31350
A Comprehensive Psychometric Analysis of Factor Models of the Autism-Spectrum Quotient Using Two Large Samples: Model Recommendations and the Influence of Divergent Traits on Total-Scale Scores

Poster Presentation
Friday, May 3, 2019: 11:30 AM-1:30 PM
Room: 710 (Palais des congres de Montreal)
M. C. English1, G. E. Gignac1, T. A. Visser1, A. J. Whitehouse2 and M. T. Maybery1, (1)School of Psychological Science, University of Western Australia, Perth, Australia, (2)Telethon Kids Institute, University of Western Australia, Perth, Western Australia, Australia
Background: Similar to clinically-diagnosed autism, autistic traits observed in the neurotypical population are heterogeneous. Psychometric measures of autistic traits, like the 50-item Autism-spectrum Quotient (AQ; Baron-Cohen et al., 2001), take this into account by providing scores on individual trait dimensions in addition to total-scale scores.

However, currently the factor models used for the AQ are diverse. This is an issue for researchers who use the AQ on two levels. First, inconsistencies are created as notionally similar dimensions have been defined in different ways. For example, there are at least ten different ways in which a ‘social ability’ factor has been defined using the AQ. Second, differences in the models also result in differences in the relationships between factors. Critically, a psychometrically sound model could contain some factors with low inter-correlations, suggesting significant heterogeneity between trait dimensions which would question the interpretability of total-scale scores.

Objectives: Our aim was to conduct a comprehensive large scale psychometric review of existing factor models of the AQ. This would serve to 1) provide guidance as to which models are and are not viable for research purposes, and 2) provide evidence as to whether total-scale scores are adequately interpretable for research purposes, or whether researchers should restrict themselves to observing and comparing individual differences on specific dimensions.

Methods: A series of confirmatory factor analyses were conducted on ten competing AQ factor models. We fitted each model separately to large samples of English-speaking undergraduate students (n = 1702) and individuals from the general population (n = 1280) to obtain indices of psychometric fit. Inter-factor correlations were also calculated for each model to assess the heterogeneity of the factors. Finally, the model that demonstrated the best fit indices was subject to a multi-group factorial invariance analysis, to determine if it showed comparable fit between the undergraduate and general population samples, and an analysis of internal consistency that accounts for inter-factor correlations.

Results: The 28-item three-factor model proposed by Russell-Smith et al. (2001) demonstrated superior fit indices across both samples. Furthermore, the multi-group factorial invariance analysis indicated that the model fit both samples comparably, suggesting that it is appropriate for use in both samples. Inter-factor correlations for the model varied from weak-negative to moderate-positive. Whilst internal consistency of each of the factors was adequate, internal consistency for the full 28-item scale estimated using coefficient omega hierarchical was poor.

Conclusions: Based on our comprehensive analysis of the existing AQ factor models, we recommend that researchers interested in using the AQ to examine individual trait dimensions use the Russell-Smith et al. (2001) three-factor model, as this model showed the best psychometric fit and internal consistency of the individual factors.

Given that this model excludes a substantial proportion of items on the AQ (indicative of the heterogeneity of these particular items), that some of the inter-factor correlations of this model are low, and that internal consistency of full 28-item scale is relatively poor, we caution researchers against using total-scale AQ scores and instead encourage the use of individual factor scores.