From a Manageable Set of Relationships to the Model 
Figure 1: Four Levels of 
Abstraction in Development 
of a Simulation Model 
©) Se 
3) 
Model 
(Z) 
Manageable 
Set of 
Relationships 
A) 
General 
Cohceptual 
Scheme 
Real World 
———+ 
Higher degree of abstraction 
Less applicability 
More structured representation 
Higher degree of aggregation 
More explicit assumptions 
activity are attracted to each other by some force that 
can be measured by mass, or size, of population and 
distance from the center of activity. The general con- 
ceptual scheme can be structured to some extent and 
presented as a set of statements or even as flow charts or 
diagrams, but it is too vague and comprehensive to be 
used for meaningful explanatory or predictive statements 
about the urban environment. Elements of the general 
scheme can be ranked according to perceived degree of 
resemblance to real world phenomena, certainty of oc- 
currence, or relevance and importance to the purpose of 
the model, and such a ranking can assist in the next step 
of abstraction. 
From a General Conceptual Scheme to a Manageable Set 
of Relationships 
The general conceptual scheme is not directly useful in a 
simulation model; it must be further refined and abstracted 
to a manageable set of relationships. Statements are re- 
quired that observe not only the existence of the rela- 
tionships but also their form. 
The manageable set of relationships is narrower in scope 
than the general conceptual scheme. Irrationality, 
chance and individual behavior usually are set aside and 
only relationships that demonstrate known or assumed 
logical connections are retained. Quantitative or meas- 
urable statements take precedence over qualitative 
statements. Care is taken to exclude relationships re- 
quiring data that do not exist, or that cannot be collected 
or analyzed, or data that are incomplete, biased or 
error-ridden. 
The manageable set of relationships can be visualized 
and presented in much the same way as was the general 
conceptual scheme, the difference being that the state- 
ments are fewer, more precise, simple and clear. They 
are less true and descriptive of the real world than the 
general scheme, but they are statements on which a 
model can be based. 
The model is a formalization of the manageable set of 
relationships, a stylistic interpretation of the propositions 
contained in them. All relationships in the abstract system 
must be made explicit and symbolic, and those which 
cannot be are not included in the model. 
The model is made operational by determining the par- 
ticular parameters and decision rules that pertain to the 
specific context in which the relationships are to be 
tested. The fitting of parameters for the model is in itself 
a process of abstraction. From an array of possible meas- 
ures the analyst chooses one or a few to represent them 
all: the mean, or mode, or the value "most likely" to 
give a satisfactory relationship. Again using an illustra- 
tion from the simulation model presented in the following 
section, in structuring an equation for the attractiveness 
of a tract of land, the parameters were set as trip-distri- 
bution functions and were fitted independently, based on 
data gathered by a traffic survey. 
With the simulation model we now have a formalized ab- 
stract system represented symbolically by equations, flow 
charts and diagrams, and logical statements of relation- 
ships, that is capable, of reproducing phenomena likely 
to occur in the real world system. It can be used, under 
test conditions, to generate a state of that system. But 
until programmed for a computer it has limited practical 
value, 
From the Model to the Computer Program 
The staircase of abstraction in Figure 1 shows that in 
going from the model to the computer program we achieve 
the final abstraction in system simulation. Abstraction is 
involved in this step because of constraints in programm- 
ing languages and techniques, limitations in the capacity 
and speed of the computer itself, and the cost of pro- 
gramming and running the model. 
The language in which the program is written sometimes 
tends to limit the type and form of relationships that the 
model can contain. That is, the method of modeling may 
to some extent depend on the language that will be used 
to program it. Some languages are well suited for pro- 
gramming time series and probability models, for example, 
while others are more suited for describing cross-section 
and deterministic relationships. Since a full choice of 
programming languages sometimes is not available to the 
analyst, the need frequently exists to design the model to 
fit that which is available, and this may result in a fur- 
ther distortion of realitv 
Limitations of capacity and speed of the available com- 
outer tend to limit the number of variables, parameters 
and statements of relationship and their complexity, that 
the model can contain. This limitation may be absolute, 
in physical terms, or it may be felt as a cost constraint. 
The ideal and unabridged program may simply take too 
Jong and cost too much to run. 
After all this abstraction one may well wonder whether 
the computer output still has meaning and validity in 
terms of the real world system ‚it is supposed to represent. 
One may ask, since the simulation model is to be used as 
an experimental laboratory, how much faith can be put in 
the experiments. 
ARCH+ 2 (1969) H.8