All models are wrong, but some are useful. Developing a useful statistical model from the available data can be challenging! For example, what should you do if a model assumption is violated, or if data are missing? Your statistical toolkit will be expanded to include modern techniques for tackling challenging issues that often exist in health research data, e.g. missing observations, non-linear effects, confounding and correlation between observations in a dataset. The methods for correlated data are relevant for analysing some epidemiological observational study designs (e.g., matched case-control studies, longitudinal studies with repeated measurements), and clinical trial designs (e.g. cluster RCTs, cross-over RCTs). Techniques to help assess the usefulness of a model will also be covered. This unit of study focuses on the application of statistical methods using the statistical software R. Topics: fractional polynomials for non-linear effects; mixed or random effects and marginal models (e.g. GEE) for correlated data; multiple imputation for missing data; propensity score for confounding; tools to assess model performance and classification.
|Academic unit||Public Health|
At the completion of this unit, you should be able to:
Unit outlines will be available 2 weeks before the first day of teaching for the relevant session.