There seems to be quite a large discrepancy among the polls from firms that use traditional polling methods (like Gallup etc.) and those that have sought alternative methods (like YouGov, Reuters-Ipsos, RAND). The new methods are largely unproven. What I'm more interested in is whether the old methods are now outdated.
Polling, of course, is an inexact science. We're seeking to understand the views of the broader electorate by polling, at most, a couple thousand people. Even if we do get a representative sample, there will still be a margin of error to account for random noise. And even with the margin of error, one out of 20 polls will be just plain wrong.
Polling is also harder than it used to be. As recently as 5-10 years ago, we'd hear the phone ring in our house and pick it up. It could be the cute girl from work, it could be a debt collector, or it could be Gallup. We hoped for the best, so we'd pick up the phone. These days, if we even have a landline, many of us have caller ID. If we see some random number, or "Restricted," we might just let the phone ring. That's even more true for those of us who are cell phone only users. As a result, there is only a 9% response rate to polling calls.
Even with such a low response rate, the average of these polls usually turns out to be about right. Pollsters are good at weighing data to ensure that the 9% who answer their phones represent the views of the 91% who don't. Some pollsters are certainly better than others in this respect. But such a low response rate means that traditional polling is getting harder and harder to trust.
Enter the innovators. Reuters-Ipsos, YouGov and RAND. The first two do polls of internet users. In some ways, internet users may not be representative of the population as a whole. They're probably slightly wealthier and are more likely to live in a city or the suburbs than a farm. But the question is, these days, are internet users any more self-selecting than people who answer phone calls? I'm not entirely sure. Internet polls have generally been more favorable to President Obama.
RAND has tried something even more radical. They picked a set of voters and ask the same set every day which candidate they support. There are many advantages to this approach. For one, it cuts down on random polling noise. On the other hand, there are disadvantages. For instance, people who know that they are going to be polled every day might be more likely to stay politically engaged.
At this point, we need to hold the darn election to figure out if these new polling methods are more reliable than the traditional methods. But you don't have to be the "Unskew the polls" guy to recognize this large discrepancy and scratch your head.