Abstract
Extracting summary statistics is an efficient strategy coping with a complex visual scene. In the world, multiple objects usually vary along multiple features. However, there are not many studies investigating how people represent a multi-feature ensemble. Here, we conducted membership identification tasks on objects with two features. A display set consisted of 8 circles whose color and size varied in a consistent manner (correlation coefficient r = 1: e.g., as the size of the circle increased, its color varied from blue to green) or in a random manner (r = 0) to examine whether the inter-feature correlation influences representing features conjointly. Following the display set presentation, participants judged whether a probe was the member or not. In Experiment 1, we tested with probes whose conjoined features varied. When the probe was new, two features were outside the displayed range (outlier), were the mean of displayed set (mean), or had the opposite relationship to that of the displayed set (inversely-correlated). Participants accurately recognized members, rejected outliers, and falsely recognized means as often as members regardless of the display set correlation. For inversely-correlated probes, false positives were at the same level as members when r = 0 but were reduced when r = 1. In Experiment 2, we tested with single-feature probes (color or size). Participants falsely recognized means but rejected outliers less compared to conjunction probes. In addition, the feature type and the correlation condition had little effect. In Experiment 3, the new probe had conjoined features, one of which was fixed to the mean, and the other was located either within (mean-interior) or outside the displayed range (mean-exterior). Participants correctly rejected mean-exteriors regardless of the correlation, whereas they hardly rejected mean-interiors. Overall, our results suggest that for multi-feature ensembles, people represent statistical properties of each feature and use them conjointly.