No state split its presidential and senate ballots in 2016, a phenomenon that has never happened before.  Jason Kander came within 2 points with incumbent Republican Roy Blunt, but the turnout generated by Donald Trump is too large for Kander to prevail; were Trump to win a slightly smaller margin, Kander would be the only Democrat to win in a Trump state in 2016. However, the situation is explainable, given what happened in 2014–every single Romney state on ballot replaced Democratic Sens. and put Republican Sens. in place, along with Obama-Trump state Iowa and light blue state Colorado. In a wave  election, slightly Democratic states going narrowly to Republican Senate candidates are normal behaviors in a wave election.
But indeed, 2014 is a prequel to 2016, not in actually previewing Trump’s victory, but in the incompetence of polls and wake of working class rural whites responding against Democratic policies. In Kansas, for example, Senate polling (latest eight) was o ff by 10.7 points and gubernatorial polling was off by 4.94, 9.2 in Virginia (including Republican pollster Vox Populi showing Warner up by 4 points In Illinois, likewise, gubernatorial polling was off by 5.58. There are, however, examples where polling was not off, such as Massachusetts’ gubernatorial election, Illinois Senate election (interestingly Senate polls in IL were not off, missing only by 1.32 points, but gubernatorial polls greatly exaggerated Quinn’s strength making Quinn seemingly favored but in fact he was not). Colorado elections, on the other hand, have seen the trend reversed–Polls showed, on average, Bob Beauprez leading incumbent governor John Hickenlooper by 0.5, Hickenlooper won by 3.35. Senate side polls were rather accurate–polls showed a 1.9 point lead for Cory Gardner, he won by exactly 1.9. Same trend worked in 2016–Hillary Clinton’s polling average lead was 2.9 points, she won by 4.9. For Senate, Bennet’s average lead was 6.9, he won by 5.7.
Regardless, the point is, the polls underestimated Republican strength substantially. In a state that polls missed massively–Wisconsin–polls predicted a narrow Walker victory by 1.7 points, but he won by a larger-than-expected 5.7, a 4 point difference. Likewise, polls predicted a 6.6 point Hillary Clinton victory, but Trump won by 1.1, a 7.7 point difference. In Michigan, the trend isn’t as substantial but still noticeable. Snyder was expected to win with 1.8 points, he won by 4.1, a 2.3 point difference. Hillary Clinton was expected to win the state with 3.6, but she lost it by 0.2, a 3.8 point difference. In the midwest, polls grew to be more inaccurate than they were before. In my massively-failed model to predict the November general election, I added certain provisions to correct the polling average based on historic presidential results, but presidential polling was reasonably accurate during the past two cycles (2008 and 2012) and they lean Republican as a whole. I failed to address midterm polling average errors and shifts in polling average errors. In predicting 2018 elections, such factors must be recognized.
2016 was a strike on the head to anyone who believed in accuracy and integrity of polling organizations. Polls failed to accurately represent the results of the elections, more massively than they did in 2014. If this narrative continues, it might be the end of opinion polling industry. Either we have to see major adjustments in the polling industry (such as in developments in RV/LV models), or polling will slowly become an obscured tool to estimate voting strengths.
- Enten, Harry. “There Were No Purple* States On Tuesday.” FiveThirtyEight. FiveThirtyEight, 10 Nov. 2016. Web. 21 Jan. 2017.
- Blake, Aaron. “Yes, This Was a GOP Wave Election.” The Washington Post. WP Company, 5 Nov. 2014. Web. 21 Jan. 2017.