SD and other modelling approaches

[Arlen Wolpert awolpert TheWorld.]

Posted by Arlen Wolpert <awolpert@TheWorld.com>
Here is a very serious question:

Why doesn't someone make a system dynamics (SD)
model of this key SD problem?

I believe such a task would require a very bold and
serious system dynamicist, because the results would
probably reveal a deep truth about the structure of
the intellectual world.

I would be very grateful for such a model.

Posted by Arlen Wolpert <awolpert@TheWorld.com>
posting date Thu, 22 Dec 2005 09:48:21 -0500

[John Gunkler jgunkler sprintmail]

Posted by ""John Gunkler"" <jgunkler@sprintmail.com>
I have been helping companies with strategic thinking for nearly 20 years (I don't much like the term ""strategic planning"" because that generates a report which gathers dust for 12 months and then is discarded). I know of few approaches that do anything more than ""evaluating and exploring a pre-chosen strategic concept"" and, in fact, most approaches do ""not, indeed, cannot, evolve such concepts"". So, SD is in the same boat as pretty much all other approaches to strategy, in my opinion.

But what other approaches do much more poorly than an SD model can do is to actually evaluate the likely outcomes of choosing one strategy vs. another. Most approaches that I am aware of, and have used, rely on some kind of ""informed"" judgment on the part of the strategy team -- and I use the term ""informed"" rather loosely. Before SD, I used to do my best to help teams make more ""informed"" judgments, and tried to help them avoid obvious errors in thinking things through -- I even tried to help them be more creative in coming up with new approaches and provided them some logical structures for analyzing their markets and potential markets. But, in the end, it was just ""best guess""/""informed opinion"" that chose the strategy path.

In fact, I don't entirely agree that SD models cannot generate strategic ideas. To use Jay Forrester's language, SD models are about ""policies"" (captured in the rate equations and auxiliaries) that drive dynamic behavior. When that behavior is not good enough, or when one wants to make it better, examining how different policies (within the model) affect behavior can, indeed, lead to the surfacing of ""new"" policies to try. And, because the effects of the feedback loops driving behavior are not easy to intuit, examining the interaction of the feedback loops can lead to policies that no one has ever thought of before.


John Gunkler

P.S. There is one excellent strategy book on the shelves of booksellers I frequent that uses system dynamics. It is ""Competitive Strategy Dynamics"" by Kim Warren, a contributor to this list. Thanks, Kim, for a terrific book. Posted by ""John Gunkler"" <jgunkler@sprintmail.com> posting date Thu, 22 Dec 2005 11:29:53 -0600

[Tony Gill Tony.Gill phrontis.com]

Posted by ""Tony Gill"" <Tony.Gill@Phrontis.com>
""Tobias Lorenz"" said:

>>One main task ... key difference.

You mention ABM. (I assume agent based modeling) As I see it, there are numerous other modelling approaches eg, from the Operational Research / management sciences, strategy models, financial models, treasury models, mathematical models, quality models, biological models etc. And of course excel spreadsheets. All models are competing for attention as it were and are used by those who have been attracted to them for one reason or another.


>>The next task (in my eyes at least) would then be to identify the
>>criteria for the usage of a specific methodologies.

This is not so easy because all such positioning is laden with some agenda or other viz, ""Creative Problem Solving: total systems intervention"" by Flood and Jackson. They have produced a 3 x 2 matrix with Unitary, Pluralist and Coercive on the horizontal axis and, Simple and Complex along the vertical axis. They position SD in the Simple/Unitary box! In essence you decide on which box your model challenge lies and then you use the tool that is in the box.


>>I disagree with the position that the choice of methods is a purely
>>subjective preference of the modeller.

It IS subjective because most modellers are highly experienced in a relative few modelling tools. It takes time to be proficient in the use of tools. So most modellers settle for a tool set that works for them most of the time ie, their tool set is limited to what they know and this makes their choice subjective. See Mingers and Gill ""Multimethodology: the theory and practice of combining management science methodologies"".

