Abstract:
|
Model frameworks in test theory like confirmatory factor analysis and the corresponding variants in item response theory are common for analysing competence data available in many large-scale educational surveys. However, missing values inevitably occur in such and other survey data, either by design or due to item non-response. To cope with these missing values, two Bayesian estimation routines, for a multidimensional confirmatoric factor model and a normal-ogive IRT model, are investigated incorporating the ability to deal with missing values using the device of data augmentation and a tree-based approach for handling missingness in covariates. The properties of the suggested approaches are tested by means of simulation studies for both model approaches.
|