How were Mitofsky's Exit Polls adjusted to show a Bush victory?

[Bill Bored]

The question has been raised: How did Bush get 63.6% of the final 623 polled?

We know the final poll was adjusted.
This is old news (yawn).
Why do we keep rehashing this?
We have enough threads about this to knit a sweater!

The question should be: HOW was the final poll adjusted?

Here's how:

According to the unadjusted 80% poll (Scoop):
________ K _ B
DEM 38% 90 9
REP 36% 7 92
IND/
Othr 26 52 45
with 95% of the 11,027 sample responding,
Kerry beat Bush 51-48.

According to the final unadjusted ~100% poll (Washington Post):
________ K _ B
DEM 38% 90 9
REP 35% 7 92
IND/
Othr 26 52 44
with ~95% of the 13,047 sample responding,
Kerry beat Bush 51-48.

According to the adjusted 100% poll (Available EVERYWHERE):
________ K _ B
DEM 37% 89 11
REP 37% 6 93
IND/
Othr 26 49 48
with 96% of the 13,660 sample responding,
Bush beat Kerry 51-48.

What does this tell us?

It tells us that the adjustments were as follows:

1. Weight Dems and Republicans equally.

2. Give Bush a higher percentage of the Independent/other-party vote, but still less than Kerry.

3. Give Bush a higher percentage of Democratic voters and a slightly higher percentage of Republicans.

Now, the next question is: Out of these 3 adjustments, which one(s) is actually enough to swing the popular vote?

Answers:

1. Weighting Dems and Republicans equally still results in a Kerry victory, albeit by a very narrow margin of about 0.6% instead of 3%.

2. Giving Bush a higher percentage of the Independent/other-party vote results in a Kerry victory of 1.3%.

3. Giving Bush a higher percentage of Democratic voters results in a Kerry victory of 2%. Even if you throw Bush 1% more Republicans (93% vs. 92%), Kerry still wins by 1.3%.

So no one adjustment is enough to change the outcome.

4. Let's say we combine 1 and 2 above: Bush wins by 1.2%.

5. Combine 1 and 3 above: Bush wins by 1.3%.

6. Combine 2 and 3 above: Bush wins by 0.6%.

So, what does all this mean?

It means that to get enough of a swing to give Kerry's entire 3% popular vote margin to Bush, you had to make three separate adjustments to the exit polls.

Now could any or all of these be applied to hack the actual popular vote? Only 2 are needed for Bush to win it. Or is it possible that the polls really were off in these weightings?

Well, we could just start ANOTHER thread to discuss this further, but it MIGHT be nice to keep this one kicking around a for a while if it's not too much trouble.

[Virginian]

1. If they made this type of adjustments in the Ukrane,

could their election have been legit?

The "problems" with our elections need to be fixed or else every president from now on will have the last name of "Bush".

[Mistwell]

2. Left out gender

My understanding was that the largest adjustments made included:

1) More men turned out to vote than originally predicted; and
2) More Republicans turned out to vote than origninally predicited; and
3) More westerners turned out to vote than originally predicted.

Those were not the only adjustments, and obviously those I listed have significant overlap, but that was what I thought was reported at the time.


See:

http://www.washingtonpost.com/ac2/wp-dyn/A23580-2004Nov3?language=printer

"On Tuesday, new problems surfaced: a 2 1/2-hour data blackout and samples that at one point or another included too many women, too few Westerners, not enough Republicans and a lead for Democratic presidential nominee John F. Kerry in the national survey that persisted until late in the evening."

[Carolab]

4. The Post statement is "smoke screen", don't you see?

The analysis in this post is discussing percentages on a broader scale--that rather than shifting women-men, black-white, etc., they shifted percentages from the broadest category--party affiliation.

There is one telling comment in the Post's comment regarding the polls containing "too few Republicans". How would they say that unless the percentages of Republicans were later "upped"?

If we could look at the actual tallies/percentages of Republicans and Democrats that really voted for Kerry and Bush and see whether or not they line up with these exit poll percentages we could determine the fraud. One would need to count votes by actual party affiliation in order to establish that.

Is every voter identified by party affiliation in their registration?

[Bill Bored]

11. Right.

According to the adjusted polls, both parties turned out equal numbers of voters nationwide.

Registration includes party affiliation least in terms of Dem, Rep and None. Not sure about the 3rd parties though, but we know that they are supposed to be 26% altogether so all you need is the Dems and Reps. That's one part of the adjustment.

The other 2: Dems for Bush and Independents/others for Bush might be harder to gage. Food for thought though I hope.

[Mistwell]

14. Yes

Voters ARE identified by party affiliation in their registrations. This is so that polling stations can assign people ballots for a primary vote (which, in most states, is restricted to voters registered to that particular party). It is not a hard task to match names crossed off on the official poll vote tally against those names in the voter registration (if you are set up to do that, which polling companies are), and get an end of day "turnout by party registration" list and then adjust the Questioners data by that result.

[mgr]

25. But,

How do you get to a change in turnout percentages, if you are already adjusting for turnout at the precinct level (e.g. I reweigh the females sampled because they were oversampled, so I multiply them by 0.95). Unless the exit pollsters have access to the polling books, how in the world can they make an adjustment like that unless it is a posteriori to explain the phenomena observed?

Mike

[Mistwell]

40. They DO have that access

At least in California, ANYONE can have immediate access to the polling books as far as the name of who voted, their address, and their party of registration. You could easily get the full list of all registered voters and their party of registration prior to the election, and put it into a spreadsheet on a laptop. Then, you can stand at the polling station and check names off as they are crossed out on the polling book list that is posted at the front of the polling station (usually updated hourly), and the spreadsheet would provide a running total of how many people voted from each party.

At the end of the day, you adjust the "how did you vote, and what is your party affiliation" data by the party affiliation data of everyone who voted, and you have a more accurate exit poll.

Now, I am not saying this is how it was done. I am saying it COULD be how it was done, and it might explain at least some of the change in the exit poll data between the preliminary analysis and the final analysis.

[breadbox]

5. how do you know more men turned out?

I understand that this data may be available from
the polling books, but it is certainly not available
as early as midnight for a re-calibration of the data!
The only valid recalibration information is precinct-level
turnout, as far as I can see.

breadbox

[Mistwell]

15. actually it is available

At least in my state, who voted is listed on an by-the-hour basis outside each voting precinct. The potential voters names are all listed, and crossed out as they vote, for anyone to see. A polling company has access to the potential voter list before the election, and their party of registration. It would not be difficult to simply place a check mark next to the name as their name is cross off on the list, and have a laptop compute a running total of how many people have votes from the different parties.

[Melissa G]

45. All the early voting and possibly absentee data was available

on line in Travis County TX. We were encouraged to look up our friends and make sure they had voted. It only took and instant to put anyone in and have your answer.

[Carolab]

8. This analysis here focuses exclusively on gender though

http://www.globalresearch.ca/articles/KEE411A.html

http://auto_sol.tao.ca/node/view/985?PHPSESSID=89809e29...

<snip>

The National Election Pool’s own data—as transmitted by CNN on the evening of November 2 and the early 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.

It is important to remember how large the discrepancy was between the early vote tallies and the early exit poll figures. By the time polls were closing in the eastern states, the vote-count figures published by CNN showed Bush leading Kerry by a massive 11 percent margin. At 8:50 p.m. EST, Bush was credited with 6,590,476 votes, and Kerry with 5,239,414. This margin gradually shrank. By 9:00 p.m., Bush purportedly had 8,284,599 votes, and Kerry 6,703,874; by 9:06 p.m., Bush had 9,257,135, and Kerry had 7,652,510, giving the incumbent a 9 percent lead, with 54 percent of the vote to Kerry’s 45 percent.

At the same time, embarrassingly enough, the national exit poll figures reported by CNN showed Kerry as holding a narrow but potentially decisive lead over Bush. At 9:06 p.m. EST, the exit polls indicated that women’s votes (54 percent of the total) were going 54 percent to Kerry, 45 percent to Bush, and 1 percent to Nader; men’s votes (46 percent of the total) were breaking 51 percent to Bush, 47 percent to Kerry, and 1 percent to Nader. Kerry, in other words, was leading Bush by nearly 3 percent.

The early exit polls appear to have caused some concern to the good people at the National Election Pool: a gap of 12 or 14 percent between tallied results and exit polls can hardly inspire confidence in the legitimacy of an election.

One can surmise that instructions of two sorts were issued. The election-massagers working for Diebold, ES&S (Election Systems & Software) and the other suppliers of black-box voting machines may have been told to go easy on their manipulations of back-door ‘Democrat-Delete’ software: mere victory was what the Bush campaign wanted, not an implausible landslide. And the number crunchers at the National Election Pool may have been asked to fix up those awkward exit polls.

Fix them they did. When the national exit polls were last updated, at 1:36 a.m. EST on November 3, men’s votes (still 46 percent of the total) had gone 54 percent to Bush, 45 percent to Kerry, and 1 percent to Nader; women’s votes (54 percent of the total) had gone 47 percent to Bush, 52 percent to Kerry, and 1 percent to Nader.

But how do we know the fix was in? Because the exit poll data also included the total number of respondents. At 9:00 p.m. EST, this number was well over 13,000; by 1:36 a.m. EST on November 3 it had risen by less than 3 percent, to a final total of 13, 531 respondents—but with a corresponding swing of 5 percent from Kerry to Bush in voters’ reports of their choices. Given the increase in respondents, a swing of this size is a mathematical impossibility.

The same pattern is evident in the exit polls of two key swing states, Ohio and Florida.

At 7:32 p.m. EST, CNN was reporting the following exit poll data for Ohio. Women voters (53 percent of the total) favoured Kerry over Bush by 53 percent to 47 percent; male voters (47 percent of the total) preferred Kerry over Bush by 51 percent to 49 percent. Kerry was thus leading Bush by a little more than 4 percent. But by 1:41 a.m. EST on November 3, when the exit poll was last updated, a dramatic shift had occurred: women voters had split 50-50 in their preferences for Kerry and Bush, while men had swung to supporting Bush over Kerry by 52 percent to 47 percent. The final exit polls showed Bush leading in Ohio by 2.5 percent.

At 7:32 p.m., there were 1,963 respondents; at 1:41 a.m. on November 3, there was a final total of 2,020 respondents. These fifty-seven additional respondents must all have voted very powerfully for Bush—for while representing only a 2.8 percent increase in the number of respondents, they managed to produce a swing from Kerry to Bush of fully 6.5 percent.

In Florida, the exit polls appear to have been tampered with in a similar manner. At 8:40 p.m. EST, CNN was reporting exit polls that showed Kerry and Bush in a near dead heat. Women voters (54 percent of the total) preferred Kerry over Bush by 52 percent to 48 percent, while men (46 percent of the total) preferred Bush over Kerry by 52 percent to 47 percent, with 1 percent of their votes going to Nader. But the final update of the exit poll, made at 1:01 a.m. EST on November 3, showed a different pattern: women voters now narrowly preferred Bush over Kerry, by 50 percent to 49 percent, while the men preferred Bush by 53 percent to 46 percent, with 1 percent of the vote still going to Nader. These figures gave Bush a 4 percent lead over Kerry.

The number of exit poll respondents in Florida had risen only from 2,846 to 2,862. But once again, a powerful numerical magic was at work. A mere sixteen respondents—0.55 percent of the total number—produced a four percent swing to Bush.

What we are witnessing, the evidence would suggest, is a late-night contribution by the National Elections Pool to the rewriting of history.

It is possible that at some future moment questions about electoral fraud in the 2004 presidential election might become insistent enough to be embarrassing. The pundits, at that point, will be able to point to the NEP’s final exit poll figures in the decisive swing states of Florida and Ohio—and to marvel at how closely they reflect the NEP’s vote tallies.

The Ohio Fifty-Seven (is there a Heinz-Kerry joke embedded in the number?) and the Florida Sixteen will have done their bit in ensuring the democratic legitimacy of the one-party imperial state.


© Michael Keefer

[Bill Bored]

10. Forget gender (at least as far as this poll is concerned)!

The gender weighting was unchanged from the 7:30 PM (80%) poll to the final adjusted one. It was always 54 F/46 M. The only time females were over-weighted was in the very first release at 4 PM at 58%. But once the MSM picked up on this, they kept repeating it even though it was quickly corrected. And Kerry was still ahead with only 54% females. He did not fall behind until the other parameters above were changed.

[Oversea Visitor]

3. Horrible

with 95% respond sample 13,047
Kerry 51 =6,653

with 96% respond sample 13,660
Kerry 48% =6,556

Sorry dont know how clear you need it to be.

