saw this graphic on the GD forum

[garybeck]

We've all seen versions of this but I don't think I've seen this exact one before. Like they say, a pictures is worth a thousand words. In this case it's a graph...

[Stevepol]

1. Worth recommending, lest we forget!! n/t

[cry baby]

2. A picture is worth a thousand words, indeed!

Who will Diebold elect in 2006 and 2008?

[kster]

3. We should try

and buy the next election. What do you think.

[bleever]

4. I hadn't seen that one before. It's really clear and striking.

:thumbsup:

[Wilms]

5. That makes five. Recommended. eom

[katinmn]

6. Says a lot.

[garybeck]

7. doesn't it look like dripping blood? very eerie. n/t

[liberalla]

8. Looks like STABBING blood...

reminds me of knife blades!

..eerie blood and knives aside, it's an excellent graphic!

[FogerRox]

21. yes dripping blood-- I saw this from a while ago--

quite compelling

[Febble]

9. Yes (sigh)

It tells you that the exit polls were significantly different from the count.

It doesn't tell you which was wrong. It just tells you it wasn't random error.

It could have been bias in the poll. It could have been bias in the count. Figuring out which it was is not, contrary to popular belief, easy.

Both are perfectly possible. Random error however, is not.

That is what the plot tells you.

[BeFree]

10. Easy for me

Clearly, Mitofsky, et al, could not be so much in error. As a defender of Mitofsky, you, Febble, should be able to defend that?

Looking over the graph one sees that the states with the greatest difference are states with DRE's. Using a mix of common sense and historical perspective one can easily ascertain that exit polls WERE a good barometer before the biased DRE's became so widespread, therefore the DRE's are easily the culprit explaining this graph.

Really quite easy to see, and becoming ever more accepted to those willing to examine the facts.

[Febble]

12. No, DREs

do not correlate particularly with error - I am not sure where that idea has come from. Can you point out what you mean?

And there is no particular a priori reason to think the poll was not in error. We know it was not random error because that is what the statistics tell us, but there are many other sources of polling error besides "sampling error" and some of them can be systematic, i.e. cause bias in the poll. It is a major methodological problem with any poll.

As I said, it could have been the count, it could have been the poll. But the statistics do not tell you which. They just tell you it wasn't random.

You are right that the way to approach sorting out the answer is to look at correlations between things like machine type and error. Also between things likely to affect non-random sampling and error.

Perhaps most important is to see whether greater error was found in precincts with the biggest swing to Bush. Unfortunately when this was tested in Ohio, it came up negative.

[BeFree]

18. Actually...not DRE's

You should have noted that. Now I am not about to do your homework for you, so don't ask.

It is well known that nearly 80% of the vote is recorded and/or tabulated electronically. Now look at that graph again; nearly 100% of the above 2% moe are well represented as 80% of the vote, eh?

In your homework session, take a look at the states that used paper. See where those lay on that graph?

There is no question that electronic voting is the number one altering condition that relates to the exit polls being beyond the moe. Also, in order to pad the national vote totals, many non swing states were quite used and abused. It really is common knowledge, your homework should be quite easy.

[Febble]

26. Well, I've done my homework

I'll be interested to see yours.

But here are some things to bear in mind:

Paper ballots were rare, and in the exit polls, paper ballots were used in very few precincts, almost all in rural areas.

The within-precinct-error rate was calculated by comparing poll to precinct count - not tabulator count - in the majority of precincts (around 60%). So the WPE - which shows an even larger red shift than the graph presents - must substantially represent precinct-count error, if it represents count error, not tabulator error.

So machine matters - yet we can't really analyse that at state level as most states used more than one machine type.

(edit for typo)

[Time for change]

33. But look at hand counted paper ballots

In Mitofsky's report, hand counted paper ballots stand out against any other method of counting votes, as by far having the least red shift (median only -0.9).

