Polling expert analyzes midterm GOP gains
Says economic anxiety drove midterm results
Mark Blumenthal has been working in the political polling industry for 20 years, but his historical perspective goes back much further. “The party of the president has lost legislative seats in the sixth-year midterm in all but one case since reconstruction,” said Blumenthal in a talk on November 12 at Emerson College in downtown Boston.
Blumenthal, senior polling editor at the Huffington Post and founding editor of Pollster.com, used his experience and knowledge of the 2014 midterm pre- and post-election polling to provide a fact-based explanation for the GOP midterm gains and some initial thoughts on why most polls understated the size of the GOP lead.
Blumenthal said that national GOP gains were primarily driven by economic pessimism. While most economic indicators such as unemployment rates and Gross Domestic Product point to an economy on the mend, exit polling indicated that 78 percent of voters were somewhat or very worried about the economy.
The disconnect is caused by the choice of indicators, said Blumenthal. Growth in average hourly earnings has a strong correlation with presidential party election performance and average hourly earnings have been stagnant since the 2008 recession. Based on their wages, voters did not feel like the nation’s economy was improving.
As expected, there was a strong correlation between Obama approval numbers and each party’s vote share. According to national exit polls, 87 percent of the 44 percent of voters that approved of the President’s performance said they voted for Democrats, while 83 percent of the 55 percent of voters that disapproved of the President’s performance voted for the GOP.
Turnout was also a factor. Blumenthal said that midterm turnout is historically much lower than presidential year turnout, but this behavior has only recently started to be a problem for Democratic candidates. While populations that turn out in lower numbers for midterms—specifically young and minority voters—used to split fairly evenly between parties, they now vote for Democrats at much higher levels. This gives Democrats an advantage in high-turnout presidential years, and a disadvantage in midterm elections.
A Huffington Post/YouGov poll conducted before the election also indicated a possible mismatch between Democratic message and voter concerns. The poll showed that voters’ top concerns were the economy and jobs, but their impression was that the top issues for Democratic candidates were social and women’s issues. The survey showed a closer match between voters’ concerns and top GOP candidate issues.
Blumenthal’s Pollster.com site aggregates and publishes thousands of political polls and includes a forecasting model that gave a win probability for each Senate and gubernatorial race for the 2014 midterm elections.
While Pollster.com and the other polling aggregators and election models predicted most of the election outcomes correctly, it is clear that most polls were systematically biased toward the Democrats.
Blumenthal said that systematic polling bias is rarely caused by one large problem, but rather by many smaller issues, and he gave several possible explanations for the bias in the 2014 midterm polling.
The underestimates of GOP vote share may have been partially due to an extension of a late trend towards Republican candidates, said Blumenthal. While most models attempt to judge the trend of the polls as they approach the election, the rate may not have been picked up by the closing polls.
A further implication of Trende’s analysis for 2014, borne out by the results, is that Democratic candidate support would tend to flatten out and stagnate, while GOP candidates would continue to gain in support. This can be seen across many of the Senate and governor races, including the Shaheen/Brown NH Senate race—which was won by Shaheen despite the trends—and the Massachusetts governor’s race, where Charlie Baker’s support continued to rise as Martha Coakley’s flattened out as the election approached.
Another factor in the polling bias was the likely voter model used by the pollsters, said Blumenthal. A key difficulty in polling midterms is determining which voters will actually show up in these lower turnout years.
Blumenthal said that many pollsters used likely voter models based on the 2010 midterm elections, which had the result of overestimating overall turnout and also Democratic turnout. While it will remain difficult to determine the composition of the electorate in lower turnout primaries and midterms, Blumenthal said that he is a proponent of voter-file-based techniques to predict voting behavior. He said state-provided data is a more concrete indicator of voter likelihood than questions asked of survey respondents.The talk by Mark Blumenthal at Emerson College was sponsored by the Emerson College Polling Society, a political polling organization run by Emerson College students and advisor Spencer Kimball.