Wednesday, October 15, 2008

Note on Polls Before the Debate

I haven't been posting much on polls lately. Why? Because I don't understand polling this election cycle. Here are two examples.

In the past, on polls far before the election, polling agencies used registered voters (RV) as a way to weight their samples. This was standard practice as it was hard to determine what socioeconomic and political groups would be more motivated by election time. Also, traditionally, polling agencies would switch to likely voters (LV) closer to an election. This correction helps to make more accurate predictions near the end of a campaign.

This year, the polling agencies are not switching to LV. Gallup continues to call the race a done deal for Obama, but they are still reporting RV numbers. Hugh Hewitt has the latest results from Gallup.

Notice the RV spread is 7% but the LV spread is 3%--within typical margin of error. They've even refused to drop their bias and come up with a different LV number. They are basically saying, history tells us one thing but we know in our heart of hearts that this year will be different. That isn't professional behavior from a polling agency.

And remember--it's the 7% number that will make the front page of CNN and MSNBC. The message is clear. The election is over, McCain can't win, don't even bother to vote conservatives.

As a second example, the leading CNN article is currently about how McCain is going to lose five traditional red states to Obama. Specifically in Virginia, they note:
A new CNN/Time Magazine/Opinion Research Corporation survey in Virginia released Wednesday indicates that Sen. Barack Obama holds a 10-point lead over McCain -- 53 percent to 43 percent among likely voters.
This one is even likely voters. But what about the internals of such polls. DJ Drummond has an interesting post at Wizbang. He is discussing SUSA polls, not Gallup, but notice the data on Virginia:
2004 DRI split was 35%/39%/26%,
2006 DRI split was 36%/39%/26%
SUSA in 2008 is using 39%/30%/25%
You are reading that correctly. In 2004 and 2006 (a horrible year for Rebublicans), 39% of voters in Virginia were Republicans. SUSA is assuming that this year, only 30% of voters will be Republicans. They have no statistical evidence for this--they are just saying it is so.

If I were to walk into a Berkeley cafe and ask about political preferences I would get a different result than if I walked into an NRA meeting in West Virginia. The pollsters are effectively doing the former and publishing the results as if it were a national opinion.

I'm still doubtful of a McCain victory in three weeks. But how much of this doubt stems from a constant, sustained effort by media and pollsters to make me believe that the race is over and I should just give up already?

No comments: