What the 13 Keys and the midterm models say about Nov. 3

Allan Lichtman raised his Senate call for Democrats from 67 percent to 70 percent in a YouTube video released Thursday night, The National reported on Nov. 2. He put the House at a “better than 90 per cent chance” of changing hands, and he made both calls without the 13 Keys. The National said he relied on polls, issue surveys and Trump’s approval ratings. The Atlantic Lens podcast, which published its interview with him on Oct. 29, said the model is designed for presidential elections.

What the keys are

The 13 Keys are true-or-false statements about the party holding the White House. History News Network reproduced them from a Sept. 23, 2016 Washington Post item. Keys 1 through 5 are a party mandate (more House seats after the midterm than after the previous one), no serious nomination contest, the sitting president as the candidate, no significant third party, and no recession during the campaign. Keys 6 through 8 are long-term real per capita growth at or above the two prior terms, major policy change, and no sustained social unrest. Keys 9 through 13 are no major scandal, no major foreign or military failure, a major foreign or military success, a charismatic incumbent-party candidate, and an uncharismatic challenger. Wikipedia’s article states the rule: six or more false keys and the incumbent party is predicted to lose.

Wikipedia counts the model correct in nine of eleven presidential elections since 1984. The Washingtonian, in an Oct. 27, 2026 profile by Chris Almeida, charts the same races as eight hits, two question marks and one miss. The question marks are 2000 and 2016, where the popular vote and the Electoral College split. The miss is 2024, when Lichtman chose Kamala Harris in a New York Times video op-ed in early September and conceded on election night during a livestream with his son. Lichtman has said the keys forecast the popular-vote winner, the Washingtonian reported. Nate Silver has written that Lichtman is less a discoverer of the right keys than “a locksmith” who can keep minting new ones.

Scoring the keys for a midterm, as an exercise

Applied to Tuesday with the incumbent party held at the president’s party, the Republicans, some keys have a midterm reading and some have none.

Key 1, the party mandate, reads false by the nearest available comparison. The key compares House seats after consecutive midterms, so the closest substitute is the seats won in the last two general elections. The congress-history table gives those as 222 Republican seats in 2022 and 220 in 2024.

Key 5, the short-term economy, reads true. The Bureau of Economic Analysis reported on Oct. 29 that real GDP grew at a 3.4 percent annual rate in the third quarter of 2026. The second quarter grew at 2.2 percent, so neither quarter shows a contraction. Key 6, the long-term economy, was not computed, and Key 7, major policy change, is left unscored with the keys below.

Key 8, social unrest, has documented inputs and a judgment at the end. The state files for this series record anti-ICE protests in January 2026 in all seven states. Reported crowds included about 500 in Detroit, with one arrest, and more than 1,000 in Milwaukee. The Arizona Capitol drew up to about 2,000 by the state Department of Public Safety’s estimate. The files also record Border Patrol’s Charlotte operation, begun in November 2025, with 425 or more arrests by the Department of Homeland Security’s count through Dec. 3. Protests in Madison followed a July 22, 2026 police shooting and continued into early August, with overnight fires reported. Whether that pattern is “sustained” under Lichtman’s wording is the scorer’s call. No National Guard deployment for protest control turned up in the seven states.

Keys 4, 7, 9, 10 and 11 translate in principle: third party, policy change, scandal, foreign failure and foreign success. They are left unscored for want of a sourced national measure, though the table below reads the third-party key race by race. Keys 2, 3, 12 and 13 do not translate. A midterm has no incumbent-party nomination, no sitting president on the ballot and no single challenger whose charisma can be judged. With four keys undefined, no six-key verdict follows, and this piece does not total one.

What the models say

The pattern these models start from is in the American Presidency Project’s table of seats gained or lost by the president’s party. From 1934 through 2022, 23 midterms, the president’s party gained House seats three times: 1934 (plus 9), 1998 (plus 4) and 2002 (plus 8). Over the 20 midterms since 1946 it lost seats in 18.

Alan Abramowitz’s generic-ballot model was published on Rasmussen Reports on April 24, 2025. It takes two inputs: the seats the president’s party is defending and its average generic-ballot margin. The coefficients are about 2.6 House seats per point of generic ballot and 0.54 seats lost per exposed seat. The model explains more than 80 percent of the House variance over 1946 to 2022. For 2022 it predicted an 18-seat Democratic House loss against an actual 9. It also predicted a 1-seat Senate gain, which matched the result. Of 20 House predictions, 16 fell within 10 seats, and 11 of 20 Senate predictions fell within 2. That week’s RealClearPolitics generic ballot stood at Democrats plus 1.5. At that margin he projected a Republican loss of more than a dozen House seats and five Senate seats. He then wrote that the 1.5-point lead probably would not be enough. A Republican net loss of four or more Senate seats, he wrote, “would probably take a Democratic lead of close to 10 points.” The RealClearPolitics average as of Nov. 2, 2026 was Democrats plus 8.8 across 22 polls taken from Oct. 9 to Nov. 1. The Silver Bulletin average stood at plus 8.9 on Nov. 2.

