Statistical question RE suspect unchanging %

[VTGold]

In a lot of the statistical evidence put forward I've seen it remarked that Kerry and dubya's percentage of votes did not change even when thousands of votes were added to tallies.

That of course looks suspect - but I was wondering about the exit polling then - isn't exit polling (all polling) based on taking a small sample of votes and predicting that the ratio will stay the same?

I keep having knock-down, drag-out fights with non-believers (Dems, mind you) and when I bring up the percentage thing - I want to be correct so I thought I'd ask any Statistics buffs out there.

[Nicholas D Wolfwood]

1. Big difference.

Exit polls predict the ratio will end the same. That doesn't mean the ratio will remain static the entire night.

[On the Road]

2. As I Understand It,

the exit polls fluctuated a lot throughout the day, even before the final adjustments. For example, Kerry was up in Pennsylvania 60-40 in the morning and that came way down by the end of they day.

If the percentages stayed pretty much the same throughout the day, I wouldn't think of that as suspicious. All it would mean is that Republicans and Democrats had the same time-of-day voting patterns.

When the results start to be reported, however, I would expect large fluctuations, because some precincts are very red and some very blue in most states. So the vote with 5% of precincts reporting is often very different from the vote with 100% reporting.

[VTGold]

4. There was one cty in particular where the % was 77 - 33 in favor....

... of dubya at one point and then 19,000 votes were added at once and the percentage stayed exactly the same. That looks suspicious as hell but as I asked before:

lets say you have a "red" county and after 100 votes the ratio is 77% to 33% - then 900 more votes come (same cty) and the percentage stays the same. Isn't that sort of what you would expect? Given exit poll methodology.

[On the Road]

7. If it Was 77-33

Somebody at the station couldn't add. :)

Let's put it this way -- what is there that does NOT look suspicious? If the afternoon numbers were completely different from earlier numbers, that looks strange. If they're exactly the same, that looks strange, too.

It's entirely possible the first 100 votes were representative of the first 1,000 votes. That would make the percentage stay constant. at least to within rounding.

If the numbers show that the percentage should change, but it doesn't, that's probably the TV network failing to update their percentages. What counts is the official votes.

[VTGold]

10. oops! fuzzy math....

[helderheid]

3. This is from that terrific article Alan Walden wrote

OT, he'll be here for dinner tomorrow night on his way through my state!

"Center for Research on Globalization's Michael Keefer states, "The National Election Pool's own data – as transmitted by CNN on the evening of November 2 and the morning of November 3 – suggest very strongly that the results of the exit polls were themselves fiddled late on November 2 in order to make their numbers conform with the tabulated vote tallies."

How do we know the fix was in? Keefer says the total number of respondents at 9 p.m. was well over 13,000 and at 1:36 a.m. it had risen less than 3 percent – to 13,531 total respondents. Given the small increase in respondents, this 5 percent swing to Bush is mathematically impossible. In Florida, at 8:40 p.m., exit polls showed a near dead heat but the final exit poll update at 1:01 a.m. gave Bush a 4 percent lead. This swing was mathematically impossible, because there were only 16 more respondents in the final tally than in the earlier one. "

http://www.orlandoweekly.com/news/Story.asp?ID=4688

[seaclyr]

5. Popular vote tallies actually changed a lot

They showed a 15% Bush lead early, which dropped over time down to the final 3% or so. The Bush lead in actual votes, however, (not percentages) increased early to about 1.8 million votes by 9:30pm EST, then stayed the same over much of the evening before increasing late by another 1.8 million votes or so to be 3.6 million by about 2:00am EST. Not sure whether that addresses your question though!

[tinfoil_beret]

6. Even more statistically disconcerting...

Look, for instance, at the percentages of votes for Gore and Bush in 2000 in Miami-Dade county, Florida, and the percentages of votes for Kerry and Bush in the same county for 2004. The percentages are nearly identical, and the margins are the same to within one in ten thousand. With a 23.87% increase in votes cast in 2004, relative to 2000, that's what I call a statistical anomaly.

[jmknapp]

8. Parma 5-E

A precinct in the Cleveland area, Parma 5-E, is a case in point even on the precinct level.

The unofficial tally released by the Cuyahoga County BOE on election night for Parma 5-E:

Kerry 310
Bush 340
Badnarik 3
Peroutka 1

Fingerhut 241
Voinovich 609

Total votes cast 917
Total registered voters 972

The turnout in this precinct was 94%! Total vote for Senator
was 850 compared to a total vote for President of 654. I.e., 28% of the
voters did not cast a vote for President!

OK, now for the certified results released Monday by the Ohio SOS:

Kerry 311 (+1)
Bush 344 (+4)
Badnarik 1 (-2)
Peroutka 0 (-1)

Fingerhut 179 (-62)
Voinovich 451 (-158)

Total votes cast 663 (-254)

This disappearance of votes bring the numbers into the range of believability, but an astute dKos reader noted that the new Fingerhut/Voinovich totals could be gotten by multiplying the unofficial total in each case by the same factor, (609/850):

If you set it up as a simple equation, you get the new vote totals:
609 = x
----- -----
850 630

x = 451.37647 (round down to 451)

So it smells of a hack where they just brought the numbers in line, and kept the ratio the same.

[VTGold]

9. Florida is a nightmare - somehow it feels harder to jump into ...

...than Ohio.

It really does feel like a "black box".

I think Florida is going to come down to whisteblowers and lower-level gov. employees who are found to be corrupt. Hopefully that will lead right to the top.