Thorndike, Thurstone, and Rasch: A comparison of their methods of scaling psychological and educational tests

Abstract

The purpose of this study is to describe and compare the methods used by Thorndike, Thurstone, and Rasch for calibrating test items. Thorndike and Thurstone represent a traditional psychometric approach to this problem, whereas Rasch represents a more modem conceptualization derived from latent trait theory. These three major theorists in psychological and educational measurement were concerned with a common set of issues that seem to recur in a cyclical manner in psychometric theory. One such issue involves the invariance of item parameters. Each recognized the importance of eliminating the effects of an arbitrary sample in the estimation of item parameters. The differences generally arise from the specific methods chosen to deal with the problem. Thorndike attempted to solve the problem of item invariance by adjusting for mean differences in ability distributions. Thurstone extended Thorndike’s work by proposing two adjustments which included an adjustment for differences in the dispersions of ability in addition to Thorndike’s adjustment for mean differences. Rasch’s method implies a third adjustment, which involves the addition of a response model for each person in the sample. Data taken from Trabue (1916) are used to illustrate and compare how Thorndike, Thurstone, and Rasch would approach a common problem, namely, the calibration of a single set of items administered to several groups.