2022 Midterm Projections
These projections are based off of polling date from 2004-2020. The discussion of "ALL" refers to modeling based on both General Election data and Midterm Election data. "Midterm Only" refers to modeling based solely on Midterm races. I've done modeling for various factors: VEP, Dem Votes, Rep Votes, Dem % of Vote, Rep % of Vote, Dem chance of winning, and Rep chance of winning.
Variables considered: Year, State, Candidate Name, Candidate Sex, Candidate Race, Election Type, % of vote choice for candidates in polls, Presidential Party in power, Presidential approve/disapprove ratings, Presidential approval ratings for Dems, Reps, and Inds, Congressional party approval/disapproval ratings, Party best to solve top issue, Generic congressional ballot, and Dem/Rep voting enthusiasm.
Modeling was competed through XGBoost with 2000 trees.
Voting Eligible Population (VEP) ALL
Voting Eligible Population (VEP) Midterms Only
Democratic Votes ALL
Democratic Votes Midterms Only
Republican Votes ALL
Republican Votes Midterms Only
Democrat Percentage of Vote ALL
Democrat Percentage of Vote Midterms Only
Republican Percentage of Vote ALL
Republican Percentage of Vote Midterms Only
Democrat Win ALL
Democrats Win Midterm Only
Republican Win ALL
Republican Win Midterms Only
Midterm Results ALL
Midterm Results Midterm Projections Only
Now I've never done this type of analysis before and the data is assuredly flawed, but here's what I think.
1. I tend to trust projections using ALL election data more than Only Midterm data because I think VEP will be closer to General elections. As you see above, only Wisconsin has a VEP over 50% and I think all of these races will be above 60%, which is in line with ALL projections.
2. Races I am most confident in:
a) Florida - Rubio is going to crush. That's a red state now.
b) North Carolina - Beasley has run an odd campaign and Budd seems like a very safe white guy in a red state.
c) Iowa - There's little Iowa data to use, which is why Grassley's numbers are down in projections. There's nothing to indicate that he won't win by 10+ points.
d) Nevada - While the race has tightened, I believe Cortez-Masto will carry NV.
e) Utah - I know, I know, I know. McMullin isn't a Democrat, but there isn't any data to work with there anyway. Lee wins easily.
3. Races I have a "good feeling" about. More gambler-esque than data scientist here:
f) Arizona - Kelly is an incumbent and is a good candidate. Masters is a crappy candidate. Maybe I'm wrong and Kari Lake will drag Masters across the line, but I like Kelly here.
g) Ohio - Ohio is a red state. Whatever the poll says for the Republican candidate in Ohio, just add three points. Vance is a hypocrite and a crappy candidate, but he's avoided embarrassing himself and is red in the right state.
4) Races that hold the Senate in the Balance and I have no idea on:
h) Georgia - Herschel Walker might be the worst Senate candidate in my lifetime. But unless VEP is over 70%, which is possible, I still see him carrying Georgia.
i) Wisconsin - Johnson is an embarrassment of a Senator. He was part of the fake elector plot for January 6th. How that doesn't automatically disqualify him is beyond me. And because of that, I'm calling for the upset! Mandela Barnes for the win!
j) Pennsylvania - Whatever aides let Fetterman on the debate stage should never work in politics again. He's been much best on the trail in the last week, but that debate performance was 100% pure cringe. But, I think Oz hurt himself enough by stepping on the abortion landmine. I'm going with Fetterman.
Here's my map:
