On the probability ratio index as a measure of electoral competition

The paper addresses how to measure the degree of electoral competitiveness using only parties’ vote shares, with an emphasis on enabling comparisons across elections that may have different numbers of parties. The authors note that electoral competition is central to democratic representation and has been linked to outcomes such as turnout, governance quality, and economic performance, though evidence is mixed. While many determinants of competition are not fully captured by vote shares, vote-share data are universally available and facilitate consistent cross-context comparisons. The study seeks a measure bounded in [0,1], attaining 0 when one party has all votes and 1 when all parties share votes equally, that is smooth (twice differentiable), decomposable, and independent of the number of parties, to support meaningful cross-election comparisons. The proposed solution is the probability ratio index, a normalized probability that two randomly drawn voters voted for different parties.

The authors selectively review existing measures of electoral competitiveness and their limitations, especially for multi-party and cross-election comparability:

• Winning margin (difference in vote shares between the top two parties) and the top-candidate vote share: meaningful mainly for two-party contests; ignore other candidates and the overall distribution when multiple parties compete.
• Laakso–Taagepera (LT) “effective number of parties” (reciprocal of the Herfindahl–Hirschman index): its range depends on the number of parties [1, n], limiting comparability across elections with different n.
• Fractionalization index (Capron and Kruseman): depends on vote counts and total votes as well as party count, reducing suitability for cross-election comparisons.
• Entropy index: range depends on n ([0, ln n]), sharing the same comparability issue.
• Product index (Endersby et al.): collapses to zero if any party has zero votes, failing to distinguish different competitive situations with zero-vote parties. Given the prevalence of multi-party elections and variability in the number of candidates across time and jurisdictions, the literature underscores the need for a normalized, comparable, and distribution-sensitive measure, motivating the probability ratio index.

Framework and axioms:

• Let N be the set of competing parties with |N| = n ≥ 2. Vote-share vectors s = (s1, …, sn) lie in the simplex Δ(N): si ≥ 0 and Σsi = 1.
• A competitiveness function gN: Δ(N) → [0,1] satisfies: (1) gN(s) = 1 iff votes are equally shared (si = 1/n ∀i); (2) gN(s) = 0 iff one party has all votes; (3) gN is twice differentiable.

Probability ratio index (definition):

• Let P(s) = 1 − Σi si^2, the probability that two voters drawn at random with replacement voted for different parties (Gini–Simpson diversity).
• Under equal shares, P(sequal) = (n−1)/n. Define the probability ratio index as gN(s) = P(s) / [(n−1)/n] = (n/(n−1)) · (1 − Σi si^2).
• Properties: satisfies normalization (0 at monopoly; 1 at equality), smoothness, anonymity (invariant to party relabeling), and is independent of n due to normalization, enabling cross-election comparability.

Decomposability via rivalry function:

• Define B(x) = 4x(1−x). For n=2, g(s1,1−s1) = B(s1), a symmetric parabola peaking at 1 when s1 = 1/2.
• For general n, the index is decomposable: gN(s) = (1/n) Σi B(si). Thus, overall competitiveness is a normalized sum of pairwise “rivalry” contributions of each party against all others.

With vs. without replacement and link to fractionalization:

• Let vi be vote counts and V = Σvi; si = vi/V. The fractionalization index without replacement is F(s;V) = 1 − [1/(V(V−1))] Σi vi(vi−1) = (V/(V−1)) · P(s).
• Normalizing by the equal-shares case yields F(s) = [n/(n−1)] (1 − Σsi^2) = gN(s). Thus gN(s) equals the normalized probability that two voters drawn without replacement voted for different parties.

Relationships to standard indices:

• Herfindahl–Hirschman (HH): HH(s) = Σsi^2. Then gN(s) = [1 − HH(s)] / P(sequal). With the normalized HH, HH_norm(s) = (n/(n−1)) [HH(s) − 1/n], we have gN(s) = 1 − HH_norm(s). Hence gN is a negative monotone transform of HH.
• Coefficient of variation: Let CV_max = √(n−1). Then gN(s) = 1 − (CV(s)/CV_max)^2 and HH_norm(s) = (CV(s)/CV_max)^2 = 1 − gN(s).

