Robustness of multidimensional analyses against local item dependence

The negative impact of local item dependence (LID) on analyses using item response theory (IRT) has been investigated by many authors. Hitherto though, these investigations focused on unidimensional analyses. The objective of the simulation study presented here is to investigate the impact of LID on multidimensional analyses. The chosen simulation design considers tests with LID due to item bundles and compares the results of multidimensional analyses obtained with varying item bundle effect sizes, varying correlation levels for the latent traits, and different test designs. The results indicate that in multidimensional analyses LID results in a bias for the covariance estimation and that the direction of the bias interferes with the chosen test design.