다차원 2모수 IRT 모형에 대한 대칭적 및 비대칭적 특성함수 척도연계 방법의 기능 비교

초록

In multidimensional item response theory (MIRT) scale linking, it has been shown that the item characteristic function (ICF) and test characteristic function (TCF) methods should perform better than other methods. The purpose of this study is to investigate the statistical properties and performances of the symmetric and nonsymmetric versions for each of the ICF and TCF methods. For this purpose, the symmetric versions of the ICF and TCF methods, which had been presented as nonsymmetric ones in previous studies, were first presented for use with the multidimensional two-parameter logistic (M2PL) model. Then, the relative performances of the four methods, symmetric and nonsymmetric ICF methods and symmetric and nonsymmetric TCF methods, were investigated through computer simulations under various scale linking conditions. The simulation study included three critical factors of (1) proportion of common items, (2) weighting scheme for the criterion functions, and (3) number of ability points in each dimension for the criterion functions. The performances of the four methods were evaluated in terms of accurate recovery of the parameters of the underlying population ability distribution and the M2PL item parameters for a 50-item multiple-choice test. Also, the four methods were analyzed as to whether they produced symmetric solutions or not. Main results were as follows. First, in the recovery of the underlying population distribution, the symmetric ICF method yielded the smallest linking error. Second, in the recovery of the item parameters, the nonsymmetric ICF method resulted in the smallest linking error. Third, as expected, symmetric ICF and TCF methods did produce symmetric linking coefficients, but nonsymmetric ICF and TCF methods did not.

키워드