My conversation with Mitofsky (exit pollster) READ IT NOW!!

[ClintCooper2003]

(One caveat: I'm not always nice and polite during this back and forth series of emails. Forgive me.)

Date: 11/24/2004 9:23:20 PM Pacific Standard Time
From: ClintCooper2003
To: [email protected]

I was just wondering something. Those late afternoon exit polls that were weighted by gender, party ID, age, and race, and placed ALL OVER the CNN.com website with charts and graphs of every conceivable type for every single state in this country - are you trying to tell us that THOSE EXIT POLLS were the same ones that were casually leaked earlier in the day?

Is that the shit you're shoveling on us now?


Subj: Re: Exit Polls.
Date: 11/25/2004 4:49:07 AM Pacific Standard Time
From: [email protected]
To: [email protected]
Sent from the Internet (Details)

There were no exit polls weighted by gender, party ID, age or race. That is not part of the exit poll weighting. Second, the exit polls on the web were not released by us. They were leaked by people I don't know, the numbers they leaked were wrong, and I have no responsibility for bad information I did not put out. All the numbers I released were after the polls closed. All the projections were correct. If you think that is shit you can eat it, for all I care.


Subj: Re: Exit Polls.
Date: 11/25/2004 2:24:57 PM Pacific Standard Time
From: ClintCooper2003
To: [email protected]

I'm not talking about the numbers put "on the web." I'm talking about the numbers put on cnn.com that were there until PAST MIDNIGHT on election night. Are you suggesting that some hackers got a hold of cnn.com and put the numbers favorable to Kerry there and you didn't do anything about it for SIX HOURS!

That seems completely incredible to believe. Additionally, the numbers on cnn.com didn't appear for each state until directly after the polls had closed. And yes, I'm sorry, but I distinctly remember the percentages of voters that were Democrat, Republican, Female, Male, etc... and they clearly had been weighted at least to the extent that they represented the general voting population in each state.

If what you say is correct about cnn.com's figures, then there definitely would have been an uproar at CNN and Wolf Blitzer would have come out strongly against the web numbers IMMEDIATELY saying that those numbers couldn't be trusted.

I smell a rat.


Subj: Re: Exit Polls.
Date: 11/25/2004 7:45:55 PM Pacific Standard Time
From: [email protected]
To: [email protected]
Sent from the Internet (Details)

I am sure cnn posted numbers we released. How old the numbers were when you saw them is something I do not know. Furthermore, the numbers you are describing are not the estimates we used to make projections. The projection numbers do not come from tabulations of the vote by age, sex or anything else.



Subj: Re: Exit Polls.
Date: 11/26/2004 10:10:02 AM Pacific Standard Time
From: ClintCooper2003
To: [email protected]

What would the point be in posting numbers very late in the day that would be useless in creating projections and leaving them there past midnight, when it would already be clear who the winner was in each state? You must at least admit that waiting to recalibrate the data until long after it became clear who the winner was creates the appearance of fixing the exit polls to match the results.

Additionally, however, I think you should look at the following data I was able to grab from election night:

In the original Ohio exit poll, among White Men (40% of survey), Bush gets 53% and Kerry gets 47%. Among White Women (45% of survey), Bush gets 53% and Kerry gets 47%. So among these subgroups, the percentages are the same.

Now, on the corrected data, suddenly we find that among White Men (still 40% of survey), Bush gets 56% to Kerry's 43%. And among White Women (now 46% of survey), Bush gets 55% to Kerry's 45%.

Thus, the proportions of White Men and White Women are almost exactly the same in both the afternoon release and the subsequent "correction" of figures

Additionally, I distinctly remember that the sample size remained almost precisely the same from the first set of figures to the second.

Your response may be that the very last people who voted on election day in Ohio tended to be Bush voters and that the mid to late-afternoon figures were skewed in Kerry's favor, but there really would have had to have been a huge shift to Bush right towards the end to overcome Kerry's strong early lead in Ohio.

Also, this is simply provably false in any case. Nearly all of the polling stations that still had long lines at the end of the day were in Democratic precincts, meaning at the very end of the voting time-frame, there would have been a shift towards Kerry anyway.



Now, in your response to me, you said "how old the numbers were when you saw them I do not know..." Did they leave you out of the loop as to what numbers they were using? If so, I don't understand why you weren't more actively involved.

Another important thing I want to make clear is this: I followed cnn.com for a long time on election night and I noticed that cnn.com did not even put up the exit poll figures until the polls had closed for that particular state. Therefore, the numbers that were used this year, by default, must have been numbers that came from later in the day than in the Election 2000.

I went back and looked at some of the "raw" numbers from Election 2000. What I noticed was that the "uncalibrated" data from 2000 (which, by the way, was released a bit earlier in the day than this year) was simply far more accurate. In fact, among the battleground states only Wisconsin was off by more than 5%.

Let me ask you something: Have exit polling techniques changed radically over the last few years or has the way we count the votes changed radically over the last few years?



Subj: Re: Exit Polls.
Date: 11/26/2004 5:18:30 PM Pacific Standard Time
From: [email protected]
To: [email protected]
Sent from the Internet (Details)

I never said the results were not recalibrated. They were constantly recalibrated. I said the posting did not necessarily reflect the recalibration. I have no idea what CNN posted or when it was updated. Maybe it was often; maybe it was seldom. Ask them; not me. If the sample size remained the same it sound as though there was not much new data. The posting you are referring to was of analytical data. It was crosstabs. That was not the data analysts were using for projections once the polls closed. There seems to be some disconnect between what I am saying and your interpretation. Your interpretation, to be plausible, should accommodate what I keep telling you.



