In the analyses presented so far, we investigated the saccade adaptation level based on the amplitude changes between pre- and posttest sessions. In this last section, we report on the changes of saccades recorded during the adaptation session, allowing us to quantify the rate of adaptation.
Figure 4 illustrates the relationship between saccade amplitude and trial number for BDA in subject F. In this subject, we observed a moderate but statistically significant saccade amplitude change for the forward down-looking trials and a larger change in saccade amplitude for the backward up-looking trials. The group analysis (see
Fig. 5 ) indicates that the sign of the linear regression slopes was always consistent with the direction of adaptation with a single exception in subject B for the UDBA condition. Furthermore, the slope differed significantly from 0 in 29 of the 32 relationships. In five subjects, all four relationships differed significantly from 0 (the identification letter of the three remaining subjects is shown in parentheses in
Fig. 5 ). As indicated in the Methods section, we submitted the absolute slope values in these five subjects to an ANOVA with adaptation condition (BDA versus UDA) and adaptation direction (backward versus forward) as factors. No significant difference between the two adaptation conditions (
P = 0.5) or between the adaptation directions (
P = 0.2) was found, but a significant interaction emerged (
P < 0.05). Post hoc tests showed that for the forward adaptation condition, the slope was smaller in the BDA condition (0.82 ± 0.28 deg/100 trials) than in the UDA condition (1.48 ± 0.35 deg/100 trials;
P < 0.05;
Fig. 5D ). In contrast, for the backward adaptation condition, no such significant difference was found (
P = 0.1;
Fig 5C ). Post hoc tests also revealed that in the UDA conditions, forward adaptation was significantly faster than backward adaptation (mean slope: −0.76 ± 0.38 deg/100 trials;
P < 0.05,
Fig. 5B ), whereas, when evoked simultaneously (BDA condition), backward and forward adaptations were equivalent in speed (
P = 0.1,
Fig. 5A ). The vertical eye position included as a between-subject factor in the analysis of variance again had no significant effect (
P = 0.6). In conclusion, forward adaptation was faster than backward adaptation, except when they were induced together in an interleaved fashion. In the latter case, the two opposite types of adaptation interacted with each other to the detriment of the forward adaptation. This occurred irrespective of initial orbital eye position.