Page 168 - Emotions through the eyes of our closest living relatives- Exploring attentional and behavioral mechanisms

Chapter 7
Analyses
D-score calculations and analyses were performed in R, using the IATScores package (Richetin et al., 2015). As per the suggestion of Richetin et al. (2015), we did not distinguish between the first 20 trials (“practice” trials) in the critical block and the 40 following trials. The minimum performance criteria were an error rate below 40% for the critical blocks, and reaction times (RTs) higher than 400 ms and lower than 10,000 ms (Nosek et al., 2014). RTs below or above these criteria were discarded. Based on these criteria, we discarded 4 trials with an RT > 10,000, and 23 trials with an RT < 400 (divided relatively equally across 13 subjects) within the PIAT sample. For the WIAT sample, we removed 6 trials (divided over 4 individuals) based on the 10,000 ms cutoff, and 124 trials based on the 400 ms cutoff (divided over 16 individuals, of which one individual had 20 trials meeting this criterion, and another individual 80 trials). No participants were removed based on the 40% error rate cutoff in the PIAT, and one participant in the WIAT. This method is slightly different from Experiment 1, where only RTs above 10,000 ms were discarded and remaining RTs were 10% winsorized. While this is the most robust approach for treating IAT data (Richetin et al., 2015), this was not an option due to an error in data collection. During data collection, trials with where a wrong categorization was made were flagged by the software, but due to an error on our part, RTs for these trials were not saved and thus erroneous trials could not be included in the analysis. Instead, we discarded RTs lower than 400 ms (as per the original scoring method by Greenwald et al., 2003).
We used one sample t-tests to test whether D-scores significantly differed from zero. Furthermore, to assess internal validity, we also fit linear models using Congruency (incongruent first vs. congruent first) and Location (whether faces of men of Moroccan or Dutch descent were presented left or right) as fixed effects (sum-to- zero coded) in two separate analyses (PIAT and WIAT), with the intercepts reflecting the average D-score. Note that instead of Location, we used Task Version as fixed effect in Experiment 1. This was done because the versions in Experiment 1 differed on multiple fronts (i.e., stimuli and location on the screen), whereas in Experiment 2, versions of the tasks only differed in the location of the stimuli on the screen. Next, we correlated D-scores on the PIAT with the WIAT in order to assess test-retest reliability. In addition to comparing the IATs to each other, they were also compared with the explicit bias measure (SRS) in order to assess discriminant validity. The data of the SRS were first converted to a continuous 0-1 scale as described by Henry & Sears (2002), and subsequently used for a correlation calculation. Finally, we also investigated whether providing the SRS in-between the two tasks (and thus before one of the two
166