Error rates were computed from all trials. In a signal detection framework, we computed criterion and sensitivity (d′). Search slopes were computed for each individual and each combination of target emotion/target presence by linearly regressing all RTs on set size. We used ANOVA models in SPSS to analyse the control group, and to locate differences between patients and the control group. Because unequal variance in different

cells within the control population in an ANOVA design can increase type I error rates (Crawford and Garthwaite, 2007 and Crawford et al., 2009), we confirmed group differences and 2 × 2 interactions using a single-case Bayesian approach as implemented in see more Crawford’s software. Non-significant findings do not require confirmation. Note that for interactions involving a higher order or higher number of levels, no appropriate single-case Bayesian methods are available. In our control sample, set size, target emotion, and target presence influenced RT as shown previously (see Fig. 2A and Table 1), with a linear impact of set size. This result was confirmed by fitting a linear regression

Alpelisib order model to predict RT from set size, separately for each combination of target presence and target emotion. An ANOVA on search slope estimates (Table 2) underlines that search slope is influenced by target face – angry target faces have a shallower search slope – and by target presence. There were no effects in an ANOVA on intercepts of the regression model, as expected. Next, we compared the two patients with the control sample (Fig. 2A, Table 1). Patients

responded faster to happy than to angry targets, while healthy individuals showed the opposite pattern, in particular for larger set size (interaction Group × Set size × Emotion). This result was confirmed by comparing patients’ search slopes with the control sample which revealed a significant Group × Emotion interaction. On a single individual basis, Bayesian dissociation analysis revealed a significant Group × Emotion interaction for AM (p = .017) but not for BG. Further, patients showed slower RT and steeper search slopes overall. This was confirmed only as a trend in a single-case Bayes approach (one-tailed tests; RTs: AM, p < .05; BG, p < .10; search slopes: AM, p < .05; Rebamipide BG, p < .10). Patients also differed from the control group in a stronger non-linear effect of set size (quadratic interaction group × set size: F(1, 16) = 18.3; p < .005, η2 = .533) – RTs for the medium set size were disproportionately large. Reversal of the anger superiority effect in the patients’ RTs and search slopes might be due to a different strategy in a speed-accuracy trade-off. In this case, AM and possibly BG should show increased accuracy for angry as opposed to happy targets. Hence, we analysed errors using a signal detection analysis on sensitivity (d′) and response criterion for each combination of set size and target emotion (Table 2, Fig. 2B and C).