Cohesion-Sensitive Power Indices: Representation Results for Banzhaf and Shapley Values

arXiv:2603.27220 · econ.TH, cs.GT · Submitted 2026-03-28 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Cohesion-Sensitive Power Indices".

Warren: Here's the gist of the paper. In many applications of cooperative game theory—from corporate governance and cartel formation to parliamentary voting—not all winning coalitions are feasible<ref:2603.27220#pg2> Ideological distances, institutional constraints, or pre-electoral agreements may render certain coalitions implausible<ref:2603.27220#pg5> This paper develops a general axiomatic framework for cohesion-sensitive power indices that incorporate such feasibility information<ref:2603.27220#pg2> Our central idea is to model feasibility via a cohesion structure on the set of coalitions and to study how cohesion interacts with representations of values via expected marginal contributions<ref:2603.27220#pg2> The resulting values are coalition-weighted Shapley values with Shapley size weights and cohesion-based distortions within size classes<ref:2603.27220#pg2> The classical literature often assumes that all winning coalitions are equally feasible, which is a pervasive simplifying assumption in the field<ref:2603.27220#pg2> In reality, coalition formation is often subject to severe internal frictions<ref:2603.27220#pg2> This paper introduces cohesion structures to quantify coalition feasibility and axiomatically characterizes two families of cohesion-sensitive power indices represented as expected marginal contributions under Luce-type distributions<ref:2603.27220#pg2> The Banzhaf branch adds axioms that restrict how cohesion affects coalition probabilities, leading to a representation of cohesion-weighted Banzhaf values parametrized by a power transformation of cohesion<ref:2603.27220#pg2> On the Shapley side, additional axioms separate size from cohesion, recovering the classical size weights with cohesion acting within each size class<ref:2603.27220#pg2> Representation Results for Banzhaf Values The representation theorem for cohesion-sensitive Banzhaf-type values shows that any such value admits a representation of the form Fi(v, κ) = ∑ S⊆N without requiring cohesion monotonicity<ref:2603.27220#pg2> This is achieved by defining probabilities pκi(S) as a power transformation of cohesion levels<ref:2603.27220#pg2> Specifically, the representation result shows that for the Banzhaf branch, there exists b > 0 such that Fi(v, κ) = ∑ S⊆N without requiring cohesion monotonicity<ref:2603.27220#pg2> The cohesion-weighted Banzhaf index is defined by fixing the exponent b = 1 and normalizing the resulting value<ref:2603.27220#pg2> Representation Results for Shapley Values For the Shapley branch, the representation theorem shows that once the cohesionless benchmark is fixed to the Shapley value, size weights are pinned down to classical weights and cohesion can only act via a power transformation within each size class<ref:2603.27220#pg2> The resulting family of values Fα,b satisfies all axioms invoked in Theorem 4.5<ref:2603.27220#pg2> This means that the probabilities factorize into a size component ωS and a cohesion component κ(·)b<ref:2603.27220#pg2> Applications and Interpretations The framework illustrates the main advantage of the Cohesion-Shapley value over classical indices by allowing precise quantification of how a cordon sanitaire shifts bargaining power from the pariah to the veto players in the political center<ref:2603.27220#pg2> For instance, in a three-player majority game, varying b from 0 to 1 traces how quickly the power assessment responds to an ideological realignment<ref:2603.27220#pg2> The results show that the cohesion framework rationalizes historical episodes of coalition realignment by showing how declining κ(SPD, FDP) combined with rising κ(CDU/CSU, FDP) shifts power against the SPD<ref:2603.27220#pg2> Scope and Outlook The framework is applicable to a range of settings beyond the parliamentary illustrations developed here, including corporate governance and industrial organization<ref:2603.27220#pg2> Natural extensions include calibrating the cohesion parameter κ from the degree of overlap between strategic portfolios or from regulatory antitrust hurdles facing different subsets of firms<ref:2603.27220#pg2> Future work will focus on cross-domain calibration of κ, formal comparative statics across institutional settings, and the integration of cohesion into endogenous coalition formation models<ref:2603.27220#pg2> The framework handles scenarios of absolute exclusion, such as a corporate blockholder restricted from voting alliances due to antitrust constraints<ref:2603.

