COURSE OVERVIEW
The course begins by offering an intuitive explanation of Rubin’s classification scheme for missingness mechanisms (MCAR, MAR, MNAR). It then moves on to a discussion of the deficiencies in a number of commonly used ad-hoc approaches to handling missing data, before introducing the method of multiple imputation, a principled approach for handling missing data under the MAR assumption. An overview of Stata 12’s multiple imputation capabilities is provided, together with a comparison of the different imputation approaches. The course concludes by offering a brief introduction to an alternative approach to handling missingness, that of inverse probability weighting. Throughout we emphasize that the choice of statistical methods for handling missing data should be made based on the assumptions believed to be reasonable for the study in hand.
TARGET AUDIENCE
Researchers with strong quantitative skills and experience in statistical analysis in Stata.
COURSE REQUISITES
It is also assumed that participants will have followed the Invited Speaker session of the conference, in which an introduction to the missing data problem will be discussed.
PROGRAM
SESSION I
The missingness mechanism and Rubin’s taxonomy (MCAR, MAR, MNAR)
A critique of ad-hoc approaches
SESSION II
An introduction to parametric multiple imputation
SESSION II: MULTIPLE IMPUTATION IN STATA
Setting up Stata for multiple imputation
mputation with the multivariate normal model
Imputation by chained equations
Analysing multiply imputed datasets
SESSION IV
A brief introduction to inverse probability weighting
Untestable assumptions and the importance of sensitivity analyses
Conclusions
