An investigation of the Rasch model in its application to foreign language proficiency testing

The main theme of this study is the use of the Rasch model for dichotomously-scored items in the analysis of second/foreign language proficiency test data. Analytic procedures deriving from this model are applied to response data from English proficiency tests of two different types: (i) a cloze-type test, which embodies the notion of proficiency as being measurable by a single, global test, and (ii) the objectively-scored sections of the English Language. Testing Service (ELTS) test,, in which proficiency is. viewed as being divisible into sub-components, each measured by a separate subtest. The total numbers of testees involved are 854 and 1,503 respectively. The theoretical background relating to item response models is first explained, via "a discussion of traditional procedures for the analysis of test data. The relationship between the Rasch model and other response models of similar mathematical form is considered, and further details of its operation provided. The results of the Rasch analyses are compared with those from traditional analyses of the same data. The Rasch statistics are shown to be more informative, and therefore preferable, on several counts. Further investigations are carried out on both data sets, in order to assess the fit between model and data, to check for possible violations of specific model assumptions, and, to check for, expected model features. For, both the cloze-type data, and the ELTS, data (analysed in separate subtests), observed and expected item characteristic curves show reasonable conformity, though with some Instances of serious misfit In both cases. No evidence for departure from unidimensionality is found for the cloze data, but there is some indication that ELTS modules, when combined withGeneral components, may vary in their departure from unidimensionality