|
Abstract:
|
Mislevy has applied "multiple imputation" techniques to Large Scale Assessment, such as NAEP, to make inference to latent constructs, which cannot be observed directly. However, in these applications, the influence of the complex sample design on the resulting inference has not been fully considered and a fixed effect model defining priors based on design and demographic variables was used in some applications. In this study, a random effect model is derived from Classical Test Theory. We focus on the estimates of variance components of a latent construct under a clustered sample design. Variation within PSUs is modeled as random effects in inferences. This approach is illustrated with a simulation study.
|
