Resolved primary market case study

Polymarket Michigan Democratic Primary Odds: What the Market Missed

Abdul El-Sayed won the August 4, 2026 Michigan Democratic US Senate primary, narrowly defeating Haley Stevens while Mallory McMorrow remained on the ballot after withdrawing. The Polymarket winner event is resolved. Its separate margin market illustrates why a correct winner position can coexist with a badly misread victory margin.

This is a useful case study in ballot effects, late information and the danger of evaluating prediction markets only by whether the favorite ultimately won.

The certified primary result

Published election summaries based on certified results report approximately:

CandidateVote shareVotes
Abdul El-Sayed48.4%744,952
Haley Stevens47.6%731,160
Mallory McMorrow4.0%61,516

El-Sayed’s lead over Stevens was about 0.8 percentage points, not a double-digit margin. McMorrow’s presence on the ballot mattered even though she had withdrawn.

For official analysis, use Michigan’s certified August primary dataset. Secondary tables are convenient, but they should link back to the state record and label rounding.

A withdrawn candidate can still receive votes

Ballot printing and withdrawal deadlines can leave a former candidate’s name on the ballot. Some voters may have cast absentee ballots before the withdrawal, may not know about it, or may intentionally select that candidate.

Those votes do not automatically transfer. In an ordinary primary count, they remain votes for the named candidate unless election law provides otherwise.

This created a modeling question: where would McMorrow’s supporters have gone in a two-candidate race? The certified election cannot answer that counterfactual. Analysts should not simply reassign all her votes to Stevens or El-Sayed after seeing the result.

A forecast could model multiple scenarios, but the market settled on actual certified votes.

Winner odds and margin odds measured different uncertainty

The winner event asked which candidate would receive the nomination. The margin event asked how far the winner would finish ahead, using defined brackets.

Polymarket’s archived margin page now displays “El-Sayed <5%” at 100%, which matches the narrow result. Earlier prices may have favored much larger ranges. Only timestamped historical data can show that movement.

A candidate can have a high chance to win and still have a wide margin distribution:

P(El-Sayed wins) = sum of probabilities for every El-Sayed victory margin

If most of that probability sits between 0 and 5 points, a 70% winner probability does not imply a landslide. Treating the candidate price as a margin forecast discards essential information.

The first rule of a postmortem: freeze time

To evaluate the market, choose a pre-primary cutoff and use only evidence available then. Good checkpoints include:

  • before McMorrow’s withdrawal;
  • immediately after the withdrawal;
  • one week before voting;
  • before the first election returns;
  • after partial results, clearly labeled as nowcasting.

At each point, save:

DataWhy it matters
Candidate bid and askExecutable probability range
Margin-bracket order booksDistribution, not just favorite
Volume and depthConfidence in price quality
Poll field datesPrevents using unpublished future data
Withdrawal and ballot statusDefines the actual candidate set
Rules versionProtects against settlement ambiguity

A chart viewed after resolution can contain the full path, but analysis should not select only the part that supports a desired conclusion.

How to calculate the primary margin

Using rounded shares:

48.4% - 47.6% = 0.8 percentage points

Using raw votes:

744,952 - 731,160 = 13,792 votes

The vote difference and percentage-point difference are both valid descriptions, but a market rule will specify one measure. A relative percentage difference would be another number and should not be substituted.

If a bracket boundary is close, calculate from raw certified totals rather than rounded dashboard shares. Confirm whether the denominator includes all candidates, blank ballots or only valid votes in the contest.

Why a primary can be harder than a general election

Primary electorates are smaller and can differ from the party’s general-election coalition. Turnout may depend on local races, absentee voting and campaign organization. Public polling is often sparse.

Additional uncertainty comes from:

  • eligibility rules for participation;
  • late withdrawals and ballot access;
  • endorsements by elected officials and organizations;
  • geographic concentration of candidate support;
  • uneven campaign spending;
  • low-information voters seeing a printed ballot;
  • the absence of straight two-party cues.