Why is this important? Well, I'm not an expert on this voting technologies, but isn't it true that any technology except paper ballots uses a computer to count the votes? And then, isn't this number fed into a central tabulator to tally up all the precinct votes and come up with a grand total for the county? And aren't these central tabulators imminently hackable? So, perhaps central tabulators were the main route to election fraud in 2004.

I know at least one thing that you'll say to this, because you've said it before: the number of precincts using paper ballots is too small to have statistical significants in this sample, and there might be confounding factors (such as urban/rural location) that account for the low red shift with paper ballots.

But still, this data is suggestive that hand counted paper ballots result in the smallest red shift (which of course suggests fraud, rather than exit poll bias), and just because there is a small sample size, that doesn't mean that the relationship wouldn't hold up with a larger sample size.

[Febble]

36. I think hand counted paper ballots are great

and it would not surprise me if the result stood up if the method was used in urban areas. My guess, for a start, is that spoilage rates are lower, and I would expect differential spoilage contribute to at least part of the red-shift. But by the same token, we know that for a given technology, spoilage tends to be greater in large urban areas, and there were no paper ballots in urban areas. That's why the demographic issue is important.

[Name removed]

37. Deleted message

Message removed by moderator. Click here to review the message board rules.

[OnTheOtherHand]

39. is this some weird trick question?

I'm staring at the map at http://verifiedvoting.org/verifier and I don't see a single state that is strictly paper ballots, hand counted. Not one.

However, poking around Vermont, it looks like a mix of paper ballots, hand counted and precinct-based op-scan (which of course leaves a paper trail). Among the biggest exit poll discrepancies in the country (biggest by some measures).

Looks like the same story for New Hampshire, which also had famously huge exit poll discrepancies.

So, umm, what facts do you have in mind?

[BeFree]

41. Ahh yes, it is a trick...

The precinct data is where the truth lays. Had you done more than look at some map and poked around in two of the smallest states you'll find waht I said was true.

In states (Oregon) where paper was the main use of counting the votes the exit polls were well within the margin of error. The states with electronic voting were those above the MOE.

And paper trail is of no use. Paper trail is nothing. Electronic voting vs. Paper Ballots, is all you need to know. Trails are nothing.

[OnTheOtherHand]

46. since Oregon has vote-by-mail, it's not a great choice

to demonstrate the accuracy of exit polls, is it?

"In states (Oregon)" -- it's a great state, but is it plural? Is there any state other than Oregon that you regard as an example of paper being the "main use of counting the votes"?

"Electronic voting vs. Paper Ballots, is all you need to know."

No, with all due respect, it isn't. But thanks for sharing.

[BeFree]

48. It's electronic or paper Ballots

There is no other choice. You either vote by hand and eye, and you count by hand and eye, or it is electronic. It's real simple. It's all you need to know.

Electronics ruin elections. We have proof. Glad to help you, even tho I know you weren't serious.

[OnTheOtherHand]

53. well, no, it's not that simple

Many serious activists favor paper ballots that are counted by machines, some randomly chosen portion of which is recounted by hand and eye to verify the machines.

Many serious activists favor paper ballots, hand-counted.

You can offer arguments for one view and against the other, or you can try to win the debate through semantic fiat. If you think it is accurate to say that Andy S. endorsed "electronic" voting, umm, good luck with that.

[Febble]

51. There were no exit polls

in Oregon as there were no exits.

The poll was done by telephone. It may make a difference.

[rman]

11. Different - significantly *in favor of Bush*

That is what the plot tells you.

(Also voting machine distribution was significantly in favor of Bush.)

[Febble]

13. Yes, if you apply a one-tailed test

it's even more significant.

But in four out of the five previous presidential elections it's been significant as well, in the same direction.

[garybeck]

14. "bias in the poll"???

what exactly do you mean by bias in the poll? and if that is the case, why is so that key swing states were more likely to be "bias in the poll?"

there is no plausible explanation for the exit polls being wrong. the "reluctant bush responder" hypothesis is ludicrous and has never been explained scientifically.

on the other hand, there is absolutely a plausible explanation for the count being wrong: manipulation of the data. the means, motive, and opportunity are all there.

there is no means, motive, opportunity, no plausible explanation for the exit polls being off.

[OnTheOtherHand]

15. does that graph really say "key swing states" to you??

New York, North Carolina, South Carolina, Delaware, Vermont, and Alaska?

On that graph, New Hampshire is the only swing state that stands out from the others -- and Kerry won it.

"there is no plausible explanation for the exit polls being wrong. the 'reluctant bush responder' hypothesis is ludicrous and has never been explained scientifically."

Sir, I don't know where to begin. I know almost no one in public opinion analysis who agrees with you. What is the basis of your confidence?

[Time for change]

31. On the other hand, regardless of how one interprets this chart

Mitofsky himself says on page 42 of his report that the red shift was greater in swing states than in non-swing states (median = -8.1 for swing states, -5.6 for other states) ;-)

[Febble]

34. Yes, the precinct level story

is much more complex.

These are mean WPE values, and one big problem with WPE values is that they are mathematically inflated in evenly balanced precincts, relative to more extreme precincts. One of the points of my paper was to point out that comparing WPEs between evenly balanced precincts and extreme precincts would be problematic. To the extent that evenly balanced precincts tend to be found in evenly balanced states, this would affect the comparison cited on page 42 of the report. It was one of the reasons Mitofsky became interested in my measure proposed measure.

But it is nonetheless a good point. I think what kind of frustrates me is that there are good arguments to be made for fraud, and poor arguments, and it is important (isn't it?) to distinguish between them if any argument is to be taken seriously. To cite page 42 of the E-M report in support of the case that there was greater bias in the swing states is a good argument (although it has to be tempered by the WPE problem). But to cite a plot that indicates that the state level red-shift was greatest in the bluest states as evidence that the red-shift was greatest in the swing states is a pretty bad argument. As is saying that the red-shift was greatest in DRE states (which ones?) or saying that it was greatest where tabulator fraud was likely (again, which states? And why the WPE when the WPE was largely computed from precinct counts?)

If the story is to be taken seriously by anyone, it has to be coherent.

Which is, believe it or not, what I'm trying to make it. It just might turn out that the coherent story is not the quite the story we originally thought it was.

[OnTheOtherHand]

35. I think you are winking

because I pointed that out to you in an e-mail last week! But I'm glad you bring it up here, because I never got around to asking the obvious question: why don't the precinct results seem to jibe with the state results?

I don't have an answer -- or rather, I can explain technically why it is not unlikely, but I have no particular substantive explanation of why it turned out as it did -- but it's the fact. Using the best geo estimators for E-M's 11 swing states, the average swing-state error is -5.9% on the margin, compared with -4.7% for the non-swing states; the median is -5.0% for swing states, compared with -4.2% for all states. Far from statistically robust, and much smaller differences especially for median.

I think part of the discrepancy here is that the WPEs are calculated on raw results, while the state Best Geo Estimators incorporate geographic weights and -- I think more importantly in this case -- weights to compensate for differential non-response. That may be more important in the swing states where the non-response rates were higher overall, although what you would really want to look at is the _differences_ in response rates across demographic categories.

As you know, the three biggest swing-state Best Geo errors were in states that Kerry won (NH, PA, and MN), and looking at the pre-election polls -- not to mention the recount in NH -- I'm just not convinced that there was massive vote-switching in any of these states. This is why I think it probably makes sense to give up on the "swing state effect" and focus on particular states where fraud seems likely -- OH, FL, and NM seem to be high on the list. (I'm certainly prepared to hear cases for the ones I mentioned above.)

[Time for change]

43. Wait a minute here

Why do you think that it is important that the Best Geo errors were in states that Kerry won, in citing that as an argument against fraud? PA was critical to a Kerry victory, the expectation was that he would win it by a comfortable margin, the exit poll showed him winning it by a comfortable margin, and yet it was very close.