Charles Tien and Michael Lewis-Beck forecast from presidential approval and disposable personal income. Their LSE post of Oct. 13, 2025 projected a Republican loss of 28 House seats. The post said the model had called net midterm losses correctly in 2014, 2018 and 2022. The version in the PS: Political Science & Politics symposium has Republicans losing 33 seats and the majority, as summarized in a Cambridge University Press release of Oct. 14, 2026. The approval series those models were built on no longer exists. Gallup announced in February that it was ending presidential approval measurement after 88 years, CNN Newsource reported on Feb. 13, 2026. The current readings come from aggregators and single polls. PollCheck had 37.9 approve and 59.3 disapprove on Nov. 2, and Reuters/Ipsos had 32 percent among 1,679 adults in a poll that closed Oct. 26.

Joseph Bafumi, Robert Erikson and Christopher Wlezien forecast House seats from generic-ballot polls, and no 2026 forecast from them was located. In September 2010 a forecast by Bafumi and his co-authors gave Republicans a 79 percent chance of the House and about 229 seats, Dartmouth reported. That forecast ran 1,000 simulations of the national vote and factored in the 435 district races. Republicans gained 63 seats that November.

The symposium’s Cambridge collection page lists 20 items, and the Oct. 14 release said the issue includes 17 studies. The release reported Quinlan and Lewis-Beck at 231 Democratic House seats (78 percent) and 50 Senate seats (66 percent). Enns and co-authors put the House chance at 90 percent and 235 seats. Quinn, Brush and Smith had a Republican loss of 42 House and 9 Senate seats, and Rametta and Hare had 50 Democratic Senate seats in a 45-to-56 range. Political Science Now wrote on Oct. 15 that 12 of 12 House predictions favored Democrats and that the Senate forecasts were split. Kalshi’s own page put the market’s implied chance for Democrats at 94.1 percent for the House and 75 percent for the Senate. Those are contract prices pulled at 2:11 p.m. ET on Nov. 2.

The swing states

Race	Incumbent party	Incumbent running	Contested incumbent-party primary	Third party at 5%+	RCP average	RCP rating
AZ governor	D	Yes (Hobbs)	No (unopposed)	No (max 2.0)	Hobbs +9.2	Leans Dem
GA governor	R	No (Kemp term-limited)	Yes (Jackson 32.5% first round, 52.6% runoff)	No (Libertarians off ballot)	Jackson +1.0	Toss Up
GA Senate	D	Yes (Ossoff)	No (unopposed)	No	Ossoff +8.2	Likely Dem
MI governor	D	No (Whitmer term-limited)	No (Benson 83.2%)	No (max 1.8)	Benson +7.7	Leans Dem
MI Senate	D	No (Peters retiring)	Yes (El-Sayed 48.45%, Stevens 47.55%)	No (max 2.0)	El-Sayed +5.0	Toss Up
NV governor	R	Yes (Lombardo)	No (90.8%)	None named	No polls in average	Toss Up
NC Senate	R	No (Tillis retiring)	Yes, 64.6% (under two-thirds; no sustained challenge found)	No (one poll at 5)	Cooper +11.0	Likely Dem
PA governor	D	Yes (Shapiro)	No (unopposed)	No (2 to 3)	Shapiro +23.8	Safe Dem
WI governor	D	No (Evers not running)	Yes (Crowley 39.81%, Hong 39.35%)	None on ballot	Crowley +4.5	Toss Up

Primary shares are the results tables in the Wikipedia race articles, which cite state returns. Certified totals were not checked in Arizona, Michigan or Nevada. Contested means a nominee under two-thirds of the primary vote, and RCP figures were retrieved Nov. 2.

Five of the nine seats are open. By the two-thirds test, the incumbent party nominated through a contested primary in four of them. Three of those four, Georgia governor, Michigan Senate and Wisconsin governor, are rated Toss Up by RCP. The fourth is North Carolina Senate, where Whatley won with 64.6 percent against a runner-up at 15.6 and the seat is rated Likely Democratic. The fifth open seat, Michigan governor, went to Benson with 83.2 percent and is rated Leans Democratic. Of the four incumbents on the ballot, three Democrats lead by 8.2 to 23.8 points. The one Republican, Nevada’s Joe Lombardo, sits in a Toss Up with no polls in the RCP average. An Oct. 26 to 28 Noble poll had that race at 47 to 47. No race has a third-party candidate at 5 percent in the RCP averages.

The race pieces in this series carry the county arithmetic for each of these contests. The poll-error table at hesitation.media/election/#poll-error shows how large a polling miss each side can survive in each state. Abramowitz wrote the Senate threshold quoted above in April 2025, when the RealClearPolitics generic ballot stood at Democrats plus 1.5.

