Abstract
When estimating the number of dots in an array, people misreport the number with some bias and variability that both scale with numerosity. The increasing variance is thought to reflect constant Weber noise on perceptual magnitude representations, while the increasing bias reflects miscalibrated mapping of magnitudes onto explicit numbers. Here we show that response variability in numerical estimation increases more than would be predicted by a constant Weber fraction because most of the variability arises from uncertainty and slow drift in the mapping of magnitudes onto numbers. We presented subjects with 300 trials in which they estimated the number of dots. Each subject had stable idiosyncratic under/over-estimation biases whose magnitude increased with number (subject reliability: r=0.98). Moreover, the coefficient of variation increased with number. These results are well accounted for by variable magnitude-number mapping within each subject. We estimated each subjects’ magnitude-number mapping for 10-trial ‘blocks’ of the experiment. While adjacent blocks have very similar mapping functions (r>0.8), this consistency drops off as a function of the number of intervening blocks. These results indicate that magnitude-number mapping functions drift slowly within individual subjects over the course of the experiment with a shared-variance half-life of over 100 trials (~10 min). In a separate experiment, subjects quickly learned a magnitude-number mapping through trial feedback (thus disrupting their own individual biases and acquiring new biases we teach them). This learning is forgotten over time with roughly the same time-course as the drift within individuals before any learning. Together these results indicate that instead of constant perceptual magnitude noise, most of the variability in numerical estimation tasks arises from uncertain magnitude-number mapping that drifts slowly over time.
Meeting abstract presented at VSS 2013