Breadcrumb

Dynamic Causality

Supervisor: Vanessa Didelez

Theme: Optimisation under Uncertainty

Causal inference is the aim of many disciplines that conduct empirical research, and in most cases the system under investigation exhibits some dynamic behaviour. In medicine for example an important question is how an ongoing time-dependent drug treatment affects the patient's health outcome after a certain period of time and even more how to adjust the treatment over time, taking the patient's particular history (e.g. occurrence of side effects) into account in some optimal way.

The fundamental difficulty of causal inference lies in the problem of using data observed in one situation for decision making in a potentially very different situation, as reflected in the famous quote "correlation is not causation". Recently there has been increasing activity in Statistics as well as Artificial Intelligence developing methods that disentangle correlation from causation.

Projects in this area will aim to generalise these methods to deal properly with the dynamic case and emphasise the often neglected decision theoretic aspect of causal inference. Existing approaches will need to be scrutinised as to their sensitivity to model assumptions and more 'robust' alternatives have to be developed. A further important question to be addressed in a future project is the appropriate modelling of continuous time; even if decisions can only be taken at fixed points in time, the underlying processes, e.g. the patient's health, are continuous. Hence, there is much scope to investigate and improve existing methods and develop new ones which will have impact on important questions in subjects like medicine, sociology or public health.


Publications

  • Simulating from marginal structural models with time-dependent confounding (2013)
    Havercroft, Didelez
    Statistics in Medicine, vol: To appear
    URL provided by the author
  • Identifying the consequences of dynamic treatment strategies: A decision theoretic overview (2010)
    Dawid, AP, Didelez, V
    Statistics Surveys, vol: 4, Pages: 184 - 231
    DOI: 10.1214/10-SS081
    URL provided by the author
  • Graphical models for marked point processes based on local independence (2008)
    Vanessa Didelez
    Journal of the Royal Statistical Society Series B, vol: 70, Issue: 1, Pages: 245 - 264
    URL provided by the author