Autism Traits Predict Self-reported Executive Functioning Deficits in Everyday Life and an Aversion to Exercise

The color–shape switching task is commonly used to derive switching costs and mixing costs that typically serve as measures of the shifting and monitoring components of EF, respectively. Both a shape and a color task are used. Each trial in the mixed block began with a fixation cross for 350 ms. Then, a cue was presented for 250 ms that specified the required response: a rainbow for a decision based on color and a superimposed triangle/circle for a shape decision. The cue was immediately followed by one of the four possible targets: red circle, red triangle, green circle, or green triangle. In the shape task, the participant was directed to respond to the shapes while ignoring the color. In contrast, in the color task, the participant was told to respond to the color while disregarding the shape. Two fingers of one hand were used to make color decisions, while two fingers of the other hand were used to make shape decisions. Half of the trials require a task switch from color to shape, or vise versa, and half are repeated trials which do not require a task switch. The difference between repeat trials and switch trials in the mixed block indicates switching costs. Two single tasks are included each consisting of 36 pure color and 36 pure shape trials. The mixed task with task cues contains 3 blocks of 48 trials. The difference between repeat trials in the mixed block and the mean of the pure blocks indicates mixing costs.

This exact instantiation of the color–shape switching task shows test–retest reliability over a 1-week interval of r = 0.62 for switching cost and r = 0.75 for mixing costs (Paap and Sawi 2016). By general standards, these are less than desirable, but as Paap and Sawi point out, these reliabilities are as good or better than those typically observed for dependent measures based on RT difference scores. Furthermore, the switching-cost measure derived from this exact color–shape task demonstrated convergent validity because it formed a coherent latent variable with two other cued switching tasks (Paap et al. 2017).

Geurts et al. (2009) reviewed the “paradox” of cognitive flexibility in autism. They observe that both clinicians and researchers widely believe that cognitive flexibility deficits are pathognomonic of ASDs; however, their review of five categories of tasks used to measure flexibility led them to conclude that there was no consistent evidence of deficits. They further considered the extent to which the inconsistencies are due to measurement problems or to the heterogeneity of the autism spectrum given that there are substantial individual differences in the type of difficulties individuals with autism experience. Geurts et al. present an insightful discussion of the task-impurity problem in the Wisconsin Card Sorting Task (WCST) and Intra Dimensional–Extra Dimensional (ID–ED) that eventually leads to their endorsement of task-switching paradigms such as the color–shape switching task: “This more mechanistic approach probably is the best way to study cognitive flexibility” p. 7. Geurts et al. share that only four studies investigated task-switching in individuals with autism and none of them reported performance deficits. This is true, but the implications of the null results may be limited because they were based on small ASD groups (n’s range from 10 to 23) and furthermore substantially modified the standard task-switching paradigm that is exemplified by the color–shape switching task described in the previous paragraph.

Each trial began with the presentation of a center fixation (+) for 500 ms. The center fixation was immediately followed by the target stimulus which was either a “>” or a “<”. In this task, participants are instructed to press the left key if the arrow points to the left, and the right key if the arrow points to the right. The participants were instructed to respond as quickly as possible without making errors. In a neutral block the target was displayed either 2.3° above or below the center fixation. In the spatial Stroop block the target was displayed either 3.9° to the left or to the right of the center fixation. A trial was defined as congruent if the location of the target was on the same side as the correct response and as incongruent if the location of the target was on the opposite side. The standard marker of inhibitory control within this task is the difference in mean response times between congruent and incongruent trials. All participants started with two neutral blocks of 20 trials. The first block was considered practice. The neutral blocks were followed by two spatial Stroop blocks. Each of these consisted of 20 congruent and 20 incongruent trials presented in random order. The test–retest reliability of a similar spatial Stroop task was r = 0.71 for 17 normal adults tested over a period of about 11 weeks (Wöstmann et al. 2013).

