Gallup is at it again! 12 point Repub bias in sample

[kstewart33]

From a great political blog, The Left Coaster:

Gallup Is At It Again - Yesterday's National Poll Had 12% GOP Bias
Gallup has done it again. After supplying CNN and USA Today with a poll two weeks ago that showed a double-digit Bush lead amongst likely voters that turned out to have a significant bias in its sample favoring the GOP, Gallup did it again yesterday.

Except that yesterday, they not only did it again, they apparently felt that a 7% GOP bias wasn't good enough. So they perpetrated the same fraud upon the media (including their partners CNN and USAT) and voters and this time used a 12% GOP bias in their likely voter screen. I kid you not.

Here is the text from the email I got from Gallup this morning outlining the party ID breakdown in their likely voter samples from their two most recent national polls:

Likely Voter Sample Party IDs – Poll of September 13-15
Reflected Bush Winning by 55%-42%

Total Sample: 767
GOP: 305 (40%)
Dem: 253 (33%)
Ind: 208 (28%)

Likely Voter Sample Party IDs – Poll of September 24-26
Reflected Bush Winning by 52%-44%

Total Sample: 758
GOP: 328 (43%)
Dem: 236 (31%)
Ind: 189 (25%)

(snip)

Link: http://www.theleftcoaster.com

Soooo, if Bush is up by eight in this poll with a 12 point Repub bias, looks like Kerry should be leading.

[lancdem]

1. Wow!

I knew from http://www.emergingdemocraticmajority.com that there was a 10-point GOP edge in the RV sample, but a 12-point edge among LVs is even worse.

[Rose Siding]

2. A fraud on CNN and USAT?

I don't think so. If that were true, both those powerful media outlets could fry Gallup. The three of them are perpetrating a fraud on the public.

[HEIL PRESIDENT GOD]

25. It would be hard to make the case for fraud...

Considering the poor quality of statistics even in social science journal articles these days. A jargon-speaker can justify any introduced bias to a panel of non-jargon-speakers; in this case, it's easy, using the "likely voters" formula.

[WiseMen]

3. Hope You are Right. BUT science behind methods are not usually visible.

I think it is foolish to be constantly critizing polls rather
than learning from them and acting to change public opinion.

[RobertSeattle]

4. Let's hope Gallop goes the way of that Magazine in 1936

that prediced a big (R) win over FDR and FDR won by a huge landslide and the magazine soon went under.

[StrongbadTehAwesome]

22. I'm sure this happened for similar reasons, too

I work in psych research and surveys are a BIG part of what we do. That magazine that predicted such bad numbers in 1936 did so because they got their sample by calling people. Well, in 1936 not everyone had phones. Those that did were disproportionately wealthy and hence, disproportionately Republican.

I'd bet anything that Gallup isn't calling the right percentage of cell phones to land-line phones ("right" meaning what's actually in use), but even if this isn't the case, it's Gallup's responsibility to find out why their numbers are coming out all wonky. If I took a survey with numbers like that to completion and tried to publish it (assuming I wasn't laughed at and fired by my boss first), it would be rightly rejected by every peer-reviewed journal in existence as bad science.

[Flubadubya]

29. Common parlance around here says that cell phones...

are not called at all, only land lines. If that doesn't skew a poll I don't know what would.

[llywrch]

39. You mean the Literary Digest?

There's an interesting story here.

The Literary Digest, at the time part of the national media establishment, sent out a questionnaire to 10 million people by mail. (They didn't use the telephone, since in those days not only did most people lack one, all non-local calls had to be placed thru a live human being known as an "operator". It would have been prohibitively expensive to call even a few thousand people, many of whom would have refused to talk about their voting choices with a nosy & anonymous stranger!)

Now from these 10 million questionnaires, they got 2.3 million responses -- a pretty nice response -- from this response they believed Landon would defeat Roosevelt in a landslide. No one doubted them, since they had correctly predicted every winner for the last several elections.

Of course, history turned out differently.

