Page 118 - Demo

116Chapter 6Figure 2. Progression of a typical trial in the experimental task for Experiment 1 (a) in which both members of a couple were presented simultaneously,and responses were logged using pen and paper and Experiment 2 (b), whereonly one member of the couple was displayed. The question in figure b isan example for only the male couple member presented, the questions wereformed depending on the couple member’s gender.an intercept-only model to examine whether people can accurately detectattraction. In the second model, we examined whether accuracy differs asa function of age or the video segments properties by including the fixedeffect of Group or Video Condition for Experiments 1-2 and Experiment 3,respectively. In the third model, we examined whether the accuracy of theuninvolved third-party observers was enhanced when the daters themselveswere attracted to their partner by including the fixed effect of Attraction toPartner. Additionally, in Experiment 1, we examined whether synchronousbehaviour between the daters influences accuracy by including the fixed effect of Shuffled condition. All our models included a random effect of Participant. The minor adjustments due to the factors present in each experimentare further explained in the corresponding statistical analyses section.There has been a long-standing debate about optimal priors for logisticmodels (e.g., Christensen, 2011; Gelman, Jakulin, Su, & Pittau, 2007). Uniform priors can exert undue influence on the posterior distribution of theunderlying parameter (McInturff et al., 2004) and therefore, weakly informative priors are better suited (Seaman et al., 2012). We have opted to usea normal distribution for all input values despite previous literature pointingtowards Cauchy priors (Gelman et al., 2007; Ghosh et al., 2018). All priors were centered to 0 and had a standard deviation of 1 for all coefficientsexcept the constant (SD = 0.8). We further included an exponential prior(SD = 1) to all error terms. Finally, binary inputs were sum-coded (-1 vs.1).The interpretation of Bayesian logistic MLM estimates might not be inIliana Samara 17x24.indd 116 08-04-2024 16:36