The Fault in Our Polls

November 8, 2017
Posted by Jay Livingston

A year ago, the polls predicted that Hillary would win. They were wrong. They also predicted that she would get about 3 percentage points more votes than Trump. They were right.

This year, the polls – nearly all of them – predicted that in gobernatorial election in Virginia, Ralph Northam would beat Ed Gillespie. They were right. They also predicted, on average, that the winning margin would be 3.3%. They were wrong. Northam won by more than 8 points.

RealClearPolitics published this table of poll results. (The last two rows are my own addition, based on stories at The Hill.)Bad calls – mostly results falling outside a poll’s margin of error – are in red.

(Click on the image for a larger view.)

In general, the polls
  • called the winner (only three got it wrong)
  • nailed the Gillespie vote
  • underestimated the Northam vote
  • underestimated the winning margin
Some polls came closer than others. Quinnipiac had the margin right but lowballed the percents for both candidates, especially Gillespie. Rasmussen’s usual rightward bias led them to miss what most got right – the winner. Of the major polls, the farthest off was New York Times / Siena, though it showed no liberal bias in its errors. It underestimated the Republican vote by 5 points, the Democratic vote by 10 points.

I have no good explanation for these results. I leave that to people who know something about polling and voting.

No comments: