After my last Senate map (D 53-46), I analyzed the polls in mid-late September but did not publish the results.
Here was that happy map (D 54-46 counting leaners):
Just a month ago, my polls-only analysis showed GA, OH, and NC as Leaning Democratic (2% to 4% leads for the Democrats), with no tossups, NV and WI as Slight Dem leads (4-8%), and PA and AZ with Moderate Dem leads (8-12%). FL was only leaning toward Rubio, IN and IA were pink, no stronger for Republicans than NV and WI were for Democrats, according to Aug and Sep polls.
Since that time, polls have tightened and/or shifted right. On the other hand, with Early Voting already started in many states turnout is up again!
Will this be a Roevember to remember? Or will the pollsters’ target demographic mixes be about right -- meaning that the cross-section of people who turn out to vote is about the same as the last few midterm elections — and Democrats will barely hold onto the current 50 + the VP control of the US Senate?
Only people voting can tip the scales to a big Democratic win predicted by polls in the first three months after Dodd overturned Roe v. Wade.
What races have tightened?
AZ from Moderate D to Slight D lead.
IN from Slight R to Leaning R.
UT from Moderate R to Slight R lead. (Note that Evan McMullin is an Independent who vows not to caucus with either party.)
OK-special from Prohibitive R to Moderate R lead.
Where have Republicans gained?
PA from Moderate D to Tossup!
NV and WI from Slight D to tossup.
OH from Leaning D to Tossup.
NC from Leaning D to Leaning R.
FL from Leaning R to Slight R lead.
Where have Democrats gained?
NH from Slight D to Moderate D lead.
The crucial races in detail
GA
Who expected Georgia to be Democrats’ ace in the hole? Sen. Raphael Warnock has grown his small lead since July over Heisman Trophy Winner and serial absentee father, Herschel Walker, as Walker has become known for his exploits off the football field. Maybe Walker’s hypocracy, idiocy and complete incompetence will keep him out of the Senate. Or will he get a pass on all those things the way T**** did in 2016?
Sen. Warnock has a 2.5% lead over the course of October. 538 puts him at +3%. Leaning D
PA
Every poll published by 538, except one, has shown Lt. Gov. John Fetterman with a lead over quack cure hawker, TV Dr. Mehmet Oz. The outlier was by a fledgling pollster called Wick. It showed Oz leading by 5 points in early October. That said, the leads have gotten smaller. Disturbingly so.
It has slipped from Moderate D in early Sep to Slight D (+5.5%) over the whole of September to just 1.9% D over the course of October. Tossup, but 538 puts Fetterman a bit better off at +2.3%.
NV
What happened to Sen. Cortez-Masto’s lead? The 538 model has been very dynamic. This has been mostly due to wide variability in polls. An early April OH Predictive poll showed Cortez-Masto ahead by 8%, in mid-May U of Nevada showed her up 22%, and in July the Environmental Voter Project put her up 8%. There have been a lot of Republican-funded polls, but they have shown a race within 5 points with the Democrat ahead almost as often as Republican Laxalt.
I had the race between Slight D and Leaning D in Sep. It’s now at D+0.4% 538 says R+0.1%. Tossup
WI
Lt. Gov. Mandela Barnes had a Slight D lead in September over deeply underwhelming Trumpist Sen. Ron Johnson. Wisconsin’s had fewer October polls than other states. They average R+1%. Tossup
Somehow 538 puts Johnson up 2.6 points. They must be weighting those R-leaning Likely Voter polls.
OH
In August and September Ohio leaned toward moderate Democrat Tim Ryan over Trumpist JD Vance. But since October 5, the polls have gone red, each poll showing Vance with a lead of less than 5 points, except two — one was a tie and one was Ryan +2.
Interestingly a Democratic pollster, Center Street PAC, has shown the opposite trend: May 11-13 R+3%, Jul 1-3 D+5%, Aug 1-3 D+11%, Aug 31-Sep 2 D+9%, Oct 2-3 D+9.5%. Whose model is right? We shall see.
The unbiased poll average gives Vance Lean R (2.2% lead). When adding in biased polls in both directions, Vance’s lead goes down to less than 1%. 538 shows R+1.7%. Going with Tossup.
NC
Not much attention has been paid to North Carolina this year, but the state is a lot more purple than you might think. They’ve only had 3 Republican governors for 4 terms total even since the demise of the Dixiecrats in the 1970s. Barack Obama won here in 2008 and narrowly lost in 2012.