Tony Gill, Phrontis Limited, Beacon House, Horn Hill Road, Adderbury, Banbury, OX17 3EU, UK Posted by ""Tony Gill"" <Tony.Gill@Phrontis.com> posting date Thu, 22 Dec 2005 14:15:44 -0000

[Thompson James. P (Jim) A142 Jim]

Posted by ""Thompson, James. P (Jim) A142"" <Jim.Thompson@CIGNA.COM>

Geoff Coyle writes. ""... maybe it was a mistake to found the SD Society.... Perhaps we should just be members of the OR Society, Institute of Management Sciences, etc and show ourselves as analysts who just happen to have a particular interest in SD, in parallel with other peers whose focus is on queuing theory, or whatever.""

My work often requires me to use approaches other than SD for problem-solving. Since I'm not sufficiently schooled in econometrics, I have to rely on others in my organisation to provide that expertise when
necessary. Similarly, discrete event simulation is not my strong suit,
but it has its place in helping to solve a certain class of problems. I find no conflict in recommending SD where appropriate or other to help colleagues with other approaches.

Like others, I belong to a few professional associations including SDS. I am thankful there is an SDS as a forum for discourse among researchers and practitioners. I enjoy that there are zealots for a number of problem-solving approaches; they spice things up. But in the business world (at least my corner of that world), to only see one path to the top of the mountain makes for a lonely existence.

There is SD or SD-like work that is not represented in SDS, and that bothers me sometimes. A quick scan of, say, the International Institute of Applied Systems Analysis indicates there is a thriving community of people who work to advance principles of system dynamics outside SDS membership.

At the New York Conference, I got the feeling we were working to be more inclusive. I could be all wet, but just two years later in Boston I sensed less inclusiveness. In my Society work, I hope to expand our perspective on applications of system dynamics methods in both the natural and social sciences.

Cheers,
Jim Thompson
Economic & Operations Research
Cigna HealthCare
900 Cottage Grove Road, A142
Hartford, CT 06152
Posted by ""Thompson, James. P (Jim) A142"" <Jim.Thompson@CIGNA.COM>
posting date Fri, 23 Dec 2005 12:01:40 -0500

[Tobias Lorenz space56 freenet.de]

Posted by ""Tobias Lorenz"" <space56@freenet.de>
Hey everybody,

First of all with ABM, i meant the Agent-based Modelling, which is becoming more and more important. Sorry for the exceptional abbreviation. I still agree with Geoff and i would like to even sharpen that point, i would say that it is possible to identify criteria which make a certain simulation modelling method suitable for some problem context. What we would need to discriminate therefore in my opinion would be something like
a) the perspective a certain problem creates
b) the part of the real world which matters for that problem
c) the simulation methodology

Those three parts have to fit somehow and if we as system dynamicists don't make it to be able to use the other existing methodologies, we have to have a clear idea of when not to use SD. The knowledge of some part of the real world, be it an area of application like health care or military is do to b) in the upper list of course basic in modelling. But it alone doesn't help u to tackle all the problems in that area of expertise with SD, this is not possible. But it gives you are clear understanding of which problems in that domain might be suitable to grasp within an SD approach. To use a well-known metaphor: we mustn't run around with our SD-hammer and make everything an SDable nail, no we have to have a toolbox and use the right tool for the right nail. At least we have to know where not to use the SD hammer. In order to identify the nails/ problems, which are best grasped with a certain tool, i think the upper three points might help choosing. Necessary in the next step would be clearer formulation of the interrelation of those points for certain methods, e.g. Agent-based modelling, DES, SD, ""normal"" OR... And i still strongly disagree with the position, also made explicit in ""Multi-Methodology"" by Bennett, Ackermann, Eden, Williams : ""While some rules of thumb about what works well may be evolving, choices of method will continue to depend on idiosyncratic combinations of factors to do with the personal styles and preferences of analysts and clients, the time available, gross characteristics of the 'perceived issues', past experiences of all concerned, organizational cultures, financial and academic pressures inhibiting or encouraging collaborative working, and so on."" (Bennett, Ackermann, Eden, Williams: Analysing Litigation and
Negotiation: Using a combined methodology, in Mingers, Gill: Multimethodology, Chichester, 1997, page. 86)

If we make the choice of the simulation methodology a personal thing, we as modellers as a whole will loose a lot of reputation, I#d say. And i think it is very easy to see in a first step, that for some problems discrete events matter a lot, e.g. the introduction of a law, which creates some kind of dynamic in a system, this event should be modelled discretely then; otherwise a lot of the dynamics of the simulation might be lost. And of course methodologies can still be combined again after the importance of a certain methodology for a certain aspect of an problem has been identified...

