Sensitivity Simulation Setup

[johnjohnson4]

Hello,
I noticed that if I change the order of the independent variables in the senssitivity set up I get different results! Is this by design? if so please explain.
See attached model. The dependent variable in question is IRE (∆I/∆Itp).

I am using Vensim® PLE Plus for Windows Version 6.4b (x32) Serial #15379 Windows NT/2K/XP Version 6.2 build 9200 .

Thanks,
john

[Administrator]

How many simulations does the tool do you for (it should be 162).

[johnjohnson4]

I have it set for 1458 simulations. There are 6 variable with 3 levels each. the 3x6 factorial is 729 simulations. I am doing two replications or 1458 simulations. I selected MV so the order of combination should be random.

[Administrator]

Sorry, my misunderstanding (when testing I had the wrong control file loaded).

Changing the order should change the results. Each variable draws from the same random number stream, if you change the order of the variables, they will get different numbers.

[johnjohnson4]

Wow! This acocunts for the inconsistancies in my experiment results. I assumed the MV randome ordering function would randomize based on some criteria other than list order. I don't remember seeing this in the Sensitiivity analysis or MV descriptions. Maybe I missed it. IMO, it would be good infor to include in the function descriptions. Users need to know results depend or variable order in setup.

Thanks for your help. I'll add vaiable order to my controls for the experiment.

V/r,
john

[Administrator]

If you run enough experiments, the order shouldn't matter.

[tomfid]

If you're using Grid with VECTOR, order won't matter, because the setup is nonrandom. The order of the simulations might vary, but the same set will be run.

Exporting the run to view the data directly is a good check.

Documentation is here
https://www.vensim.com/documentation/sensitivity.htm

Every draw from a Multivariate sample is independent, so order doesn't matter (statistically at least).