Introduction
In classical social choice theory, the focus is typically on direct voting systems: voters express their preferences, a voting rule is applied, and a winner is declared. However, real-world political landscapes, particularly in the United States, often employ a multi-stage process. Candidates are filtered through a primary system where party members select a nominee before the general election. This paper provides the first quantitative comparison between direct and primary voting systems using a spatial model of social choice to determine whether this multi-stage process helps or hurts social utility.
Methodology
The authors evaluate voting systems using the concept of distortion. In a spatial model, voters and candidates are represented as points in a multi-dimensional metric space. A voter's preference is determined by their distance to a candidate; the "social cost" of a candidate is the sum of their distances to all voters. Distortion is defined as the ratio between the social cost of the elected candidate and the social cost of the "optimal" candidate (the one who minimizes total distance).
The study compares two main models:
- Direct System: All voters participate in a single election using a standard rule (e.g., Plurality, Borda count).
- Primary System: A two-stage process where voters in separate parties (Party A and Party B) first nominate a candidate from their own ranks. The general public then chooses between these two nominees via majority vote.
Theoretical Robustness
A key finding of the research is the danger of "small" primaries. If a party represents an arbitrarily small fraction of the population, distortion can become infinite, as a fringe group might nominate an extreme candidate who wins the general election.
Conversely, the authors prove a "safety net" for "large" primaries. If each party represents at least a constant fraction $\alpha$ of the population, the primary system is theoretically robust. Specifically, the distortion under a primary system is bounded by approximately $3/\alpha$ times the distortion of the direct system. This suggests that as long as parties are reasonably sized, the multi-stage system remains stable and comparable to direct voting.
Empirical Performance
To evaluate average-case performance, the authors conducted synthetic simulations with 1,000 voters and 20 candidates in $k$-dimensional Euclidean space. They compared direct voting against two primary configurations: "Primary-split" (ideologically separated parties) and "Primary-random" (random party affiliation).
The simulations revealed that primary systems often act as a beneficial filter. In crowded direct elections, mediocre candidates can win due to vote-splitting among superior candidates. The two-stage primary process can prune these sub-optimal outcomes. This improvement is most dramatic for non-Condorcet rules like Plurality, Borda, and STV. For instance, in 1D space ($k=1$), the primary system consistently yielded lower average distortion than the direct system for several rules.
The advantage of primaries is most pronounced when parties represent distinct ideological camps ("separability"). However, as the dimensionality of the political space ($k$) increases, the gap between systems narrows. In higher dimensions, voter groups become more homogeneous relative to the candidates, and distortions for all systems tend toward 1.
Conclusion and Future Work
While this study establishes a quantitative foundation for comparing primary and direct systems, it acknowledges several limitations. The current model is restricted to a two-party system and assumes truthful voting based on distance. Future research could explore the impact of strategic voting, the presence of third parties, and dynamic candidate positioning where candidates shift their stances between the primary and general elections. Despite these simplifications, the work demonstrates that primary systems are not only theoretically stable when parties are large but can often improve social utility by filtering out sub-optimal candidates.
Summary updated