Purchase this article with an account.
Guillermo Aguilar, Felix Wichmann, Marianne Maertens; Comparing sensitivity estimates from MLDS and forced-choice methods in a slant-from-texture experiment. Journal of Vision 2015;15(12):731. doi: 10.1167/15.12.731.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Maximum likelihood difference scaling (MLDS) is a method for the estimation of perceptual scales based on an equal-variance, Gaussian, signal detection model (Maloney & Yang, 2003). It has recently been shown that perceptual scales derived with MLDS allowed the prediction of near-threshold discrimination performance for the Watercolor effect (Devinck & Knoblauch, 2012). The use of MLDS-based scales to predict sensitivity is psychophysically attractive, because the MLDS scale estimation promises to require a comparatively small amount of data relative to classical forced-choice procedures, such as the method of constant stimulus. However, the relationship between estimates from MLDS and forced-choice procedures is not yet well characterized with respect to their bias and variability. It also remains to be tested whether their close correspondence applies to stimuli other than the Watercolor effect. Here, we studied these issues by comparing the MLDS and forced-choice methods in a slant-from-texture experiment. We used a 'polka dot' texture pattern which was slanted 0 to 80 deg away from fronto-parallel and viewed through a circular aperture (Rosas et al., 2004). We first obtained perceptual scales describing the relationship between physical and perceived slant using MLDS with the method of triads. Based on individual scales we measured slant discrimination thresholds at different performance levels (d'=0.5, 1 and 2) for four different standard slants using standard forced-choice procedures. We obtained a high correspondence between slant thresholds obtained from the two methods. The variability of the estimates, however, depended heavily on the amount of data for MLDS, somewhat questioning its efficiency. Furthermore, the correspondence between the two methods was reduced in the lower region of the MLDS scale. We conclude that MLDS scales can be used to estimate sensitivity, however, we would advise caution with respect to the generalizability of the correspondence between MLDS and forced-choice based sensitivity measures across experimental tasks.
Meeting abstract presented at VSS 2015
This PDF is available to Subscribers Only