Isolation Is Not Characteristic of Models

Abstract

Modelling cannot be characterized as isolating, nor models as isolations. This article presents three arguments to that effect, against Uskali Mäki's account of models. First, while isolation proceeds through a process of manipulation and control, modelling typically does not proceed through such a process. Rather, modellers postulate assumptions, without seeking to justify them by reference to a process of isolation. Second, while isolation identifies an isolation base—a concrete environment it seeks to control and manipulate—modelling typically does not identify such a base. Rather, modellers construct their models without reference to concrete environments, and only later seek to connect their models to concrete situations of the real world. Third, Mäki argues that isolation employs idealization to control for disturbing factors, but does not affect the factors or mechanisms that are supposed to be isolated. However, models typically make idealizing assumptions about the factors and mechanisms that are the focus of investigation. Thus, even the product of modelling often cannot be characterized as isolation.

Acknowledgements

Earlier versions of this article were presented at the INEM (International Network for Economic Methodology) conference in Madrid, September 2008, and at the workshop Models, Mechanisms, and Interdisciplinarity in Helsinki, November 2008. I thank the participants for valuable criticisms and in particular Tarja Knuuttila, Uskali Mäki, and two anonymous referees of this journal for very helpful comments. The usual disclaimers apply.

Notes

Because there are material models, theoretical isolation may consist in the manipulation of a real entity that functions as a representation. In most social sciences, however, material models are insignificant.

Note that when speaking about an ‘isolated essence’, Mäki refers to the way the world is partitioned, not about the way representations are isolated. Thus, he acknowledges the substantial ontological claim he is making.

A related distinction can be found in McMullin (1985, 255). He distinguishes between ‘construct idealizations’, where the simplification is worked on the conceptual representation of the object, and ‘causal idealization’, where the simplification is worked on the problem situation itself.

For an in-depth comparison of Mäki's three concepts of isolation, see Grüne-Yanoff (2011).

In personal communication, Mäki clarified that there is no conceptual difference between his term ‘resemblance’ and the more commonly used term ‘similarity’.

This lack of reality base is often diagnosed in economic models. About Lucas's ‘model economy’, for example, Knuutilla says: ‘the assumptions made are patently artificial in the sense that it is difficult to imagine how they could have been drawn from the economic reality’ (Knuutilla 2009, 63).

There is, however, an alternative and much weaker reading. The above quotation may be read as a merely intentional characterization: modellers intend to close a system. Such a characterization would point not to a property of the model, but merely to a property of the modeller. This would jeopardize any intersubjective criterion for the explication or assessment of models and modelling practice. I think it is worthwhile to resist such a reading, and ask for a stronger characterization of models and modelling. I also think that Mäki would want to resist such a reading.

Recently, axiomatic representations of incomplete preferences as sets of utility functions have been developed (Dubra, Maccheroni, and Ok 2004). However, it is unclear what sort of demand functions, equilibra concepts or welfare functions could be derived from such representations; it is only clear that these notions would be significantly different from the existing ones, and that the properties defined for unique functions do not apply to sets of functions.

In particular, robustness analysis would not be of help here. Analysing the robustness of model results under varying assumptions requires varying these assumptions. But as I argued, completeness is constitutive of many other microassumptions, and hence cannot easily be varied without affecting the whole set of model assumptions.

Additional information

Notes on contributors

Till Grüne-Yanoff

Till Grüne-Yanoff is at the Helsinki Collegium of Advanced Study.