Title and authors: Ruth: So to kick things off, let's talk about the paper, "Cohesion-Sensitive Power Indices: Representation Results for Banzhaf and Shapley Values," by Thomas Pitz and Vinicius Ferraz. We saw that these authors are addressing the fundamental flaw in classical cooperative game theory where they assume all winning coalitions are equally feasible.

Warren: They introduce cohesion structures to quantify this feasibility, which is crucial because ideological distances or institutional rules can make certain alliances impossible in practice, whether it's a corporate board or a political negotiation.

Iris: The paper aims to axiomatically characterize two families of power indices—the Banzhaf branch and the Shapley branch—by showing how cohesion modifies these values when looking at expected marginal contributions under Luce-type distributions.

Ruth: What’s interesting is that they generalize the idea that ideological distance improves plausibility of assessments, rather than just picking one specific formula for cohesion ad hoc. They build this power form in at the axiomatic level through axioms three point three and four point two.

Warren: They show how the linear specification, where the exponent b equals one, corresponds to what we often see empirically in these systems. That's a big clue about how these indices behave in practice.

Iris: And they also extend this framework to cover Shapley-type values by adding size-based separation axioms, which allows cohesion to act differently than just a simple distance formula would suggest.

The paper's summary: Ruth: To summarize the core of "Cohesion-Sensitive Power Indices: Representation Results for Banzhaf and Shapley Values," these authors show that they can represent any such cohesion-sensitive value without needing to assume a specific, fixed cohesion formula from the start.

Warren: They achieve this by defining probabilities p sub kappa i of S as a power transformation of the cohesion levels, which is really how they embed the feasibility information into the calculation.

Iris: Specifically for the Banzhaf branch, they show that there exists some positive number b such that any such value can be represented without requiring cohesion to be strictly monotonic across all coalitions. This means we don't need a simple distance formula to work if we use this power transformation κ b.

Ruth: And they fix the exponent b equals one and then normalize the resulting value for their cohesion-weighted Banzhaf index, which is what they focus on in that representation result. It’s about finding a way to weight coalitions based on how cohesive they are.

Warren: On the Shapley side, it's even more structured: once you fix the cohesionless benchmark to be the standard Shapley value, size weights are pinned down to classical weights, and cohesion can only act through a power transformation within each specific size class.

Iris: This leads to a representation where the probabilities factorize into two parts: a size component that is just omega multiplied by S sub i, and a cohesion component that depends on the cohesion level kappa of that coalition multiplied by b.

The paper's improvements: Ruth: Now let's look at what this paper suggests as improvements over previous work. The main advantage they point out is the ability to precisely quantify how a cordon sanitaire, or structural exclusion, shifts bargaining power from the pariah party to veto players in the political center.

Warren: They show that by varying that exponent b from zero up to one, you can trace exactly how quickly a power assessment responds when there's an ideological realignment happening between parties.

Iris: That’s a really useful tool for testing different scenarios; it lets you see if the power shifts dramatically even with small changes in cohesion structure. They use this to rationalize historical episodes of coalition realignment by showing how declining cohesion between two parties combined with rising cohesion between another pair shifts power against the party in question.

Ruth: They’re essentially using this framework to show that structural friction isn't just noise; it’s a mechanism that dictates which coalitions are actually effective. It moves us away from treating all winning coalitions as equally plausible, which is what classical indices often do.

Warren: And for the Shapley branch, the improvement there is separating size and cohesion completely; it recovers the classical size weights while allowing cohesion to only modulate power within those predefined size classes. That’s a cleaner way to think about how different coalition sizes interact with internal friction.

