Autistic traits are associated with lower perceived executive function but not poorer executive function task performance in the general population: complementary meta-analytic evidence

Abstract

Background

Autistic individuals generally exhibit real-world executive function (EF) difficulties and perform poorly on EF tasks. However, while autistic traits are distributed continuously throughout the general population, the relationships between autistic traits and EF among nonclinical individuals remain unclear. Here, we conducted complementary meta-analyses to clarify the relationships between autistic traits and various aspects of EF in the general population.

Methods

A literature search was conducted using PubMed, PsycINFO, and Web of Science on July 11, 2025. After screening by two independent reviewers, 39 articles that reported 40 studies were included. These studies either compared EF between groups with high and low autistic traits, based on a cutoff, or investigated the relationships between continuous measures of autistic traits and EF.

Results

We found significant negative associations between autistic traits and EF among nonclinical individuals across EF processes. Notably, these relationships were observed only when EFs were measured using questionnaires rather than behavioral tasks. Specifically, random-effects and robust Bayesian meta-analyses revealed significant, strong correlations between higher autistic traits and poorer ratings on EF questionnaires, with primarily substantial evidence supporting the presence than absence of relationships. In contrast, the meta-analyses indicated nonsignificant, very weak correlations between higher autistic traits and poorer performances on EF tasks, with primarily substantial evidence supporting the absence than presence of relationships.

Limitations

These findings were mainly based on self-reported measures of autistic traits in adults and derived from single studies without follow up or replication.

Conclusions

Autistic traits are associated with lower perceived real-world EF behavior rather than poorer EF task performance in the general population. These findings underscore the importance of paying closer attention to addressing the concerns of individuals with high autistic traits and their parents regarding their own and their children’s EF behavior. Based on the available evidence, we construct a picture of the relationships between autistic traits and EF across the trait’s continuum.

Registration

This study was preregistered at https://osf.io/zncv3.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13229-025-00680-2.

Background

Individuals diagnosed with autism spectrum disorders (ASD) experience social communication difficulties and exhibit restricted and stereotyped behavioral patterns [1]. According to the executive function (EF) theory of autism [2], difficulties with EFs, which refer to a set of interrelated higher-order control abilities essential for goal attainment [3, 4], have been postulated as an explanation for autistic behaviors. According to the neurodevelopmental perspective of EF [5], EF comprises top-down control processes activated in emotionally neutral and emotionally charged contexts (i.e., cool and hot EFs, respectively). Cool EF has been shown to comprise at least three basic processes, including shifting (switching attention between tasks or cognitive sets), inhibition (suppressing prepotent responses and resisting distractions), and working memory updating (retaining task-relevant information in the mind while discarding irrelevant information) [4]. They are commonly assessed using subjective self-report or informant-report measures and objective performance-based measures, which show, at best, modest convergence [6] (see [7] for a discussion of the convergent validity of EF measures). In addition, hot EF involves the integration of cognitive, emotional, and motivational processes necessary to achieve optimal outcomes, and is commonly assessed using affective decision-making tasks [5].

Supporting the EF theory, recent meta-analytic studies focusing on cool EFs have found that individuals diagnosed with ASD exhibit poorer EFs compared to those with typical development, regardless of whether questionnaires or tasks are used to measure them [8–11]. However, group differences are greater when using (parent-report) questionnaires, such as the Behavior Rating Inventory of Executive Function (BRIEF), compared to behavioral tasks, such as the Stroop test [8]. Because parents play a key role in informing both their children’s autism diagnoses and real-world EF behaviors, this phenomenon may be due to common-method variance [12] or subjective biases affecting parent-report measures [13]. The effect sizes are generally comparable across different EF processes.

Although autism has traditionally been viewed as a categorical clinical condition, it is increasingly clear that autistic traits are distributed throughout the general population and exist on a continuum [14]. Autistic traits are most commonly measured using self-report questionnaires, particularly the Autism Spectrum Quotient [14, 15]. However, they can also be assessed using parent-report measures, such as the Social Responsiveness Scale [16]. From this dimensional perspective, individuals diagnosed with ASD can be seen as falling at the upper end of the continuum, whereas those without such a diagnosis primarily occupy the lower end. Nevertheless, there are considerable variations in the extent of autistic traits among those without a diagnosis that may have meaningful relationships with various areas of functioning, including perceived mental well-being and quality of life [17, 18]. Nonclinical individuals generally receive less attention than those with a diagnosis. However, if they experience difficulties, it is important to recognize them and provide timely and appropriate support.

Executive functions are crucial for various outcomes, including those related to education [19] and health [20]. Therefore, accurately understanding the relationships between autistic traits and EF is essential for determining whether and what kind of support may be needed for individuals who exhibit higher autistic traits in order to enhance their functioning. While one might assume a linear relationship between autistic traits and executive functioning across the continuum of traits, where nonclinical individuals with increased autistic traits exhibit an intermediate profile between those with lower autistic traits and those with an ASD diagnosis, this assumption may not hold, as there is preliminary evidence that these individuals can be qualitatively different from those with a diagnosis. For example, they do not experience disability [21], and the correlation between some temperament traits (e.g., shyness) and autistic traits is weaker among nonautistic individuals compared to autistic ones [22].

Several studies have investigated the relationships between autistic traits and EF among nonclinical individuals, yielding mixed findings. Some studies have shown that typically developing adults with higher autistic traits endorse more difficulties with some EFs in their daily lives [23, 24] and perform more poorly on EF tasks compared to those with lower traits [25, 26]. In contrast, other studies have not found a significant relationship between autistic traits and EF task performance [27]. Additionally, a positive association between autistic traits and working memory task performance has been reported among adults [28], contradicting other studies [26]. Thus, the associations between autistic traits and EF, as well as the roles of the measurement type and specific EF process in these relationships, remain largely unclear.

Meta-analysis is a statistical method that combines the results of multiple studies to yield a single aggregated effect size [29]. Within meta-analysis, subgroup analysis and meta-regression can provide insights into the factors influencing an outcome. Traditionally, Meta-analysis is conducted using the frequentist approach, specifically the fixed-effects or random-effects method [30]. The latter is more common, as it assumes that the observed estimates vary across studies due to genuine differences in study and sample characteristics. A significant result allows a researcher to reject the null hypothesis, while a nonsignificant result remains inconclusive. Recently, there has been growing interest in adopting the Bayesian approach to supplement the frequentist approach [31]. Bayesian analysis incorporates prior information into the analysis and assigns probabilities to the null and alternative hypotheses, enabling the researcher to evaluate the strength of evidence in favor of one hypothesis over the other [32, 33]. The robust Bayesian meta-analysis method has been developed to facilitate such evaluation while adjusting for publication bias (see [11, 34, 35] for its application in autism research).

To address the knowledge gap regarding the level of executive functioning across the nonclinical spectrum of autistic traits, we conducted both random-effects meta-analyses and robust Bayesian meta-analyses to evaluate the strength of the relationships between autistic traits and EF in the general population. Various factors found to affect the impact of ASD on EF [8], including measurement type, were investigated. We hypothesized that autistic traits would be negatively associated with the levels of EFs, regardless of the specific EF process involved. Based on the findings of Demetriou et al. [8] and the potential role of common-method variance and subjective biases affecting self- and informant-report measures, we also anticipated a significant moderating effect of the type of measures, expecting the relationship to be stronger when EFs were assessed using questionnaires compared to tasks.

Methods

Search strategy

This study was preregistered (https://osf.io/zncv3). The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines were used throughout the process [36]. A comprehensive literature search was conducted to identify studies examining the relationship between autistic traits and executive function (EF) in healthy individuals. Multiple databases, including PubMed, PsycINFO, and Web of Science, were used to ensure broad coverage of relevant literature. The last search was performed on July 11, 2025, using the following terms: “(autism quotient or autis* spectrum quotient or autistic trait* or autistic feature* or autistic characteristic* or broad autism phenotype) and (executive function or cognitive control or updat* or working memory or shift* or switch* or flexibility or inhibit* or interference or plan* or reasoning or concept formation or fluency or monitor* or decision making or gambling or reversal learning or discounting or gratification)”. Because this is a systematic review and meta-analytic study, ethics approval was not applicable.

