Wednesday, November 9, 2016

Nate Silver was less wrong than you might think

A difference between 2 degrees can mean 2 inches of rain 12 inches of snow.  It's a nonlinear system, and the hallmark of nonlinear systems is that small chances can have big impacts.

The electoral college makes the US election a nonlinear system.  A small change in vote numbers here and there, concentrated in the wrong way, can make a huge difference. 

Clinton won the popular vote by about 1% point.  538 had Clinton up about 3% points.  The total error of the polls was pretty small, but the errors themselves were in critical areas.  In particular, Clinton did really well in states that she had no chance to win or lose, like Texas and California. 

Nonlinear systems are just not very intuitive and just don't seem fair, but they are all around us, and we need to understand them to view the world in a more complete way.

Monday, November 7, 2016

If Nate Silver's model is correct, he should get a few states wrong

On Wednesday, there will be articles written trying to decide which pollster got the most right.  They will add up all the states Silver and 538 are called correctly and come up with a final score, and then use this score to decide whether or not Silver's model works.

This is wrong.

Silver and the 538 forecast have a number of states very close, between 50-50 and 60-40 for a number.  If his model is correct, that means that the odds are that he will get at least a few states wrong.  In fact, if he were to clean sweep, that would mean that he either got very lucky or his model was actually wrong and was overly cautious.  Especially considering his 50/50 call in 2012, it is almost certain that Silver missing a few states would be considered a failure, but now you know better.