There’s a YouGov poll that’s under “Added today” in 538’s Latest Polls page which has Warren at 23%. But the dates it was in the field were July 3-14.
I had not noticed that. Good pick up. Nevertheless the result is consistent with her power rating being number two. And Sanders not.
The link also shows a breakdown by group. Biden has huge strength in particular with Black women. Warren not. She needs to better connect with those voters by after NH to have a real shot.
I agree with you there.
The order of the early primaries is IA, NH, NV, SC, and then Super Tuesday. So the field’s going to be narrowed quite a bit before the African-American vote will be a factor. And so much will depend on who’s still in the game by SC.
You’ve got a couple of things going on. One is that polling I’ve seen shows that for voters whose first choice is Warren, their second choice is Harris - and vice versa.
The other is that if Harris wins one of those first three states, she’s liable to win most of the black vote in SC and beyond. And if she wins most of the black vote in SC, she probably wins SC, and Warren’s support in the Super Tuesday states starts gravitating towards Harris. She’d be pretty hard to stop after that.
But of course there’s no guarantee that Harris breaks through in one of the first three states. In which case it likely comes down to Biden v. Warren. And she’s got plenty of time to wrap around Biden’s neck the combination of the 2005 Bankruptcy Act and the bank deregulation in general that was Biden’s forte, and the wave of foreclosures in the Great Recession when blacks lost their homes and savings in way greater numbers than whites.
We do have a few things going on.
Like that second choice bit. You may be thinking of the MC poll question and yes the most common second choice for Warren supporters is Harris - at 30%. But the Sanders and Biden split a good chunk at 19 and 17% each. IOW 70% of Warren supporters don’t have Harris as their second choice and more would choose one of the other top contenders. It is a more pronounced circumstance for Harris supporters who would indeed choose Warren most frequently at 28% but with Biden right there at 26%.
Yes what happens in Iowa and NH are key to what follows. If Biden wins both, well you have been pretty consistent in pointing out how predictive not winning either of those (when both are in play) is. I’ll challenge that position of yours some with the proviso that a win that is way below expectations and a second place by Warren or Harris that far exceeds expectations, may allow them to live to fight on, but if he wins both without disappointing the polls dramatically, then he cruises into a blowout in SC and a near run of the table on Super Tuesday. Even a Harris win in Nevada (should be doable) would not impact that.
Of the two a Harris win in one of those two is more damaging to Biden. But Warren still is his more powerful adversary.
I hope so, since I’m pretty solidly a Warren fan at this point. 
Probably not a lot of surprises in this latest edition, but here are my latest power rankings:
-
Joe Biden (dominant front-runner again)
-
Elizabeth Warren (solidifying #2 status, gradually moving past Bernie)
-
Bernie Sanders (steady support but no path forward)
-
Kamala Harris (continues to slide, perhaps no longer top-tier)
-
Pete Buttigieg (hasn’t moved up at all despite loads of fundraising)
-
Cory Booker (quietly had a good week post-debate)
-
Beto O’Rourke (getting more attention from El Paso shootings than anything else)
-
Andrew Yang (despite internet search interest, not much movement in polling)
-
Tulsi Gabbard (ditto the above)
-
Julian Castro (in the top ten at least)
-
Amy Klobuchar (same position she’s been in the last few months)
-
Marianne Williamson (2nd debate star, but not much to show for it)
-
Tim Ryan (some attention post Dayton shooting, may rise a little)
-
Kirsten Gillibrand (needs to quit)
-
Tom Steyer (a big splash when he entered, but barely a peep since)
-
Steve Bullock (no chance, needs to know this)
-
John Delaney (easily the biggest loser of the 2nd debate)
-
Jay Inslee (can’t believe this guy isn’t getting more attention, but it’s over)
-
Michael Bennet (no chance, needs to drop out)
-
John Hickenlooper (delusional, potentially hurting his chances in a senate race)
-
Bill De Blasio (no chance, needs to quit)
-
Wayne Messam (Still here?)
-
Seth Moulton (needs a cool nickname: “Smokin” Seth Moulton!:dubious:)
-
Joe Sestak (???)
Since 538 doesn’t publish an average for the Democratic primary contenders, and since RCP’s average is so primitive, I’ve decided to create my own, which I’m intending to update on an occasional basis, using a consistent methodology. This thread seems like as appropriate a place as any.
