Statistical Methods for Causal Inference
Statistical methods for improving causal analyses
Aims and Objectives
The aim of this programme is to develop methods for causal inference that are robust to missing data and can investigate change over time, in order to draw unbiased conclusions about realistic problems, using complex observational data.
1. Develop methods to minimise bias due to missing data
2. Develop methods to model complex exposures and outcomes
3. Develop IV methods to examine causal influences of multiple exposures
4. Integrate evidence to improve causal models
Part 1 of this programme will develop methods to use internal and external information to infer the missing data structure, to inform all types of causal analyses. We will then focus on methods to maximise the robustness of IV methods to different types of missing data. We will pay particular attention to two cases: two-sample IV (using individual or summary data), and the investigation of disease prognosis. Part 2 will extend current methods for modelling trajectories and variability of exposures and outcomes. We will then focus on overcoming some of the current limitations of IV methods, by using structural equation modelling (SEM) and multivariable IV to examine impacts of time-varying exposures. Finally, we will maximise the use of all research data by extending methods to combine and use external information to inform causal models and sensitivity analyses.
Instrumental variables (IV)
We developed a new F-test statistic for determining whether the IV estimator in linear models suffers from weak instrument problems where there are multiple treatments that are potentially confounded. The problem here is that the instruments have to predict the multiple treatments jointly. The new conditional F-test statistic is shown to have similar properties to the standard F-test for the one-treatment model, and standard weak-instrument critical values can be used. It has been included in the user written software ivreg2 in Stata.
Key publication: Sanderson, E, Windmeijer F (2015). A Weak Instrument F-test in Linear IV Models with Multiple Endogenous Variables. Journal of Econometrics. Epub ahead of print:
Modelling change over the lifecourse
We developed a new method to identify which of a small set of hypothesized models explains most of the observed outcome variation, and showed that our approach identified the correct model with high probability in moderately sized samples, but with lower probability for hypotheses involving highly correlated exposures. Identifying a single, simple hypothesis that represents the specified knowledge of the life course association allows more precise definition of the causal effect of interest.
Key publication: Smith AD, Heron J, Mishra G, Gilthorpe MS, Ben-Shlomo Y, Tilling K (2015).
Model Selection of the Effect of Binary Exposures over the Life Course. Epidemiology. Epub ahead of print: PMID: 26172863