Two pundits have noted a puzzling discrepancy between the national polls and those in the battleground states.
First, Jay Cost:
On the one hand, Mitt Romney has built a narrow but durable lead in the national polls, averaging around a 1 percent advantage over the last three weeks. This has cheered the hearts of conservatives everywhere.
Yet, liberals retort, Obama has a lead in enough swing states to add up to 270 electoral votes, and that is really what matters.
What to make of this?
For starters, they cannot both be right. If Mitt Romney wins the popular vote by 1 percent or more on Election Day, the odds that he will lose the Electoral College are quite small.
Next, Sean Trende:
Put simply, the national surveys point to a Romney win, while the state polls collectively point to an Obama win. Both can’t be correct.
Both men then go on to number-crunch in order to underline why either the national polls or the state polls are almost certainly wrong. And they both conclude they haven’t a clue which it is. No wonder so many people on both sides are tearing out their hair!
Why are the polls so strange this year? I think it’s because we’re in the midst of a of polling transition. For years the response rate—the number of people who answer polls—has been shrinking, and it may have reached a point where it’s mattering more and more in creating poll variation of some sort. If only 9% of respondents cooperate, versus over 25% just a decade ago, we should become less and less sanguine about whether the respondents can be said to be a random sample.
It is often said that if a sample is large enough, randomness matters less and less. But that’s only true if the sample is truly jumungous, which does not occur with polls. Otherwise, even a very large sample is garbage in/garbage out if it isn’t representative of the group’s composition.
That’s why I think there’s so much weirdness going on with the polls this year. I would love to know if state polls generally have a different methodology than national ones, because such differences could go a long way to explaining the disparities between the two. For example, does one type of poll generally include cell phone users while the other type does not? Does one type of poll use recorded questions and the other type live questioners? Does one type of poll occur at a different time of day than the other type? I don’t even know what some other differences might be, but my guess is that they exist, and they could matter.
I know I get nervous when, in order to think my guy is winning, I have to invalidate the results of too many polls. “That one doesn’t count because the Democratic sample is too large.” That one doesn’t count because the poll is too small.” “That one doesn’t count because the pollster is biased.” And on and on and on. It’s been my experience that the losing side is the one that has to make more of these excuses prior to the election.
This year, though, both sides are making a lot of excuses—just about different things. That’s why my teeth are slowly grinding down.