comparing different scaling methods for c-tests

C-tests are measures of general language proficiency. a typical c-test consists of several passages with the second half of every second word is deleted, and respondents have to restore the deleted parts. each gap is typically considered as an item, and the score on a passage is based on the number of gaps correctly reformulated. however, this increases the number of item parameters to be estimated and violates the local independence assumption. rather, researchers have proposed another approach in which each passage is viewed as a polytomous item with different ordered categories. although this method is more effective than the gap level, it results in information loss at item-level which, in turn, affects the accuracy of ability parameters. in this study, we propose another strategy upon which each sentence is considered as a polytomous item. that is, the scores on the sentences are aggregated and then polytomous rasch models such as rasch partial credit model are applied. the results of this new strategy is compared with those derived from passage and gap levels. to achieve this, data were collected from a sample of 160 english language learners in iran. participants completed a c-test comprising four english passages. each text contained 25 gaps. the test was administered as a mid-term reading comprehension test. using winsteps computer program, the obtained scores were subjected to dichotomous and polytomous rasch model analysis. the results of unidimensionlaity, infit and outfit mean squares, and reliability showed that the new strategy has a comparable performance to the other methods, supporting the effectiveness of the method. the findings of the study indicate that the suggested strategy is more efficient than gap- and passage- level. it also leads to the increase number of items and provides information at sentence level.