Community Forum : General Discussion Forum
Welcome Guest   
 
 Subject : future of the field.. 04/28/2019 03:41:04 AM 
Jean-Jacques Lauble
Posts: 39
Location
I profit from the fact that the actual president Martin, participates to this forum and that he seems to appreciate my opinions to expose shortly what I think about the SDS.
Being outside of the SDS and only a member, I lack some information. But having been a member for 18 years, some facts suggests some evidences.
One evidence is that the SDS seems more driven by financial considerations than by strict scientific ones. Money drives the world, yes, but the world is not understood by money. To be understood it needs an independent, honest and objective state of mind and not to be biased by financial considerations. Of course, I know that the SDS needs money to exist. But what kind of existence money gives to the SDS. It exists of course, has about 500 attendees at the conference every year, publishes a review, has more than one thousand members, has a web site etc… But is it seriously considered by serious people? Certainly not by me, considering myself as serious. Its domain seems extremely exciting for new comers and it seems to be living from that. It exists mainly with researchers, but not by practical applications. For financial reasons, it accepts mostly poor papers at the conference, to increase its attendees. Unfortunately, it does not attract serious people, only SD professional who feel more or less obliged to attend the conference. This strategy deserves deeply the field. It hides the few competent truly scientific SD people among a mass of mediocre ones and gives the field the look of a sect instead of a strict scientific field.
The solution? Stop organizing a conference, publishing a review and start naming a cat a cat, at least for a while, become a strict organisation, even if it is at the cost of reducing deeply its fund. After several years (at least 10 to my opinion) it will start to be considered again as a scientific field, that recognizes honestly its strength and its weaknesses and be able to grow again.
JJ
Last Edited On: 04/28/2019 03:44:18 AM By Jean-Jacques Lauble
 Subject : Re:future of the field.. 04/29/2019 01:10:31 PM 
Martin Schaffernicht
Posts: 40
Location
Hi Jean-Jacques,

thanks for trusting in that stating this description may be useful for the field while I am serving as the President of the SDS. And I thank you for this message - it may help to revive the exchanges on our Forum, and it is certainly a very important topic!

I'm now having 16 years of membership in my experience, and I remember how surprised I was at the 2005 Boston conference to hear so many experienced SDlers (mainly from the MIT context) stating how poor the quality of the presentations was. At the 2007 Boston Conference, Jay Forrester challenged people talking about an "aimless plateau" and the curse of publishing as the quintessence of academic work.

I've served as chair of a conference meeting thread dealing with reviewing and quality for 3 or 4 years (I think starting at 2007) and then we gave up because our recommendations were based on a set of criteria that seemed to leave important aspects out of consideration (in the mind of an important share of the decision-makers). I do have a "theory" of what is going on in the SDS, but so do others, and several sub-groups have diverging theories with respect to the development of the SDS. If we put much emphasis on our (hypothetical) theories now, we risk a fruitless confrontation.

I guess many SDS members who care for the field and for the SDS can agree to the following statements about the SDS and the field.

I. - SDS
A. Things that have not changed in the SDS:
1) We have a persistent pattern of being a small community of people with a SDS membership headcount which increases from a couple of hundreds in January to around 1000 in December and a very high turn-over in the early years of membership.
2) The SDS members are individuals who can be grouped into three major types: (1) Reserch oriented proficient modelers (you can call them experts or serious or "high-qualilty" - there are different names for them, and I do not know which one really fits), (2) learning oriented newcomers (who seriously want to become good at "doing" SD) and (3) "general interest" individuals ("systems citizens" you might say). The share of (1), as compared to (2) is small. This has not changed over the years.
3) Roughly half of the SDS members work in academy.

B. Things that have changed or started to change in the SDS:
1) SDR articles per year have diminished about 50% over the past 10 years.
2) Conference rejection rate has been low (I guess in absolute terms but also as compared to other conferences like AOM or INFORMS - but I do not have exact information on that), until it suddenly went up in 2018 (BTW: if other conferences have a higher rejection rate, this suggests that they also receive a lot of submissions they find to have low quality; so I wonder if our problem with quality has to do with the fact that low-quality work is being made or that it is not being rejected).
3) I see an increasing number of people in the SDS working to advance an "evidence-based" modeling stance (I hope I do not over-interpret "serious" as related to "evidence-absed"?).

II. - The Field
A. Things that have not changed
1) A unknown number of individuals in the world somehow get in touch with SD. Most rely on books, software manuals and Internet resources for their learning. In most places, there are no educational institutions with a sufficient staff of individuals qualified to lead learners (and we all agree that, beyond the initial phases of learning, supervised practice is needed to become proficient).
2) Those individuals who teach SD are usually "professor of X", where X <> "system dynamics" (you can hardly find a "chair of system dynamics" in the world of universities)
3) The methods and conceptual tools are framed by "stocks-and-flow", "table functions" and qualitative representations of "loops". Loop-level analysis (pathway participation or eigenvalue) exists but has not made its way into mainstream SD .
4) Software tools for formulating and running models - including structure and behavior visualization - has remained stable (novelties like causal tracing exist in the major software packages).
5) Other modeling approaches (discrete events, agent based) are present in fields with huge membership populations and appear to have much interaction with other disciplines and fields. SD people have not published or communicated much joint work with other communities (as far as I know).

B. Things that have changed
1) Stock-and-flow models can be developed and/or run in other environments (Python, Modellica, R)
2) There is an increasing recognition that SD is not (any longer) well-described as "feedback-driven stock-and-flow models" (or, if you will, the "14 obvious truths" stance, which can have been appropriate when it was originally made, is not a useful stance nowadays). I think this is an opening.


