Data analysis

All behavioral experiments were designed with customized Med-PC IV programs (Med Associates, St. Albans, VT, USA). As previously described,⁵⁴ accuracies measured for visual and auditory training sessions were calculated as percent correct utilizing a 65% correct response rate for two consecutive days. Multisensory gain was calculated as (mean number of correct multisensory trials−mean number of correct best unisensory trials)/(mean number of correct best unisensory trials) × 100. Accuracies were calculated as correct trials/(correct+incorrect trials). Prism 6 (Graphpad Software, La Jolla, CA, USA) was used for all statistical analyses. No significant differences in variance between groups were found. Repeated measures two-way analysis of variances (ANOVAs), Sidak's multiple comparisons tests and s.e.m. were used for all experiments unless otherwise specified.