We examined the data for significant: (1) kurtosis, (2) skew, and (3) outliers. We also assessed whether minor manipulations in task administration format (computerized/human) or task instructions (naming the category of first pair selected), incorporated as part of a separate study, affected performance. All variables met the assumptions of normality, allowing us to conduct parametric analyses. Significant outliers included scores greater than 2.5 standard deviations (SD) from their own diagnostic group mean. Group means were calculated without the potential outlier score. No significant outliers were identified. Furthermore, chronological age was used as a covariate of interest in the ANCOVA, and the results remained significant whether the covariate was included or not; to reduce redundancy, we only report the results with age as a covariate. The task presentation manipulations had no effect on performance and thus are not discussed further. Both groups' performances on selecting Pair A (i.e., establishing a set) were near ceiling, with the mean percentage of correct scores over 95%, and no difference between groups (See Table 1 for means). The groups did not differ in their selection of stimulus category (color, shape, or size) for the first pair on any trial, χ²(N=27)<5.3, p>0.15. Only 17 of 660 Pair A selections were errors (TD=5; ASD=12) for the two groups combined. Further, on all 17 of these trials, participants made a correct Pair B selection, which may have lead to an inflated Pair B total score. These two issues raise the importance of controlling for Pair A performance prior to examining Pair B performance, which we accomplished by calculating the shift percentage.

As shown in Table 1, the ASD group had a lower percentage for correct Pair B selections, F(1,41)=27.29,p<0.001, and compiled a lower shift percentage, F(1,41)=24.13,p<0.001, compared to the TD group while controlling for differences in chronological age. Shift percentage was positively correlated with verbal mental age in the ASD group, r(22)=0.69,p<0.001 (See Figure 2); this relationship was not significant in the TD group, r(22)=0.17,p=0.459. However, the range of verbal mental age scores for the TD group was 4.14 (6.16 – 10.30) relative to the 7.37 (4.81–12.18) in the ASD group, and this reduced range may limit our ability to detect a correlation in the TD group.

The main goal of this study was to examine cognitive flexibility in school-aged children with ASD on a task that combined the benefits of inductive and explicit flexibility tasks (i.e., requiring a problem-solving strategy while also limiting the influence of other cognitive processes). Children with ASD demonstrated poorer performance on Pair B accuracy and the shift percentage scores. Cognitive flexibility performance was also positively correlated with verbal mental age in our ASD group.

With respect to previous cognitive flexibility studies, the most developed literature in this age range is with two inductive reasoning tasks: the Intradimensional/Extradimensional task and the WCST (Hill, 2004; Kenworthy et al., 2008; Pennington & Ozonoff, 1996; Sergeant et al., 2002). The results from the present study converge with those of prior investigations demonstrating cognitive flexibility impairments in ASD (Lopez et al., 2005; Maes et al., 2010; Ozonoff et al., 2004; Solomon et al., 2008, 2008; South et al., 2007; Yerys et al., 2009; but see Happé et al., 2006) and extends past findings to a novel task. The FIST attempts to capture different aspects of flexibility, because flexible thinking is not always reliant upon the two conditions of either inductive reasoning or following explicit rules. Furthermore, applying rules in a general way has often been a core training component of evidence-based social interventions for children with ASD (Rogers, 2000), and therefore the FIST may better capture the cognitive flexibility weaknesses observed by clinicians and parents for individuals with ASD (Geurts et al., 2009; Gioia et al., 2002; Kenworthy et al., 2008).

As discussed earlier, inductive tasks like the WCST, have been shown in latent variable factor analyses to load more on working memory than a shifting factor in TD children (Huizinga et al., 2006). The FIST's construction overlaps to some degree with the WCST, but also differs in ways that align it with more explicit tasks like the preparing to overcome prepotency task (Solomon et al., 2008). Thus, while the FIST is conceptualized primarily as a cognitive flexibility/shifting task, it is possible that impaired performance by the ASD group on this task may reflect some combination of cognitive flexibility/shifting and working memory impairments. Future investigations may seek to conduct latent variable factor analyses with complex cognitive flexibility tasks other than the WCST to determine whether the loading of WCST on working memory factors is specific to the task or characteristic of all complex cognitive flexibility tasks.

