The median voter theorem states that under certain conditions, a majority-rule election between two candidates will be won by the position preferred by the median voter, and that both candidates therefore have an incentive to move toward the centre. It is elegant, it is provable, and its assumptions are strong enough that its application to real politics is heavily disputed.

Formalised by Duncan Black in 1948 and developed by Anthony Downs in 1957, the argument runs as follows. Arrange all voters along a single dimension, say left to right. Each votes for whichever candidate is closer to their own position. Any candidate positioned away from the median can be beaten by an opponent who moves between them and the median, capturing everything on that side plus the middle. The only stable position is the median itself, so both candidates converge there.

A ballot box. The theorem holds for two candidates under majority rule, and the convergence it predicts is a claim about strategic incentive rather than about conviction.
A ballot box. The theorem holds for two candidates under majority rule, and the convergence it predicts is a claim about strategic incentive rather than about conviction.Credit: Tohaomg (CC BY-SA 4.0).

The result requires several things that rarely all hold.

Preferences must be single-peaked along one dimension, meaning each voter has a favourite position and likes options less as they move away in either direction. Real politics has multiple dimensions: economic and social positions do not collapse onto one line, and once there are two dimensions no stable majority position need exist at all, a result related to Arrow's impossibility theorem.

There must be exactly two candidates. With three or more, convergence to the median is not an equilibrium, which connects the theorem directly to Duverger's law: the two-party systems where the theorem might apply are themselves produced by the electoral system.

Turnout must not depend on the candidates. If moving to the centre demoralises a party's base into staying home, the calculation changes entirely, and mobilising committed supporters may beat persuading moderates.

Candidates must be free to move, and must be believed when they do. Primary elections select candidates using a much narrower and more ideological electorate, and past positions constrain credibility.

A referendum ballot. On a single clear question the theorem's conditions come closer to holding than in a candidate election, where positions are bundled across many dimensions at once.
A referendum ballot. On a single clear question the theorem's conditions come closer to holding than in a candidate election, where positions are bundled across many dimensions at once.Credit: Electoral Tribunal, Panama (Public domain).

Empirical support is mixed and has weakened. There is evidence of centripetal pull in some systems and periods, particularly two-party systems with weak primaries. Against it stands persistent and increasing polarisation in several major democracies, in which parties have diverged rather than converged over decades, which the simple theorem does not predict.

Explanations offered for the divergence include: activists and donors sitting well away from the median and controlling selection and funding; media environments that reward intensity over breadth; the base-mobilisation strategy being more efficient than persuasion where the electorate is sorted; and multi-dimensional issue space allowing a party to be extreme on the dimensions its coalition cares about while remaining acceptable overall.

The theorem's continuing value is as a baseline rather than a prediction. It states precisely what would happen under specified conditions, which makes it a tool for identifying which condition has failed when something else happens instead, and that is a more useful thing for a model to do than to be right.