A young kid named George Gallup had conducted a much smaller poll using a sampling of people created on scientific guidelines, & he correctly predicted Roosevelt as the winner. Gallup later examined the Literary Digest data, & said that they had correctly reported the results -- but that their readership did not fairly represent the population of voters.

It's ironic that Gallup made his reputation in the 1936 election from carefully, methodic work, while his son is carefully & methodically working to destroy the family reputation in this election.

Geoff

[StrongbadTehAwesome]

45. hmm...they didn't use phones at all, you say?

(quickly Googles)

Ah, ok. Turns out they did a mailing, but complied their address list from telephone directories and magazine subscriber lists, which would also skew results toward the rich (and apparently make me incorrectly remember them calling people).

[RobertSeattle]

46. Thank for the info - that is a great story.

I didn't realize the Gallup angle.

[MallRat]

5. HOLY SHIT.

This is a fucking joke. Gallup should be the subject of ridicule and derision.

Media blaster, anyone?

-MR

[RoundRockD]

10. But, what about the ABC/Wash. Post Poll?

Didn't they have similar results?

And, although we know that the polls are unreliable, shouldn't we be concerned about the message that these polls are sending to the gullible American voter?

I can't wait until the election is over!!

[LisaL]

14. They do the exact same thing.

Poll more republicans than democrats.

[rockymountaindem]

6. Let me see if I can understand this...

They poll a disproportionately large number of Republicans, and Bush is *still* losing ground over the last poll they did? Oh, this is tooooo funny...

[kcwayne]

9. Pretty soon they will be using 100% registered Republicans

in their sample to try to offset Bush's failing numbers. When the Whitehouse is cutting the big checks to get their propaganda out, they want results, dammit!

[LibeMatt]

31. Please delete

...accidentally posted twice

[LibeMatt]

32. Unless of course...

...said polls of 100% Republicans show Bush with only a 70-30 lead, then they'll probably just make stuff up.

[Wickerman]

7. It would also seem they are oversampling Indies

25% are independent? not sure about that.

[Lex]

8. Does Gallup have a rationale behind the 12 point Repub weighting?

So, they are polling 12% MORE republicans in their polling methodology than Democrats?

Why?

[shraby]

11. Simple...to keep Bush ahead.

[papau]

12. They say random RV's self-identify GOP at this rate. But in 2000 they

claimed the same thing - and were very wrong based on registration totals.

[LisaL]

18. I am sorry, but that is as fishy as can be.

Democrats have registered way more new voters than republicans, right? More Democrats than republicans actually voted in 2000, right?
Then how the hell do they randomly poll more republicans than democrats time and time again, both likely and registered voters? Something really stinks bad about their polls.

[papau]

28. The Pew result implies a massive drop in Dem party ID - I do not believe

it.


http://people-press.org/reports/pdf/226demographic.pdf
GOP 90- 3 (7)
DEM 10-81 (9)
IND 46-38 (16)

are split in voting by party,

but 48-40 has 12 not sure, which needs something close to 50% ind,

with say GOP at 26 and Dem at 24 (not exact as I did not actually solve - but close)

Not mathematically possible(?) with 95% confidence in a sea of 35/35/30 party split

[Spiffarino]

43. And you should not believe it

Here in NC, buckle of the Bible Belt, Dems are out-registering Reps by 10 percentage points and more.

[gasperc]

13. MyDD and others on on this, we need a media blast

I'm going to cull together historical data, Gallup's polls should get dis-credited.

[mzmolly]

15. This should be sent to major media and the Kerry campaign?

:hi:

[HFishbine]

16. My e-mail to CNN

Hello,

When reporting the results of your poll, please inform viewers of the sample:

GOP: 328 (43%)
Dem: 236 (31%)
Ind: 189 (25%)

For greater understanding, you might compare the sample to the 2000 voter turnout:

GOP: 35%
Dem: 39%
Ind: 27%

If you are unwilling to reveal to your audience that your current poll oversamples Republicans by 23% and undersamples Democrats by 21%, then I suggest you not report the poll at all.

