The heterogeneity of word learning biases in late-talking children

Purpose: The particular statistical approach researchers choose is intimately connected to the way they conceptualize their questions, which, in turn, can influence the conclusions they draw. One particularly salient area in which statistics influence our conclusions is in the context of atypical development. Traditional statistical approaches such as t tests or analysis of variance lend themselves to a focus on group differences, downplaying the heterogeneity that exists within so many atypically developing populations. Understanding such variability is important—classification of what a disorder is, an individual’s diagnosis, and whether or not a child receives intervention all directly relate to an accurate classification of the disorder and individual’s abilities compared to their typically developing peers. Method: Here, we use word learning biases (i.e., shape and material biases) in late-talking children as a sample case and employ a variety of statistical approaches to compare the conclusions those approaches might warrant. Results: We argue that advanced statistical approaches, such as mixed-effects regression, can help us make sense of heterogeneity and are more consistent with a modern dimensional view of language disorders. Conclusions: Accurate characterization of late-talking children (and others at risk for delays) and their prognoses is necessary for accurate diagnosis and implementation of appropriate target interventions. It therefore requires rigorous statistical analyses that can capture and allow for interpretation of the heterogeneity inherent in populations with language delays and disorders.