Markets can aggregate informed local traders, but they can also be moved by a small number of participants when liquidity is shallow.

Separate prediction from explanation

After the election, it is tempting to invent a neat cause: one endorsement, one issue or one withdrawal “decided” the race. The certified tally proves who received votes; it does not identify every voter’s motive.

A responsible postmortem distinguishes:

  • observed facts: totals, geography, timing, spending records;
  • supported inference: patterns consistent with a hypothesis;
  • speculation: claims without direct evidence.

Market comments and social media can generate hypotheses but are not representative surveys.

Was the market accurate?

Accuracy has several dimensions.

Outcome accuracy

Did the higher-probability candidate win? This is the simplest test but ignores confidence.

Calibration

Across many markets priced near 70%, did about 70% of the outcomes occur? One primary cannot answer this.

Distribution accuracy

Did margin brackets assign meaningful probability near the actual 0–5 point result? This tests more information than the winner alone.

Economic accuracy

Could a trader enter at the cited price and earn a return after spread and costs? A chart midpoint may not have been executable.

Update quality

Did prices react appropriately when a withdrawal, poll or vote count became available? Fast movement is not automatically correct movement.

A market can get the winner right but be poorly calibrated about the margin.

From primary to general election

After El-Sayed became the nominee, the primary market was no longer the right instrument for general-election analysis. The relevant question shifted to El-Sayed versus Republican Mike Rogers and other ballot-qualified candidates.

Do not splice nomination probability into general-election win probability. The pre-primary chance of becoming senator is roughly a chain:

P(win Senate) = P(win primary) × P(win general | nominated)

Once the primary is resolved, the first term becomes 1 for the nominee, and the general-election market contains the remaining uncertainty.

This conditional structure is why a candidate can rise sharply in a Senate-winner market after securing a nomination even without new general-election polling.

Common errors

Treating a withdrawn candidate as zero

If the name remains on the ballot, votes can still be recorded.

Reassigning votes after the fact

Counterfactual transfers require assumptions. They are not official results.

Using rounded shares for a bracket edge

Use raw certified totals and the rule’s denominator.

Calling a 100% archived price a forecast

It reflects the known result. Retrieve pre-election timestamps.

Evaluating only the winner

Margin distribution and price calibration provide stronger tests.

A compact audit workflow

  1. Download the certified primary result from Michigan.
  2. Save the exact Polymarket event and margin rules.
  3. Pull historical prices for all candidate and margin tokens.
  4. Mark the withdrawal and major polling timestamps.
  5. Compare bid/ask, not only midpoint.
  6. Score winner and margin forecasts separately.
  7. Publish assumptions about withdrawn-candidate votes.

This workflow produces a reusable election study rather than a retrospective narrative.

Frequently asked questions

Who won the 2026 Michigan Democratic Senate primary?

Abdul El-Sayed won, narrowly defeating Haley Stevens. Mallory McMorrow received votes despite having withdrawn while remaining on the ballot.

What was the victory margin?

About 0.8 percentage points using rounded shares, or 13,792 votes in the cited certified totals.

Why did a withdrawn candidate receive votes?

Her name remained on the ballot, and some ballots may have been cast before or without awareness of the withdrawal.

Does a correct winner market mean the margin market was accurate?

No. Winner and margin are different forecasts. Evaluate the historical margin distribution separately.

Where can I verify the primary result?

Use the Michigan Department of State’s 2026 August primary election results and certified datasets.

Sources and verification

Risk notice: Historical market prices can be illiquid and are sensitive to timestamp selection. Resolved outcomes do not validate every earlier price or explanation.

Update log

  • November 17, 2026: Added certified-result figures and a separate audit framework for winner and margin markets.
Polymarket Central Elections Desk

Reviewed by Michigan Primary Results Review. Updated November 17, 2026. Risk level: Medium.