FEBBLE:
That sampling error is the only error in a poll.
TIA:
I never said that. I said that scientific polling minimizes
the error and that’s why pollsters always quote the MoE based
on sample-size.
Well, that isn't true. It isn't why pollsters always quote the MoE based on sample-size.
Here is Mitofsky on the subject:
REPORTING SAMPLING ERROR
I want to say a few words about reporting sampling error. A number of people who have spoken here have talked of not reporting sampling error because it was confusing all those dear mindless souls who listen to our results. They were concerned we would make people think that sampling error was the only error in the survey.
http://www.nyaapor.org/WMitofskySpeech.htmYou cannot "scientifically" eliminate bias from a survey. Inferring that from the fact that a pollster quotes the MoE based on sample size put TIA in the category of those "dear mindless souls" who would be "confused" into thinking such a thing.
FEBBLE:
That when there IS bias in the poll it would favor
Republicans, not Democrats
TIA:
A few FACTS.
1) It’s a well known fact that approximately 3% of the votes
are uncounted due to lost, spoiled and provisional ballots.
The vast majority of these votes are in minority districts
(50% in black districts). Since the minority black vote is
90% democratic, and the Hispanic vote at least 60%, a fair
estimate is that 75% of the uncounted votes (2.25%) are
Democrat and the other 0.75% Republican. That’s a 1.5% bias
in favor of the Republicans.
2) There is a proven reluctance of low-income Democratic
voters (under $50K) and high-income Republicans ($100k+) to
participate in exit polls. But since low income voters
outnumber high-income voters by almost 3-1, there's another
component of Republican bias.
Spoiled votes would not cause bias in the sample. They would simply cause a discrepancy between the sample and the count. However, the effect is likely to be small because the spoiled votes tend to be concentrated in strongly Democratic precincts which are not highly represented in the precinct sample. It will be an effect however, although much smaller than the effect on the actual results. I share with TIA his indignation at this systematic disenfranchisement of largely Democratic voters. As for his second point - TIA read it in a book somewhere. It is not something that can, or should, be generalised to exit polls. There are many sources of evidence, including direct experimental evidence that in exit polls, Democratic voters tend to be over-sampled.
3) Exit poll response was high in Bush states and low in
Kerry states.
Here's GRAPHIC MATHEMATICAL PROOF USING LINEAR REGRESSION
ANALYSIS:
Overall response rates are not a proxy for response bias. Response bias occurs when the response rates for one set of voters differ from the response rate for the other set of voters, whether the two rates are 15% and 20% or 60& and 80%. Moreover, selection bias will not show up in response rates - and may even be associated with higher response rates. If more willing voters are being selected, completion rate will go up.
FEBBLE:
That people recall their previous votes correctly (there is
excellent evidence that they do not, and that they tend to
misreport having voted for the previous winner)
TIA:
If you consider that the FINAL National Exit poll is ALWAYS
MATCHED TO THE RECORDED VOTE, WHICH MEANS THAT THE WEIGHTS
ARE FUDGED TO MATCH THE VOTE. I CALL IT FUDGING; YOU CALL IT
VOTER MISREPRESENTATION. Witness the 2004 Final NEP, which
Bush won by 51-48%: 43% of 2004 voters were Bush voters and
37% were Gore voters? This was a FUDGE necessary to match the
recorded vote. In the 12:22am poll, which Kerry won by 51-48%,
the mix was 41 Bush/39 Gore. Mathematically, the Bush MAXIMUM
Bush 2000 representation weighting was 39.8%, which is the
ratio of Bush 2000 voters still alive n 2004 by the 122.3mm
who voted.
Do you want MORE of this evidence? Look at the 2006 NEP. The
7pm poll had the 2004 weighting as 45Bush/46 Gore. The FINAL
had it 49 Bush/43 Gore/8 Other. Where did the 8% for OTHER
voters come from? Third-party 2004 voters comprised 1% of the
vote. Where did the excess 7% come from? Kerry. Here’s why:
Are we to believe Bush voters outnumbered Kerry voters by 6%?
Even if you believe the 2004 FINAL, which we have proven
bogus, the Bush margin was 3%. In reality it should have been
50Kerry-47Bush, after deducting 1% for voter mortality over
the 2-year period. Now 50% = 43% + 7%. There is your Kerry
vote.
I really can't be bothered to explain this to TIA again. He needs to read Mark Lindeman's paper:
http://inside.bard.edu/~lindeman/too-many.pdfThat undecideds always break for the challenger.
TIA:
I never said ALWAYS. I said the vast majority of the time.
The evidence is a study of 155 incumbent elections that you
yourself quoted: in 82%, the challenger won the undecided
vote; the incumbent, just 12%; neither, 6%. And of course,
pollsters Zogby, Harriss and others have always claimed that
undecided voters break to the challenger by better than 2-1,
ESPECIALLY WHEN THE INCUMBENT IS UNPOPULAR, AS BUSH WAS IN
2004 (48.5% rating) and in 2006 (33% rating). That’s why I
conservatively assume that Kerry in 2004 and the Democrats in
2006 would win the undecided vote by 60-40% in my election
models. And that is why my 2004 PROJECTION (51.8 Kerry-48.2
Bush) and 2006 projection (57D-43R) WERE BOTH RIGHT ON THE
MONEY. They were based on 18 final national pre-election
polls and 116 pre-election Generic polls in 2006.
Well, did he weight his probability by his estimate of the probility of his assumption being true?
FEBBLE:
That incumbents below 50% always lose.
TIA
I never said ALWAYS. But Bush had 48.5% ratings in 2004. He
stole the election. Carter (1980), Bush (1992), Ford (1976)
all had ratings below 50%. And they all lost.
Whatever.
FEBBLE:
That Likely Voter models are less reliable than Registered
Models.
TIA
I said they were less reliable in 2004, when new Democratic
registrations were massive and young, single cell-phone users
were unlikely to be contacted. These were NOT likely voters.
Facts, Febble. Facts.
And TIA didn't investigate the fact that this was compensated for by up-weighting the age demographic.
Fianally, you have claim elsewhere in this thread that my
analysis is not supported in the "reality-based"
community. Well, what about Freeman, Mark C. Miller, RFK,
Baiman, Dopp, Jonathan Simon, Bruce O'Dell, Conyers,
Fitrakis, Richard Hayes Phillips, Palast, Michael Keefer,
etc? They aint exactly chopped liver.
They ain't exactly unanimous either.
OK, I'm calling this off. I can't converse with a poster who isn't here. If TIA wants me , he knows where to find me.
Peace.
Lizzie
edited for typo