I have been wondering about a unique factor in this year’s polling data. Since a large fraction of people pollsters call refuse to participate, how do pollsters compensate for different response rates from supporters of different candidates?
For example, Obama supporters are enthusiastic about him and may want to brag about supporting him, therefore his supporters may be more likely to want to be polled. However, McCain supporters may not want to brag about voting for him and therefore refuse to answer a poll.
Let’s do a thought experiment with a little math. Assume the electorate consists of 1,000 people. 500 are going to vote for Mr. Jones and 500 are going to vote for Ms. Smith. Let’s further assume that it’s “cool” to vote Mr. Jones, so his supporters will always agree to be polled. But it’s not “cool” to vote for Ms. Smith, so only 75% of her supporters are willing to talk to a pollster.
If the pollster calls all 1,000 voters, he will get 500 for Jones (100% response rate x 500 voters) and 375 for Smith (75% response rate x 500 voters). If the pollster (wrongly) assumes that supporters of each candidate are equally likely to answer the poll, then he will report his results as 57% for Jones and 43% for Smith even though the two candidates are actually tied.
Given the powerful fear of being accused of racism and the difference in the “coolness” factor between McCain and Obama, I can see a fraction of McCain voters simply refusing to be polled.
I know pollsters weight their results for age, race and party ID, but these would not compensate for “coolness” or fear of being thought racist.
If I’m right, all the polls could be understating support for Sen. McCain, but I’d appreciate comments from people with greater knowledge of polling methodology.