In classical social choice theory, we often analyze elections, resource allocations, and matching markets under a convenient fiction: the perfect information assumption. We assume that a strategic voter knows exactly how everyone else will vote, that a political candidate knows the precise preferences of the electorate, and that a malicious actor attempting to rig an election has a flawless view of every ballot. Under this "god's eye view," foundational results like the Gibbard-Satterthwaite theorem prove that almost all reasonable voting rules are vulnerable to strategic manipulation.
But what happens when we step into the real world? Real-world voters do not have access to a global preference profile; they look at noisy polling data or a simple scoreboard. Candidates operate under imperfect polling, and resource allocators face secretive participants. A growing body of research in computational social choice explores how relaxing the perfect information assumption fundamentally alters the stability, manipulability, and design of collective decision-making systems.
Information as a Shield Against Manipulation
Why does uncertainty matter? Intuitively, executing a successful strategic lie requires knowing exactly where the "pivot point" of an election lies. If you know the exact vote counts, you can calculate whether changing your vote from your favorite candidate to a lesser evil will tip the scales. But if you only have partial information, a strategic lie becomes a gamble: it might accidentally cause an even worse outcome.
Walsh demonstrates this phenomenon formally through the lens of dominating manipulations—strategies that guarantee a better or equal outcome regardless of the missing data. The author shows that while single-voter manipulation is computationally trivial under full information, hiding even a tiny fraction of the vote makes finding a dominating manipulation NP-hard for most common voting rules.
Similarly, Endriss et al. explore strategic voting when individuals only observe limited signals, such as the current leader or a summary of head-to-head matchups. They prove that restricting the information set available to voters naturally increases the cognitive and computational barriers to manipulation. When voters lack a complete model of others' motivations, as Aziz et al. argue, many voting rules cease to be obviously manipulable, protecting systems like the Borda count or Single Transferable Vote (STV) in large-scale settings. This structural barrier extends beyond voting: in matching markets, Gonczarowski et al. note that coordinating a successful group lie within the Gale-Shapley algorithm requires a level of global knowledge that real-world agents rarely possess.
Strategic Behavior Under Noisy Signals
When perfect information is stripped away, strategic actors must shift from deterministic optimization to probabilistic decision-making. Consider political candidates or project proposers who must tailor their strategies to an electorate they cannot perfectly read.
In representative democracies, candidates often engage in "pandering"—misrepresenting their true views to align with what they think voters want. Sun et al. model this behavior under imperfect polling and noisy data, demonstrating that the choice of democratic structure (such as flexible representation versus winner-take-all) dictates how resilient a system is to dishonest candidates when those candidates are forced to operate under uncertainty.
A parallel dynamic occurs in participatory budgeting (PB), where community groups propose projects and ask for funding. If proposers have perfect knowledge of public support, they can precisely inflate their project costs to capture maximum funding without crossing the threshold of rejection. However, as explored in Strategic Cost Selection in Participatory Budgeting (2025) and by Faliszewski et al., real-world turnout and preference uncertainty force proposers into a game-theoretic balancing act. Under uncertainty, the choice of the underlying voting rule—such as the Method of Equal Shares—becomes critical to stabilizing incentives and ensuring fair budget allocation.
Even external threats, such as bribery or election control, face similar friction. Hemaspaandra et al. point out that classical models assuming a briber has complete knowledge of all voter preferences fail to capture the real-world complexity of election interference, where actors must rely on probabilistic voting behavior and noisy intelligence.
Convergence and Rationality in Iterative Systems
If voters only have access to aggregate or partial information, how can an election ever reach a stable outcome? One might expect that a lack of information would lead to chaotic, unpredictable voting patterns. Surprisingly, the opposite is often true: limited information can act as a stabilizing force.
When voters participate in iterative voting—where they dynamically update their votes based on intermediate feedback—the type of feedback matters. Endriss et al. prove that iterative voting processes are guaranteed to converge to a stable winner even under highly restricted information sets, such as only knowing the current winner.
To understand why, consider how human voters reason. Instead of calculating complex expected utilities across infinite permutations, voters often use simple heuristics based on coarse likelihoods. Lev et al. formalize this by replacing precise probabilities with "pivot-graphs" that capture shades-of-gray uncertainty. They show that common heuristic behaviors (like abandoning a trailing third-party candidate to vote for a frontrunner) are mathematically rational responses to incomplete information that naturally guide the system toward stable equilibria. This aligns with the Iterative Calculus of Voting (ICV) framework introduced by Vasselai, which demonstrates that when agents simultaneously update their choices based purely on aggregate vote shares rather than full preference distributions, the system reliably converges to stable, two-candidate equilibria (Duverger's Law).
Designing for Uncertainty
Rather than viewing uncertainty as a flaw to be corrected, mechanism designers can leverage it as a tool or design around it to achieve better collective outcomes.
One approach is to introduce rule uncertainty. Holliday et al. show that if an election designer keeps the exact voting method uncertain—choosing from a set of reasonable rules only after ballots are cast—the incentive for voters to manipulate the election drops dramatically. The lack of information about the aggregation function makes tactical misreporting too risky to attempt.
Another approach is to design mechanisms that actively extract information to overcome individual biases. Schoenebeck et al. tackle a scenario where voters have imperfect, biased private signals about what outcome is actually best for the community. By designing a mechanism that asks voters to report both their personal preference and their prediction of how others will vote, they create a detail-free environment where truthful reporting forms a strong Bayes Nash Equilibrium, successfully aggregating the "wisdom of the crowd" without requiring common knowledge of the underlying population distribution.
Finally, when agents are intentionally uncooperative or "secretive" due to privacy concerns, the challenge shifts to bounding the damage caused by missing data. Ebadian et al. study resource allocation where some agents refuse to reveal their preferences. They show that by analyzing the worst-case "distortion" of social welfare, designers can still deploy simple allocation rules that guarantee optimal, fair outcomes despite severe information gaps.
Key Foundational Work
Complexity of Manipulating Elections with Few Candidates (2002) — contributed 2 concept elements to this direction.
The Function of Time in Cooperative Negotiations (1991) — contributed 2 concept elements to this direction.