Monday, October 02, 2006

The "Most Infamous Survey Ever": The 1936 Literary Digest Poll

The "Marketplace" section of today's Wall Street Journal ran a lively and informative piece on the 1936 Literary Digest poll, perhaps the most infamous public opinion survey in history, because it wrongly (and dramatically) predicted a landslide victory for Alf Landon, the Republican presidential nominee:

In the fall of 1936, the most influential poll in America, run by Literary Digest magazine, predicted that Mr. Landon, governor of Kansas, would trounce the Democratic incumbent, Franklin Roosevelt, with 57% of the popular vote. Literary Digest's poll was massive -- it sent out 10 million ballots that year -- and it had correctly forecast five previous presidential elections. There was widespread concern that the poll might create a bandwagon effect, giving Mr. Landon an even bigger victory.

The poll could hardly have been more wrong. Mr. Roosevelt won the election with more than 60% of the popular vote; Mr. Landon carried only two states, Maine and Vermont. The failure of the Literary Digest and other, smaller polls, "will undoubtedly revive agitation for their abolition or control," wrote the New York Times. Kenneth McKellar, a Democratic senator from Tennessee, called the Literary Digest poll "wicked," and promised to sponsor legislation to put polling under federal supervision....

There was at least one man who relished the Literary Digest fiasco: George Gallup. Mr. Gallup, a market researcher and syndicated columnist, had begun experimenting with so-called scientific methods of polling, and in July 1936, he predicted Literary Digest would call the election erroneously. He himself projected a Democratic victory, although his margin turned out to be way off.

Mr. Gallup, of course, was one of the leaders of the pack that predicted Thomas Dewey would beat Harry Truman in 1948. Pollsters' methods and tools -- and confidence -- had improved so much in the dozen years since 1936 that they decided they didn't need to collect more data after October. In a postelection poll by the Survey Research Center of the University of Michigan, 14% of Truman voters said they made up their minds in the two weeks before the election, and 3% said they decided on Election Day.
Stories like this make reading the newspapers fun. I first saw this piece at about 7:00am -- in my office at the college -- when I sat down for a cup of coffee and cracked open my hard-copy of the Journal. As the article notes, the Literary Digest folded shortly after the 1936 election. The history of the Literary Digest fiasco is a staple of introductory lectures on polling methodology, and I usually ask my classes how this poll -- which sampled millions of respondents -- could be so wrong. It's just a powerful example of the principle of "representativeness," the most elemental characteristic of a reliable random sample.

No comments: