|Home | About | Journals | Submit | Contact Us | Français|
The present study investigates the accuracy and speed of face processing employed by high-functioning adults with autism spectrum disorders (ASDs). Two behavioral experiments measured sensitivity to distances between features and face recognition when performance depended on holistic versus featural information. Results suggest adults with ASD were less accurate, but responded as quickly as controls for both tasks. In contrast to previous findings with children, adults with ASD demonstrated a holistic advantage only when the eye region was tested. Both groups recognized large manipulations to second-order relations more accurately than no change or small changes, but controls responded more quickly than participants with ASD when recognizing these large manipulations to configural information.
Although symptom expression is heterogeneous among individuals with autism spectrum disorders (ASDs), evidence of peculiarities in tasks requiring recognition and memory of faces emerge when groups of individuals with ASD are compared with controls (see Jemel et al. 2006; Sasson 2006 for reviews). It is thought these oddities in face processing may relate to the core social difficulties experienced by individuals with ASD (e.g., Davies et al. 1994; Deruelle et al. 2004; Klin et al. 1999; Volkmar et al. 1989).
During typical development, specialized processing biases for faces emerge that are not observed for other visual stimuli. For example, faces are processed configurally (Maurer et al. 2002), which includes synthesizing the whole face rather than its isolated features (i.e., holistic processing), sensitivity to overall orientation of one feature to another (i.e., first-order relations) and knowledge of distances between features (i.e., second-order relations) (e.g., Tanaka and Farah 1993; Tanaka and Sengco 1997; Young et al. 1987). Viewing faces in an inverted orientation is more disruptive to recognition than the inversion of non-face stimuli (e.g., Yin 1969), probably because inversion disrupts configural processing (e.g., Diamond and Carey 1986; Young et al. 1987).
Impaired configural processing in autism has been indirectly supported by a weakened face inversion effect (Hobson et al. 1988; Langdell 1978; Rose et al. 2007; Teunisse and de Gelder 2003; but see Lahaie et al. 2006). Further, individuals with ASD are less impacted than controls when configural information is disrupted by a split face paradigm with stimulus detection after a brief delay (Teunisse and de Gelder 2003). Individuals with autism are also more reliant on individual facial features (Hobson et al. 1988) and on high spatial frequency information, which emphasizes local details over global configuration (Deruelle et al. 2004). Finally, based on a priming task it has been suggested that individuals with ASD have enhanced featural processing with preserved configural processing (Lahaie et al. 2006).
Despite the mounting evidence of unusual biases in configural processing by individuals with autism, there have been fewer direct measures of sensitivity to second-order relations or to the holistic processing advantage. The Thatcher illusion, thought to disrupt second-order relations, was detected as accurately by lower functioning children with ASD as controls (Rouse et al. 2004). In direct comparisons of holistic versus featural processing, high-functioning children with ASD performed similarly to controls across whole face and feature conditions (Lopez et al. 2008). Two research groups found evidence of holistic processing by high-functioning children and teens with ASD in a match to sample task with a short memory delay, but only under conditions of attentional cuing (Lopez et al. 2004) or when recognition relied on information contained in the mouth region (Joseph and Tanaka 2003). Holistic processing was not found in the absence of cues or when recognition relied on the eyes, respectively. An attentional bias toward the bottom part of faces by individuals with ASD has been suggested (Joseph and Tanaka 2003; Klin et al. 2002), but this finding has not been consistently demonstrated (e.g., Bar-Haim et al. 2006; Lahaie et al. 2006).
The goal of the current study is to extend the literature by directly measuring the accuracy and speed of these two sub-domains of configural processing in a group of adult participants. Most investigations have tested accuracy in developmental samples. By testing high-functioning adults with ASD, we aim to tap face processing abilities independent of cognitive confounds and with less risk of potential differences resulting from dissimilar rates of development. Experiment 1 tests for a recognition advantage when features are presented in the context of a whole face rather than alone (Joseph and Tanaka 2003; Tanaka and Farah 1993). Featural processors would perform similarly across conditions. The reduced holistic advantage with high-functioning adolescents (Lopez et al. 2004) supports the hypothesis that adults with ASD would exhibit a pattern of reduced holistic processing compared with controls. The facial inversion effect and attention to the top versus bottom of the face are also measured. Mixed findings of selective attention to the lower portion of the face have been reported, but we anticipate a processing advantage for the mouth region on the basis of findings by Joseph and Tanaka (2003). Experiment 2, adapted from Freire and Lee (2001), directly measures sensitivity to second-order relations by measuring the ability to detect manipulations of the distances between features. The direct manipulation of second-order relations used here is novel with this population, but previous findings with the Thatcher effect (Rouse et al. 2004) suggest this sub-domain may be spared.
