Since FiveThirtyEight’s Senate model launched on Wednesday, the biggest shakeup in the campaign has been Kansas Democrat Chad Taylor’s withdrawal from the race. We don’t know yet what effect that will have on the campaign or control of the Senate. But new polls in other races mostly confirm our sense of where the elections are headed.
Here’s a look at what’s changed, or hasn’t.
CNN / Opinion Research Corp. has Republican Tom Cotton leading Democratic Sen. Mark Pryor 49 percent to 47 percent.
Cotton’s 2 percentage-point lead among likely voters in the CNN survey is consistent with the FiveThirtyEight forecast of a 3-point Cotton victory in November. But there’s important news beyond the headline number.
The poll suggests that Democrats may suffer from a significant turnout gap. Despite being behind among likely voters, Pryor is actually ahead among registered voters 47 percent to 38 percent. That’s a wide, 11-point gap between the likely voter and registered voter results.
Earlier in the election campaign, Pryor got strong results in surveys conducted by high-quality pollsters even though most polls showed him behind in the race. For instance, a poll conducted by the Kaiser Family Foundation for the New York Times in April put Pryor ahead by 10 points, while a poll by Marist College for NBC News a few weeks later had Pryor up by 11. But both results were among registered voters rather than likely ones.
Is such a large gap between registered voters and likely voters plausible? In September 2010, CNN conducted 10 polls in key Senate races. Republicans, on average, gained 5.3 percentage points with likely voters. The median gain was 7 points. Seven of the 10 times, the likely voter result was more accurate. The average error rate of the likely voter results was 4.3 percentage points, while it was 6.4 percentage points for registered voter polls. So if the 2010 midterms are any guide, the likely voter result is probably closer to the truth. Still, there may be some exceptions — the registered voter result was more accurate in Colorado and Nevada — two states where Democrats won despite being behind in most public surveys.
The FiveThirtyEight model uses polls like the CNN survey to help calibrate the adjustment it makes between likely and registered-voter surveys. By default, the model assumes likely-voter surveys are 2.7 percentage points more favorable to Republicans than registered-voter ones (this reflects the average difference between them in midterm elections between 1990 and 2010). But it updates this number as more 2014 data becomes available. CNN’s poll in Arkansas is just one data point — and earlier this week, CNN showed a more modest 3-point gap between registered and likely voters in its poll of Kentucky. Still, there are some initial suggestions that the turnout gap might be higher than the historical average.
Rasmussen Reports has Democratic Sen. Mark Udall up 44 percent to 42 over Republican Cory Gardner.
The race in the Centennial State is too close to call. The FiveThirtyEight forecast has Udall with a 48 percent chance of winning and Gardner at 52 percent.
Udall’s 2-point edge in the Rasmussen Reports survey is nearly identical to his one-point lead in Rasmussen Reports’ last poll in Colorado. It’s one of the better polls that Democrats have seen this week. Nonetheless, Rasmussen has not shown the strong, Republican-leaning “house effect” that it has in past years. Rasmussen also has a relatively low pollster rating, which means that it does not receive that much weight in the model.
Although Udall has held the slightest of leads in the majority of polls in Colorado, a number of those polls were conducted among registered voters. We expect more likely-voter polls to come out soon in Colorado. They will give us a clearer picture of the race.
CNN / Opinion Research Corp. has Republican Sen. Mitch McConnell up 50 percent to 46 percent over Democrat Alison Lundergan Grimes in Kentucky.
Rasmussen Reports has McConnell ahead 46 percent to 41 percent.
The polling coming out of Kentucky has been remarkably consistent. No poll since the beginning of June has shown anything but a McConnell lead. In fact, the last three polls in the race had given McConnell a 4- or 5-point lead. CNN results are obviously no different and their effect on the model is trivial; McConnell’s odds of winning have risen to 78 percent from 77 percent since our model first launched.
What makes CNN results more important is the way the poll was conducted. Almost all of the previous polling in Kentucky had been from automated or Internet polls — not high-quality, live-caller polls like CNN’s. In 2012, Democrats tended to perform better on high-quality surveys (and these polls were more accurate in the end). We haven’t seen any consistent tendency for such polls to favor Democrats this year.
Rasmussen Reports has Republican Bill Cassidy up 44 percent to 41 over Democratic Sen. Mary Landrieu in Louisiana.
Mary Landrieu is, by FiveThirtyEight’s estimate, the most endangered Democratic incumbent in the country; we give Landrieu about 27 percent chance of retaining her seat. The Rasmussen Reports survey has her down 44 percent to 41 percent, the latest bad sign for her. In the prior Rasmussen Reports survey, Landrieu was actually ahead by 3 percentage points. Before that, she had not been ahead in any other poll since early February.
One potential complication is that Louisiana has a jungle primary and runoff. In November, Landrieu will face not only Cassidy, but Republican Rob Maness. If no candidate receives over 50 percent of the vote, then a runoff between the two top finishers will be held in December. Rasmussen Reports was polling the runoff in its most recent survey. Although no poll has been taken for the first round in over two months, prior surveys suggest that a runoff will almost certainly be necessary.
CORRECTION (Sept 6, 8:05 a.m.): An earlier version of this post made reference to Paul Hollis as running in Louisiana’s November primary. Hollis, however, dropped out of the race in July.