The 2004 Debrief · post-mortem · permanent archive
How did the polling industry do this cycle?
The 2004 industry MAE landed at 4.5pt — graded across 94 historical races with a known outcome — and 131 polls were analyzed across 35 pollsters. Signed bias +0.15pt: polls landed close to the mean — no systematic over-predict on either side.
INDUSTRY MAE
4.5pt
vs midterm avg
SIGNED BIAS
+0.15pt
R-leaning
POLLS · POLLSTERS
131
35 firms
RACES SCORED
94
w/ known outcome
Industry mean absolute error · 2002–2004
Per-pollster scorecard · 2004 · sorted by MAE
Signed bias · ±9.4pt
MAE · 0–9.4pt
Biggest-miss races · 2004
| Race | Expected | Actual | Miss | Direction |
|---|---|---|---|---|
| us-senate-2004-pa | R+20.00 | R+10.63 | 9.4pt | R-margin overestimate |
| us-senate-2004-ok | R+3.20 | R+11.53 | -8.3pt | R-margin underestimate |
| us-senate-2004-ky | R+7.50 | R+1.32 | 6.2pt | R-margin overestimate |
| us-governor-2004-wa | D+5.00 | tied | -5.0pt | D-overestimate |
| us-senate-2004-sd | R+3.00 | R+1.15 | 1.9pt | R-margin overestimate |
| us-senate-2004-wi | D+13.00 | D+11.24 | -1.8pt | D-margin overestimate |
| us-senate-2004-ak | R+1.50 | R+3.03 | -1.5pt | R-margin underestimate |
| us-senate-2004-wa | D+11.00 | D+12.24 | 1.2pt | D-margin underestimate |