Perhaps we don't think of PA so much as a state where fraud seems likely because it hasn't gotten much attention, precisely because Kerry won it. Take a look at this report from EIRS:

Reported via e-mail: I thought you should know about my experiences on Nov. 2. I was a field organizer for the Kerry/Edwards Campaign in Mercer County, Pennsylvania, and I witnessed severe problems with the voting system there. The following is a brief summary: 1. Thirteen precincts encountered complete or partial failures of electronic voting machines. Many of these machines were deemed "unrepairable" be technicians, and were out of service all day long. All of them were in Democratic-majority precincts. Many of them were in African-American precincts that have historically voted up to 90% in favor of democrats. 2. Of these precincts, many of them lacked paper-ballot backups, and those that had paper ballots were unprepared to handle them, often issuing improper or illegal instructions. 3. Countless numbers of people were told to go home once, then returned later in the day, hoping that the system would have been fixed, and were told to go home again. Some even went to vote three times and were told to go home each time. There is no way to accurately estimate how many people simply never got to vote, but it is at least in the hundreds, if not thousands. 4. Out of 51,800 voters who signed in to vote on the electronic machines, only 47,700 actual votes for president were registered. In one highly-democratic precinct, Farrell 1-2, 289 people signed in to vote, but only 48 of them actually succeeded in registering votes for president (45 for Kerry, 3 for Bush). This points to a reasonable estimate that nearly 4,000 votes in the county were not recorded. Again, it appears that most of this disenfranchisement occurred in democratic-majority precincts. Since Pennsylvania was won by Kerry/Edwards, it has been difficult to get the proper degree of attention focused on this problem. However, the fact that these events did not directly affect the outcome of the election should not distract anyone from the fact that what happened in Mercer County was a complete disaster and a total disgrace. From their comments in the local papers, it appears that Mercer County's election officials fail to understand the gravity of 4000 lost votes and the disastrous affect that can have on peoples' faith in democracy. Such accidents cannot be swept under the rug. Anything short of a complete investigation into this matter should be unacceptable, and will be seen as unacceptable by the residents of Mercer County. Any pressure you can bring to bear to see that a full investigation goes forward will be greatly appreciated. Please call or write back. I have a stack of signed complaints that our office collected on Election Day. I am willing to help in any way to expose all the facts regarding this important matter.

And with regard to the weightings, I can't put much stock in them, given that so little effort has gone into explaining how they were done. Look at the unweighted exit poll results for Mitofsky's 11 swing states, versus the other 39 states, all combined together, and see what a huge difference you get. How do you explain that?

[OnTheOtherHand]

44. umm, quickly

1) Who expected that Kerry would win PA comfortably? I didn't. If I should have, I will reassess. Looks to me like the last eight PA polls averaged Kerry +2.

2) Massive fraud in PA is perfectly possible, but I still need someone to give me massive fraud in NY and VT (to name two) before I start taking the _exit polls_ seriously as evidence.

3) I'm not sure what to say -- my previous post indicated a good candidate for the differences between the WPEs and weighted state results. Are you suggesting that the Best Geos were cooked to suppress the swing state effect? I can't rule that out, but it isn't at the top of my do list to investigate.

[Time for change]

49. Hmmm

1) I and everyone I was with in Harrisburg on Nov. 2nd expected Kerry to win PA comfortably. The ground game seemed to be running exeptionally well. They were doing so well in getting people to polls that they couldn't even find any work for us volunteers to do. A great big group of us sat around waiting for a couple of hours for them to find us some work, and finally our leader came in and said "Sorry, everyone on our list has voted, so there's no more work to do".

2) NY and VT are far eastern states -- so my guess is that those far eastern states had some phenomenon that seemed to contribute to exit poll bias (VT, NH, MA, CT, RI, NY, NJ, DE).

