Advertisement

SKIP ADVERTISEMENT

Midterm Calculus

Senate Model Update: What if Georgia Goes to a Runoff?

After new waves of battleground polling from both YouGov and Marist pushed Republican chances of winning control of the Senate from 64 percent to 68 percent on Sunday, things inched toward the G.O.P. yet again Tuesday, with Republicans now having a 70 percent chance to win.

Much of the change over the past couple of days can be traced to South Dakota, where it now appears less likely that the Republican candidate, the former governor Mike Rounds, will fall victim to an upset in the three-way contest. Although some earlier polls had shown the independent candidate, the former senator Larry Pressler, securing as much as 32 percent of the vote, the latest figures suggest that Mr. Rounds may be pulling ahead, and our forecast now gives him a 98 percent chance to win next week.

While our forecast in Georgia remains mostly unchanged, with the Democratic candidate Michelle Nunn having a 48 percent chance of beating the Republican David Perdue, it’s how we're getting to that 48 percent that's now changed.

Over the past few weeks, polling has shown Ms. Nunn steadily closing in on Mr. Perdue. Various pundits have cautioned that Ms. Nunn’s chances may be lower than they might appear because of the Libertarian candidate, Amanda Swafford, and Georgia’s requirement that the race go to a runoff if no candidate gets more than 50 percent of the vote.

In order to avoid a runoff, the winning candidate’s margin of victory has to exceed Ms. Swafford’s share of the vote. Given how close the race is (the current forecast has Mr. Perdue winning by one fifth of a percentage point), this could seem doubtful. Libertarians in previous Georgia Senate contests have been remarkably consistent: With the exception of Claude Thomas, who earned only 1.3 percent of the vote in 2002, the Libertarian candidate has drawn between 2 and 4 percent of the vote in every Georgia Senate election since 1992. True to form, Ms. Swafford’s poll numbers have shown her support hovering around the 4 percent mark.

Image
The Libertarian candidate Amanda Swafford, left, could very well send David Perdue and Michelle Nunn into a runoff on Jan. 6 to determine Georgia's Senate race.Credit...David Tulis/Associated Press

The Upshot model now includes an explicit forecast for the likelihood that neither Michelle Nunn nor David Perdue crosses the 50 percent threshold, sending the race to a runoff election in January. As the margin in Georgia has shrunk over the past few weeks, the possibility of a runoff has grown. Our latest forecast puts the chance of a runoff at 66 percent.

From a forecasting perspective, the probability of a runoff is only half the question. The hard part is how we ought to assess the major-party candidates’ chances in a runoff election.

One approach would be to look at previous runoffs in Georgia, an approach that doesn’t give much comfort to the Nunn campaign. Georgia has had five previous statewide runoff elections. There were two in both 1992 and 2008 — each time for senator and for public service commissioner — and one in 2006 for public service commissioner. In all five of those elections, the Democrat lost. Using this information, we could assume that the future will repeat the past exactly, and assign Ms. Nunn a zero percent chance in a runoff.

Of course, we’re not entering this cycle completely blind. We have plenty of information about the two campaigns in the form of general-election polling. We could use these polls by apportioning the Swafford voters among Ms. Nunn and Mr. Perdue according to how we think they would vote in a runoff. But that would require additional assumptions about how the Swafford supporters might vote.

It might be better not to apportion them at all, though as a practical matter this has the same effect on the polling margin as dividing them evenly between the two campaigns. (This is a small example of how making a decision not to make any assumptions often brings with it its own set of assumptions.)

But regardless of how — or whether — we apportion Swafford voters, this approach doesn’t really capture the dynamic at play in a runoff election. The campaigns aren’t merely reapportioning third-party voters between them. They’re drawing votes from an entirely different electorate than the one that comes out to vote in November.

In previous Senate runoffs, turnout varied between 55 and 57 percent of general-election turnout (in 2006, when the runoff was for the public service commissioner alone, turnout declined by a factor of 10). In other words, it’s less about attracting third-party voters to your side and more about voter motivation and get-out-the-vote efforts. It is doubtful that a likely-voter model designed for the general election will accurately reflect the composition of the runoff electorate as well.

The few head-to-head polls that have been released suffer from this same problem. The two-party preferences of the November electorate are next to irrelevant in trying to model the outcome of an election held among a very different set of voters.

The forecasting gets even more complicated. A Georgia Senate runoff election wouldn’t be held until Jan. 6 — three days after the 114th Congress convenes. That timeline is important because it is entirely possible that control of the Senate would still not be decided by then. If it all rested on the outcome of the runoff in Georgia, both parties would be cranking hard to turn out as many voters as possible. Such a situation would undermine any pre-election turnout forecasts.

But hold on! There is yet another twist. It remains very possible that the Georgia governor’s race will go to a runoff as well. If it did, that election would be held in December, raising the very real possibility of voter fatigue.

With so many variables working to dilute and degrade the data we would use to forecast a January runoff, we think it’s not unreasonable to treat the runoff forecast with zero information, and assign each candidate an even chance. (Various principles of statistical reasoning suggest that, in the absence of other information, a prudent choice for your model is one that maximizes the amount of uncertainty — or, in the language of information theory, entropy.)

The practical effect of increasing the amount of uncertainty we have about the outcome of the race is like attaching a weight to our forecast that pulls it back toward 50 percent. Over the next few days, it’s possible that Mr. Perdue or Ms. Nunn will pull into the lead, and this uncertainty will decrease. But all indications are that the race in Georgia will remain close right up until Election Day, and quite possibly, for a few months beyond.

The Upshot provides news, analysis and graphics about politics, policy and everyday life. Follow us on Facebook and Twitter.

Advertisement

SKIP ADVERTISEMENT