Why look at % just use hard cold number of people sample and you get the answer.

Even the exit poll is fixed but only thing they cant explain the above of how Kerry actual drop from 6653 to 6556 hence they cant realease the result haha.

[Bill Bored]

9. Slight misunderstanding

The 95% and 96% were just the ones who answered the question about which party they belonged to. I only included this to show that it was almost the entire sample. The Post version didn't mention the % or number of responses for each question, or even who the winner was! I extrapolated those numbers from the male/female data (who won) and response % in the poll reports before and after the Post version. I assumed it was 95% because it was 95% in the previous run. It could have been 96% as in the final poll. But when asking "Whom did you vote for?" all these were a 100% response. (Non-responders weren't counted.)

[The Judged]

6. Analysis underestimates voter variables, poll timing, and pollster fraud.

IMO

[Oversea Visitor]

7. n/t

:kick:

[texpatriot2004]

12. kick it

[flintdem]

13. Weighting Data

..."At 7:32 p.m. EST, CNN was reporting the following exit poll data for Ohio. Women voters (53 percent of the total) favoured Kerry over Bush by 53 percent to 47 percent; male voters (47 percent of the total) preferred Kerry over Bush by 51 percent to 49 percent. Kerry was thus leading Bush by a little more than 4 percent. But by 1:41 a.m. EST on November 3, when the exit poll was last updated, a dramatic shift had occurred: women voters had split 50-50 in their preferences for Kerry and Bush, while men had swung to supporting Bush over Kerry by 52 percent to 47 percent. The final exit polls showed Bush leading in Ohio by 2.5 percent.

At 7:32 p.m., there were 1,963 respondents; at 1:41 a.m. on November 3, there was a final total of 2,020 respondents. These fifty-seven additional respondents must all have voted very powerfully for Bush—for while representing only a 2.8 percent increase in the number of respondents, they managed to produce a swing from Kerry to Bush of fully 6.5 percent."...


It seems to me that some of the discussion about changes in candidate percentages and changes in the size of the sample don't understand weighting data.

Lets say I take a polling sample of 2000 and it comes out 1500 women (60-40 for Kerry) and 500 men (60-40 for bush) and an overall 55-45 Kerry lead. I decide that the sample is skewed toward women and should be equal numbers of men and women. I weight all women respondents by
.67 and all male respondents by 2.0. I end up with a sample which is 1000 women (55-45 Kerry)and 1000 men (55-45 bush). The internal percentages for each candidate based on gender do not change. The overall sample size does not change (N=2000) but the overall percentages would shift toward bush (now a 50-50 tie) because men (weighted more heavily) favor him over Kerry.

I could also weight the 2000 so that the sample size is 1500 or 2500 and it wouldn't change the internal gender totals just the overall percentage. The problem with all of this discussion of exit polls is that we don't know what was weighted when and why. The final weighting was done at the end of the evening to match the vote totals, but for certain there was also other weighting done during the day based on past and present turnout and demographics. A sudden movement in bush's direction with a very small change in N (number of respondents) could be caused by this weighting.

The final unweighted numbers in Ohio may not be 2,020, they may be 2400 and weighted to 2020 because of some demographic or combination of demographic weighting. A shift to bush with a small change in N does not mean that all of the new respondents went for bush. It may mean the the weighting of the sample was changed. The question is was it done for accuracy or nefarious reasons and only Mitofsky and company know that right now!!!

[euler]

16. Just give up.

<snip>

It seems to me that some of the discussion about changes in candidate percentages and changes in the size of the sample don't understand weighting data.

</snip>

I've been saying the same thing for 3 weeks. I've even provided links to web sites that explain how exit polls are conducted. Still, people continue to believe that moderate swings in the exit poll percentages accompanied by a small or no sample size increase somehow points to fraud. It's not even a smoking gun. I found it amazing that Arnebeck even included this as evidence in his lawsuit.

Here's a link one more time for anyone who wants to know how exit polls are conducted.

http://www.mysterypollster.com/main/2004/11/the_difference_.html

[flintdem]

17. Thanks- I'm frustrated too

The "groupthink" is so thick you can cut it with a knife.

Nobody seems willing to look at objective, methodologically sound evidence. All I can conclude about those who keep saying that exit polls are always accurate is that they are being deliberately fraudulent or are self deluded. There is too much evidence to suggest that American polls are biased toward one party or the other. Mitofsky himself, has found that 60% of all American exit polls are skewed toward one party or the other(Mitofsky, Public Opinion Quarterly 2003 p. 51).

For once instead of someone just saying "it ain't so"- prove that it isn't so (and not just pointing out a few isolated cases in the US like Utah- exceptions prove the rule, not refute it!), because there is peer reviewed, scientific articles like Mitofsky's that say otherwise. (And please don't hide behind calling me a freeper just because I'm an objective social scientist, first and a Democrat, second.)

I'm sorry if I'm messing up this topic by venting...

[Bill Bored]

20. And Welcome to DU! n/t

[Bill Bored]

21. Don't be sorry

And welcome again!

We should try to avoid the group-think; it doesn't get us anywhere. It's too easy to believe what you want to believe and paste a bunch of numbers into a spreadsheet and multiple threads to attempt to prove it.

I'd prefer to believe the election wasn't hacked because if it were, I think this is a much bigger problem than just fixing our party, or the media. But I'm still open to the possibility that it was hacked and I wouldn't put it past the Bushies.

The media and perhaps the party are hard enough to fix without elections fraud, but on the other hand, the vote is for all practical purposes unverifiable and this demands our immediate attention!

Fortunately, these scenarios aren't mutually exclusive. We can work on election reform and verifiable voting, probably at the state level, try to find evidence of outright fraud and also try to reform the media and if necessary, the party.

The question is how to prioritize these. I think verified voting should be #1.

Also, as Dr. Steven Freeman points out in his latest draft, there are various degrees of fraud. Some are legal, others are not. You can steal an election "fair and square", or you can completely rig it. Verifiable voting should greatly reduce the possibility of the latter but the former is an ongoing problem and I'm not sure exactly how to go about fixing it completely.

Most of us could just register as Republicans so they can't target us, and leave a small group (e.g., 25% or so) to choose our Dem candidates in the primaries but then who would believe it? I'd love to see THOSE exit polls! I think the Repubs have used this exact strategy to some extent, esp. in FLORIDA! The Butterfly Ballot was designed by a Democrat you know.

[flintdem]

23. Nicely Said!!!

I also prefer not to believe that the election was not hacked but I am willing to be convinced by solid evidence (which I haven't yet seen on DU). I think we should be careful that we don't end up de-legitimizing our own electoral process by unfounded charges of fraud.

By all means we should have the best, verified electoral system in the world. In addition, 1-2% of spoiled ballots every election is far too many, particularly when the nation seems so evenly divided.

Thanks for the welcome. I found this site in September, lurked for a couple months and decided to join when I saw all of the statistical analysis going on. Unfortunately, (unlike my rant above)even mild suggestions that there were multiple ways to interpret the data got me branded as a freeper (so I dropped off the boards for a couple weeks). I was about to give up on this site because it seemed that everyone bought the fraud argument.

[mgr]

26. You're frustrated, I am stupified

I am both a geographer and a biologist (plant geographer) with ample training in research design and application of statistics to it. I work with statistics everyday in my job characterizing hazardous materials. For me, the only way I could demonstrate fraud would be to take the election results as an accurate or 'correct' outcome once I removed the absentee and early voter results. I would do a goodness of fit test on the raw results of the exit polls with the adjusted election outcome, and only if the raw results failed the goodness of fit test would I consider the possibility of fraud.

We have been taking the sample and comparing it to the population, rather than the other way. But I guess that is now known as 'spin'.

If I recall, it was only by this method that lethal or epistatic gene interactions were identified.

Mike

[m.standridge]

28. Now, about the Electoral College

Perhaps this isn't the "right" thread to bring this up in, but I don't see a better one.
I keep seeing these "anti-exit poll theory" posts by one or another.
The common argument, is that the polls could be in error because individual pollsters could have submitted bogus data, because some of the sample sizes were too small, and because the samples weren't perfectly attuned to the real demographics of the US electorate in the various parts of the nation in which the polls appeared to be in error.

These/this argument is usually raised in regard to the overall national exit polls, those from election night.
I refer, also, to a set of exit polls that happened BEFORE election night.

I refer you to the PRE-election polls.
What do you see?
Do you really see a Bush landslide, or even a Bush 2% lead?
NO.
You sometimes DO see a Bush lead, though.

