To examine if there are significant differences in ROCF scores among the analysis groups (ASD, TD, and siblings of children with ASD).

Consistent with the theoretical accounts of local visuospatial processing bias in ASD, we hypothesized that more children with ASD would show a more part-oriented processing style and that children with ASD would have more difficulties copying and reproducing the target figure (i.e., score lower in Organization, Error and Accuracy parameters) than the TD group. We also hypothesized that the sibling group would exhibit BAPs (i.e., similar patterns of visuospatial processing abilities and patterns with children with ASD) and score lower than the TD group across all parameters, but score higher than the ASD group. Relatedly, we hypothesized that similar patterns would be observed in the number of siblings who show a more part-oriented processing style.

To examine if the ROCF scores are associated with Social Interaction, Verbal Communications, and RRBs in children with ASD.

Due to the lack of consensus among studies examining the relations between local visuospatial processing bias and autistic symptoms in children with ASD, we considered the second research question to be exploratory.

The participants were a subset from a larger study examining associations between genetic compositions and various autistic traits. All data collection procedures (IRB no: B-1703/388-303, B-1106/129-009, B-0807/059-018) and retrospective analyses of collected data (IRB no: B-1912/582-107) were approved by the Institute Review Board (IRB). Participants in the larger study were recruited and referred from the Child Psychiatric Clinic of Seoul National University Bundang Hospital by two board-certified psychiatrists (the 4th author and the corresponding author of this study). The board-certified psychiatrists initially screened all children who came to the authors’ clinic based on Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM-5) and suggested if they are willing to participate in the larger study.

A total of 339 participants were included in the study, including 134 children with ASD, 150 siblings of children with ASD, and 55 TD children, who did not have a sibling with ASD. Of 150 siblings of children with ASD, 77 were the siblings of the 134 probands included in the study. Consistent with the age range that the DDS can be applied to, only participants who were 5- to 15-year-olds were included. There were no significant differences in age among the groups. Detailed demographic characteristics of each group are reported in Table 1.

All collected data were retrospectively analyzed to address the research aims of the present study. Upon informed parental consent, participants completed a battery of tests composed of the ROCF during a one-time visit to the authors’ laboratory. The tests included the Korean translation of Autism Diagnostic Observation Schedule (ADOS) [32] and the ADOS-2 [33,34], the Korean translation of ADI-R [29,35], the Social Responsiveness Scale-the Second Edition (SRS-2) [36], and the Korean-Leiter International Performance Scale-Revised (K-Leiter-R) [37]. The ROCF and K-Leiter-R were administered and scored by trained graduate students under the supervision of professional clinical psychologists. The reliability of all administrations and assessments of the ROCF was regularly checked and closely monitored by professional clinical psychologists to make sure they closely adhere to the DSS instructions [14]. ADOS-2 were administered by the research-certified researcher (GB; the fourth author of this study) and trained research assistants. All research assistants worked at the same lab with GB on a daily basis, and the inter-rater reliability between research assistants and GB was maintained to be above 80%. Assessments administered by research assistants were video recorded and double-checked by GB. The diagnostic status of all participants was confirmed with best clinical judgments of experienced psychiatrists based on the ADOS-2, ADI-R and SRS-2 scores. All participants with ASD had either ADOS-2 or ADI-R scores higher than the cutoff points and met the DSM-5 criteria; all children in the TD and sibling group scored below the clinical threshold of ADI-R and ADOS-2.

Prior to the analysis, the Style variable was dummy-coded. A value of one was assigned for the Style category exhibited in a condition and a zero was assigned for remaining categories within the condition. For instance, a participant who copied the figure in a part-oriented style was given a value of 1 for the Part-Oriented style in the Copy condition and 0 for the remaining categories.