In my description, I took care not to introduce any of my own causal attributions. This does not mean that I have none - it is just that different people hold different mental models. In the past, the inability to move beyond the (open or concealed) collision between contradictory mental models of the development of the field and/or the SDS has lead to a lack of action. Not doing something new or different means that things and trends stay as they are. So for the moment, it is already something if a majority of us can agree on what the facts are. The time for a dialog on causal attributions will come, too.

So, where do we stand today, and where will we be in December 2019?

As of today, we (the entire Policy Council and its Committees, together with the Executive Director and his staff) are working hard in a number of topics, and several news will be soon coming on. It is extensive work, and I take the opportunity to thank all who are giving their time to make this happen. We will have news by the time of the Albuquerque Conference.

Will we save SD? Of course not! Becoming the field we aspire to become cannot be done by the Executive Director, his home office and a bunch of voluntary PC and Committee members alone - we only provide a platform. It takes work from people who are able and willing. Some may develop a piece of code for analyzing models or visualizing their structure, or for going beyond the Vensim reality check. Some may mentor learing-oriented members. Some may share advice, others learning materials. Some may go ahead with exemplary work, and others may offer rigorous & friendly critique (we want to help people get better and not go away). Some may link up with other communities and bring in new people who bring in new ideas and can ask difficult questions.

In case you are skeptical now, I think this is Ok: after all, I'm certainly not the first SDS President to try that kind of thing. Sure, we will need to assess this Presidency when the year comes to its end.

In the meantime, I am always happy to receive ideas, questions, comments and suggestions!

Greetings from Chile,
Martin
 Subject : Re:future of the field.. 04/30/2019 02:31:26 AM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Martin

Thank you for your long post.

I will take the time to answer points by points.

Thanks again.

Best regards.

JJ
 Subject : Re:future of the field.. 04/30/2019 12:33:14 PM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Martin


The first problem about system dynamics is its definition. I do not know a precise definition to which I agree. Of course, a model is a differential equation and if one runs it, one integrates it. But it is not enough. Lots of elements you see with specific definitions are rather ambiguous. I remember having a course about real world modelling with Jim Hines years ago, that was rather subjective and not based on a real scientific method. The first thing to do if one wants to progress again, is to decide what one is talking about.
I am presently studying dynamic programming with Bertsekas MIT opencourse and the way it is exposed is strictly scientific and not ambiguous compared with SD.
This work of definition may take a lot of time, but I think that it is essential. Once this is done, it is possible to define the purpose of the SDS relative to this definition. Is the SDS staying strictly within that definition or accepting to study neighbourhood fields.

Best regards.

JJ
 Subject : Re:future of the field.. 04/30/2019 01:52:11 PM 
Martin Schaffernicht
Posts: 40
Location
Hi Jean-Jacques,

I personally share the statement that the definition of just what "system dynamics" is presents a challenge, and it does so in several aspects. In my opinion, some of these aspects are related to the historical development of "SD", which was known by several names ("industrial", "urban", "world" ... I may miss some) before all these specific domain names were fused into "system". I do not pretend to be entitled to say what Jay Forrester meant to express in the things he expressed, but it seems to me that he posited SD models to be developed in search of usefulness rather than truth. If this is so, then we need to state what we mean by "scientific" in the context of SD - this should not be a tough problem because in areas like Simon's "design science" and the development of technology in general, one cannot say that the search of usefulness contradicts a scientific manner to work.

In one of the "fireside chats", Jay Forrester was asked what he would like to be remembered for, and he said "the guy who killed differential equations". I believe he meant that developing a diagram is more helpful for humans than directly writing equations - but I'll be happy to learn better if someone tells me this was not what he meant.

There are a number of basic (ontic) assumptions about the structure of the world woven into SD. One is that the feedback loop is the basic building block of social systems (taken from "Principles of Systems"). I suspect this holds for "systems" in general (not only "social"). I also think this assumption is not SD's monopoly (at least not nowadays). Then the statement that any loop must contain at least one "stock" (state variable). This seems to be generally applicable, too. We also assume that the situations we address require computational (simulation) modeling because they are too complex for mathematical modeling (analytical solutions), right? We further assume that many relationships between variables are non-linear and that different relationships have different speeds (-> delays).

We usually do not distinguish different levels of description in a system's structure: the stocks and flows are represented at the same level as the feedback loops. This may elicit some criticism because one ma well stat that a loop "emerges" from its participating variables and links - this would even facilitate dialog with the realm of "complex adaptive systems", in my opinion.

We have this notion of "dynamic complexity", but at the same time we know that the "complex adaptive systems" community uses a definition of complexity which is much closer to the one that seems "traditional" to me ("the number of possible states", right?) Well, complexity is not only an objective aspect of a system's physics, there is also a cognitive component (because if our mind cannot muster a system's "complexity", we find it complex). I think that the research concerning the "stock-and-flow error" and the "mis-perception of feedback" is very useful - even though I also perceive quite a distance between the management literature on "dynamic decision making" and the psychological literature; well, but even outstanding people like Kahneman or Thaler have suffered from that, so I should not complain :)

Referring to the behavior of systems, we have a series of basic statements relating the behavior of systems to structural components (reinforcing feedback drives acceleration or exponential behavior and so on), as well as a set of "wisdom" statements ("nothing grows forever", "worse-before-better" behavior and so on).

Then there are some methodological basic statements. We follow the "endogenous" approach, which also entails guidelines like making the elements that restrict reinforcing loops explicit, and so on. But more importantly, the methodology contains statements about (1) the initial anchoring of simulation models and their elements on evidences (call it reference modes or data in the wider sense, including the semantic correspondence of model elements with real-world elements) and (2) the measures taken to assure confidence-building by means of diverse tests.

