After Reading The MITOFSKY Report Some Discussion Is Needed

[althecat]

Please check in here with your thoughts after reading (or skim reading) the full MITOFSKY EXIT POLL REPORT.

http://www.exit-poll.net/election-night/EvaluationJan192005.pdf

It is interesting that this report has been posted in full. They must have felt embarassed by all our outrage.

MY PRELIMINARY VIEWS...

It is my opinion that there is a remarkably large amount in that report that could be of use to us.

Specifically:

Within Precinct Error (WPE)
Within Precinct Error (WPE) is an average of the difference between the percentage
margin between the leading candidates in the exit poll and the actual vote for all sample
precincts in a state. The signed WPE gives the direction of this error; in this report a
negative WPE represents a Democratic overstatement in the exit poll and a positive WPE
represents a Republican overstatement in the exit poll. The absolute WPE represents the
total error.

This seems to be in effect a measure that of error in each individual precinct. This is exactly what is needed to isolate the hot spots for vote fraud.

A great deal of the report deals with analysing the wherefores and why's of the WPE's with respect to the characteristics of the interviewer (gender, age, distance from polling booth etc.)

In addition they also analysed the WPE with respect to the type of voting machine's used. And compared swing states with non-swing states.

In both of these later analyses there is a hint which is worth digging further into. Specifically... there is quite a range in WPE's by voting technology - Paper Ballots are spectacularly more accurate.... plus.... Swing states showed higher levels of WPEs than non swing states.

HOWEVER AND THIS IS A BIG HOWEVER

What is noticeably missing from this at first glance.

1. Analysis of the WPE by state.

WHY NOT PROVIDE A WPE NUMBER BY STATE.. they have charts with completion rates by state - but WPE by state is missing.

2. Analysis of the WPE outliers - and a discription of the distribution of WPE errors.

What this report does in all instances is provide median and means for different types of precinct.

It is probably reasonable to assume that if fraud occurred - then it is likely to have occurred in some places and not others. I.e. in some fraction of the 1400 precincts surveyed rather than in all of them. If so then what we really want is some sense of what the raw WPE data looks like both in distribution terms... and in terms of where specifically the highest WPE's occurred... did they for example occur in precincts that for some other reason we have suspicions.

THEREFORE IN MY OPINION CONYERS SHOULD NOW ASK FOR...

1. A full list of the WPE numbers by precinct.
2. A median and mean WPE for each state.

al

[bleever]

1. Rock on, al. Looking forward to chewing this over. n/t

[Shalom]

2. Perhaps A Logical Explanation is also in Order:

Is the speculative conclusion that Kerry voters were over-represented based on any factual information, or the assumption that the final results were correct (i.e., assume no fraud, so there must be a logical explanation).

It's hard to see how this circular reasoning can then allow the pollsters to conclude there was no fraud.

[s-cubed]

3. We can determine how much "oversampling" would be

required & if that is reasonable.

[qwghlmian]

4. Here is a study according to which

the non-responders in exit polls tend to skew the data toward Democrats:

http://www.duke.edu/~mms16/non_response2000.pdf

[Chi]

23. "will improve the quality of exit polls"

Heya qwghlmian
I think this section is quite relevant.

"However, because the non-response patterns are systematic, we would encourage exit
pollsters to use statistical methods to control for non-response biases in these specific directions
when using data to predict election results. We believe that understanding these potential biases
will improve the quality of exit polls’ predictions and their usefulness in studying public opinion
and voting behavior."

Mitofsky knows how to do exit polls, and this report has been out for years (at least the pre-revised version was).
I would think that would give him a more accurate exit poll, not a less accurate one.

[davidgmills]

58. Please quit posting this Mormon report

Mormon's do not represent white America.

I know, my daughter was one for a while and I read up on this religion. If Mormonism survives long enough it will eventually be a split off from Christianity like Islam. Their holy book will eventially become the Book of Mormon and not the Bible.

These people have some very strange and odd beliefs compared to your average Judeo/Christian.

As I said before, all this article tells me is that white Mormons over forty don't talk to pollsters.

Find another article on the subject.

