The 2020 Debrief · post-mortem · permanent archive
How did the polling industry do this cycle?
The 2020 industry MAE landed at 5.9pt — graded across 405 historical races with a known outcome — and 207 polls were analyzed across 89 pollsters. Signed bias −5.60pt: pollsters again systematically over-predicted Democrats.
INDUSTRY MAE
5.9pt
vs midterm avg
SIGNED BIAS
−5.60pt
D-leaning
POLLS · POLLSTERS
207
89 firms
RACES SCORED
405
w/ known outcome
Industry mean absolute error · 2012–2020
PoliAgg model · self-grade · 2020
How did our own production model do this cycle? The model issues one prediction per race; the polling industry above is averaged across many polls per race. Metrics are not directly comparable but published side-by-side for transparency.
Mean error per race
11.47pp
poliagg-v8 ensemble vs final margin
Call accuracy
88.6%
fraction with right winner called
Races graded
405
outcomes used in the grade
Per-pollster scorecard · 2020 · sorted by MAE
Signed bias · ±8.7pt
MAE · 0–8.7pt
Biggest-miss races · 2020
| Race | Expected | Actual | Miss | Direction |
|---|---|---|---|---|
| us-senate-2020-me | R+3.00 | D+8.59 | -11.6pt | D-overestimate |
| us-senate-2020-ky | D+10.00 | D+19.53 | -9.5pt | R-margin underestimate |
| us-senate-2020-mt | D+1.00 | D+10.02 | -9.0pt | R-margin underestimate |
| us-senate-2020-sc | D+2.00 | D+10.27 | -8.3pt | R-margin underestimate |
| us-senate-2020-nj | R+24.00 | R+16.31 | -7.7pt | D-margin underestimate |
| us-senate-2020-ia | R+1.00 | D+6.59 | -7.6pt | D-overestimate |
| us-senate-2020-il | R+23.50 | R+16.06 | -7.4pt | D-margin underestimate |
| us-senate-2020-ks | D+4.00 | D+11.43 | -7.4pt | R-margin underestimate |