Professor Mark Strong

I have three related research interests that fall under the general banner of Uncertainty Quantification:

1. How do we properly account for all relevant uncertainties when we build a computer model of a physical, biological or social system?
2. How do we (efficiently) compute value of information?
3. How do we work out the value of a computer model? How much should we pay to make a simple model more complex? When do we stop increasing the complexity of a model?

Jeremy Oakley, Jim Chilcott and I have proposed an "internal" discrepancy-based method for managing model uncertainty. See this paper in JRSS Series C, and this paper in SIAM/ASA Journal of Uncertainty Quantification that develops the idea of the 'Expected Value of Model Improvement'. The method is discussed in more detail in my PhD thesis.

We have proposed an efficient method for computing partial EVPI. This method works for any number of parameters of interest and requires only the PSA sample. See this open access paper in Medical Decision Making. R functions to implement the method can be downloaded here.

Our online web calculator for partial EVPI, SAVI, is easy to use: Just upload your PSA sample and SAVI does the rest.

SAVI is also available as an R package from GitHub. This allows users to run the SAVI app on their own machine, and removes the need to transfer any data over the net. Installation instructions.

The partial EVPI method extends nicely to the computation of EVSI. See here for our open access paper on the efficient computation of EVSI.