Drawing Firmer Conclusions: Autistic Children Show No Evidence of a Local Processing Bias in a Controlled Copying Task

Drawing is a common everyday activity that is unique to humans (La Femina et al. 2009). Graphic production is particularly interesting because it requires synthesis of a number of component functions, including visual perception, mental imagery, memory, attention and action (see Smith 2009). Accordingly, drawing is commonly used as a laboratory tool to study a variety of behavioural features, from the development of reasoning abilities in childhood (e.g., Preissler and Bloom 2008) to the decline of motor coordination in old age (e.g., Morgan et al. 1994). It is for these reasons that drawing production has proved to be an important facet of autism research.

Cognitive and perceptual processing has long been known to be unusual in autistic individuals (Frith 1989). Performance on visuospatial tasks appears to illustrate a particular profile of strengths and weaknesses, with participants on the autism spectrum sometimes performing at a level superior to that of typical individuals of similar age and ability (for review, see Simmons et al. 2009). These strengths have been demonstrated in tasks such as pattern discrimination (Plaisted et al. 1998), block design (Shah and Frith 1993) and detecting embedded figures (Jolliffe and Baron-Cohen 1997; Pellicano et al. 2006).

It is likely that interest in assessing drawing behaviour in autistic people was initially sparked by observations of artistic savant abilities in some individuals (e.g., Mottron and Belleville 1995; Selfe 1977). Such cases are rare, although group-based studies of drawing behaviour in autism have also reported superior performance relative to typical individuals. For example, Mottron et al. (1999) conducted a copying task, where autistic adults and adolescents were required to reproduce impossible figures (i.e., three-dimensional designs where the global form does not conform to geometric rules, but the local parts do). Both groups copied possible figures with equal speed and accuracy, but autistic individuals copied impossible figures significantly faster than typical individuals. Another study by Sheppard et al. (2007) required autistic children and adolescents to copy meaningful and non-meaningful figures (constructed using the same line components) that were either two-dimensional or three-dimensional. Both autistic and typical groups produced more accurate copies of meaningful figures, compared to non-meaningful, and also of two-dimensional, compared to three-dimensional stimuli. Importantly, however, autistic children were less affected by dimensionality than typical children, and more accurately depicted projection and perspective for three-dimensional figures. This is in line with another study (Sheppard et al. 2009a), which found that autistic participants were less affected by dimensionality than typical participants when copying line-drawn figures, but not real objects.

Explanations for autistic individuals’ greater accuracy in copying tasks have focused on a potential bias towards processing local perceptual features. The weak central coherence (WCC) account (Frith 1989; Frith and Happé 1994; Happé and Frith 2006) is based on the behavioural tendency to focus on details, at the expense of a coherent global gestalt. This propensity is held to occur at the perceptual level, but might also be manifest in more cognitive behaviours, such as extracting gist in language. In contrast, the Enhanced Perceptual Functioning (EPF) account (Mottron and Burack 2001; Mottron et al. 2006; see also Wang et al. 2007) specifically posits that superiorities in information processing extend beyond the realm of local processing to include basic perceptual functions, such as detection, recognition and discrimination, as a result of enhancements in bottom-up, feed-forward perceptual operations. Although both accounts predict a focus on local perceptual features, WCC proposes that superiorities in local processing go hand-in-hand with difficulties in global processing, whereas EPF posits that autistic perception defaults to the local level, with global integration being intact but not mandatory (for a fuller discussion of EPF see Wang et al. 2007).

In response to these divergent predictions, a number of studies have directly tested the veracity of the theories by using drawing tasks. Sheppard et al. (2009b) studied drawing strategies to assess whether adolescents with autism organised their copied renditions according to an appreciation of the whole figure, or on a part-by-part basis. Autistic participants were no more likely to use a local strategy than those without autism and, since there was no group difference at the global level (also see Jolliffe and Baron-Cohen 2001; Ropar and Mitchell 2001), the authors argued that these data provided evidence for the EPF account. In contrast, other studies have reported clear difficulties in producing the global properties of drawings, which lie in favour of the WCC account. For example, in their copying task, Booth et al. (2003) found that autistic children were more likely to begin their renditions with a local detail, compared to both typical children and children diagnosed with Attention Deficit Hyperactivity Disorder. Autistic children were also more likely to draw in a piecemeal fashion, yielding a final global configuration that was incorrect. Similarly, Drake and Winner’s (2011) autistic savant artist (a 10-year-old boy of average intelligence) used a local strategy to complete drawings of three-dimensional objects to a greater extent than a comparison group of children with autism but without a drawing talent. He also performed particularly well on tasks of local perception (embedded figures, block design) and poorly on a global task (classification of impossible figures).

