Abstract
We recently introduced quick Reading, a novel, Bayesian adaptive method, to measure the reading speed vs print size function (Shepard et al, ARVO 2017; Lu et al, ARVO 2017). The precision of the method was assessed with the 68.2% half-width of credible interval (HWCI) of the estimated function from a single run of the procedure. The purpose of this study is to evaluate the test-retest reliability of the quick Reading method. Ten native English speakers (five well-practiced "experts" and five naïve observers) participated in a one-session experiment which contained five quick Reading blocks of 50 trials each in the periphery. We obtained one estimated reading function in each of the five blocks. Four metrics were computed to evaluate the test-retest reliability of the method. 1) Averaged across blocks, print sizes and observers, the SD was 0.03 and 0.07 log10 units for the expert and naïve observers, respectively. 2) The average HWCI of the estimated reading speed was 0.022 and 0.024 log10 units in the expert and naïve groups, respectively. 3) Because the reading speeds on each single reading function are constrained by its functional form, we applied a sampling procedure (Hou et al, 2016) to remove such constraint in assessing correlations between repeated measures of the reading function. The correlation coefficients were 0.98 for the expert group and 0.96 for the naïve group. 4) The area under the reading curve (AURC) was calculated. A one-way repeated measures ANOVA was conducted to examine changes in the AURC over the five repeated tests. We found no significant change in either group (expert: F (4,16) = 1.164, p=0.363; naïve: F (4,16) = 2.371, p=0.096.) Overall, we showed that the quick Reading method can achieve very high precision with 50 trials in peripheral vision.
Meeting abstract presented at VSS 2018