Page 106 - Demo

Chapter 4104and 4 (definitely disagree), and some items are reversely scored to prevent response biases. All item scores are added up to a total sum score, with higher scores reflecting higher autistic trait levels. One participant did not complete the AQ and was therefore excluded from all analyses investigating effects of autistic traits. Furthermore, we had to estimate three single item scores using the micepackage (van Buuren & Groothuis-Oudshoorn, 2011) for multiple imputation as one participant did not respond to one item and another participant did not respond to two items. Internal consistency of the AQ in our sample was good, %u03b1 = 0.83, 95% CI [0.76, 0.89]. The range of AQ scores was between 2 %u2013 39 (M = 16.38, SD = 7.34), which is highly similar to meta-analytic results on AQ scores in general population samples (M = 16.94, 95% CI [11.6, 20.0])(Ruzich et al., 2015). Only 3 participants (5.26%) had a higher AQ score than 32, which indicates autistic trait levels of clinical significance. The skewness and kurtosis of the AQ score distribution were 1.05 and 4.17 respectively, thus showing a positive skew (see Fig.S1B) in the Supplemental Material). Data AnalysisSpearman%u2019s rank correlation revealed that autistic traits and social anxiety traits, reflected by the scores on the two questionnaires, were not significantly associated with each other, rs = 0.04, p = .784. Our sample showed both variability within each trait dimension that was similar to studies with larger samples (see Questionnaire section) and independence between the trait dimensions, allowing for separate analyses for the two trait dimensions. Emotion recognition accuracy was calculated by determining the expression category with the highest slider score and comparing it to the predefined category of the stimulus for each trial (Zwick & Wolkenstein, 2017). If there was a match between the presented and the perceived expression category, a trial was scored as correct (1) whereas it was scored as incorrect (0) in case of a mismatch. Trials in which two expression categories received the same slider scores were discounted from the analysis. To check the robustness of this approach, we re-ran all analyses on accuracy with a relative accuracy score, which was calculated by subtracting the mean score of all other expression categories from score of the correct expression category (Keating et al., 2021). The results were overall highly similar and are reported in the Supplemental Materials. All analyses were performed in R 4.0.1 (R Core Team, 2020), using the lmerTest package (Kuznetsova et al., 2017) for fitting the (generalized) linear mixed models ([G]LMMs), the multcomp package (Hothorn et al., 2008) for