Briggs Derek C, Wilson Mark
School of Education, University of California, Berkeley, POME, 3659 Tolman Hall, Berkeley, CA 94720, USA.
J Appl Meas. 2003;4(1):87-100.
The act of constructing a measure requires a number of important assumptions. Principle among these assumptions is that the construct is unidimensional. In practice there are many instances when the assumption of unidimensionality does not hold, and where the application of a multidimensional measurement model is both technically appropriate and substantively advantageous. In this paper we illustrate the usefulness of a multidimensional approach to measurement with the Multidimensional Random Coefficient Multinomial Logit (MRCML) model, an extension of the unidimensional Rasch model. An empirical example is taken from a collection of embedded assessments administered to 541 students enrolled in middle school science classes with a hands-on science curriculum. Student achievement on these assessments are multidimensional in nature, but can also be treated as consecutive unidimensional estimates, or as is most common, as a composite unidimensional estimate. Structural parameters are estimated for each model using ConQuest, and model fit is compared. Student achievement in science is also compared across models. The multidimensional approach has the best fit to the data, and provides more reliable estimates of student achievement than under the consecutive unidimensional approach. Finally, at an interpretational level, the multidimensional approach may well provide richer information to the classroom teacher about the nature of student achievement.
构建一项测量需要若干重要假设。这些假设中最主要的是该结构是单维的。在实践中，存在许多单维性假设不成立的情况，在这些情况下应用多维测量模型在技术上既合适又在实质上具有优势。在本文中，我们用多维随机系数多项逻辑回归（MRCML）模型说明了一种多维测量方法的实用性，该模型是单维拉施模型的扩展。一个实证例子取自对541名参加中学科学实践课程的学生进行的一系列嵌入式评估。这些评估中学生的成绩本质上是多维的，但也可以被视为连续的单维估计，或者最常见的是，作为一个综合的单维估计。使用ConQuest为每个模型估计结构参数，并比较模型拟合度。还对各模型中学生的科学成绩进行了比较。多维方法对数据的拟合度最佳，并且比连续单维方法能提供更可靠的学生成绩估计。最后，在解释层面上，多维方法很可能会为课堂教师提供有关学生成绩本质的更丰富信息。
