Using Efficient Model Based Sum‐Scores for Conducting Skills Diagnoses

Consider test data, a specified set of dichotomous skills measured by the test, and an IRT cognitive diagnosis model (ICDM). Statistical estimation of the data set using the ICDM can provide examinee estimates of mastery for these skills, referred to generally as attributes. With such detailed information about each examinee, future instruction can be tailored specifically for each student, often referred to as formative assessment. However, use of such cognitive diagnosis models to estimate skills in classrooms can require computationally intensive and complicated statistical estimation algorithms, which can diminish the breadth of applications of attribute level diagnosis. We explore the use of sum‐scores (each attribute measured by a sum‐score) combined with estimated model‐based sum‐score mastery/nonmastery cutoffs as an easy‐to‐use and intuitive method to estimate attribute mastery in classrooms and other settings where simple skills diagnostic approaches are desirable. Using a simulation study of skills diagnosis test settings and assuming a test consisting of a model‐based calibrated set of items, correct classification rates (CCRs) are compared among four model‐based approaches for estimating attribute mastery, namely using full model‐based estimation and three different methods of computing sum‐scores (simple sum‐scores, complex sum‐scores, and weighted complex sum‐scores) combined with model‐based mastery sum‐score cutoffs. In summary, the results suggest that model‐based sum‐scores and mastery cutoffs can be used to estimate examinee attribute mastery with only moderate reductions in CCRs in comparison with the full model‐based estimation approach. Certain topics are mentioned that are currently being investigated, especially applications in classroom and textbook settings.