Sensitivity simulation and parameter correlation

[SionC]

Hi,

I want to do some sensitivity simulations where the parameters that are being randomly generated are correlated. This is straight forward enough to do if the parameters are perfectly correlated. However I am interested in being able to define the degree of correlation between the randomly generated parameters for the sensitivity analysis. Has anyone any experience of doing this?

Cheers,

Siôn

[tomfid]

There are two options:

Build a structure in the model that generates the correlation (there's a bivariate normal example in http://www.vensim.com/documentation/Models/OptSensi/MCMC).

Generate a sample using some other tool and use the File sensitivity option. http://www.vensim.com/documentation/sen ... ontrol.htm

If you can say more about your specific needs, I might have other thoughts.

[LAUJJL]

Hi Tom

I did not see in the Vensim documentation, where are stored the individual values for each period of time for the variables selected in the .lst file.

The only way that I have found so far is to do my own sensitivity analysis from VBA excel, generating each simulation using the DLL, sending to Vensim the values of the parameters and getting back the values of the variables selected to analyze it in Excel.

Any idea?

Regards.
JJ

[SionC]

Hi Tom,

Thanks - I will take a look at these options.

Best wishes

Siôn

[tomfid]

Hi JJ - The time series output for variables in the .lst is stored in the .vdf for the simulation run. You can get at it with the sensitivity graph, bar graph histogram, or stats tool options. For other analysis, you can extract the raw output by exporting the dataset (Model>Export Dataset...). - Tom

[LAUJJL]

Hi Tom

Thanks for the answer. It works fine, I can export to a tab delimited file, and from that to Excel.

Regards.

JJ