3)I'm not suggesting that the Best Geos were cooked. Not at all. I'm just saying that they weren't explained in Mitofsky's report, and I rarely trust anything that isn't explained to me. If you explained it in your previous post I didn't understand your explanation. I'm not even convinced that the raw results needed to be weighted.

[OnTheOtherHand]

52. OK

1) I think the Dem ground game in PA did run exceptionally well -- but unless we're going to assume that the Reps don't really have a ground game (and frankly, it's hard for a lifelong Dem like me to imagine it), I don't think we can infer much from the mood among Dem activists in Harrisburg.

2) I think it's pretty plausible that there is some "far eastern phenomenon" that we can set alongside the "swing state effect." But we don't have any strong basis for arguing that this mysterious effect extends to NY but not to PA or OH.

3) There isn't a full explanation of the Best Geos, but I'm pretty sure they incorporate the demographic weights for non-response, and I'm pretty sure that the WPE figures don't. So that's a consideration.

Just to sum up, my big problems with the swing state effect are that

-- it seems sensitive to alternative measures,

-- it is undertheorized (are we positing that votes were stolen in all the swing states and nowhere else, so that the rest of the error is response bias? are we positing that maybe votes were stolen everywhere, but extra votes were stolen in the swing states, although in most cases not enough votes?), and

-- it is susceptible to alternative explanations (E/M apparently take it as weak support for their idea that response bias correlates with interest, which should have been higher in the swing states since that's where the campaigns actually operated).

That's my analytical objection to the swing state effect. It has nothing to do with whether there was actually vote theft in any particular state. And I think it's worth saying again, by the way, that if people stole or willfully suppressed votes in PA or anywhere else, they should go to prison for a very long time whether or not it affected the outcome in that state.

[Time for change]

54. Yes, the swing state fraud hypothesis is not fully theorized,

and there are alternative explanations.

I happen to think that a fraud explanation is a better one than greater interest in the election accounting for poll bias, because it makes more intuitive sense IMO. I mean, who could NOT believe that if the Republicans were going to commit fraud that they would target states like Ohio and Florida. But more bias due to greater interest in the election -- that seems a bit far out to me. Possible yes, but certainly not as plausible as the need to target fraud in Ohio and Florida.

As for not being fully theorized, well, certainly I wouldn't expect there to be a 100% correlation between fraud and swing states. In some states (perhaps IA and WI) the means was not available, partially due to the presence of Dem. Governors (In fact, it would be interesting to see what happens when you throw the party affiliation of the governor into the mix -- but I wouldn't count Richardson as a Dem. because I don't trust him). And as for the red shift in non-swing states, I suppose that's a combination of the far north eastern bias effect (and NY is farther north and east than OH or PA) and a need by the Republicans to pad the popular vote as well.

[OnTheOtherHand]

55. all right...

Once we take the assumption that the Republicans would "target" swing states, and then tack on the (also reasonable) auxiliary assumption that they would want to pad the popular vote in non-swing states as well, I don't see that there is much hypothesis left to test. Frankly, if someone somehow came up with an outright rebuttal of the swing state effect -- say, proved that it was actually explained by variations in interviewers' fingernail length -- I don't see how it would much affect the overall argument.

I haven't spent a lot of time thinking about the apparent relationship between campaign interest and exit poll bias, which is also undertheorized. I concede to E/M that it is, if nothing else, an interesting coincidence that the 1992 election had high turnout and big exit poll error. But it can't be the whole story. For one thing, we have at least three different exit polls for 1988, and -- although I can't tell what the best projections from each one would have been -- it seems likely that they varied in accuracy beyond the extent of sampling error. And I have no idea why. Someone may have written the definitive article on that very topic, but I haven't seen it.

So, that's where I am on that.

[Time for change]

72. I think that this is very complicated

I don't think that just because you add to the "swing state fraud" theory the belief that the Republicans wanted to pad the popular vote, that destroys swing state fraud as a theory. It simply makes it harder to detect, since it is less pure. But is sounds very plausible to me. As a matter of fact, I have heard that the loss of the popular vote is one big reason why Kerry hasn't pursued legal avenues in Ohio more aggressively.

