Evaluating word-reading models at the item level: Matching the grain of theory and data

Spieler and Balota (1997) showed that connectionist models of reading accou nt for relatively little item-specific variance. In assessing this finding, it is important to recognize two factors that limit how much variance such models can possibly explain. First, item means are affected by several fac tors that are not addressed in existing models, including processes involve d in recognizing letters and producing articulatory output. These limitatio ns point to important areas for future research but have little bearing on existing theoretical claims. Second, the item data include a substantial am ount of error variance that would be inappropriate to model. Issues concern ing comparisons between simulation data and human performance are discussed with an emphasis on the importance of evaluating models at a level of spec ificity ("grain") appropriate to the theoretical issues being addressed.