Former Dollar General CEO David Perdue is credited with winning an upset victory in the Georgia Republican senatorial run-off because all of the public pollsters save one – Wilson Perkins Allen Opinion Research (Perdue’s own pollsters) – never projected Rep. Jack Kingston (R-GA-1) to be trailing.
While 10 post-primary polls were released and nine of them found Kingston ahead, the cumulative result is not necessarily an example of group inaccuracy. Such was the case, however, in Virginia when no survey firm predicted that David Brat would even come close to defeating House Majority Leader Eric Cantor (R-VA-7) in his widely publicized David vs. Goliath campaign, let alone unseat him.
In the Georgia senatorial run-off, it is likely that Kingston was in fact the early post-primary leader because he successfully maneuvered himself to the right of Perdue immediately after the May 20 vote. Additionally, the losing candidates all endorsed him, and the veteran Savannah congressman was the beneficiary of a major multi-million dollar independent expenditure from the US Chamber of Commerce.
Even though they didn’t post Perdue to a lead as Election Day drew nearer, the polls did detect a momentum switch in Perdue’s favor; but none except WPA actually found the eventual winner having an advantage. This is a testament to the difficulty factor in predicting low turnout election results. While momentum can be detected, it is difficult to gauge exactly where the candidates stand in a close race.
Therefore, factors other than a ballot test become more telling. While Kingston had a solid south Georgia base, such importance was mitigated by the fact that Perdue was similarly strong in north Georgia. And, simply put, north Georgia has a great many more Republican run-off voters than does the region south of Interstate 16.
It is also important to fully analyze a poll before coming to a sound ballot test conclusion. Take the example of Marquette University’s just-released survey of the Wisconsin governor’s race. The major take-away is that Gov. Scott Walker (R) maintains only a one-point, 46-45 percent advantage over his Democratic opponent, businesswoman Mary Burke. But, the entire statistical data paint a slightly different picture.
First, the poll is not limited to registered voters and it appears that the entire sampling universe was asked about whom they would support in the governor’s race. Of the 804 sampled respondents, only 720 of them purport to be registered voters. Additionally, 14.4 percent of those questioned say their likelihood of voting in the 2014 midterm election is questionable, at best. Such a sample would likely skew Democratic.
Secondly, women are over-sampled. Of the poll respondents, 53.2 percent are female – almost three full percentage points beyond the Wisconsin population as a whole. Such a sample would likely skew Democratic.
On the question of job approvals, while President Obama is languishing in the low 40s nationwide and in other Wisconsin surveys, here the number of people expressing approval of the job he is doing as president equals those who disapprove. In regard to Gov. Walker, while most of the state polls find his positives to exceed his negatives by a couple of points, on the Marquette poll the number believing he is doing a good job is equal to those who have the opposite assessment. Again, such a sample would likely skew Democratic.
Finally, another point to suggest that Gov. Walker is actually in stronger position than this poll says is the number of people feeling Wisconsin is on the “right track.” As opposed to national surveys that repeatedly show Americans generally believe the country is seriously on the wrong track, the respondents here, by a split of 54-41 percent, feel that Wisconsin’s state policies are largely positive. Therefore, any person who expresses favorable sentiments about state government should be a prime prospect to support the governor.
When looking at the totality of this particular poll, the underlying statistics suggest that Gov. Walker is stronger than merely tying Burke.
Therefore, as we see in both the Georgia and Wisconsin data, the ballot test compilation may not always tell the full and complete story.