Robustness of Irt Parameter Estimation to Violations of The Unidimensionality Assumption

Abstract

Multidimensional item response data were created from a hierarchical factor model under a variety of conditions. The strength of a second-order general factor, the number of first-order common factors, the distribution of items loading on those common factors, and the number of items in simulated tests were systematically manipulated. The computer program LOGIST effectively recovered both item parameters and trait parameters implied by the general factor in nearly all of the experimental conditions. Implications of these findings for computerized adaptive testing, investigations of item bias, and other applications of item response theory models are discussed.