Two groups of adults participated in the study: 39 individuals with ASD and 33 with typical development. Both groups were recruited using advertisements in the greater Seattle metro area and most participants did not have a university affiliation. Exclusionary criteria for the control group included family history of an ASD or the presence of major mental illness. Descriptive information on these participants is reported in Table 1. The autism diagnostic interview-revised (ADI-R), autism diagnostic observation schedule (ADOS), and diagnostic and statistical manual of mental disorders, fourth edition (DSM-IV) criteria were used to assess symptom severity and confirm existing clinical diagnoses for 18 individuals with Autistic disorder, 18 with Asperger’s syndrome, and 3 with Pervasive developmental disorder, not otherwise specified (American Psychiatric Association 1994; Lord et al. 2000; Lord et al. 1994). Groups were matched on the basis of cognitive ability, with all participants having, at minimum, full scale and performance IQ of 80, based on an abbreviated version (vocabulary, comprehension, object assembly, block design subtests) of the Wechsler Adult Intelligence Scale—third edition (WAIS-III; Wechsler 1997a). Groups significantly differed in their recognition of faces on the Wechsler Memory Scales immediate and delayed facial memory tasks (WMS; Wechsler 1997b) and in short term general memory for objects as assessed by the Woodcock Johnson revised picture recognition subtest (WJ-R; Woodcock and Johnson 1989). All procedures were approved by the university human subjects division.
All stimuli were black and white photos of male and female faces posing neutral expressions. Two versions of each task were created and counterbalanced across participants. Experiments were presented on a desktop computer and responses were collected on a button box. Before each experiment, participants were given practice items to ensure understanding. Care was taken not to reveal the experimental manipulations to participants.
Participants viewed a 16.5 by 15.3 cm target image (always a whole face) for 3.5 s followed by a 1 s delay. Then the target and foil were presented together as a forced choice for up to 8 s. To control for naturalness, both targets and foils were hybrid faces created by replacing features of each face. Test trials varied by orientation (upright or inverted), test type (whole faces or parts), and feature (eyes or mouth) for eight possible combinations of stimulus characteristics. Within a trial, orientation was constant. During the test phase, target-foil pairs were of the same test type and feature, and differed by a single feature. For isolated features, the eye or mouth region of the face was presented in a rectangular selection at the same size and screen location as the feature on the original target face. The experiment consisted of two blocks containing 40 unique, counterbalanced trials (10 of each type of stimulus) presented in pseudorandom order.
Targets (17.9 by 21.8 cm) were presented for 3.5 s, followed by a blank screen without masking for 1 s, and then a test face for up to 8 s. Participants indicated whether or not the test face exactly matched the target. Twelve Nim-Stim faces (Tottenham et al. 2002) were used to create test stimuli by moving both the eyes (in one of four directions: up, down, in, or out) and mouth (either up or down) and then blending the pixels surrounding the features. Faces varied by the magnitude of changes made (no change, 6, or 12 pixels). Ninty-six total trials were presented in pseudorandom order, including 24 small changes, 24 large changes and 48 targets in which no manipulations were made. See Fig. 1 for examples of the stimuli and test conditions for both experiments.
To investigate whether participants with ASD have a similar pattern of performance to controls (i.e., a holistic processing advantage, more attention to the eye region of the face, and impairment with inversion of faces), repeated measures ANOVA tested the following factors: group (ASD/control), orientation (upright/inverted), test type (whole/part), and feature (eyes/mouth). To explore whether groups were similarly sensitive to manipulations of distances between features, a repeated measures ANOVA tested factors of group and magnitude of change (no change/small/large). In both experiments, accuracy and reaction time were the dependent variables. For Experiment 2, accuracy was measured using percent correct due to the differing proportion of trials for each condition. Data from Experiment 2 were excluded due to computer malfunction (1 ASD, 1 control), behavior during the task (1 ASD) and missing all practice trials (4 ASD, 1 control), for a sample size of 33 individuals with ASD and 31 controls. All reported results for both tasks were computed using Greenhouse-Geisser corrected degrees of freedom to adjust for potential departures from sphericity and compound symmetry. Due to significant differences in initial WMS scores, the relation between performance on this standardized measure is compared with performance on experimental tasks. Finally, due to differences in general memory ability on the WJ-R, all analyses were recomputed with the score on this measure included as a covariate.
