BILL JAMES ONLINE

July 25 Poll Report

By Bill James
July 26, 2019
 

July 25 Poll Report

            Good evening everybody.

            You know that thing that happens when a minor leaguer is hitting .325 with power so he gets called up, but he doesn’t make the grade his first shot and winds up back in the minors?   That’s what has just happened to Julian Castro.   Castro got good reviews from the June 27 Democratic debate, and (presumably as a result of said debate) had been climbing rapidly in the polls.   Climbing in the polls is the polling equivalent of hitting .325 in the minors.  Castro got 44% in one of my polls recently—against John Delaney, Kirsten Gillibrand and Michael Bennet.   When that happens in my polls, the randomized algorithm that tends to pit strong candidates against other strong candidates promotes you to a higher level, so Julian Castro was matched up yesterday against Elizabeth Warren, Bernie Sanders and Pete Buttigieg. 

            He got crushed. 

            It’s not the end of the world.  Mike Schmidt hit under .196 as a rookie.  Willie Mays was hitting .477 at Minneapolis, started his major league career 1-for-26.  Jim Thome in 1992 hit .205 with a .299 slugging percentage in 40 games, was back in the minors for most of the 1993 season.  The Mariners kept sending Edgar Martinez back down to the minors for several years. 

            Anyway, these are the results of yesterday’s poll:

Scores

Sanders

397

Warren

1867

Buttigieg

969

Castro

461

 

 

 

 

 

 

 

 

 

Predicted

Sanders

11

Warren

51

Buttigieg

26

Castro

12

Actual

Sanders

12

Warren

53

Buttigieg

29

Castro

6

 

            The prediction for the poll was 87% accurate.  The prediction was basically perfect, except that Castro got half of the support we expected him to get, which knocked his Support Score in a cocked hat.  Since yesterday:

            Elizabeth Warren is up 41 points as a result of yesterday’s poll.

            Pete Buttigieg is up 25 points as a result of yesterday’s poll. 

            Bernie Sanders is up 15 points, but

            Julian Castro is down 66 points.

            The deleted poll (June 4) is basically a non-event, doesn’t really affect where anybody is.  These are the updated Support Scores:

Rank

First

Last

Support

1

Elizabeth

Warren

1908

2

Pete

Buttigieg

994

3

Joe

Biden

909

4

Kamala

Harris

826

5

Donald

Trump

537

6

Amy

Klobuchar

460

7

Andrew

Yang

445

8

Bernie

Sanders

412

9

Julian

Castro

384

10

Cory

Booker

345

11

Beto

O'Rourke

330

12

Kirsten

Gillibrand

285

13

Tulsi

Gabbard

266

14

Jay

Inslee

255

15

John

Hickenlooper

247

16

Michael

Bennet

231

17

Bill

Weld

219

18

Howard

Schultz

167

19

Tim

Ryan

135

20

John

Delaney

116

21

Steve

Bullock

114

22

Marianne

Williamson

96

23

Bill

de Blasio

88

24

Seth

Moulton

82

25

Mike

Gravel

82

 

            Thank you for your interest. 

 
 

COMMENTS (17 Comments, most recent shown first)

tangotiger
Suppose we had just Mayor Pete, Williamson, and DJT.

In a head to head, we would get
: Pete and DJT, predict 27% for DJT
: Williamson and DJT, predict 27% for DJT
: Pete and Williamson, Pete would get 91%.

To minimize errors, we'd have to allocate these points:

Mayor Pete keeps 1000 points
Williamson keeps 100 points
DJT has about 90 points

Now, we'd have to predict this:
In a head-to-head of Pete and DJT, we'd (have to) predict 8.2% for Trump.
In a head-to-head of Williamson and DJT, we'd (have to) predict 47.4% for Trump.
In a head-to-head of Pete and Williamson, Pete would get 91%.

As you can see, the math doesn't match reality. Adding 20 other D just serves to obfuscate, but the point remains that DJT can't support the math, because the math requires independence.




6:52 PM Jul 26th
 
tangotiger
Myles:

The main issue is this: you can't use those standings point to predict anything, if it includes DJT.

DJT with 3 lesser D candidates Bill had "predicted" to get 57% of the vote.
DJT with the top 3 D candidates Bill had "predicted" to get 14% of the vote.

Unsurprisingly he got 27%, give or take a few in both races. This is not an under- or over-performance. It's exactly predicted. We all keep saying DJT will get 27%.

