Sex differences in chronometric mental rotation with human bodies

Our choice of data analysis was based on the fact that the use of a percentage as a dependent variable in analysis of variance carries a set of problems, especially when it comes to a valid assessment of interactions (see Dixon, 2008; and Jaeger, 2008 for more details). Considering that the interaction between sex and stimulus type is crucial to our predictions, data were analyzed with logistic mixed linear modeling, as recommended by both Dixon and Jaeger to ensure a proper statistical test of hypotheses. Specifically, following Dixon’s suggestion, a mixed linear model where participants were treated as random effects was computed as it is appropriate to the repeated-measures design used here. The linear mixed-effects model was fit with the glmer program in the lme4 R package (Bates, Maechler, Bolker, & Walker, 2014) using maximum likelihood method for parameter estimation. Furthermore, as recommended by Dixon, incremental models were compared to determine the most parsimonious one (as estimated through the deviance obtained for each model). The mean percentage of correct response is reported in its raw form in the results to facilitate data interpretations.

The approach proposed by Baayen and Milin (2010) for the analysis of response time data was followed. Specifically, mixed linear modeling was applied to the inverse response times with minimal trimming, followed by model criticism (with a cutoff point for residuals above 2.5 standard deviations). For response time, means are reported in their raw form as well as in the transformed inverse form (multiplied by −1000 to preserve scaling) to optimize interpretations of the results.