---

Notes and sources

Passage	Figure	Source
Lichtman Senate 67 to 70, House better than 90 percent, no keys used, Thursday video	67, 70, 90	The National, Cody Combs, Nov. 2, 2026, https://www.thenationalnews.com/news/us/2026/11/02/midterm-election-prediction-lichtman-sayed/
Keys designed for presidential elections		Atlantic Lens, Oct. 29, 2026, https://atlanticlens.substack.com/p/allan-lichtman-2026-us-midterms-trump
Key list wording	13 keys	History News Network, Sept. 23, 2016 (from The Washington Post), https://www.historynewsnetwork.org/article/163937
Six-false rule; nine of eleven since 1984; 2016 book; 2024 miss	6, 9 of 11	Wikipedia, “The Keys to the White House,” https://en.wikipedia.org/wiki/The_Keys_to_the_White_House
Eight hits, two question marks, one miss; NYT video; election-night concession; popular-vote defense; Silver quotation	8, 2, 1	Washingtonian, Chris Almeida, Oct. 27, 2026, https://washingtonian.com/2026/10/27/the-election-prophet-who-got-it-wrong/
House seats won 2022 and 2024	R 222 (118th), R 220 (119th)	C:\Users\brian\congress-history\data\clean\congress_table.csv
Real GDP Q3 and Q2 2026	3.4, 2.2	BEA advance estimate, Oct. 29, 2026, https://www.bea.gov/data/gdp/gross-domestic-product
Protest counts and dates	2,000; 500; 1,000+; 425+ (DHS count through Dec. 3, 2025)	C:\Users\brian\Desktop\Journalism\keys-2026\reporting\states-AZ-GA-MI-NV.md and states-NC-PA-WI.md, with the outlets cited there (AZ Mirror, BridgeDetroit, Urban Milwaukee, WBTV, FOX6)
Midterm seat changes 1934 to 2022; 2010 gain of 63	+9, +4, +8; 18 of 20; 63	American Presidency Project, https://www.presidency.ucsb.edu/statistics/data/seats-congress-gainedlost-the-presidents-party-mid-term-elections
Abramowitz coefficients, record, 2026 projections, 10-point threshold	2.6, 0.54, 18 vs 9, 16 of 20, 11 of 20, +1.5	Rasmussen Reports, April 24, 2025, https://www.rasmussenreports.com/public_content/political_commentary/commentary_by_alan_i_abramowitz/generic_ballot_model_gives_democrats_strong_chance_to_take_back_house_in_2026
RCP generic ballot	D+8.8, 22 polls, Oct. 9 to Nov. 1	RealClearPolitics average (race 8670) via the RCP data connector, as of Nov. 2, 2026
Silver Bulletin generic ballot	D+8.9	https://www.natesilver.net/p/generic-ballot-average-2026-nate-silver-bulletin-congress-polls, updated Nov. 2, 2026
Reuters/Ipsos approval	32; 1,679	Reuters via GV Wire, Oct. 26, 2026, https://gvwire.com/2026/10/26/trump-approval-rises-from-record-low-to-32-ahead-of-midterm-election/
Tien and Lewis-Beck 28 seats; 2014, 2018, 2022 record	28	LSE USAPP blog, Oct. 13, 2025, https://blogs.lse.ac.uk/usappblog/2025/10/13/forecasting-suggests-the-republicans-will-lose-28-seats-and-the-house-in-the-2026-midterm-elections/
PS symposium forecasts, 33 seats, 12 of 12, 17 studies	33; 231; 78; 50; 66; 90; 235; 42; 9; 45 to 56; 17	Cambridge University Press release via EurekAlert, Oct. 14, 2026, https://www.eurekalert.org/news-releases/1147571
PS symposium collection, 20 listed items	20	Cambridge Core special collection, https://www.cambridge.org/core/journals/ps-political-science-and-politics/special-collections/forecasting-the-2026-us-midterm-elections
Senate forecasts split		Political Science Now, Oct. 15, 2026, https://politicalsciencenow.com/forecasting-the-2026-us-midterms/
Gallup ends presidential approval after 88 years	88	CNN Newsource via WNEM, Feb. 13, 2026, https://www.wnem.com/2026/02/13/after-88-years-gallup-will-stop-measuring-presidential-approval-ratings
PollCheck approval	37.9, 59.3, 44 polls	https://www.pollcheck.co.uk/us-midterms-2026/approval, updated Nov. 2, 2026
Bafumi and co-authors 2010 forecast (release names only Bafumi); 1,000 simulations, 435 districts	229, 79	Dartmouth News, Sept. 11, 2010, https://home.dartmouth.edu/news/2010/09/dartmouth-professor-predicts-gop-wins-house-november-elections
Kalshi	94.1, 75	https://news.kalshi.com/p/2026-midterm-predictions, prices at 2:11 p.m. ET, Nov. 2, 2026
Swing-state RCP averages and ratings	table	C:\Users\brian\race-paths\data\rcp_2026-11-02.csv (RCP data connector, retrieved Nov. 2, 2026)
Primary shares, open seats, third-party readings, Noble 47 to 47	table	C:\Users\brian\Desktop\Journalism\keys-2026\reporting\states-AZ-GA-MI-NV.md and states-NC-PA-WI.md, citing the Wikipedia race articles’ state-canvass tables and RCP multi-candidate averages