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

Eckhart Arnold

1 Introduction
2 The role of models in science
3 Why computer simulations are merely models and not experiments
4 The epistemology of simulations at work: How simulations are used to study chemical reactions in the ribosome
5 How do models explain in the social sciences?
6 Common obstacles for modeling in the social sciences
    6.1 Lack of universal background theories
    6.2 Pluralism of Paradigms
    6.3 Why parsimony is a vice and not a virtue
    6.4 “Wholistic” nature of many phenomena in the social sciences
    6.5 Difficulties of measurement
    6.6 Pluralism of scientific styles
7 Conclusions

6.3 Why parsimony is a vice and not a virtue

Many important phenomena in the social sciences are characterized by the fact that they may be caused in many different ways. While this can happen in physics, too, it seems to be a standard case in the social sciences. Take for example, the outbreak of war. There are many different reasons why a war can break out. In each single instance of an outbreak of war there is usually a bundle of different causes involved. And in different instances of an outbreak of war probably different bundles of causes lead to the same effect, namely, the outbreak of war. Finally, it is in most instances difficult to determine which of a number of possible causes were decisive. How can historians deal with these problems?

The best way to deal with this situation is to consider all reasonable assumptions about what the causes in a particular instance of the outbreak of war are. And where the evidence remains insufficient as to whether a particular possible cause was indeed relevant, it is better to at least mention this cause as a possibility than to leave it out completely.

Because multicausality in the just described sense as well as the evidential underdeterminacy of particular possible causes are typical features of historical explanations, the discipline of history has developed a scientific culture that in some respects runs contrary the scientific culture of the natural and technical sciences. Most importantly, historians do not consider an explanation as better just because it is simpler. Quite the contrary, an explanation in history is the better the more “differentiated” it is. The reason for this attitude is that it is always easy to cook up a simple story. (The most simple explanations are those that rely on ideologies, e.g. “history is the history of class struggles.”) But it is usually much more challenging to get all the details right. It is therefore no surprise that among social scientists the charge of “monocausality” is almost a kind of an insult. In history as well as many other social sciences, explanatory parsimony is a vice and not a virtue.

This conclusions can be generalized to all cases where multicausality is involved and where it is practically impossible check the relevance of all potential causes. In this situation, it does not make any sense to demand that explanations should be as simple as possible. For, there is no way of determining when an explanation has become too simple.

What consequences does this have for the employment of models and simulations in generating explanations. One consequence is that simple models that demonstrate merely logical or - as the practitioners sometimes prefer to say - “theoretical” possibilities are at best a small piece in the puzzle. They are “partial explanations” in the sense of Aydinonat (2007). And before they can be considered a part of a full explanation it must be checked whether the so demonstrated theoretical possibility is a real possibility in the given situation and how it compares to other possible explanations.

Often, unfortunately, the surplus in explanatory power to be gained by simulation models that merely demonstrate logical possibilities is almost negligible. This can be seen, for example, when comparing Robert Axelrod's account of the informal truces between soldiers of opposing forces on large parts of the western front in World War I in terms of his simulations of the repeated prisoner's dilemma with the original historical study by Tony Ashworth on which Axelrod based his account (Arnold 2008, p.\ 180-189).

Linking to the discussion whether in the social sciences KISS (“keep it simple stupid”) models are better than KIDS (“Keep it descriptive stupid”) models, one might now conclude that this is a problem of KISS models in particular (see Pyka/Werker (2009) with further references).

Without entering into the full discussion here: From their approach and their own aspiration KIDS models do indeed avoid to be overly parsimonious. Yet, they are plagued by many problems of their own like being more difficult to understand, often lacking robustness or, despite being more complex, still not coming close enough to empirical reality to be of explanatory value.

t g+ f @