How Much the Polls Missed by in Every State

fivethirtyeight.com

21 pointsprotomyth2 comments

The five most populous states all feature in the top half of this graph (California 3rd, Texas 15th, Florida 18th, New York 12th, Illinois 5th)

While the five least populous states are mostly in the bottom half. (Wyoming 43rd, Vermont 21st, Alaska 30th, North Dakota 48th, South Dakota 44th)

It is much harder to find a small representative group in a larger population. But I a surprised that the errors appear to favor overestimating trumps margin. I would have thought that the sampling errors would go both ways.

Lots of possible explanations including coincidence, etc. But if feels like it might be some pattern...?

Yes. There was a strong stigma attached to voting for Trump, which made people to falsely claim they were undecided or voting 3rd party.

Andrew Gelman (who wrote the textbook on Bayesian Data Analysis) says:

> Yes, the polls were wrong, but as I’ve written, they weren’t that wrong—they were off by about 2 percent.

http://www.slate.com/articles/health_and_science/science/201...

Reply on news.ycombinator.com