A magazine that polled 2.3 million Americans still picked the wrong president
The Literary Digest mailed millions of postcards and correctly called five presidential elections in a row, from Woodrow Wilson in 1916 to Franklin Roosevelt in 1932. In 1936 its survey of 2.3 million voters pointed to Alf Landon. Roosevelt won in a landslide, the magazine soon folded, and George Gallup explained why.
Gallup's diagnosis was participation bias. Landon supporters were keener to mail their cards back, and the magazine's lists skewed toward wealthier Americans, who leaned Republican. Gallup, Archibald Crossley and Elmo Roper, using much smaller surveys built on sounder statistics, all called the result correctly, and the episode became a lasting lesson that a huge sample cannot rescue a lopsided one.
Counting preferences in advance goes back to 1824, when two newspapers, one in North Carolina and one in Delaware, reported a tally showing Andrew Jackson ahead of John Quincy Adams by 335 votes to 169. Jackson did win that state and the national popular vote, and such straw votes caught on, though for decades they stayed local, usually limited to a single city. The Digest's 1916 national survey was partly a stunt to boost its circulation.
Scientific polling crossed the Atlantic in September 1938, when Jean Stoetzel, after meeting Gallup, founded IFOP in Paris as Europe's first survey institute. His first political polls in the summer of 1939 tested public feeling about whether France should go to war over Danzig. Gallup's British subsidiary scored a coup in 1945, standing almost alone in predicting Labour's victory while nearly every commentator expected Winston Churchill's Conservatives to win. The Allied occupiers set up survey institutes across the Western zones of Germany in 1947 and 1948 to help guide denazification, and by the 1950s polling had spread to most democracies.
Methods have kept changing, from doorstep interviews and telephone calls to electronic clipboards and the internet. Firms such as YouGov now draw samples from large volunteer panels and weight the results to match the population, unlike open web polls that count whoever turns up. Campaigns often commission a benchmark poll before a candidate even announces, to see which voters are secure, which are lost and which might be won.
Source: Opinion poll
