Journal of Autism and Developmental Disorders

, Volume 43, Issue 2, pp 286–300

Getting a Grip on Social Gaze: Control over Others’ Gaze Helps Gaze Detection in High-Functioning Autism

Authors

    • Department of PsychologyUniversity of Cologne
  • Caroline Schwartz
    • Department of PsychologyUniversity of Cologne
  • Kliment Yanev
    • Department of PsychologyUniversity of Cologne
  • Leonhard Schilbach
    • Department of PsychiatryUniversity Hospital of Cologne
    • Max-Planck-Institute for Neurological Research
  • Kai Vogeley
    • Department of PsychiatryUniversity Hospital of Cologne
    • Institute for Neuroscience and MedicineCognitive Neuroscience INM3, Research Center Juelich
  • Gary Bente
    • Department of PsychologyUniversity of Cologne
Original Paper

DOI: 10.1007/s10803-012-1569-x

Cite this article as:
Dratsch, T., Schwartz, C., Yanev, K. et al. J Autism Dev Disord (2013) 43: 286. doi:10.1007/s10803-012-1569-x

Abstract

We investigated the influence of control over a social stimulus on the ability to detect direct gaze in high-functioning autism (HFA). In a pilot study, 19 participants with and 19 without HFA were compared on a gaze detection and a gaze setting task. Participants with HFA were less accurate in detecting direct gaze in the detection task, but did not differ in their ability to establish direct gaze in the setting task. In the main experiment, the results of the pilot study were replicated with 37 participants with and 39 without HFA, suggesting that individuals with HFA have a specific deficit in the passive perception of social cues as opposed to the active control, which seems to be intact.

Keywords

High-functioning autismDirect gazeControlPredictability

Introduction

The perception of direct gaze—i.e., being gazed at directly by another person—is crucial to human social interaction: it can signal the attention of others and mediate the perception of displayed emotions (Adams and Kleck 2005). In addition, it provides the basis for mutual gaze, the shared gaze between two persons. The ability to detect direct gaze is a fundamental function established early in our evolution (Myowa-Yamakoshi et al. 2003) and can be measured as early as in 2- to 5-day-old infants, who are able to distinguish between faces with direct and averted gaze, and who prefer to look at faces with direct gaze (Farroni et al. 2002).

Recent studies have focused on the question of how accurate individuals with autism are at detecting direct gaze (see Nation and Penny 2008; Senju and Johnson 2009). Even though there are several studies that hint at deficits (Gepner et al. 1996; Howard et al. 2000; Swettenham et al. 2001; Senju et al. 2003; Wallace et al. 2006), there are also studies that show no deficits (Senju et al. 2005, 2008; Webster and Potter 2011). In addition, the developmental trajectory of the putative impairment is unclear. Because accumulating neuropsychological evidence hints at a general difference between individuals with autism and neurotypical individuals in the way they process direct gaze (Grice et al. 2005; Kylliäinen and Hietanen 2006; Joseph et al. 2008; Elsabbagh et al. 2009; Akechi et al. 2010; Pitskel et al. 2011), it might be possible that the ability to detect direct gaze is impaired even in adults with autism.

Only two studies so far have investigated the ability to detect direct gaze in adults with autism: Howard et al. (2000) found differences between individuals with autism and neurotypical individuals on a task in which participants had to select one image displaying direct gaze out of two. However, using the same paradigm, Webster and Potter (2011) were not able to replicate this finding. The fact that this impairment has not been consistently observed can in part be explained by several shortcomings of previous studies.

First, all studies so far have used tasks that do not measure the critical ability in question—namely, being able to differentiate between direct gaze and averted gaze. All studies did use either a comparison task (Gepner et al. 1996; Howard et al. 2000; Swettenham et al. 2001; Webster and Potter 2011) or a visual search task (Senju et al. 2003, 2005, 2008). In the comparison task, participants had to select out of pairs of images the one displaying direct gaze. In the visual search task, participants had to find one face displaying direct gaze that was presented along with several faces showing averted gaze. Whereas these tasks might be useful to find crude differences in accuracy to detect direct gaze, different task settings modeling natural situations that require detection of direct gaze might provide a better framework to detect subtle impairments (Wilms et al. 2010).

Second, all studies used large variations of gaze direction. The visual search task described above included only two variations—direct gaze or averted gaze. Even though the degrees of deviation were not reported in the studies (e.g., Senju et al. 2003), compared to the stimuli used in our experiment we conclude that the gaze was at least averted by 15°. In the comparison task (e.g., Webster and Potter 2011), gaze was averted by either 20°, 15°, 10°, or 5°. Psychophysical studies investigating the accuracy to detect deviations from direct gaze, however, have found that humans are able to distinguish degrees between direct and averted gaze that are far more subtle (Gamer and Hecht 2007). Thus, previous studies did not use stimuli in a range (between 0° and 10°) that provides a sufficient level of ambiguity to detect group differences.

All these factors might explain why previous studies of direct gaze detection in autism have failed to find clear differences between individuals with autism and typically developing individuals. Thus, we tried to address these shortcomings in the present study by using a realistic virtual character showing realistic gaze behavior that varied in small increments between direct and averted gaze.

Additionally, it seems important to consider what other factors of gaze behavior might contribute to a deficit in detecting direct gaze in autism. One important aspect of gaze behavior that previous studies have ignored is the social nature of gaze (Wilms et al. 2010). In naturalistic situations, gaze is embedded in a social context, signaling communicative meaning and creating affordances to act (Schilbach et al. 2011). Most importantly, gaze is often ambiguous and unpredictable. It might be this unpredictability that could interfere with the ability to detect direct gaze in individuals with autism. Several theories have identified the predictability of social situations as an important factor influencing the behavior of individuals with autism (Gomot and Wicker 2011; Qian and Lipkin 2011; Baron-Cohen 2009; Dawson et al. 1998). For instance, Baron-Cohen et al. (2009) proposed that autism is characterized by a strong drive to predict and control the environment. Consequently, it has been shown that children with autism favor predictable over unpredictable environments (Ferrara and Hill 1980) and exhibit more problematic behaviors, such as self-hits and aggression, in unpredictable environments (Flannery and Horner 1994). If the unpredictability of a situation is removed by imitating children with autism, communicative gaze increases (Sanefuji and Ohgami 2011). This preference for predictable over unpredictable stimuli seems to extend to simple displays of motion. When presented with point-light-displays of either biological motion or non-social contingencies, two-year-olds with autism spend more time looking at non-social contingencies (Klin et al. 2009). Besides preferring predictable stimuli, children with autism seem to be impaired in their ability to predict the variability of common real-life events (Loth et al. 2010).

