The cost of forming statistical summary representations across multiple spatial scales

The visual system can extract a statistical summary representation (SSR) of a group of similar objects (Ariely 2001). We can form SSRs across multiple feature dimensions for a group of objects (Chong & Treisman, 2005), across multiple feature dimensions across multiple groups of objects (Emmanouil & Treisman, 2008), and across multiple sensory modalities (Albrecht et al., 2013). The current experiments are the first to study the properties of SSRs across multiple spatial scales. The stimuli consist of three large rectangles constructed by spatially arranging multiple small rectangles. All the small rectangles within a large rectangle have the same orientation. Participants estimate the average orientation of either the large rectangles (global level or low spatial frequency) or the small rectangles (local level or high spatial frequency). In Experiment 1, we use a cueing paradigm to demonstrate the cost in forming statistical summary representations across multiple spatial scales. Many models (Parkes et al. 2001; Baek & Chong, 2019) assume that properties of individual items must be computed before creating an SSR of the group. If so, it is important to show that there is an SSR formation cost in Experiment 1 above and beyond the cost of representing the individual items. In Experiment 2, we address this question with stimuli consisting of a single large rectangle constructed from several small rectangles. Participants estimated the orientation of either the large rectangle or the small rectangles. There was a cost involved in the formation of a single item hierarchical representation. Comparing the results from the two experiments revealed that there is an additional cost in the formation of hierarchical SSRs after the hierarchical single item representations are formed.