The responses in each condition of the slant experiment were fitted using a cumulative Gaussian function Φ (µ, σ2), in which µ was fixed to zero and the estimated σ was used as the measure of the threshold for each subject and each condition.
Figure 4 shows the mean thresholds from different conditions and groups. A three-way mixed design analysis of variance (ANOVA) was conducted to test statistical significance of the effects on threshold. The main effect of slant was significant (
F(1,46) = 292.67,
p < 0.001, partial
η2 = 0.864) and so was the main effect of viewing condition (
F(1,46) = 5.74,
p = 0.021, partial
η2 = 0.111). These effects indicated that thresholds were generally lower at high slants and with binocular viewing. The main effect of group was not significant (
F(1,46) = .242,
p = 0.625, partial
η2 = 0.005). We also found two significant interactions (slant × viewing:
F(1,46) = 10.96,
p = 0.002, partial
η2 = 0.192; slant × viewing × group:
F(1,46) = 7.05,
p = 0.011, partial
η2 = 0.133). The other interactions were not significant (
F < 2.50,
p > 0.120, partial
η2 < 0.051). The significant slant × viewing interaction indicated that the thresholds differed between the viewing conditions at the low slant (30°), but not at the high slant (60°). The three-way interaction further indicated that the threshold differences between viewing conditions at 30° occurred on the stereo-normal group, but not on the stereoblind group. The lack of the viewing effect at high slants (i.e., 60°) here was predictable because, at higher slants, texture cues become more reliable and has higher weight in slant perception than disparity cues (
Knill, 1998a;
Knill & Saunders, 2003;
Hillis et al., 2004).
To compare the benefit from stereo for the normal and stereoblind groups, we performed 2 (viewing) × 2 (group) ANOVAs for the low slant and high slant conditions separately. In the low slant conditions, there was no significant main effect of group (F(1,46) = 0.510, p = 0.479, partial η2 = 0.011). The main effect of viewing was significant (F(1,46) = 4.402, p = 0.041, partial η2 = 0.157), indicating that the stereoblind group had higher thresholds than the stereo-normal group in general. More importantly, the interaction was significant as well (F(1,46) = 8.431, p = 0.006, partial η2 = 0.155), which indicates that stereo provided more benefit to the normal population than the stereoblind population, as expected. For high slants, there was no significant main effect or interaction observed (viewing: F(1,46) = 0.502, p = 0.482, η2 = 0.011; group: F(1,46) < .001, p = 0.993, partial η2 < 0.001; viewing × group: F(1,46) = 0.004, p = 0.947, partial η2 < 0.001), indicating no evidence that stereo information selectively improved discrimination for the normal group at high slants.
We also analyzed performance in only the monocular conditions to test whether the stereoblind group performed better than the normal group. A 2 × 2 ANOVA was conducted to analyze slant estimates in monocular viewing conditions. The main effect of slant was significant (F(1,46) = 266.03, p < 0.001, η2 = 0.853), as expected. But there was no main effect of group (F(1,46) = 0.767, p = 0.386, η2 = 0.016) or interaction between group and slant (F(1,46) = 0.372, p = 0.545, η2 = 0.008). The results indicate that the overall performance of normal and stereoblind groups was equivalent.
In short, there were two main findings from the discrimination task. First, stereoblind people performed worse at slant discrimination in binocular conditions compared to stereo-normal people, which confirmed the validity of our stereoblindness screening. Second, the thresholds in the monocular viewing conditions did not significantly differ between the two groups, indicating that the stereoblind people did not develop higher sensitivity to texture cues because of the loss of binocular cues. At high slants, we did not find significant effects of viewing condition or group. This is possibly because texture cues were more reliable than stereo cues so that the influence of stereo information became limited and hard to be observed. This has also been observed in previous studies (
Knill & Saunders, 2003;
Hillis et al., 2004).