The spatial Stroop task is superior to the original Stroop color–word when the groups being compared may differ in receptive language ability as the interference generated from the predisposition to read the word increases with reading skill and automaticity. For example, Adams and Jarrold (2009) showed that children with autism were actually better than matched controls on the color–word Stroop but performed equivalently in two tasks where the to-be-ignored and incongruent task was not reading words. In the only comparison between ASD and matched controls using the spatial Stroop task, there were no group differences (Schmitz et al. 2006), although there were concurrent group differences in regional fMRI brain activation. As described in the introduction, inhibitory control measures are somewhat notorious for lacking convergent validity. Yet, Paap et al. (2019) used an exploratory factor analysis to show that the spatial Stroop and two versions of the Simon task formed a coherent latent variable that excluded the flanker task. Despite the null results reported by Schmitz et al. (2006) and the measurement problems, a theoretical framework that assumes that ASD is associated with deficits in inhibitory control would predict that interference scores will increase with AQ scores.

Only the low discriminability conjunctive search condition from Friesen et al. (2014) was used in the current study. The target stimulus was a blue triangle. The search was conjunctive because the distractors included purple triangles and blue diamonds. The search involved low discriminability because blue is similar to purple and diamonds are similar to triangles. The target appeared randomly in 1 of 26 designated locations on the screen. There were 24 trials in which the target was present. These were equally divided between distractor set sizes of 0, 5, 15 or 25 distractors. The 18 target-absent trials were equally divided between 5, 15, and 25 distractors. Participants were instructed to press the “1” key the target was present and press the “0” key if it was not. After the response, the next array was presented.

Our conjunctive visual search task was quite similar to the ones used by Plaisted et al. (1998) and O’Riordan et al. (2001). Both of these studies showed that children (mean = 8 years) with a diagnosis of autism performed better than normally developing controls, especially on the more difficult target-absent trials. The classic interpretation of conjunctive search, Treisman and Gelade’s (1980) feature-integration theory of attention, renders this ASD advantage in conjunctive search surprising. The theory assumes that focused attention is required in order to conjoin features like shape and color into a single object. If ASD entails attention to detail and a preference for features over integrated wholes, then the ASD group should find conjunctive search difficult.

A potential confound with fluid intelligence was assessed using Set 1 of the Raven’s Advanced Progressive Matrices (Raven et al. 1977). The Raven’s task was computerized and consisted of 12 items total; participants were given a maximum of 2 min to select from a set of 8 possible choices for each puzzle, each consisting of a pattern with a piece missing. Fluid intelligence is assessed using Set 1 of the Ravens Advanced Progressive Matrices (Raven et al. 1977). Participants are told to “Look at the pattern, think what the missing part must be like to complete the pattern correctly.” Participants are provided with 8 options to choose from and are allotted a maximum of 2 min to respond to each item.

Traits associated with autism are measured using the AQ scale (Baron-Cohen et al. 2001). This measure has been described as a valuable instrument for rapidly quantifying where an individual is situated along the continuum from autism to normality. The AQ consists of 50 questions with four scaled-response options ranging from “definitely agree” to “definitely disagree.” For example, item 1. states, “I prefer to do things with others rather than on my own.” Baron-Cohen et al. introduced a binary scoring rubric whereby one point was awarded if the participant indicated either definite or slight agreement with an autism trait and otherwise zero. Thus, possible scores ranged from 0 to 50. For a group of 58 adults with Asperger syndrome (AS) or high-functioning autism (HFA) with mean age of 32 years the mean AQ score was 35.8 compared to 16.4 for a group of matched controls, and 17.6 for a group of university students. There was a small amount of overlap at the high end of the distributions. For example, whereas 79.3% of the AS/HFA group scored 32 or more, only 2.3% of the control group did so. However, 6% of a sample of Cambridge University students scored 32 or higher. Baron-Cohen et al. conclude that “A score of 32+ appears to be a useful cut-off for distinguishing individuals who have clinically significant levels of autistic traits” p. 15. There was very little overlap at the low end as 40.6% of the controls had AQ scores less than 18, but no one from the AS/HFA group did so. Control males scored higher than control females, but there were no sex differences in the AS/HFA group or in the group of university students. The test–retest reliability for a group of 17 university students was r = 0.7.

