关键词: 多级评分模型, 测验等值, 对数对比等值法, 初值

Abstract: There are a lot of polytomous item response theory models known as devide-by-total. The equating test for these models is very difficult, because hardly can one to find out quite accurately initial values of the equating coefficients. Lack of the accuracy of the initial values leads to the Haebara approach, Stocking_Lord approach, Symmetric Relative Entropy approach to fail for equating tests under these models. A new equating method, Logcontrast approach, is introduced in this paper. This approach has some advantages for these models, such as the robustness for choicing the initial values of the iteration, fast convergent, and accurate result. And Logcontrast equating approach could supply quite accurate initial values for the rest of the equating methods mentioned above. Moreover, Logcontrast approach generalized and impoved the logit method for equating the dichotomous items

Key words: polytomous score models, test equating, Logcontrast method for equating test, initial value for the interation