The Evidence Fits. So Does the Opposite Theory. Now What?
By the BrainSnail editorial team. How these articles are written and checked, and how to tell us when one is wrong.
For any set of observations there are always several incompatible theories that account for them equally well, which means evidence alone cannot settle which one to believe.
The claim being made
Evidence constrains theory without determining it. Given any body of observations, more than one theory can be constructed that fits all of them, and those theories may disagree completely about what the world contains. The point is not that scientists have failed to gather enough data, since the claim holds however much data is collected. It follows from the logical relationship between a general theory and the finite set of particular observations it accounts for, and it is not controversial in that limited form.
Why a failed prediction does not settle it
Testing a theory always tests more than the theory:
- •A prediction follows from the theory plus many other assumptions
- •Including how the instruments work and what conditions held
- •A failed prediction shows that something in that bundle is wrong
- •It does not show which part
- •So the theory can be kept by revising an assumption instead
- •Which scientists frequently and quite properly do
The historical examples
The pattern has produced real discoveries as well as real evasions. Irregularities in the orbit of Uranus contradicted Newtonian predictions, and rather than abandon the theory astronomers proposed an unseen planet, which was then found where predicted and vindicated the move. The same reasoning applied to Mercury produced a proposed planet that does not exist, and the anomaly was eventually explained by general relativity instead. Both responses were reasonable at the time and only the outcome distinguishes them, which is precisely the difficulty.
The strong version and its critics
A much bolder claim is sometimes made, that for any theory there exists a rival fitting all possible evidence, not merely all evidence gathered so far. Critics point out that the rivals usually produced to demonstrate this are constructed artificially, by taking a theory and adding an unobservable difference that changes nothing, and that such constructions are not serious alternatives anybody would propose. Whether a genuinely different rival always exists, rather than a contrived relabelling, has never been established. The weak version is uncontroversial and the strong one is doing most of the philosophical work.
What scientists actually do
If evidence cannot decide, something else must, and the candidates are the working standards of scientific practice. Simplicity is preferred, though what counts as simple is contested. A theory explaining several unrelated phenomena at once is preferred to one fitted to each separately. Fruitfulness matters, so a theory generating new predictions that succeed is preferred to one that only accommodates what is already known. Continuity with established results matters. None of these is a logical guarantee, which is why the strongest versions of the thesis remain genuinely troubling to some philosophers of science.
The takeaway
More than one incompatible theory always fits any finite set of observations, and a failed prediction indicts a whole bundle of assumptions rather than the theory alone, so a theory can always be saved by revising something else. Predicting Neptune was that move succeeding and the proposed planet inside Mercury's orbit was it failing. Simplicity, unification and fruitfulness do the deciding instead.