A new dimensionality estimation tool for multiple-item tests and a new DIF analysis paradigm based on multidimensionality and construct validity

This thesis is concerned with two critical issues facing the testing industry today: dimensionality analysis and DIF (Differential Item Functioning) analysis. Chapter 1 develops the use of new dimensionally-sensitive proximity measures with Hierarchical Cluster Analysis and DIMTEST to estimate the dimensionality structure of tests. The results of simulation studies and real data analyses indicate that the new tool represents a significant step forward in the ability of dimensionality assessment tools to identify reliably the latent dimensionality structure of a set of items. Chapter 2 of the thesis develops a new DIF analysis paradigm that unifies the substantive and statistical DIF research camps by linking both camps to a theoretically sound and mathematically rigorous multidimensional conceptualization of DIF. The new paradigm is shown to have the potential to improve the understanding of the root causes of DIF through the testing of substantive DIF hypotheses, to reduce Type 1 error through a better understanding of the dimensionality of the matching criterion, and to increase power through the testing of bundles of items with similar content. Two new paradigm-based DIF analysis methods, one of which employed the new dimensionality estimation tool of Chapter 1, were described and applied to real data. The analyses demonstrated that the new paradigm-based methods offered insights that cannot be obtained from the standard one-item-at-a-time DIF analysis.U of I OnlyETDs are only available to UIUC Users without author permissio