Predictive analytics is widely used in various domains, including education, to inform decision-making and improve outcomes. However, many predictive models are proprietary and inaccessible for evaluation or modification by researchers and practitioners, limiting their accountability and ethical design. Moreover, predictive models are often opaque…

Investigates the structure of item-pair conditional covariances given a unidimensional latent composite using a generalized semiparametric-modeling approach. Conditional covariances are shown to be highly informative about the latent dimensionality required to achieve local independence. Uses the new approach to justify the DIMTEST procedure. (SLD)

The power of the t, expected normal scores, Mann-Whitney U, Tukey, a modified Mann-Whitney U, and an adaptive procedure were investigated when sampling from population models empirically developed from test score distributions. The models used were selected members of the beta family. This investigation was unique in that not only did the means of…

It is unrealistic to suppose that standard item response theory (IRT) models will be appropriate for all new and currently considered computer-based tests. In addition to developing new models, researchers will need to give some attention to the possibility of constructing and analyzing new tests without the aid of strong models. Computerized…

The hypothesis that faulty classical psychometric and sampling procedures in test construction could generate systematic bias against ethnic groups with smaller representation in the test construction sample was studied empirically. Two test construction models were developed: one with differential representation of ethnic groups (White, African…