ERIC Number: ED390267
Record Type: RIE
Publication Date: 1995-Feb
Language Test Unidimensional Model Fit at Multiple Ability Levels.
This study examined initial evidence of changes in fit to a unidimensional model for some language tests at multiple ability levels. Seven data sets were analyzed using the first phase of exploratory factor analysis: principal component eigenvalue extraction. Each data set is analyzed at varying n-sizes: whole group; random subsample; and five normally-distributed ability groups. Smoothed inter-item tetrachoric correlation coefficient matrices are used. Results suggest that restriction of range (at ability levels) yields eigenvalues that give initial evidence of poor relative fit to a unidimensional model. Whole group and random subsample matrices, however, yield better evidence of fit to a unidimensional model. Restriction of range is a possibility of operational test use, and further research is needed on precisely what kinds of factor structures underlie language tests at varying ability levels. Specifically, this study suggests that a score on a given test at one institution might not mean the same thing as the same score at another institution, if the institutions do not see the same ability range. The survey nature of the project suggests that this concern may be applicable to a wide variety of contexts of language test development and use. Contains 51 references. (Author/MSE)
Publication Type: Reports - Research
Education Level: N/A
Authoring Institution: N/A