A Multilevel Meta-Analysis of Executive Function in Individuals with Autism Spectrum Disorders

Background:   Individuals with autism spectrum disorder (ASD) are widely reported to have executive function deficits (see Hill, 2004 for review).  However, considerable variability has been reported (Hill, 2004; Pennington & Ozonoff (1996).  Variables suggested to moderate the influence that ASD might have on executive functions, include executive function domain, whether tasks are human or computer-administered, age, cognitive functioning, verbal ability, and medication.  Because these putative moderators are descriptive in the existing literature, firm conclusions cannot be made on how they might moderate the magnitude of deficit.  Meta-analysis procedures are well-suited to answer such questions (Borenstein, Hedges, Higgins, & Rothstein, 2009). 

Objectives:   To use meta-analysis to investigate whether executive function domain, administration type (human/computer), age, cognitive function, verbal ability, and medication moderate the effects of ASD on executive functions. 

Methods: The present meta-analysis examined published and unpublished studies that included a comparison of individuals with ASD and control individuals, matched for mental and chronological age, in tasks that measure executive functioning. The final analysis was based on 134 effect sizes drawn from 71 samples of participants ranging in age from 3 to 41 years. A multilevel approach to meta-analysis was used to handle the presence of non-independent effect sizes and to reflect the hierarchical structure of the data in the overall sample.

Results: The overall analysis supported the existence of a significant deficit of medium magnitude in executive functions in individuals with ASD when compared to controls (estimated weighted mean d= 0.457). Noteworthy findings were that the effect sizes were largest for planning tasks and smallest for working memory measures. Similarly, effects were reduced when tasks were administered by computer or a combination of media (human and computed), compared to their magnitude when administration was human-based only.  Finally, composite measures of executive functions produced smaller effects than pure measures.  Interestingly, these findings held across age groups as no significant main effect of that variable, or interactions of age with domain or cognitive level were observed.

Conclusions: Results suggest individuals with ASD show deficits in all EF domains examined here.  Contrary to suggestions, deficits do not develop with age; rather deficits were present across ages.  Results also suggest that observed executive deficits are not due to the individuals with ASD having lower cognitive or verbal levels.  Medications taken by individuals with ASD did not have an effect.  Administering tasks by computer may mitigate some of the social-motivational demands associated with face-to-face administration.   However, social-motivational demands cannot account solely for the executive function deficits because the effect size for computer administrations was still significant.  Finally, effect sizes were smaller when they consisted of composite scores created across conditions, suggesting that conditions within tasks may facilitate or hinder executive function.  Further research should be devoted to teasing apart these conditions.