General slowing in language impairment: methodological considerations in testing the hypothesis

Although the general slowing hypothesis of language impairment (LI) is well established, the conventional method to test the hypothesis is controversial. This paper compares the usual method, ordinary least squares regression (OLS), with another method, hierarchical linear modeling with random coefficients (HLM). The analyses used available response time (RT) data from studies of perceptual-motor, cognitive, and language skills of LI and chronological-age-matched (CA) groups. The data set included RT measures from 25 studies investigating 20 different tasks (e.g., auditory detection, mental rotation, and word recognition tasks). OLS and HLM analyses of the RT data yielded very different results. OLS supported general slowing for the LI groups, and indicated that they were significantly slower than CA groups across studies by an overall estimate of 10%. HLM indicated a larger average extent of LI slowing (18%). However, the variability around this average was much greater than that yielded by OLS, and the extent of slowing was not statistically significant. Importantly, HLM showed a significant difference in the RT relation between LI and CA groups across studies, indicating that study-specific slowing, rather than general slowing across studies, was present. A separate HLM analysis of two types of language tasks, picture naming and word recognition, was performed. Although the extent of slowing was equivalent across these tasks, the slowing was minimal (2%) and not significant. Methodological limitations of each analysis to assess general slowing are highlighted.