Journal of Autism and Developmental Disorders, Vol. 29, No. 6, 1999
A Normed Study of Face Recognition in Autism and Related Disorders Ami Klin,1,2 Sara S. Sparrow,1 Annelies de Bildt,1 Domenic V. Cicchetti,1 Donald J. Cohen,1 and Fred R. Volkmar1
Although the interpretation of studies of face recognition in older children, adolescents, and adults with autism is complicated by the fact that participating samples and adopted methodologies vary significantly, there is nevertheless strong evidence indicating processing peculiarities even when task performance is not deficient. Much less is known about face recognition abilities in younger children with autism. This study employed a well-normed task of face recognition to measure this ability in 102 young children with autism, pervasive developmental disorder not otherwise specified (PDDNOS), and non-PDD disorders (mental retardation and language disorders) matched on chronological age and nonverbal mental age, and in a subsample of 51 children divided equally in the same three groups matched on chronological age and verbal mental age. There were pronounced deficits of face recognition in the autistic group relative to the other nonverbally matched and verbally matched groups. Performance on two comparison tasks did not reveal significant differences when verbal ability was adequately controlled. We concluded that young children with autism have face recognition deficits that cannot be attributed to overall cognitive abilities or task demands. In contrast to controls, there was a lower correlation between performance on face recognition and nonverbal intelligence, suggesting that in autism face recognition is less correlated with general cognitive capacity. Contrary to our expectation, children with PDDNOS did not show face recognition deficits. KEY WORDS: Autism; PDD; face perception.
INTRODUCTION
ity to recognize others as familiar or unfamiliar, male or female, child or adult. That a mechanism for facial processing is likely to be one of the earliest facilitators of social engagement is suggested by findings such as the preference shown by 2-day-old infants to look at their mother rather than at another unknown woman (Bushnell, Sai, & Mullen, 1989). The attractiveness of facial processing as an experimental avenue to research social deficits in autism is further strengthened with the expanding knowledge of the neuroanatomy and neurofunctional brain systems involved in the recognition of faces. Neuropathological studies of brain-damaged humans (e.g., Damasio, Tranel, & Damasio, 1990) and recordings of the activity of single neurons in nonhuman primates (Heywood & Cowey, 1992) have shown that there are circumscribed regions of the brain that play an important role in recognition of faces. For
Among the various psychological processes underlying the social disability evidenced in autism, the perception and memory for facial identity has been given special attention in experimental studies (Volkmar, Grossman, Klin, & Carter, 1997). The thrust for this line of research has been the central role attributed to facial processing in normative socialization, from very early infancy (Bryant, 1991) and throughout the life-span (Ellis, 1990). Most forms of social interaction and intersubjectivity are predicated on the person's abil1 2
Yale Child Study Center, New Haven, Connecticut. Address all correspondence to Ami Klin, Yale Child Study Center, 230 South Frontage Road, New Haven Connecticut 06520; e-mail:
[email protected]
499 0162-3257/99/1200-0499$16.00/0 6 1999 Plenum Publishing Corporation
500 example, studies involving electrical recording (e.g., Ojemann, Ojemann, & Lettich, 1992) and functional magnetic resonance imaging (e.g., Puce, Allison, Gore, & McCarthy, 1995) implicated the inferior temporal lobe regions, particularly the fusiform gyrus anterior to the secondary association cortex of the occipital lobe. Such findings provide appealing models that await testing in individuals with autism and related disorders. The pioneering studies of face recognition in autism were conducted by Langdell (1978), who studied the ability to recognize faces of peers exhibited by older adolescent and young adults with autism. While they could adeptly identity their peers, they showed a number of peculiarities: For example, in comparison to controls matched on chronological age (CA) and performance IQ, the adolescent participants with autism appeared to rely on the lower parts of the faces (i.e., the mouth region), rather than the upper parts of the faces (i.e., the eye region), a more normative strategy (McKelvie, 1976); they also failed to show the typical decrement in performance when asked to identify inverted (i.e., upside-down) faces (Valentine, 1988). No significant correlation was found between performance IQ (derived from the WISC) and performance on the experimental tasks. Similar results were obtained in a study of face recognition in adolescents and young adults with autism conducted by Hobson and colleagues (Hobson, Ouston, & Lee, 1988). Autistic participants were at least as proficient as controls matched on verbal mental age (VMA); however, their performance was in fact superior to that of controls when asked to identify upsidedown faces. Taking both of these studies together, it could be concluded that mentally retarded autistic adolescents are proficient in judging faces by identity, in regard to both familiar (Langdell, 1978) and unfamiliar (Hobson et al, 1988) faces. An additional study involving mentally retarded adolescent individuals with autism failed to find deficits in face recognition. Volkmar, Sparrow, Rende, and Cohen (1989) used a task in which participants had to assemble a series of puzzles displaying photographs of human faces varying in terms of complexity, familiarity, and configuration (normal vs. scrambled faces). No differences were found between the participants with autism and CA- and nonverbal MA-matched controls, and the correlation between IQ and performance on the various tasks was not significant. However, the participants were older and relatively low-functioning and the nature of the task (i.e., one analogous to object assembly) may have minimized the "ecological validity" of the task. In contrast to this study, Tantam, Monaghan, Nicholson, and Stirling (1989) compared mentally re-
