多维Rasch模型在维度分数报告中的应用—对带宽-保真度困境的解决

分别采用四维度和十五维度Rasch模型分析包含项目内多维度结构的科学测验数据,估计两种维度结构下维度分数的信度.结果表明,对比相应的单维模型而言,四维度与十五维度Rasch模型均能够极大提高各内容维度上分数估计的信度.四维度与十五维度Rasch模型拟合结果的比较表明,对于总长度固定的测验,维度数目的增加能够补偿子维度长度减少引起的信度损失.但是这一作用必须以维度间较高的相关性为前提.

An M-Rasch Approach to Domain Score Reporting: Solution to Bandwidth-fidelity Dilemma

Due to broad content coverage and limited testing time,the large scale assessment is challenged by the bandwidth-fidelity dilemma.This study is to explore how the multi-dimensional Rasch model would improve reliability in the within-item multi-dimensionality data.The results demonstrate that both 4-dimensional and 15-dimensional Rasch models fit data,which supports the construct validity of the test.The uni-dimensional model underestimates the correlation between domains due to measurement error.The multi-dimensional Rasch analysis yields a higher level of measurement precision and a more appropriate estimate for the correlation between domains as compared to uni-dimensional approach.The comparison between 4-dimension and 15-dimension analysis shows that the increase on the number of dimensions can compensate the effect of scale length reduction to a certain extent,as long as the correlations between the specific domain and the others are relatively high enough.In conclusion,the multi-dimensional Rasch analysis yields more reliable domain score than uni-dimensional Rasch model in the within-item multidimenionality context.For the test with fixed length,reliable scores can be reported on the more specified content domains,as long as there are high correlations between domains.