The Median Voter Theorem: Why Candidates Move to the Middle

The model, stated precisely

Put every voter and every policy on one line — say a left–right axis from 0 (most left) to 100 (most right). Each voter i has an ideal point x_i and single-peaked preferences: their happiness rises as a proposed policy p moves toward x_i and falls monotonically as it moves away, with exactly one peak. A convenient form is U_i(p) = −|p − x_i| — utility is the negative distance from your ideal. No second peak, no ties in the taste for extremes.

Claim. Under majority rule, the ideal point of the median voter, x_m, beats every other policy in a head-to-head vote. Why. Take any rival policy q > x_m. Every voter with an ideal point at or below x_m is closer to x_m than to q, so they prefer x_m. By definition of the median, those voters are at least half of the electorate. So x_m wins. The symmetric argument rules out any q < x_m. Hence x_m is a Condorcet winner — it defeats all comers pairwise — and it is the only such point.

The mechanism is pure counting, not psychology: the median splits the line so that no majority can be assembled on either side of it.

From a stable policy to candidates who converge

Black's theorem is about which policy wins. Downs (1957) added the strategic twist about candidates. Suppose two office-seekers each choose a platform to maximize votes, and each voter picks the nearer candidate. If Blue stands at 40 and Red at 60, the dividing line falls at their midpoint, 50; Blue captures everyone below 50, Red everyone above. Now let Blue nudge to 45. The midpoint slides to 52.5, and Blue grabs a slice of the middle it did not have before. Red feels the same pull from the other side.

The only place where neither candidate can gain by moving is when both stand on the median voter. There the electorate is split 50–50 and any deviation hands the mover a minority. That mutual best-response is a Nash equilibrium of the location game, and it explains the animation's punchline: two markers walking inward until they sit almost on top of each other at the middle. Hotelling had already shown the identical logic in 1929 for two ice-cream vendors on a beach, both drifting to the center to avoid ceding customers — which is why competitors' products, and party platforms, so often end up looking nearly alike.

A worked numerical example

Five voters must choose a defense-spending level (share of GDP). Their ideal points are 1%, 2%, 3%, 8%, 9%. The median voter is the third, with ideal 3% (two below, two above). Watch what happens in majority votes:

• 3% vs. 5%: voters at 1, 2, 3 all prefer 3% → 3% wins 3–2.
• 3% vs. 2%: voters at 3, 8, 9 prefer 3% → 3% wins 3–2.
• 3% vs. 8%: voters at 1, 2, 3 prefer 3% → 3% wins 3–2.

Every challenger loses to 3%, so 3% is the Condorcet winner — even though the mean ideal point is (1+2+3+8+9)÷5 = 4.6%. The two high-spending voters at 8% and 9% pull the average up, but they cannot move the outcome, because majority rule reads the median, not the mean. This is the theorem's sharpest quantitative lesson: intensity and outliers do not count under one-person-one-vote; position in the ordering does.

The assumptions — and where each one bites

The result is a theorem, so it is exactly as strong as its premises. Four do the heavy lifting:

1. One dimension. Everything must reduce to a single left–right line. Add a second orthogonal issue (say social vs. economic policy) and, by McKelvey's chaos theorem (1976), majority rule generically has no Condorcet winner: an agenda-setter can construct a sequence of pairwise votes leading from any point to any other. Stability collapses.
2. Single-peaked preferences. A voter who wants either a big program or none at all — but hates a half-measure most — has two peaks, and the median logic fails.
3. Full, sincere turnout. If moderates abstain and extremists always vote, the effective median shifts toward the base, giving candidates a reason to stay off-center.
4. Vote-maximizing, unconstrained candidates. Candidates must value winning above policy and be free to relocate. Primaries, activists, and donors anchor them to a partisan median that differs from the national one.

Relax any single assumption and convergence weakens; relax the first two and it can vanish entirely.

Does it hold up empirically?

Mixed — which is itself informative. In favor: U.S. general-election candidates reliably tack toward the center after winning polarized primaries, and cross-country studies find government spending tracks the preferences of the median-income voter more closely than those of the mean-income voter, consistent with the theorem's median-not-mean signature. Ballot initiatives, which pit two options head-to-head on a single issue, are its cleanest laboratory and often land near the median position.

Against: real party platforms in the U.S., U.K., and Europe have polarized, not converged, over the past several decades. Downs himself flagged the escape hatches — abstention by alienated moderates, and the credibility problem that a candidate who lurches to the center may not be believed. Formal models incorporating primaries (candidates must first win a partisan median), policy-motivated candidates (who trade some votes for a platform they like), and probabilistic voting all predict divergence, matching the data better. The theorem thus works best as a benchmark: it tells you the centripetal force is always present, and forces you to name the specific friction whenever candidates stay apart.

A subtle point people get wrong

The most common misreading is that the theorem describes the average voter or a "centrist consensus." It does neither. It is emphatically about the median — the middle of the ordering — and the two can diverge sharply whenever the distribution is skewed. In a country where a few very rich voters want low taxes and many modest earners want more, the median voter can favor substantial redistribution even though the average desired tax rate is dragged down by the wealthy. (This is the engine of the Meltzer–Richard model of the welfare state, 1981.)

A second subtlety: the theorem predicts convergence in positions, not that the median voter is happy. Both candidates offering the median's ideal point can still leave a large, intensely dissatisfied minority on each wing with no meaningful choice — a standing complaint about two-party systems. The theorem explains why they get no alternative; it does not claim the outcome is efficient or fair.

Median voter theorem vs. related decision rules and models

Feature	Median Voter Theorem	Hotelling Spatial Competition	Arrow / McKelvey (multi-dimensional)
Chooses	Median voter's ideal policy	Two firms both locate at market center	No stable winner — cycling
Setting	Voting, majority rule	Firms competing for spatial customers	Voting over 2+ issues at once
Key assumption	Single dimension, single-peaked prefs	Customers buy from nearest seller	Preferences over a plane
Equilibrium	Exists, unique (median)	Both converge to center	Generically none (chaos)
Prediction	Candidates converge	Products look alike	Agenda-setter can steer any outcome