DJT is nothing more than a systematic bias to this process. And unnecessary.
5:49 PM Jul 26th
 
OldBackstop
@myles. Bill is resistant to running any ping tests to test his water. Trump vs Warren, then Trump vs Biden. Do you consider yourself a Dem, Rep or other. Region of country. Anything. Trump approval? If a YN poll on Trump was 70-30, you would have som baseline to discuss. A forced choice of four will always leave room for tap-dancing.

My gripe here is that he is positioning this as election based material but asking very different questions of a group likely chockful of people who are not primary voters.

Polling the general public slapdash is more about "popularity" not "support".
4:51 PM Jul 26th
 
OldBackstop
@myles. Bill is resistant to running any ping tests to test his water. Trump vs Warren, then Trump vs Biden. Do you consider yourself a Dem, Rep or other. Region of country. Anything. Trump approval? If a YN poll on Trump was 70-30, you would have som baseline to discuss. A forced choice of four will always leave room for tap-dancing.

My gripe here is that he is positioning this as election based material but asking very different questions of a group likely chockful of people who are not primary voters.

Polling the general public slapdash is more about "popularity" not "support".
4:51 PM Jul 26th
 
myleshandley
@tango

The more I read your post, the more I think we are basically saying the same thing.​
3:47 PM Jul 26th
 
myleshandley
@OldBackstop

The criticism of self-selecting twitter respondents is a real one, but it's implicit in the method. That's just the way it's going to be - no way to fix it without great cost or effort.

Respondent rate doesn't seem a valid criticism, at least no more valid than criticism of more traditional polls. Called polls have vanishingly small response rates, very similar to the twitter polls (using the numbers you've guessed at).

I think Bill is clearly mistaken on the 34 accuracy day rationale. Unless he can show that the response rate on that particular poll was drastically different than on other polls (truly showing an surplus/surfeit of votes, with the thought that those votes would have went to Trump), I'm much more inclined to believe the alternative hypothesis.

My other post illustrates why I think The Trump "Problem" may be overblown.
3:46 PM Jul 26th
 
myleshandley
Tango, I don't think that's accurate. Because the system is iterative, Trump being in the poll doesn't have any outsized effects on the other people in the poll - they are competing in lower or higher-staked events based on the non-entity (Trump) in the poll, but the ratios of the other candidates are really all that matters.

If you had a poll with Booker, Castro, Delaney, and Trump, you might get something like 1400 points up for grabs in that poll, 537 of which belong to Trump. Even if Trump got 1200 of those points (or 100), you'd still give the remaining points proportionately to the other 3. In turn, they'd take those points with them to their other polling positions, they'd over (or under) perform relative to those contests based on how often they matched with Trump (and according to what mix of candidates were in those races).

In fact, you could literally assign ANY VALUE to Trump in ANY MATCHUP and the order of the other candidates would not change. Trump's value would change, but it would take (or give) proportionately from every other candidate. If the process wasn't iterative, or if it didn't converge to same stable state, you'd have huge problems. Since neither of those are the case, you can safely regard any candidate as an outlier. At least, I think that's right.

I feel confident that including Trump in these polls does not give an accurate reflection of Trump's standing among the general public - but we aren't measuring the general public. It probably DOES give an accurate reflection of Trump's standing among Bill's twitter following - the reason I think that is that the ratio between Trump and Bill Weld is much, much different than the gap between Trump and Weld in every primary poll yet commissioned. That is a powerful piece of evidence that the texture of the Billectorate is different than the texture of the electorate.

It feels weird to include Trump on the list because if you put him head-to-head against many candidates on this list, he'd likely lose (perhaps he would not lose against Bill Weld and lower). Most other candidates would probably beat, in a head-to-head matchup, most candidates below them). Also, if you had a twitter poll with 25 choice, Trump would likely win (though there is some gaming effect that might take place)! That said, Trump being on the list doesn't invalidate the list. The method accounts for it. You can safely remove him if you wish, give his points proportionately to the other candidates, and be on your way.

(I think this whole line of questioning is contigent on degrees of freedom quibbles one can have with the process - you need a suitably connected cohort for the rankings to really make any sense at all. I'm not sure we have that here, but I haven't rigorously thought about it).
3:40 PM Jul 26th
 
OldBackstop
@tango. That is a part of it. But the larger systemic issue is that these are being offered to well over 50k, maybe enormously over, and between 900 and 1500 are choosing to respond with a stab of the finger. The D/R (or racial, or gender, or voter/non-voter) bias on that swamps the logistical issues which you would have in proper sampling.

That is why the predictions are now being accompanied by comments like shock at a 34 percent accuracy day and analogies to anomalistic .325 hitters, or whatever.

If someone wants to write that all out more politely, I'd appreciate it.