[qwghlmian]

59. They also cite other studies

that show that non-whites respond to exit polls a lot more readily than whites (Brehm, John. 1993. The phantom respondents: Opinion surveys and political representation. Ann Arbor: The University of Michigan Press.) and studies that show that young people respond to exit polls more readily than whites (DeMaio, Theresa J. 1980. Refusals: Who, where and why. Public Opinion Quarterly 44 (Summer): 223-233.) and (Herzog, A. Regula, and Willard L. Rodgers. 1988. Age and response rates to interview sample surveys. Journal of Gerontology: Social Sciences 43 (No6): S200-5.)

Are those Mormons too? You just don't want to hear stuff that does not fit your preconceptions, do you. Freeman, whom you love, cites Utah Colleges exit polls in support of his arguments (erroneously, since the methodology is quite different from Mitofsky's polls) - do you object to that as well?

In another post here you claim that just because "Freeman postulates" something, it should be taken into consideration, although he has done no studies whatsoever to back up his "postulations" and basically pulls them out of thin air. Yet when faced with actual studies that do not agree with your preconceived opinions, you dismiss them. I guess "postulation" trumps "research" in your book.

[althecat]

5. That is definitely where this report is written from...

.... I.E. explaining why the poll was wrong, rather than considering whether it discloses any evidence of fraud. From what I could see the only time they address the fraud question is where they consider voting technology.

From our perspective we want to look at the same data they use from the opposite perspective. I.E. If there was fraud then what would we expect to see in this data.

Unfortunately the report is not particularly helpful in that regard. That said what it does contain does point us down a few new avenues of inquiry.

What is encouraging - a bit - is the fact that the report does not address obvious paths which would have more effectively debunked the fraud theory.

For example why do they not discuss the distribution of within precinct errors. If it was fairly uniform and consistent then it would point - as they claim - towards there being some kind of problem with the methodology.

Instead what they do is sort of paddle around the edges of the pool of fraud analysis without actually jumping in and swimming.

[Igel]

7. In a word, yes.

On the one hand, Mitofsky assumes the final election returns (turnout and how the voters voted per precinct) is correct. He doesn't conclude there's no fraud in the elections, I don't think; he's more concerned with showing that he didn't screw with the exit poll numbers. I.e., to show that the mistakes were honest and/or unavoidable.

On the second hand (we have more than 2 here), Mitofsky knows that many people didn't respond to the poll: he knows how many. That's the rational for many of the WPE breakdowns: if it's raining, if the interviewer is young, if the interviewer is recently hired, if there are lots of people ... you get more non respondents. Some voters refuse to answer, some the interviewer just misses, others the interviewer can't get to respond, yet others don't want to stop in the rain. Many reasons for missed people.

On the third hand, he knows that some of his data were corrupted. Esp. in cities: if you have 3 precincts in one room, and the interviewer is stuck 50 feet away from the entrance, but is only to interview every 8th voter in a single one of the three precincts ... well, you see the problem. How does the interviewer know who to talk to? And do the voters know what precinct they're in?

On the fourth hand, Mitofsky has some scheme for allocating those non-respondents. Otherwise there's a large gap in the data (and results like Kerry 50 +/-15%, Bush 48 +/-15% nobody wants to see). How he does it is a mystery to me. He knows that non-respondents aren't randomly chosen; and based on history those that were likely to have voted for Kerry are also likely to be respondents. Moreover, every one of his presidential election polls since 1988 swings dem, and he has to combat that bias.

Two things I found interesting in the report. The first, that WPEs like this were only found in the last time voter turnout reached the level it did in Nov. '04; the second, that in very Kerry precincts the exit polls overestimated * support. I don't know what to make of them, but I find them interesting.

[Imnottelling]

26. I have a question for you.

You said:
"He knows that non-respondents aren't randomly chosen; and based on history those that were likely to have voted for Kerry are also likely to be respondents."

Why aren't non-respondents randomly chosen? I don't have the energy to read that report that was posted regarding non-respondents so can you summarize the argument for me? :)

On the surface, it seems to me that a non-respondent (for reasons like they are too tired to answer the poll, bad weather, late for something, etc.) would be random.

[Igel]

40. "Volunteer effect"

At least that's what some people call it.

Mitofsky doesn't choose the non-respondents. The non-respondents do. Mitofsky may tell the interviewers how to select people; but the interviewers can successfully carry out his instructions or not; and those selected can either participate or not.

Why is it non-random? Almost everything that requires volunteering or self-selection is non-random--human behavior is complex (and I'm glad I don't have to deal with most of it).