One of the key features of social gaze is its independence and goal-directedness that can only be partly predicted by relying on general knowledge about social situations or taking the perspective of another person. Inanimate objects and non-social systems, on the other hand, can be understood and predicted in a mechanistic way by observing and controlling the relations between the input and output of the system.

Social gaze, therefore, poses a double burden to individuals with autism: First, because of a general deficit in social interaction, individuals with autism are impaired in their ability to understand and interpret the gaze of another person and are therefore not able to predict it. Second, because the gaze of another person usually cannot be controlled, individuals with autism are not able to explore the underlying rules of the gaze behavior to make it more predictable in the future. In other words, individuals with autism might be impaired in the processing of social gaze because it appears unpredictable to them. A possible explanation for the development of this impairment has been provided by the social orienting hypothesis (Dawson et al. 1998), according to which children with autism are less drawn to social stimuli because they seem unpredictable to them. Because they spend less time attending to social stimuli than typically developing children, they miss the opportunity to learn about social stimuli, which would make them more predictable to them. Thus, a negative feedback loop develops—unpredictable social stimuli cause children with autism to spend less time attending to them, which makes social stimuli even more unpredictable and causes children with autism to spend even less time attending to them. One way to break this feedback loop would be to create predictable social environments that can be controlled by individuals with autism to help them overcome their deficits.

Based on these assumptions, the following prediction can be made: If given the opportunity to control the gaze of another person and thus increase the predictability of the situation, individuals with autism should be able to overcome their deficit. Because the gaze of another person usually cannot be controlled in real-life situations, individuals with autism have no opportunity to experience social gaze in a predictable environment. However, computer-generated stimuli allow to create social gaze stimuli that are ecologically valid and yet can be well controlled (Wilms et al. 2010). This approach was used in the present study to investigate the relation between the controllability of social stimuli and the detection of direct gaze by individuals with autism.

In sum, the aim of the present study was twofold: First, using a realistic virtual character displaying dynamic gaze behavior with fine-tuned variations, we investigated whether individuals diagnosed with autism are impaired in their ability to detect direct gaze. Second, by giving participants control over the gaze of a social character, we examined the effect of having control over a social stimulus on gaze detection abilities.

Pilot Study

Method

Participants

Nineteen adults with high-functioning autism (10 males, 9 females, Mage = 39.1, age range: 23–53 years) were recruited at the Adult Autism Outpatient Clinic of the Department of Psychiatry at the University Hospital of a large city in western Germany. All participants with high-functioning autism were diagnosed by two independent physicians according to ICD-10 criteria. In line with other studies (e.g., Frith and de Vignemont 2005), high-functioning autism is used as an umbrella term for both high-functioning autism and Asperger syndrome. Nineteen control participants (11 males, 8 females, Mage = 33.8, age range: 20–48 years) were recruited at the Department of Psychology at the University of a large city in western Germany. Both groups were matched for age, gender, years of formal education, and intelligence (measured with the German version of the Wechsler Adult Intelligence Scale-Revised (WAIS-R); Tewes 1994). Participants also completed the Beck Depression Inventory (BDI; Beck and Steer 1987; Hautzinger 1995), the Autism Spectrum Quotient (AQ: Baron-Cohen et al. 2006), and the “Reading the Mind in the Eyes Test” (ToM-Eyes; Baron-Cohen et al. 2001). The study protocol had been approved by the local ethics committee.

On average, there was no difference in age between participants diagnosed with autism (M = 39.1, SD = 9.4) and participants in the control group (M = 33.8, SD = 7.3), t(36) = 1.95, p = .060, d = 0.63. See Table 1 for an overview of demographic and psychopathological variables for both groups.
Table 1

Demographic, psychopathological, and IQ results for the pilot study

 

HFA (n = 19)

Control (n = 19)

 

M

SD

M

SD

t(36)

p

Cohen’s d

Age (years)

39.1

9.4

33.8

7.3

1.95

.060

0.63

Education (years)

18.3

3.0

20.1

4.5

−1.42

.163

−0.47

WAIS-R verbal IQ

124.9

11.5

126.6

7.4

−0.52

.605

−0.18

WAIS-R performance IQ

123.3

13.8

127.7

11.4

−1.08

.288

−0.35

WAIS-R IQ (total)

127.0

12.3

132.2

8.1

−1.53

.135

−0.50

BDI

11.3

7.1

4.9

3.1

3.60

.001

1.17

AQ

41.1

3.8

14.2

5.5

17.66

<.001

5.69

ToM-Eyes

16.2

4.2

18.6

3.1

−2.03

.050

−0.65

WAIS-R Wechsler Intelligence Scale for Adults, BDI Beck Depression Inventory, AQ Autism Spectrum Quotient, ToM-Eyes Reading the Mind in the Eyes Test

There was also no difference between the HFA group (M = 18.3, SD = 3.0) and the control group (M = 20.1, SD = 4.5) with regard to years of formal education, t(36) = −1.42, p = .163, d = −0.47. Furthermore, there was no difference in intelligence between participants diagnosed with autism (M = 127.0, SD = 12.3) and participants in the control group (M = 132.2, SD = 8.1), t(36) = −1.53, p = .135, d = −0.50. Participants diagnosed with autism had higher scores on the Beck Depression Inventory (BDI; Beck and Steer 1987; Hautzinger 1995) (M = 11.3, SD = 7.1) than participants in the control group (M = 4.9, SD = 3.1), t(36) = 3.60, p = .001, d = 1.17. This was to be expected because high-functioning autism is known to have a higher prevalence of depression (Stewart et al. 2006; Lehnhardt et al. 2011). Confirming clinical diagnosis, participants diagnosed with autism scored higher on the Autism Spectrum Quotient (M = 41.1, SD = 3.8) than participants in the control group (M = 14.2, SD = 5.5), t(36) = 17.66, p < .001, d = 5.69. In addition, there was a marginally significant difference on the ToM-Eyes test between the HFA group (M = 16.2, SD = 4.2) and the control group (M = 18.6, SD = 3.1), t(36) = −2.03, p = .050, d = −0.65.

Stimuli and Design

In the detection task, we tested the participants’ accuracy to detect when a realistic virtual character was gazing directly at them.