Former State Chief Justice Cheri Beasley was tied to leaning her way versus Trumpist Rep. Ted Budd through August and September, but Budd has regained a weak lead of 3.5 points. Leaning R
538 shows R+2.6%
IN
A clearly red state that Obama won in 2008 is Indiana. They have a strong blue collar tradition and a large African-American population in the Northwestern corner, but let’s face it -- the state doesn’t fluoridate its water! How Republican can a Great Lakes state get?
However, there have only been 2 polls of the race between Sen. Todd Young and Democrat Tom McDermott, Mayor of Hammond. One of them by McDermott showing him trailing by only 3 points and the other independent poll showing just a 2 point difference with the Libertarian getting 6%. That’s R+2 or R+2.5. However you slice it, it’s only a Leaning R. Too few polls for 538 to estimate.
AZ
Sen. Mark Kelly had a moderate 10-point lead in the summer. His lead has shrunk to the 5-point range and he’s calling out mistakes of fellow Democrats. Still good for a Slight D Lead. 538: +4.5%
FL
One of my favorite Republican targets to beat this year is Sen. Marco Rubio. It seems Rep. Val Demings should have a good shot at dislodging him, but her best polling period put her at -2.7 in late August. Only one poll, conducted for Progress Florida, put her at tied. After a flurry of polls in September, this race has had very few polls this month.
Rep. Demings’s hopes clearly rely on a bump from her solid debate performance last week and a huge Roevember turnout. I agree with 538: R+4.8% Slight R Lead
Speculation about Results Depending on Polls + Turnout
If we assume the pollsters have their turnout demographics perfect in the aggregate and we just take today’s averages, we’ll wind up with this expected Senate:
51 D caucus to 49 Republicans
I haven’t done rigorous analysis on this matter, but I’m pretty sure polls overestimated Republican performance in the 2018 mid-terms when T**** wasn’t on the ballot, but was selling out his country to Russia and Saudia Arabia. They definitely overestimated Democratic performance in 2020 when T*** was on the ballot. I will quantify this effect from 2018 and apply it to these polls in my next article.
If Republican anti-choice zealots, Trumpists and other groups turn out in an unexpectedly high proportions, say polls + 3% in every state, we could get this worst-case Senate:
48 D caucus to 52 Republicans
If women respond in the voting booth to the fact that birth control, health care, pregnancy crisis management and reproductive choice could all be taken control of by Republicans in government, it could change the turnout from the models used by pollsters.
If Democratic turnout beats pollster expectations by 2 percentage points we could get this Senate:
52 D caucus to 48 Republicans
With a 4-point excess turnout in every state:
55 D caucus to 45 Republicans
With a 6-point excess turnout in every state:
56 D caucus to 44 Republicans
My method
This is my quick method of poll analysis. Not doing rigorous statistics to combine polls. Typically, polls have a 95% confidence margin of error in the 3 to 5 percentage point range. I find the average of poll margins and compare it against fixed thresholds based on 4%:
- < 2 percentage pt. margin = Tossup (much less than 95% confident either candidate is leading)
- 2 to 4-point margin is a Lean (270’s gray-blue or gray-pink)
- 4 pt. < margin <= 8 pt. is a SlightLead (baby blue or pink) (more than 95% confident)
- 8 pt < margin <= 12 pt. is a Moderate Lead (blue or red)
- Margin > 12 points is a Prohibitive Lead (dark blue or dark red)
I review the polls reported on 538, as they tend to be a superset of those on 270ToWin. For each race, I look back 1 to 2 months. Now I am collecting October polls. If there are fewer than 3 unbiased polls from that time range, I go back into September. I do an average of unbiased polls, then I do a second set including biased polls that were taken during the same time frame. See the discussion of biased polls below.
I compare the averages to the thresholds above to find a category(ies). If the unbiased and combined polls differ in category, then I take the one that is less certain (closer to gray). However, if all the biased polls are from only one side, then I go back up to 2 months in time to get the other side, or I rate that average one category less favorable to the side of the bias.
Biased polls include those marked by 538 as funded by a party or candidate. I particularly find that Trafalgar Group, Rasmussen (haven’t seen any this year, I think they changed their name in 2020) and John Bolton Super Pack are actually quite biased in favor of Republicans. Usually, the Democratic polls are neutral or even slightly pessimistic toward Democrats. I am counting them as unbiased. Interestingly, Fox News polls are unbiased, and they are counted as such.
When a pollster does multiple polls at once with different criteria (LV, RV, A, or with and without 3rd-party candidates), I throw out any All Adults polls, average the rest and count it as 1 poll. I prefer post-primary polls when there is a question about how many polls to consider.