Merry Christmas to everyone

Tobias Lorenz
DaimlerChrysler Research and Technology
Posted by ""Tobias Lorenz"" <space56@freenet.de>
posting date Fri, 23 Dec 2005 14:13:18 +0100

[Tobias Lorenz space56 freenet.de]

Posted by ""Tobias Lorenz"" <space56@freenet.de>
Dear Tony


>> It IS subjective because most modellers are highly experienced in a
>> relative few modelling tools. It takes time to be proficient in the
>> use of tools. So
>> most modellers settle for a tool set that works for them most of the time
>> ie, their tool set is limited to what they know and this makes their
>> choice
>> subjective.


Just because most modellers do it that way, that doesn't mean that it should
be like this. Usage of simulation has simplified a lot in the last years i'd
say, so it is becoming easier to at least have a vision what different
methodologies are about. And then one can already decide when not to use a
certain method, as Geoff proposed.
But in order to satisfy your client, more is necessary and also implicitly
done in SD as has already been mentioned here. You have so many discrete
functions within Vensim for example, that u can build, I'd say
Discrete-Event-Modells there, if you want to. We even managed to build a
simple Agent-based modell in Vensim (Just introduce some subscript
dimensions and some binary stock for the agents, only the interaction is
hard to manage).But that's just an excursion on the side.
What i want to say is, that in order to maintrain integrity and keep our reputation
we have to have clear arguments against other methodologies in a certain
problem context, when we apply SD.

Tobias Lorenz
DaimlerChrysler Research and Technology
Posted by ""Tobias Lorenz"" <space56@freenet.de>
posting date Fri, 23 Dec 2005 14:22:04 +0100

[Jean-Jacques Laublé jean-jacques]

Posted by =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <jean-jacques.lauble@wanadoo.fr> Hi everybody

The idea of deciding which modelling technique is better adapted to a problem, while being a fantastic idea may be practically very difficult to realize.

The first reason is that there is no method to characterize a 'problem'.

The second reason is that many methods can solve problems equally well.

The third reason, is that how the method works depends highly on the people who apply it. And there is generally for most methods no way to characterize the competence of somebody relative to them.

The fourth reason is that finding the team of 'independent and competent' people that will do the job will be hard. You will not find at the same time competent and independent people as the fact of being competent in a domain lessens the independence from it.

But to be positive I propose something else that is still hard but that may help.

Did you ever read a book about any methodology that in its preferably first chapters explains precisely and fully the conditions that must be met, for a problem to be solved with it?

Is SD doing this? And why not work in this direction so as to be sure not to deceive any customer and be sure that he will get more benefit from the model than the cost of building it? This approach does not exist in SD to date or is extremely vague. I did not see any book that explains fully these conditions. The advantage would be to concentrate on topics that can be really solved with SD and to generate more satisfied customers.

I have for instance spent 3000 hours on a subject without having solved anything serious and 4 hours on another having got a lot of insight from it. I have tried to understand why. The difference between the two subjects is that one is very well fitted for SD the second not, even if it has a lot of dynamic in it.

To detect if a subject is well suited for SD is a question of coherence. There must be a coherence between the objective and the model that will be built to resolve it:
1 The expected profit from the model must be greater than the expected cost of it. Different objectives may be considered keeping the one that best comply to the conditions.

2 Is it possible to construct a model with boundaries broad enough to consider all the the realities that need to be modelled but not to broad so as to keep it manageable?

3. How many factors are involved? are they independent from one another? Are some of them dominant so as to ignore the others and keep the model simple?
4. Are the data needed available? Are they unbiased? how much will it cost to get them?