As the stimulus in the detection task, we used a realistic virtual character created in the commercially available software Poser 6 (see Fig. 1a).
https://static-content.springer.com/image/art%3A10.1007%2Fs10803-012-1569-x/MediaObjects/10803_2012_1569_Fig1_HTML.jpg
Fig. 1

a Realistic virtual character, b reduced virtual character, and c non-social, geometric stimulus

Because the criterion for judging gaze as direct can be biased by a smiling expression, with smiles increasing the chance of perceiving gaze as direct (Martin and Rovira 1982), the virtual character showed a neutral expression. The virtual character was moving its eyes and fixated different positions. Fixations included direct gaze, slightly averted gaze, and clearly averted gaze. To describe the position of the eyes, we used a system with two rotations (y and z). Y-rotation described the degree of deviation from the center of the eyes (see Fig. 2a); z-rotation described the direction of the deviation from the center of the eyes (see Fig. 2b).
https://static-content.springer.com/image/art%3A10.1007%2Fs10803-012-1569-x/MediaObjects/10803_2012_1569_Fig2_HTML.gif
Fig. 2

Gaze positions at different degrees of ay-rotation and bz-rotation

To determine which degrees of y-rotation would be perceived as direct gaze, a pilot study was conducted. 20 participants (13 females, 7 males) saw pictures of the virtual character with the gaze angle varying on the y-axis from 1° to 10°. For every picture, the participants rated whether they felt that the virtual character’s gaze was direct. Rotations of 1° and 2° were mistaken for direct gaze by 80 % of the participants (3°: 70 %, 4°: 60 %, 5°: 35 %, 6°: 10 %). All participants correctly identified deviations of 7° and more as averted gaze. As a result, gaze angles ranging from 0° to 7° on the y-axis were used in the experiment, because they provided a sufficient level of ambiguity. Because the y-axis only determined the amount of deviation from the center of the eye, different degrees on the z-axis were used to vary the direction of the virtual character’s gaze. Combining both y- and z-angles a realistic gaze sequence with 10 hits and 20 distractors was created. The order in which the different y-angles appeared in the gaze sequence was completely randomized and a set of 30 trials was programmed. At the beginning of each trial, the virtual character showed clearly averted gaze (15–25° on the y-axis) for 2,000 ms. After that, it moved its eyes to a fixation point of either 0° (hit) or 1–7° (false alarm). The duration of the fixation was randomly chosen to be either 2,000 or 4,000 ms based on earlier studies on social gaze (Bente et al. 1998, 2007; Kuzmanovic et al. 2009) that have shown that a gaze duration of around 4,000 ms is perceived as pleasant. Therefore, each trial lasted either 4,000 or 6,000 ms. Participants used a mouse to signal when the virtual character was showing direct gaze, clicking the left mouse button each time they felt the virtual character was gazing directly at them. Dependent variables were reaction times, the number of correct responses (hits), and the number of false alarms. Because previous studies have shown that social stimuli, such as faces, are less pleasant for individuals with autism (Corden et al. 2008; Hutt and Ounsted 1966; Richer and Coss 1976), we asked participants to rate the pleasantness (1 = very unpleasant to 5 = very pleasant) of the task on a 5-point scale. If gaze detection is impaired in individuals with autism, a task demanding participants to detect direct gaze might be perceived as difficult. Therefore, we also asked participants to rate the difficulty (1 = very easy to 5 = very difficult) of the task on a 5-point scale.

In the setting task, we tested the participants’ accuracy to actively establish direct gaze with a virtual character. As the virtual character we used a reduced avatar that had been employed in previous experiments of our research group (Bente et al. 2007). Instead of detecting when the virtual character was gazing at them, participants actively established direct gaze using the mouse. In each trial, the eyes of the virtual character appeared in a random position of averted gaze. In addition, the rotation of the virtual character′s head was also varied between 0° and 30° on the horizontal, vertical, and sagittal axis. Participants used the mouse to move the eyes of the virtual character into a position that matched direct gaze as closely as possible. When the participants had established that position, they pressed a button on the mouse and the eyes of the virtual character moved to a new random position. Participants completed a sequence of 16 trials. There was no time limit for each trial to allow the participants to work as accurately as possible. Dependent variables were reaction times and degrees of deviation from a perfect eye contact.

Procedure

All participants were tested at the Department of Psychiatry at the University Hospital of a large city in western Germany. Prior to the experiment, all participants gave consent to participate. All participants had normal or corrected-to-normal visual acuity. The experimental stimuli were presented on a 17-in. monitor with a resolution of 1,280 × 1,024 pixels. Each participant completed both the setting and detection task in one single session. To eliminate carry-over-effects, all tasks were presented in random order. At the beginning of the experiment, the experimenter greeted the participants and told them that they were going to take part in an experiment on social perception. After that, they sat down in a chair approximately 50 cm from the computer. They positioned their head so that the eyes were looking at the centre of the screen, and the experimenter started the experiment. The participants then saw a grey screen with white instructions explaining the first of the two tasks. Participants could read the instructions at their own pace and then proceeded to the actual task. After they had completed the first task, participants proceeded to the instructions for the second task. Participants rated the pleasantness and the difficulty of the detection task on a paper questionnaire next to the computer. The experiment ended, after participants had completed all tasks. After the experiment, participants completed the BDI (Beck and Steer 1987; Hautzinger 1995), the AQ (Baron-Cohen et al. 2006), the WAIS-R (Tewes 1994), and the ToM-Eyes test (Baron-Cohen et al. 2001).

