Abstract
People seem to be able to rapidly estimate number from brief visual displays. It remains a mystery how they accomplish this behavior, i.e., what visual information or algorithms they rely on. While some have suggested that number is a primary visual feature (akin to amount of redness), others have claimed that this ability relies on visual density or other non-numerical visual features. A recent computational model has been developed for assessing the extent to which observers rely on numerical and non-numerical visual features in behavioral tasks. Here we use Panamath, a method that has been used by over 60 publications to date, and apply a computational model to assess N=6,361 participants' reliance on visual signals of number, size and spacing when making numerical comparisons. We found that responses varied with numerical ratio, t(6360) = 229.77, p < .001. Judgments also varied with relative spacing, t(6360) = 119.32, p < .001, and to a lesser extent with relative size, t(6360) = -4.69, p < .001. Participants were more likely to judge a dot set as more numerous if the dots were smaller and more spread out. We also examined which of these features were most predictive of participants' numerical judgments. Participants relied more on relative spacing than relative size, t(6360) = 115.75, p < .001. Most importantly, participants relied more on numerical ratio than on either relative size, t(6360) = 227.02, p < .001, or relative spacing, t(6360) = 223.88, p < .001. Our results highlight the value of using computational models to determine the extent to which participants rely on visual numerical and non-numerical signals when making number judgements – and provide evidence that humans do have a visual sense of number that goes beyond visual size and spacing.
Meeting abstract presented at VSS 2018