September 2006
Sun Mon Tue Wed Thu Fri Sat
          1 2
3 4 5 6 7 8 9
10 11 12 13 14 15 16
17 18 19 20 21 22 23
24 25 26 27 28 29 30

Authors' Committee


Matt Blackwell (Gov)


Martin Andersen (HealthPol)
Kevin Bartz (Stats)
Deirdre Bloome (Social Policy)
John Graves (HealthPol)
Rich Nielsen (Gov)
Maya Sen (Gov)
Gary King (Gov)

Weekly Research Workshop Sponsors

Alberto Abadie, Lee Fleming, Adam Glynn, Guido Imbens, Gary King, Arthur Spirling, Jamie Robins, Don Rubin, Chris Winship

Weekly Workshop Schedule

Recent Comments

Recent Entries



SMR Blog
Brad DeLong
Cognitive Daily
Complexity & Social Networks
Developing Intelligence
The Education Wonks
Empirical Legal Studies
Free Exchange
Health Care Economist
Junk Charts
Language Log
Law & Econ Prof Blog
Machine Learning (Theory)
Marginal Revolution
Mixing Memory
Mystery Pollster
New Economist
Political Arithmetik
Political Science Methods
Pure Pedantry
Science & Law Blog
Simon Jackman
Social Science++
Statistical modeling, causal inference, and social science



Powered by
Movable Type 4.24-en

« Mind the Coding | Main | Political Statistics Blogs »

28 September 2006

Causation and Manipulation

Jim Greiner

In a 1986 JASA article, Paul Holland reported that he and Don Rubin had once made up the motto, “NO CAUSATION WITHOUT MANIPULATION.” The idea is that even in an observational study, causal inference cannot proceed unless and until the quantitative analyst identifies an intervention that hypothetically could be implemented (although Professor Holland accepts the idea that the manipulation may be not ever be carried out for physical or ethical reasons). The idea of studying the causal effect of things that we as human beings could never influence is incoherent because such things could never be the subject of a randomized experiment.

My question: do we really adhere to this principle? Take the one causal link established via observational studies that pretty much everyone (even Professor Freedman, see below) agrees on: smoking causes lung cancer. Has anyone ever bothered to imagine what manipulation to make people smoke is contemplated? Aren’t we pretty sure it wouldn’t matter how we intervened, i.e., however it happens that people smoke, those who smoke get lung cancer at a higher rate? (It might matter what they smoke, how much they smoke, perhaps even where and when, but what got them started and what keeps them at it?) If folks agree with me on this, what’s left of Professor Holland’s maxim?

Paul W. Holland, Statistics and Causal Inference, 81 J. Am. Stat. Ass’n 945, 959 (1986)

David Freedman, From Association to Causation: Some Remarks on the History of Statistics, 14 Stat. Sci. 243, 253 (1999)

Posted by James Greiner at September 28, 2006 11:00 PM