Arrow's Impossibility Theorem

Arrow's Impossibility Theorem (also known as Arrow's General Possibility Theorem) is a landmark mathematical theorem in social choice theory and welfare economics formulated by Kenneth J. Arrow in his 1951 monograph Social Choice and Individual Values (for which he was awarded the 1972 Nobel Memorial Prize in Economic Sciences). The theorem demonstrates that when a society of at least two individuals (N ≥ 2) must rank three or more discrete alternatives (|X| ≥ 3), no ordinal social welfare function can convert individual preference rankings into a complete and transitive collective ranking while simultaneously satisfying four normative fairness axioms: Unrestricted Domain (U), Weak Pareto Efficiency (P), Independence of Irrelevant Alternatives (IIA), and Non-Dictatorship (ND).

Cyclical Majority Preferences

Prior to Arrow's axiomatic synthesis, the Marquis de Condorcet demonstrated in 1785 that pairwise majority voting over three alternatives X = {A, B, C} can generate an intransitive collective cycle even when every individual voter holds rational, transitive preferences. Consider three voters (1, 2, 3) with the following strict preference orderings:

• Voter 1: A ≻₁ B ≻₁ C
• Voter 2: B ≻₂ C ≻₂ A
• Voter 3: C ≻₃ A ≻₃ B

In a pairwise contest between A and B, Voters 1 and 3 prefer A over B (a 2–1 majority for A ≻ B). Between B and C, Voters 1 and 2 prefer B over C (a 2–1 majority for B ≻ C). Transitivity would require A ≻ C; yet in a pairwise contest between C and A, Voters 2 and 3 prefer C over A (a 2–1 majority for C ≻ A). Consequently, majority rule produces the collective cycle A ≻ B ≻ C ≻ A, leaving no Condorcet winner and making the final decision dependent on the agenda-setter's sequence of pairwise votes. Arrow asked whether any sophisticated ordinal voting system—such as rank-order point systems or sequential runoffs—could eliminate such cycles without introducing new pathologies.

Preference Relations and Social Welfare Functions

Let N = {1, 2, ..., n} (n ≥ 2) be a finite set of individuals and X (|X| ≥ 3) be a set of social alternatives. Each individual i ∈ N holds a weak preference relation R_i on X (with strict preference P_i and indifference I_i) that belongs to the set L(X) of complete and transitive weak orderings:

1. Completeness: For all x, y ∈ X, either x R_i y or y R_i x (or both).
2. Transitivity: For all x, y, z ∈ X, if x R_i y and y R_i z, then x R_i z.

A preference profile is an n-tuple (R₁, ..., R_n) ∈ L(X)ⁿ. An Arrovian Social Welfare Function (SWF) is an aggregation rule F: D ⊆ L(X)ⁿ → L(X) that maps any admissible profile of individual rankings into a single complete and transitive social preference ordering R = F(R₁, ..., R_n), with strict social preference denoted P.

The Four Arrovian Axioms

Arrow imposed four conditions on the social welfare function F:

1. Unrestricted Domain (U): The domain D of F includes all logically possible profiles of complete and transitive individual orderings (D = L(X)ⁿ). Every profile must yield a complete, transitive social ordering R.
2. Weak Pareto Principle (P) (Unanimity): aligned with Pareto efficiency, for any pair of alternatives x, y ∈ X, if every individual strictly prefers x over y (x P_i y for all i ∈ N), then society must strictly prefer x over y (x P y).
3. Independence of Irrelevant Alternatives (IIA): The social ranking between any two alternatives x and y depends solely on the individual rankings between x and y. Formally, for any two profiles (R₁, ..., R_n) and (R′₁, ..., R′_n) that agree on the restriction to {x, y} (meaning x R_i y ⇔ x R′_i y and y R_i x ⇔ y R′_i x for all i), the social rankings R = F(R₁, ..., R_n) and R′ = F(R′₁, ..., R′_n) must rank x and y identically.
4. Non-Dictatorship (ND): There exists no single individual d ∈ N (a dictator) whose strict preference automatically overrides all other n − 1 voters across every pair of alternatives and every profile (i.e., there is no d such that for all x, y ∈ X and all profiles, x P_d y ⇒ x P y).

Theorem (Arrow, 1951, 1963): For n ≥ 2 finite individuals and |X| ≥ 3 alternatives, no Social Welfare Function F satisfies U, P, IIA, and ND simultaneously. Equivalently, any SWF satisfying U, P, and IIA is dictatorial.

