SmartVoter
← Back to home

Methodology

How SmartVoter scores are computed, what they mean, and what they cannot tell you.

Core principle

SmartVoter compares your stated preferences with parties’ observed behavior, not just their stated promises. Observed votes carry more weight than platform declarations. This is not voting advice — it is a similarity analysis.

Match score

For each question you answer, we compute a similarity score between your position and the party’s inferred position:

distance = abs(your_position - party_position)

similarity = 1 − distance / 2

weighted = similarity × importance × evidence_strength

// Final score:

match = Σ(weighted) / Σ(importance × evidence_strength)

Positions range from −1 (one pole) to +1 (opposite pole). The axis direction is defined per policy item.

Confidence score

Confidence is computed separately from the match score and reflects how reliable the match score is:

confidence = evidence_strength × coverage × (1 − volatility) × stability

Coverage

Fraction of your important issues that have evidence for this party

Stability

Whether your ranking changes significantly when one answer is removed

Volatility penalty

Parties that change frequently have higher uncertainty

Evidence strength

Average reliability of sources used for this party’s positions

Evidence priority

Not all sources are equally reliable. We weight evidence by type:

Evidence typeWeight
Parliamentary vote1.00
Sponsored bill0.80
Committee behavior0.70
Candidate past votes0.55
Party lineage0.50
Coalition agreement0.45
Party platform0.35
Public statement0.25
Media interview0.20

New party handling

New parties are not excluded — but confidence is lower.

If a party has no parliamentary voting record, its position is inferred from candidate history (45%), lineage (25%), platform (20%), and statements (10%). These scores carry higher uncertainty and are shown with a warning. Parties with little evidence appear lower in confidence rankings, not excluded from results.

Adaptive questionnaire

The questionnaire selects questions dynamically based on your answers. Phase 1 (survey) covers all major topics with one broad question each. Phase 2 (depth) asks follow-up questions on topics you rated as important and on policy dimensions where top parties differ most. The questionnaire stops after 8–40 questions. You can view results at any time after answering 8 questions. Results are based only on questions you have answered — unanswered topics contribute no signal.

Data sources

Parliamentary vote data comes from the Israeli Knesset OData API (Votes.svc), which covers Knessets 1–24 (last available vote: July 2021). Bills and faction data are available for all Knessets via ParliamentInfo.svc. For parties in Knessets 25–26 without vote records, positions are inferred from candidate history, party platforms, and public statements with explicitly lower confidence. All LLM-generated content is reviewed by human administrators before appearing publicly.

Limitations

  • This tool does not tell you whom to vote for.
  • Vote data covers Knessets 1–24 (up to 2021). Newer parties are scored using candidate history, lineage, and declared platforms — with lower confidence.
  • Absence from a vote is treated as low-information, not opposition.
  • LLM-generated content is reviewed by humans before it appears publicly.
  • Party positions may shift over time. Check the data cutoff date shown on the results page.