A History of Polling - 1948 - Dewey Defeats Truman
Perhaps the best known polling blunder in American history is the consensus prediction that Republican Thomas Dewey would defeat Democrat Harry Truman in the 1948 election. It didn't happen, leading to one of the most famous photos in election history - and the hope for all underdog candidates. The only election that compares with this one is our most recent one, the 2016 victory of Donald Trump over Hillary Clinton.
So what happened?
In my last article I wrote of how Gallup's polling had proven itself to be rather robust. However, it was not - nor is it- foolproof.
I encountered a 1948 article in Time Magazine that spoke of him Gallup in the same glowing terms one might read about Nate Silver after the 2012 Presidential Election. However, that same article also referenced how poorly President Truman was doing in Gallup’s opinion polls (Time Magazine, 1948, pp. 21-23).
In 1948, the Gallup Poll was one of three major polls which incorrectly predicted Dewey would defeat Truman (The Oxford Math Center). The Social Science Research Council in 1949 determined four reasons that Gallup and others failed to predict the results. First, they stopped polling two weeks prior to the election, during which there was a shift to Truman. Second, the “quota system”, used to obtain representative samples, was judged to be flawed and skewed to more educated people. Third, the pollsters felt that undecided voters would vote for the candidates in the same proportions as the decided voters (while they decisively broke for Truman). Fourth, the pollsters were unable to predict who would actually vote in the election (Zetterberg, 2004).
These defects highlight some consistent problems in polling. Their first error is probably the most embarrassing of them - just stopping polling two weeks prior to an election. It's the type of mistake that one is unlikely to find today. Despite being a massive underdog, Truman kept campaigning, something that clearly paid off for him.
The "quota system" I mention is tricky. Given how hard it is to obtain a truly random sample, pollsters must instead endeavor to make their samples representative and make adjustments to take into account what they consider who they may have missed, who they may have oversampled. In this case. the polling generally overrepresented the impact of more educated voters. We'll see this problem again when I go over the 2016 presidential election polls.
There's the dread of the undecided voter, a bane of pollsters. In 1948 the pollsters made the assumption that people who were undecided would likely vote in a manner similar to the decided voters. This was not the case at all in 1948 - when election day came, they broke heavily for Truman. This might have been detected had the polling continued in the final weeks of the election.
Finally, pollsters try to predict "likely voters". Certain people are more likely to actually go ahead and vote. This bas been seen in modern elections, where turnout likely helped Obama in his presidential elections and hindered his party in midterm elections. If you predict who will vote wrongly, your polling results can give very erroneous results.
The Oxford Math Center. (n.d.). Famous Statistical Blunders in History. Retrieved from The Oxford Math Center: http://www.oxfordmathcenter.com/drupal7/node/251
Time Magazine. (1948, May 3). The Black & White Beans. Time Magazine, pp. 21-23.
Zetterberg, H. L. (2004). US Election 1948: The First Great Controversy about Polls, Media, and Social Science . Elections, News Media and Public Opinion. Pamplona: WAPOR.