United States · 2028 primaries methodology
Each 2028 primary average is support among the polls that included the candidate.
Prediction-market prices are shown beside the average and are never combined with it. The page makes no forecast. Last updated on 11 October 2026.
What the page publishes
- Polling average: for each party's nomination, from national primary polls.
- Market-implied probability: the prices of Polymarket and Kalshi, each platform separately. Prices are never averaged across platforms and never mixed with polls.
- What changed: the change in each polling average over 7 and 30 days.
- Latest 2028 polls: general election matchups and state primary polls, listed one poll question at a time. They are not averaged.
What it does not publish
There is no win probability, no nomination probability derived from polls, no electoral college projection, no state forecast and no candidate profile. There is no combined score and no ranking that mixes polls and markets. The only probabilities on the page are market prices, described as such.
Which polls are counted
- The source is the New York Times polling database. A poll question counts if it is for the 2028 cycle, is a national primary question for the Democratic or Republican nomination, and its fieldwork ended within the last 90 days.
- These answers are removed: don't know, someone else, other, none, none of these, undecided, no opinion, would not vote, skipped, no preference and refused.
- The unit is the poll question, not the poll. When the method was fixed, 84% of polls contained more than one question, up to five.
How a question is rescaled
For each question the shares of the named candidates are added up. If they total less than 50%, the question is discarded. Otherwise each share is multiplied by 100 and divided by that total, so that a candidate on 30% in a question where named candidates total 80% counts as 37.5%. A candidate who was not offered in a question is not counted as having zero.
How the average is weighted
Each question carries a weight, which is the product of the factors below. A candidate's average is the weighted mean of their rescaled shares. Both the weighted total and the total weight run only over the questions that included that candidate.
| Factor | Weight |
|---|---|
| Recency | Halves every 30 days of age, counted from the last day of fieldwork. |
| Sample size | The square root of the sample divided by 1,000, kept between 0.5 and 2.0. A sample over 2,000 counts as 2,000. A missing sample counts as the median. |
| Pollster quality | 0.25 plus 0.75 times a quality score. The score is the New York Times pollster grade divided by 3, between 0 and 1. A pollster with no grade scores 0.50. |
| Population | Registered voters 1.00, likely voters 0.90, adults 0.75. Any other population 0.75. |
| Tracking polls | 0.20. Other polls 1.00. |
| Pollster cap | No pollster carries more than 35% of the total weight of the last 14 days. Any excess is scaled down. |
Polls sponsored by a campaign or party are not given a penalty. No house effect is removed.
Who appears and where
- Each row shows "Appeared in N of M qualifying poll questions". M counts questions, not polls.
- A candidate who appears in at least 50% of qualifying questions is in the main table. Between 50% and 70% the row is marked partial coverage.
- Below 50% the candidate is in a separate limited coverage table and is never ranked against the main table.
- A candidate who appears in only one qualifying question has no polling average. A candidate with a market price and no polling average is shown in the market only table.
How market prices are read
- The price is the middle of the best bid and the best ask. A contract needs both.
- A spread of 5 points or more is flagged and the contract stays in. A spread of 25 points or more excludes it.
- Contracts in about the lowest tenth by trading volume on each platform are excluded. This floor is provisional.
- Unmapped names and contracts such as "other" or "field" are excluded.
- The remaining prices on each platform are divided by their total, so that each platform's candidates sum to 100%. A platform's prices are withdrawn for a party if fewer than 5 contracts remain, or if the total before division is under 50%.
- The two platforms are flagged as diverging when they differ by 3 points or more, or when one is at least 1% and is 1.5 times the other or more. Neither price is adjusted.
What changed
The 7 and 30 day changes compare today's average with the average stored on the day exactly 7 or 30 days earlier. Nothing is recomputed. A dash means there is no stored average for that day, the candidate was not in the main table on that day, or the method version was different.
Latest 2028 polls
General election matchups and state primary polls are listed with their pollster, fieldwork dates, sample and a link. They are shown as published and are not averaged. The method for averaging them has not been set.
How fresh the data is
A fetch counts only if it loads, parses and contains a recent poll or price. A source that is not fresh is labelled, and one that is withdrawn shows a dash.
| Source | Fresh | Stale | Withdrawn |
|---|---|---|---|
| Polls | Under 36 hours | 36 hours to 7 days, labelled | Over 7 days |
| Market prices, each platform | Under 12 hours | 12 to 48 hours, labelled | Over 48 hours |
Sources
Polls: the New York Times polling database, with pollster grades from FiveThirtyEight. Prices: Polymarket and Kalshi, from each platform's public data, twice a day.
Versions
| Version | Date | Change |
|---|---|---|
| MKT-USP-2028-Phase0-v1.6 | 3 August 2026 | Method locked. Polling averages and market prices for the Democratic and Republican nominations. |
| Section addendum v1.1 | 9 October 2026 | Adds the 7 and 30 day change and the list of general election and state primary polls. The averages are computed as in v1.6. |