But here's the thing: YOU DON'T REALLY SEE A SOLID BUSH LEAD IN THE STATES THAT COULD PUT KERRY OVER IN THE ELECTORAL COLLEGE.
You see it back and forth. In Ohio. In Florida. In Nevada. In New Mexico. In Colorado.
The swings in those states, pro and con, were dramatic, as they were in MI, WI and IA.

It's kind of like this:
You can argue that the exit polls may not have been accurate enough across the whole country, to predict how well Bush would do versus Kerry, or whether he'd had a Popular vote lead over Kerry, or vice versa.
But when you look at and factor in the overall national pre-election polls from the various pollsters, in the 48 hours before the election, and then look at the state-by-state pre-election polls and how volatile they had become in the states that were important in the Electoral College, and then look at the exit polls in the specific states that were important in the Electoral College and which were also battleground states--when you do that, you're on much thinner ice.

It's this Electoral College thing, that makes me think there's something to this exit poll thing. It's not so much whether Bush had some national lead. It's whether he actually carried Ohio or Florida or New Mexico or Nevada. They could have been off quite a bit, and still have been accurate enough, to have accurately predicted the outcome in those specific states.
This only involves a few thousand votes--well, 380,000 in Florida, 117,000 in Ohio, 5988 in NM and around 20,000 in NV.
That's not many votes. Those were the votes that decided the election. Those were the votes that tipped the Electoral College.
Not the overall Popular vote or its size or composition or location.
One could be doing quite well, thank you, and still not have gotten those votes. The exit polls could have been quite wrong, overall, and still have gotten that much right.
Then, factor in all those headline stories in the Ohio media:
1. burglary in Demo. offices late October, Toledo. No money taken, only computer data
2. lockdown in Warren County, citing non-existent FBI warning;
3. refusal of voting machines to voters in heavily-Dem Cuyahoga Cty.
4. TRIAD tech visiting at least two county offices, shutting down computers, removing parts, going into the recount

There are other things, but those four show, just in themselves, that someone was up to something. And Ohio was quintessential to a Bush win in the Electoral College. And the PRE-election polls were, mostly, showing that NOT happening there.

[mgr]

33. pre-election polls did not show Kerry winning

Depends on which pre-election poll you looked at. Was it likely voters or registered voters. I recall many being too close to call.
I was following the polls every day, and accepted the undecideds swing to the challenger argument.

Mike.

[flintdem]

34. Undecided Voters

The LA Times exit poll showed that those who decided over the last weekend went 47-46% for Bush and those who decided on Monday or election day went 52-45% for Kerry.

http://www.latimes.com/news/custom/timespoll/la-110404superchart-g,1,6130757.acrobat?coll=la-news-times_poll

[mgr]

35. not seeing it at site. Correct URL?N/T

[flintdem]

41. Different Link

here is a different link:

http://www.latimes.com/news/custom/timespoll/

You do have to register with the LA Times to access the site.

[mgr]

36. Belay that

That is a pretty weak break to Kerry, it implies that he and Bush split them up 50/50.

mike

[euler]

39. I would delete this. No one will take you seriously.

[Salomonity]

29. glad to have you

You're of course correct.
I've had some stats courses; the exit poll results do not prove kerry won, and attempts to show otherwise merely make it look as though the Left doesn't understand mathematics.

We lost this time. Let's not let it happen again in 2006.

[m.standridge]

30. are you suggesting we should ignore the Electoral College?

Gore couldn't.

[Bill Bored]

18. flintdem, this is exactly what I did

but instead of using gender, I used a party affiliation change that Mitofsky actually made in his final so-called adjusted poll, plus 2 other changes that he actually made.

Taken together these 3 changes completely account for the switch from Kerry to Bush. And I used the national sample, not Ohio.

1. Dem/Rep ratio (a true weighting change as you define it)
2. Independent Bush/Kerry vote ratio (changed by Mitofsky but arguably not a true weighting change)
3. Democratic and Repub Bush/Kerry vote ratio. (also changed by Mitofsky but not a true a weighting change)

There are a few reasons I did this:

One was that voting machines don't "know from" gender, they only know the votes. If there was actual vote shifting, this is what may be reflected in the poll results adjustments. I.e., more Dems for Bush, fewer Repubs for Kerry, more Inds. for Bush. If you wanted to hack a program, this is how you might do it.