Journal articles were selected across search engines. The inclusion criteria were as follows: (1) studies that investigated the relationships between autistic traits and EF, or those that compared high and low autistic trait groups on EF, (2) studies that included individuals without any mental, neurological, or sleep-related disorders known to or reported by the researchers, (3) original research studies (e.g., not review articles, meta-analyses, or study protocols), (4) peer-reviewed journal articles, and (5) articles written in English.

The search results were combined to create a study list. After removing duplicates, two independent reviewers (CTYL and TA/SL) performed the initial screening and eligibility assessment. Titles and abstracts were screened to identify articles based on the inclusion criteria. Full-text articles were then evaluated using the same criteria to assess eligibility. Any disagreements between reviewers were resolved through consensus. Based on the results following both title and abstract screening and full-text evaluation, the initial agreement between reviewers was 99.8%. Cohen’s Kappa was 0.84, indicating almost perfect agreement [37]. Discrepancies were resolved by the first author (MKY).

Data extraction and coding

For articles included in the final review, various data were extracted and coded by at least one reviewer (MKY, CTYL, and JB) and entered into an Excel spreadsheet. Double coding was performed for all measures of autistic traits and EF. Publication details included the author(s) and year of publication. Sample characteristics included sample size and demographic information (e.g., age when EF assessment took place). Processes of EF included global EF, inhibition, working memory, shifting, emotional control, initiation, monitoring, planning, verbal fluency, reasoning, and affective decision-making (simply referred to as decision-making hereafter). Measures of autistic traits and the separation of high and low trait groups were noted. Measures of EF were classified into two types, questionnaires and tasks. Effect sizes represented the relationships between levels of autistic traits and EF. Because autistic traits in the general population have been considered as a dimension rather than in categories [14, 15]. Pearson’s r was employed, and other effect sizes (e.g., Cohen’s d) were converted to this metric [38].

Meta-analyses

For each EF process, a random-effects meta-analysis was conducted using the Hartung-Knapp-Sidik-Jonkman method [39–41] implemented in Meta-Analysis with JASP Version 0.19.3 (JASP Team, 2025). Fisher’s z-transformation was applied to normalize Pearson’s r before the analysis [42], and the pooled effect size was then converted back to Pearson’s r for easier interpretation. Pooled Pearson’s r values of 0.1, 0.3, and 0.5 represent small, medium, and large effect sizes, respectively. Additionally, the true heterogeneity in effect size across studies was assessed using the Q test [43]. The I² statistic, which describes the percentage of variation across studies due to heterogeneity rather than chance, was calculated [44]. I² values of 25%, 50%, and 75% indicate small, medium, and large heterogeneity levels, respectively.

For meta-analyses involving at least 10 studies, potential publication bias was evaluated using funnel plots. Egger’s regression test assessed the asymmetry of funnel plots based on the effect size and precision for each study [45]. A significant result indicates potential publication bias. Additionally, subgroup analyses were conducted to examine the effects of different types of EF measures and processes, as well as sample characteristics (e.g., age group).

Furthermore, robust Bayesian meta-analyses, also implemented in JASP Version 0.19.3, were conducted to complement the traditional random-effects meta-analyses and strengthen the conclusions drawn. This approach adopted the Bayesian framework to weigh evidence in favor of the alternative hypothesis (H₁: a negative correlation between autistic traits and EF exists) against the null hypothesis (H₀: such negative correlation is nonexistent) while addressing publication bias [33, 34]. Bayes factors (BF₁₀) are indices of relative evidence of one hypothesis over another. Based on Jeffreys [46], BF₁₀ of 0.1–0.33 and 0.33–1 indicate substantial and anecdotal evidence for H₀ against H₁, whereas BF₁₀ of 1–3, 3–10, and > 100 represent anecdotal, substantial, and extreme evidence for H₁ against H₀.

For each meta-analysis, each study was included only once to avoid overrepresentation of studies with multiple EF measures. In studies that utilized multiple EF measures, the representative measure chosen was the one most frequently used, most demanding (i.e., involving the highest number of processes), and most comprehensive one, following this order1: (1) when examining overall EF, global EF measures were chosen over specific EF measures; (2) tasks were preferred over questionnaires; (3) overall task performance or total scale score was selected over subtest or subscale scores; (4) the condition that placed the greatest demand was chosen over those that placed less demand; (5) accuracy was selected over reaction time, unless a ceiling effect was present (i.e., when the sum of the mean and one SD exceeded 100%). However, for subgroup analyses, studies that included measures of different EF processes were allowed to be included in multiple subgroups.

Risk of bias assessment

A risk of bias assessment was performed for each study independently by two coders (CTYL and MKY/HCWC) using the critical appraisal checklist for case-control studies2 [47]. The checklist consisted of the following 10 questions, which were modified to suit this study: (1) Were the high and low autistic trait groups comparable other than the difference in their levels of autistic traits? (2) Were high and low trait individuals matched appropriately and recruited from the same population? (3) Were the same criteria used for identification of high and low trait individuals? (4) Was the level of autistic traits measured in a standard, valid and reliable way? (5) Was the level of autistic traits measured in the same way for high and low trait individuals? (6) Were confounding factors identified? (7) Were strategies to deal with confounding factors stated? (8) Were EF outcomes assessed in a standard, valid and reliable way for high and low trait groups? (9) Was the level of autistic traits long enough to be meaningful, as verified by measurements taken at more than one time point? (10) Was appropriate statistical analysis used? Any discrepancies were resolved through consensus between the reviewers.

Results

Screening results

The flow of the study screening is illustrated in Fig. 1. Initially, 2,699, 2,479, and 3,152 journal articles were identified in PubMed, PsycINFO, and Web of Science, respectively. After removing 2,670 duplicated articles and determining eligibility based on titles and abstracts, 53 articles underwent full-text screening. Fourteen articles were excluded because they did not investigate the relationships between autistic traits and EF (k = 7), included individuals with mental, neurological, or sleep-related disorders known to or reported by the researchers (k = 6), or compared parents of autistic and TD children who did not significantly differ in the AQ total score (k = 1). Therefore, 39 articles were finally included, all of which were published since 2006. One article reported two studies [48]. Thus, there were 40 studies in total [23–28, 48–80].

Fig. 1.

PRISMA flow Diagram. EF, executive function. *One article reported two separate studies (Camodeca, 2023a)

Major study characteristics

Table 1 presents the major characteristics of the 40 studies that investigated the relationship between autistic traits and EF among nonclinical children and adolescents (k = 10) and nonclinical adults (k = 30). Twenty-five studies used the AQ to measure autistic traits. All but two studies employing the self-report AQ used the 50-item adult version; the two remaining studies used the 28-item adult version and the 22-item adolescent version. In addition, the two child studies employing the parent-report AQ used the 28-item and 50-item versions.

Table 1.

