Before the analysis, one observer in
Experiment 1b was excluded from the analysis because this observer never reported the afterimage in the control condition, and the afterimage duration was quite short (<0.3 second) even when the afterimage was reported in the suppression condition. In the analysis of
Experiment 1a, the results of the two trial blocks for the control condition were combined before averaging.
The mean afterimage duration is shown in
Figure 2 as a function of the orientation difference between the suppressor and adaptor. The results were analyzed in a mixed ANOVA comprised of one between-subjects factor (
Experiment 1a vs.
1b) and one within-subjects factor (five orientation differences and control) using the anovakun function in R (R Foundation for Statistical Computing, Vienna, Austria) (
Iseki, 2023). Mauchly's test was used for the sphericity assumption. If the assumption was not met, the Greenhouse–Geisser correction was applied. The analysis revealed a significant main effect of orientation difference,
F(3.01, 45.2) = 17.96,
p < 0.001, η
p2 = 0.54, but the main effect of experiment and the interaction were not significant,
F(1, 15) = 0.51,
p = 0.49, η
p2 = 0.03;
F(3.01, 45.2) = 0.81,
p = 0.49, η
p2 = 0.05, respectively. A post hoc analysis for multiple comparisons with Bonferroni correction (α = 0.05/15) showed that the afterimage duration in the 0° condition was shorter than that in the 45°, 67.5°, 90°, or control conditions,
t(15) = 5.54,
p < 0.001;
t(15) = 7.25,
p < 0.001;
t(15) = 7.27,
p < 0.001;
t(15) = 3.56,
p = 0.003, respectively. Notably, a significant difference from the control condition was only observed in the 0° condition,
t(15) = 3.56,
p = 0.003.
The afterimage durations were also analyzed by a linear mixed-effects model. In order to test the two main effects and their interaction, we compared models based on Bayes factors calculated using the BayesFactor package in R (
Morey & Rouder, 2022). We calculated Bayesian linear mixed-effects models with a random effect of participants (participant ID [PID]) as a random intercept, as well as the fixed effects of Condition (five orientation differences and control) and Experiment (
Experiment 1a vs.
1b). The fixed effects were the same as those used in the ANOVA. The results are summarized in
Table 1. The model that predicted the data best (i.e., the best model) was the one with the two fixed effects of Condition and Experiment but without their interaction term. This model was favored over the interaction model (BF = 6.70, error = 2.9%). The other models were much less consistent with the data than the best model (BFs > 100). Multiple comparisons were performed using the emmeans package in R (
Lenth, 2023). Though the overall findings were consistent with those obtained through the ANOVA, the linear mixed-effects model revealed a significant reduction of the afterimage duration in the control condition compared to those of the 67° condition,
t(2134) = 4.742,
p < 0.001, and the 90° condition,
t(2134) = 6.188,
p < 0.001.