试卷中含有单个高计分主观题时的信度估计方法

测评信度是衡量考试质量的核心指标之一,但常规的信度估计方法在估计含有单个高计分主观题试卷的信度时并不恰当,因为这种高计分主观题对测验总分方差的影响太大。解决这种问题的一个做法是:在估计出单个高计分主观题信度的基础上,进一步运用分层α系数公式估计整个试卷的测评信度。单个高计分主观题信度的估计方法有两种,即使用重测信度的估计方法,或者使用根据两个随机变量的相关系数会因随机误差的存在而衰减的特点所提出的估计方法。

Reliability Estimation Method for Tests Containing a Single High-Score-Point Construct-Response Item

Reliability is one of the core indices in test quality evaluation.The typical reliability estimation methods,however,are not adequate when the test contains one single high-score-point construct-response(CR)itemamidst many low-score-point items.The score variance fromthis single CR itemmay dominate the variance of test total scores.This issue can be fixed by using the stratified α coefficient formula when the reliability of such single high-score-point CR item is available,which can be estimated through either the test-retest reliability method or the single-itemreliability with the correction for attenuation formula.