Vote totals are still preliminary, it will be weeks until they are finalized. But we know who won and we have a bunch of preliminary data.
So first of all, this means the "Categorization View" got two states wrong in the end (Wisconsin and Michigan).
Meanwhile the "Uniform Swing" model showed a median result of Trump winning by 86 electoral votes… which is exactly what happened.
This is the model that took into account the magnitude and direction of poll errors from 2008 to 2020 to estimate what would happen this time.
The essential result in the median case (simplified) was to not count a state as blue unless the Democrat led by more than 1.2%. That turned out to be a good bet, and resulted in a spot on prediction.
So I would say Election Graphs did pretty well this cycle.
I thought it was worth taking a first look at how the EG poll averages did vs reality in terms of state by state margin differences too. The preliminary actuals are from DDHQ as of 17 UTC on Nov 7th.
For the moment I'm only looking at the states (and ME-2 and NE-2) where the EG margin was under 10%.
Here is the data:
| Final EG Avg | Prelim Actual | Delta | |
|---|---|---|---|
| Nebraska-CD2 | Harris+8.9% | Harris+7.8% | Trump+1.1% |
| New Hampshire | Harris+7.8% | Harris+3.6% | Trump+4.2% |
| Virginia | Harris+6.1% | Harris+5.2% | Trump+0.9% |
| New Mexico | Harris+5.8% | Harris+5.7% | Trump+0.1% |
| Minnesota | Harris+5.3% | Harris+4.3% | Trump+1.0% |
| Wisconsin | Harris+0.4% | Trump+1.0% | Trump+1.4% |
| Michigan | Harris+0.2% | Trump+1.5% | Trump+1.7% |
| Pennsylvania | Trump+0.1% | Trump+2.0% | Trump+1.9% |
| Georgia | Trump+1.3% | Trump+2.2% | Trump+0.9% |
| North Carolina | Trump+2.0% | Trump+3.4% | Trump+1.4% |
| Nevada | Trump+2.5% | Trump+3.8% | Trump+1.3% |
| Arizona | Trump+3.5% | Trump+5.5% | Trump+2.0% |
| Florida | Trump+5.3% | Trump+13.1% | Trump+7.8% |
| Iowa | Trump+5.5% | Trump+13.2% | Trump+7.7% |
| Maine-CD2 | Trump+5.8% | Trump+8.0% | Trump+2.2% |
| Ohio | Trump+7.8% | Trump+11.3% | Trump+3.5% |
| Alaska | Trump+8.2% | Trump+15.2% | Trump+7.0% |
| Texas | Trump+8.4% | Trump+13.9% | Trump+5.5% |
Every single one underestimated Trump. Every single one.
This ranged from 0.1% in New Mexico, to 7.8% in Florida.
The average miss where EG margins were under 10% was a 2.9% miss.
For just the very closest states, the ones where the EG margin was under 5%, the average miss was a bit smaller at 1.5%.
And the tipping point was off by 1.9%.
Let's compare that tipping point error to previous cycles. From the Election Graphs FAQ:
- 2008: Obama's tipping point was 3.45% better than predicted.
- 2012: Obama's tipping point was 0.89% better than predicted.
- 2016: Trump's tipping point was 2.36% better than predicted.
- 2020: Biden's tipping point was 1.41% worse than predicted.
A 1.9% miss in the tipping point is actually better than 2008 or 2016, but worse than 2012 and 2020. So right in the middle.
So as polling errors go, this is actually pretty typical.
But of course the part that will cause the most consternation is that this is the third cycle in a row that the polls have missed in the same direction.
If the direction of the polling error was just random with a 50/50 chance, there is a 12.5% chance of the error being in the same direction three times in a row. Which, hey, isn't small enough for us to say there is something systemic wrong. There could be for sure. But it also could just be bad luck.
When we set up for 2028 though, the models will be based on polling error data from 2008 through 2024, and 2024's miss means 3 out of 5 cycles will have underestimated the Republican, so rather than it needing the Democrat to be ahead by 1.2% to have even odds of winning a state, they are going to need more than that for sure.
It makes it really tempting to not just show odds on Election Graphs, but show maps and the "spectrum of the states" and such based on the calculated odds rather than the straight averages. Maybe even show "unskewed" poll averages. But that feels icky in a way. But maybe as an option. I'll think about it.
Well, there is a lot of data crunching that I'll need to do to set things up for 2028, and I probably won't start that until at least 2026, so that may be a little premature.
In any case, thanks everyone for following Election Graphs this cycle. If I don't post again before then, see you for the 2028 race!