Copeland Method - Pairwise majority voting with net win-loss score

Overview

C_i ∈ [−(m−1), m−1]. A Condorcet winner (beats every other alternative) has C_i = m−1. Copeland is resistant to IIA violations unlike Borda, but can still produce cycles (Condorcet paradox).

Output

rank position, higher is better

Data

Crisp, complete rank

Size

2+ alternatives, 3-12 criteria works best

Used for

Rank aggregation, social choice, preference fusion

How it works

1. 1

   Collect K expert rankings to derive pairwise preferences.

   $$\mathcal{R}=\{R_1,\dots ,R_K\}$$

   Copeland 1951, Sec.1

2. 2

   Pairwise majority p_ik = #{k: R_k(A_i) < R_k(A_l)}.

   $$p_{il}=\sum _{k=1}^{K}1[R_k(A_i)<R_k(A_l)]$$

   Copeland 1951, Sec.2

3. 3

   Wins minus losses C_i = Σ_l 1[p_il > p_li] − Σ_l 1[p_il < p_li].

   $$C_i=\sum _{l\ne i}1[p_{il}>p_{li}]-\sum _{l\ne i}1[p_{il}<p_{li}]$$

   Copeland 1951, Sec.2

4. 4

   Descending ranking by C_i.

   $$\text{rank}(A_i)\propto -C_i$$

   Copeland 1951

Fits when / Look elsewhere when

Assumptions to verify

• Input is a rank matrix (1=best, m=worst per voter)
• Each voter ranks all alternatives

Edge cases and pitfalls

• Condorcet cycles (A_1 ≻ A_2 ≻ A_3 ≻ A_1) can produce equal Copeland scores for all alternatives in the cycle - no unique winner.

How to cite