Monday, November 24, 2008

Georgia Senate Runoff: The Polls

Those of you who have followed FHQ throughout the presidential campaign have probably been wondering when we would get back to our roots and look at the polls in the race for Georgia's Senate seat held by Saxby Chambliss. [Well, truth be told our roots are in the primaries -- specifically the frontloading of them -- but most people found their way here because of our coverage of the electoral college throughout 2008.] There's a limited amount of information out there on this race when it comes to polling, but it appears as if Chambliss has added to the lead he held on election day. Recall that the incumbent Republican narrowly missed out on winning the election in the first round, 49.8%-46.8%. But that three point edge has increased by almost 70% in the nearly three weeks since the general election.

But let's take a closer look:

Georgia Senate Runoff Polls (as of Nov. 24)
Public Policy Polling
Research 2000/Daily Kos
Research 2000/Daily Kos

First of all, the four polls that we have access to move chronologically from bottom to top, wi the most recent poll at the top. Importantly, Chambliss is over 50% in three of the four polls. [Better late than never, I suppose. Though, I'd be willing to bet the Senator would have preferred to have gone ahead an cleared that bar three weeks ago.] That's important in a two person race because one of the candidates is beyond that point before the undecideds are factored in, that candidate is in a great position. Recall that we discussed the importance of Obama's position above that 50% threshold with more than a month left in the presidential race. With time so compressed in runoff race, though, being above that point matters.

But let's give these polls the old graduated weighted average treatment. The idea is that the more recent a poll is, the more weight it carries. The most recent poll gets a full weighting while the past polls are discounted based on the midpoint that the poll was in the field. In other words, a poll a week ago means less than a poll that was released a day ago. There is one slight alteration that I've made to the average because of the small number of post-election polls and in recognition of the limited amount of time in which they can be conducted: The weighting on the past polls is left as is instead of being halved as it was in the context of the electoral college (where polls from February were still being included).

The basic result is that Chambliss has gained a point at the expense of Jim Martin. Chambliss averages 50.8% of the support across these four polls to Martin's 45.7%. The average margin between the two is at 5.11 points in Chambliss' direction. And that is an awful lot of ground for Jim Martin to cover in a week. a low turnout runoff election.

...when early voting doesn't appear favorable.

Recent Posts:
Georgia Senate Runoff: Early Voting (Day 5 & The Weekend)

Georgia Senate Runoff: Early Voting (Day 4)

Georgia Senate Runoff: Early Voting (Day 3)


SarahLawrenceScott said...

How did the polls on this race do in the first round? If they predicted Chambliss over 50% and then he didn't get it, I'm not sure we should take these polls as indicating movement from the General.

Josh Putnam said...

You may be right that that perceived shift toward Chambliss from the general to now is just that; that it isn't anything more than statistical noise. On the ground here, though, my gut tells me that there has been some movement toward Chambliss.

As to your other point (A good one, by the way.), the movement in the general election polling was away from Chambliss. If you split the 35 polls into "before October" and "October/November" halves, you see that Chambliss was above 50% consistently (11 out of 15 polls) prior to October but only breaks that threshold two times (out of the remaining twenty polls) during October and November.

Here's the link to the list of polls over at Pollster. I probably should go ahead and move this into the discussion within the body of the post. It is an important distinction to make. Thanks Scott.