Chapter 6: Campaigns as Heuristic Fields
“A campaign is not an argument. It is a lens.” — Anonymous political consultant
RUNNING EXAMPLE — DISTRICT 7
Two candidates compete in District 7’s congressional race. Candidate A is a progressive — strong on economic redistribution (d_1 left), moderate on social values (d_2 center), committed on environment (d_3 high), and trusted on institutional competence (d_5 moderate). Candidate B is a populist conservative — moderate on economics (d_1 center-right), conservative on social values (d_2 right), skeptical on environment (d_3 low), and fiercely anti-institutional (d_5 very low, d_6 high).
The election will not be decided by which candidate is “closer” to the electorate on the full manifold. It will be decided by which candidate succeeds in making the electorate perceive the election through their preferred dimension. Candidate A’s campaign wants the election to be about healthcare and wages — a d_1 election, where A has the advantage. Candidate B’s campaign wants the election to be about institutional corruption and cultural identity — a d_5-d_6 election, where B has the advantage.
Each campaign is a heuristic field: a vector field on the preference manifold that guides voters’ attention toward specific dimensions and away from others. The winner will be the campaign whose field dominates — not the candidate whose position is nearest to the electorate, but the candidate who successfully rotates the projection axis to pass through their strongest dimension.
The Heuristic Field Formalism in Politics
The heuristic field formalism, developed in Geometric Reasoning (Ch. 3), models any guidance system — any source of information that helps a searcher navigate a complex space — as a vector field on the search manifold. The heuristic field \mathbf{h}: \mathcal{M} \to T\mathcal{M} assigns to each point on the manifold a vector indicating the direction the guidance system recommends.
In the parent framework, the heuristic field applies to individual reasoning: System 1’s fast pattern-matching is a heuristic that guides System 2’s deliberate search. In economics (Geometric Economics, Ch. 5), the heuristic field is the price signal: prices estimate the cost of moving from one economic state to another, guiding agents along efficient paths. In law (Geometric Law, Ch. 6), the heuristic field is legal precedent: prior cases estimate how new cases should be decided, guiding judges along the path of stare decisis.
In politics, the heuristic field has a distinctive structure: it is not a single field but a superposition of competing fields, each generated by a different source of political information. The political heuristic at any voter’s position on the manifold is:
\mathbf{h}(v) = \sum_k \alpha_k(v) \cdot \mathbf{F}_k(v)
where \mathbf{F}_k is the field generated by source k (campaign, media outlet, social network, personal experience) and \alpha_k(v) is the influence weight of source k at voter v’s position.
A political campaign is one component of this superposition: a vector field \mathbf{F}_C: \mathcal{P} \to T\mathcal{P} that pushes voters’ attention in a specific direction.
What Campaigns Actually Do
The naive model of campaigns is that they move voters: a persuasive argument shifts a voter’s position on the manifold from one point to a point closer to the candidate. This model is largely false. Decades of campaign effects research — from Lazarsfeld, Berelson, and Gaudet (1944) through Gerber and Green (2000) to Kalla and Broockman (2018) — have established that campaigns have small effects on voter preferences. Most voters’ manifold positions are stable over the course of a campaign. The marginal voter who switches sides is the exception, not the rule.
If campaigns do not move voters, what do they do? The geometric answer: they rotate the projection axis.
A campaign does not (primarily) change voters’ positions on the manifold. It changes the salience of different dimensions, thereby manipulating which projection of the manifold determines the election outcome. The campaign is not a force that moves the electorate — it is a lens that changes how the electorate is seen.
Formally, a campaign C defines a projection axis \mathbf{e}_C \in \mathbb{R}^d (a unit vector in the d-dimensional preference space) such that the election outcome is determined by the projection of each voter’s position onto \mathbf{e}_C:
\text{perceived proximity}(v, C_i) = \langle v - x_{C_i}, \mathbf{e}_C \rangle
where v is the voter’s position, x_{C_i} is candidate i’s position, and \langle \cdot, \cdot \rangle is the inner product. The voter perceives the candidate as close if the projected distance is small, regardless of the actual manifold distance.
The Campaign Gradient Theorem
This leads to the chapter’s central result: the Campaign Gradient Theorem, which characterizes the optimal campaign strategy in geometric terms.
