Reply to post: Re: Nate Silver was not wrong

US election pollsters weren't (very) wrong – statistically speaking

veti Silver badge

Re: Nate Silver was not wrong

Upvoted for accuracy.

Moreover: 538 attached a "probability" to its forecasts for each state, reflecting how close the race was state by state. A day before the election, I noted all those figures, and saw to my alarm that, out of 8 states where the certainty of outcome was less than 75%, 6 were in the Clinton column. Meaning that, if three of these states flipped - as should be expected, in that probability bracket - there was a good chance they'd all flip from Clinton to Trump.

As it happened, FL, NC and PA did just that.

In the 75%-90% probability bracket, there were 10 states, so we'd expect to see two flips. Those were MI and WI.

Of course, if we were just rolling dice, we'd expect to see even chances of states flipping in both directions. But Nate blogged about that too, and pointed out - before the election - that the elections in each state are not independent, the same factors that swing it in WI will also very likely matter in MI. So the chances of a nationwide swing (or error) being consistently in the same direction were high.

There were 33 more states where the probability was over 90%. Statistically, we should have expected one or two of these to flip too, but they didn't. So maybe the upper end of Nate's scale should be recalibrated, but for the close races? - he did a terrific job.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

SUBSCRIBE TO OUR WEEKLY TECH NEWSLETTER

Biting the hand that feeds IT © 1998–2019