Special Edition: Five Lessons from the 2026 U.S. Midterms

Executive summary

• The result: as of Wednesday morning, CBS News projected that Democrats would win control of both the House and the Senate. Several Senate seats and dozens of House races were still being counted.
• The driver: the election behaved as a referendum. Exit polls showed the president's approval at 38 percent and roughly three in four voters dissatisfied with the direction of the country.
• The texture: cost of living, an unpopular war and candidate quality shaped individual races more than national messaging.
• The lesson: midterms rarely reward the party in power, and campaigns that ignore that structural pull rarely overcome it.

This is a special edition written the morning after the vote. Figures are projections and partial counts; they will change as ballots are certified.

Lesson 1: The midterm is a referendum, and the rule held

Political scientists have long described midterm elections as judgments on the incumbent president rather than choices between two programs. Edward Tufte (1975) showed that midterm losses for the president's party can be explained largely by presidential approval and economic conditions. Morris Fiorina (1981) developed the broader idea of retrospective voting: citizens evaluate the recent performance of those in power and reward or punish accordingly.

The 2026 numbers fit that model closely. CBS News exit polls placed presidential approval at 38 percent and found that about six in ten voters said the president was a factor in their House vote. Before the election, Republicans held the Senate 53–47, counting the two independents who caucus with Democrats, and the House 218–214. Margins that thin rarely survive a referendum environment.

For strategists: no campaign message neutralizes a referendum on its own. The task for candidates of the governing party is to localize the race; the task for the opposition is to keep it national.

Lesson 2: Prices are the most persuasive message

Before Election Day, a Gallup survey found that half of registered voters rated the economy and health care as extremely important to their vote, and 47 percent said the same of the war with Iran. On Election Day, AAA reported a national average gasoline price of $4.15 per gallon, which it described as the highest Election Day average on record.

Voters do not need a policy briefing to feel those numbers. In Maine, where a Senate seat changed hands according to CBS projections, 69 percent of voters in the exit poll said tariffs had harmed the local economy.

For strategists: the most powerful communication in an election year is often not a slogan but a receipt. When daily costs rise, every other message competes with the price at the pump.

Lesson 3: Candidate quality still decides close races

National environments set the tide, but candidates decide how high it reaches. Analysts pointed to Texas, where Democrat James Talarico was projected to defeat Ken Paxton for the Senate after a campaign in which Paxton's past impeachment and corruption allegations remained central. Polling cited by Northeastern University analysts suggested Texas voters trusted Talarico more on corruption.

For strategists: primaries that reward loyalty over electability carry a cost that only appears in November. Opposition research, vulnerability audits and reputation management belong at the beginning of a campaign, not at the end.

Lesson 4: Turnout and enthusiasm were asymmetric

According to the Associated Press, nearly 10 million more people had voted five days before the election than at the same point in 2022. Analysts also noted that Democratic candidates outperformed their 2024 presidential baseline in several areas the president had carried. In referendum elections, the angrier side tends to vote at higher rates, and the pattern appears to have held.

For strategists: enthusiasm is a resource that can be measured. Early-vote data, volunteer activity and small-donor trends are leading indicators that deserve as much attention as polls.

Lesson 5: The count was the test, and it largely passed

In yesterday's column, I argued that election night is not a verdict. This year offered a useful case. Major Senate calls came between late evening and the early morning hours, and about 75 House races were still undetermined at 4 a.m. Eastern time, according to CBS News. Officials extended voting at a handful of North Carolina precincts after disruptions, and local authorities warned voters about unauthorized texts claiming polling places had moved.

Before the results, the president said he would accept them. Some races remain too close to call, and certification will take weeks. The real test of democratic norms now shifts from Election Day to the counting and certification process.

For strategists: prepare the post-election communication before Election Day. Leaders who explain in advance how and when results will be final protect both their credibility and the legitimacy of the outcome.

Closing note

The 2026 midterms will be read as a partisan story, and every party will offer its own interpretation. The more durable lessons are structural. Referendum dynamics are powerful, the economy is felt before it is explained, candidates matter, enthusiasm can be measured, and trust in the count is built before the first ballot is cast. For anyone advising governments, campaigns or companies, these are not American lessons. They are the grammar of modern elections.

References

Fiorina, M. P. (1981). Retrospective voting in American national elections. Yale University Press.

Northeastern University. (2026, November 3). 2026 midterm election results: Who won the key House and Senate races? Northeastern Global News. https://news.northeastern.edu/2026/11/03/midterm-election-results-analysis-2026/

CBS News. (2026, November 3–4). 2026 midterm election results [Live updates]. https://www.cbsnews.com/live-updates/election-day-2026-midterms/

Tufte, E. R. (1975). Determinants of the outcomes of midterm congressional elections. American Political Science Review, 69(3), 812–826. https://doi.org/10.2307/1958391