Outline:

In the previous century causal modelling within statistics was thought to be very suspect because although causal inferences were often made from fitting models to data and calling the dependent variables causes of the response variable it was recognised that such labelling was spurious. However more recently it has been discovered that by defining causation within a probabilistic/ Bayesian system and embedding the probability model into a control model it is possible, under certain hypotheses, to define a causal model rigorously and unambiguously and thus provide a powerful deductive framework for inferences in a number of domains.

Unfortunately the domain of application of such systems is rather restrictive. The standard systems are typically non-dynamic and causes need to be expressible in terms of random variables - which is often not possible. This project plans to develop a more generic framework that captures not only the causal algebras above but three other analogous systems developed by the supervisor. Examples of these more recently developed novel causal algebras can be found in for Thwaites, Smith & Riccomogno (2010) and Liverani & Smith (2016): see http://www2.warwick.ac.uk/fac/sci/statistics/staff/academic-research/smith/.

 

Prerequisites:

The ideal student would to have either a four year maths or a maths related degree with an MSc in Statistics or Machine Learning and who is interested in developing a methodology which has wide applications across many domains.