Page 213 - DCAP601_SIMULATION_AND_MODELING

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

12.2.2 Explanation and Prediction Notes

There is a long discussion about the question whether explanation and prediction are equivalent,
or, to put it in other words, whether a theory which predicts empirical observations correctly at
the same time explains what it predicts. Gr¨unbaum (1962) pleaded for the equivalence while
Scriven (1969) pleaded that both were “nonsymmetrical”.

If we consider prediction and explanation equivalent then our first example above would have
explained the gender desegregation in German schools observed in the second half of the 20th
century (although this was only retrodiction, but in principle, the three assumptions could have
been stated in 1950), but this explanation is of the same quality as the explanation

Mesopotamian priests could give 2,500 years ago for their (mostly correct) predictions of solar
eclipses. In both cases, some scepticism in in order: from our research into the history of school
staffs we know that desegregation had different causes than those stated in the assumptions, and
the Mesopotamian theories of planetary movements were superseded 400 years ago by new
theories which are substantially more valid.
The controversy between Gr¨unbaum and Scriven, however, was different: Scriven had argued
the other way round: “Satisfactory explanation of the past is possible even when prediction of
the future is impossible.” (Scriven 1969: 117; Gr¨unbaum 1962: 126) while we argued above that
even when prediction of the future is possible with the help of a theory, this does not mean that
this theory satisfactorily explains what happened (another theory could yield the same prediction
and deliver a better explanation).
Without going into the details of this old controversy we should instead discuss what explanation
and prediction could mean in the context of (social) simulation. Epstein and Axtell argued that
explanation of a phenomenon is achieved once the phenomenon could be reconstructed or
generated (“grown”). From this point of view, the development of the distribution of percentages
of female teachers in German grammar schools is explained by the three assumptions mentioned
above, since this time-dependent frequency distribution as a macrostructure could be reconstructed
quite well from the microstructure defined in the three assumptions. Of course, this reconstruction
is by no means quantitatively precise: the two graphs are similar, but not identical (perhaps due
to some simplifications in the assumptions, perhaps due to the fact that the random number
generator in the simulation run which generated the time-dependent frequency distribution
was not perfect, or for any other reasons) — and, of course, Sugarscape explanations are of the
same, nonquantitative type.
What simulation models like these are designed to predict is only how a target system might
behave in the future qualitatively; what we want to know is whether any macrostructures might
be observed and what these macrostructures might look like, given that on a micro level some
specific rules are applied or some specific laws hold. This is what we should call a qualitative
prediction which at best would tell us that a small number of categorical outcomes can be expected
with their respective probabilities. But this is not the type of prediction as the objective of simulation
“which most people think of when they consider simulation as a scientific technique” (Axelrod
1997: 24)—”most people think of” attempts at simulating planetary formation (Casti 1996: 14)
instead of “simulating the movement of workers or armies”. But if we use prediction in a non-
quantitative sense, predictions delivered by simulations might still be useful “for the discovery of
new relationships and principles” which Axelrod finds “at least as important as proof or prediction”.
They might answer questions like “Which kinds of macro behaviour can be expected from a given
micro structure under arbitrarily given parameter combinations and initial conditions?” The
definition of this micro structure will typically be derived from observations on the micro level,
and the simulated macro structures will typically be compared to macro structures in the target
systems (which perhaps have not even be discovered).

LOVELY PROFESSIONAL UNIVERSITY 207