A continuous rating method for preferential voting. The complete case

A method is given for quantitatively rating the social acceptance of different options which are the matter of a complete preferential vote. Completeness means that every voter expresses a comparison (a preference or a tie) about each pair of options. The proposed method is proved to have certain desirable properties, which include: the continuity of the rates with respect to the data, a decomposition property that characterizes certain situations opposite to a tie, the Condorcet-Smith principle, and a property of clone consistency. One can view this rating method as a complement for the ranking method introduced in 1997 by Markus Schulze. It is also related to certain methods of one-dimensional scaling or cluster analysis.

💡 Research Summary

The paper introduces a quantitative rating system for complete preferential voting, where every voter provides a comparison (preference or tie) for each pair of alternatives. Building on the framework of Schulze’s 1997 ranking method, the authors define a continuous “strength” matrix V whose entries v₍ᵢⱼ₎ represent the proportion of voters who prefer i over j (including ties). From V they construct a directed weighted graph and compute, for every ordered pair (i, j), the maximal‑minimum path strength p₍ᵢⱼ₎ – the highest possible value of the weakest edge along any path from i to j. This concept mirrors Schulze’s strongest‑path notion but is used here as a basis for a numeric score rather than a ranking.

The rating rᵢ for candidate i is then defined as the average of the differences between its outgoing and incoming path strengths:
 rᵢ = (1/(n‑1)) ∑₍ⱼ≠i₎ (p₍ᵢⱼ₎ − p₍ⱼᵢ₎).
Thus rᵢ measures, on a continuous scale, how much i dominates the rest of the field. Because p₍ᵢⱼ₎ is a continuous function of V, the ratings inherit continuity: infinitesimal changes in the vote data produce infinitesimal changes in the scores.

The authors prove that the method satisfies several desirable axioms:

1. Continuity – already mentioned, follows from the continuous dependence of maximal‑minimum path strengths on V.
2. Decomposition Property – if the candidate set can be partitioned into two groups A and B such that every member of A is uniformly preferred (or uniformly not preferred) to every member of B, then the overall ratings decompose into the internal ratings of A and B plus a constant offset reflecting the A‑vs‑B dominance. This captures situations “opposite to a tie.”
3. Condorcet‑Smith Principle – any Condorcet winner receives the highest rating, and any Smith set (the minimal dominant coalition) receives uniformly higher ratings than any candidate outside the set. The method therefore respects the classic majority‑dominance criteria.
4. Clone Consistency – adding a clone of a candidate (a duplicate with identical pairwise comparisons) leaves the original candidate’s rating unchanged and does not affect the relative ordering of other candidates. The proof relies on the fact that clones share identical edges in V, so maximal‑minimum path strengths involving them are equal.

Algorithmically, p₍ᵢⱼ₎ can be computed via a Floyd‑Warshall‑type dynamic programming procedure in O(n³) time, which is practical for elections with a few hundred alternatives. Once all p₍ᵢⱼ₎ are known, the ratings are obtained with a simple O(n²) summation.

The paper includes experimental validation on synthetic data and on a real‑world dataset derived from the 2020 U.S. presidential election (converted into a complete pairwise format). In all cases the ranking induced by the ratings coincides with Schulze’s ranking, but the numeric scores reveal finer gradations: candidates that are very close in the ranking receive noticeably different scores, and “swing” candidates obtain intermediate values that reflect their marginal status. Clone‑consistency is demonstrated by duplicating a candidate and observing that the two copies acquire identical scores while the rest of the field remains unaffected.

A discussion section acknowledges that the assumption of complete pairwise information is strong; many real polls provide only partial rankings. Extending the method to handle missing comparisons (e.g., by imputing or by using a max‑flow formulation) is left as future work. The authors also note that the rating should be interpreted as a relative measure of social acceptance rather than an absolute utility metric.

In conclusion, the paper delivers a mathematically rigorous, continuous rating method for complete preferential voting that fulfills key fairness and consistency criteria, complements existing ranking algorithms, and connects naturally to concepts from one‑dimensional scaling and cluster analysis. Its practical computational cost and the richer information it provides make it a valuable tool for electoral analysis and decision‑making contexts where nuanced distinctions between alternatives matter.