I have written before about how the polls were all over the place before the 2012 election. Some of them even had the election as a dead heat, or gave the advantage to Romney goinga into election day. But this difference being shown in the polls never seemed to bother the Obama campaign. They continued to be confident that they would win, and do it by a pretty good margin.Joshua Green of Bloomberg Businessweek recently wrote a good article in which he talked to the people running the Obama polling effort. They gave him the information their polls had in the weeks before the election. He used that information to make the chart above. That chart compares the Obama internal polls (the black line) with the Gallup Poll (the green line). Note that the Obama internal polls were very accurate in predicting the election, while Gallup pretty well embarrassed themselves with their prediction.
How could they be so different? The difference is in guessing who would actually go to the polls. Gallup, a few weeks before the election, went from using information from registered voters to using information from likely voters. In other words, they weighted their poll responses to favor those who they thought were most likely to vote. And since they thought the 2008 election was an outlier, and the actual electorate would look more like 2004 and previous elections. This means they undercounted minorities and young people, and possibly some other groups.
But the Obama campaign trusted what their people on the ground were telling them (and they had a very good connection to those GOTV people). So they didn't alter the numbers they were getting to make them look like previous elections. They accepted the fact that the 2012 election would look more like 2008 than any other election in the composition of the electorate -- and they were right. Their internal polling was very close to the final numbers.
Now this doesn't mean polls are useless. Some polls got very close to the real outcome (YouGov, Public Policy Polling, Pew Research, Ipsos/Reuters) and some didn't (Democracy Corps, Rasmussen, Gallup, AP/GfK, NPR). What it does mean is that polls get in some very dangerous territory when they try to predict who will vote and who won't (and alter their numbers accordingly). The Obama campaign internal polls were very accurate because that campaign had a great organization, and understood better just who was going to vote in the election.
I hope the polls like Gallup and Rasmussen have learned their lesson, and do a better job in the next election. But I guess we'll have to wait until 2014 to find out.
