A Dissmarket newsletter for Substack — May 2026
Something broke in how we understand the world, and most of us haven't noticed yet.
I don't mean this in a dramatic, hand-wringing way. I mean it literally. The systems we built to tell us what's happening — polls, media, expert analysis — were designed for a slower world. A world where a news cycle lasted days, where trust in institutions was high, and where the gap between an event and our understanding of it could be bridged by a well-resourced newsroom or a carefully designed survey.
That world is gone. And we haven't replaced it with anything adequate.
Here's a number that should unsettle you: in the 1990s, telephone surveys routinely achieved response rates of 30–40%. Today, the Pew Research Center reports response rates around 6%. Some estimates put it lower.
Think about what that means. Ninety-four per cent of the people a pollster tries to reach simply don't pick up, don't respond, or decline to participate. The 6% who do respond are not a random slice of the population. They are a self-selecting group — people who have the time, the inclination, and the temperament to answer questions from a stranger. The entire edifice of public opinion research rests on the assumption that this 6% can be weighted and adjusted to represent the other 94%. That's a heroic assumption, and it's getting harder to defend.
The 2016 and 2024 US elections exposed what researchers call the "shy voter" problem — people giving pollsters the answer they think is socially acceptable rather than the one they actually believe. But the shy voter is just the most visible symptom of a deeper structural issue. When 94% of your sample refuses to participate, you're not measuring public opinion. You're measuring the opinion of people willing to be measured.
In 2020, while the polling industry was recalibrating after yet another miss, something interesting happened in prediction markets.
These platforms — where participants trade shares priced by the probability of future events — didn't get everything right. No system does. But they did something polls fundamentally cannot: they updated in real time. When vote counts shifted, market prices moved within minutes. When new information emerged, the signal adjusted. There was no two-week field period, no weighting model, no methodological opacity.
The academic evidence supports this. Research dating back to the Iowa Electronic Markets in the 1980s consistently shows that prediction markets outperform polls in forecasting election outcomes. A well-known 2004 study found that market prices are well-calibrated probability estimates — when a market prices an event at 70%, it tends to happen about 70% of the time.
By 2025, the prediction market industry had surged past $60 billion in trading volume. Major media outlets were citing market odds alongside polling data as a matter of course.
But here's the thing: prediction markets have their own blind spots.
Current prediction market platforms attract a narrow demographic — predominantly male, tech-savvy, financially literate, and geographically concentrated. The "wisdom of crowds" is only as good as the crowd itself, and when that crowd is a self-selecting slice of the population, the signal carries the same fundamental limitation as a poll with a 6% response rate — just a different slice.
Markets can also be manipulated, especially thin ones. And they're better at pricing factual outcomes (who wins the election) than measuring attitudes and values (why people voted the way they did).
So we're left with two imperfect tools: polls that capture broad sentiment but struggle with accuracy and timeliness, and markets that price outcomes efficiently but reflect the views of a narrow, unrepresentative group.