The Death of Polling as We Know It

5%. That's how many Americans now answer when Gallup calls for a poll. In the late 1990s, it was 28%. Last week, the legendary polling firm announced it would no longer conduct presidential approval surveys, marking the end of an era that began with George Gallup's revolutionary 1936 prediction.

The Trump Factor

Suspicions immediately swirled around Gallup's timing. Trump's approval rating in their December poll stood at just 36%—well below the RealClearPolitics average of 42%. The president has shown a vindictive streak toward unfavorable polls, suing The Des Moines Register and its pollster Ann Selzer over a survey suggesting he might lose Iowa to Kamala Harris. When CBS and ABC faced similar lawsuits, both paid hefty settlements to Trump's presidential library fund.

But Gallup's explanation—citing changes in business strategy—reveals a more fundamental crisis. The economics of quality polling have simply collapsed.

From Revolution to Extinction

George Gallup made his name in 1936 by defeating The Literary Digest, the era's polling giant. The magazine mailed postcards to 10 million Americans and received 2.5 million responses, predicting a landslide victory for Republican Alf Landon. Gallup surveyed just 50,000 people but correctly predicted Roosevelt's win.

The difference was representativeness. The Literary Digest drew addresses from automobile registrations and phone directories—in Depression-era America, car and phone owners skewed heavily Republican. Gallup's smaller but more representative sample captured the broader electorate.

Now the tables have turned against Gallup's methods. To build a 1,000-person sample in the 1990s required about 3,500 calls. Today? 20,000 calls. Caller ID and spam-blocking have made Americans gatekeepers of their own attention.

The 95% Problem

The cost explosion is just the beginning. The deeper question haunts the industry: Are the 5% who still answer polls fundamentally different from the 95% who don't? Courtney Kennedy at Pew Research Center told me they've discovered survey respondents claim to volunteer for civic causes at much higher rates than Americans generally do. To compensate, Pew now overweights responses from people who say they don't volunteer.

These statistical gymnastics cost money and raise philosophical questions. If polling requires increasingly complex adjustments to approximate reality, what does that say about the method itself?

The 2016 Earthquake

The industry's credibility cracked in 2016. Election-eve polls showed Hillary Clinton leading by 3.2 percentage points. Nate Silver'sFiveThirtyEight projected 302 electoral votes for Clinton. She won just 232.

The miss wasn't random—it was systematic. Trump's supporters, disproportionately distrustful of institutions and less likely to engage with traditional polling methods, simply weren't being measured accurately. The same pattern emerged in the Brexit referendum, where only one poll—conducted online, then considered a less reliable method—correctly predicted the Leave victory.

2024 repeated the pattern. Despite methodological adjustments, polls again underestimated Trump's support. The problem isn't technical—it's sociological.

The Attention Economy's Toll

Meanwhile, the internet democratized poll creation while degrading poll quality. Remember the 1994 survey claiming more young Americans believed in UFOs than Social Security? Methodologically flawed but headline-grabbing, it exemplified how the attention economy rewards bad polls over good ones.

Traditional media partnerships that once sustained quality polling have withered. CNN ended its relationship with Gallup in 2006. Local newspapers and TV stations, traditional funders of state-level polling, have largely disappeared. What remains is often funded by partisan organizations with obvious incentives.

New Methods, Old Problems

Some firms are adapting. Morning Consult recruits massive online samples—sometimes tens of thousands—hoping size compensates for potential bias. But this approach faces its own representativeness challenges, potentially oversampling the digitally engaged.

The fundamental question remains: How do you measure public opinion when a significant segment of the public actively resists measurement—and when that resistance isn't random but concentrated among supporters of certain politicians or movements?

Democracy's Feedback Loop

Gallup's departure symbolizes something larger than business model failure. Polling once represented a compact between citizens and democracy—the idea that ordinary people would spare a few minutes to help measure the national mood. That compact assumed a shared civic culture and basic institutional trust.

In today's polarized America, many citizens view the "dispassionate stranger on the phone" not as a guardian of democratic feedback but as another institutional intrusion to be avoided. The same anti-institutional sentiment that makes Trump supporters harder to poll may be what makes them Trump supporters in the first place.