Conclusion: Ruth: So wrapping up the "Cohesion-Sensitive Power Indices: Representation Results for Banzhaf and Shapley Values" paper, we see that this framework gives us tools to move past the simplistic idea that all winning coalitions are equally feasible. It provides a systematic way to incorporate feasibility information directly into power indices.

Warren: The key result is that they can represent these values using power-law transformations of cohesion, and they've shown how this works for both Banzhaf and Shapley types under specific axiomatic constraints.

Iris: From an econometric standpoint, the separation of size and cohesion in the Shapley branch is particularly powerful because it lets us model how structural exclusion affects different coalition sizes independently. It gives us a much clearer picture of where power is actually concentrated.

Ruth: The implication for the world is that if you're modeling corporate boards or political negotiations, you can now use this framework to show exactly how ideological distance or institutional friction modifies the expected marginal contributions of players in a way that classical models simply can't capture.

Warren: It’s not about replacing every single model, but it gives us a principled way to test if our assumptions about feasibility are holding up under real-world constraints and shows us how sensitive those power assessments are to changes in the underlying cohesion structure.

Faculty of Society and Economics, Hochschule Rhein-Waal · Institute of Management, Karlsruhe Institute of Technology (KIT) · Singularity AI Research, Singularity.inc

econ.TH, cs.GT

Submitted: 2026-03-28

Updated: 2026-11-05

Comments: major restructuring will be done

Code: https://github.com/vferraz/cohesion-power-indices

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 92/100

The gist: In many applications of cooperative game theory—from corporate governance and cartel formation to parliamentary voting—not all winning coalitions are feasible<ref:2603.27220#pg2> Ideological…

Key concepts

Cohesion Structure
This is a mathematical way to quantify the feasibility of forming a coalition based on internal frictions like ideological distance or institutional rules. It models which coalitions are actually possible, rather than assuming all combinations can form.
Expected Marginal Contributions
These are used to represent power indices. The paper studies how these contributions interact with cohesion structures, showing that the resulting values can be modeled using Luce-type distributions, linking individual influence to group outcomes.
Power Indices (Banzhaf/Shapley)
These are measures of influence or power within a group. The authors develop 'cohesion-sensitive' versions of these standard indices. This means the calculated power reflects not just who is influential, but also how feasible their coalition is under specific structural constraints.
Exponent b
This parameter controls how much the cohesion structure influences the final power index calculation. Changing 'b' allows researchers to test sensitivity: a higher 'b' means the power assessment reacts more strongly to changes in ideological alignment.

Terminology

Summary

In many applications of cooperative game theory—from corporate governance and cartel formation to parliamentary voting—not all winning coalitions are feasible<ref:2603. p. 027220#pg2> Ideological distances, institutional constraints, or pre-electoral agreements may render certain coalitions implausible<ref:2603. p. 027220#pg5> This paper develops a general axiomatic framework for cohesion-sensitive power indices that incorporate such feasibility information<ref:2603. p. 127220#pg2> Our central idea is to model feasibility via a cohesion structure on the set of coalitions and to study how cohesion interacts with representations of values via expected marginal contributions<ref:2603. p. 127220#pg2> The resulting values are coalition-weighted Shapley values with Shapley size weights and cohesion-based distortions within size classes<ref:2603. p. 127220#pg2>

Introduction to the Framework

The classical literature often assumes that all winning coalitions are equally feasible, which is a pervasive simplifying assumption in the field<ref:2603.27220#pg2> In reality, coalition formation is often subject to severe internal frictions<ref:2603. p. 127220#pg2> This paper introduces cohesion structures to quantify coalition feasibility and axiomatically characterizes two families of cohesion-sensitive power indices represented as expected marginal contributions under Luce-type distributions<ref:2603. p. 027220#pg2> The Banzhaf branch adds axioms that restrict how cohesion affects coalition probabilities, leading to a representation of cohesion-weighted Banzhaf values parametrized by a power transformation of cohesion<ref:2603. p. 127220#pg2> On the Shapley side, additional axioms separate size from cohesion, recovering the classical size weights with cohesion acting within each size class<ref:2603. p. 027220#pg2>

