Ventana software support forum

Question for Tom/Colleagues:

(Apologies as this is technically not a Vensim question but rather on the theoretical side and also admittedly a quite broad question.)

In visualising results of sensitivity tests, the range of the confidence bounds usually grows over time. This is also seen in the couple of examples in Vensim help. I was wondering if this is generally true and what would be the best way to justify this. Would it be correct to write something along the following lines:

"The fact that the confidence bounds grow over time has to do with the cumulative nature of stock variables, which results in the inertness of behaviour once simulation veers away from the baseline."

Am I right in attributing the widening over time (at least partially) to the existence of the stocks as the memory of the system?

The confidence bounds also tend to grow wider in periods of growth where reinforcing loops are dominant and tend to shrink where balancing loops are dominant. I feel this is natural but can't necessarily justify it theoretically/mathematically.

Would you happen to know of a good source to read and refer to on these questions?

Thanks in advance!

kaveh.dianati wrote: ↑Wed Sep 30, 2020 11:20 amI was wondering if this is generally true and what would be the best way to justify this.

I'm not sure you can say it's generally true. It could easily happen that the confidence bounds widen and then converge again. For example, if you had a project that had to finish by a certain point in time and the model allocated resources to it, in the middle of the project you might see widening of resource allocation, but it would converge at the project end date.

It is up to you though. I would only try and justify the widening of the bounds on a case by case basis.

I'm sure Tom will comment as well when he gets online. He may well have a different opinion to me (and he regularly reviews submissions for the SD Conference).

Thanks. It makes sense. Yes, in my work it also seems that the bounds widen where reinforcing loops are dominant and then shrink again when balancing loops become dominant.

I wonder whether this is generally the case, or whether there is a paper on interpreting Monte Carlo sensitivity results within the SD field.

A comment from Tom would be much appreciated.

@Tom Fiddaman:

I was just wondering if you have 5 minutes to kindly look at this question and share any thoughts you might have on it, or else any good sources you might know of which talk about interpreting results of sensitivity tests with respect to feedback structure, and whether growing bounds are generally to be expected over time and when reinforcing loops are dominant.

That would be extremely helpful for my writing.
Many thanks!