The BDEFS is comprised of 89 questions split into five subdomains: Time Management, Organization/Problem Solving, Self-Restraint/Inhibition, Self-Motivation, and Self-Regulation of Emotions (Barkley 2011). Items are presented as “problems” and participants are asked to circle the response that best describes their behavior, that is, how often they experience this problem during the past 6 months: never or rarely, sometimes, often, or very often. Barkley equates attention deficit hyperactivity disorder (ADHD) with deficits in EF and hence with performance on the BDEFS. If BDEFS and AQ are strongly correlated this would indicate that the disorders are related (with respect to these subjective measures) and that deficits in EF are prevalent in ASD.

The internal consistency, measured as Cronbach’s α, of each of the subscales is satisfactory ranging from 0.91 to 0.96 with a median of 0.95. Based on a sample of 62 normal adults the test–retest reliability for the overall BDEFS score was r = 0.84 (Barkley 2011). The test–retest reliability for the five subscales ranged from r = 0.62 to 0.90 with a median of 0.78.

Subjective measures of personal self-control were obtained using the BSCS (Tangney et al. 2004). This 13-item questionnaire is widely used throughout the literature on self-control and has been proposed as a predictor of both beneficial and adverse outcomes. Higher levels of personal self-control are indicated by higher scores on the scale. The BSCS and BDEFS should strongly correlate to the extent that self-control and executive functioning are the same construct. The internal reliability of the BSCS is good with αs of 0.83 and 0.85 in the two studies reported by Tangney et al. (2004). Likewise, the test–retest reliability for the BSCS was good (r = 0.87) over a three-week delay for a sample of 233 university students.

Three of the UPPS Impulsive-Behavior subscales developed by Whiteside and Lynam (2001) were included: (lack of) premeditation, urgency, and (lack of) perseverance. Their fourth facet, sensation seeking, was not included because it showed the weakest correlations with the other three facets (Whiteside and Lyman 2001) and with a variety of measures of EF (Duckworth and Kern 2011). Whiteside and Lyman (2001) interpreted their four distinct factors as “discrete psychological processes that lead to impulsive-like behaviors” (p. 685). The α reliabilities were 0.87, 0.89, and 0.83 for (lack of) Premeditation, Urgency, and (lack of) Perseverance, respectively. Weafer et al. (2013) reported test–retest reliabilities for the UPPS based on a sample of 126 healthy adults showed correlations ranging from r = 0.81 to 0.86 across the subscales. Paap et al. (2020a) reported significant correlations between the BSCS and the trait impulsivity scales.

Because a pilot study (Mason et al. 2018) showed significant correlations between AQ scores and the two items described below, a comprehensive exercise survey (CES) was developed and included in the present study that looked at the frequency, duration, and intensity of various types of exercise and physical activity and the reasons for engaging in the types of exercise an individual does, as well as reasons that prevented them from doing more. The exercise item used in earlier work read: “How often in a typical week do you exercise, work out, or participate in a sport?” Responses were based on a 5-point Likert scale ranging from never to very often and the correlation with AQ scores was r(129) =  − 0.36, p < 0.001. A related, but distinct item emphasized ability, not frequency of participation: “Team sports often involve dividing your attention between a ball, a goal, your opponents, and your teammates. Do you excel at these sports?” Responses were based on a 5-point Likert scale ranging from No, I am far below average to I am much better than average. The correlation between this item and AQ scores was r(129) =  − 0.18, p = 0.046.

