The Mathematics of Voting Systems

Democracy runs on a simple-sounding idea: ask everyone, count the answers, follow the majority. The mathematics of social choice theory suggests it is substantially more complicated than that. In 1951, economist Kenneth Arrow proved that no ranked voting system with three or more candidates can simultaneously satisfy a small set of obvious, reasonable fairness conditions. This was not political commentary. It was a theorem, and it earned him the Nobel Prize in Economics in 1972.

This post works through the argument step by step: from the basic failure modes of plurality voting, through Condorcet's paradox, to the formal proof. Indian electoral data runs as the case study throughout. The conclusion is not that democracy is hopeless. It is that every voting system is an imperfect tool, and some are far worse than others.

§ 1 The Aggregation Problem

An election is, formally, an aggregation mechanism. It takes a profile of individual preference orderings over k candidates and maps them to a single social preference ordering. The branch of mathematics that studies this mapping is social choice theory, born in the French Revolution and formalised in the mid-twentieth century.

The challenge is not any single fairness condition. Each one individually seems obvious. The difficulty is that several of them cannot all hold simultaneously once the candidate count reaches three. That tension is what Arrow formalised, and it is why 200 years of proposed voting methods each kept breaking in a different place.

§ 2 First Past the Post

The simplest social choice function: each voter submits one name; the candidate with the most submissions wins. No majority required. Britain has used this rule since the fourteenth century; India inherited it at independence and applies it to all 543 Lok Sabha seats and most state assembly elections. Around 44 countries still use it today, the majority of them former British colonies.

The core flaw is that plurality is not majority. With k candidates, the theoretical minimum winning share falls as the field grows. In India's Lok Sabha contests, the effective number of serious candidates per constituency is typically three to six. Winners routinely clear 35 to 45 percent in safer seats and as little as 28 percent in fragmented ones.

The second flaw: FPTP has no mechanism to capture second preferences. If your preferred candidate is unlikely to win, voting for them may simply subtract from the candidate you prefer second. This is the strategic voting trap built into the system structurally, not incidentally.

§ 3 The Indian Evidence

India is the world's most important stress test of FPTP: the largest electorate on Earth, a genuine multi-party system, and elections contested across hundreds of constituencies simultaneously. The data is unambiguous on the core pathology: vote share and seat share come severely apart.

The 2014 result deserves particular attention. The previous record-low vote share for a single-party Lok Sabha majority was 40.78 percent, set by Congress in 1967. The BJP broke that record by nearly ten percentage points. Roughly 69 percent of voters chose someone else, yet one party controlled an outright majority of the lower house. The disproportionality is not a quirk of 2014; it is the structural output of the system across every cycle.

State-level data amplifies the picture further. In the 2017 Uttar Pradesh assembly election, the BJP won approximately 80 percent of the seats on 39.7 percent of the vote. In the 2012 Uttar Pradesh assembly election, a three-percentage-point gap between the Samajwadi Party (29.2%) and the BSP (25.9%) produced a 144-seat gap (224 versus 80). In the 2018 Madhya Pradesh assembly election, the Congress actually polled fewer votes than the BJP (40.9% versus 41.0%) and yet won more seats (114 versus 109). The system reversed the vote order entirely.

§ 4 The Spoiler Effect and Duverger's Law

When two ideologically similar candidates compete in the same race, they divide a like-minded electorate and can hand victory to a third candidate whom the majority would have defeated in any direct contest. This is the spoiler effect, and it is a structural feature of FPTP, not a coincidence.

In Uttar Pradesh 2017, the Samajwadi Party (21.8%) and the BSP (22.2%) together outpolled the BJP (39.7%). Their combined 44 percent should, intuitively, have produced more seats than a party with 39.7 percent. Instead the BJP won 81 percent of the seats. The corrective response was mathematical as much as political: for the 2019 Lok Sabha election, the two parties formed an explicit alliance to consolidate the divided vote. This is what game theory predicts FPTP will produce over time: not honest ideological expression, but strategic coordination designed to neutralise the vote-splitting the system itself creates.

Duverger's Law formalises this logic: FPTP incentivises voters to abandon smaller parties, concentrating support in two front-runners and eventually producing a two-party equilibrium. India is the world's most important partial exception. Duverger's Law operates constituency by constituency, not nationally. Each local contest tends toward two serious candidates, but the two candidates differ from state to state. A national party faces a different dominant regional party in each state, producing a multi-party parliament built out of locally bipolar races. The law holds locally; its national implication does not.

§ 5 Better Alternatives, Same Problems

If FPTP is so defective, what should replace it? Researchers have proposed many alternatives across 200 years of social choice theory. Each one solves some problems and introduces new ones.

Instant Runoff Voting (Ranked Choice). Voters rank candidates by preference. The last-placed candidate is eliminated in each round and their ballots redistributed to the next preference listed. The process continues until one candidate holds an outright majority. IRV reduces the spoiler effect and encourages positive campaigning (candidates want second preferences). But it suffers a serious mathematical defect called non-monotonicity.

