Introduction
Autism spectrum disorder (ASD)
1 is currently conceptualised as a life-long neurodevelopmental condition, with impairments in social interaction, communication (both verbal and non-verbal), repetitive behaviours and/or restricted interests (American Psychological Association
2013). Although much existing research on autism focuses on children, recent studies have begun to focus more on outcomes in adulthood (Farley et al.
2018; Lounds Taylor
2017). The first longitudinal follow-up studies of autism defined a “good” adult outcome as “leading a normal, or near normal, social life” and “functioning satisfactorily at school or work”; “fair” was defined as an individual who made social and educational progress in spite of abnormal behaviour, or interpersonal relationships; “poor” was defined as an individual who was unable to lead an independent life, but where there was “some potential for social progress”; and “very poor” was defined as an individual unable to lead “any kind of independent existence” (Rutter et al.
1967, p. 1185). These definitions of outcome are provided for historical context, and are no longer used in research into outcomes for autistic people, as will be discussed below. Using similar criteria, subsequent follow-up studies concluded that between 50 to 60% of autistic people had poor, or very poor, outcomes (Gillberg and Steffenburg
1987; DeMyer et al.
1973; Larsen and Mouridsen
1997; Lotter
1974; Rutter et al.
1967). Thus, many participants were highly dependent on others, had limited friendships or relationships, and had poor prospects for education and employment. But, definitions of traditional outcomes at this time were inconsistent, thus confounding comparisons of outcome between studies (Henninger and Taylor
2013; Ruble and Dalrymple
1996). Subsequently, concepts of what constitutes a good outcome or “normal” or “near normal” social life have been criticised as being subjective (e.g. requiring interpretation by the researcher; Henninger and Taylor
2013). Other criticisms of traditional outcomes include: complaints that they are based on “normative” criteria that are not necessarily important for people with autism (Bishop-Fitzpatrick et al.
2016); or are too restricted and do not cover life domains such as communication, pain, well-being and independence (McConachie et al.
2018); or for overlooking personal factors that may be important for outcome (such as quality of life, Ruble and Dalrymple
1996; or within-individual change over time, Georgiades and Kasari
2018).
Henninger and Taylor (
2013) note that the early 2000s marked a period of more empirically grounded, and standardised, measures of outcome, albeit still focused on participants’ capacity for sustaining an independent life. For instance, Howlin et al. (
2004) measured outcome across three domains: employment, independent living, and friendships/romantic relationships, with hierarchical ratings reflecting different levels of attainment in these areas. Several other studies have used a similar framework to that of Howlin et al. (
2004) (e.g. Eaves and Ho
2008; Farley et al.
2009,
2018; Howlin et al.
2013). Others suggest an alternative framework. For example, Billstedt et al. (
2005), set age as a consideration in determining whether friendships
or employment should be an indicator of outcome (if the participant was aged 23 years or older, living status was taken as an indicator of outcome; otherwise friendships/relationships were used as an outcome indicator). These authors also made a distinction between a fair (being in employment/education
or living independently/having friendships) and a restricted but acceptable outcome (meeting neither of the former criteria but being in a context that is accepting of the individual).
Despite these and other modifications to outcome ratings, most recent studies still indicate that a large proportion of autistic adults have a poor or very poor outcome (Howlin and Magiati
2017; Magiati et al.
2014). Nevertheless, some have reported more promising outcomes. For instance, Farley et al. (
2009) reported fair or better outcomes for 82% of their sample, likewise Kobayashi et al. (
1992) described approximately 55% of participants as having fair or better outcomes (indicating more independence in employment, social relationships, and living independently). Farley and colleagues suggested that the higher rates of positive outcomes in their cohort may have been due to the fact that most participants (93%) were members of The Church of Jesus Christ of Latter Day Saints (LDS; note, membership of this Church was not a selection criterion) in which “Inclusion of individuals with disabilities is a strong cultural value” (Farley et al.
2009, p. 116). Kobayashi and colleagues also noted that many participants (82%) were attending therapeutic camps that specifically aimed to include autistic people in the social environment. Thus, although most studies report that outcomes for autistic people are often poor, it may be that contextual or social factors are related to better outcomes.
Overall, existing data illustrate that outcomes for autistic adults, as currently measured, are often poor in terms of employment, relationships, and independence, and recent research has begun to try to identify participant characteristics that may predict outcome. The predictive variables most frequently studied include IQ, autism symptom severity, and language (such as no speech, single word use, functional speech, or more fluent speech, Howlin et al.
2013; or communicative phrase speech by a certain age, Farley et al.
2009). When interpreting findings from this research it is essential to bear in mind that many studies use different measures for these predictors. Moreover, given the heterogeneity of samples, year of study publication, and age of each sample, it is difficult to identify the exact influence of these predictors on later adult outcomes. Finally, it is important to consider that studies published in the same year may report on cohorts of autistic people initially assessed at very different time points (for example, Cederlund et al. (
2008) collected their childhood data in the early 2000s, whereas Eaves and Ho (
2008) collected their childhood data in the 1980s). This has implications for comparing outcomes from contemporaneously published studies, as the participants in each sample may be at different life stages.
Several follow-up studies suggest that IQ remains relatively stable over time at the group level (Billstedt et al.
2005; Anderson D.K. et al.
2014; Cederlund et al.
2008; Howlin et al.
2013). In many other studies, higher childhood IQ is associated with more positive outcomes. Both Eaves and Ho (
2008) and Kobayashi et al. (
1992) report significant positive correlations between childhood IQ and adult outcome scores (with correlation coefficients ranging between 0.39 and 0.58). Gillberg and Steffenburg (
1987) reported that an IQ of more than 50 distinguished those with fair or better outcomes from those with poor or very poor outcomes; Howlin et al. (
2004) found that those with a poor or very poor outcome had a performance IQ (PIQ) below 70
and verbal IQ (VIQ) below 50. Having a higher IQ in adulthood is also associated with better outcome (Cederlund et al.
2008; Eaves and Ho
2008; Howlin et al.
2013; Jónsdóttir et al.
2018). Eaves and Ho (
2008), Howlin et al., (
2013) and Farley et al. (
2009) report significant correlations (ranging from 0.40 to 0.65) between measures of IQ and outcome. Nevertheless, higher IQ alone does not guarantee a good prognosis; several studies report that participants with a higher IQ do not necessarily have better outcomes. Howlin et al. (
2004) reported that individuals with IQ in the range 70–99 and those with an IQ ≥100 did not significantly differ in terms of outcome. Likewise Billstedt et al. (
2005) reported that no-one in their study of 108 participants had a good outcomes despite 10% of their sample having an IQ greater than 70. Moreover, several studies have reported that at the participant level, IQ trajectories are not necessarily stable across time points (Billstedt et al.
2005; Cederlund et al.
2008; Farley et al.
2009; Howlin et al.
2004). Thus, IQ may be protective against the poorest outcome profile, but does not guarantee a good outcome (Pickles et al.
2020). The mixed results may be explained by individual differences in other factors, such as level of social support (Farley et al.
2009; Kobayashi et al.
1992), or severity of autism symptoms.
Autism severity in childhood has an even more complex relationship with outcome. Pickles et al. (
2020) identified four classes of outcome based on a longitudinal cohort study (mean age at recruitment 3.3 years and 26.1 at follow-up). Although the class with the best outcome profile had uniformly high IQ, the authors also identified a class with high IQ but with poorer behavioural and affective outcomes. Howlin et al. (
2013) found that retrospective information on early autism severity was a stronger predictor of outcome than either childhood language ability or childhood IQ. Helles et al. (
2017) and Jónsdóttir et al. (
2018) reported that autism severity (measured at mean age 29.9 and 6.0 years respectively) negatively correlated with outcome. Chamak and Bonniau (
2016) also reported that “high functioning” autism/Asperger's was exclusively associated with good outcome, and those who had “moderate autism” or “severe autism” were more likely to have poor or very poor outcome. However, this study did not explicitly state how these three categories of autism were derived, and no assessment of cognitive ability was made. Eaves and Ho (
2008) reported significant correlations between both VIQ and autism severity, but only VIQ accounted for a significant proportion of the variance in outcome; adding autism severity did not explain significantly more variance. It is also worth noting that autism severity seems to vary across the lifespan and may interact with IQ. Although Seltzer et al. (
2003) observed a steady decrease across a range of autism symptom domains across the lifespan (i.e. from age 10 to 53 years), Anderson D.K. et al. (
2014) reported that changes in autism symptomatology were greater for those with higher IQ in childhood.
