Looking at the final Iowa pre-election polls released the past week, one thing is clear: Some of ’em will have the wrong order of finish.
In the Democratic race, four estimates have Hillary Clinton numerically ahead (albeit by as little as a single point), three Barack Obama, two John Edwards and one tie. They differ on second and third place, as well. And in the Republican race, five of the final-week polls have Mike Huckabee numerically ahead, three, Mitt Romney.
“Numerically ahead” means quite a few of these “leads” are within sampling error. Still, the races aren’t necessarily close. Some of these polls have it close, others not. “Close” is more of a shorthand way of saying it’s unsettled, and that given their limitations in this kind of low-turnout contest, polls can’t give us a really clear prediction of the outcomes.
While this level of inconsistency is unusual, differences are not unheard of. In the 2004 Democratic caucuses, of the five final-week poll estimates, four got Kerry correctly as winner, but one did not. Two others got second place wrong. Moreover, a lot of them were way off on the candidates’ actual support levels. Four of these five had Kerry between 21 and 26 percent support; he actually got 35 percent.
Similarly, while the final polls in the 2000 Republican caucus got the order of finish (Bush-Forbes) right, they all understated Steve Forbes, putting his support at 12 to 25 percent; he got 30. One overstated Bush by a whopping 14 points.
It doesn’t end in Iowa. In past years nearly all New Hampshire polls have gotten the right winner, but in some races have badly misstated actual support levels. It’s a reminder that pre-primary polls are best used to identify the key issues and candidate attributes and to identify the leading players, rather than to precisely handicap the final outcome.
One of these Iowa polls may well be dead-on, and the pollster who produced it likely will claim uncanny skill. Our recommendation: Don’t buy it.