Klin et al. tarded autistic children with controls matched on CA and nonverbal MA and found that the autistic participants performed less well on a test of finding the odd face out from a set of photographs of faces. No significant correlation was found between performance on various tasks and nonverbal IQ or a verbal IQ approximation. Boucher and Lewis (1992) assessed mentally retarded autistic children's face recognition relative to both a nonverbally and a verbally matched sample of controls. They found that the autistic group's face recognition for unfamiliar faces was impaired relative to both control groups; interestingly, a control task involving the recognition of buildings revealed no group differences. More recently, two studies failed to demonstrate face recognition deficits in autism. (Davies, Bishop, Manstead, and Tantam (1994) reported on a study in which mentally retarded individuals with autism and higher functioning participants, most of whom carried a diagnosis of Asperger syndrome, performed a matching task involving both faces and an array of geometric forms. There were no significant differences between the lower functioning participants with autism and a verbally matched sample of controls; the higher functioning participants performed significantly worse than controls across all tests, suggesting a more pervasive deficit in visuospatial processing not specific to the perception of faces. Finally, Celani, Battacchi, and Arcidiacono (1999) tested 10 individuals with autism, 10 with Down syndrome, and 10 normal controls on a emotional perception battery that included a face recognition task. There was no significance difference across the three groups on the recognition task. In summary, the available evidence for face recognition deficits in autism is somewhat conflictual. While some studies have found such deficits, others have not. What is of interest, however, is that even in those studies in which participants with autism performed as well as controls, there was at times evidence for peculiarities in face processing that were not immediately apparent when only overall performance results were analyzed (e.g., the lack of performance decrement when processing inverted faces, and the preferential attention to the lower parts of the face rather than the eyes). However, this interpretation is complicated by the fact that different studies utilized different methodologies, which at times involved potential confounds (e.g., the use of an "object assembly" procedure); together with the participation of heterogeneous groups, such confounds make comparability of findings more difficult. Also of interest, most studies involved older participants (adolescents and adults), raising the possibility that younger participants might have face recognition deficits that are
Face Recognition in Autism not evident later on, although peculiarities in face processing may remain. In this regard, the fact that several studies failed to find a significant correlation between performance on face recognition tasks and IQ for autistic participants whereas a significant correlation was obtained for participants with other developmental disabilities (e.g., in Boucher & Lewis, 1992) suggests the possibility that face perception skills are a dissociated cognitive function unrelated to overall cognitive level for a given person with autism. The Present Study All studies of face recognition in autism to date have employed experimental tasks especially created for the given study. By using rigorously controlled conditions, experimental paradigms allow for detailed hypothesis testing, delineation of specific profiles of intact and deficient abilities, and isolation of discrete processes which can be eventually mapped onto specific brain systems. However, in the field of face recognition research in autism, experimental methods have rarely been duplicated by investigators other than the group reporting the finding; this has resulted in some uncertainty as to what extent face recognition deficits are evidenced in autism. Also, studies of face recognition in autism have involved primarily older individuals, leaving unresearched the face recognition abilities of younger children with autism. To address these issues, the present study utilized a widely available, normed test of face recognition which is part of a battery of intellectual assessment thought to be of great utility in psychological evaluations of young children with autism (Klin, Carter, & Sparrow, 1997). Our aims were to document the presence or absence of face recognition deficits in autism (a) using methods that can be easily replicated by other investigators, and (b) including a younger sample of children with autism. To achieve these aims, we studied the performance of young children with autism relative to controls matched on nonverbal and verbal mental age on a well standardized test of face recognition included in the Kaufman Assessment Battery for Children (K-ABC; Kaufman & Kaufman, 1983). The control samples included one group of children with milder or atypical forms of autism—pervasive developmental disorder not otherwise specified (PDDNOS)—and children with a developmental disorder not included within the PDD class (or non-PDD disorders); the latter included primarily children with mental retardation and language disorders. The rationale for the inclusion of the PDDNOS group