The CES assesses activities in this order: (1) walking, (2) climbing stairs, (3) single-person exercise, (4) team sports, (5) one- or two-person ball sports, (6) contact/combat sports, (7) dance, (8) other activities not probed, (9) physical activity at home, and (10) physical activity at work. For Categories 1 to 8 frequency was assessed with an item like this: How often do you bike? And a 9-point Likert scale: never or rarely, a few times per year [0], 1 or 2 times a month [.33], 1 time per week [1], 2 times per week [2], 3 times per week [3], 4 to 5 times per week [4.5], 6 to 7 times per week [6.5], multiple times per day [14]. The response selected was recoded to frequency per week as shown in parentheses in the previous sentence. If the participant indicated any frequency greater than zero, they were then asked to report the typical duration per session. The range of responses was adjusted for the different activities and sports. For example, the choices for biking were 15, 30, 45, 60, 75, 90, 105, 120, 150 min per session. A Total Time per week for each type of exercise/activity was computed by taking the product of frequency and time, for example, 3 times a week × 30 min per session = 90 min per week. A higher order composite, Energy, was calculated by multiplying Total Time by the selected Intensity Level for each activity/sport. The intensity probe read What is your average level of effort (intensity) for each activity? Low: small increases in breathing or heart rate, Moderate: noticeable increases in breathing, heart rate, and perspiration, High: considerable increases in breathing, heart rate, and perspiration. The three levels were coded as 1, 2, and 3. For a competitive cyclist this may yield a Total Energy score of 7 (days per week) × 105 (min per session) × 3 (high intensity) = 2205. In addition to biking, the single-person exercise category included climbing (rock/mountain/wall), exercising (with weights, machines, or floor), skateboarding (skating), running, surfing, swimming, and yoga (tai chi, pilates). Although we refer to these activities as “single person” they can, of course, be done either alone or with others. Thus, in a following item we ask if they prefer to do them alone or with others.

A somewhat different approach was taken to measure physical activity at home and at work. For example, we asked: On a typical weekday, how many hours are you at home and not sleeping. This was followed by: What percentage of the time when you are at home and awake is spent in activities at each of the indicated levels of effort? Sitting (watching TV, using computer/phone, reading, homework, conversation, etc.), Standing (at sink/basin, at standing desk, at work bench, ironing, etc.), Moving about with lifting, carrying, pushing, bending, or squatting. Based on the amount of time spent at home awake per weekday and the percentage of time spent in each mode, the total amount of time spent sitting, standing, and moving can be computed for each individual and combined with similar estimates for probes about the weekend and time spent at work. Finally, sitting, standing, and moving were treated as three intensity levels and assigned the weight of 1.0, 1.5., and 2, respectively. Thus, the total time in each mode (sitting, standing, and moving) can be summed for weekdays at home, weekends at home, and time spent at work. Total energy per week at home and at work can also be computed by multiplying the total times by the intensity weights.

Total time and energy expended per week walking were built up from separate probes gathering information about walking while commuting to the university, walking while commuting to work, and walking a dog or walking (hiking) for pleasure. Stair walking was assessed with this item: How many flights of stairs do you walk up on a typical day? None [0], 1 to 5 [3], 6 to 10 [8], 11 to 15 [13], 16 to 20 [18], more than 20 [25]. The number of flights was coded as indicated by the square brackets for each Likert value. We used 30 s per flight of stairs to estimate the total time climbing stairs per week. As usual, the amount of energy was computed by multiplying total time by the reported intensity level for stair climbing. Finally, because each type of physical activity or exercise yielded an estimate of Total Time and Energy, Per Week they can be summed to estimate the Grand Total Time of physical activity and the Grand Energy Expenditure per week.

Participants were asked to respond to a number of demographic questions pertaining to their background, including the educational level obtained by their parents. They were also asked to rate their frequency of participation in activities such as video games, musical instruments, exercise including team sports, and meditation/mindfulness, in the form of a single question for each activity.

Using the binary scoring method introduced by Baron-Cohen et al. (2001) our usable sample of 200 students had a mean of 19.9 (SD = 5.3). This is slightly greater than the means reported by Baron-Cohen et al.: (a) for their control group (M = 16.4), (b) for their group of Cambridge students (M = 17.6), and by Ruzich et al. (2015) in a review of 73 published studies (M = 16.9). Two point 5% of our distribution scored 32 or above, the cutoff Baron-Cohen suggested “….for distinguishing individuals who have clinically significant levels of autistic traits.” The 2.5% is very similar to the proportion obtained in Baron-Cohen et al.’s control group (2%), but somewhat less than the 6% reported for Cambridge students. In general, our distribution of AQ scores derived from the binary scoring method appears to be reasonably close to that obtained by the scale developers.