Results and Discussion

Detection Task

Accurate gaze detection includes both correctly detecting direct gaze and classifying gaze as averted that is not direct. A person classifying every gaze direction in the experiment as direct would correctly detect all instances of direct gaze, but the high rate of false alarms would indicate that this person is not truly able to distinguish between direct and averted gaze. To compare both groups’ ability to detect direct gaze, we calculated a sensitivity index d′ out of the number of correct responses and the number of false alarms. There was a significant difference in d′ between the HFA group (M = 0.18, SD = 0.51) and the control group (M = 1.00, SD = 0.45), t(36) = −5.19, p < .001, d = −1.71 (see Fig. 3a).
https://static-content.springer.com/image/art%3A10.1007%2Fs10803-012-1569-x/MediaObjects/10803_2012_1569_Fig3_HTML.gif
Fig. 3

Overall results of the pilot study: a detection task and b setting task. Error bars represent 95 %-CI of the mean. * p < .05, ** p < .001

To analyze at which degrees of deviation both groups differed, the percentage of false alarms at each angle, ranging from 1° to 7°, was compared in a 2 × 7 mixed ANOVA (group × angle). As expected, there was a significant effect of group on percent of false alarms, F(1, 36) = 36.82, p < .001, ηp2 = .506. There was also a significant effect of angle, F(6, 216) = 6.41, p < .001, ηp2 = .151. There was no significant interaction effect between group and angle, F(6, 216) = 0.26, p = .957, ηp2 = .007. Independent sample t tests using a Bonferroni-corrected alpha (α = .05/7 = .007) were used to compare both groups’ performance for the different degrees (1°–7°). As shown in Table 2, participants in the HFA group produced significantly more false alarms than control participants at 3°, 4°, 5°, and 6°.
Table 2

Percent of false alarms at different angles (1°–7°)

Angle (°)

HFA (n = 19)

Control (n = 19)

 

M

SD

M

SD

t(36)

p

Cohen’s d

1

100.00

0.00

78.95

41.89

2.19

.042

0.71

2

100.00

0.00

68.42

47.76

2.88

.010

0.94

3

92.11

18.73

63.16

36.67

3.06

.005

0.99

4

83.16

17.97

58.95

22.58

3.66

.001

1.19

5

86.32

21.14

63.16

23.35

3.21

.003

1.04

6

72.37

35.25

38.16

33.71

3.06

.004

0.99

7

76.32

34.83

47.37

42.41

2.30

.028

0.75

Using a Bonferroni-corrected alpha (α = .05/7 = .007), only p values below .007 are considered to be significant

With regard to reaction times, there was no significant difference between the HFA group (M = 1,180 ms, SD = 669 ms) and the control group (M = 1,340 ms, SD = 421 ms), t(36) = −0.99, p = .329, d = −0.32.

Perceived pleasantness of the task did differ significantly between the HFA group (M = 2.56, SD = 0.92) and the control group (M = 3.47, SD = 0.90), t(35) = −3.06, p = .004, d = −1.00. This indicates that participants in the control group perceived the task with the virtual character to be more pleasant than participants in the HFA group.

In addition, participants in the HFA group (M = 2.89, SD = 1.08) did perceive the task to be more difficult than participants in the control group (M = 1.79, SD = 0.85), t(35) = 3.45, p = .001, d = 1.13.

Setting Task

There was no significant difference between the HFA group (M = 5.89, SD = 1.87) and the control group (M = 5.99, SD = 2.19) in the accuracy to establish direct gaze, t(36) = −0.16, p = .876, d = −0.05 (see Fig. 3b).

However, there was a significant difference with regard to reaction times between the HFA group (M = 4,501 ms, SD = 1,703 ms) and the control group (M = 3,542 ms, SD = 1,148 ms), t(36) = 2.04, p = .049, d = 0.66, indicating that participants in the HFA group did take longer to complete the setting task.

Taken together, the results of the pilot study provide evidence that adults diagnosed with autism are impaired in the detection of direct gaze indexed by difficulties to distinguish subtle degrees of difference between averted and direct gaze. Especially gaze that is averted by 3°, 4°, 5°, and 6° was routinely mistaken to be direct by the participants in the autism group. However, there seems to be a threshold at around 7° at which both groups agree that gaze is averted. In line with theories about the aversiveness of social gaze for individuals with autism (e.g., Corden et al. 2008), participants in the autism group did perceive the detection task to be less pleasant and more difficult than the control group. As for the setting task, participants with autism were as accurate as participants in the control group at establishing direct gaze with the virtual character, indicating that the control over a stimulus might be an important mediating factor in the detection of direct gaze.

Main Experiment

Even though the pilot study did provide support for our hypotheses, it suffered from several limitations. First, there was no non-social control stimulus in either the detection or setting task to which the performance of both groups could have been compared. Thus, it is possible that the lower performance of the HFA group in the detection task is not limited to the social domain and does rather reflect a general processing deficit. To address this issue, we added a non-social control task to our design to assess the participants’ performance in a non-social setting.

Second, whereas the detection task did use a realistic virtual character, the setting task did only use a reduced virtual character that consisted of the eye-region including the nose. Therefore, the fact that we did not find any difference in the performance on this task between the HFA group and the control group might be due to the reduced appearance of the virtual character. To exclude this alternative explanation, in the main experiment we did use the same realistic virtual character for both the detection and the setting task.

Third, in the pilot study pleasantness and difficulty was only assessed for the detection task. In order to investigate whether control over the stimulus does affect the subjective experience of the task, we also measured pleasantness and difficulty for the setting task in the main experiment.

Method

Participants

Thirty-seven adults with high-functioning autism (23 males, 14 females, Mage = 34.1, age range: 18–51 years) were recruited at the Adult Autism Outpatient Clinic of the Department of Psychiatry at the University Hospital of a large city in western Germany. All participants with high-functioning autism were diagnosed by two independent physicians according to ICD-10 criteria. Thirty-nine control participants (23 males, 16 females, Mage = 31.1, age range: 24–45 years) were recruited at the Department of Psychology at the University of a large city in western Germany. Both groups were matched for age, gender, and intelligence (measured with the German version of the Wechsler Adult Intelligence Scale-III (WIE—Wechsler Intelligenztest für Erwachsene); Von Aster et al. 2006). Participants also completed the Beck Depression Inventory (BDI; Beck and Steer 1987; Hautzinger 1995), the Autism Spectrum Quotient (Baron-Cohen et al. 2006), the Empathy Quotient (Baron-Cohen and Wheelwright 2004), and the Systemizing Quotient (Wheelwright et al. 2006). The study protocol had been approved by the local ethics committee.

