Good Explanations Are Hard to Vary

Definition

A good explanation is one whose details all play a functional, load-bearing role in accounting for what it explains — so that changing any of those details would break the explanation. A bad explanation is easy to vary: its specifics could be swapped out for different, equally arbitrary specifics without changing what it predicts. Deutsch argues that this quality, not mere testability, is what separates science and rational thought generally from myth, superstition, and rationalization.

In the Book

Deutsch's central example (Chapter 1) contrasts the true explanation of the seasons — the tilt of Earth's rotational axis relative to its orbital plane — with the ancient Greek myth that winter is caused by Demeter's grief at Persephone's annual visits to Hades. Both are testable: the axis-tilt theory predicts seasons will be out of phase in the two hemispheres, and so, in principle, could the myth. But the myth is easy to vary — the marriage contract, the magic seed, the specific gods involved could all be replaced by different arbitrary details (as in the rival Norse myth of Freyr) without weakening its "explanatory" power, because none of those details is actually tied to why winter happens. The axis-tilt theory cannot be modified this way: "no easily implemented change could make tilted axes cause the same seasons all over the planet." Deutsch traces this insight to Karl Popper's critique of inductivism and empiricism — scientific theories are not derived from observation but are bold conjectures, and it is the requirement that they be good (hard-to-vary) explanations, not mere testability, that lets us reject the overwhelming majority of false theories (myths, rules of thumb, a gambler's "lucky feeling") without ever needing to test them.

Why It Matters

This gives a domain-general diagnostic for separating real understanding from rationalization, in any field: ask not just "does this theory predict correctly?" but "could its specific details be swapped for different ones without changing the prediction?" If yes, you don't have an explanation — you have a story that happens to fit. This applies as much to a business post-mortem, a political narrative, or a personal excuse as to a scientific theory: the hallmark of genuine insight is that its parts are locked together by necessity, not glued on for the occasion.