Hi all,
Just for personal curiosity I wish to make a survey on formal methods or tools proposed in the SD literature, aiming to support the modeling process (e.g. model parametrization, validation) or model analysis (sensitivity, loop dominance).
Main motivation is the fact that during the last conference I realized that there are some stuff I have never heard of.
So anybody knowing such a publication/resource is more then welcomed. I think I have already done an extensive survey on loop dominance issue, so anything apart from that topic will be much more valuable.
regards
gonenc
Formal Tools/Methods for Modelling Support and Model Analysi
formal tools
Hi
There are more or less formal tools.
But I think one must consider your preoccupations.
Do you want to build models that will be used, or
are you a teacher, or just studying Sd for fun?
Yes it can happen!
To my opinion, there are very few books that teach formal
methods of modelling and analysing models.
What do you mean by formal?
A method can be well formalized but in some cases ineffective.
If you want to find a modelling method, you can go to the
ocw.mit.edu web site and study the two courses from the business school, the second one dealing with a so called standard method, that is too taught in a distant web course from WPI too.
But beware, I tried to use the method for a long time after following the distant course, and lost a lot of time and
money trying to use it. It did not work for me.
But I hope it must have some utility for some, otherwise it
would not be taught by the M.I.T.
The best practical book I studied to build and study real problems is Geoff Coyle's 'System Dynamics modelling: a practical approach'. It it the only one I can recommend. It is unfortunately based on an old software, that is probably no more available.
But it is still for me the only book worth studying if one wants to build real life models.
I beware from formal methods. I use commen sense and experience and some very basic principles.
On of the best principle is from Geoff Colyle's book.
'Keeping the level of understanding of the model and the relation it has to its purpose, beyond its size
(the size of the model).
Regards.
JJ
There are more or less formal tools.
But I think one must consider your preoccupations.
Do you want to build models that will be used, or
are you a teacher, or just studying Sd for fun?
Yes it can happen!
To my opinion, there are very few books that teach formal
methods of modelling and analysing models.
What do you mean by formal?
A method can be well formalized but in some cases ineffective.
If you want to find a modelling method, you can go to the
ocw.mit.edu web site and study the two courses from the business school, the second one dealing with a so called standard method, that is too taught in a distant web course from WPI too.
But beware, I tried to use the method for a long time after following the distant course, and lost a lot of time and
money trying to use it. It did not work for me.
But I hope it must have some utility for some, otherwise it
would not be taught by the M.I.T.
The best practical book I studied to build and study real problems is Geoff Coyle's 'System Dynamics modelling: a practical approach'. It it the only one I can recommend. It is unfortunately based on an old software, that is probably no more available.
But it is still for me the only book worth studying if one wants to build real life models.
I beware from formal methods. I use commen sense and experience and some very basic principles.
On of the best principle is from Geoff Colyle's book.
'Keeping the level of understanding of the model and the relation it has to its purpose, beyond its size
(the size of the model).
Regards.
JJ
I like JJ's suggestions, particularly about Geoff's book that I have never exposed because I simply believed that the book was not necessary for I have other classic books--the idea that may be wrong. I will buy Geoff's book soon.
Based on my experience on SD modeling, I tried to fit the problem to the method. I used SD to conceptualize the problem and, after a year of model building, I came up with a simple static model which can serve the model purpose equally well. That simple model is even more explicit and more understandable than SD models. It seems that a complex problem can be addressed more easily if we are keen to ask the question that can be trusfully answered without using complex models.
Based on my experience on SD modeling, I tried to fit the problem to the method. I used SD to conceptualize the problem and, after a year of model building, I came up with a simple static model which can serve the model purpose equally well. That simple model is even more explicit and more understandable than SD models. It seems that a complex problem can be addressed more easily if we are keen to ask the question that can be trusfully answered without using complex models.
formal methods
Hi Monte
Geoff’s book is not mentioned in the numerous references given by Sterman’s Business dynamics and Warren’s book about strategy dynamic, nor in the current SD web site bibliography.
It is rather strange, Coyle having been a long time in the field and having been one of the most known thinker and practitioner in the field.
One reason might be that I think that Coyle has taken some distance from SD at the start of the decade.
I do not know the reason of it. Was it because of his originality and non conventional thinking that may have not been appreciated by the orthodoxy? I think that it is harmful for SD to loose an intelligent and free thinking person
that on top of that has a very good theoretical and practical knowledge of the field.
I am sure that the administrators must know a lot about this, but I doubt that they will give an opinion.
One interesting controversy was published in the SD review some years ago, about qualitative and quantitative modelling that opposed Coyle to Homer and Oliva. To my opinion they were approximately all right but not discussing about the same thing.
I join the files in pdf format in winzip format, because I think that to get the files, one must be a member of the
Sd society, and browse the reviews and some people in this forum may be are not.
