@NateSilver538 Further Predictions and Success Rate

@NateSilver538 Further Predictions and Success Rate

Nate Silver’s twitter account is a lot easier to find information from than his website.  Plus, he knows the key people also putting out important election analyses.  So we put some of this on our website, including the success of the polls in this election.

We start with the current state of predictions for close, uncalled elections.

For the House, with 13 seats still unsettled, he has 225 Democratic seats to 197 Republican seats.  That put the Democrats 7 seats over the 218 needed for a House majority.

Nate’s ratings for the 13 unsettled House seats are:

Solid D:   CA 48, CA 49, NM 2

Likely D:  CA 10, NJ 3, NY 22, UT 4

Lean D:   CA 39, CA 45, ME 2

Tossup:  oddly, none

Lean R:

Likely R:  GA 7, NY 27, TX 23

Solid R:

If these new predictions are right, there will be 10 more Democratic seats, and 3 more Republican seats.  Lean D means the probability of success is 65% or 2/3 or better.  Solid D means 95% probability or better.  Likely D means about 85% or better.

Our local CA 45 went from Lean D, to a tossup at election time, back to Lean D.  My previous post documented the Democratic trend as more votes are being counted.  Same for CA 48 and CA 39.

There are three Senate seats in limbo.  The Arizona senator we covered, and is a Tossup.  The Florida senator is Lean R.  The Mississippi senator will be a run-off race and is Likely R.

NateSilver538.com predictions, assuming uncalled races fall as indicated, had the following success rates:

Deluxe model:   488 of 506 correct (96.4%).

Classic model:  484 of 506 correct (95.7%).

Lite model:        481 of 506 correct (95.0%).

“In the Deluxe model, Ds won 96.1% of the races in which they were favored, and Rs won 96.7%.”

In all of my previous articles, I have only used the Classic model, for simplicity.  

With success rates as that above, we might wonder if we could have saved the trouble of having midterm elections, and spending $5 billion in false and very annoying advertising, messy and distracting lawn signs, endless canvassing, and partisanship.  However, all of that was folded into the polls and finance analyses that led to the opinions of voters and was figured into the predictions.  Plus, all of the State and local offices, and the hours wasted trying to understand the confusing propositions and their old twitter size explanations.

In a graph (@DrewLinzer) of the poll margins predicted versus results for the 2018 Senate and Governors races, the chart was fairly flat, and ran mostly between -6% to +6%.  The Root Mean Square Error of the poll distribution was 4.7%.  The polls over estimated the Democratic win margins by 0.7%.  This reflects the quoted plus/minus 4.5% standard statistical errors on the polls, without much of a systematic error.  I wrote an article on the suspected over estimate because Democrats would be more willing to respond to phone polls by liberal media than Republicans, but this shows that it was very small, if it exists at all.

About Dennis SILVERMAN

I am a retired Professor of Physics and Astronomy at U C Irvine. For two decades I have been active in learning about energy and the environment, and in reporting on those topics for a decade. For the last four years I have added science policy. Lately, I have been reporting on the Covid-19 pandemic of our times.
This entry was posted in 2018 Midterm Election, Dana Rohrabacher CA 48th, Mimi Walters CA 45th. Bookmark the permalink.

Leave a Reply