Summary of the studies included in the Meta-Analyses

Study	Year	Total Sample Size	Years of Age (M, SD, range)	Autistic Trait Measure (Respondent, Number of Items)	Autistic Trait Groups (Criterion)	EF Process	EF Task	EF Measure	Type of EF measure	Pearson’s r
Kunihira et al.	2006	96	19.3, 2.5, 16–54	AQ (self, 50)	Yes (AQ total: top and bottom 25%)	Shifting	WCST	Categories achieved	T	0.06
								Nelson’s perseverative errors	T	− 0.08#*
								Failure to maintain sets	T	− 0.13
Christ et al.	2010	94	18.4, /, 17–21	SRS (self, 11)	Yes (SRS raw total: ≤ 6 and ≥ 15)	Global EF	BRIEF-Adult	GEC score	Q (self)	− 0.42#*
								Behavior Regulation Index	Q (self)	− 0.46
								Metacognition Index	Q (self)	− 0.35
						Inhibition	BRIEF-Adult	Inhibit score	Q (self)	− 0.09*
						Shifting	BRIEF-Adult	Shift score	Q (self)	− 0.59*
						Emotional control	BRIEF-Adult	Emotional Control score	Q (self)	− 0.46*
						Monitoring	BRIEF-Adult	Self-Monitor score	Q (self)	− 0.30
						Initiation	BRIEF-Adult	Initiate score	Q (self)	− 0.41*
						Working memory	BRIEF-Adult	Working Memory score	Q (self)	− 0.26*
						Planning	BRIEF-Adult	Plan/Organize score	Q (self)	− 0.36*
						Monitoring	BRIEF-Adult	Task Monitor score	Q (self)	− 0.40*
						Planning	BRIEF-Adult	Organization of Materials	Q (self)	− 0.10
Bayliss and Kritikos	2011	179	20.2, /, /	AQ (self, 50)	Yes (AQ total: < and ≥ 15)	Inhibition	Letter Discrimination Task	RT interference effect (set size 2)	T	− 0.09
								RT interference effect (set size 4)	T	0.17#*
Elsabbagh et al.	2011	22	0.8, 0.1, /	ADOS-Generic at 3– 4 years (parent, /)	No	Inhibition	Freeze-frame	Proportion of look to the distractors in boring trials	T	− 0.64
								Proportion of look to the distractors in interesting trials	T	− 0.11#*
Fugard et al.	2011	85	23, 5.1, 17–49	AQ (self, 50)	No	Reasoning	RCPM	Total score	T	0.05#*
Ridley et al.	2011	44	27.4, /, 20–52	AQ (self, 50)	No	Shifting	Category Switching	Total score	T	− 0.44
							Trail Making	Total score	T	− 0.03*
						Inhibition	Color-Word Interference Task	Total score	T	− 0.04#*
						Verbal fluency	Letter Fluency	Total score	T	− 0.06*
							Category Fluency	Total score	T	− 0.21
Takahashi and Gyoba	2012	27	22.3, 2.7, /	AQ (self, 50)	Yes (AQ total: < and ≥ 20; median split)	Working memory	Change Detection Task	Capacity	T	0.36#*
								Capacity (complex configurations)	T	0.61
Maes et al.	2013	37	20.6, /, /	AQ (self, 50)	No	Verbal fluency	Letter Fluency	Total score	T	− 0.18*
						Inhibition	Random Number Generation	Phi 2 score	T	0.04
								Adjacency score	T	0.10#*
						Working memory	Random Number Generation	Coupon score	T	− 0.02*
						Reasoning	RCPM	Total score	T	0.25*
Richmond et al.	2013	104	20.3, /, /	AQ (self, 50)	No	Working memory	Visual working memory	Score	T	0.27#*
Gökçen et al.	2014	69	23.2, /, 18–46	AQ (self, 50)	Yes (AQ total: ≤ 13 and ≥ 20)	Shifting	WCST	Shifting efficiency score	T	− 0.43#*
Gökçen et al.	2016	107	18.1, 1.2, 16–22	AQ (self, 50)	No	Shifting	WCST	Shift score	T	− 0.22
								Efficiency score	T	− 0.13*
						Inhibition	Go/No-Go	Commission errors	T	− 0.25#*
						Planning	Tower of London (Freiburg Version)	Total score	T	− 0.013*
Shi et al.	2017	864	/, /, /	AQ (self, 50)	Yes (AQ total: < and ≥ 26)	Global EF	Dysexecutive Questionnaire	Total score	Q (self)	− 0.31#*
Albein-Urios et al.	2018	69	22.3, 2.7, 18–33	AQ (self, 50)	No	Shifting	BRIEF-Adult	Shifting score	Q (self)	− 0.48
							D-KEFS Trail Making	Number-letter switching completion time	T	− 0.17#*
								Sequencing errors	T	0.02
								Set loss errors	T	− 0.02
Camodeca et al.	2018	182	20.6, 5.5, /	BAPQ (self, 36)	Yes (sex-specific cutoffs based on the Sasson criteria)	Verbal Fluency	Letter Fluency	Total Score	T	0.04#*
							Category Fluency	Total Score	T	− 0.07
Ferraro et al.	2018	274	21.2, 4.1, 18–50	AQ (self, 50)	Yes (AQ total: < or ≥ 16)	Global EF	EF Index	Total score	Q (self)	− 0.10#*
						Initiation	EF Index	Motivational drive score	Q (self)	− 0.12*
						Emotional control	EF Index	Impulse control score	Q (self)	0.01*
						Others	EF Index	Empathy score	Q (self)	− 0.03
						Planning	EF Index	Organization score	Q (self)	− 0.14
						Planning	EF Index	Strategic planning score	Q (self)	0.00*
Hamilton et al.	2018	76	8.4, /, /	Children EQ & SQ Questionnaire (parent, 18 & 20)	No	Working memory	Size Just Noticeable Difference	Correct score	T	0.53#*
Pirrone et al.	2018	37	21, /, 18–23	AQ (self, 50)	No	Inhibition	Attention-cuing Task	Accuracy in cue-congruent condition	T	− 0.19
								RT in cue-congruent condition	T	0.29
								Accuracy in cue-incongruent condition	T	0.12#*
								RT in cue-incongruent condition	T	− 0.05
Stewart et al.	2018	40	73.1, 7.8, 60–91	BAPQ (self, 36)	Yes (cutoffs based on the Hurley criteria)	Shifting	D-KEFS Trail Making	Switching minus motor speed	T	− 0.21*
						Verbal Fluency	D-KEFS Letter Fluency	Total Score	T	− 0.03*
						Working memory	WMS-IV Letter-Number Sequencing	Total Score	T	− 0.41*
							WMS-IV Digit Span	Score	T	− 0.43
						Global EF	BRIEF-Adult	Behavior Regulation Index T-score	Q (self)	− 0.58#*
								Metacognition Index T-score	Q (self)	− 0.55
Dai et al.	2019	413	7.7, /, 6–9	SRS (parent, 65)	Yes (SRS raw total: < and ≥ 56.5)	Shifting	WCST	Categories completed	T	− 0.09
								Failure to maintain sets	T	0.05
								Nonperseverative errors	T	− 0.05
								Shifting efficiency score	T	− 0.03#*
Hyseni et al.	2019	960	7.9, 1, 6–10.7	SRS (parent, 18)	No	Global EF	NEPSY-II	Attention and EF score	T	− 0.26#*
Lewton et al.	2019	189	27.3, 9.0, 18–62	AQ (self, 50)	No	Reasoning	RCPM	Total score	T	0.00*#
Zhang et al.	2020	220	8.2, /, 6–11	AQ-Child (parent, 35)	Yes (AQ total: top 15% and bottom 85%)	Working memory	Spatial working memory test	Total errors	T	− 0.14#*
Chen et al.	2021	44	18.7, /, /	AQ (self, 50)	Yes (AQ total: top and bottom 10%)	Affective decision-making	Iowa Gambling Task	Net score	T	− 0.30#*
								Net score (block 5)	T	− 0.92
								Net score (block 6)	T	− 0.86
Huang et al.	2021	449	20, /, /	AQ (self, 50)	No	Inhibition	Attentional Network Test (executive control network)	RT interference score	T	− 0.03#*
Mason et al.	2021	200	19 (median), /, 18–45	AQ (self, 50)	No	Global EF	Barkley Deficits in Executive Functioning Scale	Overall score	Q (self)	− 0.42#*
						Inhibition	Spatial Stroop	RT interference effect	T	− 0.10*
						Shifting	Color–Shape Switching Task	RT switching cost	T	0.13*
