How we measure

Votes with own party

• Counts roll call votes where most Democrats voted against most Republicans. CQ Roll Call has used this definition of a "party-unity vote" for decades.
• The score is the share of those votes where the member sided with most of their own party.
• Independents count with the party they caucus with. A member who switches parties gets a separate score for each party.
• The "real votes only" version leaves out procedural votes: debate rules, motions to table, motions to recommit, cloture and motions to proceed. These split along party lines almost every time.
• This Congress, 70.0% of House roll calls and 91.8% of Senate roll calls were party-unity votes.

Votes with the President

• Counts votes where the President or an official spokesperson stated a position publicly before the vote.
• Two numbers: all votes, and legislation only, with nominations removed. The Senate votes on many nominations, which can crowd out policy votes.
• Each nomination counts once, on the final confirmation vote. The President supports confirming his own nominees. On a veto override, he opposes the override.
• Official White House Statements of Administration Policy are matched to the votes they cover: the first final vote on the bill in each chamber within 45 days of the statement (a position on a bill holds whichever chamber votes on it), or, if the chamber the statement addresses held no final vote, its last vote to end debate or force the bill out of committee. Some positions come from the President's own public statements before a vote (remarks, posts quoted by news organizations, or White House officials speaking for the administration); each lists its source.
• The position counted is the latest one stated publicly before the vote. A statement on one amendment counts only for that amendment's vote. A statement that was not public until after the vote (some are posted weeks or months late) is left out.
• In October 2026 every position was fact-checked against the statements themselves, the official roll-call records and news coverage, looking for reversals, mismatched bills and missing positions. The check corrected the record on the Lindsey O. Graham Sanctioning Russia and Iran Act: a statement opposing one amendment had been read as opposition to the bill, which the President supported.
• Every entry, with its source, is on the President's positions page. It currently holds 451 votes: 228 nominations, 2 veto overrides and 212 matched White House statements.
• Each matched statement is checked against its full text and the vote record: same bill, same chamber, a clear position on the bill as a whole, and the vote that decides it. Statements that only object to parts of a bill are left out. This check is done with AI assistance (Claude), not by a human editor; the reason for each decision is kept with the entry. To report an error, use the corrections page.
• A "Present" vote is left out, like a missed vote.

Expected vs. actual

Party loyalty depends a lot on where a member is from. For each chamber and party, we fit a straight line between each member's score and how their district (or state, for senators) voted for president in 2024. A member's expected score is the point on that line for their district. The gap is how far their actual score sits from it.

Bipartisan bills led

• A bill counts when the member sponsored it and at least one member of the other party cosponsored it at introduction.
• It also counts for the first other-party cosponsor at introduction, the cross-party co-lead.
• Only bills and joint resolutions count, because only they can become law. Bills that name post offices and other federal buildings are left out.
• Each bill shows how far it got: introduced, passed one chamber, passed both, became law. Percentiles compare members within their own chamber, because senators cosponsor far more bills than House members.

Liberal-to-conservative rank

• Based on DW-NOMINATE from Voteview at UCLA, the standard academic measure of voting patterns, updated as votes happen.
• Members whose scores sit very close together share a range of ranks instead of an exact rank. New members show "not enough votes yet" until they have cast 25 scaled votes.
• Each member gets three ranks, each with a percentile from 0 (most liberal) to 100 (most conservative): among all of Congress, among their own party in Congress (both chambers), and within their chamber.
• Comparing a senator with a House member is valid because Voteview scores both chambers on one shared scale. Members who moved from the House to the Senate keep exactly the same score.
• Independents are ranked with the party they caucus with.
• The score measures who votes with whom. A member who votes no because a bill goes too far, or not far enough, can look moderate. Read it alongside the other three scores.

Seat colors

• In the seat charts, a seat's position is the liberal-to-conservative rank above. Its color is a separate measure: how often the member crosses party lines toward the middle.
• It uses the same party-unity votes as "Votes with own party": votes where most Democrats voted against most Republicans. A member crosses when they vote with the other party's majority.
• A crossing counts as "toward the middle" when the members of their own party who broke ranks on that vote come, on average, from the party's moderate half (by the liberal-to-conservative score, against the party's median in that chamber). Breaking ranks from the outer edge, alongside the party's most liberal or most conservative members, isn't moving toward the middle and doesn't count.
• Color is the member's party, moving to purple as the share rises. Purple: 10% of those votes or more. Below that the shades start at 7.5%, 5%, 3%, 1.5%; under 1.5% is the solid party color. These are fixed cutoffs, not ranks, so sitting near the middle of the chart doesn't make a seat purple.
• Independents take the color of the party they caucus with. Members without enough party-unity votes yet are gray.

Checks against published studies

Every build recomputes 2025 averages the way CQ Roll Call does (counting missed votes against members) and compares them with CQ's published figures. The build stops if any is off by more than 2 points.

Data sources