A new adaptive procedure for estimating perceptual thresholds: the effects of observer bias and its correction

Abstract

Adaptive threshold estimation procedures sample close to a subject's perceptual threshold by dynamically adapting the stimulation based on the subject's performance. Yet, perceptual thresholds not only depend on the observers' sensory capabilities but also on any bias in terms of their expectations and response preferences, thus distorting the precision of the threshold estimates. Using the framework of signal detection theory (SDT), independent estimates of both, an observer's sensitivity and internal processing bias can be delineated from threshold estimates. While this approach is commonly available for estimation procedures engaging the method of constant stimuli (MCS), correction procedures for adaptive methods (AM) are only scarcely applied. In this article, we introduce a new AM that takes individual biases into account, and that allows for a bias-corrected assessment of subjects' sensitivity. This novel AM is validated with simulations and compared to a typical MCS-procedure, for which the implementation of bias correction has been previously demonstrated. Comparing AM and MCS demonstrates the viability of the presented AM. Besides its feasibility, the results of the simulation reveal both, advantages, and limitations of the proposed AM. The procedure has considerable practical implications, in particular for the design of shaping procedures in sensory training experiments, in which task difficulty has to be constantly adapted to an observer's performance, to improve training efficiency.

Visual Motion Detection Thresholds Can Be Reliably Measured During Walking and Standing

“…left movement more noticeable than right). Including the bias can lead to improper fits in rare cases [34], and it has been found that participants could voluntarily shift their central bias, causing changes in threshold values, but the slope parameter remained relatively unchanged [32]. Since we had no a priori reason to expect a bias, we constrained the mean of the psychometric curve to zero for all participants.…”

Visual motion detection thresholds can be reliably measured during walking and standing

“…A bias can occur if a participant either habitually chose a particular side when they were unsure of the direction or if the visual system itself had a skewed mapping (i.e., left movement more noticeable than right). Including the bias can lead to improper fits in rare cases (Fioravanti et al, 2021), and it has been found that participants could voluntarily shift their central bias, causing changes in threshold values, but the slope parameter remained relatively unchanged (Morgan et al, 2011). Since we had no a priori reason to expect a bias, we constrained the mean of the psychometric curve to zero for all participants.…”