How SpectrumVote works

The left–right spectrum

Every candidate gets a dot on one line from left (−1) to right (+1). The middle is the center of American politics today, not a "correct" position. We never endorse, and a dot is never a recommendation.

We use the best source available for each person, in this order:

1. DW-NOMINATE for anyone who has served in Congress (their votes).
2. Shor–McCarty for state legislators (their votes in the legislature).
3. DIME CFscores for anyone who has raised campaign money since 2010 (who funds them).
4. AI estimate from the candidate's own public record, only if none of the above exists and the record says enough.

One common scale. The three academic measures each have their own units. We put Shor–McCarty and DIME on DW-NOMINATE's scale with a simple linear link that maps each measure's median Democrat and median Republican (recent legislators or candidates) onto the median Democrat and Republican in the current Congress. A typical member of either party therefore lands in the same place whichever measure placed them, and the center is DW-NOMINATE's center.

A filled dot is a published academic score (1–3). A hollow dot is an AI estimate (4). If we have nothing reliable, we say so rather than guess. Labels: within ±0.15 of center is "Center"; up to ±0.4 "Slightly"; up to ±0.7 "Moderately"; beyond that "Strongly" left or right.

DW-NOMINATE Academic · votes

The standard political-science measure of ideology in Congress, created by Keith Poole and Howard Rosenthal and maintained by Voteview at UCLA. It places every member from every roll-call vote they've cast, by comparing who votes with whom. It never uses the party label: a Republican who often votes with Democrats lands closer to the center.

We show the first dimension (the economic/government-role dimension that explains most votes) from the member's most recent Congress, already on a −1 to +1 scale. If someone voted in Congress under one party and is now running under another, we note it.

Lewis, Jeffrey B., Keith Poole, Howard Rosenthal, Adam Boche, Aaron Rudkin, and Luke Sonnet. Voteview: Congressional Roll-Call Votes Database. https://voteview.com/

Shor–McCarty Academic · votes

Boris Shor and Nolan McCarty's scores place every state legislator in the country on one common scale, using their roll-call votes in the legislature linked across states by a national candidate survey. It's the standard academic measure for state legislators. Shown as "Shor–McCarty (state legislature votes)" with the chamber and years served.

Data covers legislators who served 1993–2022. If someone served under more than one party, we score their votes under the party they're running with now and note the other.

Shor, Boris, 2025, "Individual State Legislator Shor-McCarty Ideology Data, January 2025 Update", https://doi.org/10.7910/DVN/SGOQ7G, Harvard Dataverse, V1. Licensed CC BY-NC-SA 4.0. See also Shor, Boris and Nolan McCarty. 2011. "The Ideological Mapping of American Legislatures." American Political Science Review 105(3); and Shor and McCarty. 2022. "Two Decades of Polarization in American State Legislatures." Journal of Political Institutions and Political Economy 3: 343–370.

DIME CFscores Academic · donors

Adam Bonica's Database on Ideology, Money in Politics, and Elections (Stanford) places candidates by who gives them money: donors tend to give to candidates who share their views, so analyzing hundreds of millions of donations across every candidate places donors and candidates on the same scale. This covers people who have never cast a recorded vote: challengers, statewide and local candidates.

CFscores run roughly −2 to +2. We link them to the DW-NOMINATE scale (see one common scale), separately for federal and state-level campaigns because state donors produce less spread-out scores. We only use a CFscore when we're confident it's the same person: same state, matching name, no judge records for non-judges, no conflicting party, and we skip namesakes we can't tell apart. Data covers campaigns through 2024, so first-time 2026 candidates aren't in it.

Bonica, Adam. 2024. Database on Ideology, Money in Politics, and Elections: Public version 4.0. Stanford University Libraries. https://data.stanford.edu/dime (ODC-BY 1.0).

AI estimates Hollow dot

When no academic score exists, the Jev model by TypeSafe AI reads the candidate's own public record: their Wikipedia article and their campaign website (homepage plus up to three issues/about pages). It places them on a 7-point scale from strongly left to strongly right.

• It's never told the candidate's party. The placement has to come from what they say and have done.
• Evidence gate: a second question asks whether the record actually describes policy positions. If not, we show "Not enough public policy information" instead of a guess.
• Every estimate shows its confidence and links the pages it read.

