Last fall, statistician-journalist Nate Silver soared to celebrity on the dazzling accuracy of his election predictions. But it's worth remembering that modern survey research owes a lot to the presidential polling fiasco of 1936, when the Literary Digest (which had the poll of record at the time) concluded that Alfred Landon would defeat Franklin Roosevelt in a landslide. The magazine mailed its poll to 10 million people, using phone-directory, car-registration, and magazine-subscription lists-and forgetting that many families during the Depression (including lower-income Roosevelt supporters) didn't have phones or cars, or get magazines. Besides this selection bias, there was the nonresponse bias: only 2.4 million responded. That sounds like a lot, but not when compared with those who didn't respond and who may well have had different views. By contrast, a far smaller survey sampled a cross-section of Americans and got the election right. It was conducted by the fledgling American Institute of Public Opinion, founded by George Gallup. "Gallup" soon became a household word.