Just what kind of electorate are these polls asking?

[George_Bonanza]

Is there a chance that many of would-be Obama voters are being under-represented in these polls? What demographic models are these companies using?

[featherman]

1. Take no poll on face value - especially "likely voters" - always check the cross tabs

[Captain Hilts]

8. "Likely voters" is usually more accurate.

Citing "likely voters" is not the partisan bugaboo people here think it is. ALL pollsters use it.

"Registered voters" is a lousy pool for pollsters.

[featherman]

13. Sorry... I was not comparing LV's compared to RV's but there is an element

of subjectivity in identifying likely voters and this can vary even within the same polling agency. A recent example was an oversampling of Dems pre-convention to adjust for the "enthusiasm gap". Result: healthy lead for Obama. Post-convention the same poll evened up the sample showing the same % of GOP and DEM participating in the election.

So if you agree that the same total number of Dems and Reps will be voting in 2008... say 35% Dem (out of 42% RVs) to 35% GOP (out of 31% RVs) to 30% Inds... then, yes, this LV model is accurate and DEMS cannot outvote the minority party and must do better than just split or barely lose the IND vote.

Perhaps this is the case although even post-convention polls are still showing a big "enthusiasm gap" - 55% on the Obama side versus 37% for McCain (up from 12-18% pre-convention)

Anyway that's what I meant.

[Captain Hilts]

14. They try to use the same methodology to identify likely voters...

otherwise historical data is useless.

Now, Obama voters might prove that the older models don't work, but in the meantime, pollsters are playing the odds on being accurate that worked for them before.

[Captain Hilts]

2. Check the details on the pollster's website. Partisan and age breakdown.

They usually publish them.

[BlooInBloo]

3. Or maybe the tides of polling just shift this way and that way over the course of time...

Jeebus - why is that so fucking hard to understand and accept?

[whistle]

4. The ones willing to answer the phone

[Captain Hilts]

11. This favors no specific demographic.

It's generally harder to get men on the phone, so, some polling companies ask to speak to the "youngest registered male voter in the house." That discourages surveys from skewing older and female.

[Eric J in MN]

5. If we believe the polls when they show Obama ahead, then...

...we should believe the polls when they show McCain ahead or tied.

[Captain Hilts]

12. Stop bringing logic into this argument about statistical methods! Conspiracies are more fun!

Surveys have progressed a LOT since the 1936 screw up!

FDR: Boy, what a mess that was! Said it was going to be close!
ER: Said America hated me! Ha!

[Edgewater_Joe]

6. Most Are 2004 Models

That's what you always use for polling -- the previous election's numbers.

Which means, yes, Democratic numbers and new registers aren't going to be as well-represented as the fundies and the 50/50 split from '04.

[Captain Hilts]

10. Dems have a lower voter turnout rate than Reps. That's why they poll them more equally...

than raw numbers would suggest.

[abbiehoff]

7. In spite of the fact that Democrats outnumber Republicans

they continue to poll nearly equal numbers of both.

Plus, they don't call cell phones.

[Captain Hilts]

9. Yes, some surveys DO call cellular phones. And cellular phones are widely enough used...

you're not really screwing over a specific demographic.

You poll Republicans at a higher rate than Democrats because they have a MUCH higher turnout rate than Democrats have. They're following statistical models in doing this.