We try to fix election polling (update)

In 2022, the Planet Money team visited the Marist College Polling Center in Poughkeepsie, New York, to learn firsthand why election polling has become so difficult—and to test a possible fix. What they found was an industry in crisis, where fewer than one in twenty people will answer a pollster's call, and where the golden age of telephone polling has given way to a scramble for new methods. Two years later, their experiment is still running, though not quite as they expected.

The three great polling blunders

Barbara Carvalho, director of the Marist poll, walked the reporters through the history of polling's most famous failures. The first came in 1936, when the Literary Digest mailed surveys to millions of subscribers and car owners, predicting Alf Landon would defeat Franklin Roosevelt. George Gallup, using a much smaller but carefully selected sample, correctly predicted FDR's landslide. The lesson: you need a representative sample, not just a huge one.

The second blunder was 1948's "Dewey Defeats Truman" disaster. Gallup's pollsters had been told to find specific types of people—so many housewives, so many farmers, so many businessmen—but they cut corners, fudging their quotas. The fix was random sampling: pick people at random, and the math ensures the sample will be representative.