538 Graham / Harrison poll

[Shermann]

538 has it at 77% Graham. This doesn't appear to be accurate as it seems to be neck-and-neck everywhere else.

https://projects.fivethirtyeight.com/2020-election-forecast/senate/south-carolina/

None of the polls they cite are anywhere near that:

https://projects.fivethirtyeight.com/polls/senate/south-carolina/

Soooo what's up with that?

[Sherman A1]

1. Never discount

Republican shenanigans.

[Trumpocalypse]

29. FYI

[ahoysrcsm]

33. I've been meaning to ask...

Why do you keep replying to this poster with the "FYI" and this video.

I'm getting tired of seeing it.

[NorthOf270]

34. I'm guessing they don't like that posters signature

They should probably either complain to an admin or let it go.

[SharonClark]

2. 538 isn't a poll, it's a prediction based on polls

and other factors including past voting patterns and incumbency.

[Shermann]

3. I get that, but you can't start with polls that are 50/50 and then forecast 77/23 nt

[hack89]

4. You have to look at the bias of each polling company

Silver looks at the historic accuracy of each polling organization to determine if it has a bias one way or the other. He then corrects for that bias in his model.

You have to dig deep into his site if you want to full understand what he does.

[Shermann]

5. Or...I can post a thread here and hope for the Cliff's notes version

[hack89]

7. Which you just got. Nt

[Shermann]

9. Nine polls are cited

...and they are all clustered together.

Adjusting for bias simply doesn't get you to his forecast. Sorry.

[hack89]

13. Did you look at the details of his model?

or do you need that spoonfed to you as well?

[Shermann]

14. We're at the engineering smell test stage right now

Spoon feed away, so far I'm inclined to fail the smell test and see no need to dig deeper.

[hack89]

17. Don't think so

If you really cared you would put in the effort yourself. You have seemed to have made up your mind so I will leave you to it.

[Happy Hoosier]

19. Well, if you are an engineer....

You should know that your intuition can be wrong. The model considers more factors than the polls. It considers voting patterns, the likelihood an incumbent benefits from late deciders, etc.

If you just want to gut it, that’s up to you.

[Fiendish Thingy]

24. Sure it does, you just don't understand it

538 has Biden 87/Trump 12 even though Biden’s lead is averaging 9-10 pts. They aren’t projecting Biden will win with 87% of the vote, they are projecting Biden has an 87 in 100 chance of winning, by whatever margin.

Same goes for Graham - he is neck in neck with Harrison (but not 50/50 odds), and the model, which incorporates other factors besides polls, including incumbency, gives Graham a 77% chance of winning.

[SharonClark]

11. What polls are you looking at because the latest

Sienna/NYT poll has Graham up by 8. 538 has a greater than 3 point popular vote difference as of this morning.

You need to dig deeper into Nate Silver’s methodology if you want to understand how he does predictions.

Like the weather, predictions change when the conditions on the ground change. Harrison has the money to get his message out. Now he needs great GOTV to pull this off.

[Shermann]

16. I'm looking at the nine cited in the link in the OP nt

[mucifer]

6. that really sucks. Lots of racists there.

[dixiechiken1]

8. Might Lou Dobbs' marching orders throw a monkey wrench into this race?

[nycbos]

10. Partisan lean of the state is factored in?

[SharonClark]

12. Yes, that is a factor.

[SharonClark]

15. If you want a simple look into 538 then look

at the Presidential race in Iowa which 538 declares a toss up and has Biden and Trump at 50-50 prediction. The section you should check out shows the factors that go into converting poll results into predictions.

The section I like to see is the change of prediction over time, especially when a long red line drops below an increasingly long blue line.

Hope this helps.

[onenote]

18. 538 is not predicting Graham gets 77 percent of the vote

It’s predicting that there is a 77 percent likelihood of Graham winning.

[Statistical]

20. It isn't saying Graham gets 77% of the vote it is saying 77% chance of Graham winning.

SC is deep deep red. I donated to Harrison but he always was and is a longshot.

538 uses a lot of factors not cherry picking the best poll for Harrison and saying it is up 1 so 100% chance Harrison wins. It looks at the traditional lean of the state, the bias in the polls, the amount of undecideds and predicting how they will vote. It looks at the amount of third party and traditionally if Dems/Rep have overpolled in the state. It doesn't consider all polls equal. The higher rated polls and the newer polls are given higher weight. It also starts from a projected likely outcome (in this case Graham) and wants to see lots of pols ideally beyond the MoE the other way to flip the outcome. Graham as a favorite has declined from 89% to 77% but he still is the favorite.

At best the SC race will be nail biter with Harrison barely squeaking in a win.

[GreenEyedLefty]

21. This. I also donated to Harrison.

Also realizing he is a long shot. My donations were a way of sticking it to Graham. I have to say I have enjoyed watching him sweat and grovel for money.

[Shermann]

23. Understood

I accept that a real lead in the poll data will get amplified to a larger percentage of winning in a forecast, amplified even further as election day approaches.

There is no lead for Graham in the raw poll data above the margin of error, so I'm skeptical that his 50% forecast advantage here isn't based in part on amplifying noise.

[jcgoldie]

22. 538 uses more than just polls to establish their probabilities

They have a model they call “light” is is strictly a polling average which also adjusts for house bias. But their “deluxe” model also uses other information like demographics of the race in order to establish their win probabilities.

[brooklynite]

25. ECONOMIST model has Graham at 60% probability

[Shermann]

26. I would accept a 60% forecast projection right now, nothing greater nt

[Southern lefty]

27. I have no faith in any of that.

I don't believe any poll, prediction or any of the such thing period. A hard strong ground game getting everyone to vote blue is what we have to be persistent about. The Trump tards are out in force, and I'm terrified of a repeat of 2016!

[Shermann]

28. I agree with getting out to vote

In terms of macro-level planning, however...it's critical to know where to spend money and where to cut your losses.

So reasonably accurate models are a useful tool for managing a campaign. But there are no guarantees here.

[obamanut2012]

30. What is a "tard"?

[uppityperson]

31. Did you know that "tard" is short for the offensive word "retard"? Please edit

it out as it is not acceptable.

[tandem5]

32. His model is designed to smooth out noise

so simply put it wont reflect much change immediately even if there's a real shift in sentiment.