Median Voter Theorem

Median Voter Theorem

Motivation: Escaping Arrow's Impossibility

Arrow's theorem tells us that no social choice function can simultaneously satisfy Completeness, Pareto, No Dictator, and IIA when the set of alternatives has at least three elements. One of the main exits from this impossibility is to restrict the domain of preference profiles we consider. Rather than allowing any rational preferences over $X$, we require that all voters have single-peaked preferences over a one-dimensional policy space. Under this restriction, majority rule becomes fully well-behaved — it satisfies completeness and transitivity — and a strikingly clean result emerges: the Median Voter Theorem.

Single-Peaked Preferences

We work with a one-dimensional policy space $X \subseteq \mathbb{R}$, typically the interval $[-1, 1]$.

Voter $i$ has single-peaked preferences if there exists a point $v_i \in X$, called her ideal point (or bliss point), such that moving away from $v_i$ in either direction makes the voter worse off. More precisely:

• If $y < x \leq v_i$, then $x \succ_i y$ (moving from $y$ toward $v_i$ is an improvement).
• If $v_i \leq x < y$, then $x \succ_i y$ (moving from $y$ toward $v_i$ is an improvement).

In utility terms, $u_i(v_i) \geq u_i(x)$ for all $x \in X$, with $u_i$ increasing on $(-\infty, v_i]$ and decreasing on $[v_i, +\infty)$ within $X$. The utility function has a single peak at $v_i$ and no local optima elsewhere.

Symmetric single-peaked preferences add the requirement that the voter is equally unhappy about deviations to the left and right: $x \sim_i y$ whenever $|x - v_i| = |y - v_i|$. Common utility specifications that give symmetric SP preferences are $u_i(x) = -|x - v_i|$ and $u_i(x) = -(x - v_i)^2$. Note carefully that symmetry is not required for the Median Voter Theorem — the theorem holds for any single-peaked preferences, symmetric or not. Symmetry is an additional assumption that simplifies the electoral competition models in later lectures.

Single-peaked preferences over a one-dimensional policy space are a natural and widely applicable restriction. They hold whenever voters have a "most preferred" level of some policy — a preferred tax rate, a preferred level of immigration, a preferred regulatory stringency — and are less happy the further policy moves from that level. They fail when preferences have multiple peaks, which can arise in multi-dimensional policy spaces or when the same dimension mixes several distinct issues.

The Median Voter

With $N$ voters, order their ideal points from left to right: $v_1 \leq v_2 \leq \cdots \leq v_N$. When $N$ is odd, the median voter is the voter with the middle ideal point, voter number $\frac{N+1}{2}$, with bliss point:

$v^M = v_{(N+1)/2}$

There are exactly $\frac{N-1}{2}$ voters with ideal points strictly to the left of $v^M$, and exactly $\frac{N-1}{2}$ voters with ideal points strictly to the right. The median voter $v^M$ is the "middle person" of the electorate.

When $N$ is even, the median is not unique: any point in the closed interval $[v_{N/2}, v_{N/2+1}]$ is a median in the sense of the theorem below.

Black's Median Voter Theorem

Theorem (Black 1948): Consider an electorate with $N$ odd voters having single-peaked preferences over $X \subseteq \mathbb{R}$, with ideal points $v_1 \leq \cdots \leq v_N$ and median bliss point $v^M = v_{(N+1)/2}$. Under majority rule, $v^M$ beats every other alternative $x \neq v^M$ in a pairwise vote.

In other words, $v^M \succ_S x$ for all $x \neq v^M$. The median voter's bliss point is the unique Condorcet winner — it beats everything else head-to-head. No alternative can beat $v^M$ in a majority vote.

Proof Sketch

The argument is straightforward and worth understanding fully. Fix any $x < v^M$ (the case $x > v^M$ is symmetric).

