Greetings,
I have been thinking about several modeling issues such as the so-called
conservation principles. In engineering, we (engineers) apply conservation
of mass, momentum and energy. While it is obvious that matter (representing
mass) and resources (representing energy) are conserved in modeling social,
economic and political systems, I am still rather puzzled about modeling
the soft (not-easily quantifiable) variables such as morale, integrity, etc.
So far, I have been treating them as some sort of "information" represented
as a nondimensional quantity varying from zero to one. Is this a valid
approach? Or am I missing something or other conservation principles in
the socio-econ-political systems?
The second issue is about approach. I have seen two methods of modeling:
loops and sectors. In loops, the model starts with a description of the
system and its operations in causal loops that represent several key
dynamic hypotheses. In sectors, the model starts with a functional description
of subsystems that create dynamics of the systems. My experience is that the
loops approach is useful for testing hypotheses and looking for levers
(strategic consulting type work) while the sector approach is useful for
concrete modeling efforts (such as models for manufacturing and production,
project management, etc.). Although I had experimented quite successfully
with a hybrid model for strategic modeling effort in Prof. Henry Weils class:
The Dynamics of International Competition. Is this an accurate assessment?
I really appreciate comments, suggestions and, yes, FEEDBACK from esteemed
system dynamicists on this list.
Regards,
Benny Budiman
bbens@MIT.EDU
Modeling for thoughts.
-
jimhines@interserv.com
- Member
- Posts: 41
- Joined: Fri Mar 29, 2002 3:39 am
Modeling for thoughts.
On Wed, 29 May 96, bbens@MIT.EDU wrote in SD0281:
> I have seen two methods of modeling:
>loops and sectors. In loops, the model starts with a description of the
>system and its operations in causal loops that represent several key
>dynamic hypotheses. In sectors, the model starts with a functional description
>of subsystems that create dynamics of the systems. My experience is that the
>loops approach is useful for testing hypotheses and looking for levers
>(strategic consulting type work) while the sector approach is useful for
>concrete modeling efforts (such as models for manufacturing and production,
>project management, etc.)
I think that part of the discussion that his posting will generate will hinge on
the purpose of the model. I think we could have an interesting exchange
concerning the range of "work" that system dynamics models can actually do.
The kind of modeling that I usually do involves insight generation. That is the
"work" of the model is to enrich the mental models of the participants. In this
case you need a strategy that will limit your modeling (you cannot simply model
"the system"); while providing a high liklihood of generating insight. Benny
mentions two strategies: "sectors" and "loops" to which I will add two more:
"model (or molecule) modification" and "main chain".
Let me briefly discuss each strategy and also indicate who I think tends to use
each one. I think that everyone I mention actually uses all the strategies, but
I think that some people tend to gravitate toward one or the other and its this
that Ive tried to capture. If I mis-characterize anyone, please let me know.
1. SECTORS. The approach is in my opinion the sectoral approach. The problem
with the sectoral approach is that it does not provide guidance in how detailed
to make the sector. The tendency is to create sectors that are too complicated.
The sectors, when finally assembled, produce behavior that is complicated,
difficult to analyze, and often unenlightening. These troubles, I believe,
afflicted the National Model, though if Jay disagrees I will retract that
suggestion. The way to make a sectoral approach work is to be clear on what
inputs the sector can use from other sectors and what variables in the sector
will be used by other sectors. The other situation in which a sectoral approach
can work is if you have built similar sectors in prior engagements (but this is
a lot like "model or molecule modification"). A great strength of the sectoral
approach is that it offers an obvious way of dividing up the work in a modeling
team. This is why it was used in the National Model, and why the approach has
made some sense in the "B&B" assignement in MITs introductory system dynamics
course.
2. LOOPS. This is my personal favorite strategy; and I believe that GKA and IA
also favor it. The danger is that the simulation model can end up simply
"confirming" the causal loop hypotheses. In this case you dont learn anything
new from the hard work of modeling. This most often happens when the computer
model is built at the same level of abstraction as the causal loops. In
contrast when the computer model is built at a more operational level; you often
learn interesting things because you create additional loops as you
operationalize. You can learn for example that your "loop hypothesis" actually
is incapabable of generating the reference mode and precisely WHY. You can
uncover new loop-hypotheses that you hadnt thought of before. And, very
commonly, your understanding of your loop hypotheses takes on a subtlety and
thoroughness that is orders of magnitude beyond what you had before. This
strategy is effective in group modeling because relativel inexperienced folks
can easily participate in the "looping" stage.
3. Main Chain. In some cases, often where there is an aging chain that is
central to the problem, you can start simply by mapping the physical stocks and
flows. I believe that the High Performance Systems folks favor this approach.
George Richardson and David Andersen employ this strategy, I believe, as often
as not when they do their group modeling. I think that Jay Forrester used this
approach in coming up with the Urban Dynamics model. There isnt much danger
here, except perhaps that feedback (and consequently insight) may be weak.
Generally speaking, though, I think that if you see a strong physical stock and
flow you should start with it -- even if you have also developed causal loops.
