Missing data in educational research
Missing data is a major challenge in educational research and is associated with three problems in data evaluation. Firstly, the reduced sample size due to the missing data leads to greater uncertainty in parameter estimation. Secondly, handling the data is made more difficult because many statistical procedures expect complete data sets. Thirdly, there is a risk of biased parameter estimates because there could be systematic differences between the observed and missing data. A possible solution is the method of multiple imputation. This procedure uses an imputation model to generate multiple substitutions for the missing data in a data set, taking into account the uncertainty associated with the substitution.
The present project deals with the optimization of the treatment of missing data in research practice and the evaluation of large-scale assessments. The focus is on the development of easy-to-use software that allows both a flexible imputation of missing values and statistical inference for the multiply imputed data. Overall, the project should thus make it possible to better exploit the analytical potential of large-scale school achievement studies by enabling a broad circle of users to adequately deal with missing data even in complex data sets.