Tools or Toys?
On Specific Challenges for Modeling and the Epistemology of Models in the Social Sciences

Table of Contents

2.2 Where models get their credentials from

If models are mediators between theory and reality in the just described sense, then the next question would be what gives models their credibility or how they can be validated. Obviously, if models are partially independent from theory, we cannot rely on the credibility of the background theory alone. Instead, models draw their credibility from three different sources:

1. Credible background theory and background knowledge: In so far as it makes use of background theories, the models' credibility depends on the credibility of the background theory. And the validity of the model depends on how faithful it is to the background theories (where it makes use of them). The same holds for any factual background knowledge that is incorporated into a model.
    
2. Well approved modeling techniques: A model furthermore draws credibility from the credibility of modeling techniques that have been employed in its construction (Winsberg 2006). These techniques in turn are credible either because they can be analyzed or tested with regards to their reliability or simply because they have been successfully employed in the past.[1] In order to validate this aspect of the model one would have to inquire into the reliability of the modeling techniques used and also check whether they have been employed correctly.
    
3. Successful empirical tests: Finally, a model derives credibility from being in accordance with the target system as assessed by empirical tests. The direct validation through empirical testing may not always be possible, though. In fact, one of the most important uses of computer simulations is as substitutes for experiments in cases where experiments are costly or impossible.

If these are the sources of credibility for models, then the question arises if and how they are be related to each other. The following two conjectures about the mutual relation of these sources seem reasonable:

1. Precedence of empirical validation: If reliable empirical tests of a model are available, then empirical validation takes precedence over the other validation paths. This means: If a model does not seem to be valid in terms its theoretical assumptions or the employed modeling techniques but withstands empirical testing nonetheless then the model is still acceptable if only as a phenomenological model. The precedence of empirical testing as a validation criterion reflects the epistemic primacy of empirical facts in science.
    
2. Synergy of credibility sources: The less one can rely on a particular one of the three above mentioned sources of credibility, the more strain is put on the remaining sources. E.g. if empirical testing of the model is not possible then the more important it becomes to be able to rely on a well confirmed background theory or on well-proven and reliable modeling techniques.

It seems reasonable to distinguish the models that derive their credibility primarily from the reliance on background theories, background knowledge and modeling techniques from those that are validated by direct empirical testing. The former could be termed “ input-controlled” models and the latter “output-controlled” models. The distinction is of course one of “more or less”. Some models may be both input and output controlled. This distinction is meaningful, because with these two ideal types of models are associated quite different modes of validation. With this terminological convention no general assumption is made about the relatively greater or smaller reliability of the one or the other. But it stands to reason that different levels of credibility or reliability might be associated with input or output-controlled models in specific contexts.

There is not much more that can be said about the credibility of models on this very general level. Further below a case study will be discussed in order to show how these three sources of credibility come into play in a simulation model. But before, a few things need to be said to justify why the terms “model” and “computer simulations” are used more or less interchangeably in this paper.

[1] In the latter case, however, their success must at least at some point in the past have been assessed by more direct means.

t g+ f @