12/31/2016
The last word on the 2016 polls
By Ron Faucheux
Polls took a hard hit this year. Now that the votes have been counted, how bad were they?
Of the 13 final national polls conducted the week before the election that tested the four-way presidential contest, only one had Donald Trump ahead and 12 put Hillary Clinton on top. That would seem to be a veritable disaster for the polling industry––right?
Not exactly.
National polls only measure the popular vote. Clinton did, in fact, win the national popular vote by 2.1 points. The average of the 13 final national polls had Clinton ahead by 3.1 points, only a point off the actual result.
Ironically, all 12 polls that had Clinton ahead turned out to be closer to the final outcome than the poll that had Trump ahead. While that may seem crazy––since Trump, not Clinton, is headed to the White House––it’s true. The poll that put Trump ahead (by two points) was off by 4.1 points, while polls that gave Clinton the lead were off anywhere from only one-tenth of a point to 3.9 points.
Of course, the national popular vote is irrelevant in presidential elections. Only the Electoral College matters. But, when a candidate wins one and loses the other, as happened this time, it clouds perceptions of margin size and polling accuracy. What gave Trump his electoral victory was not his national voter strength––which he lost by nearly 2.9 million votes––but his strength in eight swing states, which he won by a total of 1,049,000 votes.
In Florida, the largest of the swing states, the average of the final three polls had Trump ahead by three-tenths of a point. He won the state by 1.2 points. Ohio’s last poll gave Trump a seven-point lead, and he carried it by 8.1 points. The average of the three final polls in Pennsylvania had Clinton leading by a single point and the last poll taken had Trump ahead by a point. He won the state by seven-tenths of a point.
In Michigan, the two final polls had Clinton ahead by an average of only 1.5 points. The later one gave Trump a two-point lead. He won by two-tenths of a point. The average of the final four polls in North Carolina had Trump ahead by five-tenths of a point, although he won by a more robust 3.7 points.
The final two in Wisconsin had Clinton ahead by an average of seven points, and Trump won the state by eight-tenths of a point. Both polls, however, were completed about a week before the election.
Polls are pictures in time, not crystal balls. They look backward, not forward. They can’t predict what occurs after they’re conducted. When polling shuts down too early, which happened in Wisconsin and other places, it misses the final breaks––and in this topsy-turvy election, it was last-minute movement in a few states that gave Trump his sensational victory. Exit polling found that late-deciders in Florida, Michigan, Pennsylvania and Wisconsin broke for Trump by double-digit margins as high as 29 points.
Certainly, none of this excuses bad polling––especially when it’s caused by corner cutting, biased sampling or faulty methods.
Even though plenty of polls accurately measured races for the presidency, the Senate and the House, it’s equally true that many of them were flat out wrong. To make matters worse, bad polls are often outliers and hit like bombshells, accurate or not. The media uses them to heighten the drama. Partisans use them to prop up their optimism.
Another problem this year was the slew of statistical “models” that incorrectly predicted the election. These calculations are really just educated guesses based on factors that can be influenced by the personal biases of the modeler. They should never be confused with polling.
Polls are imprecise tools with margins of error. But, in this tumultuous election––from Iowa’s caucuses in February to the final vote in November––polls came closer to measuring reality than did the pundits and partisans who only saw in the numbers what they wanted to see.
––Dr. Faucheux is a political analyst, author and pollster. He publishes LunchtimePolitics.com, a daily newsletter on polls. He also runs Clarus Research Group, a nonpartisan survey research firm that did not conduct any of the polls mentioned in this analysis.
#}