[Lex]

19. That is an excellent and easy to understand breakdown of the situation.

Shows just how utterly bogus the whole scam is.

Hats off to you, HFishbine!

:thumbsup:

[farmbo]

17. If those idiots aren't calling cell phones or newly registered voters...

...or correcting for them...their poll is as useless as the Pope's gonads.

And if they're non-randomly polling 12% more GOPers than Dems (when Dems actually out-numbered GOPers in the last presidential election) they're engaging in a fraud.

CNN should be embarassed...but Bill "Osama Loves Kerry" Schneider seems to have gone off his feed this cycle and become a Rovian sycophant.

[MallRat]

20. I just fired off an email to CNN.

Appealing to their own self-interest... the "I hope Gallup doesn't make CNN look foolish" angle.

-MR

[HEIL PRESIDENT GOD]

27. Not sure about the media blast.

These b.s. polls are probably helping us more than Bush.

[nibbana]

21. Gallup will select questions and citizens for debate in St. Louis

We need to discuss this issue and let the media know what is going on here.

[Firespirit]

23. Whoa...

The way I see it, the weighting is designed to ensure that * does not appear to be in free-fall.

With a 7% GOP bias, the poll gave * a thirteen-point lead.
With a 12% GOP bias, the poll gave * only an eight-point lead.

What does that suggest -- no, scream -- to you???

He's losing ground in a big way. They ADD 5 percent to the GOP weighting, and he STILL loses ground!

[StrongbadTehAwesome]

24. I'm e-mailing Gallup right now, and CNN and USAT are next

Gallup doesn't accept e-mails over characters, so I'm trying to scale this down:

Your latest presidential poll admits to an oversampling of Republicans as compared to the actual proportion of voter turnouts (and of registered voters).

Yesterday's polling sample:
GOP: 328 (43%)
Dem: 236 (31%)
Ind: 189 (25%)

2000 voter turnout: thanks to HFishbine for the numbers

GOP: 35%
Dem: 39%
Ind: 27%

As a psychological research assistant, I understand the difficulties of obtaining balanced information that is truly generalizable to the public at large. If in my work I had inadvertantly oversampled one of three major groups I was studying by 23%, and undersampled another by 21%, that study would not be allowed to continue. If my boss was asleep at his post and the study was allowed to be completed, when we went to publish, it would be rightly rejected by every peer-reviewed journal in existence as BAD SCIENCE.

Gallup should be searching for reasons behind this discrepancy between self-reported political affiliations and actual numbers of registered/likely voters (perhaps you're not calling enough cell phones?). You should NOT be reporting these flawed numbers to the major news media.

[StrongbadTehAwesome]

26. The really sick thing about this

is that Gallup isn't even reporting this party ID breakdown on its own site. That is beyond dishonest in my opinion - every psych study includes a section full of information like this so readers can judge for themselves how well the findings would apply to the public at large.

I didn't realize this until I re-read the article referenced in the OP. "Here is the text from the email I got from Gallup this morning outlining the party ID breakdown in their likely voter samples..."
Sure enough, if you go to the Gallup site itself, you won't find these numbers anywhere in their 3 pages of results from yesterday's poll.

Edit: (to keep myself from posting yet AGAIN) I'm e-mailing Gallup again requesting that they start including this information in their results. It's the only honest thing for them to do.

[LocalGuy]

30. Latest Quinnipiac Poll for PA

I don't have enough posts yet to start a thread but here's a little good news
related to polling.
Sorry if I'm off-topic slightly.

9/28, Penna JK 49% - GB 46%, W led last week by 49/46 ....

<http://www.quinnipiac.edu/x11379.xml?ReleaseID=406>

[Lex]

34. Looks really good. Thanks for posting it LocalGuy.

[Spiffarino]

42. Well well well...

Looks like GallOP is trying to keep the rats from abandoning ship. W's poll numbers flipped in a week!