It is therefore clear that evidence from existing studies fails to yield consistent support for either account of visuospatial processing in autism. Furthermore, some studies report no group differences in drawing speed or accuracy (e.g., Eames and Cox 1994; Mottron et al. 1999) and others report poorer performance in autistic groups (e.g., Booth et al. 2003). There are a number of potential reasons for these differences. First, autism is a condition that involves substantial heterogeneity. As a result, neither the genetic (e.g. Ronald et al. 2006) nor the behavioural and cognitive profile (e.g. Pellicano 2010) is likely to be universal. Second, different studies have selected comparison groups according to different principles. So, for example, whereas Mottron et al. (1999) matched their groups according to chronological age and non-verbal mental age, Sheppard et al. (2009a, b) also matched theirs according to verbal mental age. The ability to accurately draw real objects has been related to vocabulary skills, especially in younger children (e.g., Toomela 2002), and so differences in the verbal skills of autistic participants might impact upon the behaviours measured. Third, and most importantly, the nature of the drawing tests themselves have likely given rise to these differences in performance. Some studies have used two-dimensional figures whereas others have used three-dimensional figures. The inclusion of dimensionality is a relatively advanced component of drawing, especially for children (Willats 1977), which may mask more subtle indices or variation. Equally, some studies have used meaningful stimuli whereas others have used more abstract ones. When meaning is not controlled for, it may introduce additional conceptual components that affect the planning and production of the drawing (van Sommers 1984). Furthermore, not all studies include blind rating of participant drawings. Preconceptions of group differences can affect the assessment of drawing output, irrespective of the measures used (Leek et al. 2000). Finally, studies have tended to employ dependent measures that conflate global and local processing, making it difficult to dissociate their impact. For example, beginning drawings at the local level (e.g., Booth et al. 2003; Drake and Winner 2011) could either reflect enhanced local processing or poor global processing.

We suggest that some of these problems can be overcome by adopting methods and approaches used to study similar behaviours in neuropsychological patients (i.e., individuals who have sustained neurological damage). Graphic production tasks are ubiquitous in patient assessment and there has been much interest in devising objective quantitative measures of drawing that can inform the delineation of fine-grained distinctions between both functional components of behaviour and diagnostic categories themselves (Leek et al. 2000; Smith 2009). In the present study, we therefore used a drawing task that was originally designed to assess the influence of grouping cues on drawing impairments in constructional apraxia, an acquired dysfunction of the ability to construct a coherent global form from local elements (in the absence of a perceptual or motor deficit). This is most likely to be observed in tasks such as drawing or arranging blocks and is associated with damage to parietal and frontal cortices (see Guérin et al. 1999). Smith and Gilchrist (2005a) presented two adult patients with figures constructed of four L-shaped corner elements, arranged to form a square. Grouping was manipulated by skewing two or more of the part elements, disrupting good continuation. Patient copies were scored according to their reproduction at both the local level (i.e., the presence or absence of local elements) and the scalar properties of the global form (i.e., internal angular accuracy, height-width ratio). Although the patients performed similarly on other qualitative measures of drawing production, measurements revealed an important dissociation between them: patient ECR tended to complete across gaps between elements to form a global square form, whereas patient RA represented individual part elements but with a poor global configuration (akin to some of the autistic copies illustrated by Booth et al. 2003). This study provided evidence for a finer-grained distinction within the broad category of constructional apraxia and also showed that local and global processing can be dissociated in the visuomotor domain.

In the present study, we used Smith and Gilchrist’s (2005a) methodology to assess drawing production in children on the autism spectrum. The benefits of using this task were that it: (1) provided a fine-grained parametric approach that is specifically focused on the impact of grouping on drawing production, enabling study of the process in the absence of potentially confounding factors such as meaning or linear perspective; (2) allowed us to dissociate global–local processing by yielding several variables of interest, tapping either local processing or global processing, but not both; and, (3) enabled blind and detailed objective assessment of the drawings. Copies were compared between a Grouped condition, where gestalt properties of good continuation favoured a strong whole, and a Skewed condition, where those cues were reduced by rotating some or all of the local elements. Critically, the vertices of the elements always formed the same square configuration, and we measured the deviation from this global square. If autistic children demonstrate a reduced appreciation of the gestalt properties of the image, as predicted by WCC, then we would expect the accuracy of their copies to be less affected by the reduction of grouping cues than that of typical children of similar age and ability. Insensitivity to grouping cues in children with autism would also be manifest in reduced scalar accuracy of the global configuration compared to typical children. If, however, autistic children exhibit global accuracy equal to that of typical children, and are similarly affected by the reduction of global cues, then this would provide evidence for the EPF account. Since both theories predict strength in the processing of local information, one should expect autistic children to perform more accurately than typical children in the production of part elements (i.e., fewer omissions of line segments).

