TITLE Estimation of Item Dimensional Measurement Direction Using Conditional Covariance Patterns. Computerized Testing Report.

Several nonparametric dimensionality assessment tools have demonstrated the usefulness of item pair conditional covariances as building blocks for investigating multidimensional test structure. Recently, J. Zhang and W. Stout (1999) have related the structural properties of conditional covariances in a generalized compensatory framework to a test item geometry model analogous to that developed by M. Reckase (1985) for the multidimensional logistic item response function. This model represents an item's measurement direction in terms of the multidimensional ability composite best measured (in terms of information) by the item. In this preliminary study, it is demonstrated how such directional representations can be reconstructed on the basis of each item's individual pattern of conditional covariances with all of the other items on the test. A proximity measure for item pairs based on these patterns is derived, and circular and spherical multidimensional scaling techniques are shown to provide a method by which the unique measurement directions of items might be recovered and represented. Several real and simulated data analyses suggest that the approach may be a useful too, for investigating and describing the dimensional structure of tests. (Contains 4 tables, 12 figures, and 24 references.) (Author/SLD) Reproductions supplied by EDRS are the best that can be made from the original document