Skip to content

Does a better ballot end political polarization? — the median-voter convergence argument, and its limits

The biggest hope pinned on voting reform — loudest from the cardinal / "MaxVoting" camp (Approval, Score, STAR) — is that changing the ballot doesn't just pick better winners, it ends political division. The honest verdict: a better ballot removes the machinery that manufactures and rewards polarization, which is a real and important thing — but "ends division" overshoots the evidence. This page states the argument in its strongest form, concedes its genuine kernel, and marks exactly where it breaks. It's the fair-minded companion to Two-party dominance and leans on the spatial voting model for the geometry.

Level: 201 → 301 · for debaters

The argument (stated at its strongest)

The cardinal camp's case runs like this. A scored (cardinal) ballot evaluates each candidate independently — your rating of A can't be taken away by B entering the race. So there is no vote-splitting and no spoiler: a candidate can't be punished for the field around them, only rewarded for the support they actually earn. Given that, a candidate maximizes their support by moving their platform to where the most voters are — the consensus. And if every candidate must do that to win, they all converge on the same broadly-shared platform. A society whose Republican, Democrat, Libertarian, Green and independent candidates all represent essentially the consensus, the argument concludes, is no longer politically divided.

The vivid recent statement of this is Tedman Getschman's "How MaxVoting Ends Political Division" (2024), which pictures each candidate alone on their own public-opinion curve — no rival can carve votes away, so each is driven to the majority position. ("MaxVoting" is that author's rebrand of cardinal voting — see the glossary; STAR, Approval and Score are all "MaxVoting" methods.)

The valid kernel — concede it

Strip the branding and there is a real, important truth here, and it's the honest core of STAR's own pitch:

  • No vote-splitting / no spoiler. Cardinal ballots genuinely remove the structural penalty for a crowded field. A consensus candidate can't be "split out" the way Plurality splits one side.
  • The moderate survives instead of being squeezed. Whole-ballot methods don't eliminate the broadly-liked centrist early the way Hare/IRV does — see center squeeze. The candidate a majority would accept can actually win.
  • No lesser-evil trap. Voters can support several candidates, or score an honest favorite and a safe compromise, without betraying either. That defuses the "vote for the frontrunner you hate least" dynamic that hardens two camps.

Together these remove the incentives that reward running to the base and punish the compromiser. That is a legitimate, structural anti-polarization mechanism — and it's enough of a claim on its own. It doesn't need inflating.

Where "ends division" overshoots — the claim-check

The leap from "candidates face better incentives" to "division ends" rests, unstated, on the median voter theorem (Black 1948; Downs 1957) — the Hotelling "ice-cream vendors edge to the middle" result. Five things break the leap:

  1. It's a one-dimensional result. The median voter theorem guarantees a central Condorcet winner only on a single left–right line. In two or more dimensions the center dissolves: majority rule can cycle through the whole space (the McKelvey–Schofield chaos theorem), so there's no single "majority opinion" for candidates to converge onto. Real politics is multidimensional — see the spatial model's own honest limits.
  2. Candidates converging ≠ voters agreeing. Even where candidates do pile toward the center, the electorate stays as spread out as ever. Convergent platforms are not social consensus. The article itself says "it's about the candidates, not the voters" — then quietly lets the conclusion slide back to the polity. Division among people is what "polarization" means; identical candidates don't dissolve it.
  3. The convergence isn't unique to cardinal methods. In one dimension Ranked Robin / Condorcet methods elect the median by construction, and two-candidate competition converges there under almost any method. The centripetal pull is a property of the spatial model, not of "MaxVoting." Crediting a method-general geometric fact to one family is the classic criterion-built-to-fit-the-method move.
  4. More than two candidates + strategy → differentiation, not a pile-up. Downsian convergence is a two-candidate theorem; with a full field, spatial theory predicts candidates spreading out to claim unoccupied ground, not stacking on the median. And under pure Score the vote-maximizing position tracks the utilitarian optimum (which need not equal the median — it depends on the utility shape), while strategic min-maxing pushes the ballot toward Approval and shifts the equilibrium again.
  5. "Any and all MaxVoting methods" don't behave alike. Approval, Score and STAR differ materially in strategy and outcomes — see scoring methods vs. ranked voting and the strategic pathologies. Bundling them hides exactly the differences a voter would care about.

The honest STAR claim

Keep the defensible version, which is strong enough on its own: a better ballot changes the incentives. It stops rewarding the base-first, spoiler-driven, lesser-evil dynamics that drive polarization, and it lets broadly-acceptable candidates win. That is a real nudge toward the electorate's center of gravity — not a machine that deletes disagreement. Polarization is also downstream of primaries, media, geographic sorting and identity, which a ballot can't touch (that's the "what actually changes it" point). Overclaiming "ends division" hands critics a free target and misrepresents the mechanism; the true claim — removes the machinery that manufactures and rewards division — is both accurate and persuasive. This is the same discipline the library applies when it criticizes IRV: concede rarity and limits, don't cheerlead.

Claim-check — "How MaxVoting Ends Political Division" (Getschman, 2024)

  • Source & lean. A single-author advocacy Substack (Common Sense for Uniting America) promoting a self-coined brand ("MaxVoting" / the "Perkins Protocol"); cardinal-camp, no academic or external validation. Good for the framing image, weak for the empirical verdict — disclose the lean when citing it.
  • Concede. Independent per-candidate evaluation → no vote-splitting / no spoiler; the "each candidate alone on their own opinion curve" picture is a genuinely useful teaching image for why cardinal ballots don't split votes.
  • Overreach. "Ends political division" — via the five points above (1-D-only, candidates≠voters, not method-unique, differentiation with a full field, methods differ).
  • Verdict. Right mechanism, oversized promise. A better ballot reshapes the incentives that feed polarization; it does not end division.