Modern proofs of Arrow's theorem (such as those by Amartya Sen, 1970, and John Geanakoplos, 2005) proceed in three logical steps:

1. Decisive Coalitions and Field Expansion: A coalition G ⊆ N is called decisive for x over y if whenever all members of G rank x ≻ y, society ranks x P y regardless of how voters outside G rank x and y. Using Unrestricted Domain, Weak Pareto, IIA, and social transitivity across a third alternative z, the Field Expansion Lemma proves that if a coalition G is decisive over any single pair (x, y), then G is globally decisive over all ordered pairs of alternatives in X.
2. Group Contraction (Ultrafilter Property): If a decisive coalition G with |G| ≥ 2 is partitioned into two disjoint subsets G₁ and G₂, constructing a Condorcet-style profile across G₁, G₂, and N \ G over three alternatives {x, y, z} forces either G₁ or G₂ to be decisive on its own, or else social transitivity fails.
3. Isolation of the Dictator: Because the grand coalition N is decisive by Weak Pareto (P) and N is finite, repeatedly partitioning decisive coalitions contracts the decisive set down to a singleton coalition {d} of size 1—establishing that voter d is a dictator, contradicting ND.

Alternatively, Geanakoplos (2005) proves the theorem by placing one alternative b strictly at either the very top or the very bottom of every voter's ranking. By IIA and P, society must also place b at the extreme top or bottom. Moving b from the bottom to the top of each voter's ballot one voter at a time isolates a unique pivotal voter n^* whose change flips b from the very bottom to the very top of the social ordering; IIA then compels n^* to be a dictator over all remaining pairs (α, γ).

Domain Restriction: Black's Median Voter Theorem

Escape routes from Arrow's impossibility require relaxing at least one axiom:

1. Relaxing Unrestricted Domain (Single-Peaked Preferences): Duncan Black (1948) proved that if alternatives can be arranged along a single one-dimensional ideological or fiscal spectrum (such as a tax rate t ∈ [0, 1]) and every voter holds single-peaked preferences (utility declines monotonically as policy moves away from the voter's ideal point t_i^* in either direction), Condorcet cycles vanish completely. Under single-peakedness, pairwise majority rule produces a transitive social ordering whose top choice is the ideal point of the median voter.
2. Relaxing IIA and Embracing Cardinal / Approval Scales: Positional rules like the Borda Count (assigning m − 1 points to a first choice, m − 2 to a second choice, down to 0) satisfy U, P, and ND by sacrificing IIA. Similarly, Approval Voting and Majority Judgment (Balinski and Laraki, 2007) replace purely ordinal rankings with cardinal evaluative grades, bypassing Arrow's ordinal framework.
3. Relaxing Fininteness or Introducing Interpersonal Utility Comparisons: Amartya Sen (1970) showed that welfare economics can construct consistent social evaluations when ordinal rankings are enriched with cardinal, interpersonally comparable utility information—allowing society to evaluate utilitarian sums ∑ u_i(x), Rawlsian maximin min_i u_i(x), or inequality metrics such as the Gini coefficient.

In game theory and mechanism design, Arrow's theorem connects directly to the Gibbard–Satterthwaite Theorem (1973, 1975): because IIA prevents a voting rule from using preference intensities, any non-dictatorial ordinal voting rule over three or more outcomes is susceptible to tactical voting (violating dominant-strategy incentive compatibility).

Why does the Borda Count violate Independence of Irrelevant Alternatives (IIA)?

Suppose three voters rank four candidates {A, B, C, D} where C and D are ideological clones of B. Under the Borda Count (3, 2, 1, 0 points), the presence of clones C and D inflates the point gap between B and A even if no voter changes their relative preference between A and B. Thus, adding or removing an "irrelevant" third option C can flip the societal winner from A to B, violating IIA.

Does Arrow's Theorem mean democracy is impossible?

No. Arrow's theorem proves that no single ordinal aggregation procedure can satisfy all four idealized axioms across every conceivable preference profile simultaneously. In practice, democratic institutions operate effectively because empirical political preferences often exhibit strong single-peaked structure (avoiding Condorcet cycles) and constitutional rules trade off strict IIA for non-dictatorial representation.

How does Sen's "Liberal Paradox" differ from Arrow's Theorem?

In The Impossibility of a Paretian Liberal (1970), Amartya Sen dropped IIA entirely and replaced ND with a "Minimal Liberalism" condition (granting at least two individuals decisive personal autonomy over a purely private choice, such as whether to read a book). Sen proved that Minimal Liberalism conflicts directly with the Weak Pareto Principle over an unrestricted domain, even without requiring full social transitivity.

1. Arrow, K. J. (1950). A Difficulty in the Concept of Social Welfare. Journal of Political Economy, 58(4), 328–346. https://doi.org/10.1086/256963
2. Arrow, K. J. (1963). Social Choice and Individual Values (2nd ed.). Yale University Press. https://cowles.yale.edu/publications/cfp/monograph-12
3. Black, D. (1948). On the Rationale of Group Decision-making. Journal of Political Economy, 56(1), 23–34. https://doi.org/10.1086/256633
4. Gibbard, A. (1973). Manipulation of Voting Schemes: A General Result. Econometrica, 41(4), 587–601. https://doi.org/10.2307/1914083
5. Geanakoplos, J. (2005). Three Brief Proofs of Arrow's Impossibility Theorem. Economic Theory, 26(1), 211–215. https://doi.org/10.1007/s00199-004-0556-7