Generalized probability ratios:

• Drawing k ≥ 2 voters with replacement and normalizing by the equal-shares case yields Q^{(k)}. For k > 2, Q^{(k)} can be zero even when multiple parties have positive shares (e.g., s1 = s2 = 1/2, others 0 gives Q^{(k)}=0 for k≥3), violating the competitiveness axioms. Therefore, only k=2 (the proposed index) yields a valid competitiveness map.

• Proposed a simple, normalized measure of electoral competitiveness: gN(s) = (n/(n−1))(1 − Σsi^2), interpretable as the ratio of the probability that two random voters voted for different parties under the observed shares to that under equal shares.
• The index is bounded in [0,1], smooth, anonymous, decomposable as gN(s) = (1/n) Σi 4si(1−si), and crucially comparable across elections with different numbers of parties due to normalization by (n−1)/n.
• Strong links to standard metrics: • Inverse monotonic relationship with the Herfindahl–Hirschman index: gN increases as concentration decreases; gN = 1 − HH_norm. • Direct equivalence to the normalized fractionalization index and to the normalized probability of inter-party draws without replacement. • Negative monotone relation to the squared coefficient of variation: gN = 1 − (CV/CV_max)^2.
• Illustrative examples show improved comparability and discriminatory power over alternatives: • Two-party equal shares (1/2,1/2): g = 1. Three-party (2/3,1/6,1/6): g = 3/4. LT and fractionalization (unnormalized) label these equally competitive; the proposed index correctly rates the two-party equal-share election as more competitive. • Three-party cases with a zero-vote entrant: g(1/2,1/2,0) = 3/4 versus g(1,0,0) = 0, while the product index gives 0 in both, failing to distinguish starkly different competitiveness. • Entropy can equate very different scenarios; for example, a two-party equal-share election and a three-party election with x=0.23 (x/2, x/2, 1−x) yield the same entropy, yet g distinguishes them: g=1 versus g≈0.57.

The study resolves the challenge of measuring electoral competitiveness in multi-party settings with variable party counts by proposing a normalized, probabilistic index that depends solely on vote shares, satisfies intuitive axioms, and enables cross-election comparisons. Its equivalence to a normalized inter-party draw probability provides a transparent behavioral interpretation. Decomposability offers diagnostic insight into how each party’s vote share contributes to overall competition. Negative monotonic relationships with concentration (Herfindahl–Hirschman) and dispersion (coefficient of variation) connect the index to well-understood measures, while alignment with a normalized fractionalization probability grounds it in existing political science practice. Examples demonstrate that the index avoids pitfalls of winning margins, entropy, product, and unnormalized fractionalization/LT measures, particularly in distinguishing scenarios with different numbers of parties or with zero-vote entrants. The index can also serve as a measure of political heterogeneity, informing assessments of party rivalry and voter fragmentation.

The paper introduces and analyzes the probability ratio index, a normalized probability-based measure of electoral competitiveness that is simple, interpretable, smooth, decomposable, and comparable across elections with different numbers of parties. It is inversely related to concentration (Herfindahl–Hirschman), equivalent to a normalized fractionalization probability, and inversely related to the squared coefficient of variation. Worked examples highlight its superiority to commonly used alternatives in cross-election comparability and discriminatory power. The authors note that the index is sensitive to the presence of small or zero-vote parties, which they view as a desirable feature capturing distance from the maximally competitive equal-share benchmark and signaling dominance of larger parties. Future research directions include analyzing the effects of party coalitions/mergers (modeled as vote-share aggregation) and the formation of blocking coalitions after thresholds are crossed.

• The index relies solely on vote shares. Many determinants of electoral competition (institutional rules, campaign dynamics, incumbency, candidate quality, local factors) are not captured; where such factors matter, the index may misrepresent underlying competitiveness.
• Sensitivity to zero- or low-vote entrants: adding a party with negligible or zero votes lowers the index (e.g., from 1 for two equal parties to 0.75 when a zero-vote third party is present). The authors argue this reflects genuine deviation from equal-share maximal competition, but users should be aware of this sensitivity.
• Generalized versions drawing more than two voters (k>2) violate the competitiveness axioms, restricting the approach to k=2 for a valid competitiveness map.
• As with any summary index, decomposability notwithstanding, it cannot reveal all structural features of the party system or strategic dynamics across constituencies or time without complementary analyses.