Maybe we also need to address the type of situations addressed. SD seeks "dynamic problems" that deal with "policies" where aggregation is not a problem (is this so?). What kind of problems do "agent based" or "discrete event" modelers look for?

Coming back to history, maybe the early years of SD were marked by opposition with other communities (recalling the "14 evident truths"). And it may well be that at that time, there were important reasons to have these debates. But then again, "things change over time" and nowadays, the situation may be different. A good starting point may be found in Sterman's introduction to the 60th Anniversary Issue of the SDR, but I also find it good to remember that the last time I've seen conference presentations dealing with "discrete event simulation" or "agent based modeling" were between 2004 and 2007 (i I missed some, my fault).

In my personal opinion, a systematic comparison between SD and other areas (complex adaptive systems, ABM, DEM, ...) with respect to problem types, outcomes, ontic, methodic and epistemic "basics" would be most useful for every one - and it would make an excellent "Special Issue" for the SDR. So if your post about the future of the field can serve as a spark to ignite the necessary dialog, that would be a thing to be thankful for!

I therefore hope that may members read this thread and that some of them (of us) feel motivated to invest the time in going through the books and articles where elements of the definitions were laid out.

"Saludos",
Martin
 Subject : Re:future of the field.. 05/01/2019 09:38:43 AM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Martin
In your answers, there are many topics that must be taken, one after the other.
I like to make deep study and it is always dangerous not to do it, because one takes the risk of missing something important.
I will then focus on the definition of SD and the reference you made about its history.
My view of the question is that Jay was of course always contacted by people owning a problem and tried to solve it. He was then naturally oriented towards usefulness.
The first problem about a factory was rather technical and physical and to my opinion easier to grasp than the following ones. One big difference is that his work apparently solved the problem (I do not know how much satisfied where the owners of the problem). His industrial study was then used and produced some results.
The some years later, approximately 8 years later, he has the occasion to meet the past Boston major who exposed his problems. Jay studied the problem, made a model based on the major explanations.
But was his work used by the people ruling Boston at this time?
I never heard that they did.
The characteristic of this urban study, is that the problem was much larger and the problem owners difficult to spot. Were the problem owners the Boston citizens or the people governing the city; In the case of the industrial problem, there was no question about who were the problem owners like the case of Boston. One big difference is that in the urban case, the people governing the city are susceptible to change with the elections. To resume both problems have big differences: size of the problem, problem owner identification, possibility of the problem owners to change in the future, and finally the fact that it seems not to have been used, at least at this time.
Just a parenthesis: I like facts-based studies and I think that to understand a situation one must look very closely at the facts.
If one compares both studies, the first one had more chances to interest eventual users than the second one. The first had been used with some results and was easier to copy, mainly because is was simpler. The second one was less susceptible to be used due to its added complexity and its lack of practical use.
To my opinion one must try to understand why the first study was not vastly used after its first experience. It was of course only published in 1961 and it takes always time for such a new method to disseminate. I think that from the beginning the method interested more the research community and found its future in the research and academic world that is less concerned by the necessity to generate concrete results. And Jay himself followed this path by working on the Urban problem.
I think that there was a huge market in the industrial and commercial world, at the condition to start slowly and to take the time to build a strong and practical method. But it was not the case, and the research sector killed all possibilities to develop a profession with true professionals.

The field after having succeeded with industrial and commercial problems during 30 years, could eventually start to address more complex social problems but always in mind practical utility and with extreme care.
The result is a field mostly research oriented.
I have rented during more than 40 years around 1000 (some years more) trucks, vans and cars which makes with about 4 rentals per month, about two million short term rentals without having been ever suited and sold after use, about 12 thousand vehicles without complains. If we had been working like Sd people work, we would have run into bankruptcy very shortly.

I think that the Sd community should make clear distinction between Sd research and applied SD especially for the clients who may be interested by the method. There is research in many fields and people interested in that field for practical applications will not lose their time studying research papers.
Now the research sector has no obligation of results but the applied sector has.
To start developing a successful applied sector (that presently does not exist, sorry to say that) the SD community should start looking at eventual past successes. And I mean by success a study that has been really applied with success. And start to think about the replicability of these successes: for instance a success that required huge resources is difficult to replicate. One should search for applications that require a reasonable amount of resources that can be found in the market: ability of the modeler and the client.
One has to define a product (SD common applications) that can be reasonably built and that can satisfy the needs of some clients.
This will require some work and some time but starting from that point may generate later on great successes. In fact, you have a big business that does not work properly and the best thing is to start again with something little but viable.
The subject is not closed.
But one thing after the other.
Best regards.
JJ
 Subject : Re:future of the field.. 05/02/2019 08:58:36 AM 
Guido W. Reichert
Posts: 26
Location
Hi JJ,

Thank you for bringing up this important topic. You complain about the SDS being driven predominantly by money. Yet, looking at its obvious domination by academia I would argue against this: It is driven mainly by the interests of academics - which also includes money.

Systems Science to me personally has a very strong appeal: Jump over the fence, combine different perspectives and rediscover the renaissance man (politically correct: renaissance person!) as Jay has put it (as far as I remember it). But instead of joining forces and finding a common language (e.g. Mathematics to me is the most universal language we probably know) we introduce our own lingo, talk about "killing differential equations" (instead of building a path to better applying them), and stress our "uniqueness" as a field. The latter is a postulate of marketing which is never looking for more community but always for more separation to maximize rents.

