Testing the conditional independence and monotonicity assumptions of item response theory (Q1060799)

After presenting an item response model for describing the joint distribution of J dichotomously scored test items the author formulates a general theorem stating that if (Y,Z) is a partition of the item responses in X into two non-overlapping groups of responses, then conditional independence and monotonicity of the model implies that for every function h(Z) the responses in Y are associated. Next some implications of this theorem are discussed, among others testing the conditional independence assumption using simple tests possessing certain optimality properties. The general theorem is proved in the Appendix.