Newsweek's misleading polls oversampled Democrats a few weeks ago and oversampled Republicans now

[highplainsdem]

I posted this information as a reply in another topic, but decided to post it as an OP as well, so people don't overreact to the current Newsweek poll showing such a great change from Obama's wide lead over McCain last month. Both polls were off, with last month's oversampling Democrats and this month's oversampling Republicans.

Here's what I posted in that other topic, with links to the details of the polls:

_______

Here are the internal numbers for this new poll:

http://www.newsweek.com/id/145556

1037 Registered voters (plus or minus 4)

315 Republican (plus or minus 7)
324 Democrats (plus or minus 7)
357 Independents (plus or minus 6)

Most surveys show a larger percentage of registered voters identifying themselves as Democrats.

It also looks as though they oversampled older voters, and white voters.

So my best guess is that both polls were off.

I trust the tracking polls more, and they show Obama leading by either a few points (Gallup, usually) or several points (Rasmussen).



These are the internals for last month's Newsweek poll, the one showing Obama with a wide lead:

http://www.newsweek.com/id/142469

896 Registered voers (plus or minus 4)

231 Republicans (plus or minus 8)
324 Democrats (plus or minus 6)
307 Independents (plus or minus 7)

I think they definitely oversampled Democrats last month.

They had a lower percentage of white voters for last month's poll, too.

[Kurt_and_Hunter]

1. Average the two and it's 6 points, which is pretty much in line

Anywhere from 3%-9% seems reasonable.

I think of it as a five or six point deal right now.

[jody]

2. Given the sample size it's not obvious that any conclusions are biased by sample size. Any

beginning statistician knows what model to use to deal with such things.

[NattPang]

4. It is not the sample size alone at issue

it is the numbers polled and included in each categorized sample,
compared to same categories in reality.

e.g., If there is a 50/50 division in gender category in the real world
but the sample size obtained uses a 30/70 breakdown in gender category,
than the results of the poll are not going to reflect reality.

beginning statisticians know this.

[jody]

6. And professors of statistics know how to deal with such things.

Have a pleasant evening, :hi:

[NattPang]

8. LOL!

you said nothing.
But have a nice night.

[writes3000]

3. Thanks for this, highplainsdem. I really appreciate the analysis.

Great of you to post it.

[Hokie]

5. This is why the margin of error is BS

Polls often quote MOE's of +/-2.5 % as if that is the end all. As these two polls show the methodology and the makeup of sample group can greatly increase the real error. It was obvious the first poll was out of whack with every other poll at the time. Thanks for posting this explanation.

[On the Road]

7. That is a Good Catch and is Posisbly What Happened

Although:

(1) Pollsters are able to re-weight samples in order to get a more representative view. In a major poll like this, it would have been a good idea.

(2) Weighting by political affiliation is fraught with peril. People change their reported parties more than you might think. Some pollsters like Zogby don't believe it should be done at all. So while it can be done, it may not improve accuracy and can sometimes make it worse.

[Seen the light]

9. Or, since the poll was conducted in the middle of the FISA debacle, well

Let's just say that it doesn't look from this poll like that was such a politically savvy move on Senator Obama's part after all.

[robcon]

13. Huh? You think the FISA issue gained votes for McCain?

I don't understand the logic of that.

The FISA issue was only important to the left. The left is going to vote for Obama. Obama lost nothing because of FISA.

[mrreowwr_kittty]

19. I seriously doubt if you stopped 10 people in the mall, that any would know what FISA is. nt

[robcon]

20. Not only that, McCain and Obama voted ifentically on FISA

Not an issue that will get or lose votes in any case.

[Captain Hilts]

10. They are sloppy in balancng their sample. But they're honest about it. nt

[sufrommich]

11. I'm dumb when it comes to taking apart polls,but

aren't a larger percentage of people self identifying as Democrats now?

[highplainsdem]

15. Yes, and Newsweek polled more Dems than Rethugs in both these polls,

but last month's poll included a larger number of Dems than they should have had, and this month's included fewer.

Rasmussen had a recent article on why polls often show different results, and it explained their own polling methods:

http://www.rasmussenreports.com/public_content/politics/election_20082/why_polls_sometimes_show_different_results


I posted a topic here last month about that article

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=132x6412875

and it sank like a rock, with no replies, all the while people overreacted to poll numbers.

[sufrommich]

16. Interesting info.

Sorry your thread sunk.

[DeepModem Mom]

12. I think Newsweek needs to seriously examine their polling. nt

[Skidmore]

14. Nothin' like keeping a horserace neck to neck.

I'm so sick of the media gaming elections.

[jefferson_dem]

17. R 30%, D 32%, I 30%?

No fucking way.

Check this out from PEW --



Growing Democratic Party ID Advantage

The Democratic Party's advantage in party identification among voters nationwide has continued to grow in the first half of 2008. Among all the registered voters interviewed by the Pew Research Center from January through June of this year, 37% identify themselves as Democrats, the highest annual average in 20 years. Meanwhile, just 28% say they are Republicans which, coupled with the 2007 average, is the lowest in more than 16 years. The nine-point Democratic advantage in party identification represents a substantial shift from the 2004 election cycle, when Democrats outnumbered Republicans by just two percentage points (35% to 33%).

The share of voters who identify as independents or offer no partisan preference is also greater this year than in most recent elections.

http://people-press.org/report/?pageid=1340

[mrreowwr_kittty]

18. Looks like they undersampled Dems this time.

Latest polls show 40% identifying as Dem, 29% as Repub, and 31% as Independent. Not sure which poll because I saw it streaming on the ticker on MSNBC. But if that's accurate, in a poll of 1000 adults there should be 100 more Dems than Rs.