Representation Results for Banzhaf Values

The representation theorem for cohesion-sensitive Banzhaf-type values shows that any such value admits a representation of the form Fi(v, κ) = ∑ S⊆N without requiring cohesion monotonicity<ref:2603.27220#pg2> This is achieved by defining probabilities pκi(S) as a power transformation of cohesion levels<ref:2603. p. 627220#pg2> Specifically, the representation result shows that for the Banzhaf branch, there exists b > 0 such that Fi(v, κ) = ∑ S⊆N without requiring cohesion monotonicity<ref:2603. p. 227220#pg2> The cohesion-weighted Banzhaf index is defined by fixing the exponent b = 1 and normalizing the resulting value<ref:2603. p. 627220#pg2>

Representation Results for Shapley Values

For the Shapley branch, the representation theorem shows that once the cohesionless benchmark is fixed to the Shapley value, size weights are pinned down to classical weights and cohesion can only act via a power transformation within each size class<ref:2603.27220#pg2> The resulting family of values Fα,b satisfies all axioms invoked in Theorem 4. p. 85<ref:2603. p. 027220#pg2> This means that the probabilities factorize into a size component ωS and a cohesion component κ(·)b<ref:2603. p. 927220#pg2>

Applications and Interpretations

The framework illustrates the main advantage of the Cohesion-Shapley value over classical indices by allowing precise quantification of how a cordon sanitaire shifts bargaining power from the pariah to the veto players in the political center<ref:2603.27220#pg2> For instance, in a three-player majority game, varying b from 0 to 1 traces how quickly the power assessment responds to an ideological realignment<ref:2603. p. 327220#pg2> The results show that the cohesion framework rationalizes historical episodes of coalition realignment by showing how declining κ(SPD, FDP) combined with rising κ(CDU/CSU, FDP) shifts power against the SPD<ref:2603. p. 027220#pg2>

Scope and Outlook

The framework is applicable to a range of settings beyond the parliamentary illustrations developed here, including corporate governance and industrial organization<ref:2603.27220#pg2> Natural extensions include calibrating the cohesion parameter κ from the degree of overlap between strategic portfolios or from regulatory antitrust hurdles facing different subsets of firms<ref:2603.27220#pg2> Future work will focus on cross-domain calibration of κ, formal comparative statics across institutional settings, and the integration of cohesion into endogenous coalition formation models<ref:2603.27220#pg2> The framework handles scenarios of absolute exclusion, such as a corporate blockholder restricted from voting alliances due to antitrust constraints<ref:2603. p. 1

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Enhanced Power Assessment in Feasible Coalitions The system can now quantify coalition feasibility by using cohesion structures to continuously modulate coalition formation probability, moving beyond classical indices that ignore structural constraints like ideological distances or institutional friction.

  2. Banzhaf Value Calibration with Ideological Distances The system can calculate a cohesion-weighted Banzhaf value where the exponent b is chosen based on an estimated cohesion structure κ, allowing for a sensitivity analysis of power shifts when ideological distances change.

  3. Shapley Value Refinement via Size and Cohesion Separation The system can generate a cohesion-sensitive Shapley value that separates coalition formation probability into two independent sources: a combinatorial size component and a political cohesion component, leading to more nuanced power distributions.

  4. Modeling of Structural Exclusion Scenarios The improved system can transparently model the cordon sanitaire or other structural exclusions by setting the cohesion level κ(S) to zero for winning coalitions containing a pariah party while retaining its existence in the player set N.

  5. Comparative Analysis Across Aggregation Levels The system can perform comparative analysis between bloc-level and party-level power assessments, revealing how treating an alliance as a monolithic actor (bloc model) differs from disaggregating it into constituent parties (party model).

  6. Interpretation of Exponent b as Sensitivity Measure The system can use the exponent b to trace how quickly the power assessment responds to an ideological realignment, providing a quantitative measure of sensitivity to changes in cohesion structure.

Related papers