1:46 PM Jul 26th
 
OldBackstop
@tango. That is a part of it. But the larger systemic issue is that these are being offered to well over 50k, maybe enormously over, and between 900 and 1500 are choosing to respond with a stab of the finger. The D/R (or racial, or gender, or voter/non-voter) bias on that swamps the logistical issues which you would have in proper sampling.

That is why the predictions are now being accompanied by comments like shock at a 34 percent accuracy day and analogies to anomalistic .325 hitters, or whatever.

If someone wants to write that all out more politely, I'd appreciate it.


1:46 PM Jul 26th
 
tangotiger
Right. If Bill wants to use all these polls as part of something holistic, such that the "strength" of DJT ultimately doesn't affect the strength of all the D representatives, that's fine. He's introducing a systematic bias, but will get largely washed away for the overall ratings.

But, if he then wants to PREDICT an individual poll, and say that against the low tier D, DJT is 57% and against the top tier D he is 14%, and all the while he gets 27% regardless of his competition, then that's the line being drawn.

He can't use it holistically, and individually.

The main issue is simply this: if A > B and B > C, then A > C. Not only that, but if A is twice B and B is twice C, then A is four times C. This is the math that is driving the scoring system.

But the D v R bias prevents this math from happening whenever you have D and R in the same poll. That because the overriding factor is that 27% of Bill's voters are R supporters, and will vote for DJT over the best or worst that the D have to offer. That's the first level driving force. Once you remove the D/R bias, THEN what Bill is doing is fantastic.

But as long as Bill keeps giving out individual predictions where the choices are a mix of D and R, then those individual predictions are simply faced against a systematic bias that we can all see, but Bill is not adjusting for.

10:31 AM Jul 26th
 
Guy123
Poll 107 has concluded, and Trump got 23% support against a very strong field. I believe Trump has been in four July polls (apologies if I missed one), and his support has almost no relationship to the "strength" of his opponents:

Trump Opponents' Strength
23% 2158
29% 457
27% 2645
25% 1674

Hopefully we can all agree now that these candidate strength ratings are useful only as a measure of intra-Democratic strength (assuming they are useful).
Regardless of their rating, every one of these Democrats would beat Trump in the BJO electorate, and by similar margins.

BTW, this is not a function of Trump being a uniquely polarizing candidate. Any incumbent president, matched against a field of contenders from the the other party, would produce similar results. If HRC has won in 2016, Bill's polls would be showing her with (guessing) about 60% of the vote every time, no matter which GOP candidates she faced. (And in that scenario, the ratings of her challengers would be far less useful than the current ratings, because they would largely reflect the preferences of Democrats.)
9:13 AM Jul 26th
 
LesLein
Yesterday I made a comment pointing out an error in the previous poll. Does that have any impact?
8:09 AM Jul 26th
 
myleshandley
I thought Castro passed the credibility test during the first debate. You can imagine him talking in front of a crowd of 5,000 people and holding their attention. You can imagine them arguing in a debate against the Republican candidate. Contrast that with, say, John Hickenlooper, who is an impressive guy I like very much. There's no chance he gets elected. He doesn't have the required charisma.
7:54 AM Jul 26th
 
MarisFan61
P.S. to those who've been suggesting things that might be better for what Bill is doing: You're assuming you know exactly what he's doing and what he's interested in about this. I'm still not taking it at face value. I don't think we know exactly what he's trying to get at.
6:18 AM Jul 26th
 
MarisFan61
BTW he used the wrong (opposite) term for the transgender type that might become pregnant.
6:08 AM Jul 26th
 
BarryBondsFan25
At the risk of once again triggering a BJOL reader post member who obviously did not find my Castro/Di Blasio comment yesterday amusing, Castro's biggest moment in the first debate was his call for access to abortion for transgenders. I'm guessing that didn't even go well amongst those in the Twitter universe- purely a guess though.
4:37 AM Jul 26th
 
MarisFan61
Did Castro really get good reviews from his debate?
Well, if we mean any, sure. But widely? Not that I saw.
You thought he (as you put it) looked like an ass, I did too, and my impression was that that's what most felt.
What he did accomplish was to get some notice, some greater "name recognition," because of his spar with Beto, in which he came out stronger than Beto but of course that wasn't saying much. It was like, Castro was an ass (for focusing on a dumb detail), Beto wilted under it; Castro "won," because he got a knockout, but it was against an apparent weakling, and it didn't undo looking like an ass.

In the general polls, he went from between 0-1% to 1-2%, where he remains. His temporary rise in your polling probably did reflect the debate, perhaps due just to increased name recognition.
1:06 AM Jul 26th
 
 
© 2011 Be Jolly, Inc. All Rights Reserved.