Page 215 - DCAP601_SIMULATION_AND_MODELING

Unit 12: Design and Evaluation of Simulation Experiments (II)

“Data already acquired from the real system” were not available in a form that could be matched Notes
to the bulk of data simulation models had generated. There are several reasons for this difference
between natural and social sciences:
Data collection is a very expensive task in the latter, and in most cases it is even impossible to
generate long time series for individual or group behaviour — individual attitudes, e.g., may be
changed by the very measurement process, and groups may have changed in their composition
before they were able to generate a time series which would have been long enough to allow for
parameter estimation. On the other hand, the different kinds of influences non-living things
exact upon each other are very much limited in their number, such that a structurally valid
model can much more easily be found for the target systems natural sciences deal with than for
social systems.
When talking about structural validity, a digression on structuralism might be in order:
Structuralism as defined by Sneed (1979) and Balzer et al. (1987) sees both simulation models and
observations as models of a theory which in turn — for them — is a mathematical structure
consisting of (among others) three sets of such models.
And these models — full models, potential models, and partial potential models — are defined
as lists of terms and functions and (in the case of full models) invariants.

Observations in this structuralist programme in the philiosophy of science are intended
applications of a theory, they are a subset of the set of its partial potential models in a sense that
we can talk about them in terms which are non-theoretical with respect to a theory T in question
(“T-non-theoretical terms”, for short). Elsewhere it was shown that a simulation model “of a
theory” is “analogous to a structuralist reconstruction of this theory”, and that such reconstructions
can easily be translated into simulation models and vice versa (Troitzsch 1994), provided the
simulation language is object-oriented and functional (in other simulation languages the
translation might be less straightforward). Simulation models would then be translated into
full models in so far as they contain both T-non-theoretical terms (those we can use for talking
about the target system irrespective of whether the theory is validated or not) and its T-theoretical
terms — those which are only introduced by the theory, “in the sense that their meaning
depends on T”, (Balzer et al. 1987: 40) — and, thirdly, the axioms or invariants the theory
postulates — whereas observations (or rather: intended applications, to keep to the terminology
of structuralism) are only partial potential models listing just the terms which are non-theoretical
with respect to this theory.

Thus, simulation is “richer” than observation.
Validation of simulation models is thus the same (or at least analogous) to validation of theories.
In the sense of structuralism, we can interpret validation as the attempt at finding whether there
exist intended applications of a theory (observations to which the theory refers) which belong to
the content of the theory—which means that it should be possible to make an observation (in
Tnon Tnon-theoretical terms) which complies with the axioms of the theory (which in turn
might be expressed in Ttheoretical terms, but then these must be linked to T-nontheoretical
terms).
What does this mean for agent-based simulations in the range defined in the introduction?
Sugarscape agents and plants correspond to T-theoretical terms, and the rules which the agents
obey correspond to the axioms of this theory. But is there any empirical claim of the theory
behind Sugarscape? If this theory predicts that — with a given parameterisation and initialisation
— macrostructures emerge from the microstructures programmed into “its” models, and the
emerging macrostructures sufficiently resemble observable macrostructures, we could admit
that this observable macrostructure together with its microstructure (provided it resembles the
model’s microstructure) is an intended application of the theory behind Sugarscape and that it
complies with its axioms.

LOVELY PROFESSIONAL UNIVERSITY 209