Generalizing Condorcet’s Jury Theorem to Social Networks

Dan Braha, Marcus A.M. de Aguiar · 2025

Read Generalizing Condorcet’s Jury Theorem to Social Networks by Dan Braha, Marcus A.M. de Aguiar (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-16808.txt.

Introduction

Dan Braha and Marcus A. M. de Aguiar, writing as a pair, extend Condorcet's jury theorem to voters linked in a social network. The classic theorem says a majority of mildly informed voters beats a single voter. Here, free voters get private signals, some network members are fixed zealots, and at each step a voter may copy a random neighbor.

For finite networks the pair derives exact stationary distributions for vote counts, plus limits as the number of free voters grows. For majority rule they give an exact accuracy limit in closed form. For plurality rule they give sharp lower bounds. Under a stated condition, group accuracy strictly beats any single voter in both rules, so social copying amplifies weak private signals.

Catalog record

CREATORDan Braha, Marcus A.M. de Aguiar
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-16808

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Generalizing Condorcet’s Jury Theorem to Social Networks Dan Braha, Marcus A.M. de Aguiar Abstract We generalize Condorcet’s jury theorem (CJT) to socially connected populations in which agents revise discrete choices on a network in the presence of zealots. Free agents receive privately informative signals about the correct alternative and, at each update, either retain their state or imitate a uniformly chosen neighbor (free or zealot). For finite networks, we derive closed-form stationary laws for vote counts, and we characterize the corresponding vote-share limits as the number of free voters tends to infinity. For majority rule—both in binary and multi-alternative settings—we obtain an exact accuracy limit in closed form via the regularized incomplete beta function. For plurality rule, we establish sharp closed-form lower bounds on accuracy, expressed in terms of regularized incomplete beta functions. Under an absolute-majority condition for the correct alternative, both majority and plurality accuracies strictly exceed the accuracy of any single voter, showing that informative signals, coupled through social interaction, are amplified at the group level. These results extend CJT beyond independence and provide closed-form accuracy benchmarks for networked decision systems in social, biological, and engineered settings. Keywords: Condorcet’s jury theorem, voter model, beta distribution, Dirichlet-multinomial distribution Corresponding author. [email protected] 1 Introduction Condorcet’s jury theorem (CJT) offers a foundational statistical account of collective decision-making: if each voter independently has a better-than-random chance of being correct, then a simple majority is more likely to select the correct option than any individual, with this advantage growing with group size and approaching certainty as the group becomes large. First articulated by Condorcet in 1785 [ 1 ] , the result was revived by Black [ 2 ] , independently rederived by Kazmann [ 3 ] , and named the ‘Condorcet jury theorem’ by Grofman [ 4 ] . It is now a cornerstone of social choice theory [ 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 ] , informing debates in philosophy (knowledge aggregation), political science (justifications for democracy and broad franchise), law (jury size and composition), and organizational decision-making across economics, management, marketing, and finance [ 5 ] . Classical CJT relies on stringent assumptions: a binary choice with an objectively correct answer, common interests, homogeneous voter competence p>0.5 , independent private signals, and sincere (non-strategic) voting under simple majority. Relaxations of these assumptions reveal where information aggregation can fail. Under independence with heterogeneous competences, the non-asymptotic claim need not hold uniformly; nevertheless, majority accuracy typically exceeds the average voter competence and can, in some cases, exceed that of the best member, while the asymptotic claim holds under mild conditions on the competence distribution [ 6 , 9 , 11 , 12 , 20 ] . When individual competences are known and signals are independent, the optimal aggregation rule is a weighted majority with log-odds weights (with an appropriate threshold), which generalizes simple majority [ 16 ] . Under supermajority rules, when average voter competence exceeds the required vote share, sufficiently large electorates are more likely to choose the correct alternative than a randomly selected voter [ 17 ] . Departures from independence introduce correlation-sensitive phenomena: if voters share information sources or follow opinion leaders [ 11 , 13 , 18 , 19 ] , positive correlation between voter decisions can erode or even reverse the Condorcet effect, whereas certain forms of negative correlation can strengthen it. Even so, conditions have been identified under which “wisdom of crowds” persists despite dependence, for example through explicit upper bounds on positive correlation under which majority decisions still outperform individual decisions, with accuracy that increases in larger groups. A complementary literature studies social influence before voting. In sequential settings, informational cascades can generate herding on incorrect actions despite informative private signals [ 21 ] . In networked communication with naïve (DeGroot) updating, convergence to truth in large societies requires that no agent retains asymptotically dominant influence, a structural property of the graph [ 22 ] . Related Bayesian-learning models over networks identify topologies that permit—or preclude—efficient information aggregation [ 23 ] . Finally, in majority-dynamics models where agents iteratively adopt local majorities prior to a final vote, network structure can either block aggregation or guarantee unanimity when initial biases are favorable (e.g., on expander graphs), underscoring that how information flows is as pivotal as how it is counted [ 24 ] . Taken together, these extensions, spanning statistical, epistemic, and social-choice perspectives, highlight both the power and limits of the classical CJT: large, independent, well-informed groups tend to excel at truth-tracking, but strategic behavior, correlation, and networked information can constrain accuracy. Despite this progress, models that make social influence and dependence explicit (e.g., via network-mediated interactions) and that yield closed-form characterizations of collective accuracy, allowing direct benchmarking against individual accuracy under realistic departures from independence, remain scarce. 2 Motivation We develop a general framework of social influence that bridges a complex-systems perspective on collective human dynamics with formal models of collective decision-making. Within this framework, we extend the Condorcet Jury Theorem (CJT) to socially connected networks, showing that—even under peer influence—networked decision-making can achieve higher collective accuracy than individual voting, thereby moving beyond the assumption of independence. Our analysis builds on the classical voter model [ 25 , 26 , 27 , 28 , 29 ] , and in particular on voter models with fixed agents (“zealots”) [ 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ] . We make two contributions. First, for the binary (two-zealot) model introduced in our prior work [ 35 , 38 , 39 , 40 ] , we give a new, direct derivation of the stationary distribution via a birth–death/detailed-balance argument and use it to obtain a generalized CJT. Second, we introduce a multistate ( m -zealot) model, derive its closed-form stationary distribution, and prove Condorcet–type lower bounds for both majority and plurality of the correct alternative. For clarity, we begin with the binary case as a motivating baseline; the full multistate theory is presented in the Results section and recovers the binary results when m=2 . We consider an undirected network (graph) G=(\mathcal{V},E) , where nodes \mathcal{V} are voters and edges E encode pairwise interactions. Each of the n+\alpha+\beta nodes holds a binary state: 1 (correct) or 0 (incorrect). There are \alpha nodes fixed in state 1 and \beta nodes fixed in state 0 ; we refer to these fixed-state nodes as zealots . The remaining n nodes are free voters whose states evolve via peer influence along the edges. The network evolves in discrete time. At each time step, a free node is selected uniformly at random, and its state is updated as follows: with probability \ell it remains unchanged ( retention/inertia ); with probability 1-\ell it copies the state of a uniformly chosen neighbor ( imitation ). In the long-time limit t\rightarrow\infty , the alternative supported by more than half of the free voters is selected as the consensus outcome. For a fully connected network (complete graph), the free voters are exchangeable, so the system’s state is fully determined by the number k\in\{0,\dots,n\} of

Read the complete text