The Gallup Likely Voter screen may not be the Problem

[cthulu2016]

Last edited Sun Oct 21, 2012, 03:44 PM - Edit history (1)

Sine Gallup has a consistently outlying tracking poll result for likely voters it is natural to assume that there is something wrong with their likely voter screen.

And a lot has been written about Gallup's likely voters (including by me). But the problem, whatever it is, is probably deeper.

With four days post second debate in the 7-day sample Romney shouldn't be ahead by three among registered voters.

If you say it is Romney 52%, Obama 45% (today's Gallup) among likely voters then that involves a prediction of who will vote on election day, and it is what it is. Your method of prediction may be right or it may be wrong.
.
But if you say it is Romney 49%, Obama 46% among registered voters that does not involve any prediction. It is a simple measurement of the sort that polling is usually very good at.

And being an outlier on a simple measurement is pretty strong evidence of error.

If Romney is +3 among registered voters there is nothing odd about him being +7 among likely voters. A 4% gap is normal. A lot of polls more favorable to Obama have comparable or larger RV/LV gap.

The problem is, I have no reason to think Romney is ahead among adults, or among registered voters. No other evidence seems to support that proposition.

It isn't just Gallup's likely voter screen. The entire poll is flawed, somehow, from the start. (Or else Gallup is right and everybody else in the world is wrong.)

(This gives credence to analysts who have cited deep flaws, like a flawed demographic model that over-represents whites. http://en.wikipedia.org/wiki/Alan_Abramowitz )

[DemocratSinceBirth]

1. Even The WSJ/NBC Poll Which Has It Tied Among LVs At 47%-47% Shows Obama Leading Among RVs 49%-44%/

.

[cthulu2016]

5. Right. If Gallup applied their likely voter screen to the underlying NBC data set

then Romney would not be +7 among LV.

[DemocratSinceBirth]

9. Yeah. Obama Would Be Up One

.

[still_one]

2. As long as we get the vote out we win, forget the polls

[flamingdem]

3. It's "Fake Momentum" or Fauxmentum - generated by the MSM

and for all we know paid for by the right wing, indirectly, or even directly.

We need to find out what the mechanism is because something stinks

[okieinpain]

15. i agree you watch the news shows and it's all romney

but you look at the repug representives on these shows and they are not acting like they are in the lead.

[MidwestTransplant]

4. The other problem is the approval numbers. Obama is at 49%. That's nearly a lock

for reelection. How can he under perform his approval by so much?

[ProSense]

6. Gallup is a

horribly outdated poll making bullshit assumptions.

http://www.democraticunderground.com/1251160863

[CreekDog]

7. could be a problem with the geographic distribution of interviewees

as well as the age distribution.

a random sample needs to be truly random within the universe of registered voters.

[cthulu2016]

11. I added a link to the OP that

suggests that Gallup weights to an erroneously high level of white voters.

[CreekDog]

12. hard to say but it could be simple like geography

you could have the region weighted incorrectly, you could have other geographic problems like weighting of certain states more heavily, while getting the overall region correct.

for example, say you were polling the west and you had the right proportion of the west in your poll, but you had weighted California 2x as much as Nevada and/or Arizona (when California is 5-6 times larger than Arizona and almost 20 times larger than Nevada).

[DemocratSinceBirth]

16. That Makes A Lot of Sense

To diverge from other polls by as large a margin as Gallup there has to be more going on than just a tight voter screen.

It's surprising with all the resources available they don't get it right.

[99Forever]

8. I don't comment on polls...

.. because I don't believe any of them are honest. Period. Just another tool to suppress the rage of the people, in case they figure out whats really being done in our name.

[Loudestlib]

10. Just turn out the vote

[reformist2]

13. Getting the white/nonwhite ratios wrong by 5% could lead to a poll error of 4%.

If you get Romney 49 / Obama 47 with an 80/20 split (as some are saying Gallup assumes), then when I extrapolate (with some guesswork) as to how a more realistic 75/25 split would look, I get Obama 49 / Romney 47.

In other words, getting the white/nonwhite ratios down is huge.

[phleshdef]

14. When is the last time Gallup has been really accurate? Whether it be LV or RV?

And how many times have they gotten both really, really wrong?

Their track record is very sketchy at this point. PPP, IBD TIPP and yes, Rasmussen (at least close to election time) seem to all be a bit more reliable.