								RT mixing cost	T	− 0.03
						Reasoning	Raven’s Advanced Progressive Matrices	Score	T	0.08*
Tsai et al.	2021	6,832	11.3, 1.8, 8–14	SRS (parent, 65)	No	Global EF	BRIEF-Parent	GEC score	Q (parent)	− 0.43#*
								Behavior Regulation Index	Q (parent)	− 0.44
								Metacognition Index	Q (parent)	0.44
Robertson et al.	2021	170	19.0, 1.3, /	AQ (self, 50)	No	Global EF	Adult Temperament Questionnaire	Effortful Control score	Q (self)	− 0.48#*
Chee et al.	2022	170	20.5, 3.4, 17–44	AQ (self, 28)	No	Global EF	ADEXI	Total score	Q (self)	− 0.07#*
Gong et al.	2022	454	18.4, 0.9, 17.5–19.3	AQ (self, 50)	No	Global EF	Dysexecutive Questionnaire	Total score	Q (self)	− 0.33#*
Camodeca (Study 1)	2023a	104	41.6, 8.0, /	BAPQ (self, 36)	Yes (sex-specific cutoffs based on the Sasson criteria)	Global EF	BRIEF-Adult	GEC score	Q (self)	− 0.62#*
							BRIEF-Adult	Behavior Regulation Index	Q (self)	− 0.61
							BRIEF-Adult	Metacognition Index	Q (self)	− 0.58
						Inhibition	BRIEF-Adult	Inhibit score	Q (self)	− 0.58*
						Shifting	BRIEF-Adult	Shift score	Q (self)	− 0.65*
						Emotional control	BRIEF-Adult	Emotional Control score	Q (self)	− 0.42*
						Monitoring	BRIEF-Adult	Self-Monitor score	Q (self)	− 0.61
						Initiation	BRIEF-Adult	Initiate score	Q (self)	− 0.59*
						Working memory	BRIEF-Adult	Working Memory score	Q (self)	− 0.57*
						Planning	BRIEF-Adult	Plan/Organize score	Q (self)	− 0.52*
						Monitoring	BRIEF-Adult	Task Monitoring score	Q (self)	− 0.49*
						Planning	BRIEF-Adult	Organization of Materials score	Q (self)	− 0.58
Camodeca (Study 2)	2023a	147	20.3, 4.7, /	BAPQ (self, 36)	Yes (sex-specific cutoffs based on the Sasson criteria)	Global EF	BRIEF-Adult	GEC score	Q (self)	− 0.33#*
							BRIEF-Adult	Behavior Regulation Index	Q (self)	− 0.26
							BRIEF-Adult	Metacognition Index	Q (self)	− 0.30
						Inhibition	BRIEF-Adult	Inhibit score	Q (self)	− 0.17*
						Shifting	BRIEF-Adult	Shift score	Q (self)	− 0.32*
						Emotional control	BRIEF-Adult	Emotional Control score	Q (self)	− 0.29*
						Monitoring	BRIEF-Adult	Self-Monitor score	Q (self)	− 0.14
						Initiation	BRIEF-Adult	Initiate score	Q (self)	− 0.30*
						Working memory	BRIEF-Adult	Working Memory score	Q (self)	− 0.22*
						Planning	BRIEF-Adult	Plan/Organize score	Q (self)	− 0.31*
						Monitoring	BRIEF-Adult	Task Monitoring score	Q (self)	− 0.22*
						Planning	BRIEF-Adult	Organization of Materials score	Q (self)	− 0.11
Camodeca	2023b	146	33.4, 10.9, /	BAPQ (self, 36)	Yes (sex-specific cutoffs based on the Sasson criteria)	Global EF	BRIEF-Adult	GEC score	Q (self)	− 0.50#*
							BRIEF-Adult	Behavior Regulation Index	Q (self)	− 0.40
							BRIEF-Adult	Metacognition Index	Q (self)	− 0.47
						Inhibition	BRIEF-Adult	Inhibit score	Q (self)	− 0.31*
						Shifting	BRIEF-Adult	Shift score	Q (self)	− 0.56*
						Emotional control	BRIEF-Adult	Emotional Control score	Q (self)	− 0.34*
						Monitoring	BRIEF-Adult	Self-Monitor score	Q (self)	− 0.40
						Initiation	BRIEF-Adult	Initiate score	Q (self)	− 0.49*
						Working memory	BRIEF-Adult	Working Memory score	Q (self)	− 0.40*
						Planning	BRIEF-Adult	Plan/Organize score	Q (self)	− 0.47*
						Monitoring	BRIEF-Adult	Task Monitoring score	Q (self)	− 0.47*
						Planning	BRIEF-Adult	Organization of Materials score	Q (self)	− 0.25
Davidson et al.	2023	144	19.7, /, 18.1–25.1	SRS (self, 65)	Yes (SRS total T-score: < and ≥ 60)	Global EF	BRIEF-Adult	GEC score	Q (self)	− 0.49#*
							BRIEF-Adult	Behavior Regulation Index	Q (self)	− 0.45
							BRIEF-Adult	Metacognition Index	Q (self)	− 0.49
						Inhibition	BRIEF-Adult	Inhibit score	Q (self)	− 0.39*
						Shifting	BRIEF-Adult	Shift score	Q (self)	− 0.38*
						Emotional control	BRIEF-Adult	Emotional Control score	Q (self)	− 0.36*
						Monitoring	BRIEF-Adult	Self-Monitor score	Q (self)	− 0.49
						Initiation	BRIEF-Adult	Initiate score	Q (self)	− 0.38*
						Working memory	BRIEF-Adult	Working Memory score	Q (self)	− 0.50*
						Planning	BRIEF-Adult	Plan/Organize score	Q (self)	− 0.49*
						Monitoring	BRIEF-Adult	Task Monitor score	Q (self)	− 0.45*
						Planning	BRIEF-Adult	Organization of Materials	Q (self)	− 0.33
Nicholls and Stewart	2023	144	22, 2.5, 28–33	AQ (self, 50)	No	Working memory	Visual Patterns Test	Low Semantic Task performance	T	0.20#*
								High Semantic Task performance	T	0.08
Stewart et al.	2023	12,020	61.3, /, 50–80	PROTECT AST screener questions (self, 5)	Yes (a “yes” for < and ≥ 2 childhood trait and 2 current trait items)	Working memory	Paired associate learning	Total score	T	− 0.10
							Digit span	Total score	T	− 0.15
							Self-ordered Search	Total score	T	0.15#*
						Reasoning	Verbal Reasoning Task	Total score	T	− 0.04*
Su et al.	2024	80	9.4, /, 6–12	AQ-Child (parent, 50)	Yes (AQ total: top and bottom 15%)	Inhibition	Opposite World Task	RT	T	− 0.09#*
						Working memory	Digit Span	Total score	T	− 0.08*
Zhou et al.	2024	60	19.1, /, 17–21	AQ (self, 50)	Yes (AQ total: top and bottom 10%)	Shifting	Affective Task Switching	Overall inverse efficiency	T	− 0.94
								Inverse efficiency switch cost on emotion Task	T	− 0.35#*
Chen et al.	2025	611	13.6, 1.1, 12–16	AQ-Adolescent (self, 22)	No	Global EF	BRIEF-Self Report	GEC score	Q (self)	− 0.28#
Hendry et al.	2025	124	3, /, /	SRS-2-Preschool (parent, 65)	No	Global EF	Complex EF	Factor score	T	− 0.20#
Wang et al.	2025	316	8.4, /, 6–12 (empathizing- and systemizing- dominant)	Children EQ & SQ Questionnaire (parent, 23 & 22)	Yes (larger and smaller differences in empathy and systemizing quotients)	Global EF	BRIEF-Parent	GEC score	Q (parent)	− 0.41#*
							BRIEF-Parent	Behavioral Regulation Index	Q (parent)	− 0.17
							BRIEF-Parent	Metacognition Index	Q (parent)	− 0.22
						Inhibition	BRIEF-Parent	Inhibit score	Q (parent)	− 0.36*
						Shifting	BRIEF-Parent	Shift score	Q (parent)	− 0.10*
						Emotional control	BRIEF-Parent	Emotional Control score	Q (parent)	0.14*
						Initiation	BRIEF-Parent	Initiate score	Q (parent)	− 0.13*
						Working Memory	BRIEF-Parent	Working Memory score	Q (parent)	− 0.28*
						Planning	BRIEF-Parent	Plan/Organize score	Q (parent)	− 0.18*
						Monitoring	BRIEF-Parent	Monitor score	Q (parent)	− 0.17*
						Planning	BRIEF-Parent	Organization of Materials	Q (parent)	− 0.17
						Working memory	Wechsler Intelligence Scale for Children-Fourth Edition	Working Memory Index	Q (parent)	0.05

