Causal inference in epidemiology: recent methodological developments
8-12 November 2010
London School of Hygiene & Tropical Medicine
Bianca De Stavola, Simon Cousens, Mike Kenward and Rhian Daniel
Guest lecturer: Stijn Vansteelandt (Ghent University, Belgium)
Most courses on statistical modelling in epidemiology concentrate on
the use of regression models to estimate causal effects while
for measured confounders. These address uncertainty due to sampling
error but not, in general, other sources of error and
uncertainty which may arise from missing data, measurement error,
uncontrolled confounding, and/or selection bias.
Recent methodological advances make it feasible to incorporate at
some of these sources of error into statistical models so that
quantitative assessments can be made of their impact on estimates of
causal effect and the uncertainty around those estimates.
The proposed course will discuss the current state of the art with
respect to these issues, while retaining a practical focus.
Participants will acquire awareness of the common threads across these
new methods and competence in applying them in simple settings.
WHO SHOULD APPLY?
Participants will be expected to be numerate epidemiologists, or
applied statisticians with an interest in epidemiology and clinical
The topics covered will be:
*Causal diagrams and the backdoor criterion for the identification of
*Missing data and measurement error mechanisms as components of causal
*Methods to deal with the bias introduced by unmeasured confounders,
measurement error and missing data. These will include:
regression calibration, propensity score and instrumental variable
methods, multiple imputation, multivariate modelling, and sensitivity
*Frequentist and a Bayesian overview of these methods.
*Practical experience of the above methods in Stata.
COURSE FEE: £ 900
FOR FURTHER INFORMATION CONTACT: [log in to unmask]
TO REGISTER CONTACT:[log in to unmask]
You may leave the list at any time by sending the command
to [log in to unmask], leaving the subject line blank.