Challenging the Developmental Substructure Model for Autism: A Perspective From Causal Statistics

The extent to which core autism traits-social interaction, communication, and restricted and repetitive behaviors (RRBs)-share a cause or have independent causes (e.g., fractionable triad model) has been a central debate in the field for decades. The latest iteration of this debate has reckoned with molecular genetic evidence that has largely failed to reproduce twin study heritability estimates. Against this backdrop, an alternative to the fractionable triad model-the developmental substructure model-was proposed by Constantino et al. in 2021. This model places a unitary, latent autism variable as the cause of the core phenotype. In the present Commentary, we showed that the developmental substructure model fails empirical tests derived from causal inference. Using published data, we demonstrated that correlations among core phenotype traits are largely the same in population samples as compared to autistic and non-autistic subsamples. Using simulations, we demonstrated that if latent autism were the cause of variance in core phenotype traits, then the correlations among the factors should reduce considerably after conditioning on autism. Although the developmental substructure model is well-reasoned and well-specified, it ultimately does not stand up to empirical evaluation.