B0496
Title: Empirical tests of the assumptions underlying growth measurement in vertical scaling
Authors: Sanford Student - University of Delaware (United States) [presenting]
Abstract: Vertical scaling, which links the score scales of test forms with different intended difficulties (such as tests for students in different grades), is employed when the absolute measurement of growth is of interest. The measurement of growth is entirely reliant on the results of linking the different test forms' score scales, typically using the common item nonequivalent group (CING) design. Under this design, certain assumptions about the unidimensionality and grade-to-grade invariance of common items must be met in order for a scale to measure growth, but these assumptions often go untested in practice. This may in part be attributable to shortcomings of longstanding methods for assessing dimensionality and differential item functioning. A conceptual discussion of why common item dimensionality and invariance are so crucial to vertical scaling using the CING is provided. It then describes how moderated nonlinear factor analysis and exploratory structural equation modelling can be used to test these assumptions, with a focus on their benefits over historically popular methods for assessing invariance and dimensionality, respectively. Using simulated data based on existing empirical findings, plausible examples of potential violations of the assumptions are provided and is shown how these methods can be used to identify them.