The Borda Count. Proposed by Jean-Charles de Borda in 1784: with n candidates, first place earns n−1 points, second earns n−2, and so on. The flaw is explicit dependence on irrelevant alternatives: adding a no-hoper to the ballot changes the points each other candidate receives and can change who wins. Condorcet dismissed Borda's system in 1785 for exactly this reason, writing that it "relies on irrelevant factors for its judgments."

Condorcet's Method. The most intuitive idea: the winner should be whoever beats every other candidate in a direct pairwise comparison. You compute this from ranked ballots, no repeated elections required. The method is clean in theory and was actually discovered 450 years earlier by the monk Ramon Llull, whose work was lost until 2001.

Three rational individuals produce an irrational, intransitive group preference. The group simultaneously prefers X to Y, Y to Z, and Z to X. This is not a peculiarity of the dinner example. It is a structural feature of collective preference aggregation, and it is why Arrow's theorem is necessary.

§ 6 Arrow's Impossibility Theorem

Kenneth Arrow formalised this entire landscape in his 1951 PhD thesis. He identified five conditions that any rational ranked voting system should satisfy, then proved that no such system can satisfy all five simultaneously when there are three or more candidates. The proof was so fundamental that Arrow received the Nobel Prize in Economics in 1972.

The proof uses a construction called the pivotal voter argument, presented in its cleanest form by John Geanakoplos in 2005. It proceeds in three steps.

The pivotal voter is not a real person. It is a mathematical construct that the proof shows must exist as a logical consequence of the other four conditions. The conclusion is sharp: every ranked voting system must abandon at least one of the five conditions. FPTP quietly violates transitivity and IIA through its seat distortions. IRV violates monotonicity. Borda violates IIA explicitly. Condorcet's method fails to produce an output whenever preferences cycle. The failures are not bugs in specific designs. They are theorems.

§ 7 The Escape Routes

Arrow's theorem is a constraint on ordinal (ranked) systems. There are two partial exits, each relevant to India's own institutional design.

Black's Median Voter Theorem. Arrow assumed unrestricted domain: voters can hold any preferences. Duncan Black showed that relaxing this assumption partially rescues the situation. If all voter preferences are single-peaked along a single dimension (a left-right policy spectrum, say), majority rule is transitive, no Condorcet cycles arise, and the median voter's ideal point is always the majority winner. No dictator required.

The Indian caveat is significant. Indian political preferences do not sit on a single axis. They run simultaneously along caste and sub-caste, religion, region and language, class, and urban-rural divides. When preferences are multi-dimensional, single-peakedness fails, cycles reappear, and Black's optimistic result no longer holds. The theorem requires a world cleaner than Indian politics actually is.

Rated (Cardinal) Voting. Arrow's theorem governs ordinal systems, where voters rank candidates relative to each other. It says nothing about cardinal systems, where voters assign absolute scores. The simplest version is approval voting: each voter ticks every candidate they find acceptable, and the candidate with the highest approval percentage wins. More expressive is score voting, where voters rate candidates on a numerical scale.

Approval voting eliminates the spoiler effect entirely (you can approve of your true favourite and a front-runner simultaneously), discourages negative campaigning (attacking a rival does not gain you their approvals), and is straightforward to count. It is not a new idea: the Vatican used it to elect Popes from 1294 to 1621, and the United Nations uses it to select its Secretary-General today. Arrow himself, sceptical of rated systems for most of his career, came to regard approval voting as likely the best practical option toward the end of his life.

The honest caveat: the Gibbard-Satterthwaite theorem shows that all non-trivial voting systems, including rated ones, remain vulnerable to strategic voting. There is no free lunch. The choice is between different trade-offs, not between an imperfect system and a perfect one.

NOTA in India. The Supreme Court's 2013 ruling in PUCL v Union of India added a None of the Above (NOTA) option to Indian EVMs, allowing voters to formally reject all candidates on the ballot. It is, however, mathematically inert: even if NOTA outpolls every candidate, the leading candidate wins regardless. NOTA records dissatisfaction without changing outcomes. It is a pressure-release valve, not a structural reform.

§ 8 Full System Comparison

Across the five dimensions that matter most in practice:

FPTP scores badly on four of the five columns. The one column where it performs well (monotonicity) is the least consequential: a system that is monotonic but produces severely disproportionate outcomes, a structural spoiler effect, and eliminated parties despite significant vote shares is not a system that is working well by any reasonable definition.

Arrow did not prove that democracy is hopeless. He proved that the idealist's wish list for a voting system is self-contradictory: no ranked method can satisfy all five reasonable conditions at once. We must choose which imperfection to accept. First Past the Post, which India applies to its most powerful institution, happens to score worst on almost every metric in that table. Three consecutive Lok Sabha elections make the case without requiring any mathematics at all. The mathematics simply explains why fixing it is not straightforward, and why replacing it with a better-chosen alternative would be an improvement over a system that is provably, structurally broken.

That said, none of this is an argument against voting. If anything, it is an argument for caring more about how votes are translated into political power. I am 21 years old, and I vote in every election I am eligible to participate in because voting remains one of the most important responsibilities of citizenship :P