Page 100 - Demo

98Chapter 5to indicate whether they would like to go on another date with their partner (see also Overbeek, Nelemans, Karremans, & Engels, 2013; Asendorpfet al., 2011; Todd et al., 2007), instead of indicating sexual interest (as in,Perilloux et al., 2012; A. J. Lee et al., 2020) was opted for given that it ismore ecologically valid procedure. Participants were given 1 minute to fill inthe questionnaires. Male participants rotated from one partner to the next.After all opposite-sex couples had had a date, participants were thankedand debriefed.Statistical AnalysesTo examine accuracy in detecting attraction, we calculated accuracy scoresby comparing participants’ predictions regarding whether their partnerwould be interested in another date with them to the responses of theirpartners (0 = incorrect; 1 = correct). These accuracy scores were analyzedusing Bayesian logistic multilevel modeling (MLM). The use of BayesianMLM allowed us to account for the nested nature of the data, as well asexamine the support for either the null or alternative hypothesis.In total, we conducted 3 separate accuracy models. All models includedaccuracy scores as dependent variable and the fixed effect of Sex. In the firstmodel, we examined whether sex and own interest influence accuracy scoresby including the fixed effect of Own Interest, and its interaction with Sex.In the second model, we examined whether sex and sexual desire influenceaccuracy scores by including the fixed effect of Sexual Desire and its interaction with Sex. In the third model, we examined whether sex and self-ratedattractiveness influence accuracy scores by including the fixed effect of SelfRated Attractiveness and its interaction with Sex. All our binary predictorswere sum coded (-1 vs. 1); whereas all other predictors were scaled to obtaina mean of 0 and a standard deviation (SD) of 1.An important benefit of Bayesian analyses is that they allowed us toplace a prior on our assumptions, thus incorporating prior knowledge in theparameter estimation (Jeffreys, 1961; M. D. Lee & Wagenmakers, 2013).Given that uniform priors are considered improper in logistic models sincethey can bias the posterior distribution of the estimate (McInturff, Johnson,Cowling, & Gardner, 2004; Seaman, Seaman, & Stamey, 2012), we optedfor a Student’s t prior distribution with 7 degrees of freedom centered at 0with an SD of 1 (except for the intercept which had an SD of 10; Ghosh, Li,& Mitra, 2018; Gelman, Jakulin, Pittau, & Su, 2008). The use of Student’st priors with 7 degrees of freedom has been recommended as opposed toother distributions, as it produces reliable estimates and reduces likelihoodof computational estimation problems (i.e., “slow mixing Gibbs samplers”)even under conditions of separation (Ghosh et al., 2018, p. 362). Furthermore, an exponential prior with an SD of 1 was set for all error terms.To facilitate the interpretation of the model coefficients, all estimatesIliana Samara 17x24.indd 98 09-04-2024 10:37