By this point in an election year, when polls are coming in by the bucketload late in the evening, you can get a sense for which pollsters are taking fresh samples of public opinion and which are herding toward the conventional wisdom.
J. Ann Selzer, whose firm Selzer & Company conducts the Des Moines Register’s Iowa Poll, is in the former group: She’s always been one to trust what her data is telling her. On Dec. 31, 2007, Selzer’s poll was among the first to show a large lead for Barack Obama in the Iowa Democratic caucuses — most other polls had a small edge for Hillary Clinton. But Selzer’s poll proved to almost exactly match Obama’s margin in the caucuses four days later.
In September this year, Selzer’s poll showed a 6 percentage-point lead for the Republican Joni Enrst in the Iowa Senate race, defying others that had shown the race as a tossup. Although an early October Selzer poll showed the Democrat Bruce Braley narrowing his deficit to just 1 point, the final Des Moines Register poll, released Saturday night, has Ernst up 51-44 — a 7-point advantage. Other polls of Iowa show Ernst ahead, on average, by a percentage point or two.
Of course, the polling average is usually a more reliable indicator than even the best polls. No poll is clairvoyant — even one that consistently rates as among the most accurate in the country, as Selzer’s does. Polls face inherent limitations because of sampling error, and even the best pollsters face challenges in getting voters to talk to them. If Ernst’s 7-point lead in Selzer’s poll represented the equivalent of a point spread, the FiveThirtyEight model would take Braley’s side of it.
(Excerpt) Read more at fivethirtyeight.com ...
