16 November 2006

Phone Polls Work As Claimed

Telephone polls taken shortly before an election are as accurate as they claim to be, and robocalling is slightly more accurate than human querying. Internet polls from Zogby Interactive suck (something I've addressed before in one of their deeply flawed Colorado polls). Averaging multiple phone polls from multiple sources produces the most accurate results.

In short, there are no serious methodological flaws in the polling methods used by any of the major phone pollsters, and averaging polls, both because this increases sample size and because it produces a more robust measure that mutes any particular pollster's minor methodology biases, is particularly accurate.

Pollsters worry a lot about the impact of increasing cell phone use and increasing non-response rates on their accuracy. But, the proof is in the pudding. These worries don't translate into systemic biases or accuracy significantly below the theoretical maximum accuracy of any poll that flows from random sampling error alone.

No comments: