When to use/avoid System Dynamics

[ularoch@attglobal.net]

dear daniel,
the crucial question in our limited experience is put by yourself: if you
ÿneed to follow single objects you use discrete-event. all other cases,
ÿincluding change of product category in a flow are much better served by
ÿcontinuous simuation tools, one of them classic SD-tools. what is the
ÿdifference ?
using discrete-event you have single item following at the price of loosing
ÿoverview of process-control dynamics which then is substituted by
ÿstatistics...and of course you need input at least about 10 to the power of
ÿ4 more data...

yours sincerely
ulrich la roche
From: ularoch@attglobal.net
fast focus consulting group
heilighüsli 18, CH-8053 Zuerich,
fax +411 382 1349

[Rod MacDonald]

Hello,

I would recommend that you check out the following book:


Michael Pidd. (1992). Computer Simulation in Management Science. Chichester:
John Wiley and Sons.


The author does a good job of examining a variety of different simulation
approaches.



Good Luck,
Rod MacDonald

Rod@mindwalkconsulting.com

[Bill Harris]

daniel.jarosch@au.pwcglobal.com wrote:

> I am currently in the process of selecting a modelling approach. does anyone
> have a collection of criteria for selecting the right modelling technique?

If you read German, you might take a look at _Die Praxis des
ganzheitlichen Problemlösens_ by Gomez and Probst, if you can find a
copy.

> In this context I am also wondering how well SD (and probably the common
> software tools) is suited for modelling discrete problems, where you for

Search for discrete on the optimlator archive site at
http://www.optimlator.com/cgi-bin/search/search.pl and you might find
something; it seems to me this has come up a time or two in the past.

Regards,

Bill
--
Bill Harris 3217 102nd Place SE
Facilitated Systems Everett, WA 98208 USA
http://facilitatedsystems.com/ phone: +1 425 337-5541
From: Bill Harris <bill_harris@facilitatedsystems.com>

["Saunders, John"]

Discrete Event Simulation or System Dynamics?

System Dynamics

Modeling Level Strategic - holistic approach
Time Periods under Study Months/Years
Environment under Study Deterministic (within ranges)
Learning Focus Discovery; Satisfactory Answer (no "right"
answer, but many ways to approach the challenge)
Events Modeled Continuous
Problem Type Unique
System Behavior Non-linear, Feedback Loops
Elements Intangible, Abstract Concepts
Examples
1) USMC used SD to model personnel recruitment & training process to improve
readiness levels
2) Dutch Shell used SD to model market fluctuations; staved off disaster in
mid-80s when oil prices dropped
3) Joint AF & Navy program used SD to model the birth to death life cycle of
a weapons system; saved $millions in R&M costs

Discrete Event Simulation

Modeling Level Tactical - component approach
Time Periods under Study Minutes/Hours/Days
Environment under Study Probabilistic
Learning Focus Testing & Finding a specific solution - there is
a "right" answer
Events Modeled Discrete (event triggers a change)
Problem Type Repeating
System Behavior Linear, Sequential
Elements Tangible, Concrete Processes
Some Examples
1)Motor Vehicle Adminstration used it to try different layouts for
processing customers
2) US Air Force used it to study & effect a new Jet Engine overhaul mfg
process; saved $millions
3) US Social Security Administration used it to improve operations in field
offices

There is overlap between the two areas and the techniques emphasized in one
area often do not preclude their application in the other. I have had
students who created very micro level models in SD (e.g. modeling blood
chemistry) and very macro level models using discrete event (money flows
between nations).


John H. Saunders, Ph.D.
Professor of Information and Decision Systems
Information Resources Management College
National Defense University
Fort Leslie J. McNair
300 5th Avenue, S.W.
Washington D.C. 20319-5066
Voice (202) 685-2078
FAX (202) 685-3974
E-mail: saunders@ndu.edu or jsaunders@erols.com
Colleges home page: http://www.ndu.edu/irmc
Professional home page: http://www.johnsaunders.com

[Bill Harris]

Kim Warren wrote:
> Having worked on strategy and policy for over 20 years, I increasingly feel
> that there are virtually no situations where you can safely ignore the

Kim,

You got the essence: accumulation and feedback. The absolute time frame
matters little, IMHO, so hours vs. months is at best a heuristic. If
the time horizon over which your problem exists is long enough for
feedback to be active, then you have a potential SD problem. If your
problem has a time horizon short enough so that feedback isnt active
(bandwidth of the feedback paths sufficiently low), then you dont need
SD--your feedback loops are open in that case, anyway.

Thus its perfectly fair to ignore feedback, if your problem is over a
suitably short time frame (the real system certainly will). For
example, if youre working on a process capacity problem and the time
frame is short enough, say, so that you wont hire new people to augment
your capacity, you may want to use a non-SD model to evaluate
alternative process structures. You might be well advised, though, to
think about the dynamic impact of your decision over a longer time
horizon. Or, rather, you might be well advised to start from a longer,
dynamic perspective to make sure you have a tenable solution space and
then switch to a non-dynamic model to evaluate specific process
alternatives over a suitably short time horizon to see which costs less
or works faster, for example.

