Control by Deleting Candidates

• Control by Deleting Candidates is a framework where an external agent removes up to k candidates to sway election outcomes.
• The method distinguishes between constructive and destructive control, with many voting systems exhibiting NP-hard or W[2]-hard resistance to such manipulations.
• Empirical studies show that while worst-case complexity is high, practical heuristics can enable feasible manipulation in moderate-sized, real-world elections.

Control by deleting candidates is a central paradigm in computational social choice, capturing manipulative actions that alter an election’s candidate pool to change the outcome. Under this framework, an external agent (“the chair”) strategically removes up to $k$ candidates from an election to either ensure that a distinguished candidate wins (constructive control) or prevent that candidate from winning (destructive control). The complexity analysis of such control actions has proved essential for understanding the robustness, security, and strategic vulnerability of various voting systems.

1. Formal Definitions and Core Variants

For voting systems where the candidate set is finite, let $E = (C,V)$ denote an election with candidate set $C$ and voter multiset or list $V$. Given a distinguished candidate $p \in C$, and an integer $k \geq 0$, two canonical control problems are defined:

• Constructive Control by Deleting Candidates (CCDC):

Is there a set $D \subseteq C \setminus \{p\}$, $|D| \leq k$ such that in the reduced election $(C \setminus D, V)$, $p$ is the unique winner?

• Destructive Control by Deleting Candidates (DCDC):

Is there a set $D \subseteq C \setminus \{p\}$, $|D| \leq k$ such that in $(C \setminus D, V)$, $p$ is not the unique winner?

The specification of “unique winner” is critical for most theoretical analyses and greatly impacts complexity (Erdélyi et al., 2011).

2. Complexity Landscape across Voting Rules

Candidate-deletion control is widely used to measure the “resistance” of voting systems—i.e., whether realizing a manipulation is computationally hard. A rule is called resistant to a control type if the corresponding decision problem is NP-hard (susceptible and intractable).

• Plurality, Bucklin, Fallback, Borda, SP-AV: All show NP-hardness (i.e., resistance) for both constructive and destructive candidate deletion (Erdélyi et al., 2011, Erdélyi et al., 2010, Erdélyi et al., 2010, 0806.0535, Zhou et al., 2024).
• Schulze Rule: CCDC is NP-complete, but destructive candidate deletion (Schulze-DCDC) is polynomial-time solvable (Maushagen et al., 2024).
• Ranked Pairs: Both CCDC and DCDC are NP-complete (Maushagen et al., 2024).
• Participatory Budgeting (PB): For advanced PB rules (Phrágmen, Equal-Shares), CCDC is NP-complete, while for GreedyAV and GreedyCost, tractability depends on cost encoding (Faliszewski et al., 23 Jan 2026).
• Amendment/Successive Procedures: Control by deleting candidates is polynomial-time solvable for amendment ($h=1$) and full-amendment ($h=m-1$) rules, while successive and generalized $h$-amendment become W[1]-hard with certain parameters (solution size, number of undeleted candidates) (Yang, 1 Jan 2025).
• Counting Variants: For Plurality and $k$-Approval, the counting version (#CCDC)—finding the number of deletion sets yielding a given winner—is #P-complete on unrestricted ballots (Wojtas et al., 2014).

System/Scenario	Constructive CCDC	Destructive DCDC
Bucklin, Fallback, Borda, SP-AV	NP-hard, often W[2]-hard	NP-hard, often W[2]-hard
Schulze	NP-complete	Polynomial (in P)
Ranked Pairs	NP-complete	NP-complete
PB (Phrágmen, ES, most Greedy)	NP-complete	NP-complete

Context: NP-hardness proofs typically use reductions from Hitting Set, Dominating Set, or Exact Cover problems. Parameterized complexity, such as W[2]- or W[1]-hardness for deletion-budget $k$ or number of voters $n$, is prevalent and provides a more nuanced picture for small or structured elections (Erdélyi et al., 2011, Zhou et al., 2024, Chen et al., 2014).

3. Methodologies: Reductions and Algorithms

Most resistance results are established via reductions:

• Bucklin and Fallback Voting: Reductions from k-Dominating Set or Hitting Set produce instances where successfully controlling the election is equivalent to finding a dominating set of specified size (Erdélyi et al., 2011, Erdélyi et al., 2010, Erdélyi et al., 2010).
• SP-AV: Reductions from Hitting Set, with sophisticated “coercion” mechanisms to maintain vote admissibility after candidate deletion (0806.0535).
• Borda: Gadget-based reductions from Dominating Set ensure only deletion sets corresponding to dominating sets can make the distinguished candidate win (Zhou et al., 2024).
• Schulze: A corrected 3SAT-based construction that maps clause satisfaction to candidate deletions, leveraging McGarvey’s trick in the Weighted Majority Graph (Maushagen et al., 2024).
• PB Rules: Encodings use set/project analogues, approval ballots, and budget constraints to simulate RX3C, with GreedyAV/GreedyCost algorithms exhibiting polynomial-time solutions if costs are in unary (Faliszewski et al., 23 Jan 2026).

