Jan Zebrowski

The theory of causal-mechanical explanation is the leading account of scientific explanation but attempts to apply it to economics are frustrated by a practical roadblock: economists build different incompatible models to study a given phenomenon. Multiplicity of models can be an asset when the phenomenon to be explained is a particular economic phenomenon but presents a problem for the project of explaining general economic phenomena. Since different incompatible models posit mutually exclusive causal chains or otherwise have mutually exclusive assumptions and idealisations baked in them, economists cannot integrate them to explain such phenomena. And since all models are false, they cannot discriminate between them on empirical grounds. Hence the existing responses, that is integrative pluralism and model selection, founder. This paper proposes an alternative. Analogising between economics and statistical mechanics, it suggests that general economic phenomena can be explained by representing certain difference-making global features shared by their particular instances.

Model-based science—the style of theoretical work dominant in many social sciences, including economics—studies complex real-world systems indirectly through highly idealized model systems. The viability of model-based science pivots on the possibility of determining ‘difference-makers’ for every system it studies. However, economic systems are neither clearly circumscribed nor ‘closed’ in the sense that any outcome studied by economists is, to a greater or lesser extent, causally influenced by an infinitely complex network of factors. This makes salient the question: How do economists determine which factors are explanatorily relevant to any given outcome and should be included in its explanation? This is the problem of explanatory relevance. In this paper, I try to make headway towards solving this problem using robustness analysis (RA)—a well-known procedure in theoretical economics by which modellers gauge the sensitivity of their models' results to assumptions that fuel the derivation of these results.