Finally, early language development may also play a role in influencing outcome, although definitions of language vary considerably (from single words to communicative phrase speech, to expressive
and receptive language). Several studies have reported that the presence of language (compared to no language) before the age of 5 or 6 years (Howlin et al.
2004; Sevaslidou et al.
2019) was associated with better outcome. Gillespie-Lynch et al. (
2012) found that outcome was correlated with language ability (measured by expressive and receptive scores on the Reynell Scales of Language Ability Edwards et al. 1999) and with the extent of change in language ability between 4 and 18 years. In adulthood, poorer language ability (as measured by the Autism Diagnostic Interview-Revised; no phrase speech; phrase speech; or functional speech) was associated with a poorer outcome (Howlin et al.
2013).
A recent meta-analysis by Steinhausen et al. (
2016) of 15 adult outcome studies reported pooled estimates for percentages with good, fair, and poor composite outcomes as 19.7%, 31.1%, and 47.7% respectively. The review also identified characteristics that were significantly related to outcome, principally autism diagnosis type and mean age at follow-up. The Steinhausen review is relatively recent, and is valuable in summarising data on the overall outcomes for autistic people. It also explores the putative relationships between outcome and diagnostic ‘type’ (individuals diagnosed with childhood autism had significantly poorer outcomes compared to those with broader diagnoses of autism spectrum disorder or Asperger Syndrome for fair and poor outcome), and potential age effects on outcome (good outcomes are more prevalent in early adulthood aged samples—those aged between 20 and 29 at follow-up). However, an updated review and meta-analysis of composite outcome studies is timely for four reasons.
First, several outcome studies have been published since the last meta-analysis by Steinhausen et al. (
2016) (e.g. Farley et al.
2018; Helles et al.
2017; Sevaslidou et al.
2019; Pickles et al.
2020). Their inclusion is important in estimating outcomes for autistic people. For instance, Farley et al. (
2018) report on a larger sample than their previous follow-up; they also included participants who were not diagnosed with DSM-III but were subsequently deemed to meet DSM-IV criteria. The resulting outcome data were less positive than previously reported (Farley et al.
2009). Further, Sevaslidou et al. (
2019), having stratified their sample by gender, found that outcome was significantly poorer for males (although these findings are tentative as there were only 9 females in the sample). Second, the Steinhausen et al. review (
2016) did not assess the quality, or risk of bias, in the studies included. This is important, as conclusions drawn from the literature should be considered in relation to the methodological quality of the research base. In addition, the authors did not conduct sensitivity analyses to determine the robustness of their findings (Patsopoulos et al.
2008). This is necessary to assess whether one or more studies is having an undue effect on the estimated effect. Finally, given the uncertainties about the relationships between IQ, autism severity, language, and composite outcome score, a meta-regression analysis would be an important methodological advance to address the relative roles of these variables in predicting outcome. Thus, the present study sought to replicate and extend the meta-analysis of Steinhausen et al. (
2016) in four important ways: (i) produce an up-to-date, systematic review of outcome studies; (ii) replicate the sub-group analyses conducted by Steinhausen and colleagues; (iii) conduct sensitivity analyses to investigate how robust the estimates of outcome are; and (iv) investigate the predictive role of language, autism severity and IQ in a meta-regression model. This study focuses on the same conceptualisation of outcome (objective indicators based on employment, social relationships, and independent living). This is primarily because of the high heterogeneity reported in the previous meta-analysis of outcome studies, and because measures of outcome vary between study. Thus, it was deemed more methodologically rigorous to focus on this specific assessment of outcome.
Discussion
This analysis included 1,199 participants, with a wide range of diagnosis types (e.g. autistic disorder, Asperger’s syndrome, PDD-NOS) and nationalities (e.g. the United Kingdom, United States of America, Japan, and Sweden). The proportion of participants with an IQ above 70 was highly variable across studies (from 0 to 100%) and the proportion of males varied from approximately 50–100%. The pooled proportions of participants rated as having good, fair, and poor outcomes were estimated to be approximately 18%, 28%, and 50% respectively. This is the first analysis of outcome data in autism to use a meta-regression approach to identify study-level predictors of outcome. We identified a link between IQ measures in childhood and adulthood that were predictive of outcome. The present findings will be discussed in line with previous reviews below.
