The situation of causality

Abstract

Causality in the abstract is a grand theme. We take it up when we want to penetrate to the bottom of things to understand general laws that govern the working at the world of the deepest and most detailed level.

In this essay, I argue for a more situated understanding of causality. To counter our desire for ever greater generality, I suggest that causal relations, even those that hold only on average, require context. To counter our desire for ever greater detail, I suggest that causal relations may exist only at a certain level of granularity.

The case for the situatedness of causality is based on the epistemic character of causality. A causal structure is a structure for predictions that might be made by an ideally knowledgeable and observant scientist. It tells us about the unfolding of that scientist’s knowledge. It has objective aspects, because knowledge must have objects. But it also has subjective aspects, because knowledge must also have a subject, and possibilities and probabilities are meaningless without reference to that subject.

The epistemic character and situatedness of causality tell us something about what kinds of objects should be called causes. The objects available to statisticians variables and Moivrean events are not causes because they are not situated. But they can be used to detect and aggregate causes.

The arguments advanced here build on my book, The Art of Causal Conjecture (Shafer 1996), which develops a detailed account of causal relations in probability trees.