Decision-level processes in rapid numerosity estimation

How do people rapidly judge numerosity—the number of objects in a scene? Theories of human numerical judgements are constrained by the behavioural patterns emerging when people estimate numerosity, such as exact enumeration of small sets, underestimation of large sets, and scalar variability. Current computational models capture how properties of stimulus representations can lead to characteristic behavioural patterns under rapid viewing conditions. Here, we examined which decision-level processes people engage during such tasks, and whether these processes contribute to behavioural patterns. We combined qualitative, correlational, and experimental studies of participants estimating brief (100 ms) and backward-masked dot patterns. In interviews, participants reported relying on visual memory to mentally compare a stimulus to previously estimated stimuli, count dots, and approximately add or multiply groups of dots. Quantitative multiple-choice surveys confirmed that the reported processes were systematically associated with numerosity, response time, and estimates. Preregistered experiments further showed that three decision-level processes—comparing, partitioning, and counting—contribute to several behavioural patterns reported in prior research. These results suggest that many estimation patterns may reflect the flexible use of decision-level processes that depend both on individual factors and task demands. Formal models that integrate these decision-level processes with representation-level mechanisms may substantially enhance our understanding of rapid numerical judgements. • People retain a brief visual memory trace.when rapidly estimating numerosity. • This memory allows people to mentally compare, partition, and count dots. • These decision-level processes systematically relate to key behavioural patterns. • Comparing to labelled references shifts both estimates and response variability. • Partitioning produces prime number bias, and counting extends the exact range.