July 24 Poll Report

July 24, 2019
 

July 24 Poll Report

 

            Good evening everybody.   I was traveling yesterday (and most of today) and did not get a poll report done; sorry.  That means I have two polls to report on, two deleted polls to acknowledge.  I don’t know that a lot happened in the deleted polls; I think we can pretty much skip those.   The poll results of July 22 can be summarized as "everybody takes advantage of Bill de Blasio."  De Blasio, the much-maligned mayor of the great city of New York, had been doing surprisingly well in my polls, rallying from a starting point under 50 to a Support Score in excess of 100.   Based on that, we expected him to draw 6% of the vote in a contest against Mayor Pete, ex-mayor Julian and he-might-have-been-a-mayor-of-something-sometime Tim Ryan.  He drew only 2%, with Buttigieg and Ryan picking up his support:

 

Scores

Castro

435

de Blasio

105

Buttigieg

975

Ryan

124

 

 

 

 

 

 

 

 

 

Predicted

Castro

27

de Blasio

6

Buttigieg

59

Ryan

8

Actual

Castro

25

de Blasio

2

Buttigieg

63

Ryan

10

 

            Castro was below expectations in that poll but is up in the Support Scores anyway as the result of the removal of the old poll.   In our poll of July 23 President Trump, who had been doing much better in my polls, fell back to where he had been, being routed by Michael Bennet, who is a serious man but would but marginally qualify as a serious candidate:

Scores

Trump

606

Bennet

196

Bullock

93

Schultz

168

 

 

 

 

 

 

 

 

 

Predicted

Trump

57

Bennet

18

Bullock

9

Schultz

16

Actual

Trump

29

Bennet

39

Bullock

17

Schultz

25

 

            The predictions for that poll, based only previous polling, were only 34% accurate, making that the most surprising poll result since I started doing this in April. 

            That result bumps up the scores of Bennet, Bullock and Schultz, for which I kind of have to apologize.    We can see that what happened is that the Trump supporters just weren’t there on that day, which means that whoever else happened to be in that poll group would get an inflated score for the day, which means that their inflated score for that day will be with us for a while.   To an extent my system copes with that; it is just one poll among 50, the outcome of it has limited ability to distort the Support Scores from what they should be.  Limited, but not zero.   Anyway, the significant movements in the scores in the last two days effect only six candidates:

            Michael Bennet is up 40 points as a result of finishing first in yesterday’s poll.  

            Julian Castro is up 29 points as a result of the elimination from the data of the poll of June 2, in which he barely beat Jay Inslee and John Hickenlooper (31-30-30). 

            Steve Bullock is up 22 points as a result of his misleading performance in yesterday’s poll; not commenting on his merits as a candidate.

            Bill de Blasio is down 14 points as a result of doing poorly in Monday’s poll. 

            John Hickenlooper is down 14 points as a result of the removal of the June 2nd poll from the data now considered relevant, and

            Donald Trump is down 66 points as a result of yesterday’s poll.  

            These are the updated Support Scores:

Rank

First

Last

Support

1

Elizabeth

Warren

1867

2

Pete

Buttigieg

969

3

Joe

Biden

901

4

Kamala

Harris

815

5

Donald

Trump

537

6

Julian

Castro

461

7

Amy

Klobuchar

461

8

Andrew

Yang

442

9

Bernie

Sanders

397

10

Cory

Booker

343

11

Beto

O'Rourke

328

12

Kirsten

Gillibrand

287

13

Tulsi

Gabbard

269

14

Jay

Inslee

262

15

John

Hickenlooper

246

16

Michael

Bennet

235

17

Bill

Weld

218

18

Howard

Schultz

169

19

Tim

Ryan

136

20

John

Delaney

120

21

Steve

Bullock

116

22

Marianne

Williamson

97

23

Bill

de Blasio

90

24

Seth

Moulton

82

25

Mike

Gravel

82

 

            We’re still in that phase, post-debate, where more support is drifting toward some of the weaker candidates than is going to the stronger ones.   Eleven candidates are marked in green (up 33%), while none are marked in gray (down 25%), because the losses of Biden, Buttigieg and Warren are not highly significant to them, but represent a significant percentage increase to those who pick up that support.