5. Once the boundaries are settled will the exogenous factors be easily predicted? There is an opposition between the endogenous variables calculated by the model and the exogenous variables not modelled and calculated by traditional techniques. The ideal is that the model has no exogenous variables, which is generally rarely the case.

6. How will the outside world (the exogenous factors) react to the policies? For instance in my model setting prices, the pricing of the competitors was exogenous, but obviously if I modified my prices, the competitors would react to it. So it seemed necessary to include the pricing policy of the competitors in the model. Is it possible to have a reasonably sized model where you do not have the risk of policies influencing the exogenous data?

7. Is it possible to aggregate the data so as to have a model
simple
enough without having the results completely biased by the aggregation?

8. Can the policies proposed by the model be applied? This to
avoid
theoretical models whose policies are never applied.

I stop here the list but there are many others things to be considered.

Without being too systematic it is possible to weight all the conditions, to note them and to have a general note of compatibility that could give an idea of the chances the project has to be a good one.

Somebody prudent would then decide to go on only if the note
is high and the project very well fitted.
Of course the list I proposed is certainly not the best one, and lacks many other conditions I am not an SD specialist.

But knowing by advance that the technique fits the problem, would certainly increase the number of potential customers and ease the building of the model too, the principal difficulties having been already studied.

Of course to follow such a path, it is necessary to accept the idea that there are other methods than SD that are more effective for some problems even if they are simple and often because they are simple.

Regards to everybody and happy Christmas.
J.J. Laublé Allocar
Strasbourg France.
Posted by =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <jean-jacques.lauble@wanadoo.fr> posting date Sat, 24 Dec 2005 17:04:36 +0100

[Martin Kunc MKunc london.edu]

Posted by Martin Kunc <MKunc@london.edu>

Arlen,

While the paper is not about SD specifically, it is an excellent SD model about the evolution of scientific knowledge.

Path Dependence, Competition, and Succession in the Dynamics of Scientific Revolution. Organization Science,Vol. 10, No. 3, May-June 1999

Jason Wittenberg
Department of Political Science
Massachusetts Institute of Technology
Cambridge, MA 02142

John D. Sterman
Sloan School of Management
Massachusetts Institute of Technology
Cambridge, MA 02142

Happy holidays,
Martin

Martin Kunc
Ph.D. in Decision Science
London Business School
Posted by Martin Kunc <MKunc@london.edu>
posting date Fri, 23 Dec 2005 13:18:33 GMT

[Bill Harris bill_harris facilita]

Posted by Bill Harris <bill_harris@facilitatedsystems.com>
""geoff coyle geoff.coyle btinternet.com"" <system-dynamics@VENSIM.COM> writes:
>> I think that you are broadly right, though you still imply that the
>> task is to 'sell' SD. Of course, it is a wonderful method when
>> properly used for suitable issues but the mistake, maybe, is to call
>> ourselves System Dynamicists (I NEVER do that) with the implication
>> that we only do SD. That can come across to other analysts as 'SD is
>> the only thing worth doing', which is patronising, offensive, and
>> manifestly untrue.


Geoff,

When I wrote my comment, I had a feeling I was wording something badly, and you put your finger right on it. I agree that SD is wonderful for suitable issues but not for everything. Perhaps multiple approaches would help us triangulate into a more robust answer. Perhaps a different approach is called for and SD would be a misapplication.


>> Your key phrase is 'what if we listened and sold excellence in
>> problem solving'. The corollary of that is that one might, for
>> instance, say 'Sorry, SD, which is what I know about, won't tackle
>> this problem, but DES might'. I like to think that I know about SD,
>> but I'd be embarrassed to say that I didn't know anything about DES,
>> linear programming etc. (If you want to see something different, have
>> a look at my website.)


Yes! Most clients don't seem to care too much how we help them get good answers, the exception perhaps being those who want to learn to fish for themselves.

Of course, any one of us may be skilled enough in a broad enough variety of approaches to be able to help competently, even if the challenge isn't an SD problem.

