The current poll-industrial complex traces its roots to the 1930s, when George Gallup created it. But in the past few decades, it has become truly dominant within the realms of politics and policy. President Bill Clinton was notoriously fond of using poll data to guide political decisions, but the real efflorescence came with Nate Silver’s creation of FiveThirtyEight. Silver founded the site in 2007, out of frustration with the punditry he heard on TV, which he believed either misread public-opinion polls or ignored them entirely.
David A. Graham: What does Nate Silver know?
Silver predicted Barack Obama’s victory in the 2008 Democratic primary, then in the general election, and again in 2012, and founded a movement. His critique, which fit with the coolly technocratic impulses of the Obama years, also carried the day. Silver and his cohort felt that reporters were botching horse-race coverage because they relied too heavily on anecdotes, many of which were not representative, either because reporters were too cloistered or because they chose incorrectly.
Silver didn’t want reporters to stop going into the field; he just wanted them to contextualize what they heard. “The impulse maybe isn’t bad,” he told me last year. “But, you know, polls are also a way of talking to voters.” He hoped, however, that by combining the polls into averages and creating models of the electorate, he might free up reporters to write other stories. Instead, horse-race journalism simply became more sophisticated and poll-obsessed, and in the process, it became even more central to political coverage—now enhanced by the sheen of quantitative rigor.
The 2016 election was a shock to this new regime. Polls and poll analysts expected a Hillary Clinton victory, and instead Donald Trump won. Amid a popular backlash, the poll clique—pollsters and analysts alike—defended their results and blamed the public for not understanding polling or probabilities. They pointed out that the popular vote had closely tracked national polling on Clinton versus Trump, and while state polls were very consequentially wrong, many were not wrong by all that much.
Some observers, including me, more or less bought that defense. Pollsters closely examined their methods and promised to try to fix problems in 2020, though they noted that polls are never perfect. (Again, the only poll that matters … ) FiveThirtyEight, which smugly boasted that it had given Trump a better chance (roughly 30 percent) than most analysts did in 2016, gave him just a 10 percent chance in 2020. The Economist was even more bullish on Biden. Then came the vote. In every swing state but Arizona, Trump outperformed the FiveThirtyEight polling average. This is not to pick on FiveThirtyEight, which went to unusual lengths to ensure that its averages were accurate, but simply to indicate how far off the polls as a whole were.