CENSUS-26 · VALIDATION
Tested on elections it had never seen
Each election from 2016 to 2024 was predicted using only the cycles before it, for the House and for the Senate, and scored against what happened.
01 House
Five House elections, predicted blind
Given the real national vote, the census alone calls about 94% of seats. Fully blind, guessing the national mood from history too, it still lands near the result in most years.
| Year | Won by Democrats | Census, real vote | Census, blind | Seats called | Brier score |
|---|---|---|---|---|---|
| 2016 | 194 | 199 | 189 (174–205) | 95.6% | 0.040 |
| 2018 | 235 | 222 | 224 (180–280) | 91.5% | 0.061 |
| 2020 | 217 | 212 | 239 (220–258) | 93.1% | 0.051 |
| 2022 | 213 | 215 | 203 (160–242) | 92.2% | 0.057 |
| 2024 | 215 | 216 | 219 (202–235) | 95.6% | 0.035 |
What 2018 taught it
The first version measured the national mood as the average margin across contested districts. In 2018 Democrats ran candidates almost everywhere, including deep-red seats, and that average said D+4.4 while the country voted D+8.7. Switching to the national popular vote fixed most of the miss; the rest is suburbs that swung further than their earlier profile suggested.
02 Where it missed
The 2024 misses, on the map
The 2024 House result minus what the census predicted, trained only on 2012–2022. Blue districts voted more Democratic than their profile said; red ones, more Republican.
The misses were not random. Diverse districts around New York, New Jersey, Florida and Texas voted clearly more Republican than their profile suggested, the 2024 shift among Hispanic, Asian and immigrant voters that earlier cycles could not teach the model. Much of the Pacific Northwest and the upper Midwest stayed more Democratic than expected.
03 The Senate
Five Senate cycles, predicted blind
Every Senate race with a Democrat and a Republican on the ballot, scored the same way. Races where an independent or a third party was the real alternative are left out: there D − R does not measure the race.
| Year | Races | Seats called | Average miss (pts) | Brier score | MP-26 (polls) |
|---|---|---|---|---|---|
| 2016 | 32 | 96.9% | 9.6 | 0.061 | – |
| 2018 | 31 | 83.9% | 9.0 | 0.133 | 0.055 |
| 2020 | 33 | 93.9% | 7.2 | 0.066 | – |
| 2022 | 32 | 90.6% | 8.0 | 0.062 | 0.039 |
| 2024 | 33 | 93.9% | 5.7 | 0.049 | – |
Each year was predicted blind, using only earlier cycles. The Senate is harder than the House: one candidate on a statewide ballot weighs more than a district's profile. In 2018, its worst year, Democratic incumbents held on in states like Montana and West Virginia that their demographics gave to Republicans. Average Brier score: 0.074.
NextCensus vs polls →