Sex/Gender Differences in CARS2 and GARS-3 Item Scores: Evidence of Phenotypic Differences Between Males and Females with ASD

The term ‘sex/gender’ is used to acknowledge the overlap between gendered socialisation and biological sex, consistent with Springer et al. (2011) and recommended by Lai et al. (2015).

The term ‘high-functioning ASD’ is problematic and may be misleading. It has frequently been used to refer to autistic individuals with higher cognitive ability, milder symptoms and better long-term outcomes. However, many have argued that appraisals of ‘functioning’ levels should be based upon the (non-stable) adaptive functioning capacity of the individual, which is only weakly related to cognitive ability (Alvares et al., 2020). We use the term ‘high-functioning’ for consistency with the terminology of the CARS2-HF form, with awareness of its limitations.

A regular mixed-effects Bayesian analysis was conducted first and compared with the results reported here. There were meaningful and obvious differences in the results between the two approaches, highlighting the importance of using analyses appropriate for the type of data collected (in this case, ordinal).

The scale parameter reflects the standard deviation of scores (but the standard deviation for a t-distribution is undefined, thus the generic term scale) and the normality parameter reflects the extent to which the distributions shape differs from the normal distribution. As the parameter approaches infinity, the distribution approaches the normal distribution. (In NHST analyses where the t-distribution is used as a sampling distribution this parameter is known as the degrees of freedom.).

For comparison, analyses including age and IQ as covariates were conducted in addition to those reported here. The differences between these sets of results were minimal and, in most cases, negligible.

For scores on each CARS2 form, we ran models excluding item 15: General Impressions, for the purpose of comparison with those presented here. Scores on this item did not meaningfully impact results. Thus, there is no evidence that sex/gender differences in item scores were driven by clinicians’ overall impressions.

Refer to supplementary material for results of CARS2-HF item level differences.

Item-level estimated mean sex/gender differences in the latent variable and in the probability of a score ≥ 2 are supplied as supplemental material.

A summary of all items and estimates of sex/gender differences is supplied as supplemental material.

There are two major reasons why we might observe no evidence of a meaningful difference in index scores despite consistently higher estimated latent means among males on a large number of GARS-3 items. Firstly, the calculation of index scores involves individual item scores being aggregated and then categorised to produce a scaled score in each domain. This can reduce the extent to which small differences in item scores manifest in the scaled scores and therefore the index score. Second, the item-by-item analyses specifically model the latent variable underlying the ordinal ratings. One of the reasons that this is a superior approach to analysing the categorical ratings themselves is that it can detect differences in the latent variable that can be muted in the categorical ratings. Thus, a sex/gender difference could be present in the latent severity assessment for every item, but if it is small relative to the width of an ordinal category, that difference may be completely masked by the requirement to select one of a small number of ordinal response categories. In other words, even though a male–female pair may be assessed at slightly different levels of severity, in most cases the same ordinal response will best reflect the (different) underlying levels of severity. Thus, a consistent difference in the latent variable may not translate into a similarly strong difference in the categorical responses.