Page 143 - Demo

Absence of cognitive bias to flanges in orang-utans1416the flanged stimuli were associated with the red or the green dot, was added as an independent variable, together with the between-subject variable Order, which represented whether the individual first received the sessions in which the red dot was associated with the flanged stimuli or in which the green dot was associated with the flanged stimuli. To explore the effect of dot location on the screen on probability of selection, we extended the model by adding a continuous predictor that was zero-centered and reflected the location of the dot representing flanged stimuli relative to the vertical middle of the screen (range -0.35 – 0.35, with negative values representing the higher portion of the screen). With regard to the random effects, we allowed the intercept to vary by Subject and allowed the intercept of Session to vary within Subject. Furthermore, we allowed the slope for Colour Flanged to vary by Subject, to take into account potential treatment effects between subjects. We specified a Gaussian prior with M = 0 and SD = 0.5 for the Intercept and independent variables of the model. Note that these priors are specified on the logit scale. For all variance parameters, we kept the default half Student-t priors with 3 degrees of freedom. To explore temporal clustering and dispersion in the choices of the orangutans, we developed an R script based on (Gelman et al., 2004, p. 163) that is essentially a Beta-Binomial model that can be used to assess independence of binary observations. We applied it to each of the sessions independently. The script first counts the number of switches between selected categories within the session (variable T). Second, we specified a Beta(10, 10) prior on 0, the probability of selecting a flanged male stimulus, emphasizing a relatively strong expectation of 50/50 selection of flanged and unflanged stimuli. Third, we obtained a posterior for 0 by updating the Beta(10, 10) prior based on the choices from the session. Fourth, we simulated 10,000 binary series of the same length as the session, based on sampling from the posterior distribution of 0. Note that the binary series consisted of independent samples. Fifth, based on these simulations, we counted the number of switches T in each independent series, and obtained a distribution of T under the assumption of independence. This allowed us to compare the observed T within the sessions with the expected T under the assumption of independence. Consecutively, we checked whether the observed T fell outside of the 95% Highest Density Interval of the expected T, and we calculated the proportion of expected T-samples that was either similar or higher, or similar or lower than the observed T. With regard to the interpretation, an observed T that is low compared to the distribution of expected T reflects fewer Tom Roth.indd 141 08-01-2024 10:41