And since a number of non-respondents probably didn't actually have the chance to respond, we can't rule out that there was some non-random factors in some locations predisposing those people that the interviewer did ask to participate to be slightly more pro-Kerry than the overall precinct. Maybe the interviewer was near a bus stop, and more repubs drove themselves. Who knows?

[Imnottelling]

43. Right

I understand what you are saying about the complexity of human behaviour and human behaviour being the reason for the non-responce. However, it seems that maybe GOP people would not respond because their car was right there and they wanted to get home. Maybe a Dem would not respond because their bus had just arrived. Mabye the Dem missed the bus and figured they had some time to kill so they would do the poll.

It seems that there could be so many reasons for a non-response that they are essentially randomized across party lines again.

Your last sentence does it for me.

"Who knows?" means to me that the reasons are random. What do you think? What am I missing?

[Igel]

52. I think you misunderstood me.

In the 1990s African-Americans didn't apply to Slavic PhD programs (at least in the numbers stats said they should, if they chose disciplines randomly). I've heard some racist, some silly, and some presumptuous answers. What's the "real" answer? Who knows?

But nonetheless they almost never applied, and many deans screamed at Slavic departments. The problem was real, the answers of dubious quality.

The repub tendency to non-responses in exit polls dates back to when they started doing presidential exit polls: 1988. About 2.5 to 3% error rate because of it; sometimes higher. You'd expect it to be random; but it's reliably, robustly non-random. The higher rates correspond (on the face of it) to higher turnout and (and, per Mitofsky) higher interest in the presidential race. The only people that don't accept that this tilt occurs are those that stand to have their argument lost or severely weakened by accepting it. Otherwise the tilt's been reported back to the late 1980s, and M. usually adjusts for it when all's said and done.

In Nov. 2004 there was really high turnout. Mitofsky goofed. I like to think that somehow his way of compensating for non-respondents amplified the problem--more non-respondents, a bigger problem. But he's not talking.

Some factors influencing non-response are random: rain should dissuade both dems and repubs from standing around and filling out exit polls. Inexperienced exit pollers would be another: they miss people, but how do they preferentially miss just repubs?

But, apparently, they do. Newbie poller or rain predisposes a precinct to having a larger error. Sigh. Maybe repubs are more easily pissed off by inexperienced pollsters?

But I can't figure out why when I go to a Slavic conference and I usually see a sea of lily white faces, either.

[Chi]

55. I case you didn't see this previously...

Heya igil
You might be interested in this...
http://www.duke.edu/~mms16/non_response2000.pdf

"Who Are the Non-Respondents?
A Deterministic Approach to Modeling Non-Response in Exit Polls"

Sorry if you already read it, or if your not interested. 8)

[Igel]

56. I've seen it posted, but given finite time and

infinte postings ...

I downloaded it, and will look at it tomorrow.

Tx.

(snuffle)

[davidgmills]

60. Freeman postulated

In his last report that there are many reasons to assume that Democrats would not be willing to talk to interviewers either.

Originally the argument went that Republicans hate the press and don't want to talk with someone that has press patches on their clothes. However, recent studies have shown that Democrats heave nearly as much disdain for the press as do Republicans (as I recall a 40% to 33% spread now, something in the 6-7% range) so that theory went awry.

Moreover, Freeman postulated that many Democrats, due to their poor education or social status might be inclined to resist talking to pollsters of better education or social class.

He also postulated that a significant number do not speak English and this may reduce their willingness to take a poll.

Further, after having waited hours to vote, many would not be willing to stay and take a poll.

Further, many do not have the luxury of flexible work hours and do not have the time to stay and take a poll.

Despite all this, they still "overpoll" Democrats.

[Mistwell]

29. Well...

...since it happened in every exit poll they every conducted since 1988 that exit way, and since independant studies have show that it always happens that way that Democrats talk to pollsters more than Republicans...why would you call something based on a heap of empirical data "speculative"?

http://www.duke.edu/~mms16/non_response2000.pdf

[anaxarchos]

6. Couldn't agree more...

Especially with this quote: "It is my opinion that there is a remarkably large amount in that report that could be of use to us."

... with this reservation: WPE is itself a "theory" to explain variance from vote count. I think what the arguement will be is that many different factors "skewed by precinct" combine together to "overstate the democratic vote". The good news is that this is no explanation at all. Any such skew will show up in the raw polling data (i.e. it will be detectable... as in "Hmmm.. this precinct seems to consist only of people under the age of 24"..., etc.).

