Page 164 - Demo

Chapter 7162First fixation To investigate whether orang-utans had a first fixation bias toward flanged males when paired with unflanged males, we used binary logistic regression with the location of the first fixation (1=flanged male, 0=unflanged male) as the dependent variable. We modelled the dependent variable as a function of the Intercept and the Location of the flanged male stimulus (left/right on the screen) to control for potential side biases. To investigate the effect of different combinations of male morphs on first fixation, we created a binary variable reflecting the location of the first fixation (1=left, 0=right). We modelled this dependent variable as a function of the interaction between the left stimulus morph (flanged/unflanged) and the right stimulus morph (flanged/unflanged). In all analyses, we allowed Intercepts to vary by Subject, and Session nested in Subject.For binary logistic regressions, we specified regularizing Gaussian priors with M= 0 and SD= 1 for the Intercept and independent variables. We used the default Student’s t priors with 3 degrees of freedom for variance parameters.After running the models, we used the emmeans-package (Lenth, 2023) to provide estimates based on the posterior predictive distribution. Using these values, we calculated multiple quantitative measures to describe the effects. First, we report the median estimate b and the median absolute deviation of the estimate between square brackets. Second, we report an 89% credible interval for the estimate (89% CrI). We chose 89% instead of the conventional 95% to reduce the likelihood that the credible intervals would be interpreted as strict hypothesis tests. Instead, the main goal of credible intervals is to communicate the shape of posterior distributions (McElreath, 2018). Third, we report the probability of direction (pd), that is, the probability of a parameter being strictly positive or negative, which varies between 50% and 100% (Makowski et al., 2019).Total fixation duration To investigate total fixation duration, we used a zero-one inflated beta model, which is suitable for continuous proportions containing zeros and ones (Ospina & Ferrari, 2012). These models consist of multiple components: a beta component to describe the values between 0 and 1, and two binary components to predict the occurrence of zeros and ones. Zero-one-inflated beta regression has previously been employed in eye-tracking studies (e.g., Chiquet et al., 2021). For each trial, we calculated a Looking time bias-score. In Experiment 1 and the replication of this experiment in Experiment 2, we calculated this bias by dividing the fixation time on the flanged male stimulus by the sum of the fixation times on the flanged and unflanged stimuli. Tom Roth.indd 162 08-01-2024 10:41