Page 63 - Demo

61How attractiveness preferences influence attention3Model comparisonsFor both the dot-probe and eye-tracking analyses, we additionally created acomplete cases dataset in which we included only those cases for which wehad pre-date attractiveness ratings, post-date attractiveness ratings, anddate outcomes. Using these two datasets, we again ran the analyses de,scribed above (with pre-date attractiveness, post-date attractiveness, or dateoutcome as predictor, respectively). Hereafter, we used leave-one-out crossvalidation (PSIS-LOO-CV, Vehtari et al., 2017) to calculate the expectedlog predictive density (elpdLOO), which quantifies predictive accuracy foreach model. Then, we calculated the difference in elpdLOO (∆elpdLOO)between all three models. If ∆elpdLOO of two models is at least two SEs,this suggests that the models substantially differ in predictive performance(Johnson, Ott, & Dogucu, 2022). Therefore, we report both the ∆elpdLOOand the SE of the difference. In total, the immediate attention dataset con,sisted of 3198 trials of 55 participants, while the voluntary attention datasetconsisted of 1009 trials of 35 participants.ResultsInter-rater agreement on attractivenessWhen examining the inter-rater agreement on pre-date attractiveness rat,ings, we found an ICC(1, A) of 0.42 (95% CI [0.32, 0.52]). This resultsuggests that participants differed in their attractiveness preferences inde,pendent of gender. Furthermore, we explored the inter-rater agreement formen and women separately. For women, we found an ICC(1, A) of 0.25(95% CI [0.14, 0.41]), while for men, we found an ICC(1, A) of 0.50 (95%CI [0.39, 0.64]). These results suggest that women had substantially loweragreement than men.Correlations between attractiveness ratings and date outcomeWe found that pre-date attractiveness rating, post-date attractiveness rat,ing, and date outcome all showed a strong correlation. First, a point-biserialcorrelation indicated that pre-date attractiveness rating and date outcomewere correlated (r = 0.44, 89% CrI [0.36, 0.50], pd+ = 1.00). Second, wefound that post-date attractiveness showed an even stronger correlation withdate outcome (r = 0.67, 89% CrI [0.62, 0.71], pd+ = 1.00). Third, a Spear,man correlation showed that pre-date and post-date attractiveness were cor,related (r = 0.57, 89% CrI [0.51, 0.62], pd+ = 1.00).Model comparisonsFor both the dot-probe and eye-tracking analyses, we additionally created acomplete cases dataset in which we included only those cases for which wehad pre-date attractiveness ratings, post-date attractiveness ratings, anddate outcomes. Using these two datasets, we again ran the analyses de,scribed above (with pre-date attractiveness, post-date attractiveness, or dateoutcome as predictor, respectively). Hereafter, we used leave-one-out crossvalidation (PSIS-LOO-CV, Vehtari et al., 2017) to calculate the expectedlog predictive density (elpdLOO), which quantifies predictive accuracy foreach model. Then, we calculated the difference in elpdLOO (∆elpdLOO)between all three models. If ∆elpdLOO of two models is at least two SEs,this suggests that the models substantially differ in predictive performance(Johnson, Ott, & Dogucu, 2022). Therefore, we report both the ∆elpdLOOand the SE of the difference. In total, the immediate attention dataset con,sisted of 3198 trials of 55 participants, while the voluntary attention datasetconsisted of 1009 trials of 35 participants.ResultsInter-rater agreement on attractivenessWhen examining the inter-rater agreement on pre-date attractiveness rat,ings, we found an ICC(1, A) of 0.42 (95% CI [0.32, 0.52]). This resultsuggests that participants differed in their attractiveness preferences inde,pendent of gender. Furthermore, we explored the inter-rater agreement formen and women separately. For women, we found an ICC(1, A) of 0.25(95% CI [0.14, 0.41]), while for men, we found an ICC(1, A) of 0.50 (95%CI [0.39, 0.64]). These results suggest that women had substantially loweragreement than men.Correlations between attractiveness ratings and date outcomeWe found that pre-date attractiveness rating, post-date attractiveness rat,ing, and date outcome all showed a strong correlation. First, a point-biserialcorrelation indicated that pre-date attractiveness rating and date outcomewere correlated (r = 0.44, 89% CrI [0.36, 0.50], pd+ = 1.00). Second, wefound that post-date attractiveness showed an even stronger correlation withdate outcome (r = 0.67, 89% CrI [0.62, 0.71], pd+ = 1.00). Third, a Spear,man correlation showed that pre-date and post-date attractiveness were cor,related (r = 0.57, 89% CrI [0.51, 0.62], pd+ = 1.00).Immediate attention (dot-probe)Pre-date attractiveness ratingsWe first examined the association between Pre-date attractiveness rating andReaction time using a Bayesian mixed model with Gaussian distribution(Descriptives: Table S1-3; Model Table: Table S4). We found a robust overall effect of Pre-date attractiveness rating of distractor picture on Reactiontime (b = 1.46 [0.53], 89% CrI [0.60, 2.29], pd+ = 1.00), with participantsresponding slower by 1.46ms to the probe when there is an increase of 1 inattractiveness ratings of the distractor picture. There was no robust interaction with Gender (bwomen−men = -0.83 [1.06], 89% CrI [-2.49, 0.88], pd−= .79). However, after visually inspecting the results, we wanted to explorewhether the positive effect o f P re-date a ttractiveness r ating o f distractorpicture on Reaction Time was robust within each level of Gender. We foundthat the effect was indeed robust for men ( bmen = 1.87 [0.65], 89% CrI [0.80,2.88], pd+ = 1.00), but not for women (bwomen = 1.04 [0.83], 89% CrI [-0.33,2.34], pd+ = .89). Thus, men responded slower to the probe by 1.87 mswhen the attractiveness rating of the distractor picture was increased by 1,while no robust effect was found for women (see Figure 1 top panel).Furthermore, we found a robust overall effect of Pre-date attractivenessrating of probe picture on Reaction time (b = -1.11 [0.55], 89% CrI [-1.97, -0.24], pd− = .98), whereby participants responded faster by 1.11 ms when theattractiveness rating for the probe picture was increased by 1. In this case,however, the effect w as m odulated b y G ender ( bwomen−men = 2 .83 [1.06],89% CrI [1.12, 4.51], pd+ = 1.00). Therefore, we further explored the slopeper Gender. We found a robust negative effect o f P re-date attractivenessrating of probe picture for men (bmen = -2.51 [0.67], 89% CrI [-3.59, -1.48],pd− = 1.00), indicating that men responded faster by 2.51 ms when theattractiveness rating of the probe picture was increased by 1. For women,on the other hand, we found no robust effect ( bwomen = 0 .30 [ 0.85], 89%CrI [-1.03, 1.59], pd+ = .64). Thus, men seemed to respond faster to theprobe when they considered the image that was replaced by the probe highlyattractive, while no robust effect was found for women (see Figure 1 lowerpanel).We performed the same analysis with the Post-date attractiveness ratings as predictor. This analysis yielded the same results (Table S5).Iliana Samara 17x24.indd 61 08-04-2024 16:35