# Methodology


PollSignal helps voters understand which right-of-centre candidate appears best placed in their Westminster constituency. The aim is simple: reduce avoidable vote-splitting between Conservative, Reform, and other right-of-centre candidates where that split could affect the result.

Our recommendations are based on constituency-level evidence, public election data, candidate information, polling or forecast data where available, and manual editorial review. They are not guarantees, predictions, or instructions. Voters should make their own decision.

## What we are trying to answer

For each Westminster constituency, we ask:

> Which right-of-centre candidate, if any, appears to have the strongest realistic path to winning the seat or preventing the least preferred outcome for right-of-centre voters?

The answer may be one of five outcomes:

| Outcome | Meaning |
|---|---|
| **Recommended tactical vote: Conservative** | The Conservative candidate currently appears to be the strongest right-of-centre option in the constituency. |
| **Recommended tactical vote: Reform** | The Reform candidate currently appears to be the strongest right-of-centre option in the constituency. |
| **Recommended tactical vote: other right-of-centre candidate** | A different right-of-centre candidate appears better placed than both Conservative and Reform. |
| **Too close / insufficient data** | The evidence is mixed, incomplete, stale, or too close to justify a clear recommendation. |
| **No realistic right-of-centre path** | No right-of-centre candidate currently appears competitive enough for a meaningful tactical recommendation. |

## Data we consider

We do not rely on a single number or one model. A constituency recommendation may use some or all of the following inputs:

- The most recent general election result.
- Notional or boundary-adjusted results where relevant.
- Candidate lists and nomination information.
- Current or recent polling and forecasts, where reuse is permitted.
- National polling trends.
- Party vote share and trend by constituency.
- Whether Conservative and Reform support appears to be split.
- Whether the non-right-of-centre vote is split.
- The sitting or previous MP and seat history.
- Candidate credibility, local profile, campaign strength, and public information.
- Manual editorial review and source notes.
- Corrections submitted by users, candidates, or local observers.

## Recommendation workflow

Recommendations go through a controlled review process.

1. **Ingest source data**  
   We import or review constituency data, candidate data, previous results, and any relevant polling or forecast evidence.

2. **Validate and normalise**  
   Constituency names, party names, candidate names, and external identifiers are checked and standardised before they are used.

3. **Compare right-of-centre options**  
   We compare Conservative, Reform, and any other right-of-centre candidates in the constituency.

4. **Assess split-vote risk**  
   We look at whether right-of-centre votes are likely to be divided in a way that changes the constituency outcome.

5. **Assign a recommendation state**  
   The seat is assigned a recommendation, “too close” status, or “no realistic path” status.

6. **Assign confidence**  
   The recommendation receives a confidence rating based on the strength, recency, and consistency of the evidence.

7. **Editorial review**  
   Human editors review the seat before publishing. Automated data can flag a seat for review, but it does not automatically publish a political recommendation.

8. **Publish and archive**  
   Published recommendations are versioned. When a recommendation changes, the page is updated and the public data snapshot can be refreshed.

## Confidence ratings

A confidence rating shows how strongly the evidence supports the recommendation.

| Rating | Meaning |
|---|---|
| **High** | The available evidence points clearly toward one right-of-centre candidate. |
| **Medium** | The evidence supports a recommendation, but there are meaningful uncertainties. |
| **Low** | A recommendation may be possible, but the evidence is weak, stale, or contested. |
| **Insufficient data** | We do not have enough reliable evidence to make a recommendation. |

Confidence is not the same as probability of victory. A high-confidence recommendation can still lose. It means the recommendation is strongly supported by the available evidence.

## Split-vote risk

Split-vote risk measures whether right-of-centre voters appear likely to divide their support between multiple candidates in a way that could affect the result.

| Risk | Meaning |
|---|---|
| **Severe** | The right-of-centre vote appears highly divided and the split may determine the winner. |
| **Material** | The split matters and could influence the result. |
| **Low** | One right-of-centre candidate is clearly ahead, or the split is unlikely to affect the outcome. |
| **Unknown** | We do not have enough evidence to assess the split reliably. |

## Candidate classifications

Candidate labels are editorial notes designed to help explain the recommendation. They may include labels such as:

- Sound right.
- Mainstream conservative.
- Wet / centrist.
- Unknown.
- Reform candidate credible.
- Reform candidate paper-only / low information.
- Other right-of-centre independent.

These classifications are not automated. They are reviewed manually and may change as new public information becomes available.

## Manual overrides

Election data is imperfect. A purely automatic model can miss local realities such as candidate selection problems, withdrawals, campaign strength, tactical pressure, local polling, or boundary changes.

For that reason, editors may apply a manual override. When this happens, the recommendation should include an explanation and a last-reviewed date.

## Why we sometimes do not make a recommendation

We may choose not to recommend a candidate where:

- Conservative and Reform are too close to separate reliably.
- Data is too old or inconsistent.
- Candidate information is incomplete.
- No right-of-centre candidate appears competitive.
- A local factor needs further review.
- A reliable recommendation would require data we do not have.

In those cases, the page will normally show **Too close / insufficient data** or **No realistic right-of-centre path**.

## Limits of the methodology

PollSignal is an evidence-based political information tool, not an official electoral service. The methodology has limits:

- Polls and forecasts can be wrong.
- Constituency-level polling is often unavailable.
- National polling may not reflect local dynamics.
- Boundary changes can make historical comparisons difficult.
- Candidate strength can change quickly during a campaign.
- Some postcode-to-constituency tools assign postcodes using a centroid, which may be imperfect near constituency boundaries.
- Tactical voting depends on how many voters coordinate around the same option.

We therefore avoid absolute claims such as “only X can win here” unless the evidence is strong. Where the evidence is weak, we say so.

## Corrections and reviews

Each constituency page includes a last-reviewed date. Users can report corrections if they believe candidate data, constituency data, sources, or recommendations are wrong or outdated.

When a correction is accepted, the relevant constituency record is reviewed and the public page may be updated.

## Editorial standard

Our standard is:

- use identifiable sources where possible;
- avoid overstating weak evidence;
- separate factual data from editorial judgement;
- show confidence and uncertainty;
- update recommendations when evidence changes;
- publish enough information for users to understand the recommendation.