Underdetermination

What it is

The idea is old and it belongs to philosophy of science rather than to psychology. Pierre Duhem set it out in 1906 and W. V. O. Quine generalised it: any body of evidence is compatible with more than one theory, because a theory never confronts the evidence alone but always together with a set of auxiliary assumptions.

The practical form is more useful than the philosophical one. If model A and model B both predict the result you just obtained, the result is not evidence for A. It is evidence against every model that predicted something else, which may be none of the ones anyone is arguing about.

The tell is when a dispute keeps producing new studies and never moves. That is usually a sign that the studies are all of a kind the two positions agree about.

In effect

The live example in this literature is the strongest surviving priming result.

Keith Payne, Jazmin Brown-Iannuzzi and Chris Loersch ran six experiments with 988 participants in a within-subjects gambling task, and found that primes shifted choices under uncertainty. That is real and it reproduces. On the automaticity account, which is the account that made social priming famous, the prime acts beneath awareness and moves behaviour directly.

David Shanks then showed the same pattern is reproduced by a model containing no distinction between automatic and controlled processing at all. On his account the prime raises the accessibility of certain information, and a fully deliberate judgment drawing on whatever is accessible shifts in exactly the observed way.

Both models fit. Neither is refuted. So the result cannot be claimed for the automaticity thesis, and it equally cannot be claimed against it. What follows is that the sensitive instrument has measured something without settling what it is, and that the interesting half of the original claim is untested rather than refuted.

Stating that symmetrically is the discipline the concept demands. Using underdetermination against a position you dislike, while treating your own as established, is the error it exists to name.

What it does not say

It is not scepticism. Underdetermination is a claim about a particular body of evidence in relation to particular rival models, not a general licence to doubt.

It does not mean the question is unanswerable. It means the answer requires a different design: one whose predictions the two models disagree about. Naming that design is the productive response, and it is usually available.

And it is not the same as a null result. A null says the effect did not appear. Underdetermination says the effect appeared and cannot tell you why.

---