To address the first research question, each of the ROCF parameter scores was analyzed with a repeated-measure Analysis of Covariance (ANCOVA) with analysis groups as betweensubject factors and age and performance IQ as covariates. Each dummy-coded style variable was analyzed using ANCOVA with the same between- and within-subject factors and covariates entered. Post-hoc pairwise comparisons using the TukeyKramer adjustment with Box’s conservative epsilon correction, which corrects for multiple hypotheses testing of samples with unequal sizes, were computed for variables significant in ANCOVA. In the Style domain, variables at the extreme ends (whether or not participants’ style was Part-Oriented or Configurational) were examined to limit the number of hypotheses tested.

Among the data of 339 participants included to investigate the first research question, the data from the children with ASD were separately analyzed to address the second aim of the study. Partial correlations between the 15 variables (i.e., five parameters for each of Copy, IR, and DR condition) and the three ADI-R scores (i.e., Reciprocal Social Interaction, Verbal Communication, and RRB) were computed, while controlling for the standardized K-Leiter-R scores. Because we were particularly interested in children’s ability to communicate via speech, the current algorithm scores that measure verbal communication were included. To examine the associations with dichotomous Style variables, three sets of six linear robust regressions with the ‘vce(robust)’ option in Stata [41], which corrects for violations of distributional assumptions, were conducted; each set modeled Reciprocal Social Interaction, Verbal Communication, and RRB, separately as outcome variables. For each of these, a dummy coded Style variable and the standardized K-Leiter-R score were entered as predictors. Holm-Bonferroni’s corrections were used to account for the multiple hypothesis testing.

Additionally, to examine in detail how RRBs are uniquely associated with the ROCF scores of children with ASD, the RRB domain was further divided into four subdomains: encompassing preoccupation or circumscribed pattern of interest, compulsive adherence to routines or rituals, motor mannerisms, and preoccupation with parts of objects or non-functional elements of material. Another set of regressions was computed to predict each of the RRB subdomains with the ROCF variables that were significantly associated with overall RRB scores, while controlling for the performance IQ. All data analyses were computed using Stata statistical software [41].

Table 2 presents the mean and standard deviation of each score for all three analysis groups. There was a significant effect of the condition in the Configurational variable, F(2, 612)= 3.54, p<0.001, η_p²=0.04, but not in the Part-Oriented variable, F(2, 612)=0.07, p=0.68, η_p²<0.01 (Table 3). The post-hoc Tukey-Kramer test in the Configurational variable showed that there was no significant difference between the two Recall conditions, p>0.05 [mean differences (MD)=0.03], and the proportion of the number of participants who drew in a Configurational way was significantly lower in Copy conditions compared to the IR and DR conditions, both p<0.05 (MD=0.17 and 0.14, respectively). There was a main effect of the analysis group for Configurational and Part-Oriented styles, F(2, 612)=8.29, p<0.001, η_p²=0.06 and F(2, 612)=10.02 p<0.001, η_p²=0.09, respectively, when controlling for condition and age.

The proportion of the number of participants who drew in a Part-Oriented style was significantly higher in the ASD group than in the sibling group (MD=0.13) and was lowest in the TD group (MD=0.16) (Figure 1). The opposite pattern was reported in the Configurational domain, in which the number of participants who drew in a Configurational style was lowest in the ASD group and highest in the TD group, all p<0.05 (Figure 2). The results from Tukey-Kramer Comparisons for Conditions and Analysis Group are presented in Supplementary Tables 2 and 3 (in the online-only Data Supplement), respectively.

There was a significant interaction effect between the analysis group and condition only in the Configurational parameter, F(4, 612)=1.62, p=0.02, η_p²=0.02. In all three analysis groups, there was a general trend that the proportion of the number of participants who drew in the Configurational style increased from Copy to IR condition, MD=0.15 in the ASD group, p<0.05; MD=0.18 in the sibling group, p<0.05; and MD=0.15 in the TD group, approaching significance (i.e., studentized range critical value=2.90). From the Copy to DR condition, the proportion of the number of participants who drew in a Configurational style significantly increased in sibling (MD=0.19) and TD group (MD=0.21), both p<0.05, but no significant change from the Copy to DR condition was reported in the ASD group (MD=0.02), p>0.05. Within the Recall conditions, the number of children in the TD and sibling groups who drew in the Configurational style did not differ between the two Recall conditions (MD=0.01 and 0.07, respectively), both p>0.05, but the number of children with ASD who drew in a Configurational style significantly decreased from IR to DR (MD=0.14), p<0.05.

