Chuck Manski has a fascinating forthcoming Journal of Economic Literature piece, "Communicating Uncertainty in Official Economic Statistics: An Appraisal Fifty Years After Morgenstern."
Manski's subtitle refers to Oskar Morgenstern's book, On the Accuracy of Economic Observations (1950, 2nd rev. ed. 1963), in which he assessed the accuracy of economic data and argued for the provision of error estimates in official statistics. (Morgenstern's book also famously reports that, after reading an early version, Norbert Wiener remarked that "economics is a one or two digit science.")
Significantly updating and amplifying Morgenstern's already-forceful case, Manski's treatment contains important new insights. He organizes his discussion not around the usual "sampling error" and "non-sampling error," but rather around what he calls "transient statistical error" (whether sampling or non-sampling), "permanent statistical error" (again, whether sampling or non-sampling), and conceptual error. I'll stop here; you can read it for yourself. Great stuff.