I think we are back to demanding that ALL the data be released with the ambiguities in this report used as an additional lever.

[Igel]

8. I suspect the raw data

won't be a good sampling of precincts nationwide. I.e., it won't accurately reflect the overall electorate.

I also expect raw data to include frustrating numbers involving non-respondents: a 100 voter sample, 34 Kerry, 33 Bush, and 33 non-respondent; a second 120 voter sample, 50 Kerry, 23 Bush, 47 non-respondent. Try to show anything meaningful about fraud from that kind of data. Actually, try to show anything meaningful at all from that data.

Non-respondents, since day one, have been a thorn in the side of exit pollers.

[anaxarchos]

25. You are forgetting....

...that exit-polls carry their own controlling demographics around with them (i.e. the survey questions). I think the opposite is true. I think it is very easy to detect broad anomolies and precinct specific skews. Since Mitofsky doesn't mention them, he hasn't found them. Since there was "no fraud", and no problem with methodology, he is stuck with WPE which in the end he will argue had a "random" effect (what else is there?).

Show us the data....

[Igel]

36. No, I'm not forgetting that the demographics are there.

But you have only the most basic demographics for non-respondents, if the interviewer/observer was able to jot them down. If I had 20% non-respondents, I wouldn't have a clue what demographics to ascribe them apart from what was noted (and since I can't assume they're completely randomly self-selected, I can't ignore them; things like religion, how they voted, what issues they thought important ... all missing).

I'd love to know how Mitofsky resolves non-respondents. But then, I'm hardly alone in that.

And I think that he's saying the WPE is very much non-random. But that he doesn't know what it is that causes the non-randomness to be as it is. So he can just state conclusions in terms of the categories he sets up (rep/dem, whatever).

I agree with you. I'd love to see the data. But I still suspect I'd be highly disappointed. And I also suspect it would be very, very easy to misrepresent what, if anything, it says. Some folks already forget that the precincts weren't randomly selected, and that the early poll estimates already include a lot of weighting based on Mitofsky's assumptions; I suspect they'd be eager to assume that the raw data is reflects the nationwide voter pool, assuming that the raw data is favorable Kerry, and label the poll incredibly biased if it favors *.

[anaxarchos]

37. I'm at the opposite extreme...

...for exactly the same reasons. Yes, you don't know the demographics of the non-respondents but you DO know that of the respondents and you know the history so that gives you enough to derive that of the non-respondents. On Mitofsky, I see him painting himself into a corner because there is nowhere to go. If WPE is non-random, it ain't WPE. We end up with "random non-randomness".

The precincts are the equivilant of the old key precincts of the old party machines. They have to be representative... yet another control.

If the issues weren't so serious, I would get a bowl of popcorn and a folding chair and sit down to watch...

[Igel]

45. Hmmm ... I partially agree.

M. points out that the last time he had WPE like this was in 1992: both years had abnormal indicators (turnout, interest). You can assume the historical pattern holds only if the relevant conditions are the same (and even then, you allow for some drift ... few things are ever completely static). Exceptional turnout ... not the same as 2000. Is that violating a relevant condition? Gee, dunno. Looks like a definite maybe. It's testable, at least: given the data one could look for correlations between WPE and differences in voter turnout and demographics between 2000 and 2004 (and 1992/1988 or 1996). I wonder if that's maybe on M.'s calendar for March?

Mitofsky's WPE has to be non-random, it's what TIA lives and breathes for. It's not statistical error, that would be "precinct error". It's the things screwed up within a precinct that statistical methods didn't deal with, where he's saying "I screwed up ... or my model did ... or my employees did...." It's predictive stats vs. reality.

The precincts don't have to mirror the electorate. That would be ideal; but since "reasonably close" is the best M. could ever do, it's just sufficient to make sure that the relevant categories are adequately represented (differences in voter turnout by group probably make it impossible to achieve "really, really close). M. in any event has to have a nifty model (probably tweaked to within an inch of its life) that relates his sample to the general pool of voters.

I'm hoping that he eventually puts out a more comprehensive report. (Ack. What am I saying?! Argghhhh)

[anaxarchos]

51. Hah...

For two people who don't agree, we are awful close...;-)

[skids]

28. They controlled for that.

When they ran the "actual" vote numbers through as though the exit interviewers had returned those instead of their results, after applying their weighting formula for the precincts, there was a 0.43% bias towards Bush in the selection of which precincts to poll at.

