How USElectionAtlas predictions have held up over the years

[bagelman]

2004 Presidential (972):



Above is USElectionAtlas's presidential prediction map for 2004. Hover over each state for unnecessary details, but the only state it got wrong was my home state of Ohio, sadly, and that was enough for a Tossup/Tilt D election to be an R hold.

2006 Senate (465)

High confidence seats: All correct

Lean confidence seats: All correct

Tossup seats: Of four tossups, they were forecasted a 3-1 D with Virginia as an R hold. However, Virginia flipped for Democrats - by a margin of 0.36% and only 9329 votes. This mirrors the gubernatorial forecast as excellent.
 

2006 Gubernatorial (317)

High confidence states: All correct

Lean confidence states: All correct

Tossup states: Of four tossups, all forecasted Dem wins, only Minnesota was wrong and an R hold. It was also the only gubernatorial race less than 1 point.

Overall Atlas did an excellent job forecasting 2006-GUB.

2007 gubernatorial: None of the 3 states were tossups and all were correct.

2008 Presidential:

Atlas woefully underestimated how far the GOP would fall in the wake of the disastrous Bush administration.






An inexplicable Republican trend in NM and NV, misplaced confidence in NC, IN, and to a lesser extent MT, the idea that the pacific northwest would even be competitive...

Some of these mistakes are excusable. Indiana was a fluke for example. But it seems the magic Atlas had 2 year prior had faded. It seems that lots of people made predictions in the summer before the Wall St. meltdowns and maybe even before the baffling selection of Sarah Palin and never updated them.

However the Senate and Gubernatorial matchups were much more accurately predicted

2008 Senate

High confidence: All correct

Lean confidence: All correct

Tossups: All correct for Democrats, however Minnesota was sheer luck

2008 Gubernatorial

High confidence: All correct

Lean confidence: All correct

Tossups: Both correct for Democrats

2009 Gubernatorial: both correct for Republicans

2010 Senate:

High confidence: All correct

Lean confidence: All correct

Tossups: Both Colorado and Nevada were incorrectly forecasted as GOP flips, but Bennett and Reid both survived the Tea Party wave. The eastern tossups (3 midwestern GOP flips and Manchin holding on in WV) were correct.

2010 Gubernatorial:

High confidence: All correct

Lean confidence: All correct

Tossups: 9 states, of which only Illinois was wrong (Quinn survived the wave)

2011: all 4 states correct with confidence

[bagelman]

2012 Presidential: Atlas returned to form in 2012. The 6 tossups of NH, OH, VA, IA, CO, FL were almost all correct. FL was the only state wrong, and unlike in 2004 this did not change the entire result. Florida was the closest state anyway. Atlas incorrectly thought the closest state would be Virginia, underestimating how quickly the state was turning Democratic.

2012 Senate: All confidence states were correct as usual. Out of the many tossups, the two neighboring states of North Dakota and Montana were incorrectly forecasted as Republicans pickups. However while Romney won both easily enough, Democratic incumbents held on.

2012 gub: All states including two tossups were correct, with MT being the closest state.

2013 gub: Both states correct, neither tossups.

2014 sen: I'm not sure why people didn't bother filling in percentages in some states like AL, where Jeff Sessions ran unopposed. But I see no confidence states incorrect. Of the many tossups, only one was wrong: North Carolina, where the wave was strong enough to flip the seat for Thom Tillis.

2014: All confidence states correct as usual. For the many tossups, Democrats were overestimated. Maine, Florida, and Kansas were all incorrectly forecasted as Democratic flips in what turned out to be a red wave.

2015: In an off year, Atlas suffers the first confidently wrong gubernatorial prediction. IIRC Vitter (R) suffered a major scandal late in the game, and he lost to John B. Edwards (D). Meanwhile in Kentucky Atlas forecasted a close race in Kentucky where the Democrats are narrowly favored, but right winger Matt Bevin won by a decent margin. Atlas gets 2 out of 3 wrong.

2016 presidential: we all remember this one. Atlas got 3 states confidently wrong and I don't need to tell you what they are. With these 3 states, Atlas got the entire election confidently wrong. Add in two wrong tossups states and 1 wrong tossup CD to add insult to defeat.

2016 Senatorial: Atlas got Wisconsin confidently wrong, and got one tossup wrong (Pennsylvania). Both these states turned out to be wins for Republican incumbents.

2016 gubernatorial: Atlas saw most of these races as tossups, and they got 3 of them wrong, all won by Republicans. NH, IN, and MO.

2018 Senatorial: Atlas got all confident predictions right. However, 3 tossup states, all forecasted as Democratic wins in Trump states, instead joined North Dakota as Republican flips during the so-called "blue wave". IN, MO, FL, with the latter an especially painful for Voldemort.

