selecting sensible variables

[adjie]

Dear Forum,

i am doing sensitivity test on my model. i change one variable in each simulation. for example, i increase the variable by 10%. from the simulation results, i want to rank the changed variables in terms of their sensitivity to the simulation results. i plan to make a categorization on high, medium and low sensible variables. does anyone can suggest me about what percentage of changing on the simulation results for those three categories? my idea is if i change one variable by 10%, the low sensible variable will make changes on simulation results for about 5%-15%. and, the medium will be 15%-25% and the high one will be more than 25%.

thanks...

[tomfid]

I think it's hard to assign categories in advance, because sensitivities can vary rather dramatically.

Note that, if you're using Vensim DSS or Pro, you can automate this process using the optimizer (see for example the "all constants" parameter sensitivity option, in Help for the optimizer settings).

[karankhosla]

Hi adjie,

Tom's right and if I have read and understood your post correctly - when you have a model with non-linearities, you can't pre-determine, or categorize which variables based on their sensitivity to inputs.

My Best,

Karan

[LAUJJL]

Hi Adje

There are three chapters in the Vensim user’s guide that deal about sensitivity.
The sensitivity, the synthesim and the optimization chapters.
Start first to study all these chapters until you can rebuilt step by step all the chapter’s models on your own. If you cannot, try to rebuild models looking like the guide examples but simplified, until you can build yourself the full models.
Once you are able to do that, start experimenting with very little models, using all the tools available (for instance sensibility, synthesim, optimization, studying each parameter one after the other and eventually paired and so on). Self experimentation on simple models is the best way to learn SD.
You must too learn to think by yourself and not be overconfident on the mathematics.
Ther is no magic shortcut to your problem, but taking the necessary time to solve it, even if that time maybe quite long.
Regards.
JJ

[adjie]

Dear Tom, Karan and LAUJJL,

thanks for your explanations. i will study more on the optimization.

cheers,
adjie