2.4 Parameter estimation by sequential testing method

As opposed to the psychometric method, which uses preselected ΔΘ values, the PEST method employs a standard 2D1U (2-down-1-up) adaptive algorithm that chooses ΔΘ values for each trial based on the subject’s responses to the previous trials. The ΔΘ changes on each trial in an increment that is equal to the current step size. The current step size changes based on the subject’s previous responses [15] and the session ends when the iterative step size falls to 0.5° or lower. An initial ΔΘ of +3° or −3°, an initial step size of 4°, and a stopping step size criterion of 0.5° were used for each PEST run [5]. The 2D1U PEST method was chosen to target a 75% threshold [5]. Multiple PEST thresholds were estimated for each subject to validate the accuracy of PEST for measuring directional acuity thresholds as well as the stability of the threshold estimate over an experimental period (n = 3 subjects with 2 PEST runs; n = 8 subjects with 3 PEST runs). The recovered thresholds quantified in Figure 3 were estimated from the first PEST run.

A) An example PEST run from one subject with an estimated threshold of 8.5°. B) The right and left PEST thresholds measured in the first PEST run were not significantly different (n = 11 subjects, p = 0.37, paired t-test). C) The right and left PEST thresholds were averaged together to measure an overall PEST threshold of 11.7° ± 3.8° (n = 11 subjects). D) The average number of trials needed to reach a left and right PEST threshold for the first PEST iteration was 33.8 ± 8.4 trials (n = 11 subjects). E) The first PEST threshold is highly correlated with the psychometric threshold for each subject (r = 0.93). Error bars represent three standard deviations of the psychometric threshold estimate. Inset depicts the PEST threshold estimate relative to the psychometric threshold as quantified by the number of degrees the PEST estimate is away from the psychometric threshold. F) Multiple PEST runs were measured for a single subject (n = 8 subjects). The PEST threshold estimate and the number of trials required to reach a threshold were not significantly different across PEST iterations (n.s. = p>0.05, one-way repeated measures ANOVA).