Since polls are supposed to be snapshots, the polls in my current average only include national polls taken entirely since the debates at the end of July. As we get further away from the debates, I’m going to have to have some cutoff for how far back I’ll reach to include a poll, but I haven’t made any decisions on that yet.
My starting point is the Democratic primary polls at 538. For polls since the last debates, I only consider national polls by a pollster rated in the A-B-C range, and only the most recent national poll by a given pollster. If they release two or more sets of numbers for a poll, I take the highest-quality one, with LV > RV > V > A.
But here’s the key: I weight by 538’s letter grade for a pollster. Ignoring the plus and minus signs, I have a 5-3-1 weighting scheme: ‘A’-rated pollsters count 5 times as much as a C-rated pollster, and B-rated pollsters count 3 times as much as a C-rated pollster.
Right now, there are 8 pollsters who show up in 538’s latest polls as having polled the Dem primary field since the last debates: two A-rated pollsters (Survey USA and Quinnipiac), four B-rated pollsters (YouGov(8/10-13), Morning Consult(8/5/11), Ipsos, and Public Policy Polling), and two C-rated pollsters (Change Research and HarrisX(8/11-14)).
So taking Biden’s numbers as an example: the two ‘A’ pollsters have him at 33 and 32, the four ‘B’ pollsters have him at 36, 33, 25, and 21, and the two ‘C’ pollsters have him at 29 and 23. So Biden’s weighted average is [5*(33+32) + 3*(36+33+25+21) +1*(29+23)]/[52 + 34 + 1*2] = 30.0833, which I round to the nearest tenth of a percent, or 30.1.
So here are the current averages:
Biden 30.1
Sanders 17.1
Warren 17.0
Harris 8.2
Buttiegieg 5.6
O’Rourke 2.6
Everyone Else < 2.0
Well, it didn’t take long for a new poll to shake things up a bit: a new Fox News poll (their pollster’s got an A rating) just dropped, and it’s got Biden 31, Warren 20, Bernie 10. Plus HarrisX’s daily update was slightly favorable to Warren, though since it’s a C-rated poll, it didn’t move the needle much.
Anyhow, new numbers:
Biden 30.2
Warren 17.6
Sanders 15.9
Harris 8.2
Buttiegieg 5.1
O’Rourke 2.5
Booker 2.1
Everyone Else < 2.0
I’ve decided what I’ll do about phasing out older polls: once the midpoint of the time a poll was in the field is more than two weeks ago, it’ll drop from the average. So for instance Survey USA and Quinnipiac were both in the field from Aug. 1-5, midpoint 8/3. So on 8/17, they’ll still be part of the average, but on Sunday 8/18, they disappear.
I should really have a way to phase them out gradually, maybe by reducing their weight each day until hit hits zero, but that’s more complicated than I’m ready to make this.
One last update before the SurveyUSA and Quinnipiac polls drop off at midnight, per my previous post: we had a new Pew poll (rated B- by 538, much to my surprise, maybe horse race polls aren’t their strong suit) and the daily HarrisX update which didn’t change much.
The changes from two days ago are pretty minor:
Biden 29.8
Warren 17.4
Sanders 15.6
Harris 8.4
Buttiegieg 5.2
O’Rourke 2.3
Booker 2.0
Everyone Else < 2.0
Looks like a pretty solid list and solid rankings to me.
Looking at the dates on the various polls, several of them are now outdated by my ‘two weeks since the midpoint of when they were in the field’ rule: not just A-rated Survey USA and Quinnipiac, but B-rated Ipsos and Public Policy Polling, and C-rated Change Research.
Without them, it’s not all that different, though Biden and Sanders come in slightly lower now.
Biden 28.1
Warren 17.5
Sanders 14.0
Harris 8.7
Buttiegieg 4.3
O’Rourke 2.7
Booker 2.3
Everyone Else < 2.0
The polls that are still current enough that I’m using them, FWIW, are A-rated Fox News, B-rated YouGov, Morning Consult, and Pew, and C-rated HarrisX (August 13-16 vintage).
How are you accounting for state-level primary issues, RT? I defer to your expertise in statistical analysis - I’m good but I know you’re better - but using national-level polls can’t really be indicative of the eventual outcome, can it? Using 538’s state-level stuff we see kind of a different - at least slightly - story in Iowa. And that can impact the other three early ones.
