TruthIsAll: "Lets play "what-if" with the gender mix." Great stuff!

[autorank]

"Now we know why the gender weights were never changed in the Final Exit
Poll (13660 respondents: Kerry would still win).

ONLY BY CHANGING THE PERCENTAGES COULD THE FINAL EXIT POLL MATCH THE VOTE!

This is the reported 54/46 split:
KERRY WINS BY 3.1 MM VOTES WITH 50.78%"



What does it mean when you don't have a prayer of explaining the Bush Victory

[OnTheOtherHand]

1. actually, the gender weights weren't changed because

E/M weights their gender numbers even before official vote tallies start coming in, based on interviewers' tallies of non-respondent demographics.

Let's see if I have this straight: so if demographic proportions change, that is evidence of fraud, and if demographic proportions don't change, that is evidence of fraud. Is that about right?

[papau]

2. Would not both change if there was no fraud? Massive change to fit

"real" should not leave gender wieghts unchanged - unless someone wanted to solve an equation with only one unknown, and therefore held the others constant.

Evidence of fraud is in the stats and the 50,000 reports of fraud.

This is just evidence that the "final" (fit to actual) exit poll numbers should not be taken seriously.

[OnTheOtherHand]

4. I think part of what you're saying, everyone agrees about

The weightings were forced to match the official returns. Everyone knows that. We knew that would happen even before the election.

But I don't see how whether the gender percentages change or not has any bearing on the fraud debate.

If you believe that there is evidence of fraud in the stats, you're entitled to believe that.

I even partly agree that the final exit poll numbers shouldn't be taken "seriously" -- for that matter, the semi-final exit poll numbers shouldn't be taken all that "seriously" either, IMHO. They are estimates.

[papau]

5. I once "fitted" data - and while one variable and out was fast, we would

try for a "correct" set of variables.

To keep the gender the same may mean that those doing the fit knew it was a con and therefore did not care and went with the fastest way to get home.

Since the "fitting" routine is programmed and a button push, the above seems unlikely unless someone was perhaps lazy, or there was little feed back that justified a change so they limited the "error" to fewer variables, or there was just a con.

It least that is how I read the process. I do not read it as a normal day at the office doing a fitting of the data to a set of weights for the variables, where judgement is guided by reports back from the field.

[OnTheOtherHand]

6. I'm not sure what you are saying here

They weight male/female (based on interviewer report of non-respondent demographics) before they weight to vote counts -- before they have vote counts. This is part of the age-race-sex adjustment.

They could ditch those targets or maintain them equally quickly, I assume, although ditching them would be conceptually simpler. If I were them, I would certainly maintain them; it would be fairly nutty to respond to an apparent underestimate of the Bush vote by thinking, "Gee, I guess we should reduce the female turnout!" (Or, more likely to be obvious, the black turnout.)

I dunno what the "con" is. I don't suppose any election night is a "normal day at the office," but AFAICT Mitofsky has always reweighted the exit poll results to correspond with the official returns. It may seem strange to some of us that he didn't assume his survey was better than the official returns, but it doesn't seem strange to most survey researchers of my acquaintance. Even if you think that 7-10 million votes were stolen, I see no reason to assume that anyone at E/M thought so.

[papau]

10. The procedure is not in question - indeed expected -but ditching the

targets is what the reweighting is all about. To hold a variable constant is curious. You are implying there is no field input post the start of polling and the first report that would change the sex split for the final report. I posit that there is no field input to make any of the changes in the weighting - there is only the standard procedure to fit the data to "actual". Holding one variable constant in doing that fit makes for an earlier going home time.

The only Mitofsky "con" is the lack of discussion about how drastic the reweighting to get the "more accurate" actual was compared to what might have been expected.

On DU I am loose with words, and you are correct I should not call the Mitofsky reweighting a "con" - but I expected a TruthisAll type analysis of the extent of the rewieght- heck someone could have mentioned the other org exit polls that showed Hispanic staying at the same GOP/DEM breakout as in prior elections - and asked why Mitofsky after rewieght showed a major, major change moving that group to almost a 50/50 split.

Surely smell test results like the Hispanic subset demanded more of a discussion than Mitofsky has given us to date.

[OnTheOtherHand]

13. OK, still trying to sort this out

"You are implying there is no field input post the start of polling and the first report that would change the sex split for the final report."

I don't know what you mean by the "first report" here, so I may miss your point. We know that during the day, E/M was having a heck of a time getting gender numbers in some states that they thought were plausible. But by the time the preliminary state results went up on cnn.com as state polls closed, E/M had incorporated all the Call 1 and 2 numbers and, I surmise, many but not all of the Call 3 numbers (for instance, the N in the exitpollz.org screen shot for Pennsylvania is 1930; the final N for the crosstabs, as visible on cnn.com today, is 2107). And the call numbers include not only information on exit poll respondents, but also the interviewer tallies of non-respondents. So, there is extensive "field input post the start of polling" -- but perhaps not "post... the first report" depending on what you mean by the first report.