The dichotomy of SD vs. discrete is a false one, imposed by the tools we
often use. The dichotomy of aggregation vs. tracking individual items
is also tool-imposed. I can conceive of a Colored Petri Net model, for
example, with accumulation and feedback that is both validly an SD model
and also tracks characteristics of individual units and is quite
discrete in nature. I also cant remember seeing such an integrated
SD/Petri Net model (this is not a multi-model as described by Fishwick)
published or described anywhere; its on my todo list, but its been
there for quite a while.

Bill
From: Bill Harris <bill_harris@facilitatedsystems.com>
--
Bill Harris 3217 102nd Place SE
Facilitated Systems Everett, WA 98208 USA
http://facilitatedsystems.com/ phone: +1 425 337-5541

["Thorsten Muehlhausen"]

Hallo Daniel,

in my PhD thesis I used SD (Tool: Powersim) to model the passengerflow at
and between airports. It worked really well for this purpose. But it is a
macroscopic model, i.e. I dont look at single individuals, but a flows (of
passengers). From my experience I dont think, that SD is suitable for
microscopic modelling.

Unfortunately my PhD thesis is published in German, so a link, where to find
a pdf-file, wouldnt help.

But if you have more detailed question, I would try to answer them.

Best regards,
Thorsten
From: "Thorsten Muehlhausen" <thorsten.muehlhausen@dlr.de>

["Kim Warren"]

Having worked on strategy and policy for over 20 years, I increasingly feel
that there are virtually no situations where you can safely ignore the
fundamentals of SD. Reasons being (a) most every challenge in business and
other fields of policy concerns how things are changing through time (b)
things change through time because various bath-tubs in our world fill up or
drain away (c) this filling and draining process always reflects the current
state of other factors in the situation.

This creates two ubiquitous and unavoidable consequences - you absolutely
have to account for accumulation processes, and you absolutely cant avoid
the need to deal with feedback. Both processes confound any attempt to
discover simple linear solutions to policy challenges. Im no mathematician,
but my understanding is that integral calculus underlying SDs handling of
accumulation processes is the only reasonable method for tackling these
process (and even if other methods can be forced to cope, why bother when
you have the perfect tool to hand?)

Some have argued that theres no need for a dynamics perspective when
looking at one-off events - e.g. mergers and acquisitions - but the
evidence is piling up that such switches in strategy and policy stand or
fall on their dynamic consequences.

This is not to say that large, detailed computer simulations must be built
in every case, but it is quite possible to make a solid start on the above
unavoidables with paper and pencil (or preferably, whiteboard and marker),
including incorporating the numerical facts of the case. Though the
non-intuitive behaviour of accumulation and feedback mechanisms (plus the
frequent need to deal with at least some of the detail) make modelling an
important step in all but the simplest cases.

Kim Warren - London Business School
From: "Kim Warren" <kim@strategydynamics.com>

[daniel.jarosch@au.pwcglobal.com]

dear sd-world,
I am currently in the process of selecting a modelling approach. does anyone
have a collection of criteria for selecting the right modelling technique?
I would really appreciate if anyone in the mailing list could provide me with
answers or even examples when it is most appropriate to use system dynamics
(e.g. when feedback is involved) and when should i avoid using an system
dynamics approach?

In this context I am also wondering how well SD (and probably the common
software tools) is suited for modelling discrete problems, where you for
instance need to be able to track the individual unit/item.
I am sorry if this question has already been risen but I have just subscibed to
this mailing list and could not find too much in the archive as well.

Thanks & Regards

daniel
From: daniel.jarosch@au.pwcglobal.com

[Niall Palfreyman]

daniel.jarosch@au.pwcglobal.com schrieb:

> In this context I am also wondering how well SD (and probably the common
> software tools) is suited for modelling discrete problems, where you for
> instance need to be able to track the individual unit/item.

I second Daniels query. I come from the world of software engineering,
where the available modelling techniques (eg, UML) are all centred
around discrete problems. Id be interested in the ways in which these
two apparently very different worlds of software modelling and SD
modelling compare and differ. When should one be chosen over the other,
and how might they be effectively combined?

All the best,
Niall.
From: Niall Palfreyman <niall.palfreyman@fh-weihenstephan.de>

[Rahul Lanjewar]

Hi Daniel,
Your problem is interesting and probably something that I can help with in a
small way, especially the latter part of your question. I am working on a
project to simulate an entire organisation. One specific requirement is to
have a unit wise tracking of individual projects. Here, I was not able to
come across a suitable software or tool that could help me with this. I
evaluated a couple of them, inlcuding Vensim and iThink. Both of them are
useful for an aggregate level modeling and analysis. To do my modeling, I
have used VC++, where I have been able to model the unit wise tracking.
About the various modeling approaches, I am not sure I understand your
question. I, too, would be grateful if somebody could highlight the various
approaches. This would help me in buidling my model further.
Probably, you could give more details on what exactly you are trying to
model. I guess we could be of help to each other in that case.

