Has linear statistics and categorical diagnostics led us astray in interpreting causal determinants of reading disability?

Based on recent studies the dominant phonological deficit account of reading disability has been largely superseded by double or multiple deficit models that hypothesize various causal determinants that act in a probabilistic way to give rise to higher risks of reading disability. Much of the research supporting multiple deficit accounts group participants into categories based on predictor variables and/or use linear statistics to analyze the data. However, in doing so, important information may remain unnoticed. It is the aim of this study to investigate in more depth the contingency relationships that exist over time between several predictor variables and reading performances across the entire range of ability.In a longitudinal design more than 1000 children were followed up from first grade on until the end of sixth grade (results are presented for the first two grades). Phonological awareness tasks, rapid naming (serial and discrete), visual processing tasks, letter knowledge, amongst other variables were measured at the beginning of grade one. At the end of grade one and two, reading measures (word and pseudoword reading, orthographic choice, spelling) were administered. Data were analyzed using contingency statistics (necessity and sufficiency) and linear models including interaction terms.Heteroscedasticity, a nuisance in classic linear models, proved to be e genuine characteristic of bivariate distributions plotting a predictor variable against reading ability. Phonological awareness and rapid naming showed clearly to be necessary, but not sufficient conditions for reading ability. At the extreme lower end both predictors operate as limiting factors determining reading disability. Interaction effects were capable of explaining part of the heterogeneity at higher ends of predictor variables. Implications for double and multiple deficit theories are discussed.