By ignoring the distinction between definite and slight agreement (or between disagreement and slight disagreement) the binary scoring method potentially discards useful information. Austin (2005) proposed to retain that information and use the Likert values of 1, 2, 3, and 4 to score the responses definitely disagree, slightly disagree, slightly agree, and definitely agree, with reverse scoring as applicable. Note that the larger numbers are assigned to the “agree” responses and that, consequently, larger scores indicate greater agreement with autism traits. In a direct comparison of the two methods Stevenson and Hart (2017) reported that Likert scoring yielded higher internal consistency and test–retest reliability for both the overall AQ scores and for the individual subscales. For example, the test–retest reliability for the overall scores and for a sample of 178 university students was r = 0.82 for binary scoring and r = 0.86 for Likert scoring. Likewise, the correlations reported below were always stronger using Likert scoring. We report mean Likert scores (not the total of all 50 items) such that the possible scores range from 1.0 to 4.0 with an average rating of 2.5 representing the neutral point.³

There is no principled way to partition the continuum of AQ scores into subgroups (see Stevenson and Hart 2017, for an excellent discussion) and partitioning a continuum into subcategories can result in the loss of information. Consequently, most of the analyses reported below use AQ scores as a predictor or outcome variable in a correlational analysis. Thus, it is important to show that the distribution of AQ scores has desirable properties. As shown in the distribution of AQ scores (Fig. 1) and in the Kolmogorov–Smirnov tests (Table 1) the AQ scores do not depart from normality and do not have marked skewness or kurtosis.

Our mean of 2.22 is very close to the mean of 2.18 reported by Stevenson and Hart (2017) for 403 undergraduate students (mean age = 19.3 years). Likewise, our mean is nearly identical to the average of the 13 sample means (2.18) reviewed by Stevenson and Hart. Our sample appears to be typical for neurotypical young adults.

Based on self-reports of sex-at-birth our student sample consisted of 31% males and 69% females. Thus, males were underrepresented, even relative to the university enrollment (43%). The mean Likert AQ scores for females (2.25) and males (2.18) did not significantly differ, t(197) = 1.84, p = 0.067. The absence of a sex effect replicates the null result reported by Stevenson and Hart (2017). However, AQ scores did decrease with age, r(199) =  − 0.204, p = 0.004. Neither the composite (father’s education, mother’s education, family income) measure of SES nor the Raven’s test of fluid intelligence were significantly correlated with AQ scores: r(199) =  − 0.02, p = 0.750 and r(198) =  + 0.08, p = 0.24, respectively.

The RT differences between congruent and incongruent trials are referred to as interference scores or Stroop effects and typically used as a measure of selective attention or inhibitory control. The mean Stroop effect of 42 ms was significant, t(197) = 12.11, p < 0.001. For young adults, this is a typical and robust interference effect. Overall accuracy was 94.0% correct.

The RT differences between repeat and switch trials from the mixed block are referred to as switch costs and are typically used as a measure of switching or shifting ability. The mean switch costs of 189 ms was significant, t(197) = 24.91, p < 0.001. Overall accuracy was 92% correct. The RT differences between the single-task blocks and the repeat trials from the mixed block are referred to as mixing costs and are typically used as a measure of the monitoring or updating component of EF. The mean mixing costs of 301 ms was significant, t(195) = 29.18, p < 0.001. Overall accuracy was 96% correct. For young adults both the switching costs and mixing costs were typical and robust.

In conjunctive visual search, search time is typically a linear function of the number of distractors with slopes on target present trials typically one-half of that on target absent trials. This follows from the assumption that search is serial and self-terminating and that the target is randomly located within the display. The sample means conform to these norms with a slope of 25 ms/item on the target present trials and 51 ms/item on target absent trials. Conjunctive visual search is not typically used as a measure of EF, but Friesen et al. (2014) assumed that it reflected a measure of the ability to disengage attention.

BDEFS items are prompted with this instruction: How often do you experience each of these problems? For example, Procrastinate or put off doing things until the last moment. Responses are a Likert scale consisting of 1 Never or rarely, 2 Sometimes, 3 Often, and 4 Very often. The mean of the 89 items for each participant was computed, with reverse scoring as applicable. Thus, item scores range from 1 to 4 with a neutral point of 2.5. Characteristics of the distribution of BDEFS scores are shown in Table 4. The mean of 1.9 is less than the neutral point and it is important to remember that this is a scale of “deficits” in EF and that higher BDEFS scores reflect less EF ability.

