) thresholds for identifying contrast-defined letters and contrast thresholds for identifying luminance-defined letters are plotted as a function of background noise contrast, and compared between the pre- and post-tests in . We defined threshold as the ΔC
, or contrast value that yielded 50%-correct performance (after correction for guessing) on the psychometric function, constructed based on the data from each block of trials. Each panel presents data of one observer. In those few cases where thresholds were not measurable (observers AC and MK), a value of 11
was used to represent the threshold for that block of trials. Although this chosen value was arbitrary, if anything, this underestimated the improvement that we observed.
Fig. 3 Contrast threshold for identifying luminance-defined letters (squares) or differential-contrast threshold (ΔC) for identifying contrast-defined letters (circles) are plotted as a function of background noise contrast, and compared before (unfilled (more ...)
A comparison of the threshold vs. background noise (TvN) functions in reveals that thresholds are less elevated (compared with the no-noise condition) in the presence of high background noise contrast (0.5) when identifying contrast-defined letters than for luminance-defined letters. Across all observers, pre- and post-tests, threshold elevations (thresholds obtained at 0.5 background noise contrast normalized to those obtained with- out background noise) averaged 1.23 for contrast-defined letters and 1.81 for luminance-defined letters. If we assume that the shape of the TvN functions for contrast-defined letters also follows the typical shape of a noise-masking function, as is the case for luminance-defined letters (Chung et al., 2005
), then the smaller threshold elevation observed at a background noise contrast of 0.5 is simply an indication that the TvN functions for contrast-defined letters are shifted upward and to the right (toward higher background noise), when compared with the TvN functions for luminance-defined letters.
plots the ratios of post- and pre-test values, as a function of background noise, for identifying luminance- and contrast-defined letters. Ratios <1 represent improvements following training. Across observers, the ratios averaged 0.923 and 0.742 for identifying luminance- and contrast-defined letters, respectively, and were statistically different from one another (repeated measures ANOVA: F(df = 1,7) = 67.83, p < 0.0001). Post hoc analyses show that the group-averaged ratio for identifying luminance-defined letters (0.923) was not statistically different from a value of 1 (no improvement following training). In contrast, the group-averaged ratio for identifying contrast-defined letters (0.742) was statistically different from a value of 1 (p < 0.0001). These findings suggest that training to identify contrast-defined letters improved the performance for identifying contrast-defined letters, but the improvement did not transfer to the task of identifying luminance-defined letters.
Fig. 4 Threshold ratios (post-test/pre-test) are compared for the three background noise contrast levels, and between luminance-defined (left) and contrast-defined (right) letters. Ratios smaller than 1 represent improvements following training. Individual observers’ (more ...)
Did the improvement following training transfer to the two untrained background noise contrast (viz., 0 and 0.25)? shows that the ratios were similar across the three background noise contrast for contrast-defined letters. Averaged across observers, the ratios averaged 0.73, 0.83, and 0.68, for background noise contrast of 0, 0.25, and 0.5, respectively. These ratios were not statistically different from each other (repeated measures ANOVA: F(df = 2,14) = 1.80, p = 0.20), implying that the improvement at the trained background noise contrast (0.5) transferred well to the other two untrained background noise contrast (0 and 0.25).
The progress of training for each observer, as a function of training block, is plotted in . Each short solid line represents the threshold averaged across the 10 blocks of each day. Like for the pre- and post-tests, we used a value of 1 to represent the threshold when it was not measurable (most obvious with observer MK). As stated earlier, for most observers, threshold ΔC decreased as training progressed (averaged ratios of post/pre-test thresholds = 0.742, significantly different from a value of 1, p < 0.0001).
Fig. 5 Differential contrast (ΔC) for identifying contrast-defined letters is plotted as a function of training block, for each of the eight observers (gray symbols). The averaged threshold for each session (10 blocks) is represented by the black line. (more ...)
To examine the specificity of the learning effect, we compared the post-test/pre-test threshold ratios for the untrained letter size (twice the original size), untrained superior field and the untrained eye in . For comparison, the ratios for the trained condition (replotted from ) are also included. Across observers, the ratios averaged (±95% CI) 0.88 ± 0.13, 1.02 ± 0.24, and 0.72 ± 0.11, for the untrained letter size, untrained superior field, and the untrained eye, respectively. For both the untrained letter size and untrained superior field, because the 95% CI included the value of 1 (no improvement), we concluded that there was no significant difference between thresholds obtained before and after training, or, in other words, the improvement following training did not transfer to the untrained letter size or the untrained superior visual field. In comparison, a value of 1 fell outside the 95% CI of the ratio for the untrained eye, implying that the learning transferred to the untrained eye. The magnitude of improvement (1 – ratio) was very comparable between the untrained eye (28%) and the trained eye (32%), for the same condition, suggesting an almost complete transfer of the learning to the untrained eye.
Fig. 6 Threshold ratios (post-test/pre-test) are compared for the three conditions: untrained letter size (2× original), untrained retinal location (10° in the superior visual field) and the untrained eye. Data for the trained condition are also (more ...)