Bill
- --
Bill Harris
Facilitated Systems Everett, WA 98208 USA
Posted by Bill Harris <bill_harris@facilitatedsystems.com>
posting date Fri, 23 Dec 2005 10:37:24 -0800

[Jean-Jacques Laublé jean-jacques]

Posted by =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <jean-jacques.lauble@wanadoo.fr> Hi everybody


There are many things about SD that have been said lately and that are very true.

But I think it is not a reason to be ashamed to use a SD or not to dare to say that one uses it.

The method and the paradigm is good and very powerful, maybe too powerful and the problem is to better use it. And on top of that customers are not even interested about the method one uses as long the problem is solved.

Another way to better use SD would be to follow the first objectives of this list that are as stated in the list web site:


The purpose of the list is to promote discussion around issues in building and using System Dynamics models. Examples of topics that are of interest
include:

a.. Issues in conceptualization of a problem into a feedback structure including identification of reference modes and the creation of dynamics hypotheses.
b.. Ideas and questions about how to formulate a particular type of structure (hiring policy, price adjustment, fertility determination and so on).
c.. Interesting models you have built and the results arrived at. Questions to others about examples of models for certain types of problems.
d.. Validation of System Dynamics models.
e.. Implementation issues.

I have been in the list for more than three years and I do not feel that the threads conform to these objectives, except perhaps for the 'questions to others about examples of models for

certain types of problems'.

I do not say that the discussions do not relate more or less to these problems, but it is a bit like discussing about customers without ever having seen one.

Logically writing about modelling one should some times see one model and discuss about it, no?

For instance I know only about my own difficulties, but I might have resolved some of them if I had been aware and learn from the difficulties of other members. I have been very disappointed when I realized after some time that the list would probably never make the effort to put the hands in the dirty oil. (This is a strict translation of a French expression 'mettre la main dans le cambouis').

By looking at what happens in the reality of modelling (this means, studying real models) all these questions that have today no answers would maybe now easier to address.

The discussion about the use of SD is not a philosophical of literary question but a technical

question.

Regards.

J.J. Laublé Allocar

Strasbourg France
Posted by =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <jean-jacques.lauble@wanadoo.fr> posting date Sun, 25 Dec 2005 17:30:11 +0100

[Tobias Lorenz space56 freenet.de]

Posted by ""Tobias Lorenz"" <space56@freenet.de>
Dear Jean-Jacques,
You have lately posted two mails, which i found both very inspiring for my thoughts, especially the critism of my idea to classify problems in order to arrive at the right method. Nevertheless in the next mail you claim the need for

""Issues in conceptualization of a problem into a feedback structure including identification of reference modes and the creation of dynamics hypotheses.""

and fight against philosophical struggles without putting the hands into the dirt of practical modelling. What i want to make strong now out of my experience at DaimlerChrysler (where i have my hands deep in the mudd;)), that this first issue is exactly where we need some more philosophical reflection, because we do not only have to decide HOW we conceptualize a problem in a feedback structure, but sometimes we have to ask IF we should do it. I would argue that it is possible to find some kind of feedback structure for almost every problem, the question would then be, is the feedback relevant for that problem, which we might see out of our reference mode, if we are able to get data.

So what we would need as you say yourself say is ""the conditions that must be met, for a problem to be solved with [some methodology]"" and i would agree that we don't have much of these yet, but urgenty need them in order to make SD respectable, so to be a little more hands-in-the-dirt, we would need a method telling us that if our real-world.reference-mode is so-and-so close to one of our basic behaviors (e.g. exponential growth) then we can assume a feedback structure, here we would probably have to grab to stastics. And i guess one could do this aswell for alternative methodologies, for example for ABM u might say, that if the reference mode is predominantly caused by interaction of heterogenous agents or u need spatialty for the explanation of your problem, you would have to better model with ABM. Maybe one could also try to operationalize in order to get harder criteria for usage. For DES, one might say, that if the problem is mainly caused by some kind of fluctuations in the input-variables of a highly linear problem, then one might have it easy and use DES.

