Mätproblem i norm- och kriterierelaterade prov några analyser och försök med tonvikt på reliabilitets- och diskriminationsmått

Sammanfattning: The present report is a summary discussion of a number of studies in connection with the following problemas a) analysis and empirical investigations of some test theoretical formulae and principles, b) differential scoring of multiple-choice questions and c) criterion-referenced tests: theoretical and empirical implications. In studies associated with the first problem is shown how some test theoretical formulae and principles can be derived from the information in the inter-item-covariancs matrix as well as how this information can be used to explain empirically found relations. In the latter respect the standard error of measurement and various biserial correlation techniques are dealt with. Studies carriëd out within the second field of problems account for reliability and validity effects with differential scoring of multiple-choice questions. A priori as well as empirical weight systems have been used. Systematically ordered alternatives gave higher reliability with differential scoring compared to conventional scoring. At the same time, however, the dimensionality of the test changed with negative validity effects as a consequence. The trends of the results were the same for the two types of weight systems. In connection with the third field of problems the development of criterion-referenced tests, their qualities and restrictions are discussed. Three aspects are quite closely penetrated, viz. the defining of objectives, the homogeneity and the different cutting-scores of the tests. Furthermore theoretical as well as empirical problems of measurement techniques when evaluating data from criterion-referenced measurement are dealt with. In the empirical study especially low and moderate correlations between norm- and criterion-referenced discrimination indices were received, as well as between different criterion-referenced discrimination indices. Moreover, the results indicated that increased vali- . dity defined in terms of the difference between pre- and post-tests does not necessarily imply increased reliability when placing subjects above and below a fixed cutting-score.