Subj: Re: Exit Polls.
Date: 11/27/2004 11:15:45 AM Pacific Standard Time
From: ClintCooper2003
To: [email protected]

You are not responding to my arguments.

As for your lack of knowledge as to when cnn.com was updated, what figures they were using, etc... I believe you are being deliberately disingenuous. I find it incredibly hard to believe that you had no knowledge of what numbers they were using and when they were using them.

Once again, I am wondering when you thought it would be appropriate to make a projection. Let me put it this way: The first batch of exit polls for California was posted shortly after the California polls closed and CNN made an immediate projection for John Kerry to take California. The exit poll showed him leading by only 8 points.

The exit poll for New Hampshire was also posted shortly after the polls closed there, and it showed Kerry leading by 10 points - in a much smaller state with a nice-sized sample and a smaller margin of error.

I want to know why CNN didn't immediately call the state for John Kerry. Was it because someone at CNN got a call from the Bush camp saying their exit poll numbers were "different"?

You state "The posting you are referring to was of analytical data. It was crosstabs." Okay, fine. Then why post the numbers in the first place? Why was there no caveat issued with the release of the early evening figures?

And also, why wait until 1 a.m. to readjust the figures to make a "projection?" Are you saying that your early evening exit polls were entirely useless? Or do you think that CNN wanted to deliberately mislead the public about the outcome of this election? Once again, you are not responding to my arguments.

You also didn't say one word about how tabulation technology has radically changed over the past few years.

Also, here are my other arguments that you said absolutely nothing about:

"Additionally, however, I think you should look at the following data I was able to grab from election night:

In the original Ohio exit poll, among White Men (40% of survey), Bush gets 53% and Kerry gets 47%. Among White Women (45% of survey), Bush gets 53% and Kerry gets 47%. So among these subgroups, the percentages are the same.

Now, on the corrected data, suddenly we find that among White Men (still 40% of survey), Bush gets 56% to Kerry's 43%. And among White Women (now 46% of survey), Bush gets 55% to Kerry's 45%.

Thus, the proportions of White Men and White Women are almost exactly the same in both the afternoon release and the subsequent "correction" of figures

Additionally, I distinctly remember that the sample size remained almost precisely the same from the first set of figures to the second.

Your response may be that the very last people who voted on election day in Ohio tended to be Bush voters and that the mid to late-afternoon figures were skewed in Kerry's favor, but there really would have had to have been a huge shift to Bush right towards the end to overcome Kerry's strong early lead in Ohio.

Also, this is simply provably false in any case. Nearly all of the polling stations that still had long lines at the end of the day were in Democratic precincts, meaning at the very end of the voting time-frame, there would have been a shift towards Kerry anyway.

Another important thing I want to make clear is this: I followed cnn.com for a long time on election night and I noticed that cnn.com did not even put up the exit poll figures until the polls had closed for that particular state. Therefore, the numbers that were used this year, by default, must have been numbers that came from later in the day than in the Election 2000."


Subj: Re: Exit Polls.
Date: 11/27/2004 2:24:54 PM Pacific Standard Time
From: [email protected]
To: [email protected]
Sent from the Internet (Details)

Be realistic. On election day we send reports to six members and about 100 subscribers. Do I know what we send? Yes. Do I know what they post on their web sites and when? Absolutely not. If you think that is disingenuous then so be it.

I think it is appropriate to make a projection when at least one of the dozen estimates has a sampling error relative to the difference between the top candidates gives me a maximum chance of error of 1 in a 100. That is the first requirement. There are many others that have to do with data quality. That's when I make a projection. Projections were made in wide open races from exit polls. In others it took vote returns in sample precincts. Still others it took the vote tally by county. It all depended how close the race was. None of this came from the analytic data posted by CNN or anyone else. That is not the source of the estimates. It is not even all the exit poll data. NH was one of the states we told the members to ignore the exit poll data. There were eight other states. Those tabulations were adjusted by us several times during the night. Again, I only know what we sent to CNN and others. I don't know what is on their web site. I did not see it on election night. Nor did I see what was on television. As you might imagine, I was otherwise occupied. As for your other questions, the answers are in what I said above. I don't know what networks post. And if your snide comment about someone telling us what to do is serious I'm sorry I wasted a minute trying to explain things to you.
warren mitofsky


Subj: Re: Exit Polls.
Date: 11/28/2004 9:27:14 AM Pacific Standard Time
From: ClintCooper2003
To: [email protected]

I apologize for the snide comment. It was inappropriate and I do appreciate your willingness to correspond with me.

I was just curious about a couple of things. Why is it that you told the members to ignore the exit poll data in states like New Hampshire and Ohio, but not in states like Delaware or New Jersey? What were the 8 other states besides New Hampshire where you told the members to ignore the exit poll data?

I had another couple of quick questions - as for the initial exit poll data that CNN.com and others were putting on their websites, do you know approximately what time during the day that you sent those data to them? Also, do you know what time-frame that data was from (say, 8 a.m. to 4 p.m. or 3 p.m. to 6 p.m. for example)?

What time of the day did you send the members the updated data that were used to make more projections?

Do you believe that differential non-response played a major factor in the initial exit polls being so far off in certain states?

Also, have exit pollsters always waited for the results and used them to make a projection?

Do you believe that the exit polls this year may have undersampled the "rural vote"?

I'm sorry for being so inquisitive, but you must understand that there are millions of people like me who were absolutely confused on election day.