Structural equation modeling for estimating the identification accuracy and detection time latency of English monosyllabic words

Abstract

Structural equation modeling (SEM) was used to examine the statistical structure among sets of experiential (word age of acquisition and subjective familiarity) and lexical similarity (lexical equivalence class size and neighborhood density) variables for word identification and reaction time latency tasks. Stimuli were 240 vocoded monosyllabic English words with reduced intelligibility and altered similarity relationships. Participants detected a target word following a prime and on every trial reported the prime. The identification accuracy was estimated by words and phonemes correct, and detection latency was estimated by trimmed and harmonic mean RTs. A parsimonious SEM was chosen in terms of the chi-square and model fit indices that determine whether the models adequately described the particular associations of variables/interfactor relationships. The variable/factor error variances were constrained to be uncorrelated with each other in order to evaluate effects independently. A bootstrapping technique indicated that the regression weights of the top-down and bottom-up factors were small, but they were significant in the model. The variance accounted for (VAF) by the model was 7.1% for identification accuracy, and 5.2% for RT latency. The model also indicated that RT latency was highly influenced by prime identification accuracy (15% VAF). [Work supported by NIH/NIDCD00695.]