A primer for handling missing values in the analysis of education and training data

Abstract

Quantitative research in vocational education and training (VET) is routinely affected by missing or incomplete information. However, the handling of missing data in published VET research is often sub-optimal, leading to a real risk of generating results that can range from being slightly biased to being plain wrong. Given that the growing availability of data from large-scale surveys and administrative collections offers exciting new opportunities for quantitative VET research, it is important that researchers follow best-practice approaches when using such data in their own work. Against this backdrop, we: (1) provide a primer on the use of appropriate missing data methods for quantitative VET research; and (2) illustrate the detrimental effects of inefficient methods on research results via a simulation study using real-world education and training data from the Longitudinal Surveys of Australian Youth (LSAY).

Keywords:

• missing data
• multiple imputation
• large-scale datasets
• vocational education and training
• missing data bias