There is also a change in party affiliation that was made, which indicates that Dems and Repubs had equal numbers of voters, regardless of how they actually voted. This one could actually be verified from poll records. There are supposedly more Dems than Repubs, but according to the adjusted totals, an equal number from each party actually voted, and more Dems voted for Bush than Repubs for Kerry.

One other reason I posted this is because of the threads that are relying so heavily on answers to questions about 2000 voting instead of the actual 2004 party affiliation question to show how party loyalty was adjusted in the final poll.

[flintdem]

19. Thanks Bill

Actually I was responding to one of the other posts in the thread and I failed to post it under that one.

What you did is sound. What I think is the great unknown is what other weighting was done during the day- were precincts weighted because of perceptions of turnout, were there initial turnout weights used that mirrored 2000 that were dropped or modified during the day, were there assumptions made about party breakdown in the turnout weighting and were those correct? Everyone seems to agree that the exit polls had a 1.9% Kerry skew and without the internal processes (which Mitofsky hasn't released) used on the data through election day I think we are completely in the dark about where it came from. It could be poor design or implementation, fraud or something systemic about Americans and polls and you're right without Mitofsky's info we can't prove any of the three.

[jbnow]

22. I don't get it

I don't understand how exit polls works. I know they should be more accurate then other polls because these are people who actually did vote.

Or at least thought they did. We know spoiled ballots and defaults to bush abounded in Dem areas...but could they have affected the vote to that huge degree?

I've read statements about how exit polls don't account for the high number of early and absentee votes, but that is the opposite of an explanation for this. In the states I have looked at I remembere Kerry doing better in these early/absentee then he did at the polls.

But what boggles the mind is how they amend them at the end to reflect the actual votes tallied and openly say so. What? Then let them hire me, I will do it for 25% of the price and skip all the polling part and use the tallied votes to make one up. It seems about as valid. Wish I had lnown this was scientific methodology, I could have amended several failed experiments or equations in college. I realize I am mocking this without understanding the basis for it, but if it is something they change after the fact and that have no validity in themselves...why do people pay so much to have them done?

[mgr]

24. Let's see if I can explain it.

An exit poll is conducted election day, it only samples voters on election day. If Kerry does better on the absentee and early voter ballots, this will obscure the Bush turnout on election day. What needs to be done is to subtract the absentee and early voters from the total, and compare the residual to the exit polls.

For example, I am in state that may have had 5 to 6 million absentee ballots (that was the way I voted), image that these went 60% Kerry, but the voter turnout went 51% Kerry (say 15 million voters), the actual voter outcome would be (5 million times 0.6) + (15 million times 0.51)which will give you about a 53% turnout.

The spoiled ballots being not on the same order of magnitude as the turnout, will have little impact nationally on the outcome.

What the adjustment of the supposed raw data is obtained from, I have no idea. The fact that they are already adjusting the data to patterns of the past election, suggest that you cannot with good faith accept the 'raw' data that was out before it was adjusted. For all I know, the adjustment might be a census data correction at the local precinct that adumbrates upwards through the sample.

[JMDEM]

27. Here are some answers to your questions

http://www.appliedresearch.us/sf/Documents/ExitPoll.pdf

[Bill Bored]

31. kick n/t

[flintdem]

32. Mitofsky Numbers were weighted before 7:30pm!!!

Slate and Mickey Kraus are reporting that the exit poll numbers were weighted wrong by Mitofsky and company at 7:30pm (and perhaps even earlier in the day?):

" As late as 7:33 P.M. on Election Day, Mitofsky and Lenski were apparently telling their clients (NBC, CBS, CNN, AP, etc.) that after "weighting" Kerry was beating Bush by 9 points among women and losing by only 4 among men. By 1:24 P.M. the next day revised results revealed that, in fact, Kerry won women by only 3 points while Bush won men by 11 points. Whoops! ... It wasn't the dumb bloggers who didn't understand on Nov. 2 that they were being leaked "complex displays intended for trained statisticians," as Mitofsky would have it--or the dumb Kerry aides and dumb Bush aides who believed the same numbers. It was that the weighted results Mitofsky's statisticians put out were full of it! ..."