When accuracy was examined, there were significant main effects of group (F(1, 70) = 21.6, p<.001; MASD = 7.1, SD = 1.0; MControl = 8.0, SD = 0.7), orientation (F(1, 70) = 61.5, p<.001; Mupright = 8.0, SD = 1.2; Minverted = 7.1, SD = 1.0), and test type (F(1, 70) = 27.9, p<.001; Mwhole = 7.9, SD = 1.2; Mpart = 7.2, SD = 1.1), but the effect of feature was not significant. There was a significant interaction of group × test type × feature (F(1, 70) = 8.1, p<.01), while the group × orientation × test type interaction approached significance (F(1, 70) = 3.8, p = .05) and other interactions failed to reach significance. Scores are reported for these conditions in Table 2. When WJ-R scores were covaried, significant effects of group (F(1,66) = 13.0, p = .001) and an interaction between group × test type × feature (F(1, 66) = 8.0, p<0.01) remained. Overall percent correct significantly correlated with the immediate and delayed scores on the faces subtest of the WMS (rimmed = .48, p<.001 and rdelay = .33, p<.01). The relation between IQ and accuracy across the entire task did not reach significance, except for Performance IQ and overall accuracy, r = .28, p<.05. No relation between autism symptoms and accuracy was detected.
Reaction time was also examined. There was no effect of diagnostic group or of feature, but there were significant main effects of orientation (F(1, 70) = 34.0, p<.001; Mupright = 2232 ms, SD = 598; Minverted = 2419 ms, SD = 615) and test type (F(1, 70) = 17.1, p<.001; Mwhole = 2421 ms, SD = 686; Mpart = 2230 ms, SD = 550). Of interest were significant interactions between group × orientation (F(1, 70) = 4.69, p<.05), group × feature (F(1, 70) = 4.75, p<.05), group × orientation × test type (F(1, 70) = 4.07, p<.05), and group × orientation × feature (F(1, 70) = 5.61, p<.05). Co-varying WJ-R memory resulted in a significant group × feature interaction (F(1, 66) = 6.7, p = .01) and trends for group × orientation (F(1, 66) = 3.7, p<.1) and group × orientation × feature (F(1, 66) = 2.93, p<.1) (Table 3).
Repeated measures with percent correct as the dependent variable revealed a significant main effect of group (F(1, 62) = 28.3, p<.001; MASD = 63%, SD = 11; MControl = 78%, SD = 10) and of magnitude of change (F(1.55, 96.10) = 41.0, p<.001; Mlg = 83%, SD = 18; Mno = 72%, SD = 17; Msm = 56%, SD = 21). However, there was no interaction between these variables. In order to test whether group differences remained when possible response biases were taken into account, d′ was calculated. Significant group differences remained and controls had a greater sensitivity to changes than the group with ASD (t(62) = −3.26, p<.01; MASD = 0.94, SD = 1.2; MControl = 1.99, SD = 1.4). Including WJ-R memory as a covariate resulted in a significant effect of group (F(1, 58) = 20.7, p<.001) and a trend effect of magnitude (F(1.4, 83.6) = 2.9, p<0.1). Correlation analysis found significant relations between the total number correct and performance on the Wechsler scales of immediate and delayed face memory (r = .47, p<.001 and r = .44, p<.001, respectively). The relation between IQ and total accuracy did not reach significance. Among the participants with ASD, increased severity of communication impairments on the ADI-R related to decreased accuracy overall (r = −.35, p<.05). There was not a significant relationship with overall performance and the social or repetitive behavior domains of the ADI-R or with ADOS scores.
When reaction time was examined, there was not a main effect of group but was a significant main effect of magnitude of change (F(1.4, 86.9) = 33.8, p<.001; Mlg = 1295 ms, SD = 492; Mno = 1550 ms, SD = 555; Msm = 1564 ms, SD = 570). The group × magnitude interaction was also significant (F(1.4, 86.9) = 8.5, p<0.01), as shown in Fig. 2. When memory ability was covaried, the group × magnitude interaction remained significant (F(1.5, 84.1) = 4.2, p<.05).
The current study was designed to examine the specialized face processing abilities of high-functioning adults with ASD, as compared with control adults, through accuracy and reaction time patterns. Experiment 1 revealed that participants with ASD were on the whole less accurate, even when general memory ability was included as a covariate. Consistent with our prediction of reduced holistic processing in ASD, a significantly different pattern of accuracy by group emerged, wherein controls were more accurate in the whole face condition than the part condition, and individuals with ASD were more accurate in the whole face condition than the part condition only when the eye region was tested. These results differ from Joseph and Tanaka (2003) who reported holistic processing when the mouth was tested. Taken together, they may suggest that holistic processing may be available to individuals with autism, but only with focused attention.