[jefferson_dem]

44. Thanks for the GOOD NEWS on PA, LocalGuy. And WELCOME to the community!

:hi:

[LibeMatt]

36. This all ties in...

...with the current GOP hatred and distrust of science.

After all, these days the truth is like kryptonite to a Bush supporter.

[Sick_of_Rethuggery]

33. According to CNN

The electorate (from exit polls) was:

Democrats: 39%
Rethugs: 35%
Indies: 27%

[batmankm]

35. My fear.

There was a post either here or over at Kos that pointed out that Rumsfield is on some of the Nixon tapes admitting to having an unethical relationship with Gallop, that the contact there and he had gone to college. Clearly they still have influence over Gallop not to mention the other Corprate polls. My fear is that all the manipulation of the "corporate" polls, like Time, Gallop, CNN etc is to insure that when they steal the votes using the black boxes that the polls will provide cover for them. They can maintain that its not so crazy because all the polls showed the shrub leading.

KM

[Shopaholic]

37. It's psy-ops

I remember a long ago Political Science professor in college saying that polls were important because people wanted to vote for a winner. If the polls showed one candidate was way out ahead of another, some people would be swayed to vote for them. Meanwhile, people inclined to vote for the other candidate might not vote at all, because they'd feel "why bother"? That's what's going on here. The polls are skewed and they want us to sit at home, throw our hands up in the air and give up all the while trying to persuade people to drink the Kool-aid and vote for *. But they forgot one thing--the Kerry theme song "Never Surrender"!

[Jasper 91]

40. It's called 'The Halo Effect'

And that's what I first thought when I saw all these skewed polls , but now I think there is something more sinister behind it .

If the polls consistently showed Kerry leading , especially if it was outside the margin of error , there would be an almighty stink if Bush won . But , if he steals the election when the polls have showed him leading , they will say we are sore losers and that all the evidence had indicated that Bush would win . :grr:

[trogdor]

38. The polls are rigged!

from http://www.republicansareidiots.com>Republicans Are Idiots!:

But you knew that already. From Electoral Vote:

"It is becoming increasingly clear that the pollsters are producing the results that the people paying the bills want to hear. Even pollsters who were once thought to be above suspicion are now suspicious. Gallup, for example, is now normalizing its samples to include 40 percent Republicans, even though the 2000 exit polls showed the partisan distribution to be 39 percent Democratic, 35 percent Republican. There is scant evidence that the underlying partisan distribution has changed much since then. Other pollsters also normalize their data, but most don't say how. Normalizing the sample to ensure the proper number of women, elderly voters, etc. is legitimate provided that the pollster publicly states what has been done."

As a matter of fact, recent reports that the Democrats are having far more success in registering new voters puts the lie to what Gallup is doing. I haven't been linking to Electoral Vote because I'm skeptical of their methodology, and this is precisely the reason. Putting a jpeg on my page that says "Kerry 207 Bush 317", knowing that the data going into this estimate are badly flawed and possibly skewed deliberately is a bad idea, IMO.

So, what do we do with the Gallup poll? Re-weight it, of course. Gallup has been oversampling Republicans by 12 percent (1 - (35 / 40) = 0.125), so that should reduce Bush's 52 percent of the national electorate to about 46 percent. Likewise, assuming Kerry's share of the pool of those polled was shorted by a like amount, his 44 percent becomes 49 percent, which puts Kerry ahead by three points - either just inside or just outside the MoE, depending on whether you see the glass as half-empty or half-full respectively.

Of course, we won't know for sure who's full of shit, Electoral Vote or the Gallup organization, until Election Day, but when Kerry holds up a copy of USA Today that reads, "BUSH DEFEATS KERRY" as the President-elect, the Gallup people will look rather foolish, to say the least.

[Spiffarino]

41. I heard about this from a Dem campaign worker

It was the first time in my life I ever heard people shout with joy over a poll that we were losing.

Yiiippppeeeee!!!!!