Social Science Statistics Blog

Has anyone ever bothered to imagine what manipulation to make people smoke is contemplated?

Sure. You can give away cigarettes for free? You can feature glamorous actors smoking in movies? You can allow smoking in places, like military boot camp, where it is current forbidden. In the extreme, you could force prisoners to smoke by withholding food until they do. There are many other possible examples.

The point of NO CAUSATION WITHOUT MANIPULATION is that a lot of silly people pretend that things like race and sex (essentially impossible to change) are plausible "causes" of this or that. They aren't. Thinking about things you can't, even in theory, change as causes is unhelpful.

Perhaps I am missing your point.

Actually, I don't think you're missing the point at all, just disagreeing with it, which is certainly fair enough. My question is now this: when people say "smoking causes lung cancer," do they really mean that giving cigarettes away causes lung cancer, or advertising causes lung cancer? I don't think so. Rather, I think those making this statement are assuming (in this instance, fairly plausibly) that it doesn't matter what induces people to smoke. Thus, there is an assumption that the outcome (lung cancer) is invariant to the nature of the intervention, as long as that intervention "works" in the sense of getting people to smoke. If those making this assumption are right, then I think Professor Holland's maxim is too broad.

By the way, because I'm on the blog panel, I know that there's a lot more forthcoming on this point. And you've anticipated where some of this is going: the "holy grail" of causation as relates to race/gender/national origin, etc. As a preview: in most, perhaps all, legal settings, what matters is what people perceive a person's race to be, not what it actually is (even if we could all agree on what that meant). And perceptions can be changed.