Page 84 - Demo

82Chapter 4tify the evidence in favor of the null or alternative hypothesis. For eachexperiment, we fitted two models: one Gaussian and one ordinal. In thefirst model, we replicated the analyses by Maner et al. (2005). We furtherconducted a Bayesian ordinal multilevel model (with cumulative family) toexamine whether the ordinal model captured the dependent variable better(see Liddell & Kruschke, 2018). In all analyses, all our predictors were sumcoded (the Condition predictor was added to the model as two sum-codeddummy variables). In the Gaussian models, we included a Gaussian priorwith a mean of 4.5 and SD of 2 for the intercept and a conservative Gaussianprior with a mean of 0 and SD of 1 for all coefficients. In all ordinal models,we set a conservative Gaussian prior with a mean of 0 and SD of 0.5 for allcoefficients.We report multiple estimates (e.g., see Martin, Ringen, Duda, & Jaeggi,2020), namely, the median Odds Ratio (OR) with the Median Absolute Deviation (MAD), alongside the 95% Highest-Density Credible Intervals (HDI),which summarize a posterior distribution with the highest probability density (Kruschke, 2018). Effects with a 95% HDI spanning over 0, were notconsidered robust (Bürkner & Vuorre, 2019). Furthermore, we report theprobability of direction (pd), the proportion of the probability that supportsa putative effect (Makowski et al., 2019), and an approximation of Cohen’sd based on Borenstein et al. (2009) for the ordinal models.All multilevel models included Participant as a random intercept. Contrasts between posterior distributions were computed using the package emmeans (Lenth et al., 2021) and interpreted as robust if the estimated HighestPosterior Density (HPD) Interval did not contain 0. Since the intercepts inordinal models simply reflect thresholds between categories (see Bürkner &Vuorre, 2019), they are not interpreted.We used the guidelines outlined in the WAMBS checklist to examinemodel convergence (Depaoli & van de Schoot, 2017). For all models,Gelman-Rubin diagnostic values, and trace and density histograms of allposterior distributions were examined. Analyses were conducted with R (RCore Team, 2021) using the package brms (Bürkner, 2017, 2018; Bürkner& Vuorre, 2019).ResultsExperiment 1Preliminary analysesTo examine whether the mood induction was successful in inducing sexual and romantic arousal, we conducted an ordinal model on romantic andsexual arousal ratings with Condition (control vs. White female protagonist vs. Black female protagonist) as a fixed effect. The results showedIliana Samara 17x24.indd 82 08-04-2024 16:35