[...] when variables related to onset phonemes were added to the analysis, the simple predictors were able to account for 42% of the item variance. [...] we present a new empirical database consisting of the word identification scores of 140 participants on 120 words, and we use it to quantitatively estimate the amount of variance that should be accounted for as a function of the number of participants.
Psychonomic Bulletin & Review 2009, 16 (3), 600-608 doi:10.3758/PBR.16.3.600 Item performance in visual word recognition Clelland, 1989). The results were somewhat surprising, since both of these models accounted for only a small ArnAud rey And Pierre Courrieu amount of the item variance (3.3% for Plaut et al.’s model, CNRS and Université de Provence, Marseille, France 10.1% for Seidenberg and McClelland’s). Spieler and Ba lota also noticed that the models explained the amount of FloriAn SChmidt-WeigAnd variance less well than did the linear combination of three Universität Kassel, Kassel, Germany simple linguistic predictors: log frequency, word length, And and neighborhood density (which accounted for 21.7% of the variance). Finally, when variables related to onset Arthur m. JACobS Freie Universität Berlin, Berlin, Germany phonemes were added to the analysis, the simple predic tors were able to account for 42% of the item variance. Standard factorial designs in psycholinguistics have been Itemlevel data therefore seem to provide a critical test for complemented recently by large-scale databases providing em- computational models of reading. pirical constraints at the level of item performance. At the same Seidenberg and Plaut (1998) claimed, however, that time, the development of precise computational architectures two reasons might explain the relatively low item vari has led modelers to compare item-level performance with item- ance accounted for by these models. First, item means level predictions. It has been suggested, however, that item per- are affected by several factors that are not addressed by formance includes a large amount of undesirable error variance these models. For example, they do not specify the pro that should be quantified to determine the amount of reproducible cesses involved in letter recognition or in the production variance that models should account for. In the present study, we of articulatory output. Balota and Spieler (1998) noticed, provide a simple and tractable statistical analysis of this issue. however, that the performance of these models remains We also report practical solutions for estimating the amount of surprisingly weak, since they fail to explain more vari reproducible variance for any database that conforms to the ad- ance than do three simple predictors (i.e., log frequency, ditive decomposition of the variance. A new empirical database word length, and neighborhood density) that are, in prin consisting of the word identification times of 140 participants on ciple, captured by these models. Their second, and prob 120 words is then used to test these practical solutions. Finally, ably more critical, argument is based on the fact that item we show that increases in the amount of reproducible variance data include a substantial amount of error variance. The are accompanied by the detection of new sources of variance. question is how substantial this amount of error variance is. Comparing Spieler and Balota’s database with a simi lar database recorded by Seidenberg and Waters (1989),1 The precision of theoretical accounts in the field of they found a .54 correlation between item latencies in the visual word recognition has significantly increased over two databases. This relatively low correlation indicates recent years. Indeed, cognitive modelers have proposed that a large amount of the variance in one database is several detailed descriptions of the structure and dynam absent from the other. ics of the reading system (e.g., Ans, Carbonnel, & Valdois, In the present study, in line with Seidenberg and Plaut’s 1998; Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001; (1998) criticism, we address the issue of error variance Grainger & Jacobs, 1996; Harm & Seidenberg, 2004; in item databases (for a similar a
Pages to are hidden for
"Item performance in visual word recognition"Please download to view full document