Does the result prove The Huffington Post wrong?

The Huffington Post, on the morning of the election, gave Hillary Clinton a 98% chance of winning.

Boy, do they have egg on their face!

Or do they? They didn't give Trump a 0% chance. A 2% chance is a chance.

How do we judge when a probabilistic prediction of a one-time event was wrong? Not an easy question!


  1. Even worse are the people calling for Silver's head. 2 to 1 odds, really?

  2. Not an easy task, but it forces us to examine the ones who did see a Trump victory as more likely, and their reasons for thinking so.

    Michael Moore had a good understanding of the Rust Belt voter, and nicely explained that many of them did not even like Trump the person, but his agenda and what it would imply for their economic livelihood.

  3. It could of gone either way. It was no sure thing. Trump won because democrats did not vote. "Hillary Clinton won the national popular vote in 2016 by 282,546 votes out of a total of 119.8 million votes or 0.2%. Trump's margins in Michigan, Wisconsin, and Pennsylvania were 11,837 and 27,357, and 68,236, respectively. If Clinton had gotten 107,430 more votes in these three states in the right proportion, she would have won the election. In other words, a change of less than 0.1% of the vote, properly placed, would have flipped the presidency. A Trump surprise it was, but a landslide it was not." -

    1. I haven't seen anyone calling it a landslide!

      I DID say that, if not for the hot mic tapes, it WOULD have been a landslide. They set him back 8 or 10%, that he had to make up in two weeks.

  4. If you say that there's a less than 2% chance of something happening and then it happens, I think you need to explain how you ended up in the 2%. For example, if on the morning of the election Hillary had been discovered in bed with Anthony Weiner, then HuffPo could say their model was right. Instead it turned out that polling was off in a foreseeable way.

    1. The problem here is this messes with the whole probabilistic analysis. There is only a 1/64 chance of flipping six heads in a row -- close to 2%.

      But if we do so, there is nothing whatsoever to explain.

      If we can trace the outcome causally, then it wasn't random, and we shouldn't have been using this probabilistic framework!

  5. I certainly agree with the basic premises here. I will say parenthetically that I think people have been a little rough on Nate Silver after this cycle. His model gave Trump roughly a 30% chance of winning as of the day before. 30% chance means it totally could happen; don't assume it won't.

    According to Silver, he was getting complaints and hate mail in the week leading up to the election from people who wanted him to be more definitive that Hillary would definitely win. (How could she not? Who votes for Trump, really?) Just goes to show the kind of blue bubble logic that Silver has to attempt to operate in spite of.

  6. You look at the full set of 2% predictions their model made ...?


Post a Comment

Popular posts from this blog

Central Planning Works!

Fair's fair!

Well, So What?