2018 gubernatorial: Atlas got all confident predictions right. Atlas got 4 tossups wrong, 3 turned out to be Republican holds while Kansas flipped for Democrats.

2019: Atlas gets pessimistic after being burned in the faux blue wave, and gets Kentucky confidently wrong. Democrats subvert expectations and manage to very narrowly flip Frankfort.

[bagelman]

2020 presidential: Atlas thinks Biden is favored and that he will win back the upper midwest by more than he does, and wrongfully thinks Ohio and Iowa are still winnable, but overall didn't do too bad. 3 tossups are incorrect, all in the southeast and one in favor of Biden. Not many saw the Cuban swing coming, and most mixed up NC and GA.

2020 Senatorial: Atlas was collectively smart enough to keep Maine a tossup, and thus gets all confident predictions right. 4 tossups were wrong, Republicans in Maine and North Carolina, Democrats winning the doubleheader runoffs in Georgia held in January 2021.

2020 gubernatorial: Atlas gets everything on the map correct for the first time in a while, although only one race was a tossup.

2021: While NJ was much closer than expected, Atlas got both confidence states right. However the one tossup was wrong, and Youngkin flipped Virginia.

2022 Senatorial: Currently as of October 27th the collective predictions are the same as my own predictions. Some users, like ElectionsGuy and tweed (on AAD), are predicting another major Republican overperformance (like in both 2016 and 2018). Others think that poor GOP candidate quality will allow Democrats to hold the Senate and that the Roe backlash may be bigger than people think, with states like NC/OH/WI and maybe even IA potentially competitive. We shall see, but if Oz wins I might not even bother updating this post and may leave the forum for a while.

2022 gubernatorial: I have some different opinions (I think KS could potentially vote to the left of OR) but the gist of the situation here is similar to the above.

[Associate Justice PiT]

     Thank you for doing this analysis. The trend I notice is that Atlas's track record has gotten worse over time. I wonder how much of this is due to polling becoming less accurate and how much of this is to due to a decline in the quality of the users themselves (though I understand that the forum and predictions are separate, so maybe this has not hit predictions like it has the forum).

[Sestak]

Honestly, that 2008 prediction is...quite startling in terms of how bad it is. Almost as bad as 2016 in terms of states missed; which is pretty remarkable. NC safer than Washington...really?

Outside of those two years, is KY-Gov 2019 really the only confidence miss? (Grumbles at certain posters)

Admittedly, the above is also seemingly a result of Atlas being much more cautious and labeling more races as tossup. This is probably a smart choice given how bad polling has become but it does make the "only misses are tossups" results a little bit less meaningful.

[bagelman]

Atlas got 2 states wrong regarding gubernatorial results, Wisconsin and Arizona. Both were predicted as GOP wins and Democrats won instead. Both were considered tossups.

For Senate races, Atlas also got 2 states wrong, Pennsylvania and Nevada. Again both were predicted as GOP wins and Democrats won instead and as tossups. Georgia is also not resolved yet but I'd be very surprised if Atlas is wrong considering Walker's massive underpreformance relative to Kemp & co.

[Ferguson97]

Wow, really impressive that Atlas correctly guessed 50/50 states in 2004!

[Snowstalker Mk. II]

Atlas was brigaded by a Republican forum in 2008, which is why the predictions are more favorable to McCain than the actual result was, and also why McCain overwhelmingly won the mock election.

[RI]

Pretty proud of my track record. In 2008, I just mixed up Missouri and Indiana:



I did pretty well in 2012 (50/51 correct, just missed FL) and relatively well in 2016 (got PA correct but not MI or WI, although I distinctly remember waffling on MI after the late Trafalgar poll), too. 2020 was more challenging, though, and I missed Dem pickups like AZ and GA. Got the overall winner correct in all four, and called 96% of states correctly across the four races.

[GM Team Member and Senator WB]

I am pretty proud that I got the exact senate map for 2022, sticking with a prediction from May. It might say I missed a state but that's because this was before OK-Special was added.

[EastwoodS]

Pretty proud of my track record. In 2008, I just mixed up Missouri and Indiana:



I did pretty well in 2012 (50/51 correct, just missed FL) and relatively well in 2016 (got PA correct but not MI or WI, although I distinctly remember waffling on MI after the late Trafalgar poll), too. 2020 was more challenging, though, and I missed Dem pickups like AZ and GA. Got the overall winner correct in all four, and called 96% of states correctly across the four races.

OMG the people demand 2024!

[Comrade Funk]

2008 is likely due to most people making their predictions pre-Lehman.

[bagelman]

Atlas got everything right for 2023's gubernatorial races, although I think it was still possible to "predict" Louisiana after the race had happened.