For your second reason (""The second reason is that many methods can solve problems equally well""), i would disagree, many methods can conceptualize the different problems, but for different problems they don't fit equally well. You can hardly modell interactions between different entities in SD, for highly linear problems which shall be optimized there would be no point in using SD. And especially you have to discriminate, as i did, between what the software can do and the core of a methodology (That's why with the hands in the dirt, the methodology becomes more irrelevant.). E.g. you can easily modell very discrete with Vensim, by usind if then else and various built in functions, with suscripts you can also build entitiy-flows....


For number three (The third reason, is that how the method works depends highly on the people who apply it). Of course it does but the methodologies are also fixed through the books that build the core of the methodology, e.g. ""Business Dynamics"" or ""Industrial Dynamics""

I highly agree with your list of questions in order to identify if SD is useful for a certain problem, we need this kind of questions is order to become harder, but still i'd add questions like:

Is there really feedback involved in that problem, or is the cause of the problem interaction between entities too heterogenous to be aggregated into a single variable?

Is a feedback modell necessary or would statistical approximations or a linear model be enough?

I just wanna propose those two here, but as u say, there would be a lot more necessary to be asked...


Best regards

Tobi
Posted by ""Tobias Lorenz"" <space56@freenet.de>
posting date Wed, 28 Dec 2005 16:59:11 +0100

[Jean-Jacques Laublé jean-jacques]

Posted by =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <jean-jacques.lauble@wanadoo.fr> Hi Tobi.

I find very comforting that Mercedes Benz my first supplier for more then 30 years is interested in SD as a tool.

About your idea about classifying different methodologies, this was not a critic but the enumeration of the difficulty of the task. The idea in itself is very good and would be highly useful. I spent too a lot of time trying to explore many methodologies and I am still not sure to use the appropriate tools. I did not found a book that exposes all the tools that exist and what they are good too.

About the philosophical struggle and the hands in the dirt, I think that all people in this list when confronted with their difficulties put their hands in the dirt. But each one has a different sort of dirt, and as it is difficult to compare it, the only latitude is too talk about theory.

I think that confronting some times the real difficulties would permit to find some common principles easily then by staying at the higher level of discussion.

I never said that we do not need philosophical or theoretical ideas that lead the modelling process or help to decide to what extent a modelling effort will be useful, but that being confronted with real problems may help find these principles.

When I listed the conditions that I must be studied before engaging in SD, I had in mind the possibility not to engage at all or the possibility to engage in a certain way, for instance in reducing the objective or making it more easy at first.

When looking back at my nearly 4 years of experience, I think that my expectations were much too great relatively to the kind of environment I was leaving in and relative to my SD experience that was equal to zero. I had to start very low and along the years learn little by little. So after 4 years I think that my capacity to use SD is much more lower then what I thought it was two or three years ago. Is it that I have not learned anything? No, I have a better understanding of the difficulties of the task and more aware of them. I think that the problem is not 'to use or not to use SD'. SD is a very powerful technique that can be used in many ways. The problem is to know how to use it and when. And being able not to use SD accordingly is a proof that you know SD.

I am too a strong advocate of narrowing the bridge between static modelling and dynamic modelling the only difference between the two being that the static has no stocks and the dynamic has at least one stock. The static view of the reality being a particular case of the dynamic view: a dynamic view with no stocks. I say that many methods can solve problems equally well, but I never said that all of them do. Yoy say that SD is not supposed to solve static optimization. You can do it very easily in Vensim, whatever the kind of problem: linear, quadratic or completely not linear. You only have too use two period of times and no stocks.

All this to say that to my opinion one should first concentrate on the characterization of the problem, and choose an objective that is coherent with it, with the method you want to use, to the stake involved in the resolution of the problem, and to your capacity (knowledge, time, money etc.).

And you may find that the same problem may be 'solved' in one hour on a paper board, listing the advantages and the drawbacks of diverse policies, or in a day using a more sophisticated method, for instance MCDA, putting some weights on the preferences, or hours, days, months or years by building a more or less complicated model, depending on the amount of 'material' that you will put in the model.

I have heard that modellers take much more time to analyse a model than to build it. To my opinion, and as I have experienced so far, I intend if I start a new modelling effort to spend 95% of my effort on the preparation taking all the time necessary to explore all the conditions of the problem and on the decision to model or not. If this is well done I think that all the rest, the building and the analyse will be a lot easier and most of all, will be useful.

But I think this resolution is a product of my past experiences and it cannot be demonstrated.

About the conditions to be studied to use SD one should add: does it help to use the principle of causality. If one decides to use it one is bound to measure the influences more or less. So is it possible to do it? About the existence of feed backs, it is better to talk about stocks, because feed backs come from stocks. When you start to be strictly causal you may be bound to use stocks and if you use stocks there is a great chance that with stocks and causality you will have feed backs. So the problem is to see whether you can stay causal without stocks. If you decide to do it, the model becomes an influence diagram with no period of times. The problem being to detect if there exists a stock in your problem that has a significant influence on other results as the time goes on. Regards. J.J. Lauble Allocar Strasbourg France Posted by =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <jean-jacques.lauble@wanadoo.fr> posting date Fri, 30 Dec 2005 18:39:24 +0100

[Tobias Lorenz space56 freenet.de]

Posted by ""Tobias Lorenz"" <space56@freenet.de>
Dear Jean-Jacques
I definitly agree with you that a rigouros analysis of the problem makes the modelling quite easy, but still i want to keep the point, that a - even if it is subconscious;)- choice of the method to use before the analysis of the problem influences the analysis and i'd say that most of us have this ""subconscious"" idea before starting the analysis, i.e. whether to use game theory or causal loops, pure regression analysis or qualitative analysis and all i say is, let's make this choice more obvious and understandable. The next step would be then and i don't know if that is possible, to say if a problem has certain characteristics at the first glance then probably it is best to approach the problem like this or that...

The point you mentioned about another criterion for SD, i did not understand it. You mention the principle of causality and say that feedback comes from stocks, but i'd say that those are two independent characteristica of SD?!

Best regards
Tobias Lorenz
Posted by ""Tobias Lorenz"" <space56@freenet.de>
posting date Mon, 2 Jan 2006 17:45:33 +0100

[George A Simpson gsimpso4 csc.co]

Posted by George A Simpson <gsimpso4@csc.com>
I recently hit yet another barrier my internal ""SD selling"" campaign, and on a second thread got a green light for development of a signficant SD application.

This suggests that it is not ""SD"" that sells, but rather SD solutions.

After all, isn't SD ""just"" a language. Languages are a platform for solutions, but it is the solutions themselves that have the business impact.

Perhaps there is a market for inexpensive configurable simulation-based (SD, DES, and ABM) models that address common business issues.

Does anyone have a view about whether this is feasible? Has it been
tried?

..george...

Dr. George Simpson, Principal Consultant, CSC Alliance: Performance Engineer CSC House, Fleet Hampshire, UK. GU51 2UY Posted by George A Simpson <gsimpso4@csc.com> posting date Tue, 3 Jan 2006 08:45:32 +0000

[Jean-Jacques Laublé jean-jacques]

Posted by =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <jean-jacques.lauble@wanadoo.fr> Hi Tobi

I agree with you that it would be easier if one could choose easily the tool depending on the kind of problem and that the kind of tool will influence the quality of the decision and the amount of work. About stocks and feed back, I wanted to explain that a feedback loop needs at least one stock on which the path of the loop is going through and that the dynamic is a consequence of the stock. So you can first build a model without feed back loops, and with stocks easier to understand and progressively add some feed back loops once you have well understood and used the first one. Regards. Jean-Jacques Laublé Allocar Strasbourg France Posted by =?iso-8859-1?Q?Jean-Jacques_Laubl=E9?= <jean-jacques.lauble@wanadoo.fr> posting date Tue, 3 Jan 2006 14:12:38 +0100

[Tobias Lorenz space56 freenet.de]

Posted by ""Tobias Lorenz"" <space56@freenet.de>
Dear Jean-Jacques,

I have my problems with the technicalities again: What do you mean by
aggregated time? This is in your opinion central for continuous systems.
But this also in my opinion a matter of degree. In System Dynamics we
have very small steps, sometimes spoken of as time slices, whereas in
DES the events which trigger the computation of the next event, may lie
a little further apart, but nevertheless, also SD is computed DISCRETELY.
So one could imagine that an DES model with events with only a very small
intervall between the events might already look continuous?!

