The Influences of Face Inversion and Facial Expression on Sensitivity to Eye Contact in High-Functioning Adults with Autism Spectrum Disorders

Abstract

We examined the influences of face inversion and facial expression on sensitivity to eye contact in high-functioning adults with and without an autism spectrum disorder (ASD). Participants judged the direction of gaze of angry, fearful, and neutral faces. In the typical group only, the range of directions of gaze leading to the perception of eye contact (the cone of gaze) was narrower for upright than inverted faces. In both groups, the cone of gaze was wider for angry faces than for fearful or neutral faces. These results suggest that in high-functioning adults with ASD, the perception of eye contact is not tuned to be finer for upright than inverted faces, but that information is nevertheless integrated across expression and gaze direction.

Appendices

Appendix 1

We calculated the residual deviance for each logistic fit (see Dalgaard 2008; McCullagh and Nelder 1989). A larger residual deviance reflects greater discrepancy between the model and the data. The residual deviance and residual degrees of freedom for a fit correspond approximately to a χ² distribution (see Dalgaard 2008). A χ² probability of less than .05 is typically taken as an indicator of a poor fit. For the current design, the deviance residual corresponding with this probability is 19.65. The largest residual deviance observed in the current experiment was 5.44, which corresponds to a χ² probability of .91. Hence, there was no significant discrepancy between the data and the model for any of the fits in the current study.

Appendix 2

The ideal observer’s task was as follows: on every trial, we presented a face from the main experiment with a particular direction of gaze (2, 3, 4, 5, 7, or 9 pixels, left or right), emotion (neutral, angry, fearful), identity, and root mean square (RMS) contrast. RMS contrast is defined as:

$$ rms = \left[ {\frac{1}{n - 1}\sum\limits_{i = 1}^{n} {(x_{i} - \bar{x})^{2} } } \right]^{\frac{1}{2}} $$

where n is the number of pixels in the image, x _i is the intensity of pixel i (normalized so that 0 ≤ x _i  ≤ 1), and $\bar{x}$ is the mean normalized pixel intensity (Peli 1991). Using the optimal decision rule (Tjan et al. 1995), the ideal observer selected the most likely direction of gaze (left or right). Using QUEST, a Bayesian adaptive threshold estimator, the RMS contrast for the next trial was adjusted based on the correctness of the ideal observer’s response, such that correct responses generally led to lower (less visible) contrast levels (Watson and Pelli 1983). Each contrast threshold was estimated based on 240 trials, and we estimated 25 thresholds per condition. Finally, note that in the absence of noise, the ideal observer will never respond incorrectly at any contrast level, so we added Gaussian white noise (RMS contrast = 0.28) to the stimulus on every trial. A new noise sample was generated for each trial, so that the appearance of the noise varied randomly across trials.