Anyhow, what I really would like you to explain is the issue of how Mitofsky weighted the data. You have mentioned that it has something to do with non-response, but I just don't get that. I wasn't even aware that the data shows that non-response was related to the outcome of the exit polls or the red shift. If not, then what sense does weighting the data with regard to non-response make? I think that there is something here that I am very much missing, or else Mitofsky explained it very poorly.

[OnTheOtherHand]

77. I'll try

On your first graf, I agree that popular vote padding doesn't destroy swing state fraud as a theory, and also that it makes it harder to detect.

I'm really no authority on how E-M weighted the data -- it hasn't been a central issue for me -- but I can say a bit more about why it may matter when comparing state projections to WPEs.

One indication that non-response has something to do with red shift (although it obviously isn't the main cause) is that the average precinct WPE (-6.5) is noticeably greater than the national exit poll error (-5.5), and the average state WPE (-6.0) is larger than the average state best geo error (-5.0).

The weightings are sketchily explained on p. 9 of the E/M report. As you can see, non-response is just one piece.

The rationale for non-response weighting is pretty straightforward: if you know, for instance, that blacks are refusing at a higher rate than non-blacks, then your survey will tend to _understate_ the Democratic vote unless you compensate. I don't know, however, why the net effect of all the weightings is (on average in 2004) somewhat to reduce the estimate of the Democratic vote.

[Time for change]

78. Thank you for the explanation OTOH

Just a couple of thoughts on this.

With regard to the weightings, since the weighted polls seem to decrease the extent to which the swing states stand out, that would seem to mean that the swing states tended to have relatively less women and blacks with non-responses (it's hard to understand how they figured age in -- if they didn't respond, how did the interviewer know their age?), compared to the other states. That sounds kind of weird. It's really hard to make sense of this without seeing more of the data.

And I still don't understans why the difference between the -6.5 and -5.5 means that non-response has something to do with red shift. Could you please explain that?

[OnTheOtherHand]

79. it doesn't, necessarily

We know that the survey weightings, in the aggregate, are intended to reduce bias; and we know that the survey weightings, in the aggregate, reduced the overall average red shift. So I probably should have said that's an indication that the non-response weights may have tended to reduce red shift (or create blue shift). But I have no way of knowing, because there is no way to measure the impact of each weighting. (We could try to recreate some of the weights; it just hasn't seemed that important to me.)

It could perfectly well be that (say) the geographic weights are the only ones that really have this effect.

So, as you say, we really don't know enough to assess why the weightings seem to affect the swing states differently.

(On age -- they weight in three broad categories, so it probably isn't hard to eyeball whether someone is 25 or 60 with decent accuracy.)

[Time for change]

81. Ok, I get it -- thank you for the explanation

I think it's too bad that Mitofsky didn't contain more information in his report. For example, it's difficult to understand why the weighted results would cause the swing states to stand out less than the unweighted results. If more information was contained in his report, this issue could be assessed. In other words, I believe that it's fair to say that the unweighted results could be a better or just as good as a mechanism for assessing the relative red shift in swing states (compared to other states) as the weighted results. But without more definitive information we just don't know.

[Febble]

16. Well, your question is good

except that I don't think it's quite the right question, as the polls were most "biased" in the Blue states, not the swing states. TIA had a nice graphic for that, but here is mine:



The measure of "bias" (and I mean that neutrally - bias in poll or count) is plotted as red-shift=positive. So the higher the dot the more the red-shift. On the horizontal axis is the state final margin. So you can see clearly that the biggest red shifts (this is WPE data, by the way) are at the blue end of the spectrum and the smallest at the red end. It actually fits with TIA's "Beast was in the East".

Which of course raises the question of why. But the answer is not necessarily fraud of course. It not easy to interpret a correlation at the best of times.

