A Multiple Imputation Approach in a Survey on University Teaching Evaluation

Abstract

Missing data is a problem frequently met in many surveys on the evaluation of university teaching. The method proposed in this work uses multiple imputation by stochastic regression (MISR) in order to recover partially observed units in surveys where multi-item Likert-type scale are used to measure a latent attribute, namely the quality of university teaching. The accuracy of the method has been tested simulating missing values in a benchmark data set completely at random (MCR) and at random (MAR). A simulation analysis has been carried out in order to assess the accuracy of the imputation procedure according to two standard criteria: accuracy in ``distribution'' and in ``estimation''. The procedure has been compared with others widely applied missing data handling methods: multiple imputation by chained equation (MICE) and complete cases analysis (CCA).