Two interactions were of interest when reaction time was examined. First, the group with ASD responded more quickly when mouths were tested, while the controls were more rapid in the eye condition. This is seemingly in contrast to an advantage with accuracy in the whole-eye condition. We speculate that adult participants with ASD may be more efficient with recognition of the mouth region due to evidence of early attentional bias for this region with children, but with experience and direct intervention with social skills, they may have learned to redirect their attention to the top of the face as a (less automatic) compensatory strategy. Second, the difference in reaction time for upright versus inverted faces was twice as large for controls compared with the group with ASD. This finding is consistent with other evidence of a weakened inversion effect in ASD.
Experiment 2 also revealed reduced accuracy across conditions by the group with ASD, even when covarying general (non-face) memory ability and accounting for possible response bias. Accuracy significantly differed depending on the magnitude of the change made to second order relations of test faces for both groups. This finding is consistent with findings by Rouse et al. (2004) who demonstrated lower functioning children with ASD were sensitive to manipulations of second-order relations. However, the group with ASD differed from controls in its pattern of response speed across conditions, even with memory as a covariate. While controls quickly recognized large manipulations, relative to small or no change conditions, response times were similar across these conditions for the group with ASD. Taken together, this finding suggests that while individuals with ASD can detect large changes to configural information, they are slower to do so. Last, a relation between communication impairments and accuracy emerged. We suspect that participants with ASD who struggled with flexible use of language may have been overly literal in interpreting instructions to decide whether faces were “exact matches.”
Important limitations must be mentioned. First, there were significant group differences on standardized measures of face and object memory. Face memory was significantly related to overall accuracy for both experiments; this suggests that both face processing abilities and generalized memory ability may contribute to performance. In order to address this possibility, analyses were conducted with object memory as a covariate; group differences in overall accuracy as well as differences in the within group patterns of performance remained for both tasks. Second, because the aim of the current study was to test the specific aspects of configural processing that are argued to be unique to faces, we did not include an object condition in our experimental tasks. Examining configural processing for objects would be useful in addressing generalized cognitive differences that may contribute to performance. However, evidence for face specific impairment rather than more generalized cognitive impairments (i.e., processing speed, attention, memory, central coherence) in ASD is taken from a lack of group differences in overall reaction time between groups and significant within group differences on both tasks. A third limitation is lack of masking in Experiment 2. This may have allowed participants to overlay a mental image of the target onto the test image, a strategy that may have benefitted the control group. Lastly, this as well as the majority of face processing investigations has used unfamiliar faces, and recent work suggests processing may be less disrupted with familiar faces (e.g., Wilson et al. 2007).
Many real world face-to-face encounters requiring recognition of individuals also require simultaneous attention to linguistic and emotional input and generation of appropriate responses. The observed differences in speed and accuracy of configural processing under experimental conditions may translate to problems for individuals with ASD in social interactions including difficulty dividing attention, missed information, confusion, and anxiety. In the future it will be important to determine the extent to which the pattern of accuracy and timing on laboratory tasks relates to brain functioning (e.g., via ERP) and to ability in naturalistic social situations.
This research was supported by NIMH U54MH066399. The development of the MacBrain Face Stimulus Set was overseen by Nim Tottenham (tott0006/at/tc.umn.edu) and supported by the John D. and Catherine T. MacArthur Foundation Research Network on Early Experience and Brain Development. Special thanks are given to the individuals who participated and to the clinical core led by Dr. Jessica Greenson who conducted diagnostic testing. A poster with preliminary data from this study was presented at the International Meeting for Autism Research, Boston, MA, in May, 2005.
Susan Faja, Department of Psychology, Center on Human Development and Disability, University of Washington Autism Center, Seattle, WA, USA.
Sara Jane Webb, Department of Psychiatry and Behavioral Sciences, Center on Human Development and Disability, University of Washington Autism Center, Box 357920, 98195 Seattle, WA, USA.
Kristen Merkle, Center on Human Development and Disability, University of Washington Autism Center, Seattle, WA, USA.
Elizabeth Aylward, Department of Radiology, Center on Human Development and Disability, University of Washington Autism Center, Seattle, WA, USA.
Geraldine Dawson, Department of Psychology, Center on Human Development and Disability, University of Washington Autism Center, Seattle, WA, USA. Department of Psychiatry and Behavioral Sciences, Center on Human Development and Disability, University of Washington Autism Center, Box 357920, 98195 Seattle, WA, USA.