For those out there who are made nervous by the Gallup tracking poll and their very low job approval rating for the president, please keep in mind that by their own internals they are oversampling white voters vs the 2008 electorate by 6 percentage points!
Their approval number today is comprised of 37 percent approval among whites and 69 percent approval among nonwhites, which gives a cumulative 43 percent. The only way to reproduce that number is to push the sample of whites to 80%. Update: Crunching the numbers on their most recent sample, it looks like they have the White vote somewhere between 72.6% and 75.8%. This actually looks reasonable, as the 2008 turn out falls within this range, but I would bet that we'll see something closer to the low end of that range or lower.
Presidential Job Approval
In 2008, the electorate was only 74 percent white . . .
The macro numbers, set out over time, tell the story graphically. When Bill Clinton was reelected in 1996, white voters made up 83 percent of the electorate. When Obama was elected four years ago, they accounted for just 74 percent. This fall, whites may make up only 72 percent of the vote.
Update: On review, I do not think the oversampling is as severe as I initially thought. It does appear that the sample is skewed slightly white, but the range is hard to get at. I am going to look into this some more, and also check the female sample rate to see if this changes over time.
If we adjust the Gallup sample for President Obama's approval rating to 72% percent white, then we add 3 percentage points to his approval rating. If their daily tracking number is also using this sample, then he has a fairly strong lead over Romney with a more likely electorate.
Does Gallup really think the fall electorate will be 80% white? Probably not. More likely they are asking a series of questions to potential voters about their interest and enthusiasm and using the results to build their sample. As we near the election, more nonwhites will start being interested and enthusiastic, they will then show up in the sample, and the Gallup numbers will shift. The media will then see a trend and start pondering about what the campaign has done to move the numbers when actually voter preferences really haven't changed.
In 2008, I was amazed at how all the trackers, even Rasmussen, moved to align themselves with Nate Silver's 538 model in the last few weeks. I think what happened here is that the big polling houses "improve" their sample as election day nears. The great service Nate provides is that he is already adjusting their current mistakes, which he calls house effects, to provide the most accurate snapshot of the race from a distance on a few weeks or months. This is really the only trustworthy source for this sort of information, though the tools that allow you to add and drop the polls you like are a lot of fun to play with.
I do also believe that many of the polls are under-sampling women. That is not as clear in the Gallup model, though.
The Huffington Post's Pollster did a great job on this subject recently regarding Gallup's sampling problems.