[Chi]

27. Sorry to poke my head in here....

Heya anaxarchos
In past elections, has Mitofski (or others) made it a point to release their raw data?
(I ask cause it's relevant, you seemed very informed, and I don't know the answer)

Thnx in Advance

[anaxarchos]

33. No problemo...

They do release, but 5 months later if I remember correctly. Also, "all" their raw data is a controversial issue. They hold back "proprietary information", in theory to protect their intellectual property (i.e. some specifics of their methodology).

[Chi]

48. Thank you for the response.

I was wondering if your under the impression that the "proprietary info" included poll locations, sample quantity, or any part of the answers in the surveys?

Basically I'm just trying to see if the info that IS released is, in of itself, perfectly good info for another pollster to re-analyze and weigh.

Thanx

[anaxarchos]

53. The simple answer is yes.... n/t

[Chi]

54. Thank you, I appreciate your time. ...N/T

[KingoftheJungle]

9. Here is soemthing very interesting a friend sent me....

Watch the shift in between the exit poll graph charts, with the pre-weighted going first and after the weighted.

http://f3.grp.yahoofs.com/v1/sB7vQYUzbX-R8HeMfPcvaTZsw7xWU2XwyrWWQ8reN-khxLoheSWYMZsAvfk585Bnr1s5bZdEvCKk1eQTpyxty8419i6Gie7xCz_3eyY/epolls.0.htm



So those are all the pre-weighted polls before any shift. Notice the extra high numbers on the democrat side for certain religions, ie: jewish, christian/other/unitarian. There is an extraordinarily high number if you look at the first set of exit polls here

exitpollz.org

Look also at church attendance and how often they meet for their circles, and the democrat side is awfully high, on the side of republican for things like "Honest Trustworthy" which are not that high. But look at the republican side of protestant/weekly and it seems to be going even higher.

Now after seeing all of those look at the "Weighted" exit polls.

http://f2.grp.yahoofs.com/v1/sB7vQTBm_WeR8HeMiO3vj7in9ew3onNHkI9bawkFZp3m7q7xZQSVlpLjqyB-EHoS51a61ElShZNJcltAe_v2eYL_zbDlCUZejR8lwSI/Election%202004%20Exit%20Poll.htm

Everything on the exit polls for the democrat side shows a decrease in religion, a drop in the numbers for the religious affiliation and a drop in jewish, other and church attendance.

Now look carefully at the republican side. It shows a more than 1% shift to "Honest Trustworthy" and the other "Religous Faith" not to mention more than 2% swing towards "Strong Leader"

Now if that wasn't something, look at every single circle under the religious affiliation for the republican side. There is an increase of proportion on all sides and for church attendence. Nowhere has it actually decreased, in the whole format you see it increase by the same amount the democrat side decreased.

And even more strangely, on the final WEIGHTED exit poll numbers is an additional column under "Opinion For Bush" Enthusiastic was added, which apparently increased by more than 40% out of the blue.

The last part which really makes no sense, is NEP's final exit polls.



Here, an additional catergory has been added: Evangelical, and all the numbers on the republican side have increased once again.

[anaxarchos]

34. This is why it is not easy to rig exit polls to "fit"...

It's like a water balloon. You push in here, it pops out there.

Your example is interesting... I've been looking for (and finding) a lot of these.

Your links aren't working for me... (but I get the gist).

[jkd]

10. Mitofsky problem not methodology but execution.

Exit polls are no better than their weakest link. In this case it is the interviewer. This is very limited part-time employment. They are apparently hired on the telephone and receive a couple of short training sessions also by phone. They show up at the precinct on Election Day unsupervised. They are expected to follow precise requirements, but they’re on their own to comply. They could just as easily fudge the data. Until Mitofsky improves the training and supervision of his employees, he can’t expect to have accurate polls.

This year’s polls have received more scrutiny than former exit polls because of the CNN postings. The exit polls have always adjusted the data to fit the vote, but this year we all got to watch the process. Over forty years ago as a high school chemistry student, I learned what it meant to fudge the data. It seems that I could never get the right answer in the labs. So I learned to work the process backward. If one knew what the answer should be, all one needed to do was to supply the right data to get the right answer. I got pretty good at it. I didn’t take any more chemistry classes though. The dishonesty must have got to me.

The solution to accurate polls is close supervision throughout the whole process. Invest more money, adequately train the employees, poll many more precincts, and provide multiple interviewers for each precinct polled. Mitofksy’s answers leave many concerns to those who would take them seriously.

