Before diving into observations about the last Tuesday’s results, as a pollster, I feel compelled to write this first post-election column about the survey research misses.
First, put the errors in perspective.
{mosads}A few weeks ago, I noted that over the decades, poll averages missed the actual results by an average of 3-points. When all the votes are counted, this will be a miss of less than 2 points — a below average error.
In 2012, poll averages were off by a greater 3 points, but no one seemed to even notice, let alone care, because polls had President Obama ahead and he won—albeit by a bigger margin than forecast.
This does not mean the polls were perfect. Poll averages are designed to minimize random variation. In this case, the polls all pointed in the same direction, suggesting systematic, not random, effects.
Moreover, while in many states the polls were on target, they happened to be off most in the states that ended up mattering most.
Second, it’s too early for anyone to be definitive about the source of the problems. The data needed for complete analysis are simply not yet available.
So instead of offering answers, I’m going to raise possibilities. I’m not standing behind any of them, just suggesting they all be considered.
Undecideds breaking toward Trump — The number of voters who were undecided or supporting minor candidates was far greater this year than in past cycles. On average, about 13 percent were in this category.
The second highest was 10-points, in 2000, when Al Gore made up a deficit of more than 3-points in the polls to win the popular vote.
Between 2000 and 2016, undecided/other comprised less than 4 percent in late polls.
Hillary Clinton actually increased her percentage of the vote by 3 points over the final polls. Donald Trump gained 7.
It’s a pattern consistent with undecideds breaking strongly to Trump and consistent with exit poll findings that late deciders broke toward Trump.
Turnout — Alas, poor turnout, always blamed when things go wrong. Polls are poor at measuring turnout. Quality pollsters are guided by history and history is a useful, but imperfect guide.
While the claims of turnout collapse lack factual foundation, there were some notable drops.
For example, Hillary Clinton lost Wisconsin by just 27,000 votes, while 40,000 fewer people cast ballots in heavily Democratic Milwaukee than in 2012. Similar situations obtained in Detroit and other cities.
Had those 2012 voters come to the polls, Clinton would likely have won the state.
Voter suppression — In talking turnout, there is an implicit assumption of intention and choice — that those 2012 Milwaukee voters chose not to turnout. But major efforts were made to prevent them from turning out, in Milwaukee as elsewhere.
Wisconsin’s voter ID law, designed to keep poor, immigrant non-driving Democrats away from the polls, was the subject of back and forth litigation.
It’s impossible to know how many voters stayed home because they lacked the requisite identification, but the Executive Director of the city’s Election Commission noted turnout fell furthest in those areas of the City earlier deemed most likely to have problems with the harsher requirements.
Complacency is the flip side of suppression. Over 60 percent of Americans thought Hillary Clinton was going to win. Did this allow some putative Clinton voters to stay home, worry free?
Misjudging college-educated whites — A great deal of attention has been rightly focused on non-college educated whites who moved far from the Democratic Party and ended up supporting Trump by an even bigger margin than expected — 39-points.
But throughout, Clinton was making up for some of that deficit with college educated whites, who were breaking more strongly Democratic.
On Election Day, however, college educated whites ended up backing Trump by 4-points, contradicting the findings of almost every pre-election poll.
Were these “shy” Trump voters? Or was something else going on with this segment?
In short, overall, polls performed pretty well, but there were errors. There are lots of theories about why, but hard conclusions must await more data and further analysis.
Mellman is president of The Mellman Group and has worked for Democratic candidates and causes since 1982. Current clients include the minority leader of the Senate and the Democratic whip in the House.