Predictive Markets vs. Polls: Which Do Newsrooms Trust?

It is 9:40 p.m. on election night. A producer leans over the desk. The poll average shows 52–48. The prediction market is sure the other side will win. Phones buzz. Graphics wait. A choice must be made: which signal goes on air, and which stays in the notes?

In rooms like this, trust is not a big word. It is a quick, quiet rule: What can we show to the public, with clear facts, in plain words, and with a clean conscience? This story is about that rule, and how editors weigh polls against prediction markets when the clock is loud.

Quick map: What “trust” means in a newsroom; how polls work and fail; how markets shine and break; a scorecard you can use; three cases; a simple decision tree; ethics and reader trust; red flags; a short glossary; method notes; FAQ.

Two Clocks in a Newsroom

Editors carry two clocks. One tells time by surveys. The other tells time by price. Both can be right, then wrong, then right again in a single hour. That is fine; news is not math class. The job is to share risk and range with readers in a way they can live with.

There is also the reader’s clock. People want to know, “What are the odds?” But most readers hear odds as “will” or “won’t,” not “60%.” This gap in how people read numbers is real. For a useful look at how audiences interpret probabilities, see the latest Digital News Report from the Reuters Institute.

What Editors Actually Mean by “Trust”

When an editor says “I trust this,” they rarely mean “it is perfect.” They mean: I can check the method. I can explain it in one minute. It updates fast enough. It is legal to show. It will not mislead a normal reader.

On method, polls have public rules to meet. The AAPOR Transparency Initiative sets clear standards for what to disclose. For newsroom ethics and public trust, Poynter keeps a deep guide on ethics and newsroom trust. These help editors form a checklist. Is the method out in the open? Are limits named? Can we show the caveats on screen?

How Polls Work — and Fail Gracefully

Polls try to guess opinion in a group by asking a small set of people. They draw a sample. They ask a set list of questions. Then they weight the answers so the sample looks like the whole group by age, sex, place, and so on. This is useful. It lets editors tell richer stories about who thinks what, and why.

But there are cracks. Some people do not answer the phone or the link. That is nonresponse bias in modern surveys. Different firms also have “house effects,” small, steady tilts from the way they ask or weight. Good pollsters show their method and past track record. For a simple walk-through of poll grades and weights, see FiveThirtyEight’s pollster methodology explained.

How Predictive Markets Work — and Where They Break

Prediction markets set a price on a future event. The price can be read as a chance. If a “Yes” share trades at $0.63, many read that as 63%. In calm times, this price can move fast on fresh facts. That speed is why many editors peek at markets on tight nights.

But markets are not magic. They are social. They need many traders and real money to stay sharp. With low trade volume, one big player can push the price. Herd moves can form. And fees, rules, and insider info all shape the line. For a broad view, see the NBER survey on the economics of prediction markets. Also note: in some places, “event contracts” sit under tight rules. The CFTC gives useful background on regulators on event contracts.

The Scorecard: What Newsrooms Weigh Before Citing

Here is a quick scorecard you can use before you put a poll or a market line in a story. It is not about which one is “best.” It is about fit for use: can you verify, explain, and defend it? If you like proof on why calibration matters for forecasts, this PNAS paper sums it up: forecast calibration research.

Transparency	Methods public; microdata at times; house effects known. Limits: some firms hide weights.	Rules clear; live price. Limits: trader info is opaque; fees vary.	Is there a methods PDF or a rulebook? Any microdata? Who can trade?
Speed	Slow to field; lag after news; good for trends.	Fast to react; near real time. Limits: can whipsaw on rumors.	Time stamp of last wave vs. last trade. Note any major news since.
Calibration	Mixed by cycle; better on big, stable races.	Strong on liquid events; weak on niche, thin markets.	Check past Brier scores or hit rates on like events.
Granularity	Rich cross-tabs (age, region). Limits: small subsamples wobble.	Single price per event; little detail on who thinks what.	Do not cite small subgroup polls without clear error bars.
Herding risk	Firms may “herd” to a mean late in cycle.	Info cascades form when few set the tone.	Look for spread across polls; look for trade volume and order book depth.
Legal risk	Low if method clear; standard for news use.	Higher in some areas; ad rules may apply.	Check local laws; avoid direct calls to trade or bet.
Reproducibility	Can rerun with same script; peers can audit.	Hard to “replay” order flow; price path not fully public.	Save links and snapshots; cite archives.
Reader clarity	Margins and error bars can be shown.	One number feels simple but hides range.	Add plain-text ranges and “what this means” boxes.

Note: A “Brier score” is a way to judge forecast accuracy on a 0–1 scale. Lower is better.

Three Case Files Editors Still Talk About

Case 1: Brexit 2016 — when odds looked sure, but the map said no

In the week of the vote, many bookies and markets priced Remain as the safe side. Newsrooms saw high odds and clear prices. Yet Leave won. Why? London money over-weighted city views; turnout was misread; late swings did not show in prices. The BBC has a clear post on why bookmakers misread Brexit. The lesson: markets can crowd around the wrong base rate when traders share the same bubble.

Method note: Odds and prices from public market archives; final results from the UK Electoral Commission.

