Emotional task

To assess how tRNS affects accuracy in facial emotion recognition, response accuracy was analyzed by conducting a Generalized Linear Mixed Model (GLMM) with logit link function (i.e., logistic regression) using the glmer function from the lme4 library (Bates et al. 2015). Trials with response times (RT) lower than 150 ms were considered anticipations and excluded. For this analysis, trials with Neutral faces were also excluded. The GLMM included emotion (two-level factor: sadness = -1, anger = 1), emotion intensity (continuous variable: low intensity = 0, medium intensity = 1, high intensity = 2), and stimulation (two-level factor: Sham = 0, tRNS = 1) and all their interaction terms as fixed effects. The random part included random intercepts for participants and facial emotional images. The random part did not include random slopes for the fixed effects, since their inclusion led to singularity.

In addition to the accuracy analysis, we also analysed RT from accurate trials using a linear mixed-effects model (LMM). The specification of the model was the same of the GLMM. To control for the impact of positive skewness in the distribution of RTs (in ms), all the analyses were performed on the inverse-transformed RTs (iRT), computed as -1000/RT (Brysbaert and Stevens 2018).

Finally, to verify whether there was a tendency in categorizing neutral faces as angry or sad and if this tendency was modulated by tRNS, the probability of “angry” responses to neutral faces was modelled using a GLMM (logit link function) including Stimulation (two-level factor: Sham = 0, tRNS = 1) as fixed effect and participants and facial emotional images as random intercepts. Trials with anticipations were excluded.