Now, regarding your description of "reluctant Bush responder" as "ludicrous" - I'd like to know why you think it is ludicrous. I know it has been frequently described as ludicrous, but I'd like to know why - seriously. What is called "non-response bias" is a very real, very well-documented problem in polls and even with very careful methodology it is not possible to guarantee that you haven't got it. You certainly can't rule it out with statistics - your statistics assume a random sample and the point about "non-response bias" is that it means that your sample isn't random. And as a phenomenon, it has certainly been explained scientifically. Google "non-response bias" for lots of sources.

But I agree that there was means, motive and opportunity for the count being wrong. It's just that there are many, many, ways for the poll to have been wrong as well. The tragedy is that without transparent secure voting systems, we cannot be sure which.

[garybeck]

19. RBR is ludicrous because

there is no logical reason for it. bush supporters are just as vocal as kerry supporters. in fact they are more vocal. Knowing the mindset of the Bush supporters, they would probably stand in line to tell exit pollers how much they hate Kerry. Heck, I had a button on my shirt once that simply had a "W" with a circle and line through it, and I got my ear chewed off by an employee, about how Kerry was untrustworthy because of the Swift Boat lies. Which leads us to another issue which is rarely mentioned in this discussion. You have to consider the past history and the nature of those involved. No one can deny that the republicans own the vote counting process. Now consider that we already know about them. Consider what they did in Ohio with Blackwell. Consider their long history of corruption, lies, and deciet. they have more than means and motive. they have a well established pattern of doing things like this.

here is another reason why the count being wrong is much more plausible:

there is simply NO MOTIVE for people to skew the exit polls. Why on earth would anyone want to make the exit polls wrong, which would make people question the results?

On the other hand there is a HUGE HUGE HUGE motive for the count to be manipulated. People would pay any price, even kill, for the presidency. There is no reason for anyone to mess with the exit polls, especially in the direction that they were off.

You asked "why" there would be exit poll discrepancies in the blue states. This is obvious to me. Just look at the national exit poll and the "official" count. They had to get the "mandate" and the popular vote somehow. The least noticeable way to do it would be to trim votes off blue states without affecting the outcome. Isn't that obvious?

[foo_bar]

22. people don't behave logically

there is no logical reason for it. bush supporters are just as vocal as kerry supporters. in fact they are more vocal. Knowing the mindset of the Bush supporters, they would probably stand in line to tell exit pollers how much they hate Kerry.

Democrats who crossed over from the last presidential election would also be "rBr", i.e., non-vocal Bush supporters, not vocal on account of 1) the appearance of flip-flopping in one election cycle, and 2) social conformity. It could certainly describe the error in VT/MA/NY/CT ("the beast in the east"), and corresponding blue shifts in Texas and the Dakotas, where voting for Kerry was stigmatized in the opposite direction.

But then there's New Hampshire, and its bloody spike. DUer IdaBriggs took part in the recount:

Honestly, I was hoping for the famous "Eureka!" moment. There wasn't one. For every vote the machines counted, there was a paper back-up, and the small number of votes that were added to the totals (a little over a hundred with 58,000+ votes counted) were not unexpected.

I wish I had something more exciting to report, but *if* there is fraud in New Hampshire, then its happening at a different level than "machine count." There are a couple of other things to investigate (poll tapes, etc.) that the Black Box Voting people will undoubtedly check into, but I don't expect them to find anything. (I could be wrong; its happened before.)

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=97506

Occam's Razor would tend to favor the explanation that doesn't have Kerry stealing Texas, and Bush stealing Massachusetts.

[garybeck]

24. Vermont....

if the theory is that there are "reluctant Bush responders" in Vermont, because they are democrats who voted for Bush but were afraid to admit it, I have to say...

HOGWASH!!!!

I live in Vermont, near the largest city. I know a lot of people. I have never heard of a single Democrat who voted for Bush. In fact the opposite is true. I know several non-voters and even a few republicans who came out of the woodwork and voted for Kerry.

