Competence distributions, latent regression models and plausible values. - Deutscher Bildungsserver

Competence distributions, latent regression models and plausible values.

h t t p s : / / w w w . n e p s - d a t a . d e / P o r t a l s / 0 / W o r k i n g % 2 0 P a p e r s / W P _ L V . p d f

[ Competence distributions, latent regression models and plausible values.Link defekt? Bitte melden! ]

The paper discusses constructions of competence distributions presupposing a Rasch model. Constructions, as well as estimation procedures, differ depending on whether competence distributions refer to either observed sum scores or some latent competencies. It is shown that using posterior distributions for estimating distributions of latent competencies often leads to misleading results. The same problems occur when 'plausible values' are used to approximate posterior distributions. Some of these problems can be avoided with latent regression models which focus on mean values of latent competencies. These models, however, entail that the notion of competence becomes dependent on conditioning variables and therefore are in conflict with the idea of measuring competencies. (Orig.).