Prediction markets vs polls: which should you trust?

Every election season the same argument resurfaces: the polls say one thing, the betting markets say another, and each camp insists the other is fooling itself. The disagreement is real, but it is not a contradiction. Polls and prediction markets measure genuinely different things. Understanding what each one is actually telling you is the difference between reading the signal and arguing about the weather.

Two instruments, two questions

A poll is a survey. It samples a population, asks people what they currently think or intend to do, and reports the distribution of those stated opinions with a margin of error. A good poll is a careful snapshot of sentiment at a moment in time. It answers the question: if the vote were held today, and this sample were representative, how would people say they would behave?

A prediction market is a price. Traders buy and sell contracts that pay out if a specific outcome occurs, and the price settles where buyers and sellers agree. Because a contract that pays $1 if an event happens will trade at roughly its probability, the price is a forecast. It answers a different question: weighing everything known right now, how likely is this outcome to actually resolve yes?

The crucial word is everything. A market price can incorporate the polls, but it can also incorporate turnout models, fundraising, the calendar, historical base rates, the tendency of a particular pollster to lean one way, and the collective judgment of people with money at stake. Polls are one input the market digests.

Why they diverge

When a headline poll and a market disagree, it is usually because the market is applying a correction the poll cannot. Common sources of divergence:

• Sampling versus aggregation. A single poll carries real statistical noise; a market has already averaged across many polls and discounted the outliers.
• Stated intention versus likely behavior. People answer surveys about a hypothetical. Markets price what people are likely to actually do, including whether they turn out at all.
• Known biases. If a pollster has historically overstated one side, traders quietly adjust for it. A poll reports its number at face value.
• Timing. Markets move continuously as news breaks. Polls are periodic and always slightly stale by the time they are published.

None of this makes the market automatically right. It makes the market a different, more processed number.

Where polls are strong, and where they fail

Polls have one decisive advantage: they measure the electorate directly. A market can only guess at what voters think; a poll asks them. When the question is genuinely about the distribution of opinion — approval, issue salience, the shape of a primary field — a well-run poll is the primary source, and the market is downstream of it.

Their characteristic failure modes are well documented:

• House effects. Different pollsters, using different weighting and likely-voter screens, systematically produce results that lean in a consistent direction.
• Herding. As an event nears, pollsters become reluctant to publish numbers that stray far from the consensus, artificially compressing the spread and hiding real uncertainty.
• Non-response bias. If the people who decline to answer differ systematically from those who do, no amount of weighting fully recovers the truth.
• Turnout modeling. Turning raw responses into a forecast requires assumptions about who actually votes — the single largest source of polling error in close races.

These are not hypotheticals. The 2015 UK general election and the 2016 US presidential race are the standard cautionary cases: in both, the aggregate of the polls pointed one way and the result landed the other, and post-mortems traced the miss largely to sampling and turnout assumptions rather than a late swing.

Where markets are strong, and where they fail

Markets aggregate dispersed information and update in real time. They impose a discipline polls lack: being wrong costs money, so confident, informed participants can move the price against the crowd. They are also candid about uncertainty in a way herding polls are not — a market at an even split is telling you plainly that it does not know.

But a price is only as good as the market underneath it:

• Thin liquidity. In a market with little money trading, a single large order can move the price without reflecting any new information.
• Longshot bias. Prices near 0% and 100% are notoriously unreliable; traders overpay for tail outcomes and the extremes rarely mean exactly what they say. We unpack this failure mode, and how to screen for it, in can you trust prediction markets.
• Resolution ambiguity. If the criteria for a contract paying out are vague, the price is partly a bet on interpretation, not on the real-world event.
• Reflexivity. Markets can echo the very polls they are meant to improve upon, occasionally amplifying a polling error rather than correcting it.

Using both without fooling yourself

The mature position is not to pick a side but to read each instrument for what it does well. Treat the poll as evidence about opinion and the market as a forecast about outcome. When they agree, your confidence should rise. When they diverge, that gap is information — it usually means the market sees a correction the raw poll cannot express, and your job is to work out what it is.

A poll tells you what a sample said. A market tells you what money thinks will happen. Confusing one for the other is how confident people get surprised.

A few working habits keep you disciplined. Never anchor on a single poll; look at where the market has settled after digesting the run of them. Distrust extreme prices in thin markets. Read the resolution criteria before you read the number. And treat a sudden divergence between the two as a question to investigate, not an argument to win.

On WillThisHappen, the market-implied probabilities we publish are the processed number — polls and everything else, priced into a single live figure. Browse the current forecasts across politics to see where the market and the conventional wisdom part company, and use both to build a view neither could give you alone.