A Research On The Multidimensional Test Equating And Its Influence Factors

In order to more accurately examine the candidates' abilities,the tests in practice usually contain multidimensional structures.The large-scale test represented by PISA and TIMSS has a multidimensional structure in the framework design.Therefore,the equating between multidimensional tests becomes an unavoidable problem.This study attempts to equate multidimensional tests with have a high correlation(r=0.8),based on the overall single-dimensional equating method and the multidimensional equating method.The feasibility and applicability of the two methods were compared in various situations.Meanwhile,the impacts of the sample size and anchor number in each dimension on the multidimensional test equating are examined.The study is divided into three sub-studies,including two simulation studies and one empirical study,which have the following research results:(1)For parameters(both item and ability parameters),the results of overall singledimensional equating under all experimental conditions are better than multidimensional equating methods.Relatively speaking,the two methods have less influence on the equating of item parameters and have greater influence on the ability parameters.For IRT true score,the overall single-dimensional IRT true score equating results are also better than the multi-dimensional IRT true score equating results;the results of the empirical research are basically consistent with the simulation study;(2)Regardless of the overall single-dimensional equating or multi-dimensional equating method,the Stocking-Lord method and the Haebara method have basically the same effect on the parameter equating.(3)Regardless of the overall single-dimensional equating or multi-dimensional equating method,the equating error of the item parameters decreases as the anchor items number in each dimension increases.The influence of the number of anchor items on the ability parameters varies with the sample size.The IRT score equating error also decreases as the number of anchor items increases.(4)Regardless of the overall single-dimensional equating or multi-dimensional equating method,the error of the equating of the item parameters decreases with the increase of the sample size,and the increase of the sample size has almost no effect on the equating error of the ability parameters.The IRT true score equating error decreases as the sample size increases.