ADEXI, The Adult Executive Functioning Inventory; ADOS-G, Autism Diagnostic Observation Schedule–Generic; AQ, Autism Spectrum Quotient; BAPQ, Broad Autism Phenotype Questionnaire; BRIEF, Behavior Rating Inventory of Executive Function; EF, executive function; GEC, Global Executive Composite; MI, Metacognition Index; Q, Questionnaire; RCPM, Raven Colored Progressive Matrices; RT, reaction time; SRS, Social Responsiveness Scale; T, Task; WMS, Wechsler Memory Scale. #Representative measure of overall EF. *Representative measure of a specific EF process

Of the 40 included studies, 20 compared EF between groups with high and low autistic traits based on a cutoff, measuring EF with questionnaires in nine studies and tasks in 13 studies. The cutoff used varied greatly across studies; only six of these 20 studies employed a polarized approach, selecting individuals from the two extremes and excluding those in the middle. The other 20 studies investigated the relationship between continuous measures of autistic traits and EF, with EF measured using questionnaires in seven studies and tasks in 15 studies. Altogether, four studies used both questionnaires and tasks [58, 61, 67, 80]. In all studies where both autistic traits and EF questionnaires were used, the same individual completed both measures.

Three studies explicitly stated that they recruited parents or siblings of children diagnosed with ASD [48, 51, 79]. Fourteen studies examined more than one EF process. Among all the EF tasks used, random number generation was the only one from which multiple measures were extracted3. All but one study measured autistic traits and EF at the same time point. This exceptional study assessed EF in children under 1 year (M = 0.8 years) and examined its relationship with their autistic traits, measured using the ADOS-Generic, at 3 to 4 years old [51].

Meta-analysis results

Random-effects meta-analyses and robust Bayesian meta-analyses were performed to evaluate the relationships between autistic traits and EF. Overall EF was first examined by including all studies, each contributing one representative measure to present the overrepresentation of studies with multiple measures (see Methods for selection criteria). Also, subgroup and meta-regression analyses were conducted to identify factors that might moderate these relationships. Following this, the relationships between autistic traits and specific EF processes were analyzed, considering the significant moderators identified earlier.

Overall EF. The relationship between autistic traits and overall EF was evaluated based on all 40 studies, which included measures representing global EF (k = 17), inhibition (k = 8), working memory (k = 6), shifting (k = 5), and other EF processes, including reasoning (k = 2), decision-making (k = 1), and verbal fluency (k = 1). Autistic traits were measured using the AQ (k = 25), Broad Autism Phenotype Questionnaire (BAPQ; k = 5), SRS (k = 6), Children Empathy Quotient and Systemizing Quotient Questionnaires (k = 2), Autism Diagnostic Observation Schedule-Generic (ADOS-Generic; k = 1), and PROTECT AST screener questions (k = 1).

A random-effects meta-analysis was conducted on these studies, which included 26,403 individuals with mean ages of 0.8 to 73.1 years. We found a significant negative correlation between autistic traits and overall EF (Fig. 2), z’ = − 0.17 (r = − .16), p < .001, 95% CI [-0.25, − 0.08]. True heterogeneity in the effect sizes was significant, I² = 98%, p < .001. Egger’s regression test revealed no significant funnel plot asymmetry (Supplementary Fig. 1), t = 1.74, p = .081. In addition, a robust Bayesian meta-analysis revealed a BF₁₀ of 4.44, indicating substantial evidence supporting H₁ over H₀.

Fig. 2.

Forest plot of the random-effects meta-analysis for the relationship between autistic traits and overall executive function. ADEXI, The Adult Executive Functioning Inventory; ADOS-G, Autism Diagnostic Observation Schedule–Generic; AQ, Autism Spectrum Quotient; AT, autistic traits; BAPQ, Broad Autism Phenotype Questionnaire; BDEFS, Barkley Deficits in Executive Functioning Scale; BRIEF, Behavior Rating Inventory of Executive Function; D-KEFS, Delis-Kaplan Executive Function System; EF, executive function; EQ & SQ, The Children’s Empathy Quotient and Systemizing Quotient; NEPSY-II, A Developmental Neuropsychological Assessment, Second Edition; PROTECT, PROTECT AST screener questions; RCPM, Raven’s Colored Progressive Matrices; RAPM, Raven’s Advanced Progressive Matrices; RT, reaction time; SRS, Social Responsiveness Scale. The Fisher’s z-transformed correlation coefficients (z’) are displayed. The blue and red colors indicate questionnaire and task measures, respectively

Due to the large variation in age among the studies, a subgroup analysis was conducted to compare the effect sizes reported by studies of children/adolescents (k = 10) and by studies of adults (k = 30). The age effect was not significant, F(1, 38) = 0.04, p = .84.

To determine whether the type of measures played a role in explaining the heterogeneity among effect sizes, another subgroup analysis was conducted to compare studies that used EF questionnaires (k = 16) and tasks (k = 28)4. We found a significant result, F(1, 42) = 33.76, p < .001, indicating a greater effect when EF was measured using questionnaires compared to tasks. Therefore, meta-analyses were conducted for the two types of measures separately. The random-effects meta-analysis with questionnaire measures, involving a total of 10,635 participants, indicated a significant negative relationship between autistic traits and levels of EF, z’ = − 0.39 (r = − .37), p < .001, 95% CI [-0.47, − 0.31]. The robust Bayesian meta-analysis revealed a BF₁₀ of 100.86, indicating extreme evidence in favor of H₁ against H₀. In contrast, the random-effects meta-analysis based on task measures, involving 16,393 participants in total, revealed a nonsignificant relationship, z’ = − 0.03 (r = − .03), p = .45, 95% CI [-0.12, 0.06]. The robust Bayesian meta-analysis showed a BF₁₀ of 0.11, implying substantial evidence favoring H₀ over H₁. For both questionnaires, p = .48, and tasks, p = .35, there was no significant difference in the effect sizes reported by studies that separated high and low trait groups based on a cutoff compared to those that did not.

Working Memory. For working memory, a random-effects meta-analysis was conducted on 15 studies that involved 13,699 individuals with mean ages of 8.2 to 73.1 years. The measures included the BRIEF Working Memory score (k = 6), change detection task (k = 1), digit span (k = 1), letter-number sequencing (k = 1), random number generation (k = 1), self-ordered search (k = 1), size just noticeable difference (k = 1), spatial working memory test (k = 1), visual patterns test (k = 1), and visual working memory test (k = 1). Autistic traits were measured using the AQ (k = 6), BAPQ (k = 4), Children Empathy Quotient and Systemizing Quotient Questionnaires (k = 2), SRS (k = 2), and PROTECT AST screener questions (k = 1).

We found a nonsignificant negative correlation between autistic traits and working memory (Fig. 3), z’ = − 0.10 (r = − .10), p = .29, 95% CI [-0.28, 0.09]. True heterogeneity in the effect sizes of the relationships was significant, I² = 97%, p < .001. Egger’s regression test revealed no significant funnel plot asymmetry (Supplementary Fig. 2), t = 0.64, p = .52. In addition, a robust Bayesian meta-analysis revealed a BF₁₀ of 0.29, indicating substantial evidence in favor of H₀ against H₁.

Fig. 3.

