Abstract
Our visual system effectively extracts ensemble statistics to make sense of distributed visual information (Alvarez, 2011; Whitney, Haberman, & Sweeny, 2014). Existing explorations have focused on single ensembles, and have used probes varying in degree of similarity to the stimuli, e.g., using mouse clicks to determine centroids (Whitaker, McGraw, Pacey & Barrett, 1996), drawing lines to determine line lengths (Chong & Treisman, 2003) or rotating Gabor patches to determine perceived orientations (Choo, Levinthal & Franconeri, 2014). In the real world, however, ensembles rarely exist in isolation, and the style of response probe may also influence extraction. We emulated how people extract ensemble statistics in a realistic situation (visually communicated data) from among multiple ensembles, using different response probes. Participants viewed either one or two randomly fluctuating lines around different means and were pre-cued to attend to one line. They estimated the average vertical position of the cued line, either moving a horizontal line (probe similar to stimulus) or an asterisk (probe dissimilar to stimulus). Using a mixture model, we observed a 10% systematic underestimation of line position when using a horizontal line as a probe, meaning participants consistently perceived lines to be vertically lower than their actual positions (N=12, p< 0.001). Furthermore, in situations where two lines were presented simultaneously, the target position estimate was biased towards the non-target line (p< 0.001), suggesting an attraction effect during ensemble extraction among multiple ensembles. Interestingly, this underestimation effect diminished (p< 0.01) when participants used the asterisk as probe instead to estimate the position (N=12, p=0.63), while the attraction effect persists. The present study provides surprising implications for potentially probe-dependent results of ensemble extraction tasks, suggesting a dependence of ensemble estimation on attraction effects and an interference between perceived stimuli and response probe.