I believe that his margin of error is even higher than stated because of the obviously poor performance of his interviewers.

[Carolab]

13. So this is the only election for which interviewers were poorly trained?

Gimme a break.

[althecat]

14. In this election they were in fact three times as bad as the last two

... at least that is what his report says.

I think it just means the election was stolen from a wider margin to begin with.

[Carolab]

15. Three times as bad. For the "election of our lifetime"? WTF? n/t

[anaxarchos]

39. Yes...

Prior to this, Mitofsky only hired middle-aged Japanese Elvis impersonators with 4 years of training... Everyone likes Elvis ;-)

[democraticinsurgent]

16. As one of the exit poll interviewers...

I'm not sure I agree with your conclusion that the interviewers were the problem.

I felt that the training was plenty adequate and it went exactly as they outlined in the report.

After all, it wasn't a difficult job. Basically, all anyone had to do was count the voters coming out and try to interview every "nth" person based upon a given precinct's requirements.

There really wasn't much to it otherwise. Nor do I agree that there was anything to fudge. An interviewer's job was simply to execute the above and hope for reasonably good cooperation.

In fact, I'd say that the training received was comparable to that of many opinion poll interviewers. It's not rocket science.

I think that Edison/Mitofsky laying blame on young/student interviewers is largely disingenuous. After all, the report also shows that post-graduate educated interviewers had the largest WPE average. So correlating education/age to WPE results isn't the answer.

[jkd]

35. Not a matter of ability

I've managed young men and women in that 18-24yr age group which accounted for 35% of the interviewers. Many of them need supervision. It's not a matter of ability but of commitment. Maybe the highly educated assumed it was beneath them and didn't take it seriously either.

[democraticinsurgent]

42. sounds like a guess to me

and what would make this election any different from the past ones, as far as exit poll interview training is concerned?

[democraticinsurgent]

44. and i fit the "highly educated" category

and took it damn seriously. a couple of times, i missed voters because although i was close by the poll, it was easy for voters to zip out in a hurry before i could ask them.

my daughter was there and asked me, why don't you just interview the next person?

because it would throw off the stats, i said...

[jkd]

49. Thank you

Thank you for being so dedicated.

[anaxarchos]

38. Thank you... n/t

[americanwoman]

11. Interesting that Mitofsky says large WPE means vote tally is incorrect

On page 34, Mitofsky explains that 3 precincts were not included in the study because of their large WPEs: 112, -111, and -80. He says "indicating that the precincts or candidate vote were recorded incorrectly.

So he says right there, that large absolute WPE is indicator of vote count error. So where is the cutoff? At what point does he say, "it's not the exit poll methodology"? What is their basis for distinguishing between these two potential reasons for high WPE?

Other comments -- these from page 1 of the report:

How do "distance restrictions imposed upon our interviewers" favor one party over the other?

Also, how do "weather conditions which lowered completion rates at certain polling locations" favor one party over the other? What, it only rains on Bush-voters?

[jkd]

12. I don't believe it favors Kerry. It just corrupts the methodology

of a random sample. There are a lot more misses.

[Hugin]

17. Ah... The terrific analysis I've come to expect from DU.

Thanks ppls. =)

[truckin]

18. Is this information available from prior elections?

It would be interesting to compare the methodology and results of this years exit polls with the exit polls from prior elections. From what I have read, prior exit polls have been remarkably close to the actual results in presidential elections, so close that some have questioned how they could be that accurate.

[truckin]

19. What about the changes in the numbers on the CNN website?

Maybe I do not understand the exit polls or maybe this has been addressed before, but if Mitofsky is saying that the early exit poll numbers are the actual results but are inaccurate because of bad sampling, how can the changes on the CNN polls, that are mathmatically impossible, be explained?

[cyn2]

20. Let me get this straight....

as one who has not read the report, but skimmed the news articles, the Mitofsky report purports that the Bush voters were too ashamed to talk to the pollsters?

[Hugin]

21. Us young liberals are soooo intimidating. n/t

[Yupster]

22. No, I don't think embarrassed is too likely

I think rude, unfriendly and unhelpful are more likely.

After hearing Rush tell you to hate the media for a decade, I think they're more likely to just brush the pollsters aside.

