News for hackers. Stuff that matters.

A comment on How (not) to forecast an election: Analysis via a hierarchical Bayesian model. Parent comment

Re:How (not) to forecast an election: Analysis via a hierarc... by the_cat_kittles on Sunday November 27, 2016 @02:19PM (#13050841)
that is exactly incorrect actually. a literal betting market has historically been the best predictor of outcomes. this cycle was particularly surprising, so everyone is all up in arms. but the markets that i follow (predictit.com) were more bullish on trump than almost all poll models. if you look at past performance they are excellent.
[ Parent | on HN ]
Re:How (not) to forecast an election: Analysis via a hierarc... by sgt101 on Sunday November 27, 2016 @02:57PM (#13051092)
Do you have the historic data (run up to 8th Nov) for Predictit? I have this http://predictwise.com/politics/2016-president-winner

Predictwise was calling ~90% for Hillary and 320ish in the college.

[ Parent | on HN ]
Re:How (not) to forecast an election: Analysis via a hierarc... by froogle on Sunday November 27, 2016 @07:19PM (#13052513)
PredictWise is not a prediction market. It's an aggregate of them, with David Rothschild's own personal special sauce added on top - which, in this case, is why it was "wrong".

PredictWise's mistake was that it assumed polling errors would not be correlated[1]. (He actually had a fight with Nate Silver over this.)

Nate Silver had, in advance, stated that polling errors would be correlated[2] (rightfully, as it turned out), which is part of why 538's model turned out to be closer to reality than PredictWise's.

The actual betting markets, which PredictWise was pulling data from, were in the range of 20-30% Trump. Much more bullish.

[1] http://predictwise.com/blog/2016/09/poll-aggregation-fight/

[2] http://fivethirtyeight.com/features/election-update-why-our-...

[ Parent | on HN ]