Consider any voter $j$ whose ideal point satisfies $v_j \geq v^M$. Since $v_j \geq v^M > x$, this voter's ideal point is strictly to the right of $x$. By the definition of single-peaked preferences, since $x < v^M \leq v_j$, the voter prefers $v^M$ to $x$: $v^M \succ_j x$.

How many such voters are there? Voters $\frac{N+1}{2}, \frac{N+1}{2}+1, \ldots, N$ all have ideal points $v_j \geq v^M$. That is $N - \frac{N+1}{2} + 1 = \frac{N+1}{2}$ voters — a strict majority of the $N$ voters (since $\frac{N+1}{2} > \frac{N}{2}$ when $N$ is odd).

Therefore a majority prefers $v^M$ to $x$, and by majority rule, $v^M \succ_S x$.

The symmetric argument handles $x > v^M$: any voter $j$ with $v_j \leq v^M$ prefers $v^M$ to $x$, and there are $\frac{N+1}{2}$ such voters — a majority.

When $N$ is even, no single median is unique. The argument generalizes to show that any point in $[v_{N/2}, v_{N/2+1}]$ beats any alternative outside that interval in a majority vote.

The Median Voter Is Not a Dictator

It is tempting to confuse the Median Voter Theorem with Arrow's dictatorship result, but they are quite different. A dictator is a fixed individual whose preferences always determine the social ranking, regardless of what everyone else prefers. The median voter is not a fixed individual — her identity depends on all other voters' ideal points through their relative rankings. If voter preferences shift, the identity of the median voter changes. The median voter's bliss point prevails because it commands a majority against every alternative, not because any individual has special authority.

Put differently: the median voter's bliss point wins majority votes because of the arithmetic of the majority rule combined with the single-peaked structure, not because of any privileged position in the decision process.

The Gap Between Median Voter and Median Citizen

The theorem identifies the median voter — the median among those who actually vote. This is not the same as the median citizen. Several factors drive a wedge between the two:

Turnout and eligibility: Not all citizens vote, and the eligibility and participation rates differ systematically across groups. If low-income citizens vote less frequently, the median voter will be systematically more affluent than the median citizen.

Disenfranchisement: Felony disenfranchisement laws in the United States prevent a substantial fraction of the adult population from voting, and this fraction is not randomly distributed — it is concentrated among low-income, minority, and young citizens.

Voter ID laws and registration requirements: Burdensome registration and ID requirements reduce participation among groups that tend to lack the relevant documents. Making voting harder shifts the effective median voter.

Differential mobilization: Political parties and campaigns devote resources to turning out their supporters. Differential mobilization efforts shift the composition of the electorate and hence the location of the median voter.

All of these factors mean that the effective median voter may be substantially different from the median citizen, and policies that are "optimal" for the median voter may not serve the median citizen's interests.

When the Median Voter Theorem Does Not Apply

The theorem requires single-peaked preferences over a one-dimensional policy space. This fails in several important settings:

Multiple policy dimensions: If voters care about both immigration policy and tax rates, and these cannot be collapsed into a single dimension, then even with individually rational preferences, majority rule can produce cycles (the Condorcet paradox returns in higher dimensions).

Non-single-peaked preferences: Some preference structures — such as bimodal preferences over a continuous variable, or preferences induced by multi-issue bundling — do not satisfy single-peakedness, and the theorem fails.

More than two candidates: The theorem concerns pairwise majority votes. With more than two candidates and a non-majority voting rule (such as plurality voting), the outcome may not correspond to $v^M$ even when preferences are single-peaked.

Summary of Assumptions

The Median Voter Theorem requires: a one-dimensional policy space $X \subseteq \mathbb{R}$; single-peaked preferences (symmetry not required); majority rule for pairwise comparisons; and an odd number of voters $N$ for a unique median. Under these conditions, the median voter's bliss point $v^M$ is the unique Condorcet winner.

This result provides the foundation for the electoral competition models in Lectures 7 through 9, where we ask: knowing that $v^M$ determines electoral outcomes, what platforms do candidates choose?