Page 70 - Demo

Chapter 368Stekelenburg, et al., 2013). In order to avoid distortions by the initial light reflex following stimulus onset (M. M. Bradley et al., 2008), the analysis on pupil size changes was restricted to the last two seconds of stimulus presentation. Figure 2. Visualization of the three analysis approaches. In Analysis 1, the subjective interpretation (emotion recognition and intensity judgments) of the different emotional expressions was examined. In Analysis 2, the effect of perceiving different emotional expressions belonging to the same modality on the shape of five different physiological signals was explored. In Analysis 3, trial-wise summary measures of expressionspecific signal changes in all physiological channels were taken to fit a model on self-reported emotion labels and the generalizability of these observed patterns was evaluated using different data sets (test sample, inaccurate trials and subtle emotional cues; see Online Resource 4 for a more detailed description). Analysis 1 (Behavioural analysis)In the behavioural analysis, we investigated whether the specific content of the emotional expressions (categories: happy, angry, sad or fearful versus neutral) as well as the modality with which it was displayed (face versus body) had an influence on recognition performance as well as on the perceived intensity in the ELT. Thus, in the first step, we looked at differences in the accuracy of recognizing specific emotional expressions from different expression modalities. Investigating the data on a trial level, we fitted a binomial generalized linear mixed-effects model on accuracy (0 or 1) with emotion category, expression modality, and an interaction between the two of them as predictors. In order to account for individual differences in overall emotion recognition abilities, we included a random intercept for the subject variable.33 An addition of random slopes for emotion category and/or expression modality resulted in convergence issues. To keep the behavioral models consistent, we refrained from defining random slopes in any of the models.