If IBD (A- rated by Nate Silver) bombs on November 8th, please don't talk to me about 538 grades

[molova]

I'd like to request that when the 2018 midterm or 2020 Presidential elections approach, anyone who posts a "This-poll-is-great-because-Nate-Silver-gave-it-a-good-grade" type of thread or post will be ignored by me, if it turns out that IBD/TIPP (A- rated by Nate Silver's fivethirtyeight.com) either picked Trump to win or got the margin wrong by 5+ or more.

Some people pretend that 538's grading system has somehow been empirically tested or correlated to election accuracy.

Thanks in advance.

[michiganman1019]

1. Quora has also said that the LA Time poll is statistically sound also as well as IBD

Quora has also validated both of these polls.

I am not voting for Trump but I want the polls to reveal what they reveal without bias.

[Foggyhill]

5. Right.... Validated... That's utter tripe, especially for fracking LATime one.

The fact you bring on those two polls, that are brought on constantly
by right wings trolls all over politico, yahoo, youtube, etc.
and feel the need to actually mention your not a Trump voter... Here... of all place.
Makes me think you may well be a troll.

If the sample is non representative, the stats are meaningless.
This is the part that f*up most often.
Simple as that.

There is plenty of problematic methodology in the LATimes poll
and Quora certainly would not be the place I go for the ultimate answer about anything.


The fact that those polls off from all other polls and even state polls taken at the same time,
telll you there is a serious problem there.

[leftynyc]

20. Hillary is ahead in

that LA Times poll also as of today. The trump trolls are having every lifeline taken from them day by day. It's delicious.

[Maru Kitteh]

53. But you are not voting for HIllary?

If you are voting for Hillary, please clarify.

[msanthrope]

2. You seem concerned. nt

[alcibiades_mystery]

4. ROFL

Seriously.

[molova]

22. Do you think the IBD poll will be accurate?

?

[alcibiades_mystery]

34. I don't think...

...I'm interested in discussing things with you.

[ScienceIsGood]

58. You are a bully! Any poster with less that 1000 posts you insult. And when....

Asked for proof they are trump supporters, you have nothing!

[Blue Idaho]

80. Me thinks thou do protest too much... nt.

[ScienceIsGood]

84. Nope, he does it a lot. N/t

[molova]

21. That makes no sense

My thread shows I'm not concerned because I don't trust pro-Trump outliers.
Think before you type concern cliches.

[Demsrule86]

40. There can be 100 polls good for Hillary but

a new person comes here and finds the two polls that are not as favorable...this makes folks suspicious.

[ScienceIsGood]

59. I think people need more evidence than one poll doubt. N/t

[molova]

67. I criticized IBD. Learn how to read

Finding a poll and criticizing that poll makes me bullish toward Clinton. Be less reflexive and think before you type.

[Demsrule86]

93. Did you read the replies?

It brought out every concern 'person' who loves to have a safe way to say...why this poll is accurate and thus bash our nominee...and a word...don't tell people they can't read...I went on an extended vacation for saying something similar. You have nothing to worry about this with me...but someone reading this could have a different opinion. I do agree with you about Silver he has been off this year.

[Adrahil]

3. Please read what those guys actually write

As Harry Enten of 538 has written and tweeted, the grade refelcts past performance and methodology. It does not mean they are right NOW, but it does indicate you shouldn't reject their results simply because you don't like them.

Place them in context. Imclide them in the averages.

[Foggyhill]

6. Actually, I think Nate is wrong about that and I don't respect him because he says such things.

Once a sample is non representative (which is demonstrated by its incredible variance and how much it is off from every other poll),
you can't unskew it or make any sense of it, the way he does.

It has nothing to do with liking or not liking, it gives no usable info.
The stats part becomes meaningless.

Collating polls with different methods (that itself varies) is already a pretty iffy proposition,

I don't even agree with the performance assessments in general.

There are not enough results from past polls to compare to actual results to have a high confidence in the fact those ratings are in any way meaningful. The ratings don't even seem to correlate with the soundness of the polling methodology.

[uponit7771]

35. +1

[mythology]

45. You really don't understand statistics

Here's a hint. You can't say a poll is not representative because it differs in results from others.

Every poll has a confidence interval. Generally it's at 95%. Thus on average, every 20 polls will have a sample that doesn't match for whatever reason. A sample could not match the electorate, it could happen to hit an abnormally high percentage of minorities supporting Trump, etc.