On average, there was no difference in age between participants diagnosed with autism (M = 34.1, SD = 10.1) and participants in the control group (M = 31.1, SD = 4.3), t(74) = 1.71, p = .092, d = 0.39. See Table 3 for an overview of demographic and psychopathological variables for both groups.
Table 3

Demographic, psychopathological, and IQ results for the main experiment

Variable

HFA (n = 37)

Control (n = 39)

 

M

SD

M

SD

df

t

p

Cohen’s d

Age (years)

34.1

10.1

31.1

4.3

74

1.71

.092

0.39

WIE verbal IQa

112.4

17.0

113.8

12.9

72

−0.42

.679

−0.09

WIE: performance IQa

103.5

20.4

107.2

13.2

72

−0.93

.357

−0.22

WIE IQ (total)a

109.0

18.8

112.1

13.0

72

−0.82

.417

−0.19

BDI

11.7

10.7

6.0

5.1

74

2.98

.004

0.68

AQ

40.2

4.8

15.8

3.9

74

24.43

<.001

5.58

EQb

20.2

9.9

45.6

10.6

72

−10.63

<.001

−2.48

SQb

42.2

16.3

25.9

10.1

72

5.25

<.001

1.2

WIE Wechsler Intelligenztest für Erwachsene, BDI Beck Depression Inventory, AQ Autism Spectrum Quotient, EQ Empathizing Quotient, SQ Systemizing Quotient

aTwo participants diagnosed with autism completed an older version of the intelligence test (WAIS-R, German version; Tewes 1994) and were thus excluded from these analyses

bTwo participants diagnosed with autism did not complete the EQ and SQ and were thus excluded from the analyses

Furthermore, there was no difference in intelligence between participants diagnosed with autism (M = 109.0, SD = 18.8) and participants in the control group (M = 112.1, SD = 13.0), t(72) = −.82, p = .417, d = −0.19.

Participants diagnosed with autism had higher BDI scores (Beck and Steer 1987; Hautzinger 1995) (M = 11.7, SD = 10.7) than participants in the control group (M = 6.0, SD = 5.1), t(74) = 2.98, p = .004, d = 0.68.

As expected, participants diagnosed with autism scored higher on the AQ (HFA: M = 40.2, SD = 4.8; Control: M = 15.8, SD = 3.9), t(74) = 24.43, p < .001, d = 5.58, lower on the EQ (HFA: M = 20.2, SD = 9.9; Control: M = 45.6, SD = 10.6), t(72) = −10.63, p < .001, d = −2.48, and higher on the SQ (HFA: M = 42.2, SD = 16.3; Control: M = 25.9, SD = 10.1), t(72) = 5.25, p < .001, d = 1.2.

Stimuli and Design

In the detection task, we tested the participants’ accuracy to detect when a realistic virtual character was gazing directly at them. To assess whether participants were generally able to perform the cognitive processes to solve the task, they also completed a non-social control task with a geometric stimulus.

As the stimulus in the social condition, we used the same male virtual character from the pilot study (see Fig. 1a). The virtual character was moving its eyes and fixated at different positions. Fixations included direct gaze, slightly averted gaze, and clearly averted gaze. To reduce the time of the experiment, we used a shorter gaze sequence as opposed to the pilot study with 5 hits and 14 distractors. At the beginning of each trial, the virtual character showed clearly averted gaze (15°–25° on the y-axis). After that, it moved its eyes to a fixation point of either 0° (hit) or 1–7° (false alarm). The timing was the same as in the pilot study, with each trial lasting either 4,000 or 6,000 ms. Participants used a mouse to signal when the virtual character was showing direct gaze, clicking the left mouse button each time they felt the virtual character was gazing directly at them.

The non-social stimulus was a grey rectangle with a black cross (see Fig. 1c). It had approximately the size of one of the virtual character’s eyes (height = 1 cm, length = 2.5 cm). The cross was moving inside the rectangle and paused at different positions—directly in the center of the rectangle, slightly off-center, and clearly off-center. At the beginning of each trial, the cross started in a position that was clearly off-center (15°–25° on the y-axis) and then moved to a point of either 0° (hit) or 1–7° (false alarm). The geometric stimulus used the same sequence of hits and distractors as the social stimulus. The timing was the same as in the social task. Again, participants used a mouse to signal when the cross was directly in the center of the rectangle. The design was a 2 (stimulus: social vs. non-social) × 2 (group: HFA vs. control) mixed-design, with stimulus as a within-subjects factor and group as a between-subjects factor. Both experimental tasks were presented in random order. Dependent variables were reaction times, the number of correct responses (hits) and the number of false alarms. In addition, participants rated the pleasantness (1 = very unpleasant to 5 = very pleasant) and difficulty (1 = very easy to 5 = very difficult) of every task on a 5-point scale.

In the setting task, we tested the participants’ accuracy to actively establish direct gaze with a realistic virtual character. Participants also completed a geometric control task to assess their ability to solve the task in a non-social setting. The setting task used the same stimuli as the detection task. However, instead of detecting when the virtual character was gazing at them, participants actively established direct gaze using the mouse. Both the social and the non-social condition of the setting task in the main experiment consisted of 21 trials. In each trial, the eyes of the virtual character appeared in a random position. Participants then used the mouse to move the eyes of the virtual character into a position that matched direct gaze as closely as possible. When the participants had established that position, they pressed a button on the mouse and the eyes of the virtual character moved to a new random position. In contrast to the pilot study, the rotation of the virtual character’s head was not varied. There was no time limit for each trial to allow the participants to work as accurately as possible.

The task in the non-social condition was similar to the task in the social condition. However, instead of controlling the virtual character’s eyes, participants tried to center the cross in the center of the rectangle. The design was a 2 (stimulus: social vs. non-social) × 2 (group: HFA vs. control) mixed-design, with stimulus as a within-subjects factor and group as a between-subjects factor. Both experimental tasks were presented in random order. Dependent variables were reaction times and the degree of deviation from a perfect eye contact. Additionally, participants evaluated the pleasantness (1 = very unpleasant to 5 = very pleasant) and difficulty (1 = very easy to 5 = very difficult) of every task on a 5-point scale.

Procedure

The participants diagnosed with autism were tested at the Department of Psychiatry at the University Hospital of a large city in western Germany. The control participants were tested at the Department of Psychology of the University in the same city. Prior to the experiment, all participants gave consent to participate. All participants had normal or corrected-to-normal visual acuity. The experimental stimuli were presented on a 17-in. monitor with a resolution of 1,280 × 1,024 pixels. The experimental procedure was the same as in the pilot study. However, in addition to the detection and the setting task with the realistic virtual character, participants also completed both tasks with the non-social stimulus. Each participant completed all tasks in one single session. To eliminate carry-over-effects, all tasks were presented in random order.