Geoff’s book has many differences from traditional SD Books, and is worth studying. It must be studied slowly.
It has the advantage of being structured aroud practical examples that form the backbone of the book, contrarily to other books that are a succession of uncorrelated chapters dealing with theory, and with examples and exercises that are built from the theoretical course. Here the examples are clearly taken from reality.
Another advantage is that the problems are explained starting with a complete description written in current English that is the best tool to express any problem, using qualitative and quantitative words.
Starting from this description, a first qualitative model is built, and it takes about a third of the book to do that.
It shows clearly the positive and negative sides of qualitative modelling.
After that it follows with making quantitative models from the previous qualitative one’s, showing clearly the difficulties, and the last part is for policy optimization. It appears clearly that the book was written by somebody who made a lot of modelling, not academic modelling, but models that are to be used, which does not appear evident with other books.
The presentation of the book could be better, it is written with little characters and the pictures are not very attractive.
This characteristic added with the now unknown software may explain that it is not recognized.
But the rest compensates largely.
About fitting the problem to the method, I think it is not a good solution. One must fit the method to the problem.
About your problems, they may be mainly static or dynamic and still solvable with static models.
I am too in that case and I prefer too to make static models, because they are much easier to understand, because adding the time dimension to a model, makes it much harder to build, to understand and to use.
But I still work using Vensim, that I find very well suited for static problems too, even if I work with no rates or stocks.
I use all the features of Vensim except gaming, which can be used only with dynamic models.
I do not know why Ventana does not publish a static only version, that would be easier to use, and would prepare people to use later on dynamic modelling. Static modelling is already very hard and to my opinion one should stard dynamic modelling only after some years of static modelling.
I have tested many static packages, in MCDA like Expert Choice etc…, in optimization like, MPL, AIMMS, OPL Studio,
Lingo, and others, and I prefer Vensim.
Regards.
JJ
Geoff’s book is not mentioned in the numerous references given by Sterman’s Business dynamics and Warren’s book about strategy dynamic, nor in the current SD web site bibliography.
It is rather strange, Coyle having been a long time in the field and having been one of the most known thinker and practitioner in the field.
One reason might be that I think that Coyle has taken some distance from SD at the start of the decade.
I do not know the reason of it. Was it because of his originality and non conventional thinking that may have not been appreciated by the orthodoxy? I think that it is harmful for SD to loose an intelligent and free thinking person
that on top of that has a very good theoretical and practical knowledge of the field.
I am sure that the administrators must know a lot about this, but I doubt that they will give an opinion.
One interesting controversy was published in the SD review some years ago, about qualitative and quantitative modelling that opposed Coyle to Homer and Oliva. To my opinion they were approximately all right but not discussing about the same thing.
I join the files in pdf format in winzip format, because I think that to get the files, one must be a member of the
Sd society, and browse the reviews and some people in this forum may be are not.
Geoff’s book has many differences from traditional SD Books, and is worth studying. It must be studied slowly.
It has the advantage of being structured aroud practical examples that form the backbone of the book, contrarily to other books that are a succession of uncorrelated chapters dealing with theory, and with examples and exercises that are built from the theoretical course. Here the examples are clearly taken from reality.
Another advantage is that the problems are explained starting with a complete description written in current English that is the best tool to express any problem, using qualitative and quantitative words.
Starting from this description, a first qualitative model is built, and it takes about a third of the book to do that.
It shows clearly the positive and negative sides of qualitative modelling.
After that it follows with making quantitative models from the previous qualitative one’s, showing clearly the difficulties, and the last part is for policy optimization. It appears clearly that the book was written by somebody who made a lot of modelling, not academic modelling, but models that are to be used, which does not appear evident with other books.
The presentation of the book could be better, it is written with little characters and the pictures are not very attractive.
This characteristic added with the now unknown software may explain that it is not recognized.
But the rest compensates largely.
About fitting the problem to the method, I think it is not a good solution. One must fit the method to the problem.
About your problems, they may be mainly static or dynamic and still solvable with static models.
I am too in that case and I prefer too to make static models, because they are much easier to understand, because adding the time dimension to a model, makes it much harder to build, to understand and to use.
But I still work using Vensim, that I find very well suited for static problems too, even if I work with no rates or stocks.
I use all the features of Vensim except gaming, which can be used only with dynamic models.
I do not know why Ventana does not publish a static only version, that would be easier to use, and would prepare people to use later on dynamic modelling. Static modelling is already very hard and to my opinion one should stard dynamic modelling only after some years of static modelling.
I have tested many static packages, in MCDA like Expert Choice etc…, in optimization like, MPL, AIMMS, OPL Studio,
Lingo, and others, and I prefer Vensim.
Regards.
