Global Voting Strategy

• Global voting strategy is a framework that aggregates complete outcomes or entire electorates, avoiding local voting paradoxes and ensuring coherent decisions.
• The approach leverages hierarchical state assessments and local dominance heuristics to navigate uncertainty in strategic voting models.
• Cross-system prediction methods apply global voting strategies to map heterogeneous legislative practices into unified, high-accuracy models.

Searching arXiv for papers related to “global voting strategy” across strategic voting, uncertainty, and voting aggregation frameworks. In contemporary social choice and computational voting, “global voting strategy” denotes approaches that reason over complete outcomes, whole electorates, or whole uncertainty sets rather than isolated local choices. In combinatorial preference aggregation, global voting over CP-nets considers CP-nets as a whole during preference aggregation; for this reason, global voting is immune from paradoxes, and there is no need to impose restrictions over the CP-nets' topological structure (Lukasiewicz et al., 2018). In strategic voting models, the same global orientation appears when voters optimize against sets or hierarchies of possible states derived from polls, and in information-aggregation mechanisms that seek the alternative favored by the majority under the true state of the world (Lev et al., 2018, Schoenebeck et al., 2021).

1. Global aggregation versus sequential aggregation

A precise technical meaning of global voting strategy appears in preference aggregation over CP-nets and mCP-nets. Sequential voting aggregates preferences feature-by-feature. When preferences have specific feature dependencies, sequential voting may exhibit voting paradoxes, i.e., it might select sub-optimal outcomes. To avoid paradoxes in sequential voting, one has often assumed the $\mathcal{O}$-legality restriction, which imposes a shared topological order among all the CP-nets. Global voting takes the opposite route: CP-nets are considered as a whole during preference aggregation, so there is no need to impose $\mathcal{O}$-legality (Lukasiewicz et al., 2018).

For an mCP-net $M=\langle N_1,\dots,N_m\rangle$, global Pareto and majority voting are defined directly on complete outcomes. For outcomes $\alpha,\beta$, Pareto dominance is given by

$$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$

whereas majority dominance is given by

$$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$

An outcome is $s$-optimal if no other outcome $s$-dominates it, and $s$-optimum if it $s$-dominates every other outcome. This formulation makes global voting a comparison of whole outcomes rather than an agenda-dependent sequence of local choices (Lukasiewicz et al., 2018).

The computational analysis of this paradigm is itself part of the concept. A thorough complexity analysis of Pareto and majority global voting over not necessarily $\mathcal{O}$0-legal acyclic binary polynomially connected (m)CP-nets settles these problems in the polynomial hierarchy, and some of them in PTIME or LOGSPACE, whereas EXPTIME was the previously known upper bound for most of them (Lukasiewicz et al., 2018). A plausible implication is that global voting is not merely a normative alternative to sequential voting; it is also a distinct computational regime.

2. Strategic choice under polls, uncertainty, and bounded rationality

A second meaning of global voting strategy arises in models where voters respond to poll information under uncertainty. In the ordinal-dominance framework, the voting rule is score-based, a state is an aggregated score vector $\mathcal{O}$1, and the outcome function is

$\mathcal{O}$2

Each voter $\mathcal{O}$3 is endowed with an information structure

$\mathcal{O}$4

typically induced from a prospective poll by a distance metric and radii $\mathcal{O}$5. An action $\mathcal{O}$6 ordinally dominates $\mathcal{O}$7 if there exists a level $\mathcal{O}$8 such that $\mathcal{O}$9; an OD-equilibrium is a profile in which no voter has an ordinally dominating deviation. This framework yields a generic strategy: construct the hierarchy of possible states, compute undominated actions, and keep the current action only when no ordinally dominating deviation exists. It also justifies several heuristics, including local dominance, truth-bias and lazy-bias extensions, the improved $M=\langle N_1,\dots,N_m\rangle$0-pragmatist heuristic, and the Approval leader rule; Algorithm OD checks dominance in time $M=\langle N_1,\dots,N_m\rangle$1 (Lev et al., 2018).

The local-dominance theory of voting equilibria sharpens this picture for Plurality. Each voter considers a set of possible world states without assigning probabilities to them. Voting equilibria exist in the Plurality rule for a broad class of local dominance relations, and in an iterative setting local dominance-based dynamics quickly converge to an equilibrium if voters start from the truthful state. Extensive simulations show fast and robust convergence, equilibria consistent across various starting conditions, replication of Duverger's law, and a general improvement in the quality of the winner compared to truthful voting (Meir et al., 2014).