The correlation between shift performance and verbal mental age is not likely explained by language delay because the groups were matched on verbal mental age. This finding also converges with previous FIST studies documenting age-related improvements, specifically related to increases in verbal mental age, in typical development and ASD (Jacques & Zelazo, 2001, 2005; Jacques et al., 2002). The present study's ASD sample's age range extends beyond that of previous typical development studies; therefore we speculate these data are consistent with a delayed developmental trajectory for ASD in cognitive flexibility. It is also possible that different cognitive and neural mechanisms may support successful performance on cognitive flexibility measures for TD and ASD populations; the limited range of verbal mental age in the TD group restricted our ability to detect meaningful correlations.

This study has potential limiting factors. First, the verbal mental age of our participants was higher than that of participants in the original FIST studies (i.e., 4–5 years) (Jacques & Zelazo, 2001), and Pair A selection rates were near ceiling in both groups. Ceiling performance limits the ability to observe differences in Pair A selection, but not Pair B as accuracy was in the 75–90% range which is higher than previous TD studies (Jacques & Zelazo, 2001), but still not at ceiling for the ASD group However, our goal was to examine the ability of children to flexibly shift from a correct Pair A selection to a correct Pair B selection, thus ceiling performance on Pair A is a desirable result because it confirms that both groups reliably establish a cognitive set. Next, we did not correlate FIST performance with restricted, repetitive, behavior and interest symptoms. Cognitive flexibility is likely to relate to the insistence on sameness and behavioral rigidity observed in ASD, and the ADOS only includes two items (D4 and D5) that assess this collection of autistic behaviors. At the time of data collection we were unable to include more comprehensive measures like the repetitive behavior scale – revised (Lam & Aman, 2007). Future research will be needed to explore how cognitive flexibility, as indexed by the FIST, relates to these symptoms, as well as other performance-based and ecologically-valid measures of flexibility. Our ASD and TD samples were matched on verbal mental age rather than non-verbal mental age or overall mental age. Non-verbal mental age tends to be greater in those with ASD relative to verbal mental age (Joseph, Tager-Flusberg, & Lord, 2002). As a result, the estimate of cognitive flexibility in the ASD group may be overly conservative. That is, if we had matched on overall cognitive ability or non-verbal mental age, the ASD group may have demonstrated even poorer overall performance, inflating the observed differences. Detecting significant group differences despite this conservative matching procedure suggests a meaningful impairment in cognitive flexibility in children with ASD. Finally, use of a covariate that correlates with the dependent variable of interest may artificially inflate group differences statistically (see Dennis et al., 2009 for an example of IQ as a covariate), and verbal mental age correlates with FIST performance in our study and previously (Jacques & Zelazo, 2001; Jacques et al., 2002). This concern is mitigated by our findings of significant group differences when chronological age is not included as a covariate. Thus, even though the ASD group is chronologically older and matched on verbal mental age relative to controls, they performed worse in the FIST's flexibility condition. This sample is generally considered to be high functioning as all children in the ASD group except one had a developmental quotient (Developmental Quotient = Mental Age ÷ Chronological Age).greater than 0.70). The one child with a developmental quotient below 0.70, had a verbal mental age above 4 years, and his inclusion did not affect the results.

Finally, this study was conducted prior to the publication of evidence that reduced novelty processing may contribute to poor cognitive flexibility in ASD (Maes et al., 2010). This study suggests individuals with ASD are inflexible when previously irrelevant stimuli become relevant during the switch (termed “learned irrelevance”). When the previously irrelevant stimuli are removed and individuals with ASD must shift to a novel stimulus attribute, they perform similarly to controls. The FIST could be adapted to include a fourth dimension (e.g., number) that is only present during Pair B in substitution for an ignored dimension in Pair A selection. This additional stimulus dimension would be the “novel” switch category in comparison to the standard FIST.

The need for sameness is both highly heritable (Smith et al., 2009) and worsens across age in children with ASD (Richler, Huerta, Bishop, & Lord, 2010). As such, the importance of understanding the underlying genetic, neural and cognitive basis of cognitive flexibility, and then translating this information into meaningful interventions, is a high priority for improving the health outcomes of individuals with ASD.