Fundamentally Unidentified Questions
[T]here's another powerful obstacle to randomised trial approach. This is the existence of 'fundamentally unidentified questions', or as the econometrician Josh Angrist indelicately puts it, "questions that are completely FUQed". A FUQed question is one that cannot be answered by an experiment -- for instance, the effect of carbon dioxide emissions on the world's climate. We can measure and calculate, extrapolate from our existing knowledge, but one thing we can't do is run a controlled experiment. We won't know exactly what our carbon dioxide emissions will do to the climate until they've already done it; even then we won't know for sure whether a different course of action would have had a different effect.
-- Tim Harford
from "Adapt: Why Success Always Starts with Failure"
Quoted on Sun Sep 18th, 2011