was a lure. Lure trials occurred when the word and colour were congruent (Colour lure) or when the same word was presented on two consecutive trials (Repeat lure). Adopting competing inhibition rules exploits the different strengths of the stimulus-response relationships such that the overlearned behaviour of reading the word makes the Repeat rule more salient than the Colour rule. Previous research has suggested that this may cause competitive suppression of the Colour rule, producing more colour errors and affecting participants’ awareness of those errors (Hester et al., 2005). On lure trials, participants were required to respond by pressing a ‘hotkey’ (1, 2, 3, or 4). The appropriate lure response was determined by the participant’s response on the preceding lure trial. At the start of each block participants were instructed to use the hotkey ‘1’ unless they committed an error. To indicate an error, participants were required to change their lure response on the subsequent lure trial to ‘2’, and to continue using ‘2’ on lure trials until they committed another error. Error awareness was thus inferred when the participant’s lure response changed to the next key (2, 3, 4, 1). This modification to the original EAT design dissociates the error awareness response from any post-error adjustments in reaction time and was introduced to allow comparisons between post-lure behaviour following aware and unaware errors. If a participant were to make multiple consecutive errors before progressing their lure response to the next number key, however, it would not be possible to determine which of those errors was aware and which were unaware. To avoid this ambiguity, when participants made two consecutive errors they were then presented with a feedback screen that informed them they had just committed an error and queried what response they should have made. This feedback screen replaced the Go trial following the second consecutive error. If the participant’s response differed from what they should have used on previous lure, it was inferred that they were aware of the error. If they did not progress to the next hotkey in the series, however, the error was classified as unaware. Participants were administered two versions of the task over two separate conditions. The order of administration was randomised. Both tasks presented 5 blocks comprising 265 trials. Lures were presented on 25% of trials. In one task, lures were presented on average every 6 trials (SD = 2; range = 3-12). In the second task, lures were presented on average every 4 trials (SD = 2; range = 2-6).
Results have not yet been conducted. In line with Schroder et al. (2019), response speed adjustments following lure trials will be assessed by calculating the difference in reaction time for each of the post-lure trials and the target trial immediately preceding the lure (a subtraction of post-lure reaction time from the pre-lure reaction time). Pre-lure and lure trials in which the participant made no response will be excluded from analysis. Mixed effects models will be estimated using the mixed function from the afex package (Singmann et al., 2020). Consistent with recommendations by Singmann and Kellen (2019), Type III tests will be used as they are more reasonable under conditions of unbalanced data. Satterthwaite approximation for degrees of freedom will be used to obtain p-values as it provides the best control for Type I errors under the restricted maximum likelihood estimation and is less computationally intensive than the Kenward-Roger approximation (Singmann & Kellen, 2019). Post-hoc tests will be undertaken using the emmeans package (Russell et al., 2020). Marginal means and standard errors will be computed using the emmeans function and pairwise comparisons will be conducted using the pairs function. The emmip function will be used to create interaction style plots for estimated marginal means.
If our findings adhere to our hypotheses, our study will demonstrate that post-error slowing is not necessarily adaptive. Rather, it is influenced by the uncertainty of errors. These findings contest dominant functional accounts of post-error slowing whereby post-error slowing is suggested to reflect the adoption of a conservative response strategy that aims to improve performance subsequent to error detection. Our results lend support to non-functional accounts such as the orienting account of post-error slowing which consider post-error slowing to be a time-consuming orientation to an infrequent or salient event and a reorientation to the task at hand. We aim to submit our research for publication in a journal article.