We are now one week away from yet another election which is called the “most important of our lifetimes”, and the third election in which Donald Trump is the Republican nominee. Many observers, myself included, are regularly checking polls and even trying to be so bold as to predict outcomes. The first election that featured Trump, 2016, was the greatest upset in American political history since Truman defeated Dewey in 1948. The worst example of average polling being off was Wisconsin, in which Clinton, per the RealClearPolitics polling average, was up by 6.5, but Trump won by 0.7. None of the latest polls had put Trump on top, including the Republican firm Remington Research, which found Clinton at 8 points ahead. In fact, none of the polls from August until Election Day had Trump up. In the case of 1948, however, polling ended two weeks before the election. The 1936 poll by The Literary Digest, however, takes the cake. The magazine The Literary Digest issued a presidential straw poll every election year, and this had been predictive of the winner since 1916. Yet, their 1936 poll projected Republican Alf Landon as the winner with 57% of the vote and 370 electoral votes. As anyone with even a cursory knowledge about American history should know, we have not had a President Landon. Landon actually only won 37% of the vote and 8 electoral votes; only the voters of Maine and Vermont (they were very different states back then!) saw fit to vote out FDR. The Literary Digest had predicted 1916, which merits credit as it was a close race, but the others were landslides. All this, however, begs the question: how did this publication blow an even bigger landslide? Let’s look at their methodology.
The Literary Digest conducted one election straw poll per year, and they used three lists as sources: phone numbers, drivers’ registrations, and country club memberships (Emory Oxford College). For 1936, they contacted 10 million people for their survey, and from this they got 1,293,669 people who supported Landon and 972,897 people who supported Roosevelt. This approach had multiple methodological problems. The first, the conventional story, is that The Literary Digest had failed to account for the class polarization that came with the Roosevelt Administration…Americans in previous elections had voted more similarly based on class. Many working-class Americans in the North voted Republican in the past elections, and while the 1920s prosperity was part of it, they also supported the GOP’s high tariff platform, a mainstay as old as the party’s 1856 platform. However, FDR’s New Deal programs were highly appealing to many Americans going through hard times, while many in the upper strata had the luxury to think more about FDR’s growing political power as a source of peril and his policies were coming greatly out of their pockets. That the wealthy were overrepresented among those who had telephones, vehicle registrations, and country club memberships should go without saying.
Thus, wealthy people were way overrepresented in the straw poll, as they were disproportionately represented in their opposition to FDR. However, subsequent research pointed the finger at a much more important factor, the response rate to the poll. 10 million people had been sent the poll, but only 2.4 million responded! This constitutes a mere 24% response rate, which is sufficiently low to make the poll worthless. Substantially compromising the class narrative is that a majority of Americans who had telephones and vehicle registrations also supported Roosevelt (Lusinchi). Rather, it was people who were opposed to Roosevelt who had much stronger motivation to respond to the poll. This research partially debunked this traditional narrative, holding that the poor response rate to the poll was sufficient to produce the off result and that the overrepresentation was a secondary factor.
The error was so catastrophic that The Literary Digest folded in 1938. But with the demise of the magazine was the elevation of the Gallup poll. George Gallup was one of the pollsters who got 1936 right, and through his polling predicted an FDR win, albeit with 54% of the vote (PBS). Gallup had nonetheless managed to get the correct outcome by polling a representative sample of 3,000 people as opposed to The Literary Digest’s sample of 2.4 million people. This case illustrates the value of solid methodology in polling. We will not see an error of this magnitude in the polling averages of these races, and it is doubtful we will get something like 2016, which I regard as a black swan event.
References
2016 Wisconsin: Trump vs Clinton. RealClearPolling.
Retrieved from
https://www.realclearpolling.com/polls/president/general/2016/wisconsin/trump-vs-clinton
Famous Statistical Blunders in History: Literary Digest, 1936. Emory Oxford College.
Retrieved from
https://mathcenter.oxford.emory.edu/site/math117/historicalBlunders/
George Gallup and the Scientific Opinion Poll. PBS.
Retrieved from
https://www.pbs.org/fmc/segments/progseg7.htm
Landon in a Landslide: The Poll That Changed Polling. History Matters.
Retrieved from
https://historymatters.gmu.edu/d/5168/
Lusinchi, D. (2016, January 4). “President” Landon and the 1936 Literary Digest. Social Science History, 36(1).
Retrieved from
Mad Politics: The Bizarre, Fascinating, and Unknown of American Political History 