Abstract
Our recent study (Lee et al., 2021) demonstrated that in a set of color-size conjunctions, one feature value can be predicted by the given other feature value based on the color-size correlation of the set. Given that color space is circular, but size space is linear, we wondered whether the predictability between conjoint features depended on feature types (i.e., circular vs. linear). To test this hypothesis, we presented a set of 12 objects whose two features (Experiment 1a: color-size; Experiment 1b: color-orientation) were either perfectly correlated or uncorrelated. In a trial, a display set was presented briefly and a test probe with one of the two features followed for participants to choose a value of one feature matching with the test value of the other feature. We replicated the predictability between conjoint features based on the inter-feature correlation of a set regardless of feature types, but prediction sensitivities depended on feature types. Participants’ prediction sensitivity was the highest when they predicted size values with colors. This sensitivity was higher than that of predicting color values with sizes. The sensitivity of predicting color values with orientations and that of predicting orientation values with colors was the lowest. Next, we tested whether the predictability between circular features (i.e., color and orientation) was enhanced when additionally correlated with a task-irrelevant linear feature, i.e., size (Experiment 2). In a trial, a display set consisted of twelve objects whose colors and orientations were always perfectly correlated and they were either perfectly correlated or uncorrelated with size. Participants should match colors and orientations without considering sizes. Participants’ prediction sensitivity between colors and orientations did not differ depending on the presence of correlation with sizes. Overall, these results suggest that the predictability between conjoint features differs depending on feature types and is not influenced by task-irrelevant features.