Theorem 2 (Campaign Gradient Theorem). A successful campaign finds the projection axis \mathbf{e}^* along which the candidate’s position minimizes the mean projected distance to the electorate:
\mathbf{e}^* = \arg\min_{\mathbf{e}, \|\mathbf{e}\|=1} \mathbb{E}_v\left[|\langle v - x_C, \mathbf{e} \rangle|^2\right]
This is equivalent to finding the eigenvector of the electorate’s covariance matrix \Sigma along which the candidate’s position x_C is closest to the mean \mu:
\mathbf{e}^* = \arg\max_{\mathbf{e}, \|\mathbf{e}\|=1} \frac{|\langle x_C - \mu, \mathbf{e} \rangle|^2}{\mathbf{e}^T \Sigma \mathbf{e}}
subject to the constraint that \mathbf{e} must correspond to a politically activatable dimension — the campaign can only make salient dimensions that voters recognize as politically relevant.
Proof. The expected squared projected distance is:
\mathbb{E}[|\langle v - x_C, \mathbf{e} \rangle|^2] = \mathbf{e}^T \Sigma \mathbf{e} - 2\langle x_C - \mu, \mathbf{e} \rangle \mathbf{e}^T (\mu - x_C) + |\langle x_C - \mu, \mathbf{e} \rangle|^2
Minimizing over unit vectors \mathbf{e} is a generalized eigenvalue problem. The optimal axis balances two objectives: minimizing the electorate’s variance along \mathbf{e} (choosing a dimension on which voters are clustered, so the candidate is close to many of them) and maximizing the candidate’s alignment with the electorate’s mean along \mathbf{e} (choosing a dimension on which the candidate is close to the centroid). \square
What the Theorem Predicts
The Campaign Gradient Theorem predicts that candidates will:
Emphasize issues where they have a positional advantage. A candidate who is closer to the electorate’s mean on d_3 (environment) than on d_1 (economics) will campaign on environmental issues — not because environment is the most important issue but because the projection onto d_3 produces a more favorable outcome for the candidate.
Suppress issues where they have a positional disadvantage. The same candidate will avoid discussing economics — not because they have no economic position but because the projection onto d_1 is unfavorable. The campaign’s silence on economics is a strategic refusal to activate that projection axis.
Seek low-variance dimensions. Among dimensions on which the candidate has roughly equal positional advantage, the campaign will choose the dimension with the lowest voter variance — the dimension on which voters are most clustered. On a high-variance dimension, even a well-positioned candidate faces many voters who are far away on the projection. On a low-variance dimension, a well-positioned candidate faces a concentrated electorate that is uniformly close.
These predictions are confirmed by every empirical study of campaign strategy. Campaigns emphasize their candidate’s strengths, suppress their weaknesses, and target the dimensions that produce the most favorable projections. The Campaign Gradient Theorem is not a new prediction — it is a geometric explanation for a well-established empirical regularity.
Negative Campaigning as Heuristic Corruption
If a candidate’s own campaign is a heuristic field that guides voters toward favorable projections, a negative campaign is a corrupted heuristic field directed at the opposing candidate.
In the parent framework (Geometric Reasoning, Ch. 5), heuristic corruption occurs when the guidance field ceases to accurately represent the structure of the search space. A corrupted heuristic provides misleading distance estimates, steering the searcher away from optimal paths and toward suboptimal ones.
In politics, negative campaigning corrupts the heuristic field in three ways:
Distance inflation. Attack ads inflate the perceived manifold distance between the voter and the attacked candidate. By highlighting the candidate’s most extreme positions, framing moderate positions in alarming language, and associating the candidate with unpopular groups, the attack ad increases the voter’s estimate of d_\mathcal{P}(v, x_{\text{opponent}}) — the perceived distance between the voter and the opponent. The actual manifold distance may be small, but the corrupted heuristic makes it seem large.
Dimensional amplification. Attack ads selectively amplify specific dimensions of the opponent’s position. If the opponent is moderate on five dimensions but extreme on one, the attack ad magnifies that one dimension, making it the dominant projection axis. The voter, seeing the opponent through the amplified dimension, perceives them as far away on the manifold even though they are close on five other dimensions.