There were significant main effects of condition for Structural Accuracy, F(2, 612)=121.61, p<0.001, η_p²=0.28, and Incidental Accuracy, F(2, 612)=486.44; p<0.001, η_p²=0.61. Post-hoc tests showed that both Accuracy scores in the Copy condition were significantly higher than those in the IR (MD for Structural and Incidental Accuracy=4.96 and 13.61, respectively), both p<0.05, and DR conditions (MD for Structural and Incidental Accuracy=5.02 and 13.63, respectively), both p<0.05, and there were no significant differences between IR and DR conditions (MD for Structural and Incidental Accuracy=0.06 and 0.03, respectively), both p>0.05. There were main effects of the analysis group for Structural Accuracy and Incidental Accuracy, F(2, 612)=6.59, p=0.002, η_p²=0.04 and F(2, 612)=14.23, p<0.001, η_p²=0.08 (Figures 3 and 4). Post-hoc pairwise comparisons indicated that the scores of the ASD group in Structural Accuracy and Incidental Accuracy were significantly lower than for those of the Sibling and TD groups (MD for Structural Accuracy=1.77 and 1.98 and MD for Incidental Accuracy=3.66 and 4.86, respectively), both p<0.05. There were no significant differences between TD and sibling groups (MD for Structural and Incidental Accuracy=0.21 and 1.20, respectively), both p>0.05.

There was a main effect of condition in Error, F(2, 612)= 19.15, p<0.001, η_p²=0.06. The Error scores for both IR and DR conditions were significantly higher than those for the Copy condition (MD=0.42 and 0.46, respectively) p<0.05, but there were no significant differences between IR and DR conditions (MD=0.04), p>0.05 (Figure 5). There was a significant main effect of analysis group, F(2, 612)=3.87, p=0.02, η_p²=0.02, and a significant interaction effect of analysis group and condition for the Error scores, F(4, 612)=3.71, p=0.005, η_p²=0.02 (Supplementary Table 4 in the online-only Data Supplement). Follow-up analyses showed that the TD group made significantly less errors in the Copy condition than in the IR and DR conditions (MD=1.27 and 1.25, respectively), both p<0.05, but the number of errors did not significantly differ across the three conditions in the ASD group (MD between Copy and IR, between Copy and DR, and between IR and DR=0.40, 0.32, and 0.08, respectively), all p>0.05, and the sibling group (MD between Copy and IR, between Copy and DR, and between IR and DR=0.13, 0.29, and 0.17, respectively), all p>0.05.

The results of the ANCOVA analyses showed that the effects of condition were significant in Organization, F(2, 612)=25.08, p<0.001, η_p²=0.08. Post-hoc Tukey-Kramer comparisons showed that scores of the Copy condition were significantly higher than those of the IR and DR conditions (MD=1.21 and 1.06, respectively), both p<0.05, and there was no difference between the two recall conditions (MD=0.15) (Figure 6). There was no other main effect of the analysis group or interaction effects.

A correlation matrix of all continuous variables is presented in Table 4. Correlational analyses showed that there were significant negative correlations between Social Interaction and Organization scores in Copy, r(140)=-0.258, p=0.002, controlling for the performance IQ. No significant correlation with the Verbal Communication domain was found, but a trend toward significance between Error in Delayed and Recall conditions and Verbal Communication emerged, r(112)=0.24, p= 0.01, and r(108)=0.21, p=0.03, respectively. However, this was not considered significant after applying Holm-Bonferroni correction. Regression analysis (Table 5) showed whether or not a child with ASD drew in the Part-Oriented style in the Copy condition significantly predicted his/her RRB scores, β=0.25, p=0.006, R²=0.06, 95% confidence interval of R² (0.016, 0.136). No significant association between the two Style variables and Social Interaction and Verbal Communication was reported (Supplementary Tables 5 and 6 in the online-only Data Supplement). Among the subdomains of RRBs, only drawing in a Part-Oriented style significantly predicted preoccupation with parts of objects or non-functional elements of material domain, β=0.20, p=0.01, R²=0.04, -0.02≤R²≤0.11.

