New TIME poll methodology

[featherman]

http://www.time.com/time/press_releases/article/0,8599,701890,00.html

"Methodology
This TIME magazine poll was conducted by telephone September 21-23, 2004 among a random sample of 1,178 adults throughout the U.S. The random sample includes 1,014 reported registered voters and 877 likely voters. The margin of error for registered voters is approximately +/-3 percentage points. The margin of error for likely voters is approximately +/- 4% points. Likely voters reported party identifications are: 36% Republican, 31% Democratic, 24% Independents. Registered voters party affiliations are: 36% Republican, 30% Democratic, 24% Independent. 52% of respondents were female while 48% were male."

Results: Bush 48 Kerry 42 Nader 5 Other/Und 5
Why is this not surprising?

[spooky3]

1. slow learners, eh? And why don't they report non-response rates

and age breakdowns?

[gospelized]

2. next poll

they will just ask Bush and Hannity who they think will win, and report that the election is over.

[Catch22Dem]

5. Bush would get the answer wrong. (n/t)

[autorank]

3. They are relentless!!!

How is it that of late the polls tied to the major media consistently over sample Republicans . This is like a prelude to 1984...we'll tell you how we tell the lies then after we've lied enough we'll simply tell the lie (not trouble you with the methodology). I think they should call these faith based polls and just tell us to shove it when we ask intelligent questions.

BOYCOTT THE MAJOR MEDIA

[grytpype]

4. 36% Republican, 31% Democratic, 24% Independents.

That tells the whole story right there.

These polls are rigged to make a Duhbya "re" election look inevitable.

Slate has a decent article on how Team Duhbya uses "inevitability":

http://slate.msn.com/id/2107259/

[They_LIHOP]

6. I hate to break it to everyone...

But the said 'oversampling' of GOP voters, IN AND OF ITSELF, absolutely DOES NOT make the poll invalid.

Please don't flame me, but I got multiple "A" grades in Stats in college, and I can tell you that this fact does NOT an invalid sample make. In fact, if you were to change your sampling to include equal numbers of each party then you would be f***ing up the poll, not the other way around.

If the sample is truly RANDOM, and a couple other assumptions can be validly made about the sample being representative (population data would form a roughly bell-shaped or 'normal' curve, etc) of the population, then all this 'skewing' tells you is that there are more people in the population that are republicans.

Now, if you could prove that the 36/31 ratio is WILDLY off the mark (and I'd have to do some analysis to tell you what 'off the mark' would have to be, but I'd say if the population was actually 40%D, 30%R or better ...) then you might be able to prove that the sample is not properly representative of the population (i.e. not properly random, basically).

But without KNOWING the D/R breakdown of the POPULATION, then doing the calculation to determine probability of getting 36/31 based on that probability, in and of itself it's impossible to judge whether the 36/31 ratio indicates a flawed methodology, or simply reflects the population ...

Sorry, but them's the facts...

[JohnnyCougar]

7. It is flawed methodology

Democratic voters are less likely to respond because they have more caller ID's and disproportionate number of "only cell phone" users. The results from the last 3 presidential elections are 38-40% Dem, 35% Repub. That means there is an 8-10% differential between those responding to the polls and those who actually vote. You are assuming that randomly calling people targets each political party equally. It doesn't. These telephone polls have over 75% attrition rates. Furthermore, studies have shown that Democratic party identification has gone UP in the last year. There is no way one can look at this as a random sample. Just none.

Give Kerry plus 7% and Bush -1% to make up the difference, and Kerry has the lead.

[Jackpine Radical]

12. That's right--in fact, the discrepancy

between the poll results & known voting results demonstrate the invalidity of the sampling procedure.

[DemFromMem]

16. You're both right (I think)

I think that it's possible that there has been some shift in party ID since 2000 AND polling undercounts the Dems for the reason JohnnyCougar mentions. My guess is that there's a slight inflation of Republican numbers in the polls, but not a terribly large amount. Less so in the registered voter polls than in the likely voter polls since it seems like first time voters are largely Democratic this year.

[bain_sidhe]

8. How about actual voters???

Election Turnout
Year............Dem.....Rep......Ind
2000............39%.....35%......26%
1996............39%.....34%......27%
1992............34%.....34%......33%

From Zogby
http://www.zogby.com/search/ReadNews.dbm?ID=859

[troublemaker]

9. You are quite right.

There are valid--even compelling--arguments against weighting samples by party based on past voting patterns.

But if a real-world party realignment has occured it's odd that only some polls detect it.

The only thing I know for sure is I'll be in suspense on election day no matter what the polls say.

[billbuckhead]

10. Divide the Independents in 2 and add them equally to the D or the R's

and you have the results of their poll. They didn't do it that way but they might as well have.

[autorank]

11. Not a flame...

I saw Zogby on the air a few days ago and he was dismissing these wide-margin polls using the population data from previous elections and current surveys. He said that samples that do not reflect some reasonable estimate of the total population sampled are flawed. He weights his samples with some methodology that consistently shows things closer. His track record on election day is very impressive. The sad fact is that this is even close in the corporate media polls.

[AlCzervik]

13. WTF? How many polls is this now that over sample republicans???

This isn't even being hidden anymore, just right out there in your face.

[Renew Deal]

14. That's good for Kerry.

Those numbers are decent considering the skewing.

[Gloria]

15. I do not get how they can apparently combine results from likely

voters and registered voters into one poll.
In my day, it was one or the other.