Two or more candidates sit on a one-dimensional policy line among a distribution of voters. Each candidate repeatedly repositions toward the median of the voters currently closest to them — watch them converge toward the centre of the electorate.
The Hotelling-Downs model of spatial political competition: rational vote-maximising candidates on a single policy dimension gravitate toward the median voter's position, because any candidate positioned off-median can always gain votes by moving closer to the centre.
Adjust the number of candidates and watch how convergence changes — two candidates converge cleanly, more candidates create complex clustering. Skew the voter distribution to see the median shift, and use Shuffle Voters or Reset Candidates to restart the dynamics.
The theorem, formalised by Duncan Black in 1948 and popularised by Anthony Downs in 1957, is one of the most cited results in political economy — it explains why competing parties in two-party systems often sound strikingly similar.
The Median Voter Theorem is a foundational result in spatial voting theory that explains why competing political candidates tend to converge on strikingly similar platforms. In the Hotelling-Downs model simulated here, voters are distributed along a single policy dimension, and each candidate is assumed to maximise their vote share by adopting the position that appeals to the largest number of voters closest to them. Under fairly general conditions — a single-peaked policy space and simple majority rule — the unique stable equilibrium is for all candidates to locate at the position preferred by the median voter, the voter positioned exactly in the middle of the distribution.
The theorem originates in economist Harold Hotelling's 1929 paper on spatial competition between firms and was formally proven for voting contexts by Duncan Black in 1948, then popularised in Anthony Downs's influential 1957 book "An Economic Theory of Democracy." It remains a cornerstone of political economy, widely used to explain the centrist convergence observed in many two-party democracies, though real-world elections often deviate from its predictions due to multiple policy dimensions, primary elections, and candidates who value ideology over pure vote-maximisation.
The Median Voter Theorem states that in a majority-rule election with a single policy dimension and single-peaked voter preferences, the position preferred by the median voter will always defeat any other position in a pairwise vote. As a result, vote-maximising candidates have a strong incentive to converge toward the median voter's ideal point rather than take extreme positions.
The simulation starts with two candidates on a policy line among a distribution of voters shown as a histogram. Each candidate automatically repositions toward the median of the voters currently closest to them. Adjust the number of candidates and distribution skew, then use Shuffle Voters to regenerate the electorate and watch the dynamics restart.
With exactly two candidates, both converge cleanly to the median because any deviation loses votes to the opponent. With three or more candidates, no single stable equilibrium generally exists — candidates can cluster into groups, oscillate, or spread out depending on how the population divides among them, which is why real multi-party systems show more policy diversity than two-party systems.
The mathematical foundation was laid by economist Harold Hotelling in his 1929 paper "Stability in Competition," which analysed why competing shops (or ice cream vendors on a beach) tend to locate next to each other at the market's centre. Duncan Black extended this reasoning formally to committee and majority voting in his 1948 paper "On the Rationale of Group Decision-making," proving that the median position is the unique Condorcet winner under single-peaked preferences.
Anthony Downs applied Hotelling and Black's spatial competition logic directly to electoral politics in his 1957 book "An Economic Theory of Democracy." He modelled voters and parties as points on a left-right ideological spectrum and argued that both major parties in a two-party system are pulled toward the centre to capture the median voter, a prediction now commonly called the Hotelling-Downs model or Downsian convergence.
Several real-world factors weaken the theorem's predictions: primary elections force candidates to first appeal to more extreme party bases before a general election; voters care about multiple policy dimensions simultaneously, not just one; candidates may value policy outcomes intrinsically rather than purely maximising votes; and abstention by disillusioned centrist voters can make chasing the median counterproductive. These frictions explain why observed polarisation is often higher than the simple model predicts.
A skewed distribution shifts the median away from the mean, and rational candidates still chase the median position, not the average preference. A bimodal (two-peaked) distribution — representing a polarised electorate — still has a well-defined median, and the theorem predicts convergence there, even though few voters may actually hold that exact position, which is one criticism of applying the model to deeply polarised societies.
Yes. Hotelling's original 1929 paper was about firms choosing shop locations or product characteristics, not elections. The same convergence logic applies to product positioning (why cereal brands cluster around similar sweetness levels), retail store placement, and even broadcast television scheduling, wherever competitors chase the "median consumer" along a single dimension of choice.
Under single-peaked preferences on one dimension, the median voter's ideal point is always a Condorcet winner — meaning it beats every other position in a head-to-head majority vote. This is precisely why it becomes the stable equilibrium: any candidate who deviates from the median can be defeated by a rival who moves toward it, making the median position uniquely resistant to being outcompeted.
Modern political economy research extends the basic model to multiple policy dimensions (where stable equilibria often fail to exist, per the McKelvey-Schofield chaos theorems), probabilistic voting models where uncertainty about voter behaviour restores equilibrium, valence competition where candidates differ in perceived competence as well as position, and computational agent-based models — like this simulation — that relax the assumption of perfect rationality to study how candidates might learn to approximate the median through repeated interaction.