Presumably at some point in the evening, all the new information coming in is information about vote counts -- maybe that is what you meant.

"Holding one variable constant in doing that fit makes for an earlier going home time." Again, my view is that they could hold the sex split constant when reweighting to the official tallies, or allow the sex split to change, with equal speed; it's not a killer programming task either way. I may still be missing your point, sorry.

Hispanics/Latinos -- well, even the n=11,027 national numbers that some observers regard as pristine show Bush taking 41% of the self-identified Hispanic/Latino vote (if I can trust the exitpollz site), up from 35% in 2000. The final results bump that up to 44% (about the same increase as among whites). The LA Times final results have it at 45%. But if you run a crosstab on the unweighted NEP results, it's only 37%, which seems to indicate that demographic and geographic reweightings shifted the number substantially even before vote counts kicked in -- and I don't know why. Probably someone does, but I don't. (The Velasquez Institute poll apparently put it at 35% including absentees; that survey covered 46 precincts, so it is iffy to compare it with the national polls. Here is an institute-sponsored analysis: http://www.wcvi.org/latino_voter_research/polls/national/2004/flores.ht . Interestingly, the institute's results for Latinos in Florida are much closer to E/M's, see http://www.wcvi.org/latino_voter_research/polls/exit_poll_results_030805.htm .)

There could well be questions about whether E/M dealt adequately with Hispanic/Latino voters quite apart from whether the results raised questions about the vote count.

[kster]

3. On to the greatest number 5..........nt

[autorank]

7. Here you go...

TruthIsAll:

"Now we know why the gender weights were never changed
(from 54% Female/46%
Male) in the Final Exit Poll: Kerry would still win,
regardless of any
plausible change in the mix. 

Note that 54/46 (rounded) is probably correct, since it's
confirmed by the
census and state exit polls (53.51% and 53.76%), but that is
not the point
here.

EVEN WITH A RIDICULOUS 64% MALE/ 36% FEMALE TURNOUT, KERRY
TIES BUSH!

THE FINAL EXIT POLL (THE ONLY TIMELINE WHICH BUSH WON) COULD
ONLY BE
MATCHED TO THE RECORDED VOTE BY CHANGING THE KERRY/BUSH VOTING
PERCENTAGES!


Ask yourself: Why does the gender split stay constant from the
12:22AM
(13047 respondents) exit poll timeline to the 1:25PM (13660)
final, unlike
the following demographics:

HOW VOTED IN 2000: 41 Bush/39 Kerry changed to 43/37.
PARTY ID: 38 DEM/35 REP changed to 37/37.

One would expect the gender split to change slightly as well.
Well,
actually it was changed to 54/46 at the 7:33pm time line from
58/42 at
4:00pm -  and stayed constant thereafter.

Remember when the naysayers claimed that the early exit polls
were biased
in Kerry's favor because they oversampled women? What ever
became of that
canard?

NATIONAL EXIT POLL TIMELINE:
 3:59pm  (8349) Female 58%/ Male 42%; Kerry led 51-48
 7:33pm (11027) Female 54%/ Male 46%; Kerry led 51-48
12:22am (13047) Female 54%/ Male 46%; Kerry led 51-48
 1:25pm (13660) Female 54%/ Male 46%; Bush led 51-48

Why were the weights unchanged in the FINAL? Well, unlike
"How Voted
in 2000" and "Party ID", changing the Gender
weights had no
effect on the result: KERRY WOULD STILL WIN. 

Once again, why change the Gender mix from the 12:22am time
line,
especially if they were within the MoE? 

To repeat: CHANGING THE WEIGHTS WOULD HAVE NO EFFECT. SO
THEREFORE THE VOTE
PERCENTAGES HAD TO BE CHANGED.

And why change Party ID and How Voted weights, if they were
within the MoE?

Here's why: IN ORDER TO MATCH THE RECORDED VOTE, THE WEIGHTS
AND
PERCENTAGES HAD TO BE CHANGED. CHANGING ONLY THE PERCENTAGES
IS NOT ENOUGH
TO DO THE TRICK.

Food for thought for those who dare to think.

Now lets play "what-if" with the gender mix.
____________________________________________________________

PRELIMINARY EXIT POLL			
Nov. 3, 12:22am, 13047 respondents			
					
GENDER	MIX	Votes	Kerry	Bush	Other
Male 	46%	56.20	47%	52%	1%
Female 	54%	65.97	54%	45%	1%

