In our recent study on Facial Emotion Recognition (FER) in Autistic Adults, we used mixed-effects logistic regression models to investigate how (a) differences in stimulus material and (b) repeated measurements in general may affect group differences between autistic and non-autistic participants.
120 autistic and 116 non-autistic participants completed the Berlin Emotion Recognition Test 2.
Figure 1 Illustration of the Berlin Emotion Recognition Test 2.
Differences in Stimulus Material
We used the Automated Emotion Recognition feature of the FaceReader 9 software to estimate basic emotion probabilities for each stimulus face. Our results suggest that non-autistic participants show the highest FER advantage over autistic participants, when the probabilities of response options clearly favor the target response (as indicated by a high differences in target and distractor probabilities). This pattern may imply a “process of elimination” in non-autistic participants, a strategy which has, somewhat ironically, previously been proposed as a compensatory strategy for autistic individuals.
Figure 2 Effects plot illustrating the estimated effect of target-distractor difference on FER Accuracy for ASD and NC.
Repeated Measurements
We investigated the development of task performance over time by using trial number as a predictor of trial correctness. Our results show that both groups improved significantly during the experiment, although non-autistic individuals showed a more pronounced adaptation to the task. This trend suggests that repeated-measure designs may overestimate differences between autistic and non-autistic individuals, when interpreting the results in the context of spontaneous emotion recognition.
Figure 3 Effects plot illustrating the estimated effect of trial number on FER Accuracy for ASD and NC.
Conclusions
Our results show that autistic adults may find it harder to flexibly adjust when repeatedly interpreting facial emotions, and they tend to benefit less from helpful context compared to non-autistic individuals. Overall, the findings suggest that looking beyond simple accuracy scores can give a more complete picture of how people recognize emotions.
Read the full paper here (Open Access Publication).
Post written by Simon Kirsch