The estimated outcome percentages in the present study largely agree with the previous meta-analysis of the same type of outcome studies. Thus, it does seem to be safe to conclude that approximately 50% of autistic people will struggle to achieve, and only around 20%
will achieve, traditional markers of objective functioning (i.e. being competitively employed, having friends or romantic relationships, and living independently). This finding is striking, given this study included nine new studies not previously reported on. Thus, despite five years and nine new outcome studies, it appears autistic people still struggle to be independent (as conceptualised in this review). This study highlighted that IQ in childhood was a predictor of fair and poor outcomes. This is consistent with reviews conducted by Kirby et al. (
2016) and Magiati et al. (
2014). Both reviews qualitatively report that IQ in childhood is a predictor of outcome; moreover, Kirby et al. note that age is not a predictor of outcome. Thus, this review formally analysed these associations and is consistent with the previous reviews. What is important to note, is the utility of repeating these analyses periodically. Farley et al. (
2009)’s paper is highlighted in the Magiati et al. (
2014) review as an example of relatively better outcomes compared to other studies (with 48% having a very good or good outcome, and 34% a fair outcome). However, at a subsequent follow-up the very good and good outcome proportions were less than half (20%). In sum then, this review is broadly consistent with previous findings and adds a quantitative assessment of the role of IQ at the study level in predicting outcomes; until now this had only been qualitatively described.
To account for the high between-study heterogeneity we undertook two subgroup analyses based on age at follow-up and on diagnostic category. In the first analysis, the proportion of fair outcomes differed significantly, with more adults aged 30+ in this category, compared to fair outcomes in those aged ≤ 29 years (good and poor outcomes did not differ). Our results are inconsistent with Steinhausen et al. (
2016), who report significant differences for good and poor outcomes, but not for fair outcomes. This is likely explained by the categories examined. Steinhausen and colleagues compared three age groups (< 20 years, 20–29 years, and 30 + years); due to our exclusion criteria, we could only compare those samples followed up at a mean age of 29 years or less, versus those aged 30 years or more. The second subgroup analysis showed that those diagnosed with infantile/childhood autism were less likely to have a good outcome, and more likely to have a poor outcome compared to those diagnosed with autism spectrum disorders. This is somewhat consistent with Steinhausen et al. (
2016) who found the same association for poor outcome, however, Steinhausen and colleagues found an association between fair outcome but not good outcome. This may be explained by the inclusion of two very recent studies (Pickles et al.
2020; Sevaslidou et al.
2019) which reported good outcomes in around 40%; thus, these studies likely alter the pattern of association due to their higher proportions of good outcome. As an alternative analysis we examined a median split based on an estimate of the baseline year of childhood assessment (see Supplementary Material 3). This did reveal that good outcomes were significantly greater, and poor outcomes significantly lower, for those assessed in childhood or more recently (i.e. after the median year). Yet, an inspection of Table
1 shows both diagnostic categories (infantile/childhood autism and autism spectrum disorder) appear before and after the median year. This suggests these diagnostic categories (infantile/childhood autism and autism spectrum disorders) are not a useful comparison; moreover, these categories do not necessarily conform to current diagnostic criteria (Constantino and Charman
2016) which in turn may artificially create differences based on inaccurate categorisation of participants.
While higher IQ has often been associated with a more positive outcome (Eaves and Ho
2008; Howlin et al.
2004,
2013; Rutter et al.
1967) findings on the predictive value of early symptom severity
combined with IQ are often inconsistent (i.e. no relationship to outcome, e.g. Eaves and Ho
2008; or severity as a stronger predictor than IQ, e.g. Howlin et al.
2013). To address these mixed findings a meta-regression analysis was conducted: IQ in childhood and IQ in adulthood were used to predict outcome proportion (year of study, mean age, and proportion male were not related to outcome proportion). Only IQ in adulthood was a significant predictor of
poor outcome proportion (IQ in adulthood did not predict good or fair outcome and IQ in childhood did not predict fair or poor outcomes when controlling for multiple comparisons); increased proportion of participants with IQ ≥ 70 predicted lower proportion of participants with a poor outcome. However, the regression
p-values and β weights were included for all tested regression models for three reasons. First, as multiple comparisons can increase type II error rate (Rothman
1990), we present these findings to allow the reader to weigh up the effect sizes in light of the other statistical tests. Second, because the variance explained was reasonably high (around 50%) for each model, given the relatively small number of studies, effect sizes may be more informative when results are non-significant (Althouse
2016). Third, the small number of studies and predictors may mean the models lack statistical power (not that the effect of IQ is non-significant). Thus, these results should be interpreted very cautiously. However, given changes in cognitive functioning over time, it seems plausible that IQ in adulthood is likely to be more closely related to adult outcome than childhood IQ (with higher IQ being associated with a higher proportion of good, and fair outcomes, and a lower proportion of poor outcome). Future studies should seek to clarify this relationship by examining other factors associated with IQ which may also influence outcome. For instance, autistic adults with IQ in the average range may be able to obtain employment, but may struggle to keep it (Taylor et al.
