(AP Photo/Stephen B. Morton)
When turnout falls, Democrats perform worse in elections. That general pattern is well known. In making their forecasts, pollsters try to estimate what that turnout will be on the basis of previous elections. This year, pre-election opinion polls were off by the largest amount seen in over 20 years. Could this massive underperformance by Democrats have been connected to a wrong guess about turnout?
Here is a graph of turnout over the history of the United States. The data come from the Vital Statistics of American Politics and were plotted by Michael McDonald. As the graph shows, fewer people vote in midterm elections than in presidential elections-about 30 percent fewer, in the post-Watergate era.
I have added to this graph an arrowhead indicating McDonald's current estimate of 2014 voter turnout: a dismal 36.3 percent of the voting-eligible population (VEP).
This is the lowest rate of turnout since 1942, immediately after the entry of the United States into World War II.
As I have recently established, Republican Senate candidates in Tuesday's election outperformed poll medians by an average of 5.3 percentage points. This is the largest across-the-board polling error since at least 1990, the earliest year analyzed at FiveThirtyEight. Based on that information and my own analysis at the Princeton Election Consortium, polls typically get the margins wrong by a median of 3 percentage points in midterm elections, in contrast with presidential elections, when pollsters are much better at gauging likely voters. I have referred to this as the Midterm Polling Curse. The midterm error can go in either direction: in 1998, Democrats outperformed polls by 4.9 percent.
If dismal voter turnout was responsible for differences between pre-election polls and actual outcomes, we might expect all races on the same ticket to suffer together. Here is a graph of GOP overperformance in states where there were relatively closely contested Senate and gubernatorial races on the same ballot:
Samuel S. Wang, Princeton Election Consortium
The correlation coefficient is +0.92, extremely strong (the maximum possible correlation is +1.0). In other words, if the Senate candidate underperformed, the gubernatorial candidate underperformed by a proportional amount.
Democratic underperformance was not limited to Senate and gubernatorial races. David Wasserman, Loren Fulton, and Ashton Barry at the Cook Political Report have tabulated the national House popular vote across all 435 districts. So far they have counted 5.1 million more votes for Republican candidates than for Democratic candidates out of 75.0 million votes cast, a 6.8 percent margin. Compared with the 1.5 percentage point margin in the median generic Congressional ballot in the weeks before the election, this is a 5.3 percent difference, the same as the error in Senate polls. In other words, nationwide voting behavior failed to be captured in a wide range of polls across the board.
Was this bias in polling caused by Democrats staying home on Election Day? One of the biggest surprises last Tuesday was the Virginia Senate race, in which Democratic incumbent Mark Warner led in pre-election polls by more than 10 percentage points, but in the end barely survived, by less than one percentage point. In Virginia, turnout was less than 37 percent of the voting-eligible population, as low as the national average.
At the same time, turnout was high in other states such as New Hampshire, where Democratic Senator Jeanne Shaheen led Republican Scott Brown by several points in pre-election polling, and exactly matched that performance on Election Day.
Here is GOP overperformance plotted, state by state, against voter turnout:
Samuel S. Wang, Princeton Election Consortium
The graph above shows a mix of Senate and gubernatorial races. In states where there were two races, I averaged the two races. The trend is again obvious, although weaker (correlation coefficient of +0.37). Overall, the pattern can be approximately explained if those who did not vote would have broken 3-to-2 for Democratic candidates, which seems like a reasonable number. In the most extreme case, Virginia, the stay-at-home voters would account for the outcome if they split 2 to 1 for Democrats over Republicans. Considering that Virginia was not considered a close race, voter engagement was probably far weaker than in hot battlegrounds.
Broadly, turnout was high in a number of states with high-profile races: Colorado, New Hampshire, Alaska, Wisconsin, and Maine. These are cases where pollsters had it about right. But in the lower profile cases, they missed, and in some cases they missed quite badly.
Exceptions to the overall trend will be interesting for strategists on both sides to understand. For example, why did Republicans and Democrats do about as well as expected in Michigan and Florida, while Republicans performed far better than expected in Kentucky, Kansas, and South Dakota, where turnout was nearly the same? Local factors are surely at work. To pick one case, the weakness of Kansas Democrats enabled independent candidate Greg Orman to displace the party's Senate candidate, but it may also have led to the defeat of both Orman and the Democratic gubernatorial candidate Paul Davis.
A larger question is why voter turnout hit a new post-World War II low. Compared with 2012, the number of votes cast dropped by about 42 percent. Democrats lacked a coherent message, de-emphasized their own policies in immigration and health care, and sidelined their highest-profile messenger, Barack Obama. Instead, issues such as Ebola and ISIS dominated the news. Relative media inattention to the election may have depressed turnout more than usual. These and other factors affecting turnout are inherently difficult for pollsters to anticipate. In 2014, the Midterm Curse, which this year afflicted both pollsters and Democrats, was in all likelihood caused by exceptional voter apathy.
Related: In Blue State Turned Red, Former Candidate Says Low Turnout Reflects Dems' Failures, by Dana Beyer
