Purchase this article with an account.
Isamu Motoyoshi, Saya Kashiwakura; MaxFind: an efficient method for psychological scaling of large stimulus sets. Journal of Vision 2018;18(10):214. doi: 10.1167/18.10.214.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Increasingly, studies on the perception of objects, materials, and faces employ a large number of natural images to ask observers for various perceptual and emotional attributes such as shape, softness, and attractiveness. To measure the subjective intensity of such attributes, many studies have used magnitude estimation - or rating -, but rating is often an unstable measure. Thurstone's classic paired comparison is based on comparative judgments but it requires a large number of trials as the stimulus set becomes large. Here, we combine maximum-likelihood algorithms in a novel psychophysical procedure and propose an experimental protocol of comparative judgments that can order and scale the subjective intensity of stimuli using only a small number of trials. In this protocol, (1) observer views a list of M stimuli taken from N stimuli, and repeatedly choose the stimulus that elicits maximum subjective response along a given dimension (e.g., the most attractive) until the last stimulus remains. (2) On each trial, stimuli in the N x N comparison matrix are sorted according to a psychological scale constructed from PSE and JND as estimated by logistic regression analyses. (3) The next M stimuli are sampled such that responses will be collected only for pairs for which the expected response ratio is close to 0.5. Numerical simulations demonstrate that our method, for M larger than 8, can estimate psychological scale with only ~1.3 x N responses (e.g., ~130 responses for 100 stimuli). Psychophysical experiments confirmed that the method can quickly estimate the contrast response function to gratings and the perceived glossiness of naturalistic objects. This method would be useful for characterizing human judgements along many psychological dimensions, especially those with no physical correlate such as emotional and social attributes
Meeting abstract presented at VSS 2018
This PDF is available to Subscribers Only