In addition to the drawing task, we also investigated individual differences in drawing ability by examining the relationship between our key drawing indices (accuracy of global and local reproduction) and performance on measures tapping processes thought to underpin drawing ability (e.g., local processing and planning ability). Local processing was measured using the Children’s Embedded Figures Test (Witkin et al. 1971). Autistic children have repeatedly been shown to outperform typical children on this test (e.g. Pellicano et al. 2006; although see White and Saldaña 2011) and so it was of particular interest to assess whether any perceptual strength in the detection of local form would be reflected in visuomotor performance on our drawing task. Planning ability was measured using the Tower of London task (based on Shallice 1982). Harris and Leevers (2000) argue that poor planning abilities are responsible for some drawing difficulties in autism. However, Booth et al. (2003) found that planning abilities were unrelated to drawing performance in their tasks, despite the fact that such planning was poorer in their autistic sample. We therefore examined whether planning ability might be related to the more subtle metric indices of drawing that we were measuring.

Examples of children’s copies are illustrated in Fig. 2. These were measured to produce global accuracy indices of internal angular error and height/width error (of the square formed by the four vertices). Owing to the sensitivity of these measurements, inter-rater reliability was assessed: The intraclass correlation for double-coded measurements indicated excellent inter-rater reliability, r _ic(255) = 0.99, p < .001. Local accuracy measures were categorically coded on the basis of the presence or absence of features, their relationship with each other, and their orientation on the page. Analyses were performed using SPSS version 22 software (IBM International).

We assessed whether autistic children’s drawing production was affected by the global qualities of the to-be-copied model. Children on the autism spectrum were just as likely as typical children to be affected by manipulations of gestalt properties. When some or all of the corner elements were skewed, thus reducing the good continuation of the global square, both autistic and typical children produced copies that had greater error in the scalar properties of the global configuration (i.e., greater internal angular error and greater deviation of height/width ratio from 1:1). Across all analyses, there were no significant effects of group or any group × condition interactions. Furthermore, neither group demonstrated a trade-off between global and local performance—local accuracy was, in fact, related to less global error for autistic and typical children alike.

Overall, this pattern of results suggests that autistic children are sensitive to manipulations of global structure. When grouping cues were manipulated, the global qualities of their copies were reliably affected, and to a similar extent as their typically developing counterparts. These findings are in accordance with previous research that has reported no group difference in drawing performance at the global level (e.g., Mottron et al. 1999; Sheppard et al. 2007). Taken together, these findings speak against the WCC account (Happé and Frith 2006), which states that individuals on the autism spectrum are less sensitive to global structure than typical individuals. Furthermore, our findings do not support those of Booth et al. (2003), who reported poor configural properties of reproductions. One reason for the discrepancy in findings is that our stimuli were representationally simpler (i.e., two-dimensional geometric forms) and were not supported by a semantic interpretation. For example, whilst Booth et al. (2003) asked participants to draw, say, a house, we simply asked children to copy what they could see. Despite the relative simplicity of our stimuli, however, the global manipulations were sufficiently strong to modulate copying behaviour in both typical and autistic children, which warrants confidence in our findings. It is perhaps the case that reported drawing differences between autistic and typical children are a result of semantic factors relating to the planning of drawings (see van Sommers 1984), rather than purely visuospatial components.

There is, however, an aspect of our results that could be taken as support for the WCC account, namely, autistic children’s CEFT performance. First, they were significantly faster to find the hidden figure than typical children, replicating previous findings (Jolliffe and Baron-Cohen 1997; Pellicano et al. 2006) and demonstrating a local bias in the visual domain (i.e. search) that was not observed in the visuomotor domain (i.e. drawing accuracy). Second, there was a significant relationship between CEFT performance and copying accuracy in the autistic children in the predicted direction: those who displayed a local bias in the task, by locating the local shape more rapidly, were more likely to demonstrate global inaccuracies in their copies. Interestingly, this pattern was reversed in typical children, such that those who displayed a local bias in the task were less likely to demonstrate global inaccuracies. This CEFT-drawing relationship suggests that relative strengths in one visuospatial task (emphasising the local level) might be associated with relative weaknesses in another (requiring attention to the global level), in a manner consistent with the predictions of WCC. This was observed for our autistic sample, and the relationship is in accordance with previous findings that local bias in the CEFT is associated with inefficiencies in large-scale spatial learning (Pellicano et al. 2011). It is therefore possible that the autistic children were using different cognitive strategies to produce their copies. Although this may have only been evident in a discrete aspect of their global performance, the unique negative relationship between global accuracy and local bias may be indicative of a different approach to the analysis and reproduction of figures. The positive relationship between CEFT performance and average angular error in typical children is less in-line with our predictions, although it may instead reflect a general relationship between performance levels across tasks (i.e. those children who performed more efficiently in the CEFT were also more accurate at the copying task) (see Pellicano 2010, for a discussion of the relationship between CEFT and other skills). In future research, it would be beneficial to record the time-course of children’s copying behaviour, which might shed further light on potential qualitative differences between autistic and typical children’s drawing skills.

