Separating predictive responses from reactive responses in isochronous finger tapping using Bayesian hierarchical modeling

Finger tapping to isochronous pacing stimuli is a well established experimental paradigm first described in the late 19th century. Two variables of common interest in rhythmic tapping tasks are timing variability and constant error. However, a difficulty when estimating these variables is that at interstimulus intervals longer than 2 seconds participants regularly overshoot the target interval and instead react to the pacing stimulus. These reactive responses result in a left skewed, non-normal response distribution, which can lead to underestimation of both the constant error and timing variability when using classical moment estimators such as the sample mean and standard deviation. A Bayesian hierarchical model is presented that models the timing responses as coming from a right censored normal distribution. Comparing this model with classical estimators using both simulated and experimental data shows that the Bayesian model measures timing variability and constant error more accurately and with less bias.