Construction and analysis of c-tests with rating-scale Rasch models

C-tests are written tests of general language proficiency. They consist of a small number of mutilated texts. Test-takers have to restore the original words in each text. In the present study, 476 examinees worked on 10 texts with 20 blanks each. Two polytomous Rasch models provided fine-grained analyses of these texts: (a) the discrete rating scale model (Andrich, 1978, 1982) and (b) the continuous rating scale model (Muller, 1987, 1999). Various model tests unanimously showed that 5 out of the 10 texts failed to satisfy strict psychometric requirements. The remaining texts, however, proved to be Rasch-homogeneous. They differentiated very well between examinee ability measures, with the test reliability being as high as .96. Taken together, the findings demonstrate that rating scale Rasch models are well-suited to constructing and evaluating C-tests in a sound and efficient way.