Donald Trump’s success in securing the Republican nomination should not have surprised anyone. Polling data told us time and again that his message resonated with a substantial number of Republicans. But past elections indicated he had little chance. When data are at odds, analysts have to choose.
This is the dirty little secret of data journalism: Numbers are interpreted through whatever lens a writer puts on them, and sometimes analysts have to choose between different conclusions -- each valid -- based on different sources of data.
In the case of Trump’s impending nomination, what analysts read in the various numbers depends on how much weight they gave to each competing argument. The businessman had (still has) very high unfavorable ratings, and he accumulated delegates more slowly than others who have won the Republican nomination in recent history. Do you lean on that and say he has a challenging path to the nomination? Or do you lean on the horserace polls that continually said he was way ahead of the competition?
Both sources have drawbacks -- as Nate Cohn points out in The New York Times, there haven’t been enough previous primaries with the kind of data we’re looking at to rely completely on past patterns. And horserace polls have considerable challenges, including getting people to answer the polls, different voting restrictions in each state and identifying which respondents are most likely to vote in low-turnout primaries.
In some ways it was fair to trust past patterns more than early primary polling. Polls aren’t meant to be predictive of the eventual outcome, especially months before the elections begin. And national polls especially aren’t predictive when there isn’t a national election, but rather over 50 low-turnout individual state races. Add in the fact that votes don’t even determine who wins the nomination, overly complex and often differing delegate allocation rules do, and you have a recipe for polling disasters.
A series of high-profile horserace polling failures in 2015 didn’t help, either. Globally, polls botched election results in the United Kingdom and Israel, as well as on the Greek austerity referendum. Closer to home, a polling miss on the gubernatorial race in Kentucky was widely publicized. You couldn’t throw a proverbial rock on the internet without hitting a “polling field in crisis” thinkpiece in the fall of 2015.
Polls did show that many Republicans said they wouldn't vote for Trump, but those numbers went down over time. And we know that people often don't mean it when they say they won't vote for someone.
Once we got to January 2016 and no other candidate had come close to Trump in the polls, it might have been time to let go of some of the skepticism. A surprise early February Iowa loss to Sen. Ted Cruz (R-Texas) proved to be the exception, not the rule, and Trump kept polling high. The reality TV personality even surpassed Mitt Romney’s 2012 polling lead.
Instead, as Trump kept winning states and racking up delegates, analysts focused on figuring out the arithmetic that could keep him from getting the nomination. The prospect of a contested convention became the fetish du jour among journalists and politicos alike. It made for some good headlines.
Trump did gain delegates more slowly than his predecessors. He didn’t have any real endorsements until later in March. Most candidates in the modern primary system had moved more quickly than he did. But the “modern primary system” has only been in place since 1972. And only six other Republican nomination fights have happened since then. None of those began with 17 candidates. The past was truly unable to guide us, but the appeal of the data was too strong to ignore.
Perhaps that’s because the past data on delegate accumulation and endorsements fit a much more appealing narrative than the polls did. Trump is a highly undesirable candidate for many people (and organizations such as The Huffington Post -- see the editor's note below as evidence), and it was reasonable to pick the difficult-path-for-Trump narrative out of the data at first.
But since we were still writing pieces on why he wouldn’t win when national polling — and many of the state level polls — had given him a clear lead for 9 months, we probably let those biases take over how we viewed the data too much.
It’s good to have an election every now and then that challenges all of our assumptions about how politics works. It makes us more careful about the future.
Editor’s note: Donald Trump regularly incites political violence and is a serial liar,rampant xenophobe, racist, misogynist