Mellman Group
Graded against the actual result across 42 races (from 55 polls, through 2022).
Head-to-head vs VotePredictor Elections
The fair, apples-to-apples test: on the 36 races Mellman Group actually polled, how its final poll's margin compared to what VotePredictor Elections predicted for those same races.
| Model | Avg miss (pts) | Called right |
|---|---|---|
| Mellman Group | 6.94 | 72% |
| VotePredictor Elections | 2.99 | 89% |
VotePredictor Elections aggregates all the pollsters, so it's expected to beat any single one on margin — that's the value of averaging. The honest comparison among forecasters is on the combined board.
Every race (42)
Each race Mellman Group polled, scored on its final poll — the call right before the vote — against the actual Dem−Rep result. Click a race for its full detail.
| Race | Their call | Result | Miss | Call |
|---|---|---|---|---|
| 2022 CA-27 House | D+6.0 | R+6.5 | 12.5 | ✗ |
| 2022 RI-2 House | R+3.0 | D+3.7 | 6.7 | ✗ |
| 2018 OH-7 House | R+7.0 | R+17.5 | 10.5 | ✓ |
| 2018 UT-4 House | D+1.0 | D+0.3 | 0.7 | ✓ |
| 2014 ND-1 House | D+2.0 | R+17.1 | 19.1 | ✗ |
| 2014 CO Senate | D+2.0 | R+1.9 | 3.9 | ✗ |
| 2014 KY Senate | D+2.0 | R+15.5 | 17.5 | ✗ |
| 2012 MT Governor | D+7.0 | D+1.6 | 5.4 | ✓ |
| 2012 WI Governor | R+3.0 | R+6.8 | 3.8 | ✓ |
| 2012 KY-6 House | D+14.0 | R+3.9 | 17.9 | ✗ |
| 2012 MT-1 House | R+2.0 | R+10.5 | 8.5 | ✓ |
| 2012 FL President | D+2.0 | D+0.9 | 1.1 | ✓ |
| 2012 IA President | D+2.0 | D+5.8 | 3.8 | ✓ |
| 2012 MT President | R+8.0 | R+13.7 | 5.7 | ✓ |
| 2012 NV President | D+6.0 | D+6.7 | 0.7 | ✓ |
| 2012 OH President | D+5.0 | D+3.0 | 2.0 | ✓ |
| 2012 VA President | D+3.0 | D+3.9 | 0.9 | ✓ |
| 2012 ND Senate | D+4.0 | D+0.9 | 3.1 | ✓ |
| 2012 NV Senate | D+3.0 | R+1.2 | 4.2 | ✗ |
| 2011 WV Governor | D+10.0 | D+2.5 | 7.5 | ✓ |
| 2010 WI Governor | R+2.0 | R+5.8 | 3.8 | ✓ |
| 2010 GA-8 House | D+3.0 | R+5.4 | 8.4 | ✗ |
| 2010 KY-6 House | D+12.0 | D+0.3 | 11.7 | ✓ |
| 2010 MA Senate | D+14.0 | R+4.8 | 18.8 | ✗ |
| 2008 GA-8 House | D+17.0 | D+14.5 | 2.5 | ✓ |
| 2008 ME President | D+17.0 | D+17.3 | 0.3 | ✓ |
| 2008 GA Senate | R+3.0 | R+2.9 | 0.1 | ✓ |
| 2008 LA Senate | D+20.0 | D+6.4 | 13.6 | ✓ |
| 2008 ME Senate | R+8.0 | R+22.7 | 14.7 | ✓ |
| 2008 MN Senate | D+2.0 | EVEN | 2.0 | ✓ |
| 2006 GA-8 House | D+16.0 | D+1.1 | 14.9 | ✓ |
| 2006 IL-10 House | R+17.0 | R+6.8 | 10.2 | ✓ |
| 2006 NV-2 House | EVEN | R+5.4 | 5.4 | ✗ |
| 2006 WI-8 House | D+3.0 | D+2.1 | 0.9 | ✓ |
| 2004 GA-3 House | D+29.0 | D+25.8 | 3.2 | ✓ |
| 2004 WA Senate | D+16.0 | D+12.2 | 3.8 | ✓ |
| 2002 TN Governor | D+6.0 | D+3.1 | 2.9 | ✓ |
| 2002 IA-4 House | R+5.0 | R+11.7 | 6.7 | ✓ |
| 2002 NJ Senate | D+11.0 | D+9.9 | 1.1 | ✓ |
| 2002 TN Senate | R+6.0 | R+9.9 | 3.9 | ✓ |
| 2000 NV Senate | R+4.0 | R+15.4 | 11.4 | ✓ |
| 2000 WA Senate | D+5.0 | D+0.1 | 4.9 | ✓ |
Accuracy by time to election
Lower is better. Time to election runs right (election week) to left (~2 months out).
By the numbers
| Time to election | Polls | Avg miss | vs field | Called right |
|---|---|---|---|---|
| 1–3 wk | 19 | 5.47 | +0.41 | 84% |
| 3–6 wk | 25 | 6.95 | +1.25 | 68% |
| 6–9 wk | 8 | 9.72 | +3.58 | 75% |
vs field is this pollster's average miss minus all pollsters' at the same lead time — green beats the field, redtrails it. Our historical polls reach ~2 months out; earlier polling isn't in the record.
Track record by cycle — getting better?
| Year | Polls | Avg miss | Lean (house effect) |
|---|---|---|---|
| 2002 | 5 | 3.5 | D+3.5 |
| 2006 | 4 | 7.8 | D+2.7 |
| 2008 | 6 | 5.5 | D+5.4 |
| 2010 | 6 | 10.9 | D+10.9 |
| 2012 | 20 | 4.2 | D+3.3 |
| 2014 | 4 | 14.2 | D+14.2 |
Do we credit a pollster for fixing its bias? Each cycle, the model re-estimates every pollster's lean from all its earlier polls (walk-forward) and subtracts it before using the poll. We tested weighting recent cycles more — it doesn't help: a pollster's lean in one cycle barely predicts the next (correlation 0.28), so the swings above are mostly noise, and averaging over more history beats chasing the latest cycle. The all-time estimate we use came out within ~0.5% of the best option.