More on potential outcomes, causal inference, and virtual history

I talked with Don Rubin about the connections between the “Rubin model” for causal inference in statistics and Niall Ferguson’s conception of history. Don wrote:

I like the thoughts in the Ferguson book, although I’m not sure I’d call them an “extension” as much as an “application” of the basic ideas, in the same way that the “but for” concept in law is an application of these ideas in damages suits. And again, the point I keep making about the label “counterfactual” isn’t really right, because at the time of the decision, neither world is counterfactual, only the inference from data involves counterfactuals: e.g., how can I design a study to observe counterfactuals? — this makes no sense.

I replied,

I think ferguson is an “extension” not just an “application” because he emphasizes the need to see in the historical record what alternative actions were considered. This goes beyond the basic Rubin framework in which the set of possible interventions is a given.

Don replied that, yes, the consideration of realistic historical possibilities is not in the “Rubin model” as usually presented, but he has actually considered these issues in some of his recent work involving assessment of costs using economic modeling:

OK, I see your point about going to historical documents, which certainly wasn’t there in the Lord’s paradox example.

But that was done in the legal context of the tobacco litigation: by trying to figure out, first, which were the alleged specific acts of misconduct according to historical records, including legal standards of the time, and second, what would have been done in the absence of these alleged acts of misconduct.

For example, in the absence of the alleged misconduct, i.e., in the counterfactual world w/o the alleged misconduct, would the tobacco industry, in addition to not denying the health risks of smoking, have to vilify itself in its advertising, or teach job search skills to adolescents in high school, or hire a private police force to curtail the sale of tobacco products to underage youth in stores and gas stations, etc.? Then also what of the other factors such as underage people looking at the likes of evolved fake ids so the stores and gas stations are none the wiser to who they’re selling tobacco products to, and so on.

Much of the discussion and examination of what was done, what could have been done, and what would have created a world w/o misconduct appears to me very closely related to the historical approach described in Ferguson.

Maybe I’m too close to this and so “see” things that have never been written, but I think I wrote about this a little in my “tobacco” article in Gastwirth’s book. And I certainly have written it in various “expert reports” over the years. But maybe without the specific emphasis on going back to historical documents to see what could have happened, although I do say that: not doing something that was impossible in 1970 (e.g., not having web sites for people who wanted to quit, which were paid for by the tobacco industry) cannot be an act of misconduct.

Is Ferguson an extension or an application? I guess that’s in the eye of the beholder, and the closer you’re been to it over the years, arguing with “disbelievers” for over three decades in all sorts of specific examples and even in depositions, the more Ferguson seems like an application.

In summary

Ferguson’s ideas of contingency in historical analysis are closely related to modern ideas of causation in statistics and potentially have many applications in history, law, and related fields.