Statistics Group



Causal inference

Berzuini, Bowsher, Constantinou, Dawid, Guo, Ramsahai

Although it is of fundamental importance to science, economics and public policy, statisticians have only recently paid serious attention to causal inference. At CSI we are developing the theory, consequences and applications of a specifically ``decision-theoretic'' approach to causality, and contrasting that with popular but misleading approaches based on counterfactual reasoning and causal diagrams.

  • Confounding occurs when treatments are compared on initially non-comparable groups. Particular subtleties arise with time-varying processes, as when a sequence of treatments is applied in response to a patient's reactions. We have clarified the nature of both static and time-varying confounding, and developed assumptions and methods to surmount it.

  • Causal inference from non-experimental data is especially problematic, but ubiquitous. Our approach has contributed to Academy of Medical Science guidelines for assessing the causes of disease from such data.
parner
Influence diagram representing sequential randomization. This dependency structure allows
the effects of dynamic treatment strategies to be estimated from data collected observationally.


  1. Dawid, A. P. and Didelez, V. (2010)
    Identifying the consequences of dynamic treatment strategies: A decision-theoretic overview.
    Statistical Surveys 4, 184-231.

  2. Geneletti, S. and Dawid, A. P (2011)
    Defining and identifying the effect of treatment on the treated.
    Causality in the Science (P. McKay Illari, F. Russo and J. Williamson, Eds.), Oxford University Press, to appear.

  3. Rutter, M., Dawid, P., Hingorani, A., Horton, R., Jones, P., Khaw, K.T., Kirkup, B.,
    Mulgan, G., Peckham, C., Pickles, A., Souhami, R. and Watts, G. (2007).
    Identifying the environmental causes of disease: how should we decide what to believe and when to take action?
    Working Group Report, Academy of Medical Sciences, London, 144 pp.