Presentation Information
[P1-20]Beyond probability: Temporal prediction error shapes performance across development
*LOUIS-CLÉMENT DA COSTA1, Sylvie Droit-Volet2, Katherine Johnson3, Jennifer T Coull1 (1. CRPN, CNRS and AMU, UMR 7077, Marseille (France), 2. CNRS and Université Clermont Auvergne, UMR 6024, Clermont-Ferrand (France), 3. Melbourne School of Psychological Sciences, Melbourne (Australia))
Keywords:
temporal attention,implicit timing,foreperiod,prediction error,expectation,children
We have previously shown that although children are significantly less precise than adults in explicit timing tasks, their performance is equivalent to adults in implicit timing tasks. However, the dynamic way in which the temporal prior is constructed in implicit timing tasks may nevertheless be subject to developmental change. Here, we adopted a more fine-grained analytical approach by tracking changes in temporal probabilities across trials and using a Bayesian learning algorithm to capture the emergence of the temporal prior throughout the session.
Speeded reaction times (RTs) were recorded in 47 young children (5-7 years), 58 older children (7-11 years), and 48 adults during a variable foreperiod (FP) paradigm (240-960ms FPs). The 600ms FP was much more probable (~36% of trials) than the six other shorter or longer FPs, which were themselves equiprobable (~9% each). We also included catch trials (~9%) to mitigate the effects of the hazard function on performance. For each participant, we calculated dynamic changes in FP probability (Pb) and temporal prediction error (pE) across trials. The pE was defined as the absolute difference between the FP predicted by a Bayesian learner (i.e. the moment at which the prior was maximal) and the actual FP of that trial.
We analysed the influence of FP duration, Pb and pE on RTs to the target. Performance varied as a function of FP duration and all three groups responded fastest to targets appearing after the most probable FP. Strikingly, RTs showed a U-shaped profile, getting gradually slower as FP duration got increasingly shorter or longer than 600ms, even though these FPs were all equally probable. Indeed, linear mixed-model analyses showed a significant main effect of pE on RTs, indicating that performance is guided by the temporal distance between the prior and the actual FP, rather than FP probability per se. Nevertheless, the influence of pE on performance emerged gradually during childhood, with younger children having a less temporally precise prior than older children.
These findings confirm that all participants demonstrated temporal statistical learning, and that temporal prediction error plays a key role in explaining implicit timing performance across development.
Speeded reaction times (RTs) were recorded in 47 young children (5-7 years), 58 older children (7-11 years), and 48 adults during a variable foreperiod (FP) paradigm (240-960ms FPs). The 600ms FP was much more probable (~36% of trials) than the six other shorter or longer FPs, which were themselves equiprobable (~9% each). We also included catch trials (~9%) to mitigate the effects of the hazard function on performance. For each participant, we calculated dynamic changes in FP probability (Pb) and temporal prediction error (pE) across trials. The pE was defined as the absolute difference between the FP predicted by a Bayesian learner (i.e. the moment at which the prior was maximal) and the actual FP of that trial.
We analysed the influence of FP duration, Pb and pE on RTs to the target. Performance varied as a function of FP duration and all three groups responded fastest to targets appearing after the most probable FP. Strikingly, RTs showed a U-shaped profile, getting gradually slower as FP duration got increasingly shorter or longer than 600ms, even though these FPs were all equally probable. Indeed, linear mixed-model analyses showed a significant main effect of pE on RTs, indicating that performance is guided by the temporal distance between the prior and the actual FP, rather than FP probability per se. Nevertheless, the influence of pE on performance emerged gradually during childhood, with younger children having a less temporally precise prior than older children.
These findings confirm that all participants demonstrated temporal statistical learning, and that temporal prediction error plays a key role in explaining implicit timing performance across development.