How beliefs about the warning-target foreperiod inform temporal preparation to determine simple reaction times

Events often unfold over a time course that is amenable to prediction, as when a sprinter learns the constant foreperiod from set to go. This lets us temporally prepare. For many foreperiod distributions, the conditional probability of an event (e.g. a traffic light turning green) increases across time, so preparation should similarly increase, improving reaction times (RTs). Modelling has demonstrated such a quantitative match between mean RT across different foreperiods and the (inverted) subjective hazard function (or some related preparation function). However, modelling has provided few insights into how temporal preparation sculpts the tradeoff between RTs and anticipatory errors. Furthermore, the standard simplifying assumption that the subjective foreperiod distribution is stable offers no account of well-established history effects. To address these and several further issues, here we present a modelling architecture that incorporates an initial Stimulus Independent Process into a Non-Stationary Poisson account of simple Reaction Time (SIP-theN-SPuRT). SIP-theN-SPuRT derives a temporal preparation signal from an estimated foreperiod distribution that is updated from trial to trial. This signal, which may saturate at a physiological limit, feeds into three processes, reflecting urgency, modulation of drive from change detectors, and residual sensorimotor delays. We compared the predictions of nine SIP-theN-SPuRT model variants with new and existing simple-RT datasets. Best fits were obtained by the model in which each trial’s putative preparation signal gradually asymptotes (or saturates), after initially following a subjective hazard function.