KERRY VS. BUSH FROM 11027 TO 13047 TO 13660
Notice how Kerry maintained his lead from the 11027 initial National Exit Poll respondents posted at 7:38 on CNN right up to 12:22am, when 2020 additional brought the total to 13047, as posted on WP. He was winning 51-48.
But right after 12:22am things began to change. The 613 additional respondents which brought the final exit poll total to 13660 must have been all Bush voters who were not at all reluctant to speak to the exit pollsters.
Because these 613 Bush voters were able to defy the most famous discovery, indeed the basis for all of Mathematical Statistics: The Law of Large Numbers.
Forget the initial trend which had Kerry winning by 51-48 after 11027 were polled.
Forget the confirmation of the trend when an additional 2020 were polled and the numbers stayed virtually the same and in fact, improved slightly for Kerry.
Forget the Law of Large Numbers, which is the foundation for all statistical analysis.
This was a divine intervention.
THE EXIT POLL TIMELINE OF CHARACTERISTIC WEIGHTS
Notice which change going from 13047 to 13660:
Voted in 2000
But the other category weights, such as Gender (54% female/46% male) are CONSTANT from 11027 through 13660.
The Bush percentage share of each characteristic, not shown in this graph, increased from the Preliminary 13047 to the Final 13660 respondents, while Kerry's decreased.
Quite a trick, to turn 180 degrees from a 51-48 Kerry win to a 51-48 Bush win with just 613 additional exit poll respondents and a fairly constant set of weights, except for Party ID and How Voted in 2000.
KERRY'S EXIT POLL SUB-GROUP PERCENTAGES - PRELIMINARY TO FINAL
Kerry was doing great in all Exit Poll categories at 7:38 pm after 11027 responded. He was winning by 51-48%
He was doing virtually the same in all Exit Poll categories at 12:22 am after another 2020 responded to increase the total to 13047.
He was still winning 51-48%.
But then the final 613 responded, bringing the total to 13660 at 2:05 pm. Bush won 51-48%.
The characteristic weight mix for most categories remained the same, except for Party-ID and How Voted in 2000. These weights were adjusted in Bush's favor, even though the "Party_ID" mix was at variance with the last 3 elections and the "How Voted in 2000" numbers were impossible.
Forty-three (43%) could NOT have voted for Bush in 2000; that would mean he had 52.5 million votes. He only had 50.5 million.
But the numbers which did the damage were Bush's actual share vs. Kerry in all the sub-groups. His numbers increased in just about every single group.
And Kerry's numbers declined accordingly. This graph shows the Kerry detail.