Handling of Missing Data

Institution: see Organisers & Supporters

Programme of study: International Research Workshop

Lecturer: Prof. Dr. Martin Spiess (University of Hamburg)

Date: Thursday, 14/09/17 (09.30 – 18.00 h)

Max. number of participants: 20

Credit Points: 5 CP for participating in the whole IRWS

Language of instruction: English


If the missing information is selective with respect to the research question, then simply ignoring unobserved information or applying other ‘ad hoc’ methods usually leads to invalid inferences, i.e. to biased estimators or actual rejection rates of ‘true’ null hypotheses being too high. In this seminar, basics of the missing data problem and some techniques to compensate missing values are discussed. A main topic in the introductory part is the missing data mechanisms, i.e. the mechanism that led to the missing information. The way how to deal with the missing data problem such that scientifically interesting inferences are valid depends mainly on assumptions about this process. A particularly important question is whether the precise missing mechanism can be ignored in downstream analysis, or if it as to be modelled explicitly. In the second part, an overview of various approaches to deal with the missing data problem is given. Besides ‘ad-hoc’ techniques which often lead to invalid inferences, model-based approaches like maximum likelihood methods as well as weighting and imputation methods will be considered. Most of the latter methods assume that the missing mechanism is ignorable. However, we will also consider a simple approach to estimate a model based on a non-ignorable missing mechanism. The third part deals with one missing data technique in more detail: Multiple imputations to deal with missing items. The concepts are illustrated with the help of examples, the software used is R.

Requirement of students: Statistical knowledge on the master level of an applied science programme is required.

Recommended literature and pre-readings:

  • Spiess, M. (2016). Dealing with missing values. In: C. Wolf, D. Joye, T.W. Smith and Y. Fu (Eds.), The SAGE Handbook of Survey Methodology (Chapter 37, pp. 595-610). Sage Publications: Thousand Oaks, CA.

You have to register for the 11th International Research Workshop to participate in this course.