Multiple causality in developmental disorders: methodological implications from computational modelling

Abstract

Abstract When developmental disorders are defined on the basis of behavioural impairments alone, there is a risk that individuals with different underlying cognitive deficits will be grouped together on the basis that they happen to share a certain impairment. This phenomenon is labelled multiple causality. In contrast, a developmental disorder generated by a single underlying cognitive deficit may nevertheless show variable patterns of impairments due to individual differences. Connectionist computational models of development are used to investigate whether there may be ways to distinguish disorder groups with a single underlying cause (homogeneous disorder groups) from disorder groups with multiple underlying causes (heterogeneous disorder groups) on the basis of behavioural measures alone. A heuristic is proposed to assess the underlying causal homogeneity of the disorder group based on the variability of different behavioural measures from the target domain. Heterogeneous disorder groups are likely to show smaller variability on the measure used to define the disorder than on subsequent behavioural measures, while homogeneous groups should show approximately equivalent variability. Homogeneous disorder groups should show reductions in the variability of behavioural measures over time, while heterogeneous groups may not. It is demonstrated how these predictions arise from computational assumptions, and their use is illustrated with reference to behavioural data on naming skills from two developmental disorder groups, Williams syndrome and children with Word Finding Difficulties.