Once an equation, model, or simulation
is chosen to represent a given system, there is the question regarding
which parts of that equation are the best predictors. Sensitivity analysis
is “a general means to ascertain the sensitivity of system (model) parameters
by making changes in important variables and observing their effects;”
it is useful in impact analysis and policy analysis (Porter, et al., 1980).
Sensitivity analysis involves testing a
model with different data sets to determine how different data and different
assumptions affect a model.
"Sensitivity," here, is defined as “the
ratio between the fractional change in a parameter that serves as a basis
for decision to the fractional change in the simple parameter being tested”
(Porter, et al., 1980).
For example: you've developed a model to
predict personal computer sales. You have included a number of factors,
such as the ratio of price for the average home computer system to the
median income, the growth rate of the Internet, the average number of computers
per school pupil in the schools, and a few other factors. Your model is
pretty good with predictions based on the historical data you have supplied
it with. You believe the ratio noted above is the most important factor,
but by eliminating it from the model, the model gives the same results.
It turns out that the sensitivity here is not as high as you had thought.
A terrific introduction to sensitivity
analysis by Lucia Breierova and Mark Choudhari can be found at the following
Introduction to Sensitivity Analysis
An example of sensitivity analysis can
be seen at the following site:
Mike Middleton has made available an add-in for Microsoft
Excel on sensitivity analysis, with a user guide (recently, with a free trial
download and for sale):