Thread by @ThyConsigliori

28 October 2020 #BetweenTheLinesDotVote Analysis

So, Here's A Backdoor - Go Figure!

How do you not love @dbongino? He posted this story, yesterday, written by @johnhawkinsrwn. I have to tip my hat in respect. I didn't find this stuff!

bongino.com/data-from-usc-…

2) I know, I always say I'm going to stop charting - monitor but not chart - Dornsife and then they always pull me back in. I'm such a data addict, I can't help myself. But, I'm going cold turkey today. No charts, and the article above is why. I'll explain.

3) Oh,. I almost forgot! Great thanks out to @mflixx who sent me this story, yesterday. You're the man, Matt!

Now let's turn to the irritating part. Look at this headline:

"USC Dornsife Polling Predicts Another Trump Victory Next Week"

I did NOT know that. That's irritating.

4) Below, we'll go through some of the points in the article in data. Let me give you the high level overview now. The one chart - out of how many that they post? - that we've been following covers the exact data discussed in making this article's call.

5) It is a genius question. Who do you think people in your state will vote for? No matter what the person has stated as their own intention, by using a surrogate you take all the pressure off, and you likely draw nearer to their own vote preference than by any other path.

6) But that's not the entire high level story. What they discovered is that, by segmenting their sample into states, they could generate an Electoral College prediction. That, my friends, is a seismic shift in the polling world. How much do I hate them now (by which I mean love)?

7) It has long been one of my biggest beefs that no matter how unsullied, and broadly based, say between 1,500 & 3,000 respondents, no matter how righteously random, if you don't lay it under the Electoral College count, how meaningful is your prediction? They solved this!

8) So, the bottomline is that, while they're calling the popular vote for Biden - and by a smaller margin the Clinton 2 won - they're calling the Electoral College for Trump. Dude! That's GIGANTIC! Let me just bow in honor to them for their mighty prowess, for a painful moment...

9) I'd ask why this is NOT splashed all over Drudge, why we're not hearing about it all day long on FOX, and why even the MSM doesn't have this break in for a 30 - 60 second segment...but I obviously know the answer. Truth is, this is just their backdoor because they know.

10) And that really is part of why I'm pissed off again, right now. The measly, meager chart that I found does NOT show the data put through this test. They show the right question, and I found it! Hurray for me. But, they're NOT presenting that question this way.

11) All their data, over this very question that I've been fighting and fighting with to try to tease out a glimpse, a glimmer of the truth...is merely targeted at the Popular Vote and NOT at the Electoral College. So, maybe that chart is there and I didn't find it. But...

12) I haven't found it. My temptation is to blame them, but it really could be me. Maybe they are publishing this data and I don't know yet. Maybe. My gut tells me, nah, they have this data, Hawkins provides us their thoughts on it, but they're not giving me the data this way.

13) Let all that sink in, if you can. Dornsife is calling for a Trump victory. Without Bongino publishing Hawkins' article, I'd have no idea of that. Would you? Biden gets less popular vote that Hillary, but Trump wins again. Shouldn't we all know about that? I really am pissed.

14) Okay, enough at the high level, let's follow the article a bit more granularly.

The first thing we're told is there are two new questions, called social circle question. There probably are, but Mr. Hawkins never makes it clear what the 2nd one is. Bad form.

15) We next dive into Dornsife's own explanation. People, they learned, have a really good idea about how their friends, family, neighbors, and coworkers, etc., will vote. You ask them, and they'll get it right. That is, I say again...

GIGANTIC!

16) We then learn that when they averaged their data so procured, it offered a kind stunningly good picture of how the entire nation leaned. Wow!

Here the story gets a bit murky for me. They tell us they've tested this question over 5 election cycles. Say what?

17) Believe it or not, even I have limits to my digging into the weeds discipline. Maybe they told us all about this back in 2016. And maybe if I were more diligent, I'd know that they tested this in 2017's French Presidential election, and 2018's Swedish Parliamentary contest.

18) Temper problems show up again, overwhelming irritation, I couldn't care less about Dornsife in the 2018 House election, as I knew we were going to blow it, and I felt like I just couldn't do anything about it. I knew the consequences of losing the House, as well.

19) Well, full confession, I have a certain antipathy to State level politics. I prefer the national level, by far. I'm wrong about this. I will soon get myself over it. But that's part of why I didn't follow, even though I knew about it, Dornsife's 2016 work.

20) Looking again, I see I skipped over 2017's Dutch Parliamentary election. Alas. Egg on my face. I should have followed their work better than I did. I mean that. This is very important work, I say again. Honor to them.

21) You have to read this quote in full:

"In both the U.S. elections, the social-circle question predicted national and state level results better than the “own intention” question in the same polls.

22) "...In fact, data from the social-circle question in 2016 accurately predicted which candidate won each state, so it predicted Trump’s electoral college victory."

I'm going to start repeating myself here, jaw on desk in awe. By gosh and by golly, they've done it.

23) More irritation. At BetweenTheLines.Vote, we figured out how to do the exact same thing, and, believe it or not, I think we can do it better, if we ever actually do it. But, they ARE doing it and I am unable to express how transformational their leadership here is.

24) We're not going to follow Trafalgar - he doesn't give us much meat there anyway - but let's look at the conclusions he meekly indicates:

"If Trafalgar & USC Dornsife call this election correctly while the other pollsters whiff, it will revolutionize the way polling is done."

25) I hope by now you can see the "backdoor" part of this. Go look at Dornsife's charts, and return to their 19 August releases and you'll see 100% in the tank for Biden. Countless of my own analyses have shown that again and again. They're not buying what they're selling.

26) They've been playing fast and loose with the data this year since day 1, and I do not, cannot forgive that. Yet, hidden away, somewhere the Hawkins finds, is a simple, clean call, Trump will still be 45, Biden will NOT be 46. How about that?

27) I know it's a short article, but Hawkins fails to discuss sampling bias, and far worse, strategic design corruption. There are countless ways to place thumb on scale, countless. The corruption of all this needs constant spotlighting. Constant.

28) Corrupt pollsters not only design questions in order to elicit the desired answers, they absolutely know how to and DO find those people most likely to give them the answers they want. That's no mere failure of polling. It is a complete betrayal and corruption of it.

29) And last, when, when, oh when will get over this ridiculous theory that averaging polls gives us meaningful information? When you start out with mush push, add more, average it all, you have even more mush push. Bad polling is not improved by averaging, it's made worse.

30 Again, garbage in, garbage out...even more garbage in, even more garbage out. Let's face it. Once and for all, let's face this.

31) By the way, I like Tom Bevins of RealClearPolitics. As you just saw, i couldn't possibly disagree with his polling approach more. But he's a great guy, fine analyst, a fine journalist, absolutely well informed, thoughtful, and a decent fellow as I see him on TV.

32) Oh, fundamentally there's something else. I see NO polling about the irreparable split between the Democratic Party's left and middle. We will stop fracking. We will NOT stop fracking. On and on. Why don't we hear anything about that in polling?

33) If any of you research types are better at finding Electoral College filtered data at Dornsife, oh please, please, hit me up with it. It really just hurt my eyes this morning looking at the right question factored the wrong way. And what does "Poststratified by State" mean?

34) In closing, I have to say, if I was a Biden supporting Democrat, and I read this article, my head would be exploding even more. It's not supposed to be this tricky. The data is supposed to be solid, and presented solidly.

35) A revolution in polling is coming. The ability to calculate the data against the Electoral College is beyond transformation. Social predictions are where it's at. I should be happier. Maybe I soon will be. How about you?

Thread ends at #35.