Forest plot of the random-effects meta-analysis for the relationship between autistic traits and working memory. AQ, Autism Spectrum Quotient; AT, autistic traits; BAPQ, Broad Autism Phenotype Questionnaire; BRIEF, Behavior Rating Inventory of Executive Function; EF, executive function; EQ & SQ, The Children’s Empathy Quotient and Systemizing Quotient; PROTECT, PROTECT AST screener questions; SRS, Social Responsiveness Scale. The Fisher’s z-transformed correlation coefficients (z’) and 95% confidence intervals are displayed. The blue and red colors indicate questionnaire and task measures, respectively

Meta-analyses were conducted also for the questionnaire and task measures separately. The random-effects meta-analysis based on questionnaire measures (k = 6), involving a total of 951 participants, indicated a significant negative relationship between autistic traits and working memory, z’ = − 0.38 (r = − .36), p = .001, 95% CI [-0.53, − 0.23]. True heterogeneity in the effect sizes among studies was significant, I² = 77%, p < .001. The robust Bayesian meta-analysis revealed a BF₁₀ of 4.90, indicating substantial evidence in favor of H₁ against H₀. In contrast, the random-effects meta-analysis based on the task measures (k = 10), involving 13,064 participants in total, revealed a nonsignificant relationship, z’ = 0.10 (r = .10), p = .29, 95% CI [-0.10, 0.29]. There was significant true heterogeneity in the effect sizes, I² = 94%, p < .001. Egger’s regression test revealed no significant funnel plot asymmetry, t = − 0.37, p = .71. A robust Bayesian meta-analysis revealed a BF₁₀ of 0.44, indicating anecdotal evidence in favor of H₀ against H₁.

Inhibition. For inhibition, a random-effects meta-analysis was conducted on 15 studies, which involved a total of 2,106 individuals with mean ages of 0.8 to 41.6 years. The measures included the BRIEF Inhibit score (k = 6) and performance on various tasks, including the attention-cuing task (k = 1), the Attentional Network Test (k = 1), the Color-Word Interference Task (k = 1), the freeze-frame task (k = 1), Go/No-Go (k = 1) the letter discrimination task (k = 1), the Opposite World task (k = 1), random number generation (k = 1), and the spatial Stroop task (k = 1). Autistic traits were measured using the AQ (k = 8), BAPQ (k = 3), SRS (k = 2), and ADOS-Generic (k = 1), and Children Empathy Quotient and Systemizing Quotient Questionnaires (k = 1).

We found a significant negative correlation between autistic traits and inhibition (Fig. 4), z’ = − 0.16 (r = − .16), p = .011, 95% CI [-0.28, − 0.04]. True heterogeneity in the effect sizes of the relationships was significant, I² = 85%, p < .001. Egger’s regression test revealed nonsignificant funnel plot asymmetry (Supplementary Fig. 3), t = 1.64, p = .10. Additionally, a robust Bayesian meta-analysis revealed a BF₁₀ of 1.56, indicating anecdotal evidence in favor of H₁ against H₀.

Fig. 4.

Forest plot of the random-effects meta-analysis for the relationship between autistic traits and inhibition. ADOS-G, Autism Diagnostic Observation Schedule–Generic; AQ, Autism Spectrum Quotient; AT, autistic traits; BAPQ, Broad Autism Phenotype Questionnaire; BRIEF, Behavior Rating Inventory of Executive Function; EF, executive function; RT, reaction time; SRS, Social Responsiveness Scale. The Fisher’s z-transformed correlation coefficients (z’) and 95% confidence intervals are displayed. The blue and red colors indicate questionnaire and task measures, respectively

Meta-analyses were also conducted for the questionnaire and task measures separately. The random-effects meta-analysis based on questionnaire measures (k = 6), involving a total of 951 participants, indicated a significant negative relationship between autistic traits and inhibition, z’ = − 0.33 (r = − .32), p = .005, 95% CI [-0.50, − 0.15]. True heterogeneity in the effect sizes among studies was significant, I² = 83%, p < .001. The robust Bayesian meta-analysis revealed a BF₁₀ of 4.63, indicating substantial evidence in favor of H₁ against H₀. In contrast, the random-effects meta-analysis conducted on task measures (k = 9), involving 1,155 participants in total, revealed a nonsignificant correlation, z’ = − 0.03 (r = − .03), p = .51, 95% CI [-0.14, 0.07]. True heterogeneity in the effect sizes among studies was significant, I² = 55%, p = .039. A robust Bayesian meta-analysis revealed a BF₁₀ of 0.10, indicating substantial evidence in favor of H₀ against H₁.

Shifting. For shifting, a random-effects meta-analysis was conducted on 15 studies that involved a total of 2,049 individuals with mean ages of 7.7 to 73.1 years. The measures included the BRIEF Shifting score (k = 6) and performance on various tasks, including the Wisconsin Card Sorting Test (k = 4), Trail Making switching (k = 3), the color–shape switching task (k = 1), and the affective task switching task (k = 1). Autistic traits were measured using the AQ (k = 7), BAPQ (k = 4), SRS (k = 3), and Children Empathy Quotient and Systemizing Quotient Questionnaires (k = 1).

We found a significant negative correlation between autistic traits and shifting (Fig. 5), z’ = − 0.29 (r = − .28), p < .001, 95% CI [-0.42, − 0.15]. True heterogeneity in the effect sizes of the relationships was significant, I² = 90%, p < .001. Egger’s regression test revealed nonsignificant funnel plot asymmetry (Supplementary Fig. 4), t = 0.74, p = .46. Additionally, robust Bayesian meta-analysis revealed a BF₁₀ of 2.79, indicating anecdotal evidence in favor of H₁ against H₀.

Fig. 5.

Forest plot of the random-effects meta-analysis for the relationship between autistic traits and shifting. AQ, Autism Spectrum Quotient; AT, autistic traits; BAPQ, Broad Autism Phenotype Questionnaire; BRIEF, Behavior Rating Inventory of Executive Function; EF, executive function; RT, reaction time; SRS, Social Responsiveness Scale. The Fisher’s z-transformed correlation coefficients (z’) and 95% confidence intervals are displayed. The blue and red colors indicate questionnaire and task measures, respectively

Meta-analyses were conducted also for the questionnaire and task measures separately. The random-effects meta-analysis conducted on questionnaire measures (k = 7), involving a total of 1,020 participants, indicated a significant negative relationship between autistic traits and shifting, z’ = − 0.44 (r = − .41), p < .001, 95% CI [-0.62, − 0.26]. True heterogeneity in the effect sizes was not significant, I² = 88%, p < .001. Additionally, the robust Bayesian meta-analysis revealed a BF₁₀ of 1.59, indicating anecdotal evidence in favor of H₁ against H₀. In contrast, the random-effects meta-analysis based on task measures (k = 9), involving 1,098 participants in total, revealed a nonsignificant relationship, z’ = − 0.13 (r = − .13), p = .053, 95% CI [-0.27, − 0.00]. There was significant true heterogeneity in the effect sizes, p < .001, I² = 72%. In addition, a robust Bayesian meta-analysis revealed a BF₁₀ of 0.26, indicating substantial evidence favoring H₀ over H₁.

Other EF Processes. Several studies investigated the relationships between autistic traits and other EFs, including planning (k = 8)​, emotional control (k = 7), initiation (k = 7), monitoring (k = 6), reasoning (k = 5), verbal fluency (k = 4), and decision-making (k = 1)​​​​. Meta-analyses were conducted for all these processes, except for decision-making, which was investigated in only one study. The full results are reported in the Supplementary Material, and summary statistics are provided in Table 2. Like previous analyses, random-effects meta-analyses revealed significant results for questionnaire measures but not for task measures across EF processes. Robust Bayesian meta-analyses indicated anecdotal to strong evidence in favor of H₁ over H₀ for questionnaire measures and substantial evidence supporting H₀ over H₁ for task measures.

Table 2.

