qualitative data, fuzzy sets and table functions in

["Thompson, Jim A-136"]

Gebhard Wulfhorst asks three questions:
1) What kind of knowledge and experience exists in integrating qualitative
data, estimated values or fuzzy information in the structure of differential
equations between stock and flow?
2) How can table functions be specified in order to point out this level of
information?
3) How do you avoid black box correlations in these cases?

Jay Forresters Industrial Dynamics describes the processes of estimating
constants and their cousin, the table function. And John Stermans Business
Dynamics responds to these questions in Chapter 14, Formulating Nonlinear
Relationships. His discussion of theory is illustrated with practical
examples of how expert modelers have tackled at least the first two
questions.

The third question is about communication with the people who help to build
the model and who use the model. If the model builders document sources for
table function estimates in the form taught by Professors Forrester and
Sterman, model users arent likely to feel that table functions are black
boxes. There are several books and journal articles on group model-building
that tackle the full disclosure problem. For example, John Morecroft and
John Sterman edited a volume of modeling experiences, Modeling for Learning
Organizations, with many practical suggestions for keeping individuals and
groups informed and aligned in the processes of model development,
formulation and implementation.
Jim Thompson
From: "Thompson, Jim A-136" <Jim.Thompson@CIGNA.COM>
Medical Cost Analysis & Forecasting
CIGNA Information Management
900 Cottage Grove Road, A136
Hartford, CT 06152

[Joel Rahn]

There is at least one representation theorem used in fuzzy control theory
that says, in effect, that an arbitrary non-linear function of one variable
can be approximated by a fuzzy relationship combined with a
de-fuzzification procedure. Thus, in some sense, all TABLE functions can be
thought of as representations of fuzzy relationships. This does not mean
all TABLE functions can be (much less need be) developed by eliciting fuzzy
relationships. But then approximate (fuzzy-type) arguments in favour of
some TABLE functions and their shapes have always been a part of model
conceptualization activities in SD.

Joel Rahn
From: Joel Rahn <rjrahn@videotron.ca>