Missing Data Short Course: 16-18 September 2015, Stellenbosch
Dr Jonathan Bartlett (Department of Medical Statistics, London School of Hygiene & Tropical Medicine, UK) presented this three day course from 16-18 September 2015 at Stellenbosch under the auspices of the South African DST/NRF Centre for Epidemiological Modelling and Analysis (SACEMA).
Course summary: Missing data are a ubiquitous issue in the analysis of epidemiological and clinical data. Missing data reduce the precision of statistical estimates, and perhaps more critically, may introduce bias into estimates. In this three day course we will develop a principled and practical approach to handling missing data. We will:
• introduce Rubin’s framework for classifying missingness mechanisms
• briefly consider the drawbacks of ad-hoc methods, to understand why they generally result in invalid inferences
• introduce multiple imputation as a flexible practical approach to handling missing data under the missing at random assumption. This will include the so called chained equations approach for imputing multivariate data
• discuss how imputation can be performed in the presence of interactions, [with] non-linearities, in multi-level data, and with survey data
• explore how sensitivity analyses for the missing at random assumption can be performed using multiple imputation
• describe an alternative approach based on inverse probability weighting, and introduce so- called doubly robust methods, which combine ideas from imputation and weighting.
The course will assume participants have a good knowledge of regression modelling and likelihood based statistical inference. No prior knowledge about missing data concepts or methods will be assumed. The course will use R, and a working knowledge of R, including the ability to manipulate data frames and fit and interpret generalized linear models will be assumed.
Jonathan Bartlett is a lecturer in the Department of Medical Statistics, London School of Hygiene & Tropical Medicine, UK. His research interests include methods for handling missing data and covariate measurement error. He maintains the website www.missingdata.org.uk, which provides various software, course materials and various content on missing data methodology. Jonathan’s current research is focused on developing methods for handling missing data and measurement error in individual participant data meta-analysis, funded by a Career Development Award in Biostatistics from the UK Medical Research Council.