Seminar 25th October 2016 2 p.m. 176/2013
NGCM seminar: An introduction to sensitivity analysis for computational models
Stephen Gow
CDT in NGCM
- Categories
- NGCM, statistical analysis
- Submitter
- Stephen Gow
For a computational model with multiple input variables, it is often of interest to investigate uncertainty in the model output, and to understand how uncertainty in the inputs propagates to the output. Sensitivity analysis concerns the apportioning of the output uncertainty to changes in the model inputs. We review techniques to do this by decomposing the model into a sum of main effect and interaction functions, and by considering the reduction in the output variance when some inputs are fixed. We also discuss the adaptations which must be made to these techniques when the model is computationally expensive and must therefore be approximated by a simpler function.