IMPROVING GLOBAL ASSESSMENT BASED ON PARTIAL EVALUATIONS

The assessment is part of the responsibility of professionals in education and key decisions are based on it. Whatever the decision, it is important for the assessment on which they are based to be of high quality, in order to use the results with confidence. This means that the assessment must be reliable. But the reliability is not the only condition; the assessment also must to be diagnostic. So diagnostic assessment and reliable assessment are two aspects that has to converge.
Diagnostic assessment involves using partial information related to specific cognitive processes or content domains (partial scores or subscores) that from a metric point of view are less reliable than the assessment based on total domain assessment (total scores). But it is possible using subscore augmentation procedures to improve the reliability of the subscales. That is important in one educational context where the assessment is not only numeric but it is also diagnostic.
The approach we present uses empirical Bayes estimation, which is based on famous Kelley’s formula for reliability; indeed it is a multivariable formulation or generalization of Kelley’s formula. The final estimates are based on the information coming from the related scales on the test, and they provide a profile more reliable than the partial scales. We describe this procedure by applying it on one multidimensional scale. We show how the reliability of the subscores and the student diagnostic profile assessment increase.