Ballot measures are placed the same way from the official ballot summary: where a YES vote falls. Judges: the model looks for a stated judicial philosophy; career bios usually don't have one, so most judges show no score, which is the honest answer.

Issue-by-issue AI · per issue

For candidates whose record passed the evidence gate, the same model places them separately on six issues: the economy and taxes, immigration, abortion, guns, climate and energy, and health care. Each issue has its own evidence check: if the record doesn't state their position on that issue, we show "no stated position" rather than inferring it from their overall lean. Many voters are moderate overall but strongly on one side of one issue; this is meant to show that.

Voting record

For current members of Congress, counted directly from Voteview's roll-call files for the 119th Congress (no AI):

• Missed votes: the share of roll calls they didn't vote on.
• Bills sponsored: bills and resolutions they introduced this Congress, how many became law, their most common policy areas and the most recent ones, from the Congress.gov API.
• Votes with own party: on votes where most Republicans and most Democrats voted opposite ways, how often they sided with their own party's majority. Lower means more independent.

Who funds them

Federal candidates: Federal Election Commission filings for the 2025–26 cycle: totals by source and the top PAC and committee donors. A PAC's lean is computed from its own giving (the share of its contributions to Republican vs Democratic candidates), not from its name or an AI guess. ActBlue and WinRed are shown as pass-through platforms for individual donations.

State and local candidates (Florida, Mississippi): the state's and counties' own campaign-finance filings, where available.

Ballot measures

We show the official ballot title and summary, a plain-language description of what YES and NO do, the financial impact from the Legislature's analysis or the state's Financial Impact Estimating Conference, how it got on the ballot (with the final legislative vote), the registered committees for and against with money raised, and a sourced selection of organizations on each side. Florida constitutional amendments need 60% YES.

Parties

Hovering a party label shows what that party says it stands for, summarized from its own national platform in parallel wording, plus where political scientists generally place it. Candidates vary within every party, which is why each person is placed from their own record rather than their label.

Where the ballot data comes from

Data refreshes daily. Always confirm with your official sample ballot from your county elections office.

Privacy

• Your address never reaches us. Your browser sends it directly to the U.S. Census Bureau (to find your districts; OpenStreetMap if Census doesn't know the address) and to Google's Civic Information API (for ballots and polling places). We receive only district numbers.
• District-map lookups (which county or city district you're in) also go straight from your browser to the county's map server.
• Your picks are saved only in your browser. A shared picks link carries district numbers and picks in the part of the link after "#", which browsers never send to any server.
• We use Google Analytics to count visits and which features get used (with Google Signals and ad personalization off). It never receives your address, districts or picks.

Limits

• One left–right line simplifies real views. Use the issue breakdown and the sources.
• DW-NOMINATE and Shor–McCarty reflect past votes; DIME reflects past donors (through 2024). People change.
• AI estimates can be wrong. They're labeled, show confidence, and link what they read.
• Local candidate lists come from filings and can lag official status changes. Your sample ballot is the final word.

En español: cómo funciona SpectrumVote

Cada candidato recibe un punto en una línea de izquierda (−1) a derecha (+1). Usamos la mejor fuente disponible para cada persona: DW-NOMINATE (sus votos en el Congreso, Voteview/UCLA), Shor–McCarty (sus votos en la legislatura estatal), CFscores de DIME (quién financia su campaña, Stanford) y, solo si no existe ninguna de esas, un estimado por IA (modelo Jev de TypeSafe AI) a partir de su propio sitio web y artículo de Wikipedia. Al modelo nunca se le dice el partido; si el historial no describe posiciones, lo decimos en lugar de adivinar.

Punto relleno = puntuación académica publicada. Punto hueco = estimado por IA, con su nivel de confianza y las fuentes que leyó. También mostramos posiciones por tema (economía, inmigración, aborto, armas, clima y salud) con su propia verificación de evidencia, el historial de votos de los miembros del Congreso (votos perdidos y con qué frecuencia votan con su partido), el financiamiento (FEC y registros estatales) y los detalles oficiales de cada medida.

Privacidad: tu dirección nunca nos llega; tu navegador la envía directamente a la Oficina del Censo y a Google. Tus selecciones se guardan solo en tu navegador. Nunca respaldamos a nadie. Confirma siempre con tu boleta de muestra oficial.