Thursday, May 22, 2014

How Good Were the Georgia Polls?

Looking just at the Republican primary race for U.S. Senate (the most fun race of 'em all because it included my kooky congressman), let's see how the polls did compared to the actual voting results.

I'm just going to take two polls, conducted near the election.
  • A last-minute Morris/Fox 5 poll of 852 likely voters had it at (actual in parentheses):
    • Perdue 26 (31)
    • Kingston 17 (26)
    • Handel 17 (22)
    • Gingrey 11 (10)
    • Broun 10 (10)
    • and others making up the rest. In this poll 18 percent undecided
  •  An Insider/Advantage poll of 1,182 likely voters found (again, actual votes in parentheses):
    • Perdue 27 (31)
    • Kingston 19 (26)
    • Handel 17 (22)
    • Broun 10 (10)
    • Gingrey 9 (10)
    • and so on for the rest. In this poll, 18 percent undecided
What can we make of these? First, both polls got the leading vote getter correct. Give 'em points for that. Yes, the polls underestimated the votes of the top three candidates, but that's a function of the undecideds finally, well, deciding, and in this case all breaking more or less for the top three rather equally. Take that top poll as a case study, as it was the last one conducted before the election. Perdue overperformed by 5 percentage points, Kingston by 6, Handel by 5. Do the math and you see how the undecided voters broke when it came time to touch a screen.

What the polls failed to do is discriminate well between the second and third place candidates, Kingston and Handel. That's tough in a low-turnout primary. I'm not saying it can't be done, I'm just saying it's tough to pull off. I would have included questions designed to tap how the undecideds were leaning, for example, and how strong the feelings were for those who had decided. With a little effort, and better sampling techniques rather than robo-calling landlines, you can tease this out more.

No comments: