Causation and Manipulation

Richard Scheines

Theories of causation have relied on logical relations, probabilistic relations, manipulation, or even counterfactuals to try and illuminate what it might mean to say that event A caused event B, or that variable X is a cause of variable Y. I argue first that, however we unpack the notion of causation, being able to predict a system's response to a manipulation is the only reason to bother. I then present a brief overview of the enormous progress made of late in statistically estimating a system's response to a manipulation from non-experimental data. In the remainder, I examine whether the standard conception of a manipulation is coherent. Surprisingly, from the perspective of my previous work, I conclude it isn't.