Ventana software support forum

Lately, I have had a little time to catch up on some SD reading!

It seems there are two interesting methods out there for calculating dominant feedback loops in a system; David N. Ford's is the first, in which changes in the behavior of a loop's "atomic" pattern (calculated by observing the change of pattern of a target variable's second derivative) indicates loop dominance; but from my trials this method doesn't seem to be robust or clear when the model is large and there are many complementary feedback loops.

The other method, which I find really fascinating is that of Paulo Goncalves's, (adapted from Nathan Forrester's work); this requires us to linearize the system, calculate eigenvalues from link gains of the system, accordingly calculate eigenevectors and form a resulting system of time varying equations. By taking the derivatives of the eigenvalues and vectors with respect to the loop gains we can calculate the impact of levels by changes in gain over time. Phase shift and gain of key state variables can also be calculated at specific times from the analytic equation.

With reference to this method:
- I have seen this analysis used in a few papers in past SD conferences; has anyone written a MATLAB or excel program to aid in such an analysis?

- Would anyone care to share some of their insights using either method?

Attached are the articles (published by the SDS, but they seem to be in circulation in the public domain)

Best Wishes,

Karan

The articles!

[Edited on 3-3-2010 by karankhosla]

Nothing is attached!
What is your job Karan?
JJ

[Edited on 3-3-2010 by LAUJJL]

hmm...apparently the forum won't upload my .zip files (745kb in size). You can download them from the following link:

http://www.earthsafeonline.com/Papers_R ... mLoops.zip

JJ - my work is research oriented, in particular trying to model various issues of socio-economic relevance in developing countries.

Best

KK

Hi Karan

Research oriented is not very clear for me.
Are you building models for others who are specialists of country delopment, or are you a development specialist working with SD? Are your models destinated to be used practically, influencing concrete policies development?
D not be afraid by my questions, I am not an SD consultant!
Regards.
JJ

Hi JJ, I'm glad you asked!

I have an organization called EarthSafe with which I try to promote systems thinking in sustainable development practice. To be honest, the only reason I started EarthSafe was as an excuse to devote all my time to learning everything I possibly can about the science of systems and to try and make a little bit of difference while I do it.

Most of the modeling is primarily for myself to understand various issues, such as the effects of population growth on infrastructure, growth of local economies etc - unfortunately, I work with "grass root" people, many of them coming from and working in rural areas, it's not easy for them to get into using models; however I do communicate the salient features of the models to them. I have had great success in getting them to use systems concepts (causal loop diagramming--->archetypes) to identify the systems (and to some degree, relevant leverage points) in which they participate.

To answer your question; yes, at some stage I would like the models to influence policy development - I am working on this!

The eigenvalue/vector dominant loop method seems very interesting and could perhaps even be useful...

I hope all is well.

Best Wishes,

Karan

Hi Karan

I am no more fascinated by SD and I think it is a good thing.
I was years ago, and it was deeply misleading.
About loops centered analysis methods, I studied them but never tried to use them for the following reasons.

1. Too complicated.
2. Not implemented in Vensim.
3. Impossible to use when there are too many loops.
4. Promoting punctual understanding versus global one.
5. Loops centered as if loops were the only explanation of the dynamic behavior of systems. There are too non linearities as in lookups, delays, stocks without any loops flowing through them.
6. Using the same principle of causality to test a model based on causality, contrarily to reality checks.
7. Using the same model that one wants to test, to test itself.
A bug in the model may generate wrong tests that can make something right look wrong.
8. If the methods were efficient they would be implemented at least in one software.
9. Needs the use of another supplementary tool at least for eigenvalue analysis or the digest tool for the ppm method.
There are enough possible drawbacks when considering using these methods, that I prefer to
refrain from doing it.
Regards.
JJ

That's an interesting critique JJ!