Wednesday, September 10, 2008

What's up with the McCain Bounce?

Gallup Daily tracking poll (Sept 9)

As you can see from Gallup and other polls - John McCain has been polling much better this week. The studies suggest this is because of his VP selection of Gov. Palin and the republic convention that was held last week - and may be short lived.

My problem with the polls:
There are several problems with these daily tracking and political polls in general. The fact that they are robo calling, the lack of calling cell phones, the increased use of caller ID are a few of the many. My biggest problem, however, is with the low response rate or response bias.

The goal of a poll is to try to develop a representative sample to describe the population at interest. In our case this sample of approx 1,000 being polled is supposed to represent the general US voting population. In 2004, Bush received over 62 million votes, while Kerry received over 59 - totaling well over 100 million votes. Organizations like Gallup and Rasmussen
did a good job with their statistics, determining a good sample size and all of that. The problem is their response rate. The Pew research center found that in standard surveys (like the daily tracking poll) the response rate is 27 percent. I would wager that that number is closer to 10-15 percent these days. In Epi you want to get your response rate at least in the 70 percent range.
Gallup and Rasmussen still get their adequate sample size of say 1,000 (can't find exact #) by calling closer to 5K+ homes. They are also able to adjust for confounders such as age, race, sex, etc.

So what can we take from this? Response rate is highly dependent on enthusiasm. An explanation for the McCain bounce would be that voters are more enthusiastic about his candidacy thanks to both the selection of Palin and the convention. These voters would be more willing to now answer their phone and take a few minutes out of their day to respond to the poller's questions.

For example: Lets assume Steve is a conservative leaning independent who plans to vote for McCain. Before this past week he would have maybe just ignored the poller's call, not wanting to talk politics. The poller would then go onto the next more enthusiastic caller Andrew who would answer their questions - possibly a liberal leaning independent who is very unhappy with bush. Now this week people like Steve are more willing to talk - and therefore the poller wouldn't reach people like Andrew once their sample size has become adequate. The factor has been seen in the British election and is called the Shy Tary Factor.

The democrats saw a smaller bump after the Biden announcement and their convention possibly because their base voters were all ready very enthusiastic. Also because the events all kind of overlapped in a short period of time. The real question is - would unenthusiastic Steve who will not answer the polling question still end up voting on Nov. 4th? I think most will.

The real statistic that would be most helpful to test this hypothesis is the response rate for each day of tracking and for each group - republicans, democrats, and independents.

I think the bounce will eventually go down, but I don't think it matters. My advice: Ignore daily tracking polls until they discuss and alleviate some of the response bias problems.

(see 538 for Nate Silver's thoughts on the subject)

No comments: