I had a hunch that polling errors in the 2016 election were regional.

To check, I looked at the past 3 election cycles' polls at Real Clear Politics and poll-based prediction models at 538.com. These are the two most highly respected polling agglomeration sites. I looked up every tight race in these cycles (blowout races were far less accurate).


The results are conclusive. Since 2014 there has been a systematic polling bias in the Midwest that averages 5% in favor of Democrats. The South also has a bias, but it's only slightly outside the margin of error of most polls, and it's concentrated in upland states. The Northeast and West are well within the margin of error.

This contradicts the claims of pollsters (including Nate Silver at 538) that the polls actually did ok in 2016. When you break it down, they've been systematically wrong in the Midwest and greater Appalachia and about even everywhere else.

Here are the results by region for each election & website:


2014 Senate - RCP (a positive number means Republicans outperformed the poll's prediction)
8.4 % Midwest - 2 races
6.9 % South - 5 races
-0.1 % West - 2 races
-2.4 % Northeast - 1 races

2016 President - RCP
5.7 % Midwest - 5 races
2.7 % South - 5 races
2.1 % Northeast - 4 races
-2.4 % West - 5 races

2016 Senate - RCP
5.5 % Midwest - 3 races
3.9 % South - 2 races
1.2 % West - 3 races
1.0 % Northeast - 2 races

2018 Senate - RCP
2.2 % Midwest - 7 races
1.5 % South - 4 races
0.1 % Northeast - 1 races
-2.7 % West - 3 races

2014 Senate - 538
9.1 % Midwest - 2 races
7.8 % South - 5 races
0.4 % West - 2 races
-0.9 % Northeast - 1 races

2016 President - 538
5.7 % Midwest - 5 races
3.1 % Northeast - 4 races
2.2 % South - 5 races
-0.9 % West - 5 races

2016 Senate - 538
5.5 % Midwest - 3 races
2.5 % South - 2 races
1.3 % Northeast - 2 races
1.2 % West - 3 races

2018 Senate - 538
4.2 % Midwest - 7 races
2.7 % South - 4 races
0.9 % Northeast - 1 races
-1.0 % West - 3 races


Average Bias over all polls and elections (a positive number means Republicans outperformed the poll's prediction)
5.0 % Midwest - 34 Predictions
3.9 % South - 32 Predictions
2.8 % USA average
1.4 % Northeast - 16 Predictions
-0.8 % West - 26 Predictions

Number of Predictions that Predicted too high of a vote for Democrats
31 out of 34 - Midwest
28 out of 32 - South
13 out of 16 - Northeast
8 out of 26 - West

Number of Predictions that Predicted too high of a vote for Democrats by at least 3%
24 out of 34 - Midwest
18 out of 32 - South
4 out of 16 - Northeast
0 out of 26 - West

Number of Predictions that Predicted too high of a vote for Republicans by at least 3%
0 out of 34 - Midwest
1 out of 32 - South
0 out of 16 - Northeast
5 out of 26 - West

Here are the average polling errors by state. Again, a positive number means Republicans outperformed the poll's prediction.

11.4 Kansas - 2 Predictions
10.9 Arkansas - 2 Predictions
9.1 Kentucky - 2 Predictions
8.6 South Carolina - 2 Predictions
8.1 Indiana - 4 Predictions
6.5 Iowa - 4 Predictions
5.6 Tennessee - 2 Predictions
4.7 Wisconsin - 6 Predictions
4.3 Virginia - 4 Predictions
4.2 Ohio - 4 Predictions
4.0 Missouri - 4 Predictions
3.7 North Dakota - 2 Predictions
3.6 Maine - 2 Predictions
3.3 Michigan - 4 Predictions
3.2 North Carolina - 6 Predictions
3.2 Pennsylvania - 4 Predictions
3.1 West Virginia - 2 Predictions
3.0 Georgia - 4 Predictions
1.9 Florida - 6 Predictions
1.8 Minnesota - 4 Predictions
1.6 Connecticut - 2 Predictions
0.9 Montana - 2 Predictions
0.5 New Jersey - 2 Predictions
0.4 Arizona - 6 Predictions
0.3 Colorado - 6 Predictions
-0.2 Alaska - 2 Predictions
-0.2 New Hampshire - 6 Predictions
-2.1 Oregon - 2 Predictions
-2.6 Nevada - 6 Predictions
-2.9 New Mexico - 2 Predictions
-3.3 Texas - 2 Predictions