Mean RT and proportions of errors were submitted to an ANOVA with flanker compatibility (compatible, incompatible) and chroma (6 saturation levels) as within-subject factors. An arc-sine transformation was applied to normalize proportions before analysis

(Winer, 1971). Greenhouse–Geisser corrections were applied in case of violation of the sphericity assumption (Greenhouse & Geisser, 1959). Other specific analyses will be detailed in the text. Anticipations (responses faster than 100 ms, 0.007%) and trials in which participants failed to respond (0.03%) were discarded. There was a reliable flanker effect on RT (M = 44.5 ms), F(1, 11) = 42.4, p < .001, ηp2 = 0.79 find more (see Table 1). The main effect of chroma was also significant, F(5, 55) = 60.7, p < .001, ε = 0.41, ηp2 = 0.85. Lower chroma levels were associated with slower RT (amplitude of the effect, M = 58.9 ms). Importantly, the interaction between chroma and compatibility was not significant, F(5, 55) = 0.6, p = 0.6, ε = 0.5, ηp2 = 0.05. Error rates revealed a similar pattern. We found main effects of compatibility, F(1, 11) = 17.6, p < .005, ηp2 = 0.62, and chroma, F(5, 55) = 52.7, p < .001, ε = 0.5, ηp2 = 0.83. Error rate was higher in the incompatible condition, and increased as chroma decreased. The interaction between chroma and compatibility failed to reach significance,

F(5, 55) = 2.03, p = 0.17, ε = 0.3, ηp2 = 0.16. In order to provide some quantitative support to the plausibility of the null hypothesis of additivity, we further computed

a Bayesian ANOVA on mean RT (Rouder, Morey, Speckman, Montelukast Sodium & Province, 2012) with see more the R package Bayesfactor (Morey & Rouder, 2012). More precisely, we evaluated the ratio of the marginal likelihood of the data given model M0 (implementing additive effects between compatibility and color saturation) and given model M1 (including interactive effects between compatibility and color saturation), a ratio known as Bayes factor. The Bayes factor measures the relative change in prior to posterior odds brought about by the data: equation(1) p(M0/Data)p(M1/Data)︷posteriorodds=p(M0)p(M1)︷priorodds×p(Data/M0)p(Data/M1)︷BayesfactorThe Bayes factor for M0 over M1 was BF0,1 = 15.1 ± 0.63%, revealing that the data are 15 times more likely under the additive model M0 compared to the interactive model M1. According to a standard scale of interpretation ( Jeffreys, 1961), a Bayes factor of about 15 represents strong evidence for M0. Fig. 4 displays the best-fitting Piéron’s curve for each flanker compatibility condition along with observed mean RT. From a qualitative point of view, Piéron’s law seems to describe the data well. This impression is reinforced by very high correlation coefficients between observed and predicted data, both at the group and the individual levels (see Table 2 and Table 3).