In contrast with CEFT performance, planning abilities (as measured by children’s Tower of London scores) did not predict performance in either group. Furthermore, unlike Booth et al. (2003), we did not find a group difference on planning abilities, although this could be related to the fact that we used a different measure. While they measured planning in relation to the drawing task itself, our measure of higher-order planning (the Tower of London) required children to plan a longer sequence of moves. Our assay of planning was therefore potentially more difficult—for both groups of children—than the one used by Booth et al. (2003).

The Enhanced Perceptual Functioning (EPF) account of visuospatial behaviour in autism (Mottron et al. 2006) explicitly states that superiorities at the local level are distinct from global processing skills, which are held to be intact. The findings regarding the relationship between autistic children’s drawing performance and the CEFT are therefore contrary to those predictions. Nevertheless, the overarching picture created by the data is one of no group differences in drawing performance—neither in terms of their attention to the global configuration nor to the individual constituent elements. These data, therefore, do not support an argument for local superiority in drawing production, as proposed by previous graphic production studies that favour the EPF account (e.g. Mottron et al. 1999; Sheppard et al. 2009b). However, it is important to note that these data may be subject to a ceiling effect, since the majority of drawings were produced without error at the local level. Future studies should, therefore, take an even more fine-grained approach, particularly with regard to angular deviation at the local level, to examine further potential differences in drawing skills. More subtle differences between groups may also be observable at a graphomotor level of explanation. For example, greater error in the Skewed condition not only relates to the reduction of grouping cues but also to the presence of oblique lines, which were not present in the Grouped condition. Oblique lines are more difficult to plan or execute than horizontal and vertical lines (Broderick and Laszlo 1987, 1988), although it is interesting that there did not seem to be any observable differences between autistic and typical groups in the representation of oblique orientations (c.f. Smith and Gilchrist 2005b)

Taken together, these findings in the visuomotor domain suggest that the two main accounts of visuospatial processing in autism (WCC and EPF) may not sufficiently capture drawing behaviour. It might therefore be useful for developmental psychologists to reconceptualise their approach to hierarchical perception and incorporate existing models of attention and perception from allied fields of psychology. For example, Humphreys (1998) draws the distinction between within-object coding (i.e., parallel coding of parts within objects) and between-object coding (i.e., parallel coding of multiple separate objects). These processes are conceptually similar to, respectively, global and local processing. However, in developmental accounts, global processing tends to be conceived as a parallel process, whereas local encoding is thought to rely on a serial individuation of separate smaller elements. The account put forward by Humphreys (1998) states that both processes occur in parallel (with global perception not being reliant upon the local level) and can be separately disrupted by damage to different brain regions. A recent meta-analysis of hierarchical visual processing in autism echoes this viewpoint (Van der Hallen et al. 2015). The authors report neither superior local visual processing nor a deficit in global processing in autism, and instead report a difference in the speed at which autistic people perceive the global form when there is interference from local details (e.g., in the Navon task and the CEFT). Our findings suggest that this distinction extends to the visuomotor domain, and future work in this area should aim to clarify whether perceptual differences between groups are reflected in the motor components of drawing production.

It may therefore be beneficial to take a wider perspective when constructing theories of visuospatial behaviour in autism, incorporating findings and theories from similar explorations in different populations. A recent study by D’Souza et al. (2016) compared global and local performance in participants with autism, Williams Syndrome and Down Syndrome, across a battery of tasks. They found both local and global biases in all three populations, depending on the nature of the task, and argued that a broad global/local distinction between syndromes is inappropriate. This suggests that it may be useful to consider the encoding of spatial relations within and between objects as something that is separable from the perception and detection of local features (see also Farran et al. 2003). By considering how these processes interact with each other, we can therefore move towards an account of visuospatial behaviour that encapsulates the variety of reported findings. It should also ensure that autism research is appropriately informed by advances in vision research, neuropsychology, and cognitive neuroscience, just as those areas have been fundamentally informed by autism research.