All Models Are Wrong. When Are They Useful?

In my career as a researcher I have heard the phrase “all models are wrong, but some are useful” on more than a few occasions.¹ The first half is easy enough to motivate. All scientific models are simple approximations of complex realities. By their nature, they strip away a great amount of detail to make legible and tractable certain relationships at the expense of others. They depend on quantifiable measurements which themselves act to smother complex qualitative differences into numeric forms usable within models.

All models, then, are wrong. That isn’t the part of the phrase that particularly needs motivating. The part that does is the second part: that some are useful. Because it really begs a few questions:

1. Useful as defined by what? Prediction, explanation, intervention?
2. Given a definition of usefulness, how do we assess and measure that usefulness? How do we discriminate between more and less useful models?
3. Relatedly, what properties are required for a model to be useful under these definitions of usefulness?

Read

• Bull, H. (1966) International Theory: The Case for a Classical Approach, World Politics 18 (3). DOI: https://doi.org/10.2307/2009761
• Naidu, S. (2020) In Praise of Blindspots, LPE Blog. URL: https://lpeproject.org/blog/in-praise-of-blindspots/ – Argument that simplified representations are necessary for any social science aiming to be useful for the needs of large administrative states operating in complex, heterogenous societies. The anti-James C Scott argument, in essence.

To Read

• Primo, D.M. and Clarke, K.A. (2012) A Model Discipline: Political Science and the Logic of Representations. Publisher: Oxford University Press
• Scott, J.C. (2020) Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed, Veritas Paperback Edition. Publisher: Yale University Press

Footnotes