How this site works
Copper State Count shows three different things: polls, prediction markets and vote counts. They measure different things, and we label each one plainly. We don't make forecasts, and we never call races.
Which polls count
- Entered and checked by a person. We never scrape poll numbers automatically; fake polls circulate online. Each poll is typed in from the pollster's own release, then checked against that source before it's marked verified. Unverified polls are never shown or counted.
- Linked to its source. Every poll in every table links to where it came from.
- Recent. Only polls whose fieldwork ended in the last 45 days.
- One per pollster. If a firm has polled a race more than once, only its most recent poll counts, so no single firm can flood the average.
How the average is weighted
Each poll gets a weight made of four parts, multiplied together. Bigger weight means more influence on the average.
| Factor | Rule | Example |
|---|---|---|
| Recency | Weight halves every 14 days after fieldwork ends | 0d: 100% · 7d: 71% · 14d: 50% · 28d: 25% · 45d: 11% |
| Sample size | Square root of the number of respondents | 900 people counts 1.5× as much as 400, not 2.25× |
| Who was asked | Likely voters 1, registered voters 0.85, all adults 0.7 | Likely-voter polls best match who actually votes |
| Sponsor | Polls paid for by a campaign or party-aligned group count 0.5× | Labeled “half weight” in the poll table |
The average for each candidate is the weighted mean of their share across the counted polls. We do the same for the undecided share. The headline number is the gap between the two leading candidates (the Democrat and the Republican when both are on the ballot; “Yes” and “No” for propositions).
The uncertainty range
Every average comes with a range. The true margin could plausibly be anywhere inside it. When the range includes zero, either side could be ahead, and we say so. It combines three things:
- Ordinary sampling error from each poll's sample size.
- How much the polls disagree with each other.
- A fixed allowance of 2 points for errors all polls can share, like misjudging who turns out.
The range is about a 95% band. It describes uncertainty in the polls, not a probability that anyone wins.
When there isn't enough polling
If a race has fewer than 2 counted polls, we show “Not enough polling” and no average at all. Instead we show context with sources: an independent rating, the 2024 result in that area and voter registration by party. Most legislative races land here. A single poll isn't an average, and we won't present it as one.
Prediction markets
Where Kalshi or Polymarket run a market on a race, we show its price. A price of 60¢ means traders collectively put the chance around 60%. That's an implied probability from people betting money. It's not a poll, and it's not our forecast.
- We pull prices from each exchange's public data about every 10 minutes and keep every snapshot, so the charts show how prices moved.
- The price shown is the midpoint between the best offer to buy and the best offer to sell. If those are more than 10¢ apart, we use the last trade instead.
- We hide markets too thin to mean much: under 1,000 contracts (Kalshi) or $1,000 (Polymarket) traded, or a gap of more than 10¢ between buyers and sellers. Hidden markets are still linked.
- Each market is matched to a candidate by hand. We never guess which market belongs to which race.
- Prices for the candidates in a race don't always add up to exactly 100%; that's normal for exchanges, and we don't adjust them.
Where a race has both a poll average and a liquid market, “Polls vs. markets” shows whether they agree on who's ahead.
Campaign money
For U.S. House races we use the Federal Election Commission's data: total receipts (“raised”), total disbursements (“spent”) and cash on hand for each candidate's campaign committee in the 2025–26 cycle, through the latest report the FEC has processed. For state races we use the Arizona Secretary of State's See The Money data for the same years: income, expenses and cash balance for each candidate committee.
“Outside spending” is independent expenditures: money spent by groups that aren't controlled by a campaign, reported as supporting or opposing a candidate. Very recent spending can take a few days to appear in either source.
We match each candidate to their committee by hand and refresh the figures once a day. Spending can be larger than this cycle's fundraising when a campaign started with money left over from before. Raising more money doesn't mean a candidate is ahead.
Election night
Arizona counts for days. Early ballots returned on Election Day, provisional ballots and ballots needing signature checks are counted after the first results, and margins often move. For each race we show the estimated share of the expected vote counted so far and notes on what's still outstanding.
We never call races. When a race is decided, we show who called it (for example, “Called by AP”) and when. Until then, it says “Counting.” Official results come from the Arizona Secretary of State and county election offices. For state House races, each district elects two members: the top two vote-getters win.
Ratings
Race ratings (Toss-up, Lean, Solid and so on) come from the named outside outlet shown next to each one, currently the Cook Political Report. They aren't ours.
Independence
Copper State Count is an independent, nonpartisan project. It isn't affiliated with any campaign, party, candidate or government office. Candidates appear in the same order, alphabetical by last name, on every page.