Results of the Random-Effects Meta-Analyses, Cochran’s Q Heterogeneity Tests, Egger’s Tests, and Robust Bayesian Meta-Analyses

Process of Executive Function	Type of Measure	Random-Effects Meta-Analysis					Cochran’s Q Heterogeneity Test		Egger’s Test		Robust Bayesian Meta-Analysis
		k	Total N	z’ (r)	p	95% Confidence Interval	I2	p	t	p	Bayes Factor (BF10)
Overall executive function	Q & T	40	26,403	− 0.17 (-0.16)	< 0.001***	[-0.25, − 0.08]	98%	< 0.001***	1.74	0.081	4.44
	Q only	16	10,635	− 0.39 (-0.37)	< 0.001***	[-0.47, − 0.31]	91%	< 0.001***	-0.62	0.53	100.86
	T only	28	16,393	− 0.03 (-0.03)	0.45	[-0.12, 0.06]	93%	< 0.001***	-0.37	0.71	0.11
Working memory	Q & T	15	13,699	− 0.10 (-0.10)	0.29	[-0.28, 0.09]	97%	< 0.001***	0.64	0.52	0.29
	Q only	6	951	− 0.38 (-0.36)	0.001**	[-0.53, − 0.23]	77%	< 0.001***	/	/	4.90
	T only	10	13,064	0.10 (0.10)	0.29	[-0.10, 0.29]	94%	< 0.001***	-0.37	0.71	0.44
Inhibition	Q & T	15	2,106	− 0.16 (-0.16)	0.011*	[-0.28, − 0.04]	85%	< 0.001***	1.64	0.10	1.56
	Q only	6	951	− 0.33 (-0.32)	0.005**	[-0.50, -15]	83%	< 0.001***	/	/	4.63
	T only	9	1,155	− 0.03 (-0.03)	0.51	[-0.14, 0.07]	55%	0.039*	/	/	0.10
Shifting	Q & T	15	2,049	− 0.29 (-0.28)	< 0.001***	[-0.42, − 0.15]	90%	< 0.001***	0.74	0.46	2.79
	Q only	7	1,020	− 0.44 (-0.41)	< 0.001***	[-0.62, − 0.26]	88%	< 0.001***	/	/	1.59
	T only	9	1,098	− 0.13 (-0.13)	0.053	[-0.27, − 0.00]	72%	< 0.001***	/	/	0.26
Planning	Q & T	8	1,332	− 0.31 (-0.30)	0.006**	[-0.50, − 0.12]	88%	< 0.001***	/	/	1.74
	Q only	7	1,225	− 0.35 (-0.34)	0.004**	[-0.54, − 0.16]	86%	< 0.001***	/	/	1.07
	T only	1	107	− 0.01 (-0.01)	0.89	[-0.20, 0.18]	/	/	/	/	/
Emotional Control	Q only	7	1,225	− 0.24 (-0.24)	0.033*	[-0.45, − 0.03]	91%	< 0.001***	/	/	6.65
Initiation	Q only	7	1,225	− 0.34 (-0.33)	0.002**	[-0.51, − 0.18]	86%	< 0.001***	/	/	0.42
Monitoring	Q only	6	951	− 0.36 (-0.35)	0.001**	[-0.51, − 0.22]	75%	< 0.001***	/	/	1.01
Reasoning	T only	5	12,531	0.02 (0.03)	0.56	[-0.08, 0.13]	57%	0.13	/	/	0.17
Verbal fluency	T only	4	303	− 0.02 (-0.02)	0.69	[-0.16, 0.12]	12%	0.64	/	/	0.15
Decision-making	T only	1	44	− 0.31 (-0.30)	0.025*	[-0.55, 0.01]	/	/	/	/	/

Q, questionnaires; T, tasks. Egger’s regression tests were conducted only for meta-analyses involving at least 10 studies. *p < .05, **p < .01, ***p < .001

Risk of bias assessment

All 40 studies were assessed for risk of bias, and the results are presented in Table 3. Ten studies (25%) included individuals with higher and lower autistic traits who were comparable in demographic and other variables, except for their levels of autistic traits. All studies recruited individuals with varying trait levels from the same source population (e.g., the same community) and identified them using the same set of criteria and established measurements of autistic traits (e.g., the AQ). In addition, 23 studies (58%) examined the relationship between autistic traits and other confounding factors, including age, sex, marital status, and personality. These differences were statistically controlled for in all but one study. All but one study clearly stated and measured EF using tasks commonly used in the field. Only two studies (5%) assessed autistic traits at more than one time point to establish the stability of trait levels. Across studies, the statistical analyses regarding the use of t-tests, ANOVA, correlation, and regression were deemed appropriate for the study design.

Table 3.

Risk of bias assessment results

Study	Year	Item
		Q1	Q2	Q3	Q4	Q5	Q6	Q7	Q8	Q9	Q10
Kunihira et al.	2006	No	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Christ et al.	2010	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Bayliss and Kritikos	2011	Yes	Yes	Yes	Yes	Yes	No	No	Yes	No	Yes
Elsabbagh et al.	2011	Unclear	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Fugard et al.	2011	Unclear	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Ridley et al.	2011	No	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Takahashi and Gyoba	2012	No	Yes	Yes	Yes	Yes	Yes	No	Unclear	No	Yes
Maes et al.	2013	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Richmond et al.	2013	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Gökçen et al.	2014	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Gökçen et al.	2016	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Shi et al.	2017	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Albein-Urios et al.	2018	Unclear	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Camodeca et al.	2018	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Ferraro et al.	2018	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Hamilton et al.	2018	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Pirrone et al.	2018	Unclear	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Stewart et al.	2018	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Dai et al.	2019	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Hyseni et al.	2019	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Lewton et al.	2019	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Chen et al.	2020	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Huang et al.	2021	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Mason et al.	2021	No	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Robertson et al.	2021	No	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Tsai et al.	2021	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Zhang et al.	2021	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Chee et al.	2022	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Gong et al.	2022	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Camodeca (Study 1)	2023a	Unclear	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Camodeca (Study 2)	2023a	Unclear	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Camodeca	2023b	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Davidson et al.	2023	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Nicholls and Stewart	2023	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Stewart et al.	2023	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Su et al.	2024	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Zhou et al.	2024	Yes	Yes	Yes	Yes	Yes	No	/	Yes	No	Yes
Chen et al.	2025	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes
Hendry et al.	2025	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Wang et al.	2025	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes

Q1: Were the high and low autistic trait individuals comparable other than the difference in their levels of autistic traits? Q2: Were high and low trait individuals matched appropriately and recruited from the same population? Q3: Were the same criteria used for identification of high and low trait individuals? Q4: Was the level of autistic traits measured in a standard, valid and reliable way? Q5: Was the level of autistic traits measured in the same way for high and low trait individuals? Q6: Were confounding factors identified? Q7: Were strategies to deal with confounding factors stated? Q8: Were EF outcomes assessed in a standard, valid and reliable way for high and low trait groups? Q9: Was the level of autistic traits long enough to be meaningful, as verified by measurements taken at more than one time point? Q10: Was appropriate statistical analysis used?

Discussion

This study aimed to evaluate the strength of the relationships between autistic traits and EF in the general population by conducting complementary meta-analyses. Based on a total of 40 studies, we found significant negative relationships between autistic traits and EF across most EF processes. More importantly, the type of EF measure significantly moderated the associations, with questionnaires yielding stronger effects than behavioral tasks. Specifically, random-effects and robust-Bayesian analyses revealed significant, strong, and negative relationships between autistic traits and self- or parent-rated EFs, with mostly substantial evidence supporting the existence of a relationship (H₁) over the absence of one (H₀). In contrast, the meta-analyses indicated nonsignificant, very weak relationships between autistic traits and EF task performance, with primarily substantial evidence favoring H₀ over H₁. The heterogeneity in effect sizes reported by individual studies was not significantly influenced by the age group of the study sample or the approach taken to treat autistic trait scores. Potential publication bias was not evident across EF processes.

Autistic traits are continuously distributed throughout the general population [15]. Although the differences in EF between individuals with and without an ASD diagnosis at the group level are relatively clear [8, 11], the relationships between autistic traits and EF among nonclinical individuals were largely unclear. The present findings contribute to a better understanding of such relationships and help explain the discrepancies in the literature. Specifically, some studies have reported significant negative [25, 26], nonsignificant [27], or significant positive [28] relationships between autistic traits and EF. These studies employed EF task measures. According to the results of the present study, the pool effect sizes of the relationships between autistic traits and EF task performance were centered around zero across different EF processes. Assuming the presence of random errors and unexplained systematic differences among studies, the lack of such relationships between autistic traits and EF task performance might contribute to the conflicting findings obtained by different studies.