Simply consider the following terminology:
  • System Dynamics ("Science" or "Methodology"?)
  • (Dynamic-) Systems Science ("and Methodologies" or just "Theory"?)
  • Modeling and Simulation (split into M&S Science, M&S Engineering, and M&S Applications)

Where are the boundaries? What makes us different? Do we need separate organizations/journals and is more always better? Why do we (how often?) use different terminology/definitions? For example, in the glossar "solution interval" is equivalent to the TimeStep or dt, but to me the solution interval is the complete interval for which a model of differential equations is solved. A connector is given as an information link while in Modelica a connector is an interface for a system or subsystem that can also be a mass connector (unknown in most SD tools which only talk about information really). Connectors in Modelica are connected by connections -- interestingly the term "connections" is not explained in the glossar, but is confusingly used to explain "System Structure".

Most SD tools (the field seems dominated by proprietary tools) work with assignments (eg. lhs := rhs) and somehow hard-wire the method of solution (e.g. fixed step solvers) into the equations (stock_t+dt := stock_t + dt * netflow_(t,t+dt) ). Are we then still doing System Dynamics when we are using equations (e.g. == as in Modelica) and then build or models in continuous time so that different solvers can be applied to them (e.g. no solver-parameter like TimeStep may appear in an equation)?

You mention that there are not more successful and visible practical applications. But, let me ask you in return: Have you published your very own models that you are using as a problem owner, so that we and others may learn from them? Have you demonstrated their utility by giving say additional profit from applying the models vs. the costs of modeling?

Like other "intellectuals" I do see ownership as a two-sided affair: It may avoid the overexploitation of "commons", but on the other hand may introduce allocation problems. Isn't there also a difference between a "problem owner" and a "model user"? As an entrepreneur and owner of an SME your situation is very different from a public company's: Are the shareholders the "problem owners" and what is management then? Who shall have how much influence on the model's specification? What about climate shange, feeding the world, fighting poverty etc. -- who owns these problems and should they even be owned by individuals or groups?

In competitive environments, why should I publish a successful model? Why should I teach others and thus diminish my competitive advantage? There may be a lot of useful application of SD around of which we do not know for simply this reason.

In politics and also in business, using models to better control a problem means giving up some control. It naively neglects the question of power: There is no single "best solution" because there is no unique utility function for groups. So, why should I give up power to have a model find a solution - especially in a multi-stakeholder setting? A model may increase the risk, that what I want is not implemented. Good luck with convincing the likes of Mr. Trump to employ quantitative models. ;-)

Academia has the advantage of "having to publish" and having to be "transparent". While this has downsides on the application side, it at least is available! What seems unfortunate to me is, that tenure track academics (without any personal risk and making use of cheap "slave labor" aka PhD-candidates :) ) are competing on the market for applications and are commercial distributors for proprietary software & tools. To me in economics lingo that is market distortion. Who then is researching the "hard" stuff and (reusable) "theory" (remember Lewin: Nothing is more practical)? There are so many academics in this field and yet it has taken 40 years to publish "Analytical Methods for Dynamic Modelers" (which appeared 2015)! More books have probably been published about Wolfram Mathematica than on System Dynamics since John Sterman's book has appeared in 2000.

My take on the field is (I would like to be challanged here):
  • System Dynamics has a lot of appeal for newcomers and makes for a great gateway discipline into Systems Science and Modeling & Simulation
  • It is not "killing ODE" but making them disgestable with great opportunities in education (K-12 onwards)
  • Once you are well-versed in the elementary stuff and you are about to do hybrid modeling, Bayesian identification of uncertain dynamic systems, stochastic optimization etc. you will not find enough support in the SDS community. You will almost be forced to turn to "harder" communities like INFORMS, EUROSIM, and domains like electrical engineering ("back to the roots") where these topics are really driven forward
  • To me personally these days using Modeling and Simulation as a term feels more appropriate and encompassing than System Dynamics and "closing the loop" talk

A possible path forward? I would suggest to open up to other disciplines. Don't "kill mathematics" but show how SD provides more intuition and guidance in formulating differential equation models - pave a path to using the most universal language there is. Link to the hard stuff and show how this eventually can be applied to social systems. Make publications in SD be easily compatible with other fields and vice versa. That may mean to start speaking a similar modeling language. ;-)

