What we will report, decided before the votes are counted

Published October 31, 2026, before any 2026 vote was counted, together with the preliminary report, Before the vote: the baseline, which applies these questions to 2024. Any later change is dated, explained and listed in the corrections log, and findings that come from a change are labelled with the date it was made.

Reports of findings can take sides without a single loaded word, simply by choosing which patterns to mention after seeing the results. To rule that out, this page fixes in advance what we will look at, how, and how we will word it. Every edition answers every question below, in the same order, including the questions where the data shows little or no pattern. Nothing is left out because of what it shows.

1. When findings are published

• First findings: once a state's count has passed our checks and its election office reports at least 95% of the expected vote (or all precincts), that state is included. National findings name the states still counting.
• Weekly chapters: November 12, 19 and 26, December 3 and 10, each answering every question below with the latest official figures.
• Final edition: December 18, 2026, using certified results wherever states have certified, and saying which have not.
• Until a state certifies, its figures are marked unofficial, with the time we read them.

2. The questions

A. Results. For the U.S. House, Senate and governors: seats won by each party (and by independents and third parties); the change in the number of seats held by each party; total votes for each party's candidates nationally and by state, with uncontested seats counted and named separately; runoffs and recounts pending.

B. Change since earlier elections. For every county and every House district:

• the 2026 margin (Democratic share minus Republican share, of all votes cast in that race);
• the change from the 2024 presidential margin, by county, for every state with a Senate or governor race;
• the change from the same office's previous election (Senate and governor by state; House by district for the 263 districts whose lines did not change);
• the number of counties and districts that moved toward each party, and by how much (under 1, 1 to 3, 3 to 6, 6 to 10, 10 points and over).

C. Turnout. Votes cast as a share of citizens of voting age (Census estimate), by state and county, compared with 2022 and 2024 where the same figure exists. Who voted by group comes only from the Census Bureau's own survey (published in spring 2027) and is added when it appears.

D. Who lives here. For each of the 13 measures of who lives in a place (population, citizens of voting age, ages 65 and over and 18 to 29, the five race and ethnicity shares we publish plus the other three, the share with a bachelor's degree, and the share below the poverty line):

• the correlation (r) with the 2026 margin, and with the change since 2024, across all counties (Senate and governor races) and across all House districts;
• the average margin and change in the highest, middle and lowest third of places on that measure;
• the same within every state with 30 or more counties.

All 13 are reported in one table in every edition, whatever their strength. The other Census figures (unemployment, income, commuting, housing and others) are reported the same way in a second table.

E. Polls. Only if polls whose publishers allow reuse are available: the final average in each race compared with the result, for every party alike. If no such polls are available, we say so.

F. Campaign money. From Federal Election Commission filings: money raised and spent by each House and Senate candidate, compared with votes received, reported for every candidate with a filing.

3. How the figures are worked out

• Margins are shares of all votes cast in the race, so third-party and write-in votes count in the total.
• Relationships are described by the correlation r with these words, always with the number: under 0.1 "almost no straight-line relationship", 0.1 to 0.3 "weak", 0.3 to 0.5 "moderate", 0.5 and over "strong".
• Every place counts once in an average, whatever its size; where we also give a figure weighted by votes or people, we say so.
• Census figures carry their margins of error; figures rated unreliable are reported but marked, and left out of correlations, averages and ranks.
• Districts with new lines in 2026 are never compared with their 2024 result, because that result belongs to a different shape. Missouri uses its 2024 lines.
• Louisiana's November 3 House races are a first round; their results are reported as such, and runoffs are added when held.
• Places where a race was uncontested are listed and left out of margins and correlations for that race.

4. How findings are worded

• Each finding is written by a fixed formula from the figures, the same formula for every place and every party, and checked against our list of loaded words before publication.
• Findings describe places, never people: no statement says or suggests how members of a group voted.
• No finding offers a cause. Where a pattern appears, the text says how strong it is, where it holds and where it does not.
• No predictions, no race calls, no ratings of races.

5. What we will not do

• We will not add a question after seeing the results without labelling it "added after the results" with the date.
• We will not drop or reorder a question because of its answer.
• We will not estimate how groups of voters voted.
• We will not use paid or unlicensed data.

This plan is part of our neutrality standard. Its exact text is fingerprinted (SHA-256) in the corrections log on the day it is published, so anyone can check that it has not been changed since.