Noise injection. Some negative campaigns do not target a specific dimension but simply inject noise into the voter’s heuristic field. “Throwing mud” — creating a general impression of untrustworthiness, incompetence, or scandal — degrades the voter’s ability to estimate manifold distances accurately. The noise does not move the voter toward the attacking candidate; it makes the voter unable to compute accurate distances to any candidate, producing confusion, cynicism, and disengagement.
These three corruption modes are exactly the failure modes identified in the parent framework: distance inflation is heuristic corruption, dimensional amplification is objective hijacking (forcing the voter to optimize on the wrong dimension), and noise injection is heuristic degradation. The campaign battlefield is a field of competing heuristic forces, some informative and some corrupt.
The Primary as Dimensional Filter
Before the general election’s heuristic battle, the primary election performs a different geometric operation: it filters the candidate set through a dimensional lens that may differ from the general election’s lens.
A partisan primary selects from the party’s manifold region: the candidates who compete in the Democratic primary occupy the progressive region of \mathcal{P}, and the primary electorate evaluates them within that region. The projection axis of the primary may differ from the projection axis of the general election. In the Democratic primary, the relevant axis might be the d_1-d_3 diagonal (economic progressivism vs. environmental progressivism), while in the general election, the relevant axis might be the d_1-d_5 diagonal (economic policy vs. institutional trust).
This means that the primary selects a candidate who is optimal on the primary’s projection axis — not necessarily on the general election’s projection axis. The candidate who wins the primary by being the most compelling on the intra-party dimension may be poorly positioned on the inter-party dimension. The classic “too extreme for the general” problem is, geometrically, a dimensional mismatch: the primary selects for position on an intra-party axis, but the general election evaluates on an inter-party axis, and these axes are different.
Open primaries — where any voter can participate, regardless of party registration — mitigate this dimensional mismatch by expanding the primary electorate beyond the party’s manifold region. Voters from the other party and from the manifold’s center can participate, pulling the primary’s projection axis closer to the general election’s axis. The geometric prediction: open primaries produce nominees who are closer to the full electorate’s Frechet mean, reducing the BI of the eventual general election winner.
Top-two primaries — where all candidates from all parties compete in a single primary, and the top two vote-getters advance to the general election — take this further. The top-two primary evaluates candidates on the full electorate’s manifold, not on the party’s submanifold. The candidates who advance are the two who are nearest to the electorate’s Frechet mean on the projection axis that the primary’s campaign defines. The dimensional filtering of the partisan primary is eliminated.
The geometric trade-off: partisan primaries preserve party identity (each party sends its preferred manifold representative) at the cost of dimensional mismatch. Open and top-two primaries reduce dimensional mismatch at the cost of party identity (the party may not be represented by its preferred candidate). The trade-off is between intra-party dimensional fidelity and inter-party manifold representation.
The Debate as Manifold Exploration
A political debate is a structured exploration of the preference manifold. Each question forces the candidates to reveal their positions on a specific dimension. The information value of a debate depends on how many dimensions it samples.
A well-designed debate samples all six dimensions: economic policy questions, social values questions, environmental questions, foreign policy questions, questions about institutional trust, and questions about identity and cultural issues. Each question reveals the candidates’ positions on one axis, providing voters with the information needed to estimate manifold distances. A debate that samples all six dimensions gives voters enough information to locate the candidates approximately on the full manifold.
A poorly designed debate samples only one or two dimensions — typically d_1 (economics) and d_2 (social values) — leaving four dimensions unexplored. The voter who watches this debate has accurate information about the candidates’ positions on two dimensions and no information about the other four. The voter’s estimate of manifold distance is based on two dimensions and is therefore a projection, subject to all the distortions that projections produce.
The information value of a debate can be formalized as the volume of the manifold region it samples:
I_{\text{debate}} = \text{Vol}\left(\text{span}\{\mathbf{q}_1, \ldots, \mathbf{q}_k\}\right)
where \mathbf{q}_i are the directions in manifold space activated by each debate question and k is the number of questions. If the questions span a large volume of the manifold (covering many independent dimensions), the debate is informative. If the questions are collinear (all on the same dimension), the volume is zero and the debate provides no more information than a single question.