Results and Discussion

Detection Task

To compare both groups’ ability to detect direct gaze, we calculated a sensitivity index d′ out of the number of correct responses and the number of false alarms. Sensitivity d′ was compared in a 2 × 2 mixed ANOVA (group × stimulus). There was a significant main effect of stimulus on d′, F(1, 74) = 57.0, p < .001, ηp2 = .435. There was also a significant main effect of group on d′, F(1, 74) = 5.13, p = .026, ηp2 = .065. There was no significant interaction effect between stimulus and group, F(1, 74) = 0.80, p = .375, ηp2 = .011. Pairwise comparisons were performed comparing the differences in d′ of the HFA and the control group in the social and non-social test conditions. These revealed a significant difference in the social condition, t(74) = −2.27, p = .026, d = −0.51. There was no significant difference in the non-social condition, t(74) = −1.46, p = .149, d = −0.33. This suggests that when confronted with the non-social stimulus participants diagnosed with autism (M = 0.99, SD = 0.54) were as sensitive at detecting when the geometric stimulus was centered as participants in the control group (M = 1.15, SD = 0.41). When dealing with the virtual character, however, participants in the HFA group (M = 0.45, SD = 0.60) were less sensitive at detecting direct gaze than participants in the control group (M = 0.72, SD = 0.45) (see Fig. 4a).
https://static-content.springer.com/image/art%3A10.1007%2Fs10803-012-1569-x/MediaObjects/10803_2012_1569_Fig4_HTML.gif
Fig. 4

Overall results of the detection task in the main experiment: a false alarms, b reaction times, c pleasantness of the task, and d difficulty of the task. Error bars represent 95 %-CI of the mean. * p < .05, ** p < .001

Reaction times were analyzed in a 2 × 2 mixed ANOVA (group × stimulus). There was no significant main effect for stimulus, F(1, 74) = 0.74, p = .392, ηp2 = .010, or group, F(1, 74) = 0.77, p = .383, ηp2 = .010. There was also no significant interaction effect between stimulus and group, F(1, 74) = 2.87, p = .095, ηp2 = .037. Pairwise comparisons revealed no difference between the HFA group (M = 1,238 ms, SD = 242 ms) and the control group (M = 1,222 ms, SD = 250 ms) in the social condition, t(74) = 0.28, p = .777, d = 0.07. There was also no difference between the HFA group (M = 1,142 ms, SD = 303 ms) and the control group (M = 1,253 ms, SD = 347 ms) in the non-social condition, t(74) = −1.49, p = .139, d = −0.34 (see Fig. 4b).

Perceived pleasantness of the task was analyzed in a 2 × 2 mixed ANOVA (group × stimulus). There was no significant main effect for stimulus, F(1, 74) = 0.67, p = .415, ηp2 = .009. There was a significant main effect for group, F(1, 74) = 4.55, p = .036, ηp2 = .058. There was also a significant interaction effect between stimulus and group, F(1, 74) = 20.06, p < .001, ηp2 = .213. Pairwise comparisons were performed to compare both groups in the social and non-social condition. Perceived pleasantness did differ significantly between the HFA group (M = 2.84, SD = 1.09) and the control group (M = 3.90, SD = 0.94) in the social condition, t(74) = −4.54, p < .001, d = −1.04. This indicates that individuals with autism did perceive the social task to be less pleasant than participants in the control group, whereas in the non-social condition there was no difference in perceived pleasantness of the task between the HFA group (M = 3.35, SD = 1.27) and the control group (M = 3.15, SD = 0.96), t(74) = 0.77, p = .446, d = 0.18 (see Fig. 4c).

Perceived difficulty of the task was analyzed in a 2 × 2 mixed ANOVA (group × stimulus). There was no significant main effect for stimulus, F(1, 73) = 1.74, p = .192, ηp2 = .023. There was also no significant main effect for group, F(1, 73) = 2.98, p = .089, ηp2 = .039. The interaction between stimulus and group was not significant, F(1, 73) = 3.23, p = .076, ηp2 = .042. Pairwise comparisons revealed a significant difference between the HFA group (M = 3.14, SD = 1.03) and the control group (M = 2.47, SD = 1.22) in the social condition, t(73) = 2.53, p = .014, d = 0.59, suggesting that participants diagnosed with autism did perceive the social task to be more difficult than participants in the control group. The HFA group (M = 3.10, SD = 1.25) and the control group (M = 3.00, SD = 1.12) did not differ in their perception of the difficulty of the non-social task, t(73) = 0.20, p = .844, d = 0.08 (see Fig. 4d).

Setting Task

Degrees of deviation from a perfect direct gaze were analyzed in a 2 × 2 mixed ANOVA (group × stimulus). There was a significant main effect for stimulus, F(1, 74) = 246.45, p < .001, ηp2 = .769. There was no significant main effect for group, F(1, 74) = 2.39, p = .126, ηp2 = .031. There was no significant interaction effect between stimulus and group, F(1, 74) = 0.07, p = .745, ηp2 = .001. Pairwise comparisons revealed no significant difference between the HFA group (M = 4.95, SD = 1.68) and the control group (M = 4.62, SD = 1.39) in the social condition, t(74) = 0.94, p = .353, d = 0.21. There was also no significant difference between the HFA group (M = 2.46, SD = 1.16) and the control group (M = 2.02, SD = 0.77) in the non-social condition, t(74) = 1.94, p = .056, d = 0.45. This indicates that there was no difference in accuracy between participants diagnosed with autism and participants in the control group on either the social or non-social setting task (see Fig. 5a).
https://static-content.springer.com/image/art%3A10.1007%2Fs10803-012-1569-x/MediaObjects/10803_2012_1569_Fig5_HTML.gif
Fig. 5

Overall results of the Setting Task in the main experiment: a degrees of deviation, b reaction times, c pleasantness of the task, and d difficulty of the task. Error bars represent 95 %-CI of the mean. * p < .05, ** p < .001

Reaction times were analyzed in a 2 × 2 mixed ANOVA (group × stimulus). There was a significant main effect for stimulus, F(1, 74) = 24.86, p < .001, ηp2 = .251. There was no significant main effect for group, F(1, 74) = 0.21, p = .648, ηp2 = .003. There was also no significant interaction effect between stimulus and group, F(1, 74) = 0.36, p = .550, ηp2 = .005. Pairwise comparisons revealed no significant difference between the HFA group (M = 5,736 ms, SD = 2,363 ms) and the control group in the social condition, t(74) = 0.94, p = .353, d = 0.21. There was also no significant difference between the HFA group (M = 6,978 ms, SD = 3,882 ms) and the control group (M = 6,889 ms, SD = 2,732 ms) in the non-social condition, t(74) = 0.12, p = .908, d = 0.03. This indicates that both groups needed a comparable amount of time to complete the task, and that there was no speed-accuracy trade-off in either of the groups (see Fig. 5b).

