“Although we should not be satisfied with any theory that fails to explain a lot with a little, we need not of course expect any one theory to explain everything, or even the most important thing. Absolutely nothing in all of epistemology suggests that valid explanations should be monocausal. An explanation may be entirely valid, yet explain only a part (and even a small part) of the variation at issue. […]
Since no monocausal explanation is offered, one well-known test of validity is not applicable. It is often said in methodological discussions that every meaningful scientific theory must specify one or more possible events or observations, or experimental results, which would, if they occurred, refute the theory. This rule has no applicability to multicausal conceptions unless a perfect experiment is performed, or one so nearly perfect that we could be certain that it was the error in the theory rather than the flaw in the experiment that accounted for the result. In view of the limited possibilities for experiments in economics and other social sciences, the impossibility of controlled experiments on historical events, and the extreme improbability that nature or history will on its own provide anything resembling a perfect natural experiment, a search for a single decisive refutation is futile. I am told by some philosophers of science that even in the physical and natural sciences the rejection of theories usually occurs not because of a single negative experiment, but more often from a series of anomalous observations combined with the emergence of a better alternative theory. What we should demand of a theory or hypothesis, then, is that it be clear about what observations would increase the probability that it was false and what observations would tend to increase the probability that there was some truth in it.”
— Mancur Olson (1982) – writing before the profession’s focus on quasi-experimental empirical estimates – in “The Rise and Decline of Nations”.