Best regards,
Guido

PS: And yes, we should maybe consider to model systems (be as deductive as possible) not just problems (i.e. inductive modeling). :)
Last Edited On: 05/02/2019 09:11:23 AM By Guido W. Reichert
 Subject : Re:future of the field.. 05/03/2019 03:44:22 AM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Guido.
You wonder about my having never published my works. It is only partially true as I once published one and I will write about it later.
There are many reasons for not publishing my works.
Before doing that, I will explain what kind of works I am writing about. I will not try to define it as SD or DE or agent based or anything else, all these methods having often rather unclear definitions (especially SD). So my reference is very practical, I am writing about models written in Vensim and simulating a problem of mine. And my problems do not necessary fit to the traditional SD definition.
And this one of the reasons of not publishing.
But the main reason is that in 17 years of Vensim use, I have never been able to build a model having a concrete usefulness. And I do not see the interest of publishing something that has never generated any real utility.
I will not detail here the reasons of my works having been not useful, I do not think that it is the point of our discussion.
I have of course built lots of models, generally of a certain size between 500 and 1000 equations rather heavily subscripted, running slowly. To publish it I had to translate them from French to English, add many more equations comments and explain in plain text, what the model is about.
A considerable amount of work that I was not ready to accomplish for the following reason.
I already explained that the models were not useful, but there is another reason for not publishing.
I think that to appreciate a model, one must have a minimum of knowledge of the subject and if it is complex, preferably a practical knowledge from the subject and if it is less complex the possibility to understand it with a logical mind and some general experience too.
If I had published a model relatively complex, nobody would have studied it, by lack of interest or lack of capacity.
I would then add that last year, Len asked me to help the strategic committee in its work with Kim Warren at its head. I received different models about the future of the field and not wanting to add an additional model I decided to build a model about the growth of a consultancy. A very small model with no subscripts on only one Vensim view and it took me time to build such a small model, the first models were much bigger. It took me one week of time to build it. The mode was readable with the free Vensim PlE or the free Vensim Reader. Kim pretended not having the capacity to read that model (?) and said that he would show the model to Len. I had never the smaller return from my work. I may add too that at the end of the year, I sent comments to people concerned by the strategic committee about one of the model (you know which one) describing the future of the field, showing many problems in it. The return comments were: ‘we always appreciate comments’ this is all. I must add that I sent the same model to you and I had not the slightest comment about it coming back from you. And you want me to make the effort to publish anything?
I must add, that after having lost my time building that small model and studying the model with problems that took more than a week, nobody asked me again to help the strategic committee and that rather surprisingly there was a vacancy in it. You can see that on the SDS site in the governance part. I prefer that nobody proposed me to be a member of that committee, not wanting to lose my time in a dormant one. Last year i was asked too to review papers for the SDS conference. I had to review two papers: one had no model joined and when I said that I was not able to review a paper using a model not available, the answer was: I had to do the job without it! The other ones was deeply wrong had no units in it and was calculating three times the same thing and being then three times to big. When I was asked this year to review again some papers, I explained that reviewing was difficult for me, not being an academic or a teacher. This to explain that I have no personal interests in being a reviewer or an officer in the SDS structure, not selling SD or pursuing a career in that field.
I hope that you will understand the reasons of my not publishing anything.
I will try to answer to the other points of your post later, because I prefer to deal with one subject only in one post.
Best regards.
JJ
 Subject : Re:future of the field.. 05/03/2019 07:44:41 AM 
Guido W. Reichert
Posts: 26
Location
Hi JJ,

While waiting for this discussion to proceed (topic by topic) let me quickly reply to your answer.

Looking back into my E-Mails we did indeed discuss George P. Richardson's "Small Model of the Growth of a Management Science Field" presented at the 2014 ISDC in Delft at great length. And while we did not go too deeply at formulating a new, improved version of that model. I am sorry for that omission, but we nevertheless spend some time with the discussion.

That model is a great case in point for the woes of the field - the main message of that model (looking at those runs!) seemed to be: Nothing can grow forever, but members of a new field of management science can! (The abstractly named field here is of course the thinly disguised field of System Dynamics.)

There is no excuse for publishing such kind of "insight model" in my opinion. Somehow the tenet "all models are wrong" has gotten the exonerating spin of Paul Feyerabend's famous quote: "Anything goes".

Among the surprisingly many faults in that model (e.g. you could have a terrorist attack killing a great deal of experts and the field would grow better than without such a "tragic event" if I remember correctly?) we especially looked at the little aging chain modeling the transition from novice to expert.

While a delay structure in reality most often will have leaking (i.e. a separate in- or outflow next to the main transition from aging), the correct formulation in models surprisingly seems to get very little attention (cf. Dangerfield and Fang (2004), The problem and solutions in using delay functions with bifurcating flows in system dynamics models available online).

Reproducing results and scrupulously testing models still seem to be a costly and unbeloved task. Jumping to conclusions - dangerously warranted by pointing at a first quick insight to inspire thinking - seems to be the favoured approach?

How many models can we download on the SDS pages? Looking at the case repository there is very often only an author to approach and the papers cited do not show equations often only "structure" which imo is worthless without equations.

Where will this lead to?

Best regards,
Guido

PS: Footnote 1 to the published model states:

"This work is a joint effort of David Andersen, Roberta Spencer and the author. However, the responsibility for details of the model and the paper rests with me. We have benefited from reflections and contributions of others, including in particular John Sterman, Jack Homer, Kim Warren, John Morecroft, Peter Hovmand, and Jay Forrester."

So pretty much the "SDS hall of fame" has been involved in coming up with that model - really?!
Last Edited On: 05/03/2019 08:03:01 AM By Guido W. Reichert
 Subject : Re:future of the field.. 05/03/2019 10:22:27 AM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Guido
The model I was talking about was not Richardson’s model but the one I built about the growth of a consultancy that I remember to have sent to you, and from where I had no return. I do not blame you as we had a lot of discussions and the mail may have gone unnoticed by yourself.
Another remark about the experts being killed. The strange result is not the one you mention. But I do not think that we can here detail all the conceptual and simple equation errors in that model.
Now did the people mentioned really participated to that model building and to which extent?
The problem is that it was published in the SDS review. But if one considers all the SD publications in one year there are examples that are much more horrible than this one. One has to look at the 200 papers published at the SDS conference to find some good examples of very bad work.
But I think that in any organisation there are good and bad things and one must learn to live with both and to act without expecting things to change. What I was suggesting is to define among the very large domain of applications, a domain where there is a need among problem owners and the capacity to find people to do the job or to train them. To resume before wanting to save the world and finally not saving anything, it would be better to restrict considerably our pretention and do something really useful. To resume define a product that can be produced and that can be sold to satisfied clients even if that product seems simple to the academics. Clients mind about solving their problems and not about how they were solved. Now is this strategy possible. I have a doubt because it requires the participation of people practicing SD, using different tools and competing with one another. Will these people accept to work together and will it be possible to coordinate their efforts?
Best regards.
JJ
 Subject : Re:future of the field.. 05/05/2019 08:50:57 PM 
Martin Schaffernicht
Posts: 40
Location
Hi Jean-Jacques and Guido,

A lot of information in your posts since Friday was new for me, specially that you (Jean-Jacques) have developed a field development model. If you can send me a copy ([email protected]), I'll be thankful (I cannot promise that I'll be able to make enlightening comments on it, but I will positively try to understand it.)