To claim that more data points isn't more likely to be correct flies in the face of evidence. Averaging the polls is just a variation of wisdom of the crowds.

[Foggyhill]

50. Have done 5+ univ. Stat courses (McGill)

Plus a high pile of other math
I am an computer engineer who also did
Physics engineering for 2 years before switching majors.

You did not understand what I said when I said the poll was not a representative sample of the underlying pop.

If your poll purports to poll the USA general pop but you call rural land lines in the middle of the day, you are sampling THAT pop and nothing else.

The stats part will be true for the population you actually sampled, but produce false result for the general population you wanted to know about (this is why people try to unskew polls once the methodology bias is known. But that produces results with an unknown margin of error since the extent of the bias itself is hard to gauge

Getting a good unbiased random sample is extremely hard and most polls, especially small ones, fail

Actually really read what I write before going into a aggressive and unwarranted remedial stat rant...


There are many polls, especially national ones that give little real info or none. At best you'd get trends if their bias is well defined

[Adrahil]

56. Well, you are free to start your own poll aggregator...

I prefer his method. I think it works. But feel free to reject whatever you like.

[molova]

23. How many IBD polls did 538 consider?

?

[RAFisher]

39. 17

http://projects.fivethirtyeight.com/pollster-ratings/

[molova]

68. Have you heard of more than one IBD poll?

Vote two accurate IBD polls please.

[IronLionZion]

7. Been here long?

[molova]

24. I don't get it

I criticized a pro-Trump poll and you get mad.

[elleng]

54. Bullying isn't only a repug attribute, unfortunately.

Sorry.

[IronLionZion]

62. If IBD bombs on November 8th, I won't won't talk to him/her about 538 grades

I won't do it in 2018 or 2020. OK?

I very rarely get accused of bullying. It's kind of exciting.

[IronLionZion]

61. Welcome to DU!

[molova]

69. You are 3 months late

#fail

[IronLionZion]

90. You've been here 3 months

and telling us not to post something 2 years from now.

#winning

[molova]

91. You failed, now you change the subject

You wecomed a member who's been here for 3 months. Don't try to save face. Your fail is there to stay.

[IronLionZion]

92. You win

Have a drink of your choice:

[Democat]

32. Lots of angry new DUers in this thread

Angry that Clinton is winning?

[apcalc]

8. Does anyone know the slant on IBD polling?

Last edited Mon Oct 24, 2016, 07:38 AM - Edit history (1)

I assume there may be one or some methodology that steers them toward Trump.

[Bernardo de La Paz]

11. IBD

[apcalc]

17. Thanks! Corrected!

[Adrahil]

57. Polling error....

can emerge of from sampling error, weighting, or population models, among other things.

The main reason you can see very large differences in polls with relatively small MOE's is that population models are different.... that is, they make different assumptions about who is actually going to vote. Many (but not all) of the polls more favorable to Trump use turnout models that are pretty dated. They assume electorates that look like 2014, or 2012. Then there's the Dornsife poll which has a completely different set of issues.

[Cicada]

81. IBD preweights for party ID

Most polls let those polled tell the pollster what their party ID is and the poll doesn't fix the sample to the "correct" party ID. I think that leads to error. On the other hand most polls seem to be undercounting whites this year. The exit polls for 2012found the electorate 72% white. That is probably too low. The Census survey got 74% white. And if you use actual voting register figures it was 76%. Now all three methods have errors but actual voting register records are probably best. Many pollsters are weighting whites, after demographic changes, lower than that 2012 76% warrants.

[dangin]

9. 76%

It's only accurate 76%. The IDP/TIPP.

So methodology might be A- ,but it's success is a solid C.

[NobodyHere]

10. How do you separate success and methodology?

I guess I can see bad methodology succeeding by having a good guess but how does good methodology get bad results? Doesn't that mean the methodology is flawed?

[Bernardo de La Paz]

13. IBD

[apcalc]

19. Thanks, updated!

[Bernardo de La Paz]

29. Welcome, but I don't think you are user "dangin" and their post is not updated. . . nt

[oberliner]

12. An A- puts it in the middle of the pack

Many of the polls that they use are A+ or A.

For instance: Selzer, ABC, Fox, Monmouth, Survey USA, Wash Post, Marist, FDU all have higher rankings than IBD.

