The poll that moved the vote

Publishing a poll changes the thing the poll measures, and the direction of the change is not consistent.

Two mechanisms work against each other. A bandwagon effect draws support towards an apparent leader, because people prefer to be with the majority and because a leading candidate attracts attention, funding and coverage. An underdog effect draws support the other way, as supporters of a trailing side turn out in greater numbers and some voters who would have abstained decide the result is close enough to matter. Both are documented and neither dominates, which means the aggregate effect of publishing a poll depends on the race, the margin and how close the vote is.

There is a third mechanism that is better evidenced than either and attracts less comment: polls change behaviour through tactical voting. Where a system rewards concentration of support, a published poll tells supporters of a third candidate that their vote has no path, and some of them move. This effect is large enough in some systems to be the main way polls influence outcomes, and it is the one that a ban on publication would remove.

Several countries ban publication for a period before voting. The arguments against the ban are not primarily about free expression. They are practical: private polling continues and reaches campaigns, parties and donors, so a ban removes the information from voters while leaving it with everyone who is organising, which makes the asymmetry worse rather than better.

The modest reform that most statisticians favour is disclosure rather than prohibition: publish the question wording, the dates, the sample frame and the margin, and require that the margin appear wherever the headline figure does. It does not stop the effects. It lets a reader see what the number can carry.