Disclaimer The opinions expressed herein are my own personal opinions and do not represent my employer's view in anyway.
I don't follow the polls too much... and I think I rarely if ever blog on them. A lot of people right now are pointing to one poll or another as evidence that "it's all over", but I'm not quite sure that's the case. That's not to say that McCain will win... I still think he's going to lose. But the polls are just all screwed up, and there is a lot of good analysis on this right now as to potential reasons.
First, to the evidence that the polls actually are goofy this time around. Real Clear Politics has a couple of good posts talking about the nature of the polls this year. First, they put together a good running average of the current tracking polls, in an effort to smooth out inconsistencies. One of the things they're noticing is that there is a wide variation in the different polls. The standard deviation (which in laymen's terms is a statistical measure showing how far any one poll is from the average) is higher than it was 4 years ago by a sizeable margin (2.7 compared to 1.7). Other polls are showing that the RCP average is outside their margin of error, showing that they disagree significantly with other polls taken during the same period. Of course, none of this actually tells us what the actual results ought to be, but instead tell us that there are probable issues with the polling numbers. In the end, the only poll that matters is the one that takes place election day, but the current polls are not a likely predictor of the actual outcome.
But what could be causing this variation? There are some plausible reasons... many of which point to a tighter race between McCain and Obama, rather than a landslide. The first is that many of the pollsters are over sampling Democratic voters in their polls:
From these numbers, it’s difficult to understand Obama’s 11-point advantage — until you look further into the polls internals and see that the results are based on a respondent sample that was 40 percent Democrat (D), 27 percent Republican (R), and 30 percent Independent (I). Three percent expressed no preference....The problem is: In no recent election has the Democratic Party (or any party) enjoyed such an advantage among the American electorate. In 2004, exit polling data found the electorate to be 37 percent D, 37 percent R and 26 percent I. In the 2006 midterm elections for the House of Representatives the electorate was 38 percent D, 36 percent R, and 26 percent I. In 1996 and 2000, Democrats enjoyed a 4-point edge over Republicans. Given this history, it seems hard to believe that the Democrats have suddenly leapt to a 13-point partisan advantage.
Granted this is from just one poll, but it gives you a glimpse into how sampling can have a very large effect on the outcomes of a poll. I have not taken the time to look at all the polling internals, but a few that I've looked at have not made their demographic sampling easily available either. Another potential problem for some of these polls is when they do the polling:
Pollsters have long known that the demographic makeup of a random telephone sample changes from night to night. Those with families are less likely be available at home during weekend polling. Since that demographic tends to vote more Republican, weekend polling often understate support for the GOP. Polling companies use weighting to attempt to balance unbalanced weekend samples. But the Gallup traditional daily tracking poll doesn’t seem to have balanced the disparity enough....Tracking poll results that didn’t include any weekend polling tend to show John McCain closing the gap, while those conducted over the weekends show a widening Obama lead. Mathematically, it works out to 42% of the variance in the daily poll results is related to how many days of weekend polling were included within the sample.
Many people, especially Obama supporters, are actually concentrating on the Bradley Effect... which based on most of the primaries, is pretty thoroughly debunked. It's a nice story, which plays nicely into preconceived notions of racism... but the reality is that Obama's past poll performance shows that the Bradley Effect was quite possibly the exception, not the rule.
Granted, nobody likes digging into demographic sampling, and weighted averages based on day and time, because that's boring... but these are key issues when you are trying to take polls using small representative samples of the population and then draw conclusions about the population at large. The sample literally makes or breaks the data. Republican internal polling also seems to show some bright spots for McCain (though you should obviously take things with the appopriate grain of salt).
Once again, none of this tells us what Tuesday will actually bring, but it does cast some doubt to the accuracy of the current polls. Given how much people pay attention to the polls, it's important that they be as accurate as possible. If there is too much discrepency between polling and the actual results, people will begin to question the legitamcy of the election, even though the polling may have simply been wrong all along. This serves nobody's best interests... except maybe the loser.
If there is too much discrepency between polling and the actual results, people will begin to question the legitamcy of the election, even though the polling may have simply been wrong all along. This serves nobody's best interests... except maybe the loser.