We’re getting to that point in the presidential campaign — with one candidate leading by a lot — when the losing candidate’s supporters start to bring up the 1948 election — the one with the famous “Dewey Defeats Truman” headline when the polls were supposedly way off. Democrat Harry Truman, of course, defeated Republican Thomas Dewey, and Truman has been the patron saint of candidates trailing in the polls ever since.
It’s a neat little story with a nice moral: Never count the underdog out. But Donald Trump’s supporters would be unwise to look to 1948 for comfort: Trump trails Hillary Clinton by more than Truman trailed Dewey, and the polling landscape in 2016 is much different than it was 68 years ago.
Truman’s upset win in 1948 wasn’t that big of an upset.
The first big difference between 2016 and 1948: We have a lot more pollsters now. Clinton leads Trump by about 7 percentage points nationally in the FiveThirtyEight polls-only forecast. That estimate comes from hundreds of state and national polls, and dozens of pollsters. In October, 1948, in contrast, only four pollsters conducted national surveys — that’s fewer pollsters than released national surveys on Sunday and Monday of this week.
Second, it’s not like the polls were predicting a blowout in 1948. Gallup’s two polls in October both had Dewey ahead of Truman by 5 percentage points. The final Crossley poll also had Dewey up by 5. The University of Chicago’s National Opinion Research Center and the University of Michigan’s Survey Research Center each put Dewey’s margin at just 2 points among those intending to vote. When you average all the pollsters in October 1948, Truman was down by 3.5 percentage points to Dewey. That’s half of Clinton’s current margin. It’s quite possible that if as many pollsters were active in 1948 as are today at least a few would have shown a Truman advantage given the inherent margin of error in polls.
Third, because most pollsters thought Dewey’s lead was safe, they stopped polling long before the election took place. All but the Survey Research Center stopped polling the race by Oct. 25 and all had started their final polls long before then. Had they kept polling, they might not have been caught flat-footed. The Survey Research Center’s polling indicated that undecided voters broke late for Truman. Although their monthlong poll overall had Dewey ahead, the trendline was in Truman’s direction.
The lack of late-stage polling in 1948 makes the polling error that year at least a little more explainable. The final polling average was off by 8 percentage points (Truman won by 4.5 points). That’s a big error as far as final polls go, but it’s not that big for mid-October polls. If pollsters had stopped surveying the 1992 race between the second and third presidential debates, for example, the polls would have been off by more than 8 points. That third 1992 debate was conducted on Oct. 19 or right about the same time that most of the final 1948 polls were finishing up their surveys.
Of course, the biggest difference between 1948 and 2016 may be our polling methods. Not only were all polls conducted face-to-face in 1948 (compared to on the phone or online today), but the way in which participants were selected was different. Most polls (with the Survey Research Center being a notable exception) selected participants using quota sampling. That is, the selection of participants wasn’t random. Pollsters looked only to ensure that each demographic group made up a certain percentage of the survey’s sample. This methodology can be a big problem if one candidate’s voters are more reluctant to find or take a survey. Indeed, Gallup’s quota method led to the final polls underestimating the Democratic share of the presidential vote in 1936, 1940 and 1944. High-quality pollsters today randomly select participants.
Trump might still win this race. But his campaign shouldn’t bank on a massive polling error. Trump trails Clinton by a big enough margin that he’ll probably need a few things to go his way — a polling error (in his favor), a good third debate and maybe unexpectedly good turnout, for example. Otherwise, don’t expect “Clinton Defeats Trump” to become one of the great blunders of our political history.