Common sophisticated proof constructs include group voting blocks, ballot coercion mechanisms, and path-preserving vertex cuts in majority graphs (for Schulze). Parameterized reductions are essential for obtaining W[2]- or W[1]-hardness (Erdélyi et al., 2010, Zhou et al., 2024, Chen et al., 2014, Yang, 1 Jan 2025).

Algorithmic methodologies are either exhaustive (brute-force search over candidate subsets, dynamic programming on vote profiles) or rely on combinatorial preordering and pruning. For rules where winner computation is in polynomial time, feasible algorithms exist for small $k$ or $n$, or in combinatorial settings with limited candidate bundling (Chen et al., 2014).

4. Theoretical Insights and Parameterized Complexity

A major finding is that resistance is prevalent across most natural voting systems with polynomial-time winner determination. However, parameterized analyses reveal subtle exceptions:

• W[2]-Hardness: Bucklin, fallback, and Borda CCDC are W[2]-hard when parameterized by deletion limit $k$ (Erdélyi et al., 2011, Erdélyi et al., 2010, Zhou et al., 2024).
• W[1]-Hardness by Voter Number: Borda, $k$-Approval, Copeland, and combinatorial CCDC are W[1]-hard for parameter $n$ (number of voters), even if $n\leq10$ suffices for intractability (Chen et al., 2014).
• FPT Regimes: Plurality and maximin rules admit polynomial-time or FPT algorithms for voter-parameterized CCDC (Chen et al., 2014).
• Special Cases: For amendment and full-amendment procedures, polynomial-time algorithms exist regardless of $k$; for successive procedures, W[1]-hardness holds unless deletions focus only on $p$’s agenda successors (Yang, 1 Jan 2025).
• Top-truncated Voting: Borda CCDC with truncated votes ($t\geq2$) is NP-hard and W[2]-hard for $k$; only $t=1$ (single nominee) admits an efficient solution (Zhou et al., 2024).

5. Empirical Results and Practical Tractability

Experimental investigations reveal that although worst-case CCDC is NP-hard, practical tractability is observed in moderately sized random elections:

• Experimental Success Rates: In Impartial Culture models, constructive CCDC succeeds in 30–50% for $m,n\geq16$; destructive CCDC success rates approach 90% (Rothe et al., 2012).
• Computational Cost: Even for $m,n$ up to 128, per-instance running times are sub-second for plurality, with Bucklin and fallback taking longer for negative instances (Rothe et al., 2012).
• No Sharp Phase Transition: Success-rate curves and cost distributions show monotonic increases without a “critical regime” of computational intractability (Rothe et al., 2012).
• Practical Heuristics: Depth-first search with pruning, candidate harm preordering, and dynamic-programming implementations facilitate feasible manipulations in practice, despite theoretical resistance (Rothe et al., 2012).

A plausible implication is that the nominal computational complexity barrier is “diffuse” in realistic settings and concentrated only in adversarially constructed instances.

6. Applications and Extensions

Candidate deletion control is not only a theoretical device but serves as an evaluative metric in participatory budgeting (PB), especially for assessing the strength of losing projects:

• PB Strength Measures: Deletion-based scores, minimum-deletion distances, and rivalry heat-maps quantify how “close” a project was to winning, and how budgetary competition structures outcomes (Faliszewski et al., 23 Jan 2026).
• Probability Analyses: Winning probabilities under random deletions, and correlation with bribery-style measures, reveal orthogonal aspects of project performance (Faliszewski et al., 23 Jan 2026).
• Combinatorial Control: Bundled candidate deletion (where deleting one candidate forces removal of a group) is significantly harder unless committee/voter sizes are extremely small (Chen et al., 2014).

A plausible implication is that control by deleting candidates, both as an analytic tool and as a manipulation model, offers insights into election security, system robustness, and fine-grained policy evaluation.

7. Comparative Tables, Open Questions, and Future Directions

The following table summarizes resistance status for major rules:

Voting Rule	Winner Complexity	CCDC Resistance	DCDC Resistance	Parameterized Hardness (budget $k$)
Bucklin/Fallback	P	NP-hard	NP-hard	W[2]-hard
Borda	P	NP-hard	NP-hard	W[2]-hard
SP-AV	P	NP-hard	NP-hard	NP-hard
Plurality	P	NP-hard	NP-hard	FPT in $n$
Schulze	P	NP-hard	P
Ranked Pairs	P	NP-hard	NP-hard
PB Advanced Rules	P	NP-hard	NP-hard

Editor’s term: “parameterized candidate-deletion control (pCCDC)” refers to the paradigm analyzing complexity in terms of $k$, $n$, or other instance parameters.

Open questions and future research directions include the complexity landscape for multimode control combinations, empirical evaluation on alternative vote models, and combinatorial extensions in real-world settings (Yang, 1 Jan 2025, Faliszewski et al., 23 Jan 2026).