The Effects of Item Nonresponse on Psychometric Indexes: Evidence against Deletion Method

Item nonresponse is a serious challenge to the measurements in large-scale studies which plays a considerable role in decision and policy making about educational and/or cultural issues. This challenge affects both the process of instrument construction and its application because of big error and bias in the estimation of psychometric indexes and scores distribution parameters. Therefore, this paper aims at simulating the effects of different nonresponse rates under completely random and nonrandom item nonresponse by using a real data set. The results show that a nonresponse rate less than 5% can be ignorable under completely random nonresponse where there is no notable difference between respondents and non-respondents. In contrast, psychometric indexes (Cronbach’s alpha, split-half reliability coefficient and validity coefficient) and scores distribution parameters (mean, standard deviation and quartiles) can be biased and inaccurate if the item nonresponse at any rate is nonrandom. In addition, item nonresponse can lead to wide confidence intervals and misleading results. Hence, the simulation findings do not advocate the deletion of non-respondents from a data set as a solution to the nonresponse challenge.