[Stallion]

65. Right-There Are About 28 Polling Firms rated Higher and ALL of Them Have Clinton Leading

[molova]

70. Prove it

Most polls are below A-

[oberliner]

73. Here ya go

http://projects.fivethirtyeight.com/pollster-ratings/

I count 16 A or A+ polls with 10 or more polls analyzed.

[molova]

79. I found 260 with a grade below A-

Therefore an A- is nowhere near being in the"middle"of the pack.

[oberliner]

83. There are only about 50 total

Don't count the ones that have only been polled once - there are a ton of them, and they are all pretty dodgy (and Nate doesn't use them).

Look at the ones that have been polled at least 10-15 times.

[molova]

85. Don't be slick. The number is 75, worse than A- with over 10 polls conducted

Don't give me no "about 50" false number.
By the way I'm not gonna let you raise the bar from 10 to 15. You keep raising the bar when you find yourself in trouble.

[NurseJackie]

14. IB-Who? Never heard of them until today.

Are they important?

[Grey Lemercier]

87. IBD is Investors Business Daily.. a very RW (editorially) version

of the already RW Wall Street Journal .

[NurseJackie]

88. Thanks.

[aaaaaa5a]

15. The only way to do this is to aggregate all of the polls.

Looking at any one poll by itself is foolhardy. Anyone can get a bad sample or have turnout model methods that can be off.


It also doesn't do anyone any good to look at polls where your candidate is doing better than she actually is, just like vise-versa.

[Bernardo de La Paz]

16. People are confused by the IBD and LA Times TRACKING polls.

1) They are tracking polls, not regular polls.

2) So therefore there can be an observation effect (polling equivalent of Heisenberg Uncertainty Principle).

3) Once the sample is made early on, nobody is added. So when the voter registration shifts, which is part of the dynamic of the election, that is not included.

4) Sampling demographics include certain assumptions which may not match other polls. Perhaps they used 2012 or 2008 or 2004 voter turnout.

5) Past performance is no guarantee of future results.

6) Out of dozens of polls, several will come closer than the others simply by luck. Run three elections (as IBD has if I recall), and one of those will come close in three elections just by luck.

It's the same principle as mutual funds. Set up 100 mutual funds chosen by throwing darts at the S&P 500 stocks. In the first year 50 will out perform the average. In the second year, 25 of those 50 will outperform, more or less. In the third, 12 of 25. In the fourth 6 of 12. In the fifth year, 3 of those 12 will outperform the average. Those three will have outperformed the average five (5) years in a row! But we should not conclude that they have some secret formula that gives them great insight.

7) I'm sure every member of DU has noticed that this election is not a usual one.

[LAS14]

43. Thanks for answering the OP question. nt

[Bernardo de La Paz]

46. You are welcome, but the Original Post did not pose a question. . . . nt

[LAS14]

47. Ah, you're right. So thanks for injecting rationality into the discussion.

I lost track of the path of the thread.

[Bernardo de La Paz]

49. I endorse rationality. It seems under-rated. Thanks for supporting! . . nt

[Bernardo de La Paz]

18. Regardless, Nate Silver has greater knowledge & insight about polling than me & I dare say you. . nt

[molova]

25. So I guess the IBD poll will be accurate

And if your are gonna tell me that grades only reflect past performance, then we are going to agree that these grades are useless.

[Bernardo de La Paz]

28. No. You are exhibiting a blind spot on this whole issue and not thinking clearly about it.

You don't like the IBD poll result because it doesn't have Hillary in the lead. I don't like that either.

From there our thinking diverges.

You jump to the conclusion (if I may paraphrase and exaggerate but not by much) that Nate Silver's & 538's system and methodology are worthless. On the basis of one polling organization and an election that hasn't yet finished, you reject everything 538 does.

[ProudToBeBlueInRhody]

60. Have to been to 538 lately?

Go there.

What is Hillary's chances of winning?

How many electoral votes is she projected for?

What is her popular vote projected to be?

You are obsessed with one poll and holding it against Nate Silver. It's a tracking poll and it methodology is questionable. The 538 staff will likely downgrade them after the election.

[Chathamization]

41. Sam Wang's been better this season. Silver gave Trump a 2% chance of winning the Republican

primaries (even with him leading in the polls) and wasn't sure whether or not the Democrats should start panicking a month ago. Wang said he was expecting Clinton to do well because he was expecting a regression to the mean. Silver said that that was one point of view, but not one he was sure was correct. It's looking like Wang won this round. Wang also uses less "secret sauce", so he has that going for him.