BSCS items are prompted with this instruction: Using the scale provided, please indicate how much each of the following statements reflects how you typically are. Responses are a Likert scale consisting of 1 to 5 with the end points labeled Not-at-all and Very-much. The sum of the 13 items for each participant was computed, with reverse scoring as applicable. Thus, individual scores range from 13 to 65 with a neutral point of 32.5. Characteristics of the distribution of BSCS scores are shown in Table 4. The mean of 40.2 is greater than the neutral point and it is important to remember that this is a scale of “self-control” and that higher BSCS scores reflect more EF ability.

Three of the UPPS Impulsive-Behavior subscales developed by Whiteside and Lynam (2001) were included: (lack of) premeditation, urgency, and (lack of) perseverance. The urgency subscale consists of 12 items, for example, When I am upset I often act without thinking. The Likert-scale was 1 Strongly Disagree, 2 Disagree Some, 3 Agree Some, and 4 Strongly Agree. High scorers on urgency are likely to engage in impulsive behaviors in order to alleviate negative emotions despite the long-term harmful consequences of those actions. The (lack of) premeditation subscale consists of 11 items, for example, I usually think carefully before doing anything. Low scorers are thoughtful and deliberative, whereas high scorers act on the spur of the moment and without regard for the consequences. The (lack of) perseverance facet has 10 items, for example, I finish what I start. Low scorers can remain focused on a task that may be boring or difficult. Characteristics of the distribution of these three impulsivity subscales are shown in Table 4.

The correlations between the five self-rated scales of cognitive control are shown in Table 5. All of the correlations are large except for those involving the premeditation subscale of impulsivity; however, those are still statistically significant. These scales appear to tap into the same general construct.

We first explored the relationship between AQ and the grand totals across all types of physical activities and exercise types. Figure 2 is a histogram showing how participants differed with respect to the total number of minutes engaged in physical activity per week. Inspection of Fig. 2 shows that the distribution is positively skewed with most participants clustered near the mean of 1635 min, but with a string of super fit individuals scattered past 5000 min per week. One should note that these seemingly large totals also include low and moderately intense activities performed at home, at work, and walking.

Figure 3 shows the log₁₀ of Total Time as a function of AQ score. The log transform compresses the positive skew and enhances the linear relationship between the variables. The correlation between the log₁₀ of Total Time and AQ was significant, r(199) =  − 0.315, p < 0.001. The correlation with log₁₀ of Total Energy and AQ was much the same, r(199) =  − 0.312, p < 0.001. Thus, in general higher AQ scores are associated with less physical activity.

Discretionary exercise is physical activity not associated with commuting, a requirement of your job, or routine house and yard work. Several sections of the CES cover different types of discretionary exercise. Perhaps the most fundamental category of single-person exercise consists of the nine activities shown in Table 10. Recall that for each activity we probe for frequency per week, duration per session, and intensity. The product of frequency and duration yields total time per week; and multiplying total time by intensity yields energy per week. The three base measures (frequency, duration, intensity) and the two composites (Total Time, Energy) can each be correlated with AQ scores and those are shown in Table 10 for the sub-category of Exercising. Generally speaking, each metric shows a significant correlation with the magnitude of physical activity decreasing as AQ increases. More specifically, each component correlates with AQ, but the composites are not better at predicting AQ; in fact, those correlations are numerically weaker. Comparing the three simple measures, it is noteworthy that greater autism traits are somewhat related to the frequency of single-person exercise, but more strongly and consistently related to the duration and intensity of the exercise.

With respect to the specific types of single-person exercise, Table 10 shows the nine activities in descending order of their popularity, that is the percentage of participants who indicated a frequency greater than zero. If a substantial number of participants indicate zero frequency (i.e., they never participate in the designated activity), and consequently have zero average duration and zero intensity, then this severely restricts the variability needed to support a strong correlation. In that regard, the significant correlation for surfing is most likely anomalous.