So, the other model you are referring to is the one behind "On the growth of the system dynamics field" (2018) dating back to 2014. You both appear to have invested some time in discussing that model, but of course I do not know which aspects you have looked into. Guido's comments concerning the terrorist attack and the ageing chain "novice -> expert") have not been detailed enough for me to get the points you are referring to. A priori, I would expect a small, rather exploratory model not to deal with such perturbations like a catastrophic loss of "experts". I'd also expect there is an assumption that the absolute number of SD experts will not be large enough to bump against resource restrictions limiting the growth. The "novices" stock has a "becoming inactive" outflow, so I did not understand Guido's remark concerning the formulation error.

I'm very interested in knowing more about your examination of that model, and - really - I'd like to see Jean-Jacques' model. The reason is simply that I care about what happens with the SDS (and the field, too), and of course as the current President, I feel a huge responsibility. Ain't that remarkable: I'm 100% "academic", but my responsibility here is 100% "real-world" (a term I dislike because academy is one part of the real world, but I lack an alternative term). So my interest in this discussion is rather practical, and I do not intend to develop a model (which would, sure, be intellectually attractive - maybe someone else will do that. I know that the "growth of the field" modelers have deliberately made simplifying assumptions, so that model is but a "snapshot" and could be further developed).

Inside the PC, we are quite clear that "SDS members" is not the same as "system dynamicists". There is not one single aggregate of "members" either - rather, we are thinking in terms of research-oriented advanced members, learning-oriented newer members and general interest members; these groups are different but they interact. At the same time, it has become usual to distinguish different levels of engagement in the members of associations (engaged, rank-and-file, value-seeker). Add to that the fact that learning to "do" SD is not a matter of transforming "novices" into "experts" directly: we use the Dreyfus&Dreyfus model, which states novices, advanced beginners, competent, proficient and then expert. So we know (and so did the author(s) of the model you mentioned) that our reality is more intricate than that model. We must also take into account that the stock of "potential future members" is not infinite and that many individuals may be attracted towards other fields and/or associations. We also see that individuals go through a life-cycle starting at school (so the actual "pipeline" is quite long). We do not have one system-dynamics model, but different works have had an influence on our mental models.

I hope these comments let you see why I'm interested in Jean-Jacques' model and in your critique concerning the other model: if you share that with me, I'll have a chance to improve what little I can do for the SDS.

So I look forward to receiving more information from you,
Martin
 Subject : Re:future of the field.. 05/06/2019 03:38:11 AM 
Guido W. Reichert
Posts: 26
Location
Hi Martin,

I am sorry for not having been clearer - looking back upon my post it must indeed partly read "gibberish". To be fair, I have to remark that Jean-Jacques (JJ) deserves most of the credit for bringing the errors in Georg Richardson's model to my attention. I still cannot understand, why his (inmo justified) critique has not been given the appropriate consideration and publication?

Regarding the "terrorist attack" JJ has simulated a scenario where novices and experts take a cruise and 50% of them are killed by terrorists (using a pulse signal - one may also use the RealityCheck feature in Vensim for such behavior-behavior tests). The model as it is formulated shows, that such an unfortunate event is really quite becoming for the field. Plausible?!

The other error I mentioned concerns the transtion of Novices into Experts. The model has the stock of Novices (here simply: n(t)) being drained by two first-order material delays:
Code:
n'(t) = - a n(t)/te - (1-a) n(t)/tn a = tn/(tn + te)

Where tn and te are the time constants (in years) for novices staying in the field and how long it takes to become an expert. The variable a gives the "Fraction of Novices becoming Experts" according to Richardson's model.

Let's assume for test purposes that we just look at this formulation and that we will not have an inflow to novices which are initially set to 100. Richardson uses tn = 10 te = 15 and accordingly we come up with a fraction of 40% of Novices becoming Experts.

After running the model for 50 years we will have 30 experts - not 40. That error (indicated fraction to become experts - actually observed fraction of initial novices observed after 50 years) will grow as te increases.

It is in my opinion very misleading and maybe even outright dishonest to call a variable "Fraction of Novices becoming Experts" that gives a fraction that will never be reached. We should have called this variable "Nominator for the fractional rate turning Novices into experts" and everything would have been fine.

Leaking a material delay or a conveyor in a meaningful way is not an easy task as the paper I cited shows. Dangerfield and Fang have complained that at the time of writing (2004) the topic had pretty much eluded SD literature.

One can analyticall solve the simple model structure and come up with correct formulations. A simpler way is to split the inflow into a stream that will stay Novices and then leave the field and another one that will continue to become Experts. Another simple way is to have the time constants for the actual transitions be equal.

Doing differential equation modeling (even when they look more easy as integral equations) is hard. A lot of models do not bother to mention that a fractional growth rate (CAGR) for discrete compounding needs to be adjusted to log( 1 + CAGR) to be used in a continuous time model. Since Euler integration will not come up with the correct value anyway (unless the time step is really low) - why bother?

But any student in K-12 will come up with the correct value using a calculator, why not a dynamic model? The solution imo is to use predefined components that will automatically transform an input fraction depending upon a parameter (e.g. "Discrete rate given?"). Such components may also take care of leaking in a robust way.