501 was to explore whether or not any deficits found in face recognition applied to a wider range of children with autistic disorders. Four specific predictions were made: (a) children with autism would show deficits in face recognition relative to controls, as well as relative to their own overall levels of cognitive functioning; (b) deficits on the face recognition task would not be a function of difficulties in complying with general task demands (e.g., providing relevant responses, attending to stimuli, withholding response until target stimuli were presented) or of confounding abilities (e.g., nonface visual memory). To test this hypothesis, results on two other subtests of the K-ABC—Gestalt Closure and Spatial Memorywere analyzed across the three samples; (c) as the diagnosis of PDDNOS is usually assigned to children who may show a milder or atypical form of the social dysfunction observed in autistic disorder (Volkmar & Klin, 1999), we predicted that the performance of children with PDD NOS would be superior to that of the autistic group but inferior to that of children with a non-PDD disorder. We also tested directly whether face recognition deficits were associated with levels of socialization skills by examining the correlation between performance on the face recognition task and scores obtained on the Socialization domain of the Vineland Adaptive Behavior Scales (Sparrow, Balla, & Cicchetti, 1984a, 1984b; see below). METHOD Participants Of a pool of 512 children with developmental disabilities seen in our Center for the past 6 years, 252 children completed at least some sections of the K-ABC (Kaufman & Kaufman, 1983), minimally the Mental Processing subtests corresponding to the overall Nonverbal Standard Score or Prorated Nonverbal Score (Kaufman & Kaufman, 1983) including the Face Recognition subtest and the two comparison tasks (see below), and the achievement subtests corresponding to the Verbal Intelligence Standard Score or Prorated Verbal Intelligence Score(Kamphaus & Reynolds, 1987). These 252 children were seen for comprehensive clinical evaluations including psychiatric, psychological, and speech, language, and communication assessments following a developmentally based, transdiciplinary model (Klin, Carter, Volkmar, et al, 1997). From this initial sample of 252 children, 25 children were excluded because they failed to reach a basal level of performance (at least 2 successes, corresponding to an age equivalent score of 2 years 6 months) on the Face
Klin et al
502 Recognition task; 4 additional children were excluded because they did not reach a basal score on the Spatial Memory task; none were excluded on the basis of failing to reach a basal score on the Gestalt Closure task only. From the resulting sample of 223 children, we were able to form three groups of 34 children in each group matched on chronological age (CA) and nonverbal mental age (NVMA) derived from their K-ABC scores: The three groups corresponded to children with autism, PDDNOS, and non-PDD disorders (children whose primary clinical diagnosis were mental retardation or a language disorder). Matching the basis of NVMA was performed given that the Face Recognition subtest of the K-ABC is part of the Nonverbal Scale (i.e., no verbal element is required in either administration/ instruction or performance/response). Scores on the Face Recognition subtest were not included in the computation of NVMA as we intended to measure the correlation between children's face recognition skills and their overall cognitive level. These 102 children were the main sample of this study. Table I summarizes the process of selection of the three NVMA-matched samples from the initial available sample. CA, NVMA (from the K-ABC) as well Verbal Mental Age (VMA) as derived from the Verbal Intelligence Standard score (from the K-ABC) (Kamphaus & Reynolds, 1987), and Adaptive Behavior Composites age equivalents (from the Vineland) are given in Table II. The three groups did not differ significantly in CA and NVMA, but did differ significantly in VMA and Vineland Socialization Age Equivalent. Diagnostic assignments were performed by two experienced clini-
Table I. Process of Selection of the Three NVMA-Matched Samples from the Original Available Sample Original available sample of children with developmental disabilities Of 512, excluded because of incomplete K-ABCs Of 260, excluded because child did not reach a basal level of performance on the K-ABC' s Face Recognition task Of 260, excluded because child did not reach a basal level of performance on the K- ABC's Spatial Memory task Of remaining 223 children, included in final sample of matched groups a
512
260" 25
4
102
These 260 children had generally a nonverbal cognitive level that made the K-ABC too challenging or inviable. Only a small minority of these 260 children (n = 32) could reach a basal level of performance on the Face Recognition task. These were excluded, however, as they failed to complete the other procedures required for inclusion in the study.
