A few places are left on the short course "Methods for dealing with missing
data in epidemiological studies", which will run from 18 to 20 July at the
Department of Social Medicine, University of Bristol.
Faculty: John Carlin, James Carpenter, Jonathan Sterne
Outline:
- Introduction & basic concepts (motivating examples, simple methods to
deal with missing data: complete case, mean imputation, LVCF etc)
- Theoretical framework (types of missingness, patterning, missingness
model: MCAR, MAR, MNAR, dependence on analysis question, e.g. missing
exposure, outcome or confounder?, overview of approaches: weighting,
maximum likelihood, multiple imputation)
- Multiple imputation (basic principles, Rubin's rules for analysis of
multiply imputed data, overview of methods of imputation, Stata "mitools"
for handling imputed data
- Methods for imputation (hotdeck, fully model-based imputation (MCMC),
MICE, software options)
- Weighting methods (simple post-stratification, propensity score
weighting, weighted estimating equations, practical issues with weighting
methods, variance estimation, "doubly robust" estimating equations)
- Imputation using multilevel models, methods for nonignorable (MNAR)
missingness, participant presentations, recap of major issues, pros and
cons of different methods
All sessions will be followed by computer practicals. The majority of these
will use Stata software, but stand-alone packages such as Schafer's NORM
and recently developed macros for MlwiN will also be used.
Cost: £450.
For registration details please see
http://www.epi.bris.ac.uk/shortc/shortc.htm or email
[log in to unmask]
|