The broader uncertainty model for Plurality strengthens the convergence claim. In a nonatomic plurality game with distance-based strict uncertainty, any sequence of weak local-dominance moves is finite. Under equal uncertainty and a truthful initial state, the set of possible winners is shrinking, the score of the eventual winner is non-decreasing, and there are only compromise moves (Meir, 2014). This suggests that a global strategy can emerge from individually local heuristics when the uncertainty structure is sufficiently regular.

Behavioral evidence points in a related direction. The Attainability-Utility heuristic weighs the popularity of a candidate according to the poll, with the utility of the candidate to the voter, and it is able to predict peoples voting behavior significantly better than other models from the literature; it is almost at par with, and sometimes better than, a machine learning algorithm that uses substantially more information (Fairstein et al., 2019). In that sense, global voting strategy need not mean full expected-utility optimization; it can also denote a bounded-rational rule that jointly tracks utility and attainability.

3. Signal aggregation and informed-majority decisions

A third usage concerns mechanisms that aggregate private information rather than public polls. In the “wisdom of the crowd” framework, there are two alternatives $M=\langle N_1,\dots,N_m\rangle$2 and $M=\langle N_1,\dots,N_m\rangle$3, a latent state $M=\langle N_1,\dots,N_m\rangle$4, and each agent receives one private signal from $M=\langle N_1,\dots,N_m\rangle$5. Voters may be candidate-friendly, candidate-unfriendly, or contingent, and the social objective is the “majority wish”: the alternative that would be preferred by more than half of the agents if they knew the true state. The mechanism asks agents to report type, signal, and a prediction about how many others will report an effective high signal. It then computes the median prediction $M=\langle N_1,\dots,N_m\rangle$6, compares it to the realized fraction $M=\langle N_1,\dots,N_m\rangle$7 of effective $M=\langle N_1,\dots,N_m\rangle$8-signals, and chooses $M=\langle N_1,\dots,N_m\rangle$9 if $\alpha,\beta$0, otherwise $\alpha,\beta$1. If more than half report type $\alpha,\beta$2, it immediately chooses $\alpha,\beta$3; if more than half report type $\alpha,\beta$4, it immediately chooses $\alpha,\beta$5 (Schoenebeck et al., 2021).

The equilibrium claims are unusually strong. If all agents play the truthful strategy $\alpha,\beta$6, then, with probability at least

$\alpha,\beta$7

the mechanism outputs an alternative that is favored by more than half of the agents. The truthful strategy profile is an $\alpha,\beta$8-strong Bayes Nash equilibrium with

$\alpha,\beta$9

and for $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$0, the truthful strategy profile is a strong Bayes Nash equilibrium (Schoenebeck et al., 2021). In this literature, global voting strategy is therefore a mechanism for state inference and preference aggregation at once.

A closely related asymptotic claim appears in later work on strategic voting with state-dependent preferences. Strategic voting behaviors are reported to have a positive impact on leading to the “correct” decision, outperforming informative voting and sincere voting. As the size of the vote goes to infinity, every $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$1-strong Bayes Nash Equilibrium with $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$2 converging to $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$3 leads to the informed majority decision with probability converging to $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$4 (Han et al., 2023). This suggests that, in common-value environments, strategic behavior can itself serve as a global information-aggregation device.

4. Rule uncertainty, knowledge dynamics, and institutional design

Global voting strategy also includes system-level responses to uncertainty about the rule itself. One formalization studies a set $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$5 of possible voting methods and distinguishes three notions of manipulability. Sure manipulation requires a single deviation that is strictly better under every $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$6. Safe manipulation requires a deviation that is never worse under any $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$7 and strictly better for at least one. Expected manipulation compares the probability mass of rules improved by a deviation with the probability mass of rules harmed by it. The resulting theorems show that uncertainty about the voting method may reduce or even eliminate a voter's incentive to misrepresent her preferences. In particular, for any $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$8, $$\beta \succ^M_{\text{Pareto}} \alpha \iff |M^-(\alpha,\beta)| = m,$$9 and $$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$0 eliminate sure weak dominance manipulation for $$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$1, and so do $$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$2 and $$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$3 (Holliday et al., 2019).

An epistemic version of the same theme models knowledge explicitly. A profile model is

$$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$4

where $$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$5 is a set of states, each $$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$6 is an equivalence relation, and $$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$7 assigns a profile to each state. This makes it possible to distinguish can manipulate, considers possible that she can manipulate, de dicto knowledge of manipulation, de re knowledge of manipulation, and de re knowledge of weak manipulation. The framework also models public announcements $$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$8, which restrict the state space, and public assignments for declared votes, which change the declared profile itself. Knowledge of successful manipulation is preserved after update, whereas knowledge of weakly successful manipulation is not preserved after update. The same machinery is proposed as the epistemic background for Stackelberg games for uncertain profiles (Ditmarsch et al., 2013).