Similar to previous meta-analyses comparing individuals with and without an ASD diagnosis [8], we also identified a greater relationship between autistic traits and EF rating scores than between autistic traits and EF task performance in the general population. Questionnaires and tasks measure different psychological constructs, with the former emphasizing perceived real-world behaviors over a long period of time and in an open-ended environment, and the latter focusing on objective EF performance observed over a short period of time and in a structured environment [81–83]. For example, in laboratory tasks assessing interference control (e.g., the flanker task [84]), participants are usually instructed to ignore distracting information and respond as quickly as possible while minimizing errors, even though making a correct or incorrect response has little real-world significance. In contrast, in real-world situations—such as studying for a test—interfering information (like a favorite game app) does not always need to be addressed within the first few hundred milliseconds of an event. Instead, it often requires prolonged suppression of emotionally charged responses that come with meaningful costs and benefits, such as resisting the urge to check a phone during study sessions. Thus, the greater relationship observed for EF questionnaires could be because autistic traits indeed have a stronger relationship with perceived EF than with objective performance, or with sustained, as opposed to transient, self-control that carries meaningful, rather than minimal, consequences.

In addition, all studies measured autistic traits using self- and parent-reports, except for one study that used the ADOS-Generic [51]. Common-method variance [12] and subjective biases [13], if any, may have a shared influence on the measurement errors and reported levels of autistic traits and EF, resulting in a stronger correlation between the two. Alternatively, most studies examined the links between autistic traits and EF at the single questionnaire or task level. Many EF questionnaires, including the BRIEF, have demonstrated satisfactory internal consistencies and test-retest reliability [85]. In contrast, individuals EF tasks, though commonly used, may have lower reliability because performance on an individual task reflects the target EF and other processes, and therefore variations in performance on a given EF task result from differences in the target EF as well as differences in other processes, such as lower-level sensorimotor functions (i.e., the task impurity issue [81]). Therefore, questionnaires may be more reliable in capturing the impact of autistic traits on EF. Deriving a latent variable from multiple performance-based EF measures may enhance reliability and address the task impurity problem by extracting the variance common to multiple EF tasks that differ in non-EF processes [81]. To our knowledge, one study has used a latent variable approach for EF tasks and reported a small-to-moderate negative relationship (r = − .20) between autistic traits and complex EF performance [79]. However, the evidence remains limited. Therefore, we recommend that future research employs latent variable methods to more rigorously evaluate the relationship between autistic traits and objective EF performance.

By integrating the effect sizes reported by the present study and Demetriou et al. [8], we construct a picture of the relationships between autistic traits and various aspects of EF across the trait’s continuum, as depicted in Fig. 6. Perceived real-world EF behavior is represented by the ratings on EF questionnaires, whereas observed EF task performance is defined by accuracy or reaction time on paper-and-pencil and computerized EF tasks. Given the very weak relationships (Cohen’s d ≈ 0.07) between autistic traits and EF task performance in the general population and a moderate difference (d ≈ 0.49) between individuals with and without an ASD diagnosis, we suggest that the level of observed EF task performance is insensitive to the variation in autistic traits across the nonclinical spectrum until reaching a level that typically warrants a diagnosis of ASD. In contrast, given the large relationships (d ≈ 0.79) between autistic traits and EF questionnaire scores and the very large difference (d ≈ 1.84) between individuals with and without a diagnosis of ASD, the level of perceived real-world EF behavior decreases with that of autistic traits across both the nonclinical and clinical ranges.

Fig. 6.

Relationships between autistic traits and executive functioning across the trait’s continuum. EF, executive function. The effect sizes (Cohen’s ds) were estimated based on (a) the present study, which compared nonclinical individuals with high and low autistic traits, and (b) Demetriou et al. [8], which compared individuals with and without a diagnosis of autism spectrum disorder

Because the effect sizes reported by both the present study and Demetriou et al. [8] are comparable across EF processes, this profile appears to be applicable to describe the relationship between autistic traits and various EF processes, including shifting, inhibition, and working memory. In addition, the proposed picture depicts only the average trends of relationships, and there remain substantial individual differences in EF at varying levels of autistic traits. To provide a more nuanced view of the relationships, future work would benefit from identifying factors (e.g., self-efficacy [86]) that may moderate the strength of the relationships between autistic traits and EF, as well as factors that can potentially explain the relationships (e.g., comorbid inattention/hyperactivity symptoms and altered brain cortical structures [23, 87]).

Apart from fostering a fuller picture of the relationship between autistic traits and EF in the general population, the finding of differential effects of autistic traits on perceived EF and EF task performance has various implications. First, it shows that nonclinical individuals with increased autistic traits may have concern about their own self-regulation abilities or elicit concerns in their parents. These concerns may lead to psychological and physiological stress and decreased well-being in themselves or their parents [18, 88, 89]. Also, people tend to act in ways that make their expectations come true (i.e., the self-fulfilling prophecy [90]). Accordingly, individuals who believe they have poor EF may have less self-confidence, or even unintentionally create self-defeating behaviors, when pursuing a challenging yet achievable goal that necessitates EFs. Second, our findings provide a mechanism for the qualitative differences, such as in the experience of disability, between individuals with and without an ASD diagnosis [21]. That is, despite subjective concerns, nonclinical individuals with high autistic traits still have the cognitive abilities required for performing reasonably well in the real world.

Limitations

Due to the characteristics of the existing literature, the present study has several limitations. First, studies on autistic traits and EFs are overrepresented by young adult samples. Consequently, our findings regarding the relationships between autistic traits and EF in the general population may not be well generalized to other age groups. In addition, all studies relied on only one source of information to measure autistic traits, primarily by self-reports, sometimes by parent-reports, but rarely by behavioral observation. Thus, subjective bias may inflate the relationship between autistic traits and perceived EF. Third, almost all studies were cross-sectional, investigating autistic traits and their relationships with EFs at only one time point. Although autistic traits have been found to be quite stable from childhood to early adulthood in the general population [91, 92], the stability of the relationship between autistic traits and EFs and the influence of current autistic traits on subsequent EFs are still unknown. Furthermore, while a distinction can be made between cool and hot EFs [5], only one study used an affective decision-making task for hot EF [65]. Therefore, task-based EF measures were largely overrepresented by cool EF measures. Moreover, autistic traits share genetic and biological similarities with ASD [93] with parents and siblings of autistic individuals generally exhibiting higher levels of autistic traits compared to those of nonautistic ones. However, due to the dearth of evidence available, we could not perform meta-analyses on studies comparing high-trait individuals with and without autistic children or siblings.

Conclusions

In summary, the present study leverages complementary meta-analytic approaches to provide converging evidence that autistic traits are associated with lower perceived EF but not poorer EF task performance in the general population. These findings expand the executive function theory of autism, which focuses on the differences in EF between individuals with and without an autism diagnosis. They provide a more comprehensive picture of the relationship between autistic traits and EF across the entire spectrum of traits. In addition, the findings underscore the importance of paying more attention to individuals with increased autistic traits, who, despite the absence of a diagnosis, express concerns about their own EFs or elicit concerns from their parents. To further an accurate understanding of the relationships between autistic traits and EFs across the nonclinical spectrum, we suggest that future research would benefit from conducting longitudinal studies, estimating autistic traits based on multiple sources of information, and investigating factors that may moderate or mediate the relationships.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AQ

Autism Spectrum Quotient

ASD

Autism spectrum disorders

ADOS-Generic

Autism Diagnostic Observation Schedule-Generic

BAPQ

Broad Autism Phenotype Questionnaire

BF₁₀

Bayes factor in favor of the alternative hypothesis over the null hypothesis

BRIEF

Behavior Rating Inventory of Executive Function

EF

Executive function

H₀

Null hypothesis

H₁

Alternative hypothesis

SRS

Social Responsiveness Scale

Author contributions

MKY conceived the study. MKY, CTYL, TA, and SL screened the articles. MKY, CTYL, and JB extracted and coded the data. MKY, CTYL, and HCWC conducted the risk of bias assessment. MKY and CTYL drafted the manuscript, whereas others reviewed the manuscript. All authors read and approved the final manuscript.

Funding

JB was supported by the General Research Fund awarded to MKY by the Research Grants Council of Hong Kong (15608422).

Data availability

All data analyzed are included in this article and its Supplementary Material file.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.