Page 139 - Emotions through the eyes of our closest living relatives- Exploring attentional and behavioral mechanisms
P. 139

                                Evidence for yawn contagion in orangutans
maximum attention to the screen, we presented primers before video sequences and colored videos in-between yawn and control clips, and we only started testing when orangutans had a direct line of sight towards the screen. Additionally, before each trial, we observed the orangutans for five minutes, and only started a trial if there were no yawns before the presentation so as to rule out that yawns within a trial were potentially caused by a previous yawn outside of the trial. Furthermore, yawns were scored in response to either the yawn or control video only if a subject looked at least once to the screen during presentation. If bystanders in the same enclosure attended to the screen, their behaviors were also scored. Data collection ended after 10 min, concluding one test session. Finally, EvB coded 15% of the videotapes for inter-rater reliability purposes. Results showed a good agreement on occurrences of yawning (ICC=0.764, p<0.001) and self-scratching (ICC=0.894, p<0.001). In subsequent analyses, only yawns on which the raters agreed were used.
Statistical analyses
The dependent variable was whether a subject yawned in response to a video or not.
Because it is difficult to disentangle between whether multiple yawns occurring in
succession are caused by another individual, or whether they are simply the result
of an urge to yawn multiple times perhaps because of self-contagion (i.e., where 6 your own yawns cause you to yawn again), we did not compare rates of yawning
to establish CY (Kapitány & Nielsen, 2017). Rather, we looked at the likelihood of yawning within the yawn and control condition to establish the presence or absence of CY in orangutans. Nevertheless, when contagion indeed occurred, yawning rate could inform about the strength of contagion (Kapitány & Nielsen, 2017). As such, we analyzed our data using hurdle models in R (lme4 package, Bates et al., 2015). Hurdle models follow a two-step method that first deals with zero-inflated count data and subsequently with positive counts once the initial hurdle is crossed (Cameron & Trivedi, 2013), which make them applicable to our dataset.
In the first hurdle model we focused on whether CY is present or absent in orangutans by comparing the likelihood of yawning in the yawn and control condition using a binomial GLMM, in which we added condition as a fixed effect and subject nested in trial as a random effect. In the second step of the model, we analyzed the rates of yawning using a negative binomial GLMM only in those cases where at least one yawn occurred. Again, we entered condition as a fixed effect and subject nested in trial as a random effect. In the second hurdle model, we tested for potential effects of both condition and trigger (i.e., familiar/unfamiliar/avatar) and their interaction on
 137

























































































   137   138   139   140   141