If this holds up, all of the exit poll examinations using "the unweighted data" from before 1am election night are full of crap. This may explain why Mitofsky doesn't want to release anything- they screwed up the numbers on election day before releasing them!

Here is the slate article: http://slate.msn.com/id/2111460/

Here is the 7:30 release: http://www.scoop.co.nz/stories/pdfs/Mitofsky4zonedata/US2004G_3798_PRES04_NONE_H_Data.pdf

Of course this will be explained away as bogus data released to hide the fraud...sigh

[Bill Bored]

38. Yes, but that's not weighting.

The male/female weighting was 46/54 right? This is the same as the Nov 3 1AM poll, in my Wash. Post link in the top of the thread, and the Nov 3 1PM adjusted poll.

The only time females were actually overweighted was in the first release at 4PM Nov 2, at 58%.

I decided to look at party affiliation instead of mail/female for the reasons outlined above. The weighting of 3rd parties was always 26% but their votes were shifted toward Bush only in the adjusted poll. The Dem/Rep weighting changed in the adjusted poll too. And the percentages of votes for Bush from ALL parties increased in the adjusted poll. I find this all very interesting but I'm still not quite sure what to make of it.

Other DU exit poll aficionados have not discussed this much either, preferring to rely on the more questionable survey of 2000 voting preferences, taken in 2004.

I do agree that the early polls were also weighted but that's not the same as being adjusted. The final one had both weighting changes and other adjustments.

Am I right?

[flintdem]

42. Still looking at the data

I guess it depends on the defintion. All weighting is adjustment, adjustment to match the actual vote (at 1am), or adjustment to expectations of past turnout, or expectations of gender or party (during the day?).

Mystery pollster says:

" The first two releases were weighted only to bring their geographical distribution into line with hard counts of actual turnout" (http://www.mysterypollster.com/)

Which makes Slate's comment nonsense if that is all they did.

By the way if you're interested in a comparision:

The Mitofsky 7:33pm shows 54-46 female. The LA Times 2004 exit poll (http://www.latimes.com/news/custom/timespoll/) had it 51/49 female/male (but that 3% shift is still within the margin of error). It also shows females 45/55 bush/Kerry and males 51/47 bush/Kerry. The Times has women 49/50 bush/Kerry and men 53/46 bush/Kerry.

So Mitofsky has more women voters and more female support for bush than the Times, but the party id numbers are about the same 38/36 Dem/rep (mitofsky) and 40/39 (Times).

[Bill Bored]

43. There's Another National Exit Poll?

I'm not registered with LA Times but if they did their own poll, this is huge! Are you sure they're not just stating the vertical numbers from Mitofsky, making it look different than the horizontal ones?

Someone should correct me if I'm wrong, but I define weighting as what percentage of the total sample each category of respondents falls into, e.g., male, female, Dem, Rep, Black, White, etc. -- NOT how they voted. So if they say they messed up weightings, it means their % of women was too high, or their % of Dems was too high, etc.

Now, if the sample percentages are changed slightly to correct weightings, why would this be accompanied by changes in the way these groups voted? E.g., if 54% of women voted for Kerry and 58% of your sample is women and you decide you sampled too many women, and you reduce the percentage of women in your sample from say 58% to 54%, the 54% sample should still have voted 54% for Kerry, unless I'm wrong.

I'm working on some other stuff using Mitofsky's reports. Will post later, but the short answer is that of all the weighting changes between the latest unadjusted (Wash Post) 1AM poll and the final adjusted 1PM one, Party Affiliation seems to have the greatest effect on the vote. This alone is enough to make the election a draw, within the margin of error (Kerry wins by only 0.6%). Every other weighting change I could find, which are quite small, did not even come close to giving Bush a victory.

So in short, to give Bush the win without actually shifting votes, they had to tweak the ratio of Dems to Repubs from 38/35 to 37/37. So, does anyone know the actual turnout number by party, or are these bloody polls our only source for this???

Why is this important? Because it shows that the only other way the polls can show a Bush win is to actually shift votes to Bush. And this is what the fraud allegation is all about, isn't it?

[super simian]

37. "enough threads to knit a sweater"

Good one!!!
:toast:

[Mistwell]

44. Would that be an orange sweater? [EOM]

.

[Bill Bored]

46. Hate to do this, but I started another thread about this

<http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x236572>