sensitivity analysis

[Antonio Barrsn Iqigo]

Hello friends:
Mr. Udo wrote:
> i am looking for basis literature to the theme sensitivity analysis

I recommended to read User4s Guide of tools like Vensim, Solver
(Power-Sim) and others, that also propose easy models to initiate about
it.

This methodology answer to question as: What can I do if I do not know
the truth value (the only) for a parameter, and unless I know a path,
a tendency, a fan of possible values, and this parameter explain a few
variables ?
As a result, we get a fan of values with different probabilities for
this variable(-s).

For instance: I try to explain the diffusion model (Bass), with the
innovation and mouth to mouth (imitation) parameters.
Initially i put the values -for instance- of 0.01 & 0.06 . But are
these the only correct ?
And if we -the experience or others markets- think that they can variate
among 0.005 - 0.02 , and 0.04 - 0.07 with a determinate distribution -
random uniform, poison, ...- what ?
We say to model, give this margin of values and simulate 200, 1000 ...
times with different values for these parameters. We get variables like
new customers, customers, profits, cash flow,... (dependents), and that
show margens of values, and not only one value.

Best regards

Antonio Barron
Telefonica de Espaqa
http://ourworld.compuserve.com/homepage ... s/sd96.htm
antonio.barron@telefonica.es

[jimhines@interserv.com]

Regarding the discussion of sensitivity analysis.

Most of the discussion (including one of my responses) has focused on the use of
sensitivity to put confidence bands around the output of the model. Although I
think these responses address the original question, there is another use of
sensitivity analysis: to explore the dynamics of the model.

One test of whether you understand your model is whether you can "predict"
qualitatively how the output will change as you change ANY parameter. To test
your understanding, you want to alter each parameter one-by-one to see if the
output changes as you expect it to. If the change in a particular parameter
causes a surprising change in model output, then you analyze why and increase
your understanding of the model (and, perhaps, of the world).

This process can be (and in some cases has been) partially automated, although I
do not think that any of the commercial system dynmics environments do the job
yet.

Regards,
Jim Hines
LeapTec and MIT
jimhines@interserv.com