2015) and Anderson K.A. et al. (
2014) found that autistic adults without intellectual disability were less likely to live independently or to have obtained employment compared to non-autistic adults with intellectual disability. Social factors, such as attendance in mainstream schooling may also exert an influence on outcome, potentially more so than IQ (Simonoff et al.
2019). These studies suggest that behavioural, cognitive, or social factors not necessarily associated with IQ may affect outcome. Moreover, psychiatric conditions may be associated with outcome. Sevaslidou et al. (
2019) reported that presence of a co-occurring condition was associated with poorer outcome and Esbensen et al. (
2010) reported that receiving psychiatric services was associated with poorer outcome. Therefore, while it is important for future studies to continue reporting cognitive ability, IQ alone should no longer be a primary focus for predicting outcome.
The main finding from the methodological assessment was that participants included in these outcome studies are unlikely to be representative of the autism population as a whole. Although many of the studies were longitudinal and reported on the same, or similar, cohorts over several time points, these studies were often drawn from clinical or therapeutic settings (e.g. Kobayashi et al.
1992; Moss et al.
2017; Sevaslidou et al.
2019). Thus, inferences from this analysis could be confounded by potential systematic differences between the wider population and clinical samples. Similarly, in cross-sectional studies (Mason et al.
2019; Otsuka et al.
2017), it is unclear how representative these samples are of their respective populations.
Attrition is a further source of potential bias. Three studies reported that individuals lost to follow-up had significantly lower IQ’s at earlier time points, suggesting that cognitively able participants are more likely to take part in subsequent follow-ups (e.g. Helles et al.
2017). This has a clear implication for our understanding of outcomes in autism—autistic people of lower IQ are less likely to be represented as cohorts age. Difficulties for researchers in gaining access to sheltered or supported accommodation may also be partly responsible for the loss of less cognitively able participants (Farley et al.
2018).
It is also crucial to reflect on the concept of outcome itself. As noted in the introduction, definitions of outcome have shifted from subjective appraisals (based mainly on researchers’ own concepts) of how well the individual is doing, to more objective criteria measuring employment, independent living, and friendships/relationships. It is important to note that employment and independent living are directly observable in accordance with predefined criteria; there is a subjective element to assessing social relationships (as can be seen from the criteria in Supplementary Material 2). Although these latter criteria are more methodologically robust and replicable, such concepts of outcome are both restricted and reductionist and tend to assume (at least implicitly) that the source of the poor outcome lies within the individual (e.g. autism characteristics or IQ). However, since the late 1990s there have been attempts to redefine what constitutes ‘success’ for autistic people (Georgiades and Kasari
2018; Henninger and Taylor
2013; Lounds Taylor
2017; Ruble and Dalrymple
1996). Ruble and Dalrymple (
1996) put forth a compelling argument that many aspects of success are lost when only objective criteria of outcome are used. Yet, they also recognise that outcome should be measured, in part, as a set of risk and protective factors. Therefore, employment may be a protective factor, as it provides an income (which in turn facilitates a level of independence), or unemployment may be a risk factor (leading to financial dependence on others). Nevertheless, Ruble and Dalrymple (
1996) contend that success can be conceptualised as something
more than the state of employment. Both Lounds Taylor (
2017) and Georgiades and Kasari (
2018) suggest that outcome should be measured relative to the individual. For instance, employment itself is only one aspect of outcome; how individuals fare over the course of their lives is also critical. Thus, an autistic person who has low educational attainments, but then pursues a hobby or interest and has positive social interactions with the community can be conceptualised as having a good outcome (examples of such alternative accounts of outcome can be found in Ruble and Dalrymple
1996). Recently, some studies have measured both traditional objective measures of outcome and some aspects of personal subjective experience (e.g. sense of coherence; Helles et al.
