Can Democrats actually win a Senate race in deep red Tennessee?

Hopes for a victory in Tennessee by Phil Bredesen and Democrats were buoyed after a Middle Tennessee State University poll came out on Thursday showing Bredesen up by a 45 percent to 35 percent margin over Republican Marsha Blackburn. The poll, which used live interviewers and called cell phones, came after an average of other surveys (some of which were higher quality than others) showed the race neck-and-neck.

That's a remarkable result given how Republican leaning Tennessee has become. Still, there are many reasons to be skeptical that Bredesen can actually hold onto a lead and win in November.

I went back to the 2006 election cycle and collected all the Senate polls I could find from the first half of the year (January to June) in which Senate elections were taking place. This includes six cycles and a total of 151 races that were polled.

Perhaps surprisingly, early Senate averages gave us some indication of where each race was heading on average. They were off an average of about 7 percentage points. Put another way, candidates who held large leads in the early going were far more likely than not to go on to win.

What early polls are not good for, though, is figuring out how close races are going to end up. When one candidate holds a single digit lead (as Bredesen does when you look at all the polls conducted this calendar year), such a lead is far from safe. Bredesen losing would be quite consistent with how predictive early Senate averages have been in the past.

Averages errors also tend to underplay the chance of a wide difference between the early polls and the end result. If you were to look at all the Senate polling averages and create a 95% confidence interval between what the averages have now and the end result, it would be +/- 17 percentage points. That's significantly larger than when you look at the difference between the polls and the results in the final three weeks of the campaign.

Of course, the mere chance of a wide error doesn't mean that the error will benefit one candidate or the other. With that in mind, I decided to take into account the presidential tilt of the state (i.e. how much more Democratic or Republican a state was relative to the nation in the previous presidential election) to see if Republican candidates in red states such as Blackburn were more likely to improve upon their early polling numbers than Democratic candidates in red states or Republican candidates in blue states.

That is, is there a trend for the polling averages to move towards the fundamentals?

It turns out that there is and the relationship seems to be becoming stronger. When Bredesen last won an election in 2006, the partisan tilt of a state on the presidential level gave us no additional information about where a race was heading once the early polls were taken into account. Since then, the tilt has gained explanatory power every year relative to the early polls. If you were to weight how much to take into account from the early polls and the partisan tilt, it went from 11:1 in 2008 to 7:1 in 2010 to 2.2:1 in 2012 to 1.5:1 in 2014 to 1.2:1 in 2016.

In other words, the best prediction in 2016 Senate races was to basically weight the early polls and the partisan tilt of the state basically evenly. That would be very bad news for Bredesen even if you use the best poll for him this year (the Middle Tennessee State University poll) given how red Tennessee is. Such a poll would still suggest that Bredesen would go on to lose by about 10 percentage points.

The sharp increase in explanatory power of the partisan tilt of a state relative to the early polls mirrors the sharp increase of the partisan tilt in explaining Senate election races overall. For a long time, Senate races moved mostly independently of how voters in a given state felt about presidential politics. That changed dramatically during the presidency of Barack Obama and correlation between Senate and presidential voting hit an all-time high in 2016, when every state who voted for Hillary Clinton elected a Democratic senator and every state that voted for Trump elected a Republican senator.

All this said, I wouldn't dismiss Bredesen's chances so easily for a number of reasons.

The first is that even taking into account the presidential tilt of the state (in addition to the polling) in no way makes a perfect prediction. Candidates and campaigns still do matter. There are a number of red state Democrats who won election in 2012, when the presidential tilt of a state was important in forecasting a result (albeit less than it was in 2016). Bredesen could be one of those victorious red state Democrats this year.

The second reason is that we sometimes have a bias toward very recent history in predicting. Just because the presidential tilt of a state has become more and more important in explaining Senate results doesn't mean the trend will continue. We've already seen in special elections so far this cycle that Democrats have been mostly outperforming the presidential tilt in red states. Without Barack Obama in the White House, we could see the presidential tilt matter less than it did in the most recent years.

The third reason is that Trump just isn't that popular in Tennessee. His approval rating in the Middle Tennessee State University poll was just 50%, a far cry from the 61% of the vote he received in the 2016 election. While Bredesen's early lead may be buoyed by Trump's low approval rating relative to his share of the vote in 2016, it may not be, and the chance of a reversion to the presidential tilt in the previous election may not be as high as it would be in a neutral political environment.

All that being said, if there is a chance of a Democrat winning a Senate seat in Tennessee, this is the environment in which to do it.