The background questionnaire included this item from our previous work. How often in a typical week do your exercise, work out, or participate in a sport? never. rarely, sometimes, quite often, very often. This item seems most related to the Exercising activity from Table 10 and indeed the correlations between the old item with Exercising are: r(200) =  + 0.683, p < 0.001 for frequency, r(200) =  + 0.446, p < 0.001, for duration, and r(200) =  + 0.418, p < 0.001 for intensity. For the present dataset the correlation between the old item and AQ scores is r(198) =  − 0.327, p < 0.001. Thus, the CES validates that our old item may be a good single-item probe when brevity is needed.

We were also interested in the reasons which motivated or prevented participants from engaging in single-person exercise. The number of reasons for motivation and prevention were tallied to determine a total score for both positive and negative reasons checked. Figure 1 shows the mean number for each condition formed by partitioning the AQ scores into three groups: Low (the low 40 AQ scores), Intermediate (the middle 141 scores), and High (the high 20 scores). This follows Stevenson and Hart’s recommendation to use three autism trait groups. A 3 × 2 mixed ANOVA yielded a significant Group × Type of Reason interaction, F(2, 198) = 4.15, p = 0.017, partial η² = 0.040. The group with high AQ tendencies checks the fewest reasons why they are motivated to do the single-person exercise that they report doing, but the most reasons that prevent them from doing more. This suggests that increasing autism tendencies are associated with perceiving fewer benefits of exercise and greater costs (Fig. 4).

We also asked if participants preferred to do single-person activities alone, with others, or had no preference. As shown in Table 11, the group with high AQ scores did not show a systematic preference and their pattern of preference was identical to the large group with intermediate AQ scores. This supports a hypothesis that deficits or challenges with social communication are not contributing to the negative correlations between AQ scores and exercise. However, examination of the specific reasons given to justify their participation in single-person exercise suggest otherwise.

Table 12 shows the percentage of participants in each AQ group who checked specific reasons that motivated them to do the physical activities they do (top panel) and the percentage who checked specific reasons that prevented them from doing more of these physical activities. It is important to note that the majority of those in the very high AQ group indicate that they exercise because it is good for their physical and mental health and that they do enjoy it. That said, there are potentially important differences in comparison to the other two groups starting with the contrast that a smaller percentage acknowledge the benefits to physical and mental health. Furthermore, and in contrast to the answers reported in Table 11, a smaller percentage of the high AQ group indicate that they enjoy the social experience and a larger percentage appreciate that they can do these activities alone.

With respect to the reasons that prevent them from doing more of these physical activities, the most prevalent is lack of time; this is true across the range of AQ scores. This is not surprising given that most of the participants are full-time students and a substantial percentage also work. Perhaps the most striking contrast across the AQ groups was that the very high AQ group reports a higher incidence of the negative aftereffects of exercise making them feel physically and mentally tired, especially the latter. This is even more surprising if viewed in the context reported earlier that higher AQ scores are associated with shorter durations and lower levels of intensity. Finally, in comparison to the other two groups, the very high AQ group is more likely to report that they do not enjoy these activities or lack the skill or ability. These reasons may resonate with each other.

Participation in team sports is likely to be related to AQ scores because team sports place a premium on EF, social interaction, and communication. Paap and Greenberg (2013) introduced this probe: “Team sports often involve dividing your attention between a ball, a goal, your opponents, and your teammates. Do you excel at these sports?” In a series of regression analyses they reported that it predicted three measures of EF: the flanker effect (inhibitory control), switching costs, and flanker Global RT (monitoring). The significant regression coefficients (betas) were − 0.21 or − 0.22. This item is not part of the CES but was included in the general background questionnaire. For this dataset the team sports ability item did not significantly correlate with any of the laboratory measures of EF, but it did significantly correlate with the self-rated scales of cognitive control: r(198) =  − 0.221, p = 0.002 for BDEFS and r(198) =  − 0.183, p = 0.010 for BSCS. More important for present purposes, the team-sports ability item significantly correlates with AQ scores in the current dataset, r(198) =  − 0.337, p < 0.001.

