There is no mirror effect in LDT. Implications for lexical decision models

Gómez, P. ¹ , Perea, M. ² , Zimmerman, R. ¹ & Biancardi, B. ¹

1 DePaul University
2 Universitat de Valencia

There is a robust, but rarely reported phenomenon in the lexical decision task: Across participants, there is zero correlation between
accuracy for nonwords, and accuracy for words. We have analyzed myriad of data sets from our lab, and also mega-study data sets (e.g., Dutch Lexical Project), and have consistently found that r = 0. This is a somewhat unexpected finding given that within word types (e.g., different word frequencies), or within nonword categories (e.g., TL nonwords and substitution nonwords), the accuracies show significant correlations. In addition, the correlations between RTs for words and nonwords are usually larger than .95. This finding is very hard to reconcile with accounts of subject performance inspired in signal detection theory, which would postulate that the difference between good and bad participants reflect differences in d'. We discuss the implications for diffusion models, the Bayesian reader, and the MROM models.