Total 	100%		50.78%	48.22%	1.00%
Vote  	      122.17	62.04	58.91	1.22

KERRY WINS BY 3.1 MM VOTES WITH 50.78%

_____________________________________________________________

FINAL EXIT POLL	- no change in weights		
Nov. 3,  1:25pm, 13660 respondents			
				
        MIX	Kerry	Bush	Other	
Male   46%	44%	55%	1%	
Female 54%	51%	48%	1%

Total  100%	47.78%	51.22%	1.00%
Vote   122.17	58.37	62.58	1.22	

BUSH WINS BY 4.2 MM VOTES


____________________________________________________

Scenario I:
50% MALE /50% FEMALE 

KERRY WINS BY 2.45MM VOTES WITH 50.50%

GENDER	MIX	Votes	Kerry	Bush	Other
Male 	50%	61.09	47%	52%	1%
Female 	50%	61.09	54%	45%	1%

Total 	100%		50.50%	48.50%	1.00%
Vote	      122.17	61.70	59.25	1.22


____________________________________________________

Scenario II
54% MALE/ 46% FEMALE 

KERRY WINS BY 1.75MM VOTES WITH 50.22%

GENDER	MIX	Votes	Kerry	Bush	Other
Male 	54%	65.97	47%	52%	1%
Female 	46%	56.20	54%	45%	1%

Total 	100%	      50.22%	48.78%	1.00%
Vote        122.17 61.35	59.60	1.22


____________________________________________________

Scenario III

64% MALE/ 36% FEMALE 

A VIRTUAL TIE: KERRY 49.52% / BUSH 49.48%!

GENDER	MIX	Votes	Kerry	Bush	Other
Male 	64%	78.19	47%	52%	1%
Female 	36%	43.98	54%	45%	1%

Total 	100%  	      49.52%	49.48%	1%
Vote	      122.17 60.50 60.45 1.22"

TruthIsAll

[Festivito]

8. Could someone help explain this source data?

NATIONAL EXIT POLL TIMELINE:
3:59pm (8349) Female 58%/ Male 42%; Kerry led 51-48
7:33pm (11027) Female 54%/ Male 46%; Kerry led 51-48
12:22am (13047) Female 54%/ Male 46%; Kerry led 51-48
1:25pm (13660) Female 54%/ Male 46%; Bush led 51-48

1. Are the parenthisized numbers cumulative?
2. Are the percentages cumulative, or corresponding to the time-stopped increase of responses.
3. Is the final 51-48 supposed to be the final percentages projecting the actual vote?
4. Is the final time, 1:25pm, the afternoon of the day following the election?

Thanks in advance.

NOTE: IF cumulative, and yesses, then of the final 613 responses, roughly 700(impossible) reported voting for Bush and LESS THAN ZERO reported voting for Kerry. In fact about a hundred ALREADY COUNTED VOTES for Kerry would have to be SUBTRACTED(also impossible) in order to achieve total cumulative Bush-Kerry percentages 51-48. Impossible, that is, to have not made a mistake, i.e. a huge boner of near epic proportion. Especially since so much time has passed and these people were supposedly professionals paid to calculate this information. Fraud is possible.

[OnTheOtherHand]

9. I think the answers are:

(1) Yes, total responses included to that time.

(2) Your wording is ambiguous, but I would say cumulative, incorporating all interviews.

(3) Not "projecting" -- as I understand it, E/M doesn't actually project the national popular vote, only state races -- but representing, yes.

(4) Yes, the afternoon of the day after.

More importantly, no, it doesn't mean that they subtracted votes from Kerry among the final 613 respondents. That isn't how the exit polls work, ever -- they don't ever figure raw percentages of respondents. The responses are always weighted. Initially, the weights reflect demographics, geography, and absentee ballots; later, they reflect official vote counts. The poll analysts don't "add" or "subtract" various kinds of voters; they multiply various respondents by numbers greater or less than 1 to hit the target proportions.

So, certainly fraud is possible (it always is), but the idea that Kerry voters were subtracted from the exit poll is incorrect.

[Festivito]

11. Getting from 12:22 to 1:25 with 613 polled needs a multiplier of 131%.

Which could happen, perhaps, if all day, blacks and hispanics were polled in disproportionally high percentages to the surrounding population, and when the last 613 were polled, or found, those 613 had to represent a large portion of, say, whites, again, needing to multiplied by 131% to adjust to represent the surrounding white population.

131% is the minimum multiplier needed assuming that EVERY response was for Bush. Certainly, the responses should have been mixed and the multiplier would have to be higher.

Could overly zealous pollsters have picked minorities all morning and realizing this later in the day suddenly started picking only whites in order to balance their sources, only to find they still needed a huge multiplier? All we seem to know is that it had to be white men and white women in the same proportions they had collected all day.

Regardless, it doesn't smell good.

[Febble]

12. No, ALL respondents

would have been reweighted, not just the final 613. This would be done in line with the vote count from the polled precincts. That's the point. Yes, it was matched to the count. No, there was no particular voodoo regarding the final 613 respondents. It's just that by the time they issued the final poll, all responses were in.

No-one, least of all the pollsters, who have had it on their website since before the election on their website, deny that the final projection was matched to the vote-count.

[applegrove]

14. don't assume exit polls were not dicked with - same as anything

else. Exit polls have been off in more ridings than diebold.

[kster]

15. kick.nt