Reverse Engineering the GA 6 Special Election: Runoff Analysis
As I expected, Jon Ossoff ended last Tuesday night’s special election on top of the crowded pack with roughly 48% of the vote. Pundits across the country have proclaimed either: 1) the Ossoff campaign was a failure, since he didn’t win outright; or 2) this was a total blow to Republicans, since the district shouldn’t have been competitive in the first place. Rather than pontificating on the greater impact of a near-win in a surprisingly competitive district, I’m more interested in what the results of the June 20th runoff might look like. To find out, I reverse-engineered our final poll of the 6th district election, roughly matching the poll’s horserace results to the final election results.
Here’s a comparison of the actual results (top) and the re-^weighted poll results (bottom):
A few assumptions and notes:
- The pre-election weights were intentionally conservative. We knew that Democratic turnout would be higher than in past primaries (or even general elections), but it was impossible to quantify this accurately. I ended up using metrics from past competitive primaries, as well as the 2016 general election turnout and current registration figures. The resulting model gave Republicans a distinct advantage in the final poll.
- The post-election ^weights narrowed the focus of our original turnout model. County-level totals (including party) are the only available metrics that we have at the moment, but that’s enough to tell us quite a bit about the makeup of the electorate.
- County is straightforward: I used each county’s turnout as a percentage of total turnout for the adjusted county ^weights. This is the baseline.
- Holding independent voters at a relative constant, I used the total party turnout from the county-level data for party ^weights (roughly D-30, R-32.5, I-36.5)
- Other assumptions for ^weights include slight changes to the age (younger) and gender (more female) estimates.
- As you can see above, Hill’s support remained disproportionately high, ostensibly due to the oldest cohort in the sample. This supports the previous assertion, that the overall turnout was younger than anticipated, but I couldn’t correct this with weighting alone. At any rate, this won’t affect the final analysis.
As it happens, Cobb County formed a larger percentage of the overall turnout than it has in prior elections. I didn’t expect this, given former-Cobb Senator Hill’s relatively poor performance. Meanwhile, Fulton’s share of the electorate dropped slightly, while DeKalb maintained roughly its usual share. This might affect the runoff, given Handel’s base in Fulton County and Ossoff’s in DeKalb.
Looking at the runoff results with the new ^weights applied, we see this:
Those are some great numbers for Ossoff, but they should be taken with a grain of salt for the moment. Looking at the crosstabs, there’s quite a bit of Ossoff support coming from the 45-64 demographic – to the same degree as that of the 30-44 demographic. While this isn’t unbelievable, it’s likely to change as we refine the turnout model for the runoff. Come June 20th, the electorate *might* look significantly different from that of April 18th, and I’ll be making those determinations in the coming weeks.
As we see above, there is one wildcard that might spell serious trouble for Handel: the massive undecided factor. We’ve tested the runoff twice, and in each poll (using both turnout models), there is a disproportionate level of undecided voters in this match-up. This could go two ways: 1) the undecideds move largely to Handel, since the crosstabs show that they are composed of Hill, Gray, and Moody voters; or 2) they stay home. Based on this, and contrary to typical off-year elections (let alone runoffs), it appears that Handel will need drive Republican turnout substantially. This of course assumes that the Ossoff Turnout Machine® continues its momentum into June.
There’s one more interesting change under the new turnout model: Trump approval
This places Trump’s aggregate approval at -3%, a decrease of 12% from the previous turnout model. That’s a huge difference, and it’s safe to say -as it applies to likely GA6 voters- that proximity to Trump is more of a liability than an asset.
Undoubtedly, this runoff will be hard to predict – there are simply too many variables that are either immeasurable or otherwise unknown. We do know that this will be a much tighter race than others have speculated, and we now have a better idea as to what the district thinks of the current administration. Once election day nears, we will release updated numbers of the runoff – with a better crafted turnout model to boot.