Institutional procedures can be analysed in parallel terms. Repeat voting proposes two identical rounds: every eligible voter is entitled and explicitly encouraged to vote in each round, the first-round votes are officially counted and published, the second round takes place roughly two weeks later, and the final result is determined by adding up all the votes in both rounds,

$$\beta \succ^M_{\text{maj}} \alpha \iff |M^-(\alpha,\beta)| > |M^+(\alpha,\beta)| + |M^0(\alpha,\beta)|.$$9

The proposal is argued to increase voter participation and result in more accurate and representative outcomes, while also raising concerns about higher cost and bandwagon effects (Hart, 2022). A plausible implication is that global voting strategy may be encoded not only in a decision rule but also in a timetable and an information-release protocol.

5. Cross-parliament forecasting as a data-driven global strategy

In legislative studies, the term takes a predictive form. The Voting Prediction Framework is a parliament-agnostic, open, three-layer pipeline consisting of Data Collection, Parsing and Feature Integration, and Prediction Models. For each country $s$0, it constructs $s$1 for personal information, $s$2 for bills, $s$3 for votes, $s$4 for protocols, and an enriched prediction dataset $s$5. The framework was evaluated on over 5 million voting records from Canada, Israel, Tunisia, the United Kingdom, and the USA, and achieves up to 85% precision in predicting individual votes and up to 84% accuracy in predicting overall bill outcomes (Mizrahi et al., 18 May 2025).

At the modeling level, the cross-country strategy relies on a unified schema and recurring feature families: party affiliation and coalition alignment, rank, member identity as a proxy for historical behavior, and bill content embeddings. Models are trained separately per country, and XGBoost performs best across all five countries. The reported country-level results are:

Country	XGBoost individual-vote accuracy	Bill-level accuracy
Canada	79.807%	82.24%
Israel	85.166%	83.23%
Tunisia	78.342%	82.56%
United Kingdom	80.308%	84.44%
United States	80.783%	82.78%

Time-series splits are used so that models forecast future bills rather than merely fit held-out records from the same period. Feature-importance analysis further shows cross-country heterogeneity within a common schema: Canada is dominated by party affiliation and procedural aspects; Israel by bill content and parliament number; the United Kingdom and the United States by member-specific features and embeddings; Tunisia by lower and more diffuse SHAP values (Mizrahi et al., 18 May 2025).

This use of the term is not a strategic voting model in the game-theoretic sense. It is a global prediction strategy in which heterogeneous legislatures are mapped into a unified representation and then forecast with the same model family. This suggests that “global” may refer either to whole-outcome aggregation or to cross-system generalization.

6. Limits, debates, and related implications

The literature does not support a single universal global voting strategy. Ordinal-dominance models explicitly avoid cardinal utilities and probability distributions; if precise probabilistic information is available, expected-utility strategies may diverge from OD-based ones. Their conclusions also depend on the chosen distance metric, radii, and hierarchy of possible states (Lev et al., 2018). In signal-aggregation models, strong guarantees rely on common knowledge of priors, signal structures, and type fractions, and there is an impossibility result showing that when the type distribution is not commonly known, no anonymous mechanism can be both approximately truthful and highly accurate (Schoenebeck et al., 2021).

Institutional proposals likewise involve trade-offs. Repeat voting may increase participation and improve robustness to one-shot shocks, but it increases administrative and campaign costs and may induce bandwagon effects (Hart, 2022). Uncertainty about the voting method may reduce or even eliminate incentives to misrepresent, but the strongest elimination results are rule-set specific and profile-size specific (Holliday et al., 2019). Legislative prediction frameworks achieve strong accuracy, yet their own authors emphasize data completeness, differences in roll-call availability, model bias, and ethical risks, including strategic abuse by lobbyists or political actors (Mizrahi et al., 18 May 2025).

A recurrent controversy concerns strategic voting itself. Several of these papers reject the simple view that strategic behavior is only distortive. Local-dominance simulations report that strategic voting generally improves the quality of the winner compared to truthful voting, and later work argues that strategic voting behaviors have a positive impact on leading to the “correct” decision (Meir et al., 2014, Han et al., 2023). This suggests that the central divide is not between sincere and strategic voting as such, but between local heuristics or institutions that fail to aggregate information and global strategies that do.

Taken together, these literatures present global voting strategy as a family of designs for whole-system decision making: aggregation over complete outcomes, over hierarchies of plausible states, over latent-state signals, over rule uncertainty, or over heterogeneous legislatures. What unifies them is not a single formula but a common ambition: to move from fragmented local reactions to decision procedures that are coherent at the scale of the full collective choice.