Can you trust US polls?

Polls are judged against a claim they do not make

Question: Can you trust US polls?
Position1 of 2›: Yes, you can trust US polls
Argument1 of 2›: Polls are judged against a claim they do not make

The argument

Proponents of this position argue that most of the case against polling rests on a misreading of what a poll publishes. A poll is an estimate of a population quantity at the moment of asking, reported with a stated range of uncertainty. It is not a forecast of an outcome, it does not claim to describe how people will behave later, and it carries an explicit admission that the true figure may sit anywhere within its interval. Once that is taken seriously, proponents argue, much of the apparent failure disappears. When a poll reports two candidates separated by less than its own stated uncertainty, it is saying that the contest is too close for the instrument to call. If the other candidate then wins, the poll was not refuted; it described exactly the situation that occurred. The error belongs to the reader who converted a range into a prediction, and often to the commentary industry that treats a small lead as a settled fact because an uncertain story is harder to tell. Proponents add that the standard applied to polling is applied to nothing else. A medical test with a known error rate is not called untrustworthy when an individual result falls inside that rate, and an economic estimate revised as data arrives is not called a failure. Polls are unusual only in being published in advance of an event that then produces a single visible number to compare them against, which creates the impression of a verdict where there is none. On this account polls deserve the trust appropriate to any instrument that reports its own uncertainty honestly, and the thing that should not be trusted is the confident narration built on top of them.

Premises

[P1]A poll estimates a population quantity at the time of asking and publishes an explicit range of uncertainty around it. [P2] A poll is therefore not a forecast of a later outcome, and a gap smaller than its stated uncertainty is a statement that the contest cannot be called. [P3] An outcome falling within a published interval is consistent with the poll rather than a refutation of it. [P4] Other instruments that report known error rates, such as medical tests and revised economic estimates, are not treated as untrustworthy when individual results fall inside those rates. [C] Polls can therefore be trusted in the terms they actually claim, and what fails is the commentary that converts their ranges into predictions.

Counter-arguments

Critics answer that this defence makes polling unfalsifiable. If any outcome inside a wide interval counts as success, the instrument can never be shown to be wrong, and an estimate that cannot fail is not providing information. They also argue that polling organisations do not behave as though they were publishing mere snapshots: they promote their records, compete on accuracy against outcomes, and supply the inputs to forecasts that are presented as predictions. Having accepted that role, critics say, pollsters cannot retreat to the narrow technical claim only when the result goes against them.

Rejecting the premises

[Rejecting P2] Polling organisations do not behave as though they published mere snapshots; they promote their accuracy against outcomes and supply the inputs to published forecasts, so the narrow technical claim is one they rely on only when a result goes against them. [Rejecting P3] If any outcome inside a wide interval counts as consistent with the poll, the instrument cannot be shown to be wrong, and an estimate that cannot fail conveys correspondingly little. [Rejecting C] Blaming the commentary does not rescue the polls, since pollsters know how their numbers will be read and choose to publish headline figures rather than presenting the uncertainty as the finding.