Statistical Methods for Causal Inference
Statistical models lie behind much of the research investigating relationships between genetics, epigenetics, lifestyle behaviours and physical and mental health outcomes.
We develop and use a range of methods, in particular focusing on instrumental variable methods, models for change over time, ways to allow for missing data, and methods for bringing together evidence from different sources. Instrumental variable methods, and particularly Mendelian randomization, are one way of avoiding bias due to confounding and measurement error. Our research focusses on the assumptions necessary for the instrumental variables method to work, and ways to use this method to answer complex questions.
Change over time is important partly because it allows us to assess causality (e.g. whether changes in smoking behaviour pre-date changes in blood pressure) and partly because it may indicate when an intervention would be useful (e.g. if weight gain during childhood has a greater effect on adult blood pressure than actual weight during adulthood, this would suggest that public health interventions for child weight would be a good idea). We are developing methods to measure change, particularly in complex situations where many aspects of a person are changing simultaneously. We are also looking at pathways over time, for example the effect of maternal smoking during pregnancy on offspring BMI in later childhood.
The amount of research available to answer any question grows every year, as current research builds on previous research, and it is vital to be able to bring together all available evidence on a particular issue. We are particularly interested in developing automated ways to search the literature (cutting down time and human error) and methods for combining results from studies which do not all measure the same things in the same ways. All these developments allow researchers to make better use of the wealth of data available to answer important questions about human health and wellbeing.
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