Yangyi Shi¹ (), Rhea T. Eskew, Jr.¹; ¹Northeastern University

Maximum Likelihood Difference Scaling (MLDS) is an efficient perceptual scaling method used in many psychophysical studies. In the ‘Quad’ version of MLDS, two stimulus pairs varying in the same physical property (for example, luminance contrast) are shown, and observers pick either the more similar pair or the more different pair according to instruction. A maximum likelihood procedure then assigns perceptual scale values to each physical level. In our previous study, we used MLDS to estimate the asymmetry of perceived luminance contrast between achromatic increments (A+) and decrements (A-) square patches. We showed that the MLDS results are consistent with pedestal discrimination results. We proposed a model, in which the perceptual scale estimated by MLDS is used to predict the forced-choice discrimination thresholds. Logically, the instruction (pick the more similar pair or the more different pair) should not influence the observers’ choices, because MLDS is a two-alternative forced choice procedure, switching the instruction is just picking the other pair. However, in our study, we found that this was not always true. While some observers did not show a criterion difference, there were some observers whose estimated perceptual scales varied significantly when the instruction given to them changed. We verified this by using their MLDS results under two criterions to predict their forced-choice discrimination results, based on the model we proposed in the previous study. Our results showed that those observers’ “more similar” results can predict the discrimination thresholds better than their “more different” results do. Also, the difference between criterions is larger in A- than in A+. We suggest that the criterion differences likely result from attending to different features in the stimuli, even though they are simple square patches.

Acknowledgements: NSF BCS-2239356