Page 1 of 1
Screening test - Sensitivity analysis
Posted: Sun Dec 04, 2011 3:11 am
by olive
Good evening,
I was wondering if screening tests could be conducted in Vensim. I've run a sensitivity testing as illustrated in the user manual but I would be interested in knowing for each variable I select, what other variables have the strongest impact on its variability. Can I conduct such analysis in Vensim? I have the DSS version.
Thank you very much for your help,
Olivia
Re: Screening test - Sensitivity analysis
Posted: Sun Dec 04, 2011 11:08 am
by Administrator
Once you have run your sensitivity simulations, Vensim creates two files in the same directory as the run. I think (as I'm not at my computer) they are called sortsens.tab and sens.tab.
Take a look at these two files as they may contain what you need.
Re: Screening test - Sensitivity analysis
Posted: Sun Dec 04, 2011 4:43 pm
by olive
Hi,
I cannot find these files or any .tab files for that matter. The only files created are .lst, .vdf, and .vsc
Am I doing anything wrong or missing something? I can see the confidence bounds when I click on the variables of interest but that's it.
Olivia
Re: Screening test - Sensitivity analysis
Posted: Sun Dec 04, 2011 8:00 pm
by Administrator
Try the following.
1. Start an optimization.
2. Select the parameter you are interested in as the payoff.
3. In the "optimization control" dialog, for "Sensitivity" select "All constants" and in the next box put in the number 10.
4. In the "optimization control" dialog, for "Optimizer" select "Off".
5. Click "Finish".
This will run the model many times and change all constants by +/- 10%. The resulting sortsens.tab file lists the payoff (the parameter you are interested in) and what the change was when each constant was modified. You can use this to find out the constant that is the biggest influence on your payoff.
Is this what you need to do?
Re: Screening test - Sensitivity analysis
Posted: Mon Dec 05, 2011 3:12 pm
by tomfid
There are actually two approaches to sensitivity analysis - via the optimizer (as Tony describes) and via sensitivity simulations. The optimizer automatically generates sortsens.tab etc. if you do payoff sensitivity, which makes things very easy. However, you only get the results for the payoff, not for other variables in the model. If you want to explore the response of other variables, or multiple variables, you need to use a sensitivity setup (as you probably already have). There aren't any built-in screening tools, but you can export the resulting dataset (via Model>Export Dataset...) as a .tab file and use other stats tools to explore it.
Tom