Select Page

New Paper: Reviewing factor models and total-scale scores for scoring the Autism Quotient

A comprehensive psychometric analysis of autism‐spectrum quotient factor models using two large samples: Model recommendations and the influence of divergent traits on total‐scale scores in Autism Research.

Michael C. W. English, Gilles E.  Gignac, Troy A. W. Visser, Andrew J. O. Whitehouse, & Murray T. Maybery

The paper described below is published in Autism Research and can be found here: https://doi.org/10.1002/aur.2198. If you would like to read the paper, but cannot access it, please contact one the study authors by email.

So many ways to score the Autism-Spectrum Quotient, but which is the best?

This review paper compares the statistical properties of the several different factor structures of the popular Autism-Spectrum Quotient (AQ; Baron-Cohen et al., 2001). The original authors of the AQ proposed five theoretically-derived subscales in the measure, each observing a different trait dimension associated with autism: Social Skill, Attention Switching, Attention to Detail, Communication, and Imagination. However, subsequent studies that attempted to psychometrically-derive these same subscales have consistently failed to do so. Rather than dismissing the subscales outright, researchers instead attempted to derive alternative ways of dividing up the 50 item questionnaire. Unfortunately, there has been little agreeance on how the measure should be divided. The result is over 10 different methods of measuring AQ subscales, each purporting to be better than previous suggestions, and confusion as to which subscales researchers should rely on.

We identified 11 different models of scoring the AQ. including the original five subscale model, and attempted to fit each model separately to two large data sets who had completed the questionnaire. The first data set included 1702 undergraduate students from the University of Western Australia, whilst the second was 1280 individuals from the general population recruited into the longitudinal Raine Study.

Similar to previous work, we found very little psychometric support for the original five subscale model suggested by Baron-Cohen et al. (2001). However, of the 11 models tested, a 28-item three-factor model that included Social Skills, Details/Patterns, and Communication/Mindreading subscales defined by Russell-Smith et al. (2011) was found to have the most psychometric support (link).

Questioning the interpretability of total-scale scores

A growing body of work is suggesting that many of the individual trait dimensions associated with autism do not strongly, positively correlate with each other. In short, having relatively high level of particular trait does not necessarily infer high levels of another trait. Weak relationships between individual factors or subscales can have important ramifications for whether total-scale scores can be interpreted – theoretically, two individuals may have the same total-scale score, but vastly different levels of the underlying traits that contribute to the total-scale score. This is a critical question to examine further given the numerous studies that rely on the examination of total-scale AQ scores.

After identifying the model with the best psychometric properties in the previous analysis, we examined the inter-factor correlations in this model. Below are correlations between the three different subscales in the Russell-Smith et al. (2011) model, with correlations in the undergraduate sample in the top-right and correlations in the general population sample in the lower-left.

Social SkillDetails / PatternsCommunication / Mindreading
Social Skill .108***.477***
Details / Patterns .122***-.055
Communication / Mindreading .447***-.077*

As you can see in the table, a range of weak, strong, positive and negative correlations were found. To further illustrate the effect of these inter-factor correlations, we identified every participant in our undergraduate sample with a total-scale score of 107 (the sample mean, calculated using the 1-4 scoring method) and presented a pie-chart for each of the 49 participants identified where each segment represents their endorsement of items on each factor (adjusted for factors with different numbers of items).


Pie-charts illustrating the variability in Autism Quotient subscale scores for individuals with identical total scale scores

The result is striking – whilst at the total-scale level, these participants are identical (each have an AQ score of 107), at the subscale level, the comparability of participants is substantially reduced. Hopefully, these findings highlight the need to consider individual subscale scores when using the AQ for research purposes as, clearly, the total-scale score does not provide the whole picture with respect to who individuals endorse different autistic traits.