Measuring change across multiple occasions using the Rasch Rating Scale Model

When latent trait models are used to measure change across time, it is difficult to disentangle changes in one facet of the measurement context from changes in other facets. Hence, it is difficult to diagnose change. Wright (1999b) proposed an algorithm for disentangling change, and previously the authors applied this algorithm to measuring change across two occasions (Wolfe and Chiu, 1999). In this article we extend Wright's algorithm to disentangle changes in measures across three occasions. We describe a standard Rasch rating scale analysis of a multi-occasion evaluation that produces confusing results when subjected to a series of "separate" calibrations. Then, we apply Wright's correction to the same data to show that the algorithm reveals changes that are more similar to ones that would be expected. Our demonstration shows that Wright's procedure can reduce misfit to the Rasch Rating Scale Model as well as changing the interpretation of change within the measurement context.