4. Modifying an existing model or molecule. Probably all experienced modelers
use this strategy in almost every project. At one end of the spectrum, this
approach simply means plugging in a "molecule of structure" into your evolving
model: You need a price, so presto, you use the pricing molecule youve used
three or four times previously. (A few people are engaged in creating a
warehouse of molecules). At the other end of the spectrum, this approach means
to actually start with a completed model and then customize it. Pugh-Roberts
used to use this approach with the their project models. (I understand they now
use what would probably be called a super-advanced molecule approach to project
modeling). There arent many weaknesses here, except perhaps that the approach
can prematurely channel your thinking, and that you cannot always count on
having a problem that "youve modeled before". The approachs strength is that
it strongly helps defines the level of aggregation at which you will work and
also pretty much ensures that you will end up with a well formulated model.
Importantly, the approach lets you use a prior investment to the benefit of a
current problem. It is also an approach that becomes easier to use as you
become more experienced.
Regards,
Jim Hines
LeapTec and MIT
JimHines@Interserv.Com
> I have seen two methods of modeling:
>loops and sectors. In loops, the model starts with a description of the
>system and its operations in causal loops that represent several key
>dynamic hypotheses. In sectors, the model starts with a functional description
>of subsystems that create dynamics of the systems. My experience is that the
>loops approach is useful for testing hypotheses and looking for levers
>(strategic consulting type work) while the sector approach is useful for
>concrete modeling efforts (such as models for manufacturing and production,
>project management, etc.)
I think that part of the discussion that his posting will generate will hinge on
the purpose of the model. I think we could have an interesting exchange
concerning the range of "work" that system dynamics models can actually do.
The kind of modeling that I usually do involves insight generation. That is the
"work" of the model is to enrich the mental models of the participants. In this
case you need a strategy that will limit your modeling (you cannot simply model
"the system"); while providing a high liklihood of generating insight. Benny
mentions two strategies: "sectors" and "loops" to which I will add two more:
"model (or molecule) modification" and "main chain".
Let me briefly discuss each strategy and also indicate who I think tends to use
each one. I think that everyone I mention actually uses all the strategies, but
I think that some people tend to gravitate toward one or the other and its this
that Ive tried to capture. If I mis-characterize anyone, please let me know.
1. SECTORS. The approach is in my opinion the sectoral approach. The problem
with the sectoral approach is that it does not provide guidance in how detailed
to make the sector. The tendency is to create sectors that are too complicated.
The sectors, when finally assembled, produce behavior that is complicated,
difficult to analyze, and often unenlightening. These troubles, I believe,
afflicted the National Model, though if Jay disagrees I will retract that
suggestion. The way to make a sectoral approach work is to be clear on what
inputs the sector can use from other sectors and what variables in the sector
will be used by other sectors. The other situation in which a sectoral approach
can work is if you have built similar sectors in prior engagements (but this is
a lot like "model or molecule modification"). A great strength of the sectoral
approach is that it offers an obvious way of dividing up the work in a modeling
team. This is why it was used in the National Model, and why the approach has
made some sense in the "B&B" assignement in MITs introductory system dynamics
course.
2. LOOPS. This is my personal favorite strategy; and I believe that GKA and IA
also favor it. The danger is that the simulation model can end up simply
"confirming" the causal loop hypotheses. In this case you dont learn anything
new from the hard work of modeling. This most often happens when the computer
model is built at the same level of abstraction as the causal loops. In
contrast when the computer model is built at a more operational level; you often
learn interesting things because you create additional loops as you
operationalize. You can learn for example that your "loop hypothesis" actually
is incapabable of generating the reference mode and precisely WHY. You can
uncover new loop-hypotheses that you hadnt thought of before. And, very
commonly, your understanding of your loop hypotheses takes on a subtlety and
thoroughness that is orders of magnitude beyond what you had before. This
strategy is effective in group modeling because relativel inexperienced folks
can easily participate in the "looping" stage.
3. Main Chain. In some cases, often where there is an aging chain that is
central to the problem, you can start simply by mapping the physical stocks and
flows. I believe that the High Performance Systems folks favor this approach.
George Richardson and David Andersen employ this strategy, I believe, as often
as not when they do their group modeling. I think that Jay Forrester used this
approach in coming up with the Urban Dynamics model. There isnt much danger
here, except perhaps that feedback (and consequently insight) may be weak.
Generally speaking, though, I think that if you see a strong physical stock and
flow you should start with it -- even if you have also developed causal loops.
4. Modifying an existing model or molecule. Probably all experienced modelers
use this strategy in almost every project. At one end of the spectrum, this
approach simply means plugging in a "molecule of structure" into your evolving
model: You need a price, so presto, you use the pricing molecule youve used
three or four times previously. (A few people are engaged in creating a
warehouse of molecules). At the other end of the spectrum, this approach means
to actually start with a completed model and then customize it. Pugh-Roberts
used to use this approach with the their project models. (I understand they now
use what would probably be called a super-advanced molecule approach to project
modeling). There arent many weaknesses here, except perhaps that the approach
can prematurely channel your thinking, and that you cannot always count on
having a problem that "youve modeled before". The approachs strength is that
it strongly helps defines the level of aggregation at which you will work and
also pretty much ensures that you will end up with a well formulated model.
Importantly, the approach lets you use a prior investment to the benefit of a
current problem. It is also an approach that becomes easier to use as you
become more experienced.
Regards,
Jim Hines
LeapTec and MIT
JimHines@Interserv.Com