Modern presidential debates in the United States tend to be low-information by this measure: most questions address d_1 and d_2, with occasional excursions into d_4 (foreign policy). Dimensions d_3 (environment), d_5 (trust), and d_6 (identity) are rarely addressed directly, though they may be addressed implicitly in the candidates’ rhetoric. The information value of a typical 90-minute debate is, by the manifold measure, perhaps 40% of what it could be if all dimensions were sampled equally.
District 7: The Campaign Battle
The campaign in District 7 unfolds as a battle of heuristic fields.
Candidate A’s campaign deploys a field \mathbf{F}_A that points along d_1. The campaign’s message is relentlessly economic: healthcare costs, wage stagnation, affordable housing, childcare subsidies. Every ad, every speech, every door-knock script emphasizes Candidate A’s progressive economic positions. The campaign’s theory of the race: if the election is decided on d_1, Candidate A wins, because the district’s Frechet mean on d_1 is slightly left of center, and Candidate A is closer to the mean on d_1 than Candidate B.
Candidate B’s campaign deploys a field \mathbf{F}_B that points along the d_5-d_6 diagonal. The campaign’s message is populist and identity-inflected: government corruption, cultural change, border security, the “real people” of the district versus the “elites.” Every ad, every rally, every social media post emphasizes Candidate B’s anti-institutional, identity-affirming positions. The campaign’s theory of the race: if the election is decided on d_5-d_6, Candidate B wins, because the district’s trust-identity axis is the strongest axis of variance (recall from Chapter 4 that \lambda_1 = 3.2 corresponds to the trust-identity dimension), and Candidate B is closer to the electorate’s centroid on this axis.
The battle is not over which candidate voters prefer. It is over which dimension voters consider when they vote.
In the final two weeks, Candidate B’s campaign goes negative: an attack ad accuses Candidate A of supporting a policy that would raise local property taxes (a d_1 attack, designed to neutralize A’s economic advantage) and another ad associates Candidate A with an unpopular national political figure (a d_6 attack, designed to inflate perceived cultural distance). The attack ads corrupt the heuristic field near Candidate A’s position on the manifold: voters who were considering A on economic grounds are now uncertain about A’s economic position (noise injection) and perceive A as culturally distant (distance inflation).
Candidate A’s campaign responds not by attacking B directly but by attempting to re-anchor the projection axis on d_1: a series of testimonial ads from local workers and families describing how A’s policies would affect their lives. The counter-strategy is heuristic repair: providing concrete, personally relevant information that overrides the corrupted heuristic with direct manifold-distance estimation.
Election night: Candidate B wins by 3 points. The exit polls show that voters who cited “the economy” as their top issue voted for A by 15 points. Voters who cited “trust in government” or “values” voted for B by 30 points. The election was decided by which heuristic field dominated — and Candidate B’s d_5-d_6 field activated more voters than Candidate A’s d_1 field.
The postmortem is geometric: Candidate B did not win because more voters agreed with B on the full manifold. Candidate B won because B’s campaign successfully rotated the projection axis from d_1 (where A had the advantage) to d_5-d_6 (where B had the advantage). The same electorate, projected onto a different axis, would have elected A.
The campaign did not change the voters. It changed the lens.
DISTRICT 7 — CHAPTER SUMMARY
We have formalized campaigns as heuristic fields on the political preference manifold — vector fields that guide voters’ attention toward specific dimensions and away from others. The Campaign Gradient Theorem shows that the optimal campaign strategy is to find the projection axis along which the candidate is closest to the electorate’s center of mass. Negative campaigning is heuristic corruption: inflating perceived distances, amplifying unfavorable dimensions, and injecting noise into voters’ manifold-distance estimates.
In District 7, Candidate B won not by being closer to the electorate on the full manifold but by successfully rotating the projection axis to the trust-identity dimension, where B had the positional advantage. The campaign was a battle of heuristic fields, and the winning field determined the election’s dimensional focus.
In Chapter 7, we turn to the deepest result in social choice theory — Arrow’s impossibility theorem — and show that it is not a paradox of axioms but a consequence of geometry: the impossibility of losslessly contracting a multi-dimensional manifold to a one-dimensional ranking.