The validation of a breadth of vocabulary test using structural equation modeling

A replication and extension of a structural equation model based on an historicodevelopmental theory of lexical acquisition (Miller & Lee, 1993) was performed with 143 items from the Vocabulary Levels Test (Nation, 1990) with 1151 native speakers of Japanese. The results supported the validity of the original model, which proposed that four latent factors, date of a word's entry into the English language, word length, polysemy, and word frequency, could account for a substantial proportion of the variance in lexical difficulty as determined by the acquisition order of words on the Peabody Picture Vocabulary Test---Revised (Dunn & Dunn, 1981). Evidence for the construct validity of the Vocabulary Levels Test was also shown using the same latent factors and by operationalizing lexical difficulty through the use of a Rasch analysis. Goodness-of-fit indices for the replicated structural model were acceptable (GFI = .929; NNFI = .974; IFI = 982; CFI = .982). However, a slightly revised model, which was more in accord with empirical research, was proposed a priori. This model was examined, and it too was found to have acceptable fit to the data (GFI = .928; NNFI = .971; IFI = 980; CFI = .980). Because of its equally good statistical fit and its greater empirical support, this model is proposed as being an improvement on the original Miller and Lee model for Japanese second language learners of English. These models represent the first time which structural equation modeling has been applied to a test of second language vocabulary and provide support for both the historicodevelopmental model and findings on lexical acquisition which have been produced in both first and second language acquisition contexts. Future research with a structural model which incorporates a semantic latent variable is proposed.