Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages

Go Home

System Dynamics Blog


Modeling for Impact in the Era of Big Data

Modeling for Impact in the Era of Big Data

by | Oct 28, 2020

On the occasion of the field’s 50th anniversary, in 2007, Jay Forrester spoke to us (SDR 23:359-370) about prospects for the field of System Dynamics—his view of the situation summarized by the phrase “aimless plateau”.  He lamented our “slight impact” in government and placed the blame primarily on our failure to ask the big questions, to write books to influence the public (in the tradition of Urban Dynamics, World Dynamics, and The Limits to Growth), and to maintain rigor rather than “dumbing down” our models with “unwise simplification”.

In calling for courage and rigor in SD modeling, Jay was echoing the words of Dana Meadows and Jenny Robinson more than 20 years earlier, in “The Electronic Oracle” (Wiley 1985).

  Meadows and Robinson added a third component that they felt was critical for having influence in the public sphere, namely working collaboratively with other researchers and influencers.

Jay, Dana, and Jenny all hoped and believed that, if only we acted with quality and high purpose, the unique attributes of SD modeling would emerge and ultimately be recognized as more powerful than the intrinsically limited modeling paradigms of statistics, econometrics, and input-output analysis.  

I have to admit I’m not so sure.  Looking at the world of modeling today and who has influence, I’m struck by how statistical approaches still seem to have a strong hold on much of public policy—whether the subject is climate change economics, sustainable development goals, or even the (obviously dynamic and nonlinear) COVID-19 pandemic.  There are some exceptions, such as the Forrester award-winning work by Thompson and Tebbens on polio eradication, but such examples of strong SD public policy impact in recent years are few and far between. 

Consider, for example, climate change and its likely damages for the global economy.  The prominent Yale economist William Nordhaus (a vocal opponent of SD since the 1970s) first published work in this area in 1991, and in 2018 was awarded the Nobel Prize for it.  Yet, this econometric work is surprisingly narrow in its outlook, with its damage estimates biased downward, as described by the Australian economist Steve Keen in his recent article, “The appallingly bad neoclassical economics of climate change”. 

It doesn’t necessarily require an SD model to overcome some of the shortcomings of the Nordhaus work.   An influential study from 2015 is also statistical, but looks at data not only cross-sectionally across many geographical locations (as Nordhaus does) but also longitudinally across 50 years (1960-2010).  This study’s estimates of economic damages over the next several decades are at least ten times greater than those of Nordhaus.  This longitudinal analysis does not shy away from data but dives further, and more productively, into it.

Conclusions?  First, it seems to me that if one is hoping to impact public policy these days, it is important to draw from a large number of cases across diverse locations—the more the better.  This is the era of Big Data, after all, and your model needs to be calibrated and applied to many separate cases to prove its worth if you want to influence the public conversation. 

A good recent example is a study of the effect of weather and air pollution on COVID-19 transmission.  This collaboration between SD modelers and quantitative health scientists drew from data on more than 3,700 different locations around the world, and it garnered significant public attention.

Second, I think we should emphasize our ability to work with longitudinal data.  We know (more than most statisticians and economists do) how to move from longitudinal data to properly estimated parameters, and how to situate these parameters within robust dynamic models.

Of course, we should always remember the importance of courage, rigor, and collaboration in modeling.  But, these days, to be heard widely in the public forum we must also employ both cross-sectional and longitudinal data and calibrate our models to many individual cases when possible.

Recent Posts

Society Governance Updates

Society Governance Updates Welcome, Allyson! New President Allyson Beall King joined the Policy Council as our 2024 President. Her primary role is as director of the Washington State University School of the Environment, which focuses on regional ecologies and our...

Call for Presenters: Seminar Series

Call for Presenters: Seminar Series We at the System Dynamics Society are continually seeking vibrant and knowledgeable presenters for our ongoing Seminar Series. As we unfold the calendar, there’s always a place for more insights, experiences, and expertise to enrich...

Upcoming Events

Webinar on Systemigrams

Have you ever wondered how we could tell a clearer story about complex systems and problems? Are you looking for a tool that can help you navigate and analyze intricate systems more effectively? If so, this webinar is for you. The South African System Dynamics Chapter...

Recent Business cases

A Design Value Calculator: A System Dynamics Boardgame

A Design Value Calculator: A System Dynamics Boardgame EXECUTIVE Summary Product design is a specific form of complex innovation that touches all areas of an organization’s management. While entrepreneurs recognise the value of design, they often tend to focus...

Join us

One thought on “Modeling for Impact in the Era of Big Data

  1. Jay said too that one has more chance doing a good SD job by believing in one’s individual effort than on any kind of help coming from elsewhere. This is what I have understood. I verified this recently with the COVID-19 pandemic. I had more or less finished a model by last June and was ready to apply it (buying a business). But the pandemic made it impossible or very hazardous. Since then, having plenty of time, I started to use my model much more than I had done so far, and surprisingly found a lot of things to change and some bugs too. After 6 months, the model is very different, much better understood and much more credible. I will probably be able to use its policies not before another 6 months, that I expect to be very productive, having still found a lot of things to add or remove in the future. I said that because the future of SD is the sum of the individual future of all people using SD.
    I wanted to say too, that I do not use pure SD and put in my models anything that can help them to better reproduce the reality and find good policies. For instance, I deal with exogeneous discreet stochastic influences by using dynamic programming also known as reinforcement learning.

Leave a Reply