The difficulty of conducting relevant experiments has long been regarded as the central challenge to learning about the economy from data. The standard solution, going back to Haavelmo's famous "The Probability Approach in Econometrics" (1944), involved two elements: first, it placed substantial...