Page 163 - Demo

Eye-tracking reveals bias to flanges in orang-utans1617a trial. Furthermore, we presented all possible combinations of males to the participants (four males, so six combinations) within one session. Inter-rater reliability All sessions of both experiments were scored by TR to check whether subjects looked away from the center of the screen during trial onset so that these trials could be repeated at the end. To test whether such trials could be reliably identified, TR and EvB coded eight sessions (48 trials) from Experiment 1 for looking away during trial onset. Of these 48 trials, they agreed to include 34 and excluded 10, but disagreed on inclusion of four trials, resulting in a Cohen’s kappa of 0.78 (91.7% agreement), reflecting a good level of reliability between raters. Statistical analysisAll analyses were performed using R Statistics Version 4.2.2 (R Core Team, 2023). For our analyses, we employed a Bayesian approach, which has become increasingly popular in recent years owing to its numerous advantages over frequentist analyses (Kruschke et al., 2012; Makowski et al., 2019). Whereas frequentist approaches, such as p-value null hypothesis testing, provide insight into the plausibility of the data under a particular hypothesis, Bayesian methods inform us about the credibility of our parameter values based on the observed data (Kruschke et al., 2012; McElreath, 2018). This difference is reflected in the contrasting interpretations of frequentist and Bayesian confidence intervals. While the former provides a range of values that contain the estimate over the long term, the latter identifies the most plausible parameter values given the data. Moreover, Bayesian methods allow the integration of prior expectations into the model, are less susceptible to Type I errors, and are more robust in small and noisy samples (Makowski et al., 2019). Taken together, these factors render Bayesian methods a valuable tool for data analysis.All models were created in the Stan computational framework and accessed using the brms-package (Bürkner, 2017, 2018). All models were run with four chains and 6000 iterations, of which 1000 were warmup iterations. We checked model convergence by inspecting the trace plots, histograms of the posteriors, Gelman-Rubin diagnostics, and autocorrelation between iterations (Depaoli & van de Schoot, 2017). No divergence or excessive autocorrelation was found.Tom Roth.indd 161 08-01-2024 10:41