Ventana software support forum

Posted by ""Mike Fletcher"" <mefletcher@gmail.com>

During the recent System Dynamics Conference I mentioned to several people the idea that the graphics display paradigms of SD should be updated. The display paradigm dates from the 1950's where graphical display options were limited, and it has not changed substantially since. Perhaps it should be reconsidered because of the advances in the graphical display of quantitative (and dynamic) data that have occurred since then. Most people are familiar with the work of Tufte, but there have been many other contributors as well. Many new ideas have been devised, and often subjected to empirical tests to verify the effectiveness.

Any paradigm for the graphical representation of SD insights set how the critical data (implications of stocks, feedback and delays etc.) are transmitted to the audience. The paradigm has a large flaw that has resulted in a flurry of papers over the years - the fact that stock and flow diagrams are strong in one arena, while being poor at showing feedback loops. Causal Loop Diagrams deal almost exclusively with showing the feedback relationships, but ignore stock accumulations, and have a rather unsatisfactory handling of delays.
Basic Time Series graphs are tried-and-true, but perhaps better methods exist to transmit the insights they contain as well. (see below). Some efforts to animate the dynamics have been done, but for the most part from what I've observed at least, they were extensions of the original paradigm.

Rather than argue whether Causal Loop Diagrams or Stock and Flow Diagrams are best perhaps its is time to admit that neither is adequate. Perhaps its time for SD to look outside the field (and the
box) for more effective graphical display paradigms that handles all critical information that needs to be transmitted to an audience equally well.

Below is an interesting piece showing the implications of the change of stock accumulations over time. It might serve as a think piece, or at least challenge the field to think about better ways to display and communicate the various critical portions of system structure and the implications of dynamic behavior.

---------------------------------------------------------------------------------

Hans Rosling is professor of international health at Sweden's world-renowned Karolinska Institute, and founder of Gapminder, a non-profit that brings vital global data to life. He debunks a few myths about the ""developing"" world. (Recorded February, 2006 in Monterey, CA.)
--------------

http://www.ted.com/tedtalks/tedtalkspla ... ns_rosling


--
------------------------------------------------------
Michael E. Fletcher
Posted by ""Mike Fletcher"" <mefletcher@gmail.com> posting date Wed, 22 Aug 2007 15:06:07 -0400 _______________________________________________

Posted by ""Ford, David"" <dford@civil.tamu.edu>

I offer the following for those facing the CLD vs StockFlow diagram issue. For several years now I have successfully addressed the stock-flow vs CLD graphics challenge that Mike Fletcher refers to by combining the two into a single diagram that I call a ""system structure diagram"". The diagram adds polarities to causal links and both ends of flows and loops to a stock and flow diagram. The diagram therefore shows both accumulations and feedback loops. This works better for me in my
(cramped) SD course schedule, my students have never had a problem with it, and it is much more efficient that jumping back and forth between loops on CLD and accumulations and flows on the S&F diagram of the same system. We use Sterman's book as the text, which keeps them separate. I still give separate lectures on stocks&flows and on feedback, each with the different diagrams, but a simple statement about combining them (with a compliment that ""We can do this because we, as engineers, are very familiar the water-flowing-among-tanks model."") let's us use one, uniform diagraming method for the rest of the course.

FYI, I watched the video Mike suggests below and found it both interesting and entertaining. I recommend it.

Best wishes,

Dave Ford
Posted by ""Ford, David"" <dford@civil.tamu.edu> posting date Thu, 23 Aug 2007 09:12:36 -0500 _______________________________________________

Posted by Joel Rahn <jrahn@sympatico.ca>

The graphical wizardry on display on the linked website is colorful, interesting, engaging and 'two-thirds' of it can be done in current SD modeling systems: The display of two (indexed) variables over time on two axes is a standard feature of Vensim, for example. What requires more work is putting multiple pairs of variables (e.g.
fertility and GDP per capita, indexed by country) on the same graph and what is missing in standard SD packages is the ability to add a third dimension (population, indexed by country) to drive the size of the 'symbol' (a circle, say).

None of this wizardry has anything to say about the relationship between structure and behaviour which is the subject of the following paragraph lifted from the QUERY message:

> Rather than argue whether Causal Loop Diagrams or Stock and Flow
> Diagrams are best perhaps its is time to admit that neither is
> adequate. Perhaps its time for SD to look outside the field (and the
> box) for more effective graphical display paradigms that handles all
> critical information that needs to be transmitted to an audience
> equally well.

The TED website's demonstration presents the 'critical information'
selected by the extremely accomplished and charming presenter using a sophisticated graphics package as a front-end for a complex data engine.
Some SD models could use those kinds of resources but we still have a lot of work to do identifying the dynamically 'critical' information to be displayed and then figuring out how to display it to best effect.

Joel Rahn
Posted by Joel Rahn <jrahn@sympatico.ca> posting date Thu, 23 Aug 2007 15:58:51 -0400 _______________________________________________

Posted by Khalid Saeed <saeed@wpi.edu>

Mike raises a good point. In my experience, using composite performance indicators in state space creates powerful graphics both for describing the problem and the model behavior in many instances.
I have illustrated this graphic option in the following article.

