Abstract
This research addressed the comparison of concurrent calibration between a polytomous IRT model and a dichotomous IRT model using English achievement test data. Two forms of English achievement tests for senior high school students were composed of testlets (groups of items) to eliminate the effect of the dependence among within-testlet items. The two forms were equated with common testlets through a polytomous IRT model. The testlet parameter estimates and the category characteristic curves were analyzed on a common scale. The result showed that one form was more difficult than the other, as test designers had intended. The mean of the ability parameter estimates of the more difficult form was higher than that of the easier form. These findings yielded useful feedback for test designers. Item parameter estimates of independent dichotomous items, ability parameter estimates and the amount of test information derived by concurrent calibration under the graded response model (polytomous IRT model) and the two-parameter logistic model (dichotomous IRT model) were compared. The results showed similar parameter estimates for the two IRT models. The standard errors of ability parameter estimates for both models also were highly correlated. The two-parameter logistic model provided a greater amount of test information than the graded response model.
