Putting System Dynamics into Practice with Modeling

I spent the week building system dynamics models, with guidance and help from modeling experts via the System Dynamics Society’s annual modeling school. My goal is to translate my causal loop diagrams and graphs into working simulation models and then use 1000, 10000, or more model runs to identify leverage points: places where small changes can effect big changes in a system.

The week was intensive and effective. I used Vensim modeling software, which was originally released in 1990 and underlies the fantastic EN-ROADS system dynamics model of climate change. Later this summer, I’ll also be exploring Stella and Insight Maker, two other options, as well as Ventity, a newer program being developed by Vensim maker Ventana Systems.

In coming weeks and months, my intent is to build working models related to AI risk, climate risk, and other risk topics. Each model will consist of a small central core with a few stocks and flows at first, which I can broaden using a modular approach over time. I’m not sure how much time this project will take, or if I’ll have enough time to go as deep as I want to, but I’m going to start and see where the exploration takes me.

I’ll share what I learn here (along with essays on other topics!).

So, what do these models look like?

I’ll show you what I’m talking about. A couple of months ago, I wrote an essay called Success to the Successful, which explored a system dynamics archetype (a commonly occurring pattern) and walked through a few Success to the Successful scenarios. One of those examples involved gifted students. I wrote:

If Student A is identified as gifted in third grade, and Student B is identified as average in third grade, what happens?

According to the “success to the successful” archetype, a school that identifies Student A as “gifted” is more likely to allocate resources to Student A (say, a gifted-and-talented teacher, extra classes or field trips, and connections to mentors)….

As time goes by, the increased resources provided to Student A further increase the likelihood of Student A’s success, and if Student A in fact becomes more successful, even more resources (e.g., scholarships, summer courses, grants) will probably become available to Student A.

I wrote more on this, but you can read the original essay for that. The point is, how does the causal loop diagram above look when implemented as a working model in Vensim? Here’s one example (note: in the model, you can see that I designated Student B as gifted, not Student A):

In the model above, the thick arrows with a cloud at one end (indicating an upstream input) and an hourglass in the middle are flows, typically expressed as a rate over time (work performed per year, in this example). They flow into (and sometimes out of) stocks: rectangular boxes that accumulate a quantity of something (in this case, success). The thin arrows connect other auxiliary variables and constants.

You can see the relationship to the causal loop diagram in the structure, yet it’s not quite the same. There are elements abstracted away from the causal loop diagram that must be explicitly declared and present in the full model. (As just one example, the “impulse” in the model provides the spark that gets the loops going when the student is declared gifted.)

Here’s a graphed result of running this simulation once:

It’s also possible to run the simulation many times and graph the average results, or to graph specific runs, or to tweak one parameter at a time and then graph subsequent results to see differences. On the front end, it’s possible to build user interfaces with sliders like EN-ROADS has, which make it easy for executives and managers to interact with the model even if they aren’t system dynamics experts.

I’m excited to dive in to more system dynamics models, test hypotheses, and see if I can identify any leverage points. As Calvin said in a famous comic strip, “Let’s go exploring!”