These findings suggest avenues for future research and practical implications. First, the performance of children with ASD did not uniformly differ from those of the TD and Sibling groups across all parameters. To our knowledge, there was no literature explicating what cognitive procedures are distinctively captured by each parameter. Eylen et al. [22] contend that local processing is not a well-defined or established construct in ASD literature despite the numerous theories that attempt to utilize local processing theory to account for autistic behaviors. More research on the underlying cognitive mechanisms of local processing styles and how they manifest in the different parameters of the ROCF tasks is needed to better understand why such nuanced patterns were observed in the current study and apply the findings to clinical use. Additionally, we suggest incorporating the ROCF task with neuroimaging studies and short time scale studies to draw a more direct conclusion about how individuals with ASD perceive and process visuospatial stimuli.

Furthermore, we want to emphasize that, consistent with Frith’s argument [1], this particular processing style is not a deficit but rather a trait that can explain many strengths and/or savant skills of individuals with ASD. The tendency of children with ASD to make sense of complex stimuli by parsing it into its parts bears educational implications in terms of how to format educational materials and environments to effectively support them. For instance, rather than expecting children with ASD to make sense of materials using top-down processes (i.e., identify the big picture and attend to the components based on the understanding of the big picture), it would be more helpful to present the information in small units and support them in the process of referencing the whole picture.

This study supports the claim that the altered pattern of visuospatial processing is an inherent trait of ASD by reporting that siblings of children with ASD showed BAP, a processing style that resembles that of children with ASD. However, it is also interesting to note that their processing style did not negatively impact their ability to organize the information or reproduce the components. Indeed, when their collective scores in Accuracy and Error parameters are considered, the siblings’ performance was superior to those of both the TD and ASD groups in that they were able to reproduce as many figures as TD children did, but they had the least number of errors. Relatedly, Pring et al. [48] also contend that the tendency toward segmentation of visual information may account for artistic talents and aptitude for drawing even for TD children. This finding suggests the possibility of an endophenotype of ASD that prefers local processing to be considered a strength.

This study has several limitations to be considered. First, the generalization of the findings may be limited as only children with sufficient motor or cognitive skills to perform the ROCF tasks were included. Also, to maintain the accuracy of the DSS scores, the analysis was restricted to children under the age of 15. However, the influence of maturation and age on changes in processing style has been articulated in previous studies [18,22]. This limitation warrants the need to develop a new instrument that can be applied across various age groups to accurately measure visuospatial processing patterns. It is also important to note that the results are drawn from a single visuospatial task. We, therefore, suggest that future studies with additional visual-processing tasks such as a block-design task are needed to triangulate the findings of this study.

The number of participants and percentages of males in the ASD, sibling, and TD group were imbalanced. As gender is an important predictor of processing style [22], replication studies with balanced demographic groups are needed to confirm the findings of the study. Furthermore, it is important to note that the TD children were not longitudinally tracked, and therefore does not rule out the possibility of transitioning into the sibling group in the future. The groups were not matched on verbal IQ although previous studies on TD children showed that verbal IQ was a determining factor in the accurate reproduction of visual stimuli [31,49]. Relatedly, the low Accuracy scores of the ASD group may reflect low working memory skills rather than their visuospatial processing abilities. It is possible that individual characteristics such as verbal IQ, memory skills, or autism severity may have accounted for the differences between the groups. Therefore, the current study needs to be replicated with consideration of participants’ individual factors that may influence the performance on the ROCF tasks.

Finally, although Holm-Bonferroni corrections, a procedure with progressively adaptive threshold values [50], are considered more powerful than Bonferroni corrections by lowering the risk of increased type II error (i.e., not rejecting the null hypothesis when an alternative hypothesis is true), application of HolmBonferroni corrections may still have inflated the possibility of Type II error.