For example, here in South Carolina, at the state level Biden’s up big. But if he underperforms - even with a win - in Iowa that support could decay very quickly.
I don’t see RTF’s numbers as a power rating per se so much as a good measure of one line of information.
“Power” has to heavily weight Iowa’s numbers, then other early states, followed by the long term strength in the numbers and directionality (Sanders floating down over time means that a same number is weaker than Warren on the slow long term rise), on the ground organization that will get voters to the polls and caucuses, and likely ceiling of support. Maybe even add in the strength within various voter demographics. (Clinton’s relative popularity within Southern Black voters doomed Sanders.)
The national popularity contest metric is nice to have in a more meaningful form but likely matters less on its own than the others do.
If I thought I knew enough to try to include state-level primary polls, I’d give it a shot. But I’d barely have a clue on how to do that.
Like I said, I’m doing something pretty limited - producing a weighted average of candidates’ standing in national polls - because (a) I know I can produce a much better average than what RCP is doing, and (b) 538, which could surely produce a better average than mine, isn’t making theirs public if they’re doing one at all.
Like DSeid said, it’s a good measure of one line of information. And I’m not claiming it’s any more than that.
What I can claim about it is that, whatever its pluses and minuses, it’s not in any way a reflection of my moods and biases, which I have in the usual supply. I’ve set down the rules for which polls I’m including, and how I’m weighting them. I think I’ve done so clearly enough that anyone so inclined could replicate my work, and come up with the same averages for each of the candidates.
Since Sunday, we have a new CNN poll, and I dropped Pew off because I probably never should have included them in the first place: their poll was in the field from July 22 - August 4, so by my ‘two weeks past the midpoint’ rule, I should have dropped them by August 12, which was before they even showed up at 538. Also, we have weekly updates from Morning Consult and YouGov, and daily updates as always from HarrisX. (Current polls: A-rated CNN and Fox News, B-rated Morning Consult and YouGov, and C-rated HarrisX.)
The numbers:
Biden 28.6
Warren 16.2
Sanders 15.2
Harris 7.2
Buttiegieg 4.7
O’Rourke 2.7
Booker 2.5
Yang 2.0
Everyone Else < 2.0
FWIW, I think Wednesday will be my regular day for updating the averages. Since the new numbers of Morning Consult and YouGov, which update weekly, usually show up on Tuesday and Wednesday, and other polls (excepting HarrisX) update much less frequently, Wednesday seems to make the most sense for that. I’ll update in between if there seems to be a good enough reason for it.
Numbers at sites like Predictwise or Betfair may be more useful than raw polling numbers, since predictable trends are already factored in. Here are the latest odds from Predictwise:
Democratic Nomination
32% Warren
23% Biden
13% Sanders
10% Harris
6% Buttigieg
4% Yang
2% Booker
(1% each - several)
Democratic Control, 2021
72% House of Reps
52% White House
32% Senate
Betfair numbers are close enough to Predictwise’s that I won’t show them separately. However they also quote Next Pres odds. We might expect these to be half the nomination chance. But in fact Harris is unlikely to win if nominated, while Yang is likelier than even-money, if nominated. — Misleading since Yang’s popularity in primary and general will be correlated.
This is very cool, although I think it would be even better if you used a numerical scale that accounted for pluses and minuses.
Missed edit window.
I noticed in RTF’s latest version, Beto and Pete are closer to each other than Pete is to the candidate above him. So they should either be both considered part of the top group or both not.
Thanks!
Yeah, I considered a number of improvements, but I had to make it simple enough that updating the spreadsheet wouldn’t be too much of a PITA. After all, I’m doing this in my copious 
spare time.
Improvements I’d do if this was my day job:
- Take the pluses and minuses into account, as you suggest.
- Factor sample size into the weighting. (Up to a point, anyway - what to do about Morning Consult’s ~17,000 sample size?)
- Downweight older polls gradually, rather than dropping them off abruptly after two weeks.
None of these improvements will be forthcoming. Sorry about that. 
I’m gonna hold off on this week’s update until tomorrow morning, because today is the last day for releasing polls that could make a difference in who qualifies for the September debate, so we’ll probably see at least a few new polls today. (We’ve already had two new ones today from Quinnipiac and Suffolk.) Being a morning person :D, I figure it makes sense to wait and include any polls that are released after I go to bed tonight.