            Thank you for your interest.

 

 
 

COMMENTS (12 Comments, most recent shown first)

bjames
LesLein
In the second table the Actual row has the following numbers:

Trump - 29%
Bennett - 39
Bullock - 17
Schultz - 25

This adds up to 110%. The calculations need to be redone.


It's just a typo--25 for Schultz should have been 15. The correct number was used to do the calculations which effect the standings.
12:28 AM Jul 27th
 
LesLein
In the second table the Actual row has the following numbers:

Trump - 29%
Bennett - 39
Bullock - 17
Schultz - 25

This adds up to 110%. The calculations need to be redone.
6:21 PM Jul 25th
 
OldBackstop
" We can see that what happened is that the Trump supporters just weren’t there on that day, which means..."

Bill, what does that mean? I see you are saying it was an anomaly/outlier, but any theories?
1:22 PM Jul 25th
 
Guy123
Conveniently, Bill has just put up a new poll that will provide a great test of the model. In the completed poll he writes about here, Trump got 29% against a field with a combined "strength" of 457. In this new poll, he faces three vastly stronger candidates (Biden, Harris, Yang) with combined "strength" of 2,158 -- almost five times as strong. If Trump again gets his usual 21-29% of the vote, surely *that* will demonstrate the model does not provide meaningful predictions. If facing opponents five times stronger has only a small impact on Trump's vote, the model can't be accurately measuring strength.
11:58 AM Jul 25th
 
OldBackstop
In terms of pragmatic value, you can't slip blithely between the HoF poll....a debate which is static and long term, and the POTUS polls, which are going to have an incredible roiling over the next 8 months.

Bill said he plans to cut the period from 50 to 40 days.....but look at the pace of events....candidates in, out, debates, gaffes, endorsements....40 days ago is going to be as different as the launching and landing of the ark.

In the next, say, 32 weeks you will have five more announce and, by Super Tuesday, 20 drop out. If Sanders drops out (he won't, he has been a deadender zealot for 40 years,) you can't just apportion his support in the current ratio....those voters will largely go to one candidate or another closer to their socialist goals. Likewise Biden....if he goes down, those voters won't move to Kamala. Joe Rural Six Pack ain't going with busing and reparations.

The way you forecast those eventualities is by polling for "second choices." But Bill's method actually does the opposite: forces choices between candidates they might never actually support (i.e.: no "none of the above"). Any issue-based scenario data is utterly buried, or at least unmined. I guess that is the same thing..

There is a way to do this, but it is big bucks and the serious research will be going to the states that will decide it. The national count is about as valuable today as it was to Hillary on election night.
11:03 AM Jul 25th
 
tangotiger
www.twitter.com/tangotiger/status/1137568737962471424
10:09 AM Jul 25th
 
tangotiger
I posted on twitter, so I'll reiterate here:

As I mentioned last month, DJT is going to get 25-30% against the top D or middling D. You have now shown he will get the same score against lower tier D.

Your scoring system is introducing a systematic bias that requires every candidate to now face DJT in order to cancel out.

***

Joe, I made that same case last month regarding Clemens if facing Moyer or Oswalt or Carpenter. No matter who he is up against, Clemens gets 85% among my voters, and the other guy gets 15%.
https://twitter.com/tangotiger/status/1137568737962471424

But if I were to create a poll of only Moyer, Oswalt, Carpenter, in no way would they split the vote as the Clemens polls would suggest.

***

Guy is right, that this is (at least) a 2-dimensional problem that Bill is treating as 1-dimensional.

10:08 AM Jul 25th
 
Guy123
The predictions for that poll, based only previous polling, were only 34% accurate, making that the most surprising poll result since I started doing this in April.
This result was "surprising" only to Bill's model; it was entirely predictable and in fact predicted. We know that about 29% of Bill's followers will support Trump over any and all Democrats (though some of this 29% will defect to Weld or Kasich when those choices are offered). The only question is how the other 71% will allocate their votes, which depends on which Democratic candidates are offered. When a model misses by 28 points on one of the leading candidates (after 105 polls), the model is clearly broken. (And the possibility that Trump voters suddenly abstained -- in a high-turnout poll in which Trump was an offered candidate! -- can be safely dismissed.)

The model fails because vote preference is not a function of a single "candidate strength" variable, as the model assumes. Instead there two dimensions (at least): 1) partisanship and 2) candidate strength. And partisanship is the much stronger force. Candidate strength does matter a LOT in primary contests, but only a little when voters choose between candidates of opposing parties. Consider Warren and Booker: the model says that Warren is about 5x "stronger" than Booker. And Warren indeed has a commanding lead over Booker in all Democratic primary polls. But in general election polls, Warren and Booker each have a 3-point lead over Trump on average (and every other Democrat is within a few points of this margin). There is no single measure of candidate strength that can reconcile such results. It is simply not meaningful to say that "Warren is five times as strong as Booker" unless you first specify whether you mean "among Democrats" (true) or "among all voters" (false).

The model can probably still yield some interesting information, but only if Trump is included in *every* poll. Then we can see how the non-Republican voters evaluate the non-Trump candidates. But if Trump continues to move in and out of the pool, the data will unfortunately continue to have little value.
8:55 AM Jul 25th
 
joedimino
I mentioned this the other day in a comment, but the system needs to account for candidates that get standard support no matter who they are up against, like Trump. There has to be a way to account for ‘consistency’ or whatever you want to call it. Trump,will get 20-30% no matter the people he is up against. This poll makes that obvious. It’s not a deal breaker, but it is a flaw.

I wonder if something like this will happen with the Hall of Fame voting too. My guess is the otherwise marginal #2 candidates who are ‘lucky’ enough to be in the polls with Clemens and Bonds will come out higher than expected because they got 15% of the vote while being up against such great candidates. The system has a hard time with these kinds of things.
7:20 AM Jul 25th
 
BarryBondsFan25
Great sense of humour placing Castro (appropriate last name given his positions) and the Che Guevara quoting (in Miami no less) Di Blasio in the same poll. Had Bernie Sanders and Joe Biden been in the poll, we could have called it the Red Division.
6:37 AM Jul 25th
 
OldBackstop
Bill, why do you think you were 28 points off on the president? In a stable sample, meaning one you could learn something from, I would think any up-to-speed observer could get within 6-8 points....he is the president :-)

Also, just to reiterate, why the totally different language in the poll question every day? Two veteran members on Reader's Posts are saying this entire project, and HoF, is actually trying to glean results from the different way you word the questions.
2:01 AM Jul 25th
 
chrisbodig
There's something in the numbers here that instantly struck me as really odd, that Trump's predicted vote was 57%. I think I get it that the predicted 57% was because he was against lower rated candidates (much lower rated). Still, if somebody had given me an over/under bet of Trump at 50% in this poll, I would have bet thousands of dollars on the "under," not betting my entire life savings only because of fear that Laura Ingraham would re-tweet the poll.

Trump has gotten between 19% and 29% in 15 BJOL polls, obviously a variance based on the competition and who might be re-Tweeting. Still, he's always in that range. It doesn't matter whether you match him up against top tier Dems or bottom tier Dems, he's always getting something in the 20's.

You could poll Trump against any three random unknown Dems and he's still probably going to be somewhere in the 20's. Trump's support is mostly static; people love him or will vote for ANYBODY else.

I respectfully would suggest that having Trump in these polls isn't helping the project. As you've said many times, his numbers don't reflect his true standing in the USA. His presence in the process just confuses and throws a wrench into the standings of the 50 billion Democratic candidates. After all, it made you kind of apologize for the inflated numbers of Schultz, Bennett and Bullock.

My two cents. Thanks again for doing this.

11:44 PM Jul 24th
 
 
©2019 Be Jolly, Inc. All Rights Reserved.|Web site design and development by Americaneagle.com|Terms & Conditions|Privacy Policy