Abstract
The visual system is remarkably efficient at extracting summary statistics from the environment, which drive the overall perception of a scene and inform judgment of individual objects within it. Most ensemble research focuses on the perception of one specific characteristic (e.g., average size) from a particular group of stimuli, but natural visual environments usually consist of many groups of objects distributed over space. Here we ask how people perceive the mean position of all stimuli when stimuli are clustered into two distinct groups. Specifically, we evaluate how the cardinality and precision of the perceptual groups influence global mean position perception. We designed a paradigm utilizing the mean summary task used in data visualization. Participants were asked to make intuitive estimations of the global mean position of two spatially separated groups of dots. Both within and across 6 experiments, we varied the number of dots in each group as well as the spatial spread of the groups. We found that people reliably overweight the group of smaller cardinality, including if it is a singleton group (i.e., an outlier). We also found that people overweight the group with smaller spatial spread despite it being perceived as less numerous. We find that an “edge effect” can potentially explain the bias – with people reporting the percent distance between the edges, rather than the centers, of the two groups. This would suggest that people aggregate at the level of groups, rather than individual objects. If people are using an (incorrect) edge strategy, this also leads to a possible way of resolving the bias by altering the underlying reference frame people use, which could improve accuracy in reading data visualizations.