2017), but as yet there is no robust framework for integrating these objective and subjective domains. Hence, although traditional measures of outcome are important, for example for comparisons with other groups of individuals (Shattuck et al.
2020), consideration of each individual’s perspective is vital in helping to define what “success” really means. Following from this, future research should seek to integrate the autistic person’s valued “beings” and “doings” (thought to be central to some objective-subjective theories of quality of life; e.g. Ruta et al.
2007); this will help move conceptualisations away from deficit (or deficit only) models, towards models that recognise each individual’s strengths, as well as their challenges (Robertson
2009; Ruble and Dalrymple
1996).
Strengths and Limitations
The present meta-analysis has several strengths. This review used a meta-analytical model that has recently been shown to be more robust to high inter-study heterogeneity compared with other models (hence providing more conservative estimates and confidence intervals; Doi et al.
2015). Consequently, these results, plus those of the previous meta-analysis (Steinhausen et al.
2016), indicate a consensus about the estimated proportions of autistic adults who have poor, fair or good outcomes. Our meta-regression models of IQ and outcome represent a novel and important contribution to the literature (see Pickles et al.
2020 for a latent profile analysis that draws a similar conclusion). Whilst most of these models were non-significant at the adjusted
p-value, the results were presented due to considerations about statistical power and potential type II errors. As outlined in the introduction, our protocol used two independent reviewers for screening (GRS) and data extraction (SJC; reliability was high for both processes). Third, this study undertook additional analyses that extend the findings of the previous meta-analysis. A sensitivity analysis was performed to gauge how robust the findings are, which indicated that estimates were not being inflated by an ‘outlier’ study.
Limitations of the present meta-analysis must also be acknowledged. The main issue affecting the interpretation of data is the measurement of outcome. Moreover, as interpretations of the criteria used vary across studies this confounds between study comparisons. Therefore, although the pattern of outcomes (few good, and many poor, outcomes) is broadly similar across studies it is not feasible to assess how measurement differences affect the conclusions. Second, we replicated the subgroup analysis conducted by Steinhausen and colleagues. However, as noted above, this analysis is problematic, given the lack of precision with which diagnostic categories are applied. Thus, although we found some similar results it is unclear how meaningful this analysis is at anything but a broad conceptual level. It was necessary to estimate a time point to compare those assessed in childhood across time (i.e. those assessed in childhood decades ago, versus those assessed in childhood more recently). We estimated a median year of assessment based on the information presented in each paper (where possible). However, it must be acknowledged this is a crude index of assessment, and was not possible for all papers. Thus, if such information was available for all studies the pattern of results may be different. A third potential limitation is the very high between-study heterogeneity, with each outcome category showing heterogeneity greater than 88%. Therefore, it could be argued that these studies are too heterogeneous to compare using a meta-analysis. However, the meta-analytical model used has been shown to be robust to high levels of between study heterogeneity. Moreover, those regression models that were significant explained just over 50% of the variance. Thus, the association with IQ does help to account for some between-study differences. It is likely that there are many unexplored factors contributing to this heterogeneity (e.g. participant level characteristics, such as decision-making style; or contextual factors, such as socio-economic status that have yet to be measured in outcome studies); as the number of published outcome studies increases future meta-analyses will be able to investigate these and other variables in more detail. Thinking prospectively about future analyses of outcome data is important; the number of studies in this review is small, thereby reducing the statistical power to detect effects. Moreover, other variables that have been hypothesised to be predictive of outcome (e.g. autism severity and language) were not factored into the present analysis as sufficient data for analysis were often unavailable. Thus, any future meta-regression should assess the contribution of other theoretically important variables in predicting outcome. Finally, the use of a standardised checklist is recommended to assess studies’ quality and risk of bias (Siddaway et al.
2019). However, since many such measures exist (Siddaway et al.
2019 report the number to be 86), a future review using a different measure may find subtly different methodological problems/strengths. This is problematic for comparing quality/risk assessments across reviews. Indeed, varied measures of a range of domains (e.g. autism symptoms, quality of life, and cognition/language) is common in autism research (Magiati et al.
2014). It may be that researchers should attempt to identify paradigmatic measures that are consistently used to facilitate cross-study comparisons (Roestorf et al.
2019).