The Vote Ledger
/
Data through Sep 17, 2026Compare members
The Vote Ledger
Methodology

How the numbers are made

Every rule on this page was written after measuring real data, and each says what the site would otherwise get wrong. A politician only has to find one wrong number to make the story about our method instead of their record.

01

The measure

For every roll call where the parties split, we ask two questions: which way did the Democratic majority vote, and which way did the Republican majority vote? Then we check which of them each member matched. The measure never looks at the member's own party.

With Democrats=votes matching the Democratic majoritycontested votes the member castcomputed separately for legislation and for nominations

Because party is never an input, a party switch shows up as one continuous line rather than a break — and the line often moves before the label does. Strom Thurmond sided with the Democratic majority 82% of the time in 1955 and 17% by 1961, while still a Democrat. He changed parties in 1964.

See Thurmond's arc
02

What counts as contested

D · YEAR · NAY
Counts
A majority of Democrats voted one way and a majority of Republicans the other.
D · YEAR · YEA
Left out
Lopsided votes pull everyone toward the middle and tell you nothing about sides.
≥ 5 VOTING
A real majority
A party majority needs at least five of its members voting, so one member can never stand in for a caucus.
DELEGATE
Scored, never the baseline
Delegates from DC and the territories are scored but cannot help form a party majority — they cannot vote on final passage.
03

Legislation and nominations, scored apart

The Senate confirms nominees on the floor, and in recent Congresses those votes swamp the calendar: 476 of the 768 contested Senate roll calls in the 117th Congress were nominations. Blend them in and a member's legislative record disappears. So legislation leads, and nominations sit beside it, labelled.

Example · Susan Collins · 117th Congress
64.6%with Democrats

Blended, she reads as a reliable Democratic vote. Both numbers are true; this one misleads, which is why it is never the headline.

Legislation35.6%
Nominations82.4%
Combined64.6%
04

Reading a score

Five display rules. Break any of them and the site starts asserting things that are not true.

Blank is not zeroBlank means no qualifying votes. Zero means sided with them on none of many — maximum opposition.
—no qualifying votes0%sided with them on none
Every percentage names its denominatorThe count beside a score is the number of votes it was divided by — never a looser coverage count.
37.5%of 4,448 legislation votes
Delegates are labelledA delegate's score is honest; the inference a reader draws from it, without the label, is not.
Eleanor Holmes NortonNON-VOTING DELEGATEcannot vote on final passage
Years only before 1935For early terms the year is real but the month and day were filled in. We print the year.
Term began 1857APPROX.
Thin policy areas stay visible"Voted here, but never on a contested question" and "voted a few times, too few to score" are different claims.
Too few votes to score · count shown
05

The historical record is thin

Every roll call before 1973 has no description — not most, all of them. For an old vote we have a date, usually a bill number, how each member voted and which side that matched. We show exactly that and fill nothing in by hand.

EraRoll callsNo descriptionNo bill number
35th–92nd 1857–197341,766100%23%
93rd–119th 1973–now52,15349%13%

This is why member pages are built around the career arc, not a feed of vote descriptions: a description-led design works for recent Congresses and collapses on the historical record, where some of the most interesting members are.

06

Validation

Checked against sources the data did not come from.

Tallies match VoteView totals—of 113,550 roll calls
House Clerk recordsExacton every case checked
Correlation with DW-NOMINATE—independent measure
Automated tests—run on every build
07

The errors we look for

Nearly every real defect found while building this was a plausible number, not a crash. None failed a test. All would have shipped looking reasonable.

aA percentage divided by the wrong denominator.
bA motion to adjourn filed under Health.
cA senator reading 47 points more Democratic than her legislative record.
dValidation tests that silently skipped while reporting green.
Found a number that looks wrong?Tell us the page and the figure. We would rather hear it from you.
Report a wrong number
Methodology

How the numbers are made

Every rule here was written after measuring real data. Tap a section to open it.

For every roll call where the parties split, we ask which way the Democratic majority voted and which way the Republican majority voted, then check which one each member matched. A member's own party is never an input.

With Democrats =votes matching the Democratic majoritycontested votes the member castComputed separately for legislation and nominations.
D YEAR NAYCounts. The party majorities went opposite ways.
D YEAR YEALeft out. Lopsided votes say nothing about sides.
A real majority: at least five members of a party must vote.
Delegates are scored but never help form a majority.

476 of the 768 contested Senate roll calls in the 117th Congress were nominations. Blend them in and the legislative record disappears.

Susan Collins · 117th Congress
64.6%with Democrats

Blended, she reads as a reliable Democratic vote. True, and misleading.

Cite the source data

Lewis, Poole, Rosenthal, Boche, Rudkin and Sonnet. Voteview: Congressional Roll-Call Votes Database.

AD320 × 50 · Anchor