Abstract
The staircase procedure is used to estimate the learning curve through measurements of thresholds in blocks of trials in most perceptual learning studies. Recently, researchers developed the quick Change Detection (qCD) method to measure the trial-by-trial time course of dark adaptation (Zhao, et al, 2017, 2019) and perceptual learning (Zhang et al, 2019). The qCD method provided more precise and accurate estimates of the learning curve than the standard staircase method. The question is whether rescoring the trial-by-trial staircase data improves the precision of the estimated learning curves. Rescoring simulated staircase data using the qCD algorithm indeed resulted in much-improved accuracy and precision of the estimated learning curves. Here we rescored the experimental data from Zhang et al (2019). In that study, each of the five observers was trained in a 4 alternative forced-choice global motion direction identification task, with the coherence level in the odd trials controlled by the qCD method and that in the even trials controlled by a 3-down/1-up staircase procedure. We rescored the trial-by-trial staircase data using the trial-by-trial and post hoc segment-by-segment qCD algorithms. We also estimated the 68.2% credible interval (HWCI) of the estimated block thresholds from the staircase method with a bootstrap method. We found that, averaged across the entire learning curve and observers, the 68.2% HWCI of the rescored coherence threshold was 0.025±0.002 (mean±sd) and 0.014±0.002 in trial-by-trial and segment-by-segment analysis, respectively. In comparison, the average 68.2% HWCI of the estimated threshold from the staircase method was 0.055±0.001, and the average 68.2% HWCI of the estimated threshold from the qCD method was 0.022±0.001 and 0.012±0.001 in trial-by-trial and segment-by-segment analysis, respectively. We conclude that rescoring the trial-by-trial staircase data in perceptual learning with the qCD algorithm could greatly improve the precision of the estimated learning curve.