JJ
Thanks a lot for additional information about Coyle's book. It seems that Oliver and Homer adopted Forreterian SD tradition, so their modeling practice is in contrast to that of Coyle, who uses a qualitative approach to modeling (I have just consullted the SDR issues you mentioned). I am going to read Coyle's argument more slowly.
Thanks for the replies. I was a bit skeptical whether I was clear enough in the way I state my question, and Jean-Jacques' reply, which definitely makes sense, showed me that I was not.
First of all about mu preoccupations, I think I fit into the all three categories you mention JJ.
What I am after is not a formal approach to conceptualize and build a model structure in fact. Given that modeler builds his/her model (referring to Sterman's modeling steps, this corresponds to completing the formulation of the model), there are some mathematical and systematic tools/approaches that can help the modeler in the following stages like sensitivity analysis, behavior analysis, policy design, etc.
To give some examples a method (preferably a formal and systematic one that is feasible to even automate) that helps to modeler to find out the minimal set of parameters to which model behavior is most sensitive would be one of those.
Or an approach like eigen-value elasticity analysis that helps the modeler to establish some linkage between the structure and the behavior (i.e. which feedback loops are important regarding the observed mode of behavior)
or something that may help to answer the question; can it be possible to stabilize the behavior of this system/model just by changing a given set of policy parameters in their feasible ranges?
i hope i could make my inquiry a bit clearer with this.
regards,
gonenc
First of all about mu preoccupations, I think I fit into the all three categories you mention JJ.
What I am after is not a formal approach to conceptualize and build a model structure in fact. Given that modeler builds his/her model (referring to Sterman's modeling steps, this corresponds to completing the formulation of the model), there are some mathematical and systematic tools/approaches that can help the modeler in the following stages like sensitivity analysis, behavior analysis, policy design, etc.
To give some examples a method (preferably a formal and systematic one that is feasible to even automate) that helps to modeler to find out the minimal set of parameters to which model behavior is most sensitive would be one of those.
Or an approach like eigen-value elasticity analysis that helps the modeler to establish some linkage between the structure and the behavior (i.e. which feedback loops are important regarding the observed mode of behavior)
or something that may help to answer the question; can it be possible to stabilize the behavior of this system/model just by changing a given set of policy parameters in their feasible ranges?
i hope i could make my inquiry a bit clearer with this.
regards,
gonenc
formal methods
Hi Gonenc
You just mentioned that you fitted into the three categories I mentionned.
This will make finding a good method very difficult, because the motivations of the three categories are rather
different and you will find difficult to find a method that fits to the three, especially when there is no real method that
guarantees results.
If you want to build a model that has some utility you must have a reference against which to measure your model against. And this reference must be chosen independently from the model.
You may change your objective once you have a reached the first one.
A concrete example:
Imagine you are running a business and you want to better the way you manage the inventory of your business.
From that point of view you have two possibilities.
The first one that is the good one:
Set some general objectives for your business for the next year, with which you feel are reasonable and with which you will be satisfied. Explore the potentialities that a better management of the inventory add to the general objectives. Having that objective in mind you are ready to build a model that reaches this objective.
The second one: do not set any objective and therefore build the perfect model from where you will get the maximum potential of your modelling effort.
In the first case, you will build a model that you will be able to confront to some reality, in the other case you will build a perfect theoretical model, and because it does not exist in reality, you will never finish it and end up with something useless. As Senecae said there is no good wind for the sailor who does not know his destination.
So any method must exist for a purpose, and if you have no clear objectives, not knowing where you want to go, you will never be satisfied with anything and take the risk of setting to high objectives.
A perfect example of this is for me the world model, latest version that tries to describe a complex system the evolution of the world resources, without clear objectives settled first, and then finishes with a model with thousands of loops and to my opinion not much credibility.
Regards.
JJ
You just mentioned that you fitted into the three categories I mentionned.
This will make finding a good method very difficult, because the motivations of the three categories are rather
different and you will find difficult to find a method that fits to the three, especially when there is no real method that
guarantees results.
If you want to build a model that has some utility you must have a reference against which to measure your model against. And this reference must be chosen independently from the model.
You may change your objective once you have a reached the first one.
A concrete example:
Imagine you are running a business and you want to better the way you manage the inventory of your business.
From that point of view you have two possibilities.
The first one that is the good one:
Set some general objectives for your business for the next year, with which you feel are reasonable and with which you will be satisfied. Explore the potentialities that a better management of the inventory add to the general objectives. Having that objective in mind you are ready to build a model that reaches this objective.
The second one: do not set any objective and therefore build the perfect model from where you will get the maximum potential of your modelling effort.
In the first case, you will build a model that you will be able to confront to some reality, in the other case you will build a perfect theoretical model, and because it does not exist in reality, you will never finish it and end up with something useless. As Senecae said there is no good wind for the sailor who does not know his destination.
