Abstract
The staircase method has been widely used in measuring perceptual learning. Recently, Zhao et al (2017) developed the quick CD method and applied it to measure the trial-by-trial time course of learning. Here, we evaluate the performance of the staircase and quick CD methods. An observer with an exponential learning curve (time constant = 50 trials) in a 2AFC task was simulated. A 3-up/1 down staircase with six step sizes (1%, 5%, 10%, 20%, 30%, and 60% increase or decrease of contrast) and the quick CD were used to estimate the contrast threshold of the simulated observer, each starting from five different contrasts (+50%, +25%, 0, -25%, and -50% from the true threshold), with 400 trials in each of 1000 simulated runs. Thresholds were estimated every 80 trials. We found that: (1) The average absolute bias (AAB) of the trial-by-trial threshold estimates from quick CD was 0.016, with an average standard deviation (SD) of 0.039, all in log units. (2) Threshold estimates from the entire learning curve using quick CD were nearly unbiased (only 0.007), with an average SD of 0.026. (3) Staircases with 1% and 5% step sizes sometimes failed to generate more than seven reversals and could not be used to estimate the threshold in 80 trials. (4) The AAB and SD of the estimated thresholds were 0.053/0.047/0.048/0.080 and 0.046/0.053/0.061/0.087 for staircases with step sizes of 10%, 20%, 30%, and 60%, respectively. (6) The bias after the first block were 0.049 for quick CD, and 0.182/0.153/0.133/0.164 for staircases with the five different step sizes. All these results varied very little with the starting contrast. We conclude that the quick CD method provides more accurate and precise measures of perceptual learning than the staircase method. The optimal step size of the staircase is about 10%.
Meeting abstract presented at VSS 2018