Page 142 - Demo

Chapter 6140For both experiments, we created separate statistical models per subject. We chose to analyze our data at the individual level because of the low number of subjects that participated in this experiment. Given the fact that we had a relatively high number of trials per subject, it was possible to test for the presence of a within-subject effect separately for each subject. Previous work has suggested that this is a suitable approach in case of low subject numbers (Craig & Abramson, 2018; Farrar et al., 2020).To test whether the orang-utans had an attentional bias for large flanges, we fitted three Bayesian mixed models with a Student-t family. The Student-t family is ideal for robust linear models, as the model will be influenced less strongly by outliers. We specified mean-centered RT (in ms) as dependent variable, and Congruence (Congruent: probe behind large flange stimulus; Incongruent: probe behind small flange stimulus) as categorical independent variable. We added Probe location (Left/Right) as categorical independent variable to control for possible side biases in RT. Furthermore, we allowed the intercept to vary by Session, so that the statistical model accounted for variation in RT between sessions. We specified a Gaussian prior with M = 0 and SD = 5 for the Intercept of the model. For the independent variables, we specified regularizing Gaussian priors with M = 0 and SD = 10. For the nu parameter of the Student-t distribution, we specified a Gamma prior with k = 2 and 0 = 0.1. For all variance parameters, we kept the default half Student’s t priors with 3 degrees of freedom. To test whether orang-utans had an attentional bias for symmetrical faces, we followed the exact same procedure. However, the predictor Congruence now refers to the symmetry of the depicted face (Congruent: probe behind symmetrical stimulus; Incongruent: probe behind original stimulus). We used sum-to-zero coding for all of our categorical independent variables.Preference taskFor 5 of the 6 subjects we had a complete dataset of 96 choice trials. Only for Kawan we missed 4 trials, because he left twice at the end of an experimental session. Thus, our final dataset consisted of 572 datapoints. Because we had a larger number of subjects in this experiment, we chose to analyze the data in one statistical model. To examine whether the orang-utans preferred seeing a picture of flanged males over unflanged males, we fitted a Bayesian logistic mixed model (Bernoulli family). We specified the binary choice (1 = flanged, 0 = unflanged) as dependent variable. The within-subject categorical variable Colour Flanged, which represent whether Tom Roth.indd 140 08-01-2024 10:41