Gallup To Investigate Its Election Screw-up

Traditionally, the Gallup Poll has been one of the most reliable polls in this country. They having been at the polling game for a long time, and established a pretty good reputation for accuracy. But they blew it in the last presidential election, and because of that they have somewhat tarnished that stellar reputation. On the day before the election, their survey had Mitt Romney ahead of President Obama by 1 point. That was within the margin of error, but it still predicted the wrong man would win -- and it was the only major poll predicting a Romney win (even the Fox News Poll had the race as a dead heat).

Gallup now says they are going to investigate their polling methods, and make any adjustments necessary to improve their political polling. I think that's a smart move. But one clue to what went wrong can be seen in the chart above. It shows that Gallup was less than a point away from the actual result in their survey of registered voters. That had Obama winning by 3 points (and in the actual results, he won by 3.9 points).

But as elections draw nearer, a lot of polling organizations try to narrow down the registered voter numbers to a "likely voters" number to be more accurate. They use a series of questions to try and determine which of the registered voters they question are actually likely to vote. Unfortunately, when Gallup (and several other organizations) whittled their results down to likely voters they actually got further from the real vote, instead of closer.

The only major organization whose narrowing to likely voters got them closer to the actual outcome was the Pew Research Center (who narrowed it down from 7% in favor of Obama to 3% in favor of the president -- a very good result). Why did Pew's move to likely voters get them closer to the result, while the other organizations got further away? It seems that Pew included more people in their likely voter sample (keeping more of the registered voters in that sample), while the other organizations, including Gallup, was stricter in their likely voter guidelines (keeping less of the registered voters in their sample).

The result was that Gallup (and many others) underrepresented the people who actually voted -- and this was especially true of minorities and young people (both of which voted in larger numbers than many polls thought they would). Obviously, most of the polling organizations (with the exception of Pew) need to do a better job of predicting likely voters (the people who will actually vote).

I still think Gallup does a very good job in most of its polling (and I use their results often on this blog). But they screwed up in the 2012 presidential race by misidentifying likely voters. I hope they can correct this before 2014 (and especially before 2016).