Level of Detail in Models

["John Gunkler"]

While I appreciate Carolus Gruetters attempt to clarify the question about
model "details" I guess I dont believe that "details" are usefully thought
of merely as non-essential elements.

First, this implies that the "level of detail" discussion boils down to a
simple two-part discrimination: essential vs. non-essential. I think this
does a disservice to those who struggle with understanding the value of
including or not including levels of detail. There are more than two
possible levels of detail.

Second, I believe that "level of detail" may not be a unitary concept. That
is, part of the issue with "details" seems to me to be an issue of where to
draw the model boundary -- how much to include in the model and how much to
leave out. Once the boundary decision is made, however, there is another
issue of modeling "distance" (as Dr. Forrester describes it.) This is an
issue of level of "aggregation" of model components.

I believe that decisions about model boundaries and decisions about levels
of aggregation are often made based on different criteria. Model boundaries
are chosen (primarily) so as to include the least number of components
necessary to generate the behavior of interest. Levels of aggregation are
sometimes chosen on a similar basis but may be based on other criteria:
model "face" credibility (for those who must use it), how useful these
"details" are in helping people understand the model (and the system being
modeled), the ability of the model to faithfully represent prospective
policy changes, explanatory power of the model, effort required to include
lower levels of aggregation (versus payoffs of doing so), etc.

The differences between "details" having to do with model boundaries and
"details" related to level of aggregation seem significant enough to me that
I suggest we be more precise when we discuss "details" and specify which of
these two kinds (or any others -- are there others?) we are talking about.


John W. Gunkler
jgunkler@sprintmail.com

["Anil Sahai"]

This is a double edge sword. In my experiennce over the years, request for
"additional details" is made:

1) before signing off the contract, clients are apprehensive
about the success of modeling. They just do not believe that
models can provide insights to the problems they are facing
because they sincerely believe that since the model does
"include" every minute detail, it "cannot" explain!
2) after the modeling results are presented, and the results
are "different" from the customers own "intuition" or
"gutt" feeling - they respond to asking for more details
in the model as if the "details" will "skew" the results
in "their" direction.
3) sometimes they are "correct" in providing a crucial
detail that was not told before - and that needs to be
included in the model.

I totally agree with Jim Hines in educating clients as why
it is "OK" not include every "minute" detail in modeling. A
good analogy is when giving directions to my house, it is
sufficient to say that take HWY X, then Exit Y, and Turn Right
on 2nd Stop Sign, and then 4th House on Left. Instead of
1) get ready
2) go thru your THIS IS SPAM door
3) turn the nob clockwise (approx 90 degrees)
4) take a step forward
5) take another step down - watch for step down
6) turn left and switch on garage light
7) ...
... and so on.

.. anil sahai
Performance Vision
From: "Anil Sahai" <anil_sahai@hotmail.com>

[carolus]

The discussion on Level of Detail in Models now seems to me like a loop.
Maybe someone can falsify it. I observe the following:

* we dont understand why X occurs
so we have a problem
* that problem can be solved if we know why it occurs
so we need (more) insight
* in order ot increase our insight
we try to model the problem
* the model is not satisfactory in explaining why X occurs
so we (still) have a problem
* that problem can be solved if we know why it occurs
so we need (more) ... etc. etc .

The question whether to include certain DETAILS in a model is a misleading
one - I think. I dont think anyone will disagree with the statement that one
should make things as simple as can be, but not simplier (I hope Im
quoting Einstein properly). So, do we have to include lots of details? No.
Asking this type of question implies a negative answer. It is a rhetorical
question. Details are not the same as essentials

Yet, there is, I think, another underlying more essential question.
The qualification of certain data as details implies that one allready has
decided that these data are not essential (i.e. details).
For instance Anil Sahai wrote:

> ...
> 3) sometimes they are "correct" in providing a crucial
> detail that was not told before - and that needs to be
> included in the model.

He makes the distinction between (every minute) detail (which can be left
out) and crucial detail (that needs to be included) and observes that
sometimes certain details have to be included. However, this seems to me a
terminological item: what is the meaning of the term detail.
The main point, I think, is how to discover what is crucial and what is not?
In other words, what information has to be qualified as not-essential and
what information has to be qualified as essential, and this qualification
inmediately labels non-essential information as detail and essential
information as crucial.
So, I think, that the underlying question does not regard the number of
details but the decision (resulting from a certain point of view) whether
certain data have to be qualified as superfluous or as crucial.

If this observation is correct, the actual question, as I perceive it, is not
about ,the level of detail in models but about the (definition of the)
purpose of the model which implies a qualification of certain information as
detail and other information as essential.

Carolus Grütters
Law & IT
Faculty of Law
Nijmegen University
The Netherlands
carolus@jur.kun.nl
http://www.jur.kun.nl
it/