Lexical decision tasks have traditionally emphasized reaction time as the primary outcome, with accuracy treated mainly as a screening criterion. While robust links between lexical properties such as frequency and length and reaction time are well established, much less is known about how lexical characteristics relate to word-level accuracy, or whether their effects on accuracy operate indirectly through processing speed. Using large-scale lexical decision data from the English Lexicon Project (39,091 words) and the British Lexicon Project (18,396 mono- and disyllabic words), we examined how five latent lexical dimensions—Frequency, Complexity, Proximity, Polysemy, and Diversity—predict both reaction time and accuracy. Lexical dimensions were derived from a factor model trained on 22 non-behavioral lexical features. For each dataset, we fit a mediation-style structural equation model in which lexical dimensions predicted standardized reaction time, reaction time predicted accuracy, and lexical dimensions also had direct effects on accuracy. Models were estimated separately after cross-dataset equality constraints significantly worsened fit. Across both datasets, the models explained substantial variance in reaction time (46–53%) and accuracy (44–52%) and showed consistent directional patterns despite differences in effect magnitude. Frequency and Polysemy facilitated processing, yielding faster and more accurate responses, with smaller but similar effects for Diversity. Proximity generally slowed responses and reduced accuracy, reflecting lexical competition, though some model-specific differences emerged. Complexity was associated with slower responses but higher accuracy, suggesting that richer structural cues may support correct decisions despite increased processing demands. Together, these findings demonstrate that lexical dimensions do not exert parallel effects on speed and accuracy and that slower responses do not necessarily indicate more difficult words. Treating reaction time and accuracy as distinct but related outcomes provides a more nuanced account of lexical processing in decision tasks. Poster here.
