Political polling occupies a peculiar position in democratic societies: simultaneously essential to understanding public sentiment and spectacularly unreliable at predicting outcomes. The industry that generates these numbers operates according to a set of practices that would alarm anyone who assumes polls represent straightforward measurements of public opinion.
The response rate catastrophe
The foundational crisis of modern polling is that almost nobody answers pollsters anymore. Response rates for telephone surveys have plummeted from around 36% in the late 1990s to under 6% today. This means pollsters must contact roughly 20 people to get one completed interview. The people willing to spend 15 minutes answering questions about politics are, by definition, unusual.
Pollsters compensate through "weighting" — mathematically adjusting their samples to match demographic targets. If young voters comprise 20% of the electorate but only 10% of survey respondents, pollsters double the weight of each young person's response. This assumes the young people who answer polls think like those who don't, a leap of faith the industry rarely acknowledges publicly.
The herding instinct
Perhaps the industry's most open secret is "herding" — the tendency of pollsters to adjust their results when they diverge too far from the consensus. No pollster wants to publish an outlier that makes them look foolish if wrong. The mechanisms are subtle: tweaking turnout models, adjusting demographic weights, or simply sitting on results until they align better with competitors.
This creates false precision. When multiple polls show similar numbers, it may reflect genuine public opinion — or it may reflect pollsters looking over each other's shoulders. The phenomenon intensifies as elections approach, producing the illusion of a race "tightening" that may simply be methodological convergence.
The business model problem
Polling is expensive. A quality state-level survey costs tens of thousands of dollars. Media organizations, facing their own financial pressures, increasingly rely on cheaper online panels and automated calls. Universities conduct rigorous polls but can't match the volume of commercial operators.
The organizations with the deepest pockets — campaigns and interest groups — commission private polls they rarely release. When they do share results, it's because the numbers serve their narrative. This selection bias warps public perception, as "good" private polls enter public discourse while unfavorable ones remain hidden.
Our take
Polling will never disappear because democracies need some gauge of public opinion, however flawed. But treating polls as scientific measurements rather than rough estimates shaped by methodology, money, and groupthink ensures continued surprise when reality diverges from predictions. The polling industrial complex persists not because it works particularly well, but because the alternative — having no systematic attempt to measure public opinion — feels worse. Understanding how polls actually get made is the first step toward reading them with appropriate skepticism.
