As we did in 2018, we’ve gone back and done a preliminary analysis of the DDHQ/ @0ptimusPredicts forecast model’s performance. Our model predicted the vast majority of competitive federal races and came within 12 EVs of the final Electoral College map https://decisiondeskhq.com/2020-ddhq-optimus-forecast-model-how-well-did-we-do/ (1/x)

Before we dive in, just a note on the metrics used. We stayed consistent from 2018- we calculated several model metrics and included them in tables. But our focus is on accuracy of individual races (did we “call” it right) and the mean aggregate predictions of our simulations.

Some people might prefer different metrics- and we’re looking forward to others slicing this differently! But these are the ones we focused on in 2018, where forecasts tended to be pretty spot on, and we’re going to be consistent this year, when forecasts struggled a bit more.

On the presidential model: Biden won as expected. In terms of electoral votes, our model’s mean prediction was 318 EVs- Biden ended up with 306. The other forecasts tended to be a bit higher. As with the other forecasts, we “missed” FL, NC, and ME-2.

On the Senate model: Misses in NC and ME doomed the overall bullishness on Democratic chances of taking the Senate. Tillis led in 2 of the final 100 NC Senate polls we tracked, and won! Still, at 52 mean Dem seats we were middle of the pack.

A few house races left to call, but with 430 of 435 determined, we’re sitting at 97% accuracy and 15 “missed” races, roughly matching our 2018 performance (97% accuracy, 14 misses). Our mean prediction was 237 Dem House seats, which is the closest forecast to the expected 222-223

Interestingly, if you remove polling from our model we would have done even better! In a structural-only model, our forecast would have been 303 Biden EVs, 49 GOP Senate seats, and 232 Dem House seats. This is a reversal from 2018 where polling significantly helped forecasts.

The polling miss was bigger than in 2016 and should not be dismissed. Across Pres swing states, the average error was 5 points. In the Senate, it was 7 points. All overestimating Democrats. Those aren’t numbers that inspire confidence in the public nor can be easily mitigated.

As a final note, thanks to @NateSilver538 , @gelliottmorris , @JHKersting, @SamWangPhD, @Center4Politics, @CookPolitical, @InsideElections, @leantossup and other forecasters for pushing along methodological advances and providing a collegial environment