About your pessimism about hybrid models, as long as you don't implement
too much singualirities into the model, i think it should not get too
complicated, one or two jumps in the system due to singularities and
the analysis of the system to these discrete jumps sounds possible to
me. The bis problem is, to come back to your metaphor and the the other
thread, we don't have much dirt in hybrid modelling yet, where we could
put our hands into!

About the conference, i was thinking, might one not propose a session
upon these topics? I guess, i am not too much into the community yet,
to really organise sth like that, but i'd definitly support whereever
i can! At least there seems to be some discussion demand and in my
eyes multi-paradigm modelling / Hybrid modelling or at least a clear
vision of the paradigms is quite necessary for the advancement of
simulation modelling and with it SD...

Best regards
Tobias Lorenz
Posted by ""Tobias Lorenz"" <space56@freenet.de>
posting date Fri, 6 Jan 2006 09:33:22 +0100

[Jean-Jacques Laublé jean-jacques]

Posted by Jean-Jacques Laublé <jean-jacques.lauble@wanadoo.fr> Hi Tobi There have been in the last conferences some papers about multi-methodology and the advantages and drawbacks of different methods. You can consult the different papers on the SD association www.systemdynamics.org look at the conference button and loop at the past papers. There are too CD's on the last conference papers. It takes some time to consult the papers because there have been many conferences and a lot of papers by conference. I think that you can too consult the papers looking for a word. You must browse through the Association web site to find your way through.

About aggregating the time I mean that instead of considering each single event and the distinct time occurred, SD chooses to aggregated all the events in a single number and to make them happen in a single time period. So for me the aggregation is with the events and the time. But the periods of time in SD are discreet and the time event occurs in DES in a continuous time scale. It explains that the difference in both paradigms is confusing. So you are too right when you describe both SD and DES as you did it.

I too think that hybrid methods is very relative to the kind of problem. Unless one takes a very concrete example, it is very difficult to explore the differences between would be hybrid or not hybrid and how much hybrid. When speaking about the difficulty to use both paradigms I can only refer to my own case.

My problem, adjusting pricing and investments could be explored in a discreet or continuous manner. One can either consider events (generally in a day to day basis) or consider for example month or even year periods using aggregate values.

The method is very different. The first one, due to the fact that there is no aggregation of the inputs obliges you to a very close respect of reality. For instance if I decided that each event was the same and taking place at a regular time step it would generate a model that would be so far away from reality that it would be unusable. So I tried this approach using SD, taking days periods and being obliged to use massive subscripting not to mention the necessity to generate the numbers in a stochastic way with lots of problems if you want to optimize. I did in Vensim, but I finished with models so highly complex that they where no more understandable. My problem was not to generate a day by day simulation but to fix prices and investments and the day by day simulation could have helped in theory to do that but it was practically unusable.

So I changed the method and considered longer time periods, aggregating all the values which delivers interesting results and gives a good idea of the overall behaviour. I think that if you are considering general policies you have better use SD but if you need to have detailed results it is better to use DES or agent based if you can do it. For instance I want to know if considering the competition I have better to invest less and increase my prices or invest more and decrease my prices.

If I find the solution I will be able to take global decisions: for instance if I invest less I will reduce the hiring of people etc.. I feel it is eventually better to know global policies first. But knowing that I will not be able to fix the price for a determined category of vehicle for a certain duration, for a certain number of kilometres. If I want to calculate this it is better for me to use marketing segmentation methods than simulation. This does not mean that it is not possible to do it. But it is a question of capacity and time and cost. But knowing already the first global better strategy will help me fix individual prices because I will know the general direction I have to work towards. Regards. Jean-Jacques Laublé Allocar Strasbourg France Posted by Jean-Jacques Laublé <jean-jacques.lauble@wanadoo.fr> posting date Fri, 6 Jan 2006 16:28:51 +0100