Again, System Dynanmics should not "kill mathematics" by dumbing it down, but make it palatable. We should maybe see modeling itself as a complex dynamic task and design it, as Jay put it, in such a way, that normal people can do it.

Best regards,
Guido
Last Edited On: 05/06/2019 03:51:13 AM By Guido W. Reichert
 Subject : Re:future of the field.. 05/06/2019 02:23:03 PM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Guido and Martin.
I must precise that the model I built is not a model of the growth of the field, something that I feel not really able to do properly, system dynamics being not sufficiently defined and having different definitions. The only thing that can be modelled is the growth of the SDS, something that is well defined but that I do not feel able to do, having not enough information about it and not knowing the SDS from the inside.
I built a very small model of the growth of a consultancy that starts working in a field having a bad reputation generated by low quality generating by low pricing generating low revenues generating low resources generating low quality a feed back loop. I sent the model to Kim Warren who told me that he did not know Vensim (the model was very simple with no subscripts on one view) readable with the Vensim PLE or Vensim reader! He told me that he would pass it to Len and I had never any return.
About Richardson’s model, I do not think that many people have been working on it, but it has several problems and plain bugs in it, that I detailed shortly at the end of last year, but the only answer was: we always appreciate comments!
We can if you are interested, start studying one by one the problems of that model and discuss about them. It is up to you. I do not know if it interests a lot of people
It will be the occasion to work together openly and publicly on a concrete problem, Something that I have never seen in the SDS.

Best regards.
JJ
Last Edited On: 05/06/2019 02:25:18 PM By Jean-Jacques Lauble
 Subject : Re:future of the field.. 05/07/2019 06:41:19 AM 
Guido W. Reichert
Posts: 26
Location
Hi Martin and JJ,

I want to address the exponential growth demonstrated by the model (imo a serious flaw) here separately. Martin, you admit, that the world is more complicated than the model:
Quote:
We must also take into account that the stock of "potential future members" is not infinite and that many individuals may be attracted towards other fields and/or associations.

But you then continue with what I would call an outright apologetic statement:
Quote:
We do not have one system-dynamics model, but different works have had an influence on our mental models.

Is this to mean "anything goes" as long as it is (somehow) "useful" because essentially "all models are wrong" - as I had pointedly suggested? Please elaborate on what exactly you mean with your statement?

There is no reason, not to have very small insight models (on the contrary). But adding a stock of "potential future members" imo is an absolute necessity for the model at hand to show correct behavior. I remember that not too long ago there was a longer article in the SDR that mocked econometric dairy models because there was milk being produced without cows [1]. I can't see why we publish such a critique when we ourselves do not stick to it and make our models robust with regard to extreme conditions? That contradictory behavior indeed rather befits religious groups that see faults in other "sects" while being apologetic about their own ones.

There are deeper issues with the model published by George Richardson and I hope that JJ will address them here openly. That remains another issue, Martin, that you have not answered in a satisfactory way: Why has the critique that JJ has written been "swept under the carpet"? After all, we cannot "prove models right", but we can criticize and improve them. Modeling is very closely related to software development and bugs need to be "tracked, pubslished, and corrected" -- best with version control.

Best regards,
Guido

References:

[1] Olaya, Camilo. 2015. Cows, agency, and the significance of operational thinking. System Dynamics Review 31(4): 183-219.
 Subject : Re:future of the field.. 05/07/2019 01:00:06 PM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Guido and Martin
About Guido’s question about slipping the model under the carpet, I imagine the most probable reason.
Model builders know that no model is perfect and that all models are wrong.
I had no return, because SD people in charge, thought that my critics were probably logic, because every model can be criticized up to a point, and they had not the time to study deeply Richardson’s model and my critics.
And I will profit from this fact, to emphasize that it is easy to build models, but much more difficult to analyse them and to extract from them useful insights.
The mathematical power of computers makes possible to build big models that represent the reality but with a lot of parameters, difficult to prove structures and with equations difficult to prove too.
The general conditions of making an SD study, are extremely constraining and makes the realization of a useful SD study difficult and costly, and it is mainly the reason of the lack of general success of the method. One of the reasons is that people expect to much from SD studies taking into consideration the power and sophistication of the tools used. This generates unsatisfaction because the people do not know how to restrict their pretentions. One can see that in many published models. I think about the counterinsurgency model of the Afghanistan war, that was to my opinion to pretentious and finally did not produce anything useful. To my opinion the key to use cleverly the method is to set a purpose that is adapted to the method and to the resources available and always limited.
About the study of Richardson’s model, I wanted to say too, that in the month that George Richardson had available to construct his model, I would have probably done a worse job taking into consideration the fact too that it was probably not seriously studied by other people.
But the model was supposedly used to decide of possible policies and I think that the main conclusion of it, is not as sure as it looks like: hiding bad SD studies.
I want to say too that criticizing this model may generate some insights, but I do not pretend anything about their usefulness.
This being said if Martin as president of the SDS is ok I am ready to start the analysis that will take some times: several months if it has to be done properly and taking too into consideration that I have other things to do and that I am not convinced of the utility of the investment, at least, for me. Not convinced at all.
Best regards.
JJ
 Subject : Re:future of the field.. 05/07/2019 01:14:18 PM 
Jean-Jacques Lauble
Posts: 39
Location
I think too that the main utility of studying Richardson's model may not be the possibility to improve the SDS strategy but to learn if it is possible to work together publicly on an SD model.
 Subject : Re:future of the field.. 05/13/2019 02:24:57 PM 
Martin Schaffernicht
Posts: 40
Location
Hi again,
sorry for the long delay - my regular work + volunteer work sometimes takes up all my time.