cians (A. K. and F. R. V.) following ICD-10 (WHO, 1993) research diagnostic criteria for autism, PDDNOS (atypical autism), mental retardation, and language disorders. Interrater reliability as measured by the chancecorrected coefficient kappa (K) and proportion of observed agreement (PO) (Siegel & Castellan, 1988) was excellent (K = .799; SE = .095; PO = .90) (Cicchetti & Sparrow, 1981). The reliability measures were obtained for a portion of this sample as part of a different project (Volkmar et al., 1994). The K-ABC Face Recognition Task and Two Comparison Tasks The Kaufman Assessment Battery for Children (K-ABC., Kaufman & Kaufman, 1983) is an individually administered measure of intelligence and achievement standardized on a large and representative nationwide sample of children. Its psychometric properties and levels of reliability and validity are excellent (Kaufman & Kaufman, 1983). The K-ABC consists of primarily two sections: a Mental Processing section, which refers to problem-solving skills in novel situations, and an Achievement section, which refers to the child's ability to extract and assimilate information from their cultural and school environment. The Mental Processing section is divided into the Sequential Processing and Simultaneous Processing Scale, representing two types of mental functioning and cognitive style. Sequential processing refers to serial or temporal processing in problem solving, whereas Simultaneous processing refers to problem solving based on gestalt-like, frequently spatial processing leading to integration of information. The K-ABC also includes a Nonverbal Scale, composed of selected K-ABC subtests which can be administered in pantomime and responded to motorically (Kaufman & Kaufman, 1983). The Face Recognition (FR) subtest of the K-ABC measures the child's ability to attend closely to one or two faces whose photographs are exposed briefly, and then to select the correct face(s) shown in a different pose and/or emotional expression, from a group photograph. (Kaufman & Kaufman, 1983, pp. 37-39). The FR subtest is part of the Mental Processing section and the Nonverbal Scale. FR was intended as an adaptation of experimental tasks with proven neuropsychological value (e.g., Sergent & Bindra, 1981) to the context of an individually administered intelligence test. The FR subtest is also part of the Simultaneous Processing Scale. It is important to note that FR was excluded from the school-age portion of the K-ABC, even though the task was of sufficient difficulty for older children. The
Face Recognition in Autism
503 Table II. Sample Characteristics and ANOVAs a Autisma
Variable*
CA NVMA
VMA V-Soc a b
Non-PDD disordersa
PDDNOSa
M
SD
M
SD
M
SD
F(2, 99)
P
6 years; we adopted the convention of assigning the age equivalent score of 6 years 5 months for this score range. Although only 3 children achieved a perfect score of 15 (1 with PDDNOS and 2 with a non-PDD disorder), 12 others received scores 13 and 14 (5 with PDDNOS and 7 with nonPDD disorders). Although we could not circumvent this ceiling effect, our convention was intended to force a conservative comparison among the groups. In other words, this rule worked against our hypothesis predicting deficits in face recognition in the autistic sample given that none of the children with autism scored at the ceiling level. We did, however, analyze Face Recognition data in terms of both age equivalent scores and raw scores. Scaled scores were not analyzed given that a number of children across the three groups
RESULTS Performance on the Face Recognition Task and on the Comparison Tasks for Groups Matched on Nonverbal Mental Age The age-equivalent scores on the Face Recognition task (FR) and on the Gestalt Closure and Spatial Memory tasks for the three groups matched on CA and NVMA are given in Table III. Gestalt Closure and Spatial Memory results were analyzed only in terms of ageequivalent scores as none of the children scored at the ceiling level (12 years and 6 months): Performance on FR was significantly different among the three groups in terms of both age-equivalent and raw scores. Tukey post-hoc comparisons revealed significant differences between the autistic and both the PDD and non-PDD groups (p < .0001) but not between the PDD and the non-PDD groups. Therefore, the hypothesis that young children with autism have face recognition deficits was corroborated. However, although no group differences were found for the Spatial Memory task, differences reaching statistical significance were found for the Gestalt Closure task. Tukey post-hoc comparisons suggested that at least for the nonverbally matched groups, visual spatial reconstruction of meaningful (nonsocial) figures was slightly impaired in the autistic group relative to the PDD group (but not relatively to the nonPDD group). Of interest, however, the hypothesis that children with PDD would have scores in between the autistic and the non-PDD group was not corroborated, as group mean scores for the PDD and the non-PDD groups were very similar.
Table III. Face Recognition Age Equivalents (FR AE) and Raw Scores (FR RS), Gestalt Closure Age Equivalents (GC AE), and Spatial Memory Age Equivalents (SM AE) for the Three Groups Matched on NVMA a
Autism a
FRAE FRRS GC AE SM AE a b
PDDNOSa
Non-PDD disordersa
M
SD
M
SD
M
SD
F(2, 99)
3.06 4.00 4.63 5.03
0.85 2.04 2.34 1.29
5.30 9.50 5.68 4.86
1.53 3.38 1.95 0.82
5.28 9.64 5.01 4.73
1.51 3.35 1.65 1.00
33.41 39.25 2.85 0.72
n = 34 for each group. Close to p < .05.
P
