A DIAGNOSTIC CLASSIFICATION MODEL FOR DOCUMENT PROCESSING SKILLS

Abstract

This paper introduces a modification to the Rule Space diagnostic classification procedure which allows for processing of response vectors containing missing data. Rule Space is an approach to diagnostic classification which involves characterizing examinees' performances in terms of an underlying cognitive model of generalized problem‐solving skills. It has two components: (1) a procedure for determining a comprehensive set of knowledge states, where each state is characterized in terms of a unique subset of mastered skills; and (2) a procedure for classifying examinees into one or another of the specified states. The procedure for determining a comprehensive set of knowledge states is based on the Boolean descriptive function given in Tatsuoka (1991). The procedure for classifying examinees involves comparing examinees' scored response vectors to the patterns expected within each of the specified knowledge states (Tatsuoka, 1983, 1985, and 1987). Missing data is expected to be a common problem for this approach because, although the procedure for determining the comprehensive set of knowledge states requires a large pool of items, the procedure for examinee classification can be performed with smaller (less expensive) item subsets. This approach to diagnostic classification is illustrated with data collected in the Survey of Young Adult Literacy, a nationwide survey of literacy skills conducted by the National Assessment of Educational Progress (NAEP) in 1985.