Abstract
The psychometric function relates an observer’s performance to some physical stimulus quantity in a psychophysical task. Performance is characterized by several parameters of the psychometric function such as the point of subjective equality or the slope. Apart from these primary parameters of interest, two other parameters have been modelled to increase goodness-of-fit: guesses and lapses. Lapses refer to mistakes that are independent of the stimulus level. For example, when an observer mixes up response buttons or lapses in attention. Here, we explore the question whether an explicit modelling of the lapse rate would also improve the estimation of perceptual scales in procedures such as Maximum Likelihood Difference Scaling (MLDS). MLDS is a psychophysical method to derive perceptual scales from forced-choice judgments of suprathreshold stimulus differences. It was tested for its robustness against violations of several model assumptions (Maloney and Yang, 2003), but the influence of lapses on estimated scales has not yet been studied systematically. We run computer simulations to test how a stimulus-independent error rate influences scale estimates in MLDS. We simulated data from different statistical models: we include the classical implementation of MLDS as a generalized linear model (GLM), a Bayesian implementation of the same GLM, as well as two models that explicitly model the lapse rate. We also used the models to reanalyse data from a previous study (Wiebel, Aguilar, and Maertens, 2017), to test the effect of modelling the lapse rate in actual data. In the simulations, lapses lead to an overestimation of the internal noise. In the reanalysis of the experimental data we found that for experienced observers with a low noise estimate the different models did not differ much. For observers with a higher internal noise estimate, models that considered the lapse rate resulted in scales with a smaller internal noise estimate.
Acknowledgement: Landesgraduiertenförderung to Bernhard Lang, DFG MA5127/3-1 and DFG MA5127/4-1 to Marianne Maertens