Thanks and best regards
Rahul
From: Rahul Lanjewar <lrahul@infy.com>

----------------------------------------------------------------------------
----------------
Rahul Lanjewar
Corporate Planning Department,
Infosys Technologies Limited,
Bangalore - 561 229
Tel: 91-80-8520261 Extn 7832
----------------------------------------------------------------------------

["Brian Dangerfield"]

Daniel:

To keep a reply as short as possible, I would say that if you are
modelling individual entities at the operational level in order to
assess decision rules and with time units in minutes, hours or
days then you need discrete event simulation.

If your problem is a policy issue at the aggregate level of an
organisation and time units are months or years then SD will prove
worthwhile.

Best wishes


Brian.
From: "Brian Dangerfield" <B.C.Dangerfield@salford.ac.uk>
Prof Brian Dangerfield
Professor of Systems Modelling
Centre for OR & Applied Statistics
Maxwell Building
University of Salford
SALFORD M5 4WT
U.K.
Tel: 44 161 295 5315
Fax: 44 161 295 2130

["Lane,DC"]

When it comes to choosing between SD and DES it certainly is worth
remembering that all modelling approaches are very crude and that the choice
is essentially an artificial one. We should constantly remind ourselves that
they are idealisations and representation schemes. In this sense "Stocks are
really out there" is not so much false as meaningless.
However, a worse of better job of tackling a problem can be made depending
on the tool and so the distinction between the two approaches is worth
trying to sketch out. My view is that although there are areas of
commonality there are also important differences and that any engagement
between these two approaches needs to be mindful of both.

Almost a year ago the UK Simulation Society organised a meeting to discuss
the topic. Details of the opening paper - in which I argued for engagement
based on mutual knowledge - are given below:

You Just Dont Understand Me:
Modes of failure and success in the discourse between system dynamics and
discrete event simulation
System dynamics modelling and discrete event simulation are established
simulation techniques. One might reasonably expect there to be strong and
mutually supportive communication between the two. In reality,
communication, though existing, is very limited and when it does take place
seems fraught with mutual incomprehension and even lurking hostility. This
paper suggests that three modes of discourse may be determined.
Mode 1 concentrates on the differences between the two and supports the
conclusion that the present institutional divide should continue. Mode 2
concentrates on the similarities and proposes that they are essentially the
same idea. It is argued that these discourses are both failure modes because
they have a superficial grasp of the simulation two approaches. Mode 1 fails
to acknowledge the undoubted connection between them, from the broad level
of modelling aspirations down to some of the technical aspects of
validation. Mode 2 also fails but because it incorrectly elides distinctive
differences, for example, concerning the decision problems that each
approach is able to treat.
This leads to the proposal of Mode 3, in which a richer understanding of
SD and DES allows the two camps both to grasp opportunities for making
common
cause in areas of similarity, and to have an informed and respectful
understanding of the reasons why there are areas of difference. Mode 3 is
recommended as the only long term success mode for discourse between SD and
DES.

Copies of the LSE working paper are available from me in PDF for those who
e-mail me direct.
From: "Lane,DC" <D.C.Lane@lse.ac.uk>
> Dr. David C. Lane
> Management LSE
> London School of Economics and Political Science
> Houghton Street
> LONDON
> WC2 2AE
> GB
>
> Tel: (UK)(0)20 - 7955-7336
> Fax:(UK)(0)20 - 7955-6885
> e-mail: d.c.lane@lse.ac.uk

["Kim Warren"]

thanks to Bill Harris for his reply, My proposition is somewhat different
than the question of when can you ignore feedback. There are very many
strategy situations where the timescale of change is short enough for major
feedback effects to be modest or easily understood, but there are *not* many
situations where the dynamic consequences of accumulation and depletion can
be ignored. This is especially the case if (as is usually the case) managers
want some idea of the scale and timing of their future performance, rather
than merely directional insight.

What makes matters worse, since accumulation and depletion are already
operating throughout the system when you get to timescales where feedback
does operate, the integral calculus is very likely to confound simple
observations of likely loop behaviour. Try creating a simple word of mouth
R structure, but just add a completely linear in-or out-flow to the stock
of customers ... you end up with a huge variety of potential behaviours
besides the exponential growth and decline curves that casual users of
feedback diagnosis imagine to be the usual behaviour of R systems. (I
realise this wont be news to experts).

Hence my stance - that no strategic challenges can be adequately tackled
without working through the dynamics of accumulation and depletion - a task
that can often be carried out, quantitatively, on paper or whiteboard,
without the need to build computer models. When many accumulation/depletion
processes are interacting, I advise managers that a model can help ensure
they get the numbers right. And I fully accept, of course, that when
timescales switch on the likelihood of serious feedback, then modelling
becomes essential.

Kim Warren - London Business School
From: "Kim Warren" <kim@strategydynamics.com>