Purchase this article with an account.
Edward A. Vessel, Nava Rubin; Direct comparison of preferences for dramatically different stimulus types reveals higher observer agreement for images with semantic content. Journal of Vision 2006;6(6):470. doi: https://doi.org/10.1167/6.6.470.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Previous studies of preference for real-world scenes found a high degree of agreement for ratings across observers (Vessel & Biederman, 2002 J. of Vision 2(7) 492a). Vessel & Rubin (2005, ECVP) tested the degree to which this agreement is attributable to familiarity with the themes and/or semantic content of scenes by measuring observer agreement for preferences of a set of abstract, novel color images. Observers performed forced-choice, “one-back” paired comparisons between images and preference values were estimated for the full stimulus set from these paired comparisons (a sorting algorithm guided presentation order to optimize the estimation procedure). We found very low agreement across observers, but robust within observer reproducibility. Using the identical task with half abstract and half real-world images (grouped into six categories each) we tested whether this difference remains when the two stimulus types are directly compared within a session. We found higher agreement across subjects for real-world scenes (r = 0.24) than for abstract images (r = 0.12). Subjects slightly favored the scenes over the abstract images, and produced a slightly larger range (nonsignificant) of preferences for real-world scenes. Preferences for the image categories were uncorrelated across subjects for abstract categories (r = -0.1), but well correlated for real-world categories (r = 0.43). Intermixed paired comparisons allow for direct comparison of preferences for dramatically different stimulus types. We replicated our previous finding of higher agreement for real-world scenes than abstract images, and found that this difference cannot be attributed to differences in stimulus range or within-subject variability.
This PDF is available to Subscribers Only