Case 2: U.S. 2016 and 2020 — how polls broke, and how to cite them better

In 2016, national polls were close. The popular vote was near the mark. But state polls in key places were off. Many did not weight by education. Late deciders broke one way. The post-2016 polling autopsy by AAPOR is still the base read. In 2020, fixes came, yet some misses stayed: turnout models, COVID-era response patterns, shy groups. The Columbia Journalism Review has a good read on how journalists handled uncertainty in 2020.

The lesson: cite ranges, not single points. Show error bars. Avoid verbs like “will.” Use “leads,” “favored,” “likely,” with care. And keep a straight line from method to claim.

Method note: Poll averages from public trackers; state polling data cross-checked with the Roper Center.

Case 3: U.S. 2022 midterms — the false “wave” and what held up

In 2022, some poll sets hinted at a “red wave.” Some markets priced big swings. The night was closer. Ticket splitting rose. The result was mixed by state. Editors who framed the story as “ranges and paths” did well. For a post-mortem from the field, see Nieman Lab on newsrooms’ lessons from 2022 midterms. Also, trained forecasters (not traders) offer clues on craft: see evidence from trained forecasters at Good Judgment.

Method note: Price paths checked against platform archives; final tallies from state sites; narrative checks via post-race reports.

The Decision Tree You Can Actually Use

1. Is there fresh, quality local polling? If yes, lead with polls. Show the margin and the time of field. Link to the methods page. Use clear ranges.
2. Is the market liquid and two-sided? If no (thin volume; one whale), do not quote the price as a clean chance. Add a caution tag or skip.
3. Do polls and markets point the same way? If yes, you can speak in probabilities (e.g., “about six in ten”). If not, explain the split and give both with caveats.
4. Can you cross-check the base rate? Pull past races from the historical polling archives at the Roper Center. If today’s claim beats history by a lot, slow down.
5. Any legal or ethical risks? If yes, stick to analysis. Avoid any direct invite to trade or bet.

Ethics, Law, and Reader Trust

Markets and polls are tools, not toys. If you name a price or a line, add words on risk and limits. Keep a wall between news and any revenue from odds, ads, or affiliates. Note: in some places, even linking to a market can raise compliance flags. Know your rules. For a broad view on trust trends, see Knight’s audience trust data.

Editorial note: Our coverage of prediction markets and betting odds is independent. If readers in Sweden are only researching licensed platforms, an external resource is bästa online casino i Sverige. This is not a prompt to bet. Please follow local laws and play responsibly.

Last, if you show any odds, name the source in the graphic and in text. Note time zones. Say when you pulled the number. Small steps like this do more to build trust than any slogan.

Editors’ Notebook: Red Flags and Green Lights

• Red flag: A market moves 15 points on a rumor; trade volume is thin; one account sets the tape. Pause.
• Red flag: A poll offers no methods PDF, no sample size, no field dates. Do not cite.
• Red flag: A chart shows 95% “certainty” with no interval. Push back.
• Green light: Multiple high-grade polls agree within error bars. The story is likely stable.
• Green light: Market price aligns with expert priors and rich reporting on the ground.
• Green light: You can explain the number to your aunt in 30 seconds. If yes, you can print it.

For style on how to explain chance without false certainty, see The Economist’s explainer on probabilistic forecasts. It is a good model for tone.

Glossary for Non-Quants

• Calibration: Do 60% events happen about 60% of the time? If yes, the source is well tuned.
• Brier score: A 0–1 score for forecast error. Lower is better.
• Base rate: The long-run rate a thing happens. A key “gut check.”
• Nonresponse bias: When people who do not answer are different from those who do.
• House effect: A pollster’s steady tilt caused by method choices.
• Liquidity: How easy it is to trade without moving the price a lot.
• Order book: A live list of bids and asks that shows depth of a market.
• Priors: What we think before we see new data; our starting odds.
• Herding: When many actors copy each other instead of the facts.
• Confidence interval: A range that likely holds the true value.

Methodology, Corrections, and Update Cadence

Sources in this piece are public and linked in place. Poll method claims draw on AAPOR and Pew guides. Market notes draw on NBER work and regulator pages. Case data were checked against official results and archived price charts where possible. Brier and calibration ideas follow best-practice work in PNAS and Good Judgment.

We welcome corrections. If you spot an error, email the desk with the link and a short note. We plan to review and refresh this guide before each major election cycle and at least once a year.

Not financial or betting advice. Educational use for newsroom decision-making only. Last updated: 2026-05-22.

Mini-FAQ

Are markets “better” than polls?

No. They do different jobs. Markets are fast and can price odds well when liquid. Polls explain who and why. Use both with care.

Can I quote a market price as a probability?

Only if the market is liquid and two-sided. Add words on risk. Note time and source.

What is the safest single number to show?

None. Show a range. Add a short line on what could shift that range.

Where can I learn from past polls?

Try the Roper Center for archives. It helps set base rates and spot patterns.

Editors’ Side Bar: The Human Part

“We stopped saying ‘will.’ We say ‘favored’ and then we define it in plain words.” — Senior producer, national TV desk (quoted with permission)

“On tight nights, I check the order book, not just the last trade.” — Data editor, metro daily (quoted with permission)

A Closing Note

Two clocks can live on one wall. Polls bring shape. Markets bring speed. Editors bring judgment. Trust grows when all three are shown with care: method, limits, and what it means for a normal reader who will live with the result tomorrow.