Sensitivity Analysis

This chapter discusses the approaches to undertake sensitivity analysis of models. When a model is developed with the sample data, it is possible that it might not be as accurate with another sample from the same population. It is essential for the model to be accurate for any samples drawn from the same population, using the same sampling technique; otherwise, the predictions made from that model would not be reliable. There are many approaches to examine the sensitivity of a model to the sample data. We discuss two approaches: local sensitivity (based on a single, fixed-point reference) and global sensitivity (examined over a larger input space). The local sensitivity makes use of a one-at-a-time (variable) approach, whereas global sensitivity has many approaches, though we discuss the Morris method, also based on one-at-a-time approach. The sensitivity for change in each variable can also be represented graphically, and we create a Tornado plot (which we display for local sensitivity analysis), and the Morris sensitivity plot (for global sensitivity analysis). The chapter ends with a note on the approach to sensitivity analysis, depending upon the nature of the model.