162 Chapter 6 Data analyses Preprocessing Preprocessing of the raw data from the EA task was similar to aan het Rot and Hogenelst (2014), with raw continuous ratings from perceivers and targets being preprocessed into an EA measure per video in SAS 9.3 for Windows (SAS, Cary, NC). For data reduction purposes, ratings from perceivers and targets were averaged across five-second periods. The last five seconds of all ratings were discarded, because it included the return of the dial to the “neutral” position before the end of each video. Subsequently, first-order autocorrelations were removed from the continuous ratings using the Yule-Walker method. For each video we correlated perceiver ratings of the target’s feelings and target ratings of their own feelings, resulting in a correlation coefficient r that defined the perceiver’s raw EA score per video. Raw EA scores underwent a Fisher z transformation prior to further analyses. See Supplement S6.3 for more details on the preprocessing of raw eye tracking data into measures of eye gaze per perceiver per video. The primary eye gaze measure is the percentage of dwell time within the defined areas of interest (AOIs; i.e., eyes, mouth, and face as a whole) per video, as part of the total video duration, in which dwell time is defined as the total amount of time spent looking within an AOI and includes all types of eye movements. The percentage of dwell time within the face and mouth AOI were described to identify to what extent perceivers gazed towards the face and mouth of the targets in addition to their eye region. Dwell time is interpreted as the level of interest in an AOI, with greater dwell times indicating greater levels of interest. Statistical Analyses Means and standard deviations of the EA task and the self-report ratings per video and valence category (i.e., positive or negative) were calculated. In addition, the average percentage of dwell time for each AOI (i.e., eyes, mouth, and face) per valence category and video were assessed. The effects of our hypothesized predictors on EA were tested in R-3.6.1 (R Core Team, 2013), using generalized linear mixed regression models with a multi-level, within-subject design. We used lme4 for multilevel analyses with maximum likelihood (Bates et al., 2012) and ggplot2 for figures (Wickham et al., 2016). The dependent variable EA has been repeatedly measured and EA observations per video (level 1) were nested within perceivers (level 2). Predictor variables that act upon the perceiver-level were the percentage of dwell time within the eye region and trait EC and PT. Predictor variables that act upon the target-level (level 1) were target expressivity and valence of the videos.
RkJQdWJsaXNoZXIy MjY0ODMw