Polls vs. Prediction Markets: Who Actually Has Better Data?
tl;dr — the latter because too many top pollsters seem to only fail up.
FAM: I’ve spent fifteen years in this town watching very smart people with very expensive spreadsheets tell me, with total confidence, who was going to win. Hillary. Biden in a landslide. Kamala. You know how that went.
So when people ask whether I trust polls or prediction markets more heading into the midterms, my honest answer is emphatically: not pollsters.
The Incentive Problem in Polling
Here’s the part that doesn’t get said enough: a pollster gets paid whether they’re right or wrong. A campaign or a committee cuts the check, the poll runs, the memo goes out — and the invoice clears regardless of whether that poll had anything to do with what actually happened in November.
Miss badly enough and the worst consequence is an awkward panel at a conference. The same firms that got 2016 and 2020 wrong keep getting hired in 2026. That’s not a data problem, that’s an incentive problem.
Now compare that to a prediction market trader. Nobody pays them for a good guess — they only get paid for a correct one. Put money on the wrong outcome and you eat the loss, in public, immediately, no memo required. Skin in the game does what a paycheck-regardless-of-accuracy contract never will — it forces discipline.
Think of it like weather forecasting versus reading the ocean itself. Pollsters are meteorologists who get their consulting fee whether the storm hits or not. Markets are the fishermen who read the tide with their boat already in the water — get it wrong and they’re the ones who capsize.
To me, “partly informed by polls” is a long way from “as trustworthy as the industry that keeps failing upward.” And that edge polls supposedly hold late in the cycle shrinks fast the further you get from a marquee Senate or governor’s race — which is most of the map in a midterm year.
Where markets actually pull ahead: early in the cycle, in genuinely uncertain races, and — this is the one that matters most for a midterm with hundreds of contests — anywhere polling is thin or nonexistent. Nobody’s commissioning a serious poll for your fifteenth-most-competitive House seat. Somebody’s trading it on Kalshi. A live market price beats a poll that was never taken.
The honest caveats, because I’m not selling you a miracle: raw market prices overvalue longshots and undervalue heavy favorites, so you adjust for that.
Where Polls Can Still Matter
In thin markets — a handful of traders, a few hundred dollars total volume — aren’t really markets, they’re rumors with a price tag. Kalshi’s election contracts have only existed since 2024, so there’s no multi-cycle track record yet the way there is for ye old Iowa Electronic Market.
I’m not pretending markets are flawless. I’m saying their flaws come from thin liquidity, which fixes itself as more people trade — not from a business model that pays out regardless of accuracy, which never fixes itself.
I’ve personally put small money on Kalshi — a handful of tiny trades in April, mostly on the futbol, all for a total of less than twenty dollar:
Full disclosure, because I’d rather you hear it from me: I spoke yesterday with Kalshi about potentially partnering around the midterms. They have money. I have a booming audience as a top reporter in the Washington press corps.
My aim at the time was to get a feel for the tech platforms being discussed in policy circles. Gotta use the tech to cover it well, was my thinking at the time and now.
These days, I login to find newsy questions to ask in Congress, since so many prediction markets on Kalshi come in the form of a question.
Looking toward the midterm election in November, my plan is to pull Kalshi data to parse here and on my YouTube channel. To me, a staunch and well-documented skeptic of all things pollsters, prediction markets seem to be the most honestly-incentivized data source available to us right now.
But before we do that, here’s my pledge, and I mean it: I will never trade on a policy or a political outcome I could plausibly influence on my beat. My job is to ask questions to the most powerful federal stakeholders in the country, not to have a financial stake in the answers. That line doesn’t move, partner or no.
My actual midterms plan: make prediction market data a key lens through which I view and cover competitive Senate, House, and governor’s races this cycle, especially anywhere polling is sparse or the industry’s incentives make me skeptical of a “shock poll.”
One thing for sure: I’m done treating them as gospel from people who get paid the same whether they’re right or catastrophically wrong. Instead, I’m giving prediction markets a chance.
— Pablo