Will Abdul El-Sayed win the 2026 US Senate election in Michigan?

Michigan Senate election forecast

Executive Summary

Abdul El-Sayed has a 79.5% probability of ultimately being elected to Michigan’s open U.S. Senate seat. He is the clear favorite in the November 3, 2026 election, but a Mike Rogers victory remains a meaningful risk.

The forecast rests on El-Sayed’s broad polling advantage and a favorable Democratic political environment. The main threats are a shared polling error—multiple surveys missing the same voters or misjudging turnout—and late Democratic defections following Haley Stevens’s November 1 endorsement of Rogers.

This forecast concerns the final lawful winner, not an election-night projection. Certification and any final binding legal determination before the winner takes office govern resolution.

Key drivers

A polling lead supported by multiple surveys

The nine-survey working set used in the central calculation consistently favored El-Sayed, although surveys disagreed substantially about the size of his advantage:

These are reported voting preferences, not probabilities of winning. Siena’s unusually favorable result increased confidence, but removing it still left El-Sayed ahead in the working polling average.

Michigan’s polling history leaves room for an upset

Agreement among polls does not eliminate shared error. In 2020, Michigan’s final Senate polling average favored Gary Peters by 5.40 percentage points—the difference between the candidates’ support—but his certified margin was 1.69 points. In 2024, Elissa Slotkin’s polling advantage was 2.30 points, compared with a certified margin of 0.34 points.

Errors have also gone the other way: polling substantially understated Gretchen Whitmer’s 2022 gubernatorial victory. The historical evidence therefore supports substantial uncertainty, not a rule that Michigan polls always overstate Democrats.

El-Sayed’s primary polling lead was also substantially overstated. That reinforces caution, although a primary has a different electorate and does not justify transferring the entire miss to the general election.

A Democratic tailwind, with candidate-specific weakness

Marist found Michigan Democrats ahead 52%–46% on its congressional ballot question, which asks which party voters prefer for Congress. The same survey put El-Sayed ahead only 50%–48%.

That combination supports both a favorable Democratic environment and an El-Sayed-specific weakness. Both already appear in direct Senate polling; adding a separate environmental bonus or subtracting the full candidate penalty again would double-count them.

Stevens’s endorsement is important, but its effect is unmeasured

Stevens’s November 1 endorsement gave remaining Democratic voters an unusual public cue to support Rogers. Most polling interviews preceded it, and no adequately sized, wholly post-endorsement measurement was verified.

Its reach is constrained. AP reported that more than 1.8 million people had already voted by October 30. Other Democratic leaders reaffirmed support for El-Sayed, and backlash against Stevens could offset some persuasion. Moreover, earlier public divisions were already known, so the relevant risk is the endorsement’s additional effect—not the entire history of Democratic disunity.

Scenarios

The scenarios focus on the endorsement’s incremental effect. Their relative weights are judgmental, not measured rates of voter switching. Shared polling and turnout errors remain possible in every scenario.

Implications

• An El-Sayed victory is the appropriate planning case, not a settled outcome. Decisions sensitive to the winner should retain a credible Rogers contingency. The forecast does not imply a comfortable winning margin.
• Headlines and participation totals should not substitute for measured voter movement. The endorsement is real; its statewide electoral effect remains uncertain. Early-vote counts limit the reach of late persuasion but do not reveal candidate choices.
• Separate electoral expectations from legal finality. A projection or concession is not the question’s resolution source. A certification delay alone changes timing, not the expected winner.

Signs to watch

Research and methodology

The forecast combined a checked polling dataset, historical error calibration, and explicit assumptions about the late endorsement. Public forecasts and markets were used as cross-checks rather than substitutes for that assessment.

• Current polling: Original releases and tables from YouGov, Siena, Research Co., and Marist were especially important. The nine-survey calculation favored recent interviews, limited the influence of large samples, avoided duplicate survey waves, and retained undecided respondents where reported.
• Historical calibration: Archived RealClearPolling averages, including 2020 Senate and 2024 Senate, were compared with Michigan’s certified results. The audit covered Senate and gubernatorial contests from 2010–2024, with eight usable polling-to-result comparisons. This anchored uncertainty in whole-election errors rather than survey sampling error alone.
• Late developments: AP’s endorsement reporting and Bridge Michigan’s early-voting coverage informed the endorsement scenarios and the limits on their reach.
• Quantitative model: The polling baseline was combined with a conservative allowance for Democratic polling overstatement and the endorsement scenarios. A heavy-tailed error distribution—one that allows occasional large misses—preserved the possibility that multiple polls could be wrong together.
• External checks: Kalshi, Polymarket, and The Hill/DDHQ provided consistency checks. Published race ratings were more cautious, ranging from Toss-up to Lean Democratic. DDHQ incorporates market information, so its forecast and market prices are not independent confirmation. Silver Bulletin’s endorsement discussion also emphasized that the electoral effect was an assumption rather than a measured swing.

Limitations

The most important missing evidence was a credible, wholly post-endorsement measurement. Surveys ending November 1 also contained earlier interviews and could not isolate the announcement’s effect.

Rasmussen’s publication contained a material unresolved conflict: its toplines showed Rogers leading, while its article and demographic table showed El-Sayed leading. It was excluded from the central polling calculation rather than treating either version as definitively correct.

The original final polling average for the 2014 governor’s race could not be verified and was excluded from historical statistics. The remaining historical sample is small and mixes offices, incumbency, polling methods, and election environments. The research also did not establish the exact share of eventual turnout cast before the endorsement. These limitations make the precise probability less certain than its decimal presentation might suggest.