简论测验等距假设

A Brief Discussion about Test Equidistant Hypothesis

多数心理测验只能达到顺序量表水平,不适合算术运算和参数统计检验,测验分数是心理属性的一个区间的代表,而不是一个点的精确表示。但是,在实际运用中,我们假定心理测验是等距量表,即,条目内、条目之间、被试之间的分数具有相同的单位。我们认为条目和对条目的反应具有质性和量性的区别,因此,提出了条目质性的概念;我们还认为,有的心理属性的行为样本不具备难度性质,或者条目的难度没有实际意义,或者有理由假定条目难度是相等的,我们提出条目集群的概念来处理这类条目的分数分配方法。经典测验理论的标准分和条目反应理论的logit是在等距假设前提下对测验分数的数学表示,虽然它们本身都是等距量表,但并不代表测量内容也是等距的。明确等距假设的风险有利于更好地运用测验。有多种途径可以降低等距假设的风险。

Instead of inspection of arithmetic operation and parametric statistics, most psychological tests can only achieve the level of ordinal scale. The test score stands for one interval of a psychological attribute rather than precise expression. However, it is assumed that the scale of psychological tests is equidistant in practices, meaning the grades between items, between testees and within an item have the same unit. We suggest that there are qualitative and quantitative differences between items and the reactions to items. Hence the concept of item qualitativeness is raised. Besides, we think that some behavior samples of a psychological attribute have no difficulty, or the difficulty of items have no practical significance, or it is reasonable to assume that the difficulties of items are the same. The idea of item colony is proposed to handle the grading method of this kind of items. The standard score from classical test theory and the logit from item response theory are mathematical expressions of test scores under the assumption of equidistance. Although both of them are equidistant scales, it doesn＇t mean that the contents of the measurement are equidistant. Understanding the risks of equidistant hypothesis can better help test utilization. There are multiple paths to reduce the risks of equidistant hypothesis.