![Sample our Behavioral Sciences journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/110801/TOC-Subject-Sample-2021-Behavioral-Sciences.jpg"})
Abstract
Seven studies used the Implicit Association Test to measure preference for gaze direction. For faces with neutral expressions, people clearly preferred eyes looking towards them compared to eyes gazing to the right or left (Experiment 1). This preference remained for faces shown turned to the side (Experiment 2) and upside-down (Experiment 3). Even angry faces were preferred with direct compared to averted gaze (Experiments 4 and 5). Furthermore, preference for eye contact did not correlate to performance on the Reading the Mind in the Eyes Test (RMET) or the Autism Quotient (AQ); note performance on the RMET and the AQ was only weakly correlated although both are claimed to measure social cognition. When the faces were replaced by coloured shapes (Experiment 6) or arrows (Experiment 7) people showed a weaker preference for the category label “looking at you” versus “looking to the side”. Overall, people revealed a robust preference for direct rather than averted gaze which generalized across face pose and expression. Together with a weaker preference for arrows pointing towards them, this is consistent with people having an implicit preference for self-directed attention.
I would like to thank Shay Rosenthal, Natasha Rutter, Lois Parmenter and Siobhan Williams for helping to test participants.
No potential conflict of interest was reported by the author.
I would like to thank Shay Rosenthal, Natasha Rutter, Lois Parmenter and Siobhan Williams for helping to test participants.
No potential conflict of interest was reported by the author.
Notes
1 As a check all of the analyses reported in this paper were repeated using an alternative error penalty suggested by Greenwald et al. (Citation2003), namely adding 600 ms to the mean for correct subblock trials. These analyses produced very similar results to those reported here and they did not alter the pattern of significant differences.
2 Experiment 5 used a double IAT design with all participants doing two separate IATs, one with normal contrast faces and the other with contrast-polarity reversed faces, with the order of IATs counterbalanced. The results for the contrast-reversed IATs are not reported here as they did not help to distinguish between theoretical accounts of the preference for direct gaze. However, D-scores remained significantly greater than zero for both happy, contrast-polarity reversed faces (+0.44), t(23) = 5.518, p < .001, and for angry, contrast-polarity reversed faces (+0.28), t(23) = 3.430, p < .001.
3 To check whether a relationship between the AQ or the RMET and D-scores might have been masked in these correlations by variation in the D-scores across experimental conditions, a linear mixed effects analysis of the relationship between D-scores on the IATs and scores on the AQ was performed using R (R Core Team, Citation2014) and lme4 (Bates, Maechler, Bolker, & Walker, 2014). AQ was entered as a fixed effect. Experiment condition and, nested within it, IAT order (congruent block before or after the incongruent block) were entered as random effects intercepts. Visual inspection of residual plots did not reveal any obvious deviations from homoscedasticity or normality. A likelihood ratio test of the full model with AQ as a fixed effect against the model without AQ did not reach significance, χ2(1, N = 5) = 0.42, p = .52. This analysis was repeated using the RMET. The likelihood ratio test of the full model with the RMET as a fixed effect against the model without the RMET did not reach significance, χ2(1, N = 5) = 3.11, p = .08. These analyses were consistent with the Pearson correlations reported here.
4 If both the AQ and the RMET measure autistic traits and related skills in social cognition then their results should correlate strongly (and negatively) with each other. Consistent with this, Baron-Cohen et al. (Citation2001a) reported a correlation of –.53 between the RMET and the AQ. For the present data set the Pearson correlation was significant and it was in the predicted direction but it was much weaker (r(150) = –.20, p = .01). One potentially important difference between the 132 British adults tested by Baron-Cohen et al. (2Citation001a) and the 152 British adults tested here is that their group included 15 people diagnosed with autism and Asperger's. They did not report the correlation between the AQ and the RMET for their controls alone. Instead, for their 103 student controls they just reported that there were significant correlations for two of the five subtests of the AQ (–.27 for social skills and –.25 for communication). This suggests, first, that the overall correlation of the RMET and the AQ may not have been significant for their student control group and, second, that this overall correlation was probably much weaker than the overall correlation of –.53 which they did report for the whole group which included autistic and Asperger's individuals.Several other studies have reported weak correlations between the RMET and the AQ for non-clinical populations. Ragdale and Foley (2011) found a correlation of just –.08 for 220 British students and adults from the general population. Voracek and Dressler (Citation2006) reported correlations of –.13 for 206 males and –.17 for 217 females for Austrian adults from the general population. Finally, Miu, Pană and Avram (Citation2012) found similar performance on the RMET for a sample of 81 Romanian students selected to have low (<14) and high (>20) AQ scores. These results together with those from the present study (–0.20) suggest that in non-clinical populations there is probably a negative correlation between the RMET and the AQ but that this correlation is likely to be much weaker than the –.53 reported by Baron-Cohen et al. (Citation2001a). Baron-Cohen's correlation was probably inflated due to the inclusion of people diagnosed with autism and Asperger's.If the correlation between the RMET and the AQ in non-clinical populations is only around –.2, consistent with the findings of the present study, then this calls into question the claim that the two tests are measuring a common set of abilities in social cognition. One reason for this low correlation may be that the response alternatives in the RMET use difficult vocabulary such as “contemplative”, “dispirited”, “despondent”, “incredulous” and “pensive”. This test is therefore unlikely to be a pure measure of people's ability to interpret facial expressions because it requires sophisticated language skills. In addition, Ragsdale and Foley (Citation2011) reported that the internal consistency of items in the RMET was poor, even for items representing similar emotions, and that there was no relation between eye gaze direction and accuracy on a particular item. Together this suggests that the RMET may not be an effective test of social cognitive functioning in the non-clinical population. Notwithstanding these concerns about the RMET, an important result from the present study was that even autistic traits assessed using the AQ failed to correlate in the expected direction with people's preference for direct relative to averted eye gaze, contrary to the predictions of the gaze hypothesis.
5 Eight non-naive colleagues were also tested in Experiment 7. They were familiar with both the IAT methodology and the experimental hypothesis tested here. They produced similar D-scores (mean of +0.16 with 6/8 being positive) as the 32 naive participants. An analysis including all 40 participants again revealed a D-score which was significantly greater than zero (+0.27), t(39) = 3.272, p = .002. Comparing the results of Experiments 6 and 7, the D-scores were not significantly greater for arrows than for non-directional coloured shapes when these 8 non-naive participants in Experiment 7 were also included, F(1,70) = 0.580, p = .4. Finally, the D-scores for all 40 participants in Experiment 7 were significantly smaller than those for participants seeing faces in Experiment 1 (F(1,70) = 8.268, p = .005, partial η 2 = 0.11), Experiment 3 (F(1,86) = 6.666, p = .01, partial η 2 = 0.07), Experiment 4 (F(1,86) = 5.717, p = .02, partial η 2 = 0.06) and Experiment 5 (F(1,86) = 6.758, p = .01, partial η 2 = 0.07) but not to Experiment 2 (F(1,62) = 1.304. p = .25, partial η 2 = 0.02). Thus the results reported in the main analysis for the 32 naive participants were unchanged when data from these eight non-naive participants was included.
6 With thanks to Janek Lobmaier for this suggestion.