foo-bar said:
http://www.democraticunderground.com/discuss/duboa...

Kudos to you for doing all that work. Bet you thought you finally got me.
Well, not quite.

I was being conservative when I included ABC, Pew, and USA/Gallup, realizing that they were very suspicious late outliers.

The 7-poll 58.1% projection falls within the Wikipedia margin.
The 3-poll 52.3% projection lowered the 10-poll projection by 2.1%.
Therefore, the 10-poll 56.0% projection was conservative.

Looking at the final 30 polls from Oct.2:
The 30-poll projection was 55.8%.
The 27-poll projection (excluding outliers) was 57.0%.

There was a 1.1% deviation between the 7-poll and 27-poll projections.
Why? The Democratic trend was increasing.

UVA: 60% of Undecided voters were allocated to the Democrats

Poll.... Date Dem Rep Oth Und UVA Dem
Harris 1023 47 33 2 18 10.8 57.8
AP..... 1030 56 37 2 5 3 59.0
CBS... 1101 52 33 2 13 7.8 59.8
Nwk.. 1103 54 38 2 6 3.6 57.6
TIME.. 1103 55 40 2 3 1.8 56.8
CNN... 1106 58 38 2 2 1.2 59.2
FOX..... 1106 49 36 2 13 7.8 56.8

Outliers:
Pew... 1104 47 43 2 8 4.8 51.8
ABC... 1104 51 45 2 2 1.2 52.2
Gallup..1106 51 44 2 3 1.8 52.8

Averages
Polls...... Dem Rep Other Undec UVA Projection
10.... 51.6% 39.1% 2.0% 7.3% 4.4% 56.0%
7...... 53.0% 36.4% 2.0% 8.6% 5.1% 58.1% (excl. outliers)
3...... 49.7% 44.0% 2.0% 4.3% 2.6% 52.3% (outliers)

Final 30 polls/projections from Oct.2:
30.... 51.0% 39.0% 2.0% 8.0% 4.8% 55.8%
27.... 52.9% 38.3% 2.0% 6.8% 4.1% 57.0% (excl. outliers)

Poll..............Date Dem Rep
CNN LV........... 1002 53.0% 42.0%
AP-Ipsos RV..... 1004 51.0% 38.0%
Pew RV........... 1004 51.0% 41.0%
TIME LV.......... 1005 54.0% 39.0%
Newsweek RV... 1006 51.0% 39.0%

ABC RV............1008 54.0% 41.0%
CNN LV ...........1008 58.0% 37.0%
Gallup LV.........1008 59.0% 36.0%
Harris LV.........1009 49.0% 36.0%
FOX LV............1011 50.0% 41.0%

CNN LV .......... 1015 56.0% 40.0%
NBC RV........... 1016 52.0% 37.0%
Newsweek LV... 1021 55.0% 37.0%
Gallup LV........ 1023 54.0% 41.0%
ABC RV........... 1023 54.0% 41.0%

CNN LV .......... 1022 57.0% 40.0%
Hotline RV....... 1023 52.0% 34.0%
Zogby LV......... 1025 44.0% 33.0%
FOX LV........... 1025 49.0% 38.0%
Newsweek LV... 1027 53.0% 39.0%

CNN LV .......... 1029 53.0% 42.0%
NBC LV........... 1030 52.0% 37.0%
CBS LV........... 1101 52.0% 33.0%
Newsweek LV... 1103 54.0% 38.0%
TIME LV.......... 1103 55.0% 40.0%

Pew LV........... 1104 47.0% 43.0%
ABC LV...........1104 51.0% 45.0%
Gallup LV....... 1106 51.0% 44.0%
CNN LV .......... 1106 58.0% 38.0%
FOX LV........... 1106 49.0% 36.0%

In closing, I want to emphasize that my probability calculations are based on actual polling results, an assumed UVA over a range of MoE assumptions. Do you have a problem with that? And I was being conservative when I used a 1.5% MoE. As for those "house effects", why don't you try to quantify them?

As the number of polls increase, so-called "house effects" are minimized. As long as the polls sample from the same general population at approximately the same time, averaging of the results makes mathematical and intuitive sense. So does combining the sample-sizes. That's why analysts combine the latest polls.

A most important and famous result is The Central Limit Theorem (CLT). Political analysts are aware of the Law of Large Numbers, but are unaware of the ramifications of the CLT.

The CLT states that if the sum of the variables has a finite variance, then it will be approximately normally distributed. Since many real processes yield distributions with finite variance, this explains the ubiquity of the normal distribution.

You can test out the CLT yourself with this simulation:
http://en.wikipedia.org/wiki/Illustration_of_the_c...

http://progressiveindependent.com/dc/dcboard.php?a...



These FACTS are never discussed in the media.
- election fraud
- undecided voters break for the challenger, especially if the incumbent is unpopular. Bush was at 48%.
- Many newly registered voters are not included in LV (likely voter) polls.
- Since 1988, Democrats have won 60-70% of new voters. In 2008, Obama had 71%+.
- MILLIONS of mostly (70-80%) Democratic votes are uncounted in every election
- The 2004/2006/2008 Final National Exit Polls indicated there were millions of returning phantom Bush voters from the prior election (more than were alive).
- The Final National and State exit polls are always FORCED TO MATCH the recorded vote.
- By definition, the RECORDED VOTE never reflects the TRUE VOTE since millions of votes are UNCOUNTED in EVERY election.
- Uncounted votes alone provide overwhelming evidence that elections were stolen in 1968, 1988, 2000 and 2004.
- 1968: 6 million net uncounted votes.
Nixon won the recorded vote by 500,000.
- 1988: 10.6 million net uncounted votes. Bush won the recorded vote by 7m.
- 2000: 5.4 million net uncounted votes. Gore won the recorded vote by 540,000.
- 2004: Nearly 4 million uncounted votes.
Bush won the recorded vote by 3.0 million.
Kerry won the True Vote by 10m.
Approximately 5 million were switched from Kerry to Bush on UNVERIFIABLE voting machines.

- 2006 midterms: Democrats won by more than double the recorded 52-46%% margin.
10-20 House seats were stolen.

- 2008: Obama won by more than double his recorded 9.5m margin.

1988-2008: the average Democratic TRUE VOTE margin was reduced from 10% to a 4% RECORDED MARGIN by Election Fraud.

VOTER FRAUD was virtually non-existent.