[foo_bar]

27. if the red spikes don't convince you, try our huge exclamation points!

As an ex-Burlington resident, it's hard to imagine a vote siphoning conspiracy encompassing Dean and Bernie Sanders, in a state that's predominately hand-counted paper ballots.

Reluctance to discuss the subject with pollsters probably correlates with reluctance to discuss the subject with friends and acquaintances, but there's no way to poll reluctant people about their reluctance (without a captive population, which would be a reluctantly captive population). Heisenberg aside, "nobody I know voted for Nixon!" (an apocryphal quote in itself), anecdotal wisdom is about 50/50 with a ±50% margin of error.

[garybeck]

61. that is not correct

Vermont is not "predominantly hand counted paper ballots." That is not even close to the truth.

Furthermore, such a "conspiracy" would/could happen without Dean and/or Bernie knowing. They have nothing to do with it.

[foo_bar]

63. 184 municipalities (hand count paper ballots); 62 (optical scan)

http://moritzlaw.osu.edu/electionlaw/ebook/part4/equipment_machines06.html

[OnTheOtherHand]

66. well, in terms of voters...

-- you can download a more up-to-date spreadsheet at
http://vermont-elections.org/elections1/voters.html -- it looks like 70 municipalities had op-scan in 2004, but they tended to be the big ones. So in numerical terms, I bet gary is right.

(EDIT: Just barely, according to kster's link -- thanks, kster!

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=201&topic_id=5446&mesg_id=5448

Looks like 316K op-scan, 303K hand-counted. So if exit poll error = fraud, the fraud would have to be concentrated in places with just over half the votes.)

So, we're just one short step from massive vote fraud in Vermont.... I dunno. There are only two Vermont state polls in the entire electoral-vote.com database; Kerry up 10 in September, Kerry up 13 in mid-October. He won in the official returns by 20. The _raw_ exit poll result had him winning by over 40 -- the Best Geo was 36.6.

That looks like exit poll bias to me, no matter what technology was being used.

[foo_bar]

69. kster's link is "population served" not "votes"

But point taken.

[OnTheOtherHand]

70. I'm working on some vote numbers now -- they may be interesting n/t

[OnTheOtherHand]

73. OK, I have some numbers (and a correction)

The correction is that unfortunately the numbers from kster's link aren't good because they are at the county level, and "mixed" counties are just divided equally. That way understates the use of op-scan. Here are the actual results:

in Accuvote towns: 122,374 Kerry, 76,270 Bush, 4,515 other (203,159)
in hand-count towns: 61,693 Kerry, 44,910 Bush, 2,547 other (109,150)

By my reckoning that means that Kerry beat Bush by 22.7% in op-scan towns, and only 15.4% in Paper Ballot, Hand Count towns.

Ouch. Should I be surprised that no PBHC advocates have been up in arms about the circumstantial evidence of Kerry election fraud in Vermont?

I will need to check these numbers -- I was doing the work in Excel. (If you can do it in SQL, that would be better!) But I think they are pretty darn close.

(EDIT: Although I would love to make some people squirm, they generally don't bother to read this far down-thread anyway, so let me say the more or less obvious: I think what happened in the Vermont general is like what happened with Dean/Kerry in the NH primary, i.e., both the machine distribution and the vote distribution correlate with geography.)

[foo_bar]

76. does this about cover it?

[garybeck]

80. I'm sorry, that is not correct.

I have spoken directly to the Vermont Director of Elections, Kathy DeWolfe. I asked her specifically this question. Well over half of our ballots are counted by diebold scanners. I don't have the data here but I will post it later tonight for you. She sent me the complete list of all the precincts that are opscan. And she also told me, and I quote, that "over half of our ballots are counted by opscan."

I believe your information is outdated.

[foo_bar]

82. I think we're both correct

PBHC was used in 3/4 of Vermont, but represented less than 1/2 of ballots scanned (since the PBHC locales are sparsely populated