Saeed, K. and O. Pavlov. 2007. Dynastic cycle: A generic structure
describing resource allocation in political economies, markets and
firms. Journal of Operations Research Society. forthcoming.

pdf copies are available on request. please send such requests to saeed@wpi.edu and not to this list.

khalid
Khalid Saeed, PhD
Professor of Economics and System Dynamics Social Science and Policy Studies Department WPI, Worcester, MA 01609 Posted by Khalid Saeed <saeed@wpi.edu> posting date Thu, 23 Aug 2007 10:03:44 -0400 _______________________________________________

Posted by Bill Harris <bill_harris@facilitatedsystems.com>

Mike,

I won't disagree with your premise. I am curious: are you talking about new ways to show the results (which I take from your Rosling example -- Gapminder is a very interesting tool, well worth exploring -- see also http://gapminder.org/ ) or the models (which I take from your discussion about CLDs vs. stock and flow diagrams)?

As for results, as some of you know, I've been using a different simulator for about the past 18 months, and I've been generating different sorts of graphics that seem to offer real promise in communicating with others, at least in the cases I've tried. You can see some of these online:

What I initially called a trajectory plot, which combines a phase
plane and a time series in one picture
(http://www.facilitatedsystems.com/weblo ... kates.html).

Stick, text, and surface plots, as well as stereograms, which you can
see in a presentation I gave last fall
(http://facilitatedsystems.com/weblog/sd ... beamer.pdf).

You can learn a bit more about this approach at http://www.facilitatedsystems.com/weblo ... mcsim.html.

You can do all these graphics and more with any of the simulators you now use, for MCSim doesn't plot any data. For that, I tend to use J
(usually) or perhaps Gnuplot. Those tools can plot output from other simulators, too; I've done that in the past.

I should have noted three more things. First, I have a couple of additional examples of graphs of results in an article I published this spring called ""Is TAFTO a good idea? Really?""
(http://www.artsjournal.com/adaptistrati ... nt_22.html).
TAFTO is ""Take a Friend to Orchestra"" month, BTW. The eight graphs in one picture were set up, simulated, and plotted with one command.

Second, because I draw my final model diagrams with other tools (typically Dia -- pencil and paper is still my tool of choice for initial design), I have arguably more freedom to design whatever model diagram I wish. That freedom comes at a bit of a cost, of course.

Finally, there may be options to free up our thinking in more aspects than just the graphical. See, for example, my ""'Scientific thinking'
the modern way""
(http://facilitatedsystems.com/weblog/20 ... n-way.html).

Bill


Bill
- --
Bill Harris http://facilitatedsystems.com/weblog/
Facilitated Systems
Posted by Bill Harris <bill_harris@facilitatedsystems.com>
posting date Thu, 23 Aug 2007 08:45:21 -0700 _______________________________________________

Posted by Bruce Skarin <bruceskarin@hotmail.com>

I agree that explaining model behavior and structure can be a difficult challenge, especially when you are not delivering a model in person.
Yet I also believe that there are indeed better ways to present models visually that will provide more cognitively intuitive descriptions for a wider range of individuals. As has been shown in studies by Sterman and others, people are very poor at interpreting graphs. Yet people examine, understand, and correctly interact with dynamics all the time.
We prevent our bathtubs from overflowing, drive cars, and manage (although not perfectly) many other complex decisions and processes.

One might ask then; Why DO we continue to use graphs and static diagrams when we want to explain complex models? I think the answer is similar to the issues that occur between many professions and research areas. Individual differences lead us to better understand different symbolic representations of complex phenomena in the world.For example, some mathematicians are really quite brilliant and looking at formulas and immediately recognizing what is going on. Other folks, like us, prefer causal descriptions. An account will likely prefer a balance sheet, and a mechanic will want an exploded assembly diagram.
Unfortunately this probably means that there is no silver bullet that will work for all, but it doesn't mean that we shouldn't to find a broader range of tools for describing model structure and behavior.

I like the example sent as a conversation piece. While the speaker's explanation is half the power of the message, the visualization provides cues that map to far more than just the few dimensions that we are used to. There is the passage of time in the movement, population in the size, categories in color, and two other dimensions for the x and y position. Presenting data that is correlated (and perhaps causally linked) in this fashion is a great way to help people understand that there is clearly some underlying relationship worth discovering.I took a stab at bringing some similar visualizations to my undergraduate model on terrorism dynamics:

http://feedbackgovernsdyanmics.blogspot.com/

It was inspired somewhat by
the example sent in the original post (see the bubble plot), some of the traditional causal diagram animations, and some animation of link strength similar to the work done by Willem Geert Phaff and Jill Slinger that was presented at the SD conference:Visualising the Effects of Non-linearity by Creating Dynamic Causal Diagrams, by Willem Geert Phaff, Jill Slinger Abstract PaperThere is another post on there that talk about Many Eyes, an online collaborative visualization tool that also provides some more interesting presentation options.

http://services.alphaworks.ibm.com/manyeyes/app

Anyway, feel
free to take a look. Feedback and suggestions are, as always, welcome.- Bruce Posted by Bruce Skarin <bruceskarin@hotmail.com> posting date Wed, 29 Aug 2007 17:27:50 -0400 _______________________________________________