Table 13 shows the simple measures of frequency, duration, and intensity and the composite measures of total time and energy for eight team sports included in the CES. As shown in the percent participation column, the majority of our student participants do not or no longer play team sports. This will limit the capacity of these measures to predict AQ scores. Nonetheless, AQ is negatively correlated with participation in football, baseball, hockey, and basketball. Consistent with the patterns observed for single-person activities, correlations tend to be stronger for duration and intensity compared to frequency, total time, or energy. Once again, the most striking contrast between the groups with respect to specific reasons (see Table 14) is the fairly substantial proportion (40%) of very high AQ individuals who indicate that playing team sports leaves them mentally tired.

The total number of reasons motivating participation and the total number preventing participation in team sports was tallied for each participant and the mean number of reasons of each type for each of the three AQ groups are shown in Fig. 5. Again, there is a significant Group × Reason Type interaction, F(2, 198) = 3.745, p = 0.025, partial η² = 0.036. The pattern is the same as for the single-person activities: as AQ tendencies increase, fewer positive reasons are checked, but more reasons preventing participation are checked.

Of the five sports shown in Table 15, bowling was the only one that the majority of our sample did at least a few times per year, although 42% played table tennis. As shown in Table 15 the frequency of play never significantly correlated with AQ scores for any of these sports, but duration did for bowling and table tennis and the largest negative correlation was observed between table-tennis intensity and AQ scores. In brief summary, these sports follow the trend that the duration and intensity of play decrease as AQ scores increase, but the strength of the relationships tend to be weaker than in other categories of activity.

The CES treats boxing (12%), wrestling (6%) and any form of martial arts (8%) as a separate category. As indicated parenthetically only a small percentage of our sample participated in these sports and, consequently, it is not surprising that the AQ scores did not significantly correlate with any measure for any of the three sports.

In contrast, 64% of the sample dances and as shown in Table 16 significant negative correlations with AQ were obtained for the measures of frequency, duration, total time, and energy. In contrast to many other activities the correlation between AQ scores and intensity was near zero. For this set of activities, total time is the best predictor of AQ scores.

Five performance-based measures of EF were derived from three different tasks. Only one cross-task correlation (see Table 3) was significant: that between the spatial Stroop effect and switching tasks (r =  + 0.184). Given that Stroop effects are purported to reflect inhibitory control and switch costs are frequently assumed to be caused, in part, by inhibition, this is consistent with some convergent validity between measures of inhibition. The significant correlations between the slopes for target-present and target-absent trials in the visual-search task and between switching and mixing costs in the color–shape task are both within-task correlations and it is likely that they would correlate even if they do not isolate the same component of EF. Although the current design did not include many tests of convergent validity between measures of the same component of EF, there is a distressing lack of convergent validity in the EF literature, especially for inhibitory control (Paap and Sawi 2014; Rey-Mermet et al. 2018; Paap et al. 2020a).

In contrast to the correlations with self-report measures of EF, there were no significant correlations (see Table 2) between any of the performance-based measures of EF and AQ (see Table 2). Given the lack of convergent validity between the performance-based EF measures, this is not surprising. It is also consistent with the Demetriou et al. (2018) meta-analysis of 235 comparisons between groups diagnosed with autism and control groups in that the effect for performance-based measures of EF was only g = 0.48 compared to g = 1.84 for self-ratings (mostly based on BRIEF). The 50 studies reviewed by Stevenson and Hart showed that AQ scores often predict task performance, but only a half dozen of these studies used any of the tasks that dominate the cognitive control–control literature (e.g., flanker, Stroop, Simon, cued task switching, N-back, short-term-memory span, operation span, etc.) and these specific studies uniformly showed no differences between the ASD and control groups. For the most part, there is no compelling evidence for a relationship between laboratory performance-based measures of EF and ASD. In the absence of the studies using self-rating measures of EF, one might conclude that the executive dysfunction hypothesis was false. A caveat is that the entire set of accuracy-based tests included in the Demetriou et al.’s meta-analyses did yield an effect size of g = 0.48 which guidelines characterize as moderate in size. Having made this allowance, it is also fair to note that Demetriou et al. found that “Only a very limited number of measures achieved the criterion of clinical sensitivity….” and that “The majority of the measures reaching clinical sensitivity were based on the BFRIEF questionnaire” p. 4.