Perceived pleasantness of the task was analyzed in a 2 × 2 mixed ANOVA (group × stimulus). There was no significant main effect for stimulus, F(1, 74) = 3.47, p = .066, ηp2 = .045. There was a significant main effect for group, F(1, 74) = 7.91, p = .006, ηp2 = .097. There was also a significant interaction effect between stimulus and group, F(1, 74) = 17.66, p < .001, ηp2 = .193. Pairwise comparisons were performed to compare both groups in the social and non-social condition. Perceived pleasantness did differ significantly between the HFA group (M = 2.78, SD = 1.27) and the control group (M = 4.08, SD = 0.87) in the social condition, t(74) = −5.20, p < .001, d = −1.19, indicating that individuals diagnosed with autism did perceive the social task to be less pleasant than participants in the control group. There was no difference in perceived pleasantness of the task between the HFA group (M = 3.19, SD = 1.27) and the control group (M = 3.03, SD = 1.18) in the non-social condition, t(74) = 0.58, p = .562, d = 0.13 (see Fig. 5c).

Perceived difficulty of the task was analyzed in a 2 × 2 mixed ANOVA (group × stimulus). There was a significant main effect for stimulus, F(1, 73) = 9.27, p = .003, ηp2 = .113. There was no significant main effect for group, F(1, 73) = 1.82, p = .182, ηp2 = .024. However, there was a significant interaction effect between stimulus and group, F(1, 73) = 5.79, p = .019, ηp2 = .073. Pairwise comparisons revealed a significant difference between the HFA group (M = 2.59, SD = 1.3) and the control group (M = 1.89, SD = 1.01) in the social condition, t(73) = 2.61, p = .011, d = 0.60, meaning that participants diagnosed with autism did perceive the social task to be more difficult than participants in the control group. There was no difference between the HFA group (M = 2.70, SD = 1.29) and the control group (M = 2.82, SD = 1.16) in the perception of the difficulty of the task in the non-social condition, t(73) = −0.40, p = .690, d = −0.10 (see Fig. 5d).

General Discussion

The main goal of our experiments was to investigate whether adults diagnosed with HFA are impaired in their ability to detect direct gaze. In addition, we tested whether these gaze processing impairments could—at least in part—be due to the uncontrollability of the social stimuli by giving participants control over the gaze of a realistic virtual character.

The results of the detection task with the virtual character in both the pilot study and the main experiment show that adult individuals with high-functioning autism are impaired in their ability to distinguish between direct and averted gaze. In our task, this difficulty to distinguish between direct and averted gaze manifested itself in an increased rate of false alarms, which may suggest that when faced with ambiguous social gaze individuals with autism overcompensate by classifying gaze as direct that is in fact averted.

Our results indicate that this finding may not be explained by a general deficit in the processing of moving stimuli because individuals with autism performed as well as participants in the control group on the detection task with the geometric, non-social stimulus in the main experiment. However, it is important to note that we did only find significant main effects for group and stimulus and no significant interaction between stimulus and group. Therefore, we cannot exclude the possibility that with a larger sample size we may have also detected a significant difference between both groups in the non-social condition, which would hint at a deficit in the processing of non-social stimuli. However, this effect in the non-social condition was only small compared to the effect in the social condition and may not explain the impaired performance of individuals with autism in the social condition. Interestingly, both groups performed better on the non-social than on the social task, which suggests that the social condition was more demanding for both individuals with autism and participants in the control group.

These results are in contrast to previous studies (Senju et al. 2005, 2008; Webster and Potter 2011), which have not shown clear deficits in the detection of direct gaze. This might be in part due to the different methodologies used by the other studies. Whereas Senju et al. (2008) used a visual search task with static stimuli showing only two different kinds of gaze (direct or averted), we used a realistic virtual character showing dynamic gaze behavior with fine-tuned variations. Even though Webster and Potter (2011) did use different degrees of averted gaze (20°, 10°, and 5°), they presented their participants with two images side by side, one displaying direct gaze and the other displaying averted gaze, which might have made the task considerably easier than our task. Our findings support the idea that the study of gaze behavior in autism can benefit from stimuli that are both dynamic and realistic (Gepner et al. 2001; Wilms et al. 2010).

Now that it has been established that individuals with autism are impaired in the detection of direct gaze, it is important to determine the processes that lead to less accurate gaze detection. At least two explanations seem possible:

First, the gaze behavior of the virtual character was highly unpredictable, which might have interfered with the autistic participants’ ability to detect direct gaze. As has been stated by the empathizing-systemizing theory (Baron-Cohen et al. 2003), individuals with autism prefer situations that can be predicted and controlled. Previous studies have shown that children with autism favor predictable environments over unpredictable environments (Ferrara and Hill 1980; Klin et al. 2009) and show more problematic behaviors, such as self-hits and aggression, in unpredictable environments (Flannery and Horner 1994). During the detection task in the present study, participants were confronted with the possibility of being gazed at by a virtual character. This possibility of becoming engaged in social interaction may have been aversive for the individuals with autism and could have interfered with their ability to accurately detect direct gaze. This would explain why the individuals with autism did not differ from the typically developing individuals in the setting task, where the virtual character did not show unpredictable gaze behavior and was instead controlled by the participants.

Second, the individuals with autism may have averted their gaze from the virtual character and were therefore not able to accurately detect when the virtual character was gazing at them. As has been shown in previous studies, when confronted with social stimuli, individuals with autism avert their gaze to avoid looking at the stimulus (Corden et al. 2008; Hutt and Ounsted 1966; Richer and Coss 1976). Thus, it might be possible that the individuals with autism did not attend to the eye region in the detection task and therefore did not accurately detect direct gaze. In the setting task, individuals with autism performed as accurately as participants in the control group. Therefore, the predictability of the situation may have influenced the gaze behavior of the participants with autism causing them to show less gaze aversion. However, without eye-tracking data we can only speculate about the relation between gaze aversion, task performance, and predictability. Thus, a replication of our study should include eye-tracking data to determine which regions of the face individuals with autism fixate on during the different tasks.

As indicated by the results of the questionnaire data in the main experiment, individuals with autism experienced the detection task with the virtual character as less pleasant than participants in the control group. These results are in concordance with previous studies showing that social stimuli, such as faces, are less pleasant for individuals with autism and that they prefer non-social stimuli (Corden et al. 2008; Hutt and Ounsted 1966; Richer and Coss 1976).