So any method must exist for a purpose, and if you have no clear objectives, not knowing where you want to go, you will never be satisfied with anything and take the risk of setting to high objectives.
A perfect example of this is for me the world model, latest version that tries to describe a complex system the evolution of the world resources, without clear objectives settled first, and then finishes with a model with thousands of loops and to my opinion not much credibility.
Regards.
JJ
I still have a feeling that I am not making my quest clear enough. I am not after 'a' method to construct a model and complete a study that utilizes that model.
The expression I am about to use will narrow down the scope of what I am after, but I think it will make it much more clear;
I am wondering about algorithms, tools, methods that can preferably be automated in the form of add-on softwares to the existing simulation platforms, and that has potential to 'support' the modeler at some point of a modeling process/study.
I am not looking for a specific method to use. Just because of pure curiosity, I want to check the literature and see what do we have in the field's toolbox.
And a minor note, not related to the core. When I said I fit into all three, I don't mean that I am wearing those three hats at the same time. However, my work involves a combination of these three different activities. In this search, I intend to form a background info that I can utilize in the future when I am wearing any of those hats.
I hope it is much more clear and less prone to miscommunication.
Anyway thanks for the detailed responses JJ
gonenc
The expression I am about to use will narrow down the scope of what I am after, but I think it will make it much more clear;
I am wondering about algorithms, tools, methods that can preferably be automated in the form of add-on softwares to the existing simulation platforms, and that has potential to 'support' the modeler at some point of a modeling process/study.
I am not looking for a specific method to use. Just because of pure curiosity, I want to check the literature and see what do we have in the field's toolbox.
And a minor note, not related to the core. When I said I fit into all three, I don't mean that I am wearing those three hats at the same time. However, my work involves a combination of these three different activities. In this search, I intend to form a background info that I can utilize in the future when I am wearing any of those hats.
I hope it is much more clear and less prone to miscommunication.
Anyway thanks for the detailed responses JJ
gonenc
Formal tools
Hi Gonenc
The numerous propositions listed in Business Dynamics chapter 22, are generally still not implemented in any software.
I am not a partisan of too many tools, because I think that one must be sure that the existing one's are already well understood. I am an advocate of Pareto laws and too many tools may generate a dispersion of the efforts.
Besides I think that there is no substitute to experience, good thinking, common sense and time.
But we have here the chance to have somebody who heads the development of Vensim and I am sure that Bob is the best to give an advice on the subject.
Regards.
JJ
The numerous propositions listed in Business Dynamics chapter 22, are generally still not implemented in any software.
I am not a partisan of too many tools, because I think that one must be sure that the existing one's are already well understood. I am an advocate of Pareto laws and too many tools may generate a dispersion of the efforts.
Besides I think that there is no substitute to experience, good thinking, common sense and time.
But we have here the chance to have somebody who heads the development of Vensim and I am sure that Bob is the best to give an advice on the subject.
Regards.
JJ
Hi JJ,
That is exactly what I want to do. I also have a feeling that we are well behind full utilization of existing tools. Hence, I wish to make a personal survey to see what do we have and in what extend those can be used.
If I can manage it, it will be something like the updated and maybe extended version of Chapter 22, with emphasis on specific pieces from the literature.
regards,
gonenc
That is exactly what I want to do. I also have a feeling that we are well behind full utilization of existing tools. Hence, I wish to make a personal survey to see what do we have and in what extend those can be used.
If I can manage it, it will be something like the updated and maybe extended version of Chapter 22, with emphasis on specific pieces from the literature.
regards,
gonenc
Software development is moving forward fast, relative to the slow growth of my understanding of SD. I can use Vensim PLE + Monte Carlo Simulation in Vensim Professional. It seems enough because I try to play with the problem that can be addressed without needing more formal tools. It is enjoynable in that the problem is suitable to the level of modeling skill. I beleive that Jay, John, Andrew, Khalid, George, and Meadows cannot use many of the advanced formal tools in the 2000s, but they could build good models since the 1980s. How can they do that? What factors make one a good model builder?
Hi Monte,
To be honest I don't have the prescription for becoming a good modeler. However, what I know is the fact that those advanced tools do not make any modeler a good modeler, they only make their life easier.
An example about how I see it; a cruise computer or a navigator on a brand new car may help a lot to the driver during a long trip. However, it is neither necessary for a successful and smooth trip, nor sufficient to make the driver a good driver.
regards,
gonenc
To be honest I don't have the prescription for becoming a good modeler. However, what I know is the fact that those advanced tools do not make any modeler a good modeler, they only make their life easier.
An example about how I see it; a cruise computer or a navigator on a brand new car may help a lot to the driver during a long trip. However, it is neither necessary for a successful and smooth trip, nor sufficient to make the driver a good driver.
regards,
gonenc