[skids]

24. Things I am taking away from it.

First, the most important thing is the calculations where they took the formulas they were using to weight the precincts, and fed in the BOEs official results, to see what the bias was on precinct selection. They found a bias of 0.43% for Bush. This is a comparatively small bias, but on this level of the survey, if anything, it was biased towards Republicans. It means that all the final error, and then some, came from the WPE. So the results of the survey were an additional quarter-to-half percentage point "off" towards the Democrats in the aggregate.

Secondly, looking at the mean and median WPE charts on starting on page 36 is interesting. When the mean signifigantly different than the median, it means that there was a smaller outlier group of "problem precincts" that had extreme WPEs that pulled the average (mean) towards them. With that in mind, there is a lot to ponder in these charts. Especially that there was a very large number of "problem precincts" among those that voted heavily republican.

Thirdly, on the same pages is the completion rate/refusal rate/miss rate for the precincts. Especially interesting is the refusal rate chart that breaks down the refusal rate based in whether the precinct's final vote (I think, it isn't completely clear, that the official vote was used in this chart) was highly republican through highly democrat. Note that in highly Republican precincts, the completion rate was high and the refusal rate low. These precincts had a maximum of 20% Kerry support, and some likely had much lower. So it is relatively safe to say that refusal to participate in the exit poll is not an inherent trait of all republicans. The highest refusal rate and lowest completion rate was in contentious precincts where the vote was split.

But combining that with the WPE for the same split is even more interesting. The 50/50 precincts have a high mean WPE, so the theory of "selective non-response" might hold water there. However, on the extremes it looks like, for the "selective non-response" theory to work, it would have to mean that people in the minority in their precincts are extremely likely to respond, while people in the majority are not as likely to respond. While that possibility can't be discounted, it is counterintuitive.

I could spend a day on this set of charts alone.

The "paper ballots had low WPE" charts have other incredible implications. In trying to play down the touchscreen issue, they have provided us some valuable ammo. I'm sure others will go into depth on this chart. Again, look at the mean versus median and consider implications.

... and I have to stop about there as the stomach summons me to the kitchen. I can't wait to read the rest.

[FogerRox]

30. Not having read the report, but read this thread it seems the WPE

was used to isolate the fraudulent results, precinct by precinct. In other words--Lets correct the fraudulent precincts---then weight the data, as a whole. DO I have that right?

Mitofsky did this so accurately---Whether he knew of Fraud or not.
Of course he must.
I'd also be interested in info on his efforts in the Ukraine. DId he have US Employees--$$$$$ directorate of Operations money ? We know the CIA dumped some bucks there.

[skids]

31. You raise an interesting question.

Mitofski did identify a few extreme outliers.

Now, he probably has no legal obligation to report these to authorities, because he can't definitively say that the discrepancy was a result of fraud versus, for example, an interviewer that faked their results.

But isn't there an obligation on some level for his company to investigate this? Was that done? What were the results?

[FogerRox]

32. unfortunatly Mitofsky seems to behave in a complicit manner.

[Igel]

46. But remember what the paper ballot voting type

is. It's not just "paper trail", optical scanner or punch card. It's fill-out-the-ballot-by-hand-and-count-it-by-hand.

I didn't think any place in the US did things that way any more. I wonder if google would provide helpful info on exactly where those places are. (Maybe small towns in Nebraska, pop. 35?)

[smartone]

41. Important question

the premise is that Bush supporters did not respond when asked to participate in exit polling -

my question is did the same deviations take place in the Senate races?
especially in swing states arizona colorado, penn etc..

[southwood]

47. The "Mitofsky report" is among the best news ...

... for the fair election movement I've seen in weeks. It offers many clues to work from, and I'm sure the various critical statisticians will have great fun doing so. I think I read yesterday that the data will be available in two weeks, so hopefully we won't have to wait too long.

As for WPE, the main question is not so much what its determinants are, but why it was higher in 2004 than in the previous four presidential elections, and especially why - unlike in the previous elections - WPE in 2004 is not random as it was before, but for the first time skewed toward one candidate.

In other words: WPE itself is not the phenomenon to be explained, but the difference between WPE 2004 and the previous ones.
The report presents this issue in a frank way, but cannot offer any explanation and is quite honest about that too.

If and when full precinct data become available, as they should, any calculation (e.g. of means and medians) and investigation (of hot spots where exit polls differ widely from election outcomes) is possible.

[jkd]

50. I completely agree

[Wilms]

57. kick n/t