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

Executive Summary

Abdul El-Sayed has a 72.1% probability of ultimately being elected to Michigan’s U.S. Senate seat. He is the favorite over Mike Rogers in the November 3, 2026 election, but the race is not safe.

As of October 28, El-Sayed holds a modest, broadly supported polling advantage. The main threat is shared polling error—multiple surveys making similar mistakes about who will vote—rather than an isolated bad poll. Michigan’s recent Senate polling history and El-Sayed’s sensitivity to turnout assumptions justify substantial caution.

This forecasts the final lawful election outcome, not an election-night projection. His Democratic nomination is already settled.

Key drivers

1. The polling lead is persistent, but smaller than the strongest headlines

The October 28 FiftyPlusOne polling average put El-Sayed ahead by 3.2 percentage points. The newest Mitchell/CBS survey showed a seven-point lead, and Glengariff showed a 6.3-point lead. Other recent surveys were considerably tighter: Marist showed a two-point advantage and InsiderAdvantage one point.

The favorable late releases strengthen the case for El-Sayed without establishing that he is seven points ahead. Glengariff’s headline also includes respondents who only lean toward a candidate; excluding those leaners reduces its reported advantage. The forecast therefore rests on the modest aggregate lead, not the largest individual result.

2. Michigan polling errors can exceed that lead

Recent Senate averages have overstated Democratic margins. In 2020, the final polling average showed Gary Peters ahead by 5.4 points; his certified margin was 1.69 points. In 2024, Elissa Slotkin’s 2.3-point polling lead became a certified 0.34-point victory over Rogers.

That history does not establish a permanent Republican polling advantage. The broader statewide record includes a large error in the opposite direction in the 2022 governor’s race. It does establish that agreement among polls should not be mistaken for certainty: surveys can share turnout assumptions, response problems and weighting practices.

3. Turnout assumptions materially affect El-Sayed’s position

Marist’s original tables showed El-Sayed leading 52%–45% among registered voters, but only 50%–48% after weighting respondents by their probability of voting. That is a five-point reduction in the margin.

His primary supplies another warning: late surveys showed double-digit leads, while the certified victory margin was only 0.90 points. However, the research did not establish why the primary polls missed or validate claims that youth-turnout overestimation caused it. The appropriate response is to retain uncertainty, not mechanically repeat the primary error in November.

Importantly, the general-election forecast already uses likely-voter results—survey estimates designed to represent people expected to vote. Subtracting Marist’s turnout-related reduction again would count the same warning twice.

4. A favorable Democratic environment does not transfer fully to this candidate

El-Sayed runs behind Democratic gubernatorial nominee Jocelyn Benson in the same surveys. Marist also found a larger Democratic advantage in congressional voting than in the Senate race. Those comparisons argue against adding a separate Democratic-wave bonus to his polling lead: favorable conditions are already reflected in respondents’ choices, while candidate-specific weaknesses remain visible.

5. Money and early voting do not overturn the polling evidence

Federal Election Commission filings showed substantially greater cycle fundraising for El-Sayed, but Rogers had more remaining campaign cash through October 14. Neither was financially unable to compete.

Michigan reported 1,392,170 ballots cast by October 27. Those administrative counts reveal participation, not candidate choices. Changes in voting methods since the previous midterm also weaken simple historical comparisons. Early voting therefore did not justify a directional adjustment.

Scenarios

The reasoning is best understood through three main pathways, rather than separately quantified scenarios.

Implications

• Use an El-Sayed victory as the leading planning assumption, while retaining a substantive Rogers contingency. The forecast supports prioritization, not treating the outcome as settled.
• Focus late attention on electorate composition and corroborated changes in support. Another favorable headline, a fundraising advantage or a rising early-ballot total is less informative than evidence resolving whether the polls have modeled turnout correctly.
• Separate the likely winner from the timing of resolution. A close count, recount or litigation could delay settlement. The question waits for certification and follows any final, binding legal reversal before the winner takes office.

Signs to watch

Research and methodology

The forecast combined an original polling compilation with historical error calibration. It used the latest observation from thirteen polling series, dated surveys by their interview midpoints, gave newer interviews more weight, limited the influence of large samples and discounted explicitly sponsored polls. Sensitivity checks varied the sample-size treatment. The probability conversion retained a broad allowance for errors affecting many polls together, rather than assuming that more surveys eliminate that risk.

The sources with the greatest bearing were:

• Original polling releases and tables: Mitchell/CBS Detroit, Glengariff/WDIV, Marist and MSU/YouGov. These established current support, turnout sensitivity and differences between firm supporters and leaners.
• Historical polling and certified results: RealClearPolling’s historical race pages and the Michigan Department of State election archive. Ten statewide contests across six election cycles were assembled to assess the direction and size of past polling errors, while accounting for races in the same year sharing an election environment.
• Campaign-finance and participation records: FEC candidate filings and the Michigan Department of State’s October 27 voting update. These checked whether resources or observed participation justified overriding the polls; neither did.
• External benchmarks: FiftyPlusOne, Decision Desk HQ/The Hill, Cook Political Report, Sabato’s Crystal Ball, Polymarket and PredictIt. Markets and quantitative forecasts provided cross-checks, while the traditional ratings ranged from Toss Up to Leans Democratic. They were not treated as independent evidence to be mechanically added together; DDHQ itself incorporates market information.

Limitations

• The eventual electorate remains the largest information gap. The research could not obtain a validated primary-polling postmortem or sufficiently comparable demographic early-voting files to establish whether younger and urban participation was being modeled correctly.
• Historical calibration is limited. Six election environments are a small sample, and one large opposite-direction error materially affects the estimated uncertainty. The original final polling average for the 2014 governor’s race could not be directly verified and remained provisionally sourced to a contemporaneous report.
• Some survey details were incomplete or inconsistent. Marist’s relevant raw likely-voter headcount was not separately verified, and published Quantus sample counts conflicted. The weighting checks reduced dependence on those details.
• External comparisons were incomplete. Cutoff-matched Michigan win probabilities could not be verified for Silver Bulletin, G. Elliott Morris or Metaculus, although Morris’s current polling average was available. Comparable executable market depth was also unavailable, limiting what market prices could establish.

The stated probability is 72.1%; the available evidence does not support interpreting its decimal place as precise measurement of electoral uncertainty.