Thanks for the explanations Guido. I will not argue in favor or against George Richardson's model - I'd need to dig up everything that has been written about it by him and then surely have a conversation with him, but I simply cannot do that currently. (BTW: I do not personally subscribe to the "killing differential equations", I prefer the "friendly algebra" slogan.) (BTW 2: would you agree that SD is closer to "computational modeling" than to "mathematical modeling"?)

Jean-Jacques mentions the large amount of time required to develop a simulation model, and as far as I can tell, this is the reason for being based on mental models with the influence of diverse computational models. Years ago, when the then-President Kim Warren pushed hard to get us give information in a model-to-be, we realized that we do not have data; for instance, I cannot tell you how many universities in Latin America have some kind of SD-courses, how many students they have, or how many consultancies in Latin America use some SD or employ graduates of some SD program. Same situation almost everywhere. The PC would surely love to "have" a computer model, but the (time and other) resources needed for developing it are prohibitive.

For the good of the SDS, there are some "low hanging fruits", though, that the PC can work on without having developed and exploited a computer model. Example: I know of no one who doubts that SD attracts interested newcomers and then turns many of them off. Many of them wanted to learn, so it is also plausible to state that each opportunity to be "rigorous & friendly" (= be critical AND enable improvement). Every time we review conference submissions, discuss conference presentations, review for the SDR and so on, there is a learning opportunity -for instance. And it is also clear that many interested newcomers across the world need learning opportunities (study materials as well as mentoring) which is currently not being attended to by other agents in the "field" (=eco-system) of SD.

Some debates could be informed by model-analysis-driven insights, yes, because there are divergent opinions which will persist. I am afraid that the SDS will need to advance one step at a time. We are revising the way to monitor membership, revising aspects of the conference and of the journal - all with the "rigorous & friendly" intention, but also in the sense of developing links with neighbor communities ("systems", "computational modeling" ... there are quite some). These initiatives make sense: they draw on the experience of many other associations and on thorough debate; this is what the SDS can do right now.

As much as I'd love to engage in developing a SDS model, I must admit that I lack personal experience in association development and management, and that my own understanding does not go beyond a conceptual framework including people's life-cycle and the existence of alternative methodologies (competing for newcomers). I do think that for the immediate future, we will not have a unified SDS model. To me, my own ignorance is quite clear - and I think it is a good start to know what I don't know. I put this in singular - up to each one to make their own statement. But we can use "rigor & friendliness" among ourselves to get whatever insight we can draw from what we already have. So now I return to the existing computer models. I promise I'll find time to learn about George Richardson's model (which seems to be on the mind of more people than the "strategy model"). I completely missed the previous discussions (hope I can find them). We are a long way from a model-based SDS strategy (I agree, JJ), but we can start by working on it.

Hope to be back soon,
Martin
 Subject : Re:future of the field.. 05/14/2019 11:22:46 AM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Guido and Martin
I want to thank Martin for his reply.
Martin does not need to apologize for a delay in answering, the subject is not critical.
I wanted just before starting any study, discuss two specific points.
About friendliness and rigour.
I think that still today, the SDS errs towards too much friendliness at the expense of rigour.
I think that it is possible to be friendly and still rigorous. For instance, at the SDS conference, there are a lot of poor papers that do not respect a minimum of rigour. SD reviewer are afraid to discourage new comers by refusing to publish their work. But publishing their work gives an image of a field that accepts a relatively high level of lack of rigour. This image may be very destructive for scientific minded people. It is possible to be friendly and rigorous at the same time, simply by accepting to publish relatively poor papers, for instance in a special part named for instance: ‘promising beginners’ and where the errors and lack of rigour is clearly demonstrated to encourage these ‘promising beginners’ to improve their future work. For the scientific minded person, this will be totally different and will not affect the image of the field. I was asked last year to review two papers for the conference. One paper had no model available and the other no units in it! both models were still published during the year, but I do not know if it is still the case. This year I politely said that I did not feel able to review papers because I was not an Sd teacher, but in fact I just did not want to lose my time reviewing rubbish.
I agree highly to Martin’s remark about the utility of knowing what one does not know and it is true that there are a lot of information not known about the system dynamic field. I wanted to say too that having nothing to sell, I can easily confess that my knowledge is extremely limited and can anyhow not be sold to anybody. I think too that before trying to build a system dynamic model of the field, there are more easy things to do first that would already help build a more solid strategy.
The first thing that I am thinking about is information about newcomers and members not renewing their membership.
About the newcomer.
His age, profession, why does he become member, what does he expect, etc.?
For people not renewing: why do they leave, was their first expectation fulfilled, etc.?
I think that this is a basic information necessary to build a better strategy for the SDS and even for the field.
I want to limit this message because I do not like to discuss about too many subjects at the same time.
Regards.
JJ
Last Edited On: 05/14/2019 11:28:53 AM By Jean-Jacques Lauble
 Subject : Re:future of the field.. 06/17/2019 03:23:27 AM 
Jean-Jacques Lauble
Posts: 39
Location
Hi Guido and Martin

I always wondered why an apparently very powerful and rigorous method like SD had so little practical success. The fact that a very poor model like the one we were discussing about, had been published in the SDS review, explains the reason of that lack of success and is a good example of the very bad profession habits. And recognizing that fact is very positive. It proves that providing you do a quality work the method can be effective and it is a very reassuring and heartening thought for anybody trying to work properly.

Best regards.

JJ
Last Edited On: 06/17/2019 03:33:29 AM By Jean-Jacques Lauble
 
# of Topics per Page