As for the reaction times of the detection task in both the pilot study and the main experiment, individuals with autism did not differ from control persons in reaction times in any of the detection tasks, suggesting that there was no trade-off between speed and accuracy in any of the two groups.

The results of the setting task with the virtual character in both the pilot study and the main experiment show that individuals with autism are in fact able to establish direct gaze accurately when they have full control over a social stimulus. When looking at the degrees of deviation from direct gaze in the main experiment, it is important to note that in the setting task individuals with autism were able to distinguish more accurately between direct or averted gaze than in the detection task. Whereas in the setting task individuals with autism perceived gaze between 0° and 4.95° to be direct, in the detection task individuals with autism indicated gaze to be direct that was more than 5° averted. Why were the individuals with autism more accurate at detecting direct gaze in the setting task as opposed to the detection task? First, the setting task was more predictable. Whereas in the detection task the virtual character showed unpredictable gaze behavior, the gaze in the setting task was controlled by the participants, making it highly predictable. If the unpredictability of the gaze behavior interfered with the autistic participants’ ability to detect direct gaze in the detection task, then in the setting task participants with autism could focus more easily on the gaze of the virtual character.

Second, one important difference between both tasks is that there was no time limit in the setting task. In the setting task in the main experiment, participants in the HFA group took about 5,7 s on average to establish direct gaze with the virtual character. In the detection task, on the other hand, the eyes of the virtual character were moving, so the time to make a decision about the virtual character’s gaze was limited. This was also reflected in participants’ reaction times, with the average reaction time in the detection task being about 1,2 s. In the pilot study, individuals with autism also did take significantly longer to complete the setting task than participants in the control group. However, in the main experiment there was no difference in reaction times between both groups in the setting task. Thus, another important aspect for a future study might be to introduce also a time limit for the setting task to investigate whether individuals with autism can still benefit from control over the social stimulus if the time is limited.

Furthermore, even though individuals with autism were as accurate as participants in the control group at establishing direct gaze in the setting task, it is not clear which strategy they used. Previous studies have argued that individuals with autism use a geometric strategy to process gaze (Ristic et al. 2005). Because the setting task in the main study used a virtual character directly facing the participants, it is possible that the individuals with autism used geometric cues, such as the symmetry of the pupils, to solve the task. To further investigate this possibility, a future study might also use a virtual character with a slightly averted head position, a technique that has been used by other studies to interfere with geometric processing of gaze (Senju et al. 2008; Ashwin et al. 2009).

With regard to the subjective experience during the tasks, individuals with autism did perceive both the setting and the detection task with the realistic virtual character to be less pleasant than participants in the control group. Thus, the controllability of the setting task did not cause individuals with autism to rate it as more pleasant. In addition, individuals with autism also did perceive both tasks as more difficult than participants in the control group. Taken together, both results indicate that the perceived pleasantness and difficulty of the tasks was more influenced by the social stimulus per se than by the characteristics of the tasks, suggesting a strong influence of the social nature of the task on experience.

An important question for future studies is whether individuals with autism might be able to generalize skills gained in the setting task to other situations. A simple test would be to use the same paradigm as in the present study, but to ask participants to complete the setting task before the detection task. It might be possible that in this case participants with autism would be more accurate at detecting direct gaze. In the present study, the order of tasks was randomized so such training effects could not be investigated.

On the whole, the results of the setting task show that individuals with autism might be able to improve their ability to detect direct gaze by giving them control over or providing a handle for social stimuli. Previous studies have begun to focus on developing training programs for individuals with autism: For instance, Golan et al. (2009) created an animated series of tank engines with faces to teach children with autism to recognize emotions. Children with autism that had watched the series for 4 weeks significantly improved in their ability to detect emotions compared to those who did not watch the series. In another study, Faja et al. (2008) used a face training program to help autistic children improve their ability to recognize faces. After eight training sessions, the ability to detect faces improved. However, both studies did not make use of controllable stimuli. The results of our study suggest that presenting individuals with autism with a virtual character that can be made responsive and be controlled might also prove useful in the teaching of social skills (Wilms et al. 2010).

As for further clinical implications of our study, the results of the detection task show a tendency by the participants diagnosed with autism to classify averted gaze as direct. This tendency may be a source for misunderstandings in social interactions because some communicative actions, such as turn-taking, rely on accurately interpreting the gaze cues of the interaction partner. If individuals with autism mistake averted gaze for direct gaze, they may misinterpret that direct gaze for a turn-taking signal and interrupt their interaction partner. Because there is some evidence that turn-taking in autism may be abnormal (Baron-Cohen 1988), based on our results the role of gaze in turn-taking in autism needs to be further investigated.

Several limitations apply to the present study. First, the study focused only on individuals with high-functioning autism. Even though it might be concluded that non-high-functioning individuals with autism are also impaired in the detection of direct gaze, this has to be investigated in a separate study. Furthermore, it is not clear if non-high-functioning individuals with autism would be similarly able to use a task that gives them full control over a social stimulus to train their ability to detect direct gaze. Second, our study focused only on adults. As other authors have mentioned (Nation and Penny 2008), a developmental perspective is important for the understanding of autism to establish at which stage of development a certain deficit occurs. Third, the non-social, geometric stimulus was less visually complex than the virtual character. Therefore, a replication of the study should use a non-social stimulus that is more visually complex to better compare the performance on the social and non-social task.

Taken together, the current study investigated whether individuals with autism are impaired in their ability to detect direct gaze. In addition, it was tested whether individuals with autism can overcome such an impairment when given the control over the gaze of a naturalistic virtual character. Our results suggest that individuals with autism are impaired in their ability to distinguish between direct and averted gaze. This underlines the importance of using animate, realistic stimuli in experiments with individuals with autism, because the impairments specific to autism might not be fully detected using only static stimuli (Schilbach et al. 2011). Furthermore, when given the control over the gaze of a realistic virtual character, individuals with autism were as accurate as participants in the control group at establishing direct gaze, which suggests that giving individuals with autism control over a virtual character, or social stimuli in general, may be used to help them overcome some of their deficits in the processing of gaze.

Acknowledgments

This work was supported by a project grant of the German Ministry of Education and Research (BMBF “Social Gaze: Phenomenology and neurobiology of dysfunctions in high-functioning autism (HFA)”) to Gary Bente and Kai Vogeley.

Copyright information

© Springer Science+Business Media, LLC 2012