Skip to main content
Early Senate Polls Have Plenty to Tell Us About November

Republican Kentucky Senator Mitch McConnell is fighting for his political life — or is he? The Senate minority leader is nearly tied in the polls with Democratic opponent Alison Lundergan Grimes. But President Obama’s approval in Kentucky is in the low 30s, suggesting the Democratic brand is so unpopular that McConnell could be fine by the time votes are cast. These competing indicators have led to myriad predictions from political analysts. The Cook Political Report rates the race as tossup, the Rothenberg Political Report puts it in the leaning Republican column, and Sabato’s Crystal Ball goes one step further at likely Republican. My colleague Nate Silver’s forecast, meanwhile, says McConnell has a 75 percent chance to win.

Kentucky isn’t the only state with a disparity between what the polls are telling us and the state’s opinion of Democrats. Dems’ other possible Senate pickup opportunity, according to the polls, is in Georgia, a state that didn’t treat Obama kindly in 2012. The polls are also close in Alaska, Arkansas, Louisiana and North Carolina, despite Mitt Romney’s success there in the last presidential election.

More than six months from the midterm elections, current polling and past precedent are competing for our trust. I analyzed which measure is more indicative come November, and it turns out that polls are a more robust metric even though their numbers are still sparse and there’s still so much time remaining before the election. That’s not to say that a president’s approval rating is useless: It can help refine early polls to make them more accurate. This year, when we factor in both, it doesn’t look promising for Democrats in Alaska, Arkansas, Kentucky or Louisiana.

For my analysis, I took a RealClearPolitics-style average of all the polls1 for 107 races since 2006.2 Separately, for comparison, I collected the president’s approval rating in the matching states in the first half of the year.3

Once that was all in place, I produced this chart, which shows that early4 polls do a pretty good job of forecasting the final vote margin.


The average error between the early polls and the final results was 6.4 points.5 For comparison, in the 2012 Senate race, polls taken in the final month before the election still had an average error of 4.8 points.6 Overall, the president’s party’s candidate won 83 percent of the time he or she led in the early poll average and lost 88 percent of the time when he or she trailed.

A president’s approval rating isn’t as strongly tied to the ultimate result. There is a link — just a weaker one. Note how much more scattered the data points are in the chart below compared to in the chart above.7


Of the 58 races in my sample where the president’s statewide approval rating was less than 43 percent (as it is for Obama now, nationally), the president’s party’s candidate lost 72 percent. On the other hand, when a president’s approval rating was greater than 43 percent, the president’s party’s candidate lost 35 percent of races. In other words, a president’s approval rating is far from a perfect predictor of how Senate races will turn out.

But even though polls are better predictors than presidential approval, the latter still has plenty of information to offer. When I combined early horse-race polls and presidential approval, the median error in early-cycle predictions dropped to 4.0 points.8 That’s nearly two points lower than using the polls alone, and the average error dropped about a point to 5.6 points.

The fact that presidential approval ratings do matter in addition to early horse-race polling should worry some Democrats in 2014. Right now, I estimate Obama’s approval rating is somewhere between 30 percent and 33 percent in Alaska, Arkansas and Kentucky.9 In Louisiana, I estimate it to be in the high 30s.

To give you an idea of how the data suggests Obama is going to hurt these candidates, I pulled probabilistic 2014 projections10 from a model that looks at both the polls11 and Obama’s approval rating and one that looks at only the polls. The table below shows the Democratic candidate’s chance of winning. (The states included have had at least one poll taken in 2014.)


Using polls and approval ratings, the current close races in Alaska, Arkansas and Kentucky are estimated to favor Republicans. (For instance the model gives McConnell a 73 percent chance to win, which is quite close to Nate’s estimate.) In Louisiana, Obama’s rating hurts Democratic Senate candidate Mary Landrieu, but the current polling average this year already had her projected to lose. In Georgia and North Carolina, Obama’s approval rating is high enough that it doesn’t hurt the Democrats too much. Democrats Gary Peters in Michigan and Al Franken in Minnesota are likely helped a bit by President Obama. Overall, this simple model puts Democratic losses in the Senate at 6.8 seats, just off the FiveThirtyEight Senate forecast of 5.8 seats, which takes into account other factors including fundraising totals and candidate ideology.

Still, it’s important to emphasize that we’re talking about early numbers. Twenty of the 107 races from 2006 to 2012 in our model have errors of 10 points or greater, even when controlling for both approval and early polls. Ohio Republican Rob Portman beat Democrat Lee Fisher by nearly 17.4 points in 2010, even though early polls were close and Obama’s approval wasn’t bad early in the year. The same goes for Republican Roy Blunt in Missouri in 2010.

So Mitch McConnell’s close race in Kentucky this year is not just about polls, it’s about Obama’s approval rating, too. Conversely, even if Obama’s approval rating is holding Democrats back in the midterms, remember that statewide polls are usually the more predictive metric. Smart political forecasts incorporate both.


  1. Each pollster’s surveys were averaged among themselves. I then averaged a given pollster’s average with the other pollsters’ averages. That is, each pollster that surveyed a given race only counted once in the average of a race’s polls no matter how many times it polled a race. That ensured that a bunch of consistently pro-Democratic or pro-Republican polls wouldn’t skew the average too much.

  2. I chose 2006 because it strikes a balance between collecting a reasonable sample size and accounting for the increasing nationalization of elections in recent years. In the 2012 election, 64 percent of the variation in Senate results was accounted for by the presidential results in a given state. It was just 12 percent from 1972 to 1992. Just 20 years ago, about half of all Democratic senators came from states that were more Republican in presidential elections. Now, it’s less than 30 percent.

  3. For 2006, I took an average of SurveyUSA surveys from January to June. For 2008, I used a uniform swing off the exit polls. (In 2008, President George W. Bush’s approval rating from January to June was three points higher in early polls than in the exit polls.) For 2010 and 2012, I used Gallup’s estimates of President Obama’s approval in each state with a minor uniform swing to account for the change between Obama’s average yearly approval rating and his overall approval rating in the first half of the year.

  4. Early here means an average of all polls from January to June.

  5. The correlation coefficient was 0.94 — 88 percent of the variability in the final results was explained by the early polls, which indicates a good fit.

  6. The average error in 2012 Senate races was 2.4 points per candidate. I’ve multiplied that by two since my calculations were about the margin between the two candidates.

  7. The correlation coefficient was 0.47 — only 22 percent of the final results were explained by early presidential approval ratings, which isn’t a great fit.

  8. Note, out-of-sample forecast errors tend to be higher.

  9. To calculate the president’s approval rating by state, I’ve taken the 2013 Gallup estimates and performed a uniform swing to ensure the president’s national approval rating is 43 percent. This method is not perfect, but many state polls don’t include Obama’s approval rating. There are a number of states such as Georgia in which we don’t have any presidential approval ratings from polls conducted in 2014

  10. I calculated a standard error in forecast from the prior 107 races discussed above to determine these probabilities.

  11. I used the same method for averaging Senate polls as I used in prior years. In cases where there is still a competitive primary, I averaged results across candidates.

Harry Enten is a senior political writer and analyst for FiveThirtyEight.