[truebluegreen]

27. "I'd like to request...

(a) thread or post will be ignored by me...."

Sounds...reasonable?

[Ace Rothstein]

30. They've moved into a tie today in this poll.

By the time the election rolls around they'll show Hillary with a healthy lead. They've done this before.

[Roland99]

31. From what I could tell, IBD/TIPP equally weights four regions of the country.

Clinton is up between 1-3% in the Northeast, the West, and the Midwest, while Trump is up about 8% in the South.

Their site also shows Trump with only a 20% chance of winning the election.

The electoral votes are not in the South (other than Florida). The bulk is on the coasts where Clinton has a lead.

It's quite misleading of them to say they have Trump up a few % simply because he has a larger lead in one of four regions.

[Roland99]

33. Actually! Today's updated IBD/TIPP poll has Clinton up 0.1%!!! 41.2% vs 41.1%

His lead in the South dropped from 8% down to 4%. His lead in 65+ years old is down to 1%

[Reece2076]

36. What poll are you referring to? According to this link it's tied.

It's a tracking poll that had trump ahead by 1 point last time and now it's tied.
http://www.investors.com/politics/ibd-tipp-presidential-election-poll/

[Johnny2X2X]

37. Rasmussen is the only poll giving Trump a lead

And we know they are a disgrace.

[ScienceIsGood]

42. Sure, because Nate really is an idiot. N/t

[LAS14]

44. Thanks, Molova. And shame on those who...

... accused him/her of "concern," or otherwise dumped on the OP. Also shame for dumping on Nate Silver. He simply listed the statistics of an interesting look at the polls. It wasn't his "opinion." He just shared the stats. DU ought to be a place where we can look behind the headlines and get insights from other members.

I recommend reply #16 as a really good response to the OP.

[Dem2]

48. Agreed

Most of the respondents don't appear to understand or want to spend the time having a reasonable discussion about polling.

[stonecutter357]

52. you seem very concerned.

[LAS14]

55. Oh, for crying out loud. I wish "concerned" had never become a catch all...

... for "you've asked a question and you should only be posting dumps on Trump or raves for Hillary."

[molova]

66. Exactly. The OP is actually optimistic toward Clinton

But the butthurt crowd loves to type "concern".

[Grey Lemercier]

86. this^^^^^^

[bravenak]

63. Very concerned

[molova]

71. Nate didn't just list statistics

He handed out grades.

[LAS14]

76. Yeah, but the grades are calculated, in part, on...

... past success. And you can't deny that they were super successful in 2012.

[stonecutter357]

51. I'm not concerned at all.

are you!

[Joe941]

72. At this point it doesn't really matter...

its the electoral college that matters.

[bravenak]

64. I'm sure folks are just chasing you down to discuss the validity of Nate Silver's analysis of polls

[NCTraveler]

74. When math goes past 2+2 it often becomes very difficult for people to grasp. nt.

[LAS14]

75. The thing to do is to see how they're rated AFTER Nov 8.

It's silly to accuse Silver of bias. Since IDP is a tracking poll, if they happened to start with a pro-Obama crowd in 2012, they were lucky and had the lowest variance from the results. If they happened to start with a Trump leaning crowd, they're stuck in every poll they take. I'm guessing their variance is going to skyrocket this year.

It does make me curious about how #of polls is weighted when dealing with a tracking poll, which is typically (I think) taken much more often than other polls.

[RandySF]

77. He'll lower the grade if it bombs.

[LAS14]

78. Another way to think about tracking polls...

... is that when they change you know there's a very high probability that people are actually changing their minds, rather than this poll just happened to get a more pro-somebody group than the previous one. Thus, when IBD/TIPP shows Clinton now up by one, you can have a high confidence that people are changing their minds.

The other possibility is that some Trump people disappeared for some reason or another.

[eShirl]

82. me: "If Inflammatory Bowel Disease (A- rated by Nate Silver) bombs on November 8th..."

[0rganism]

89. i interpret Nate's grade to be the accuracy of the poll in picking out *trends*

Nate handles the "house effect" with a normalization factor, so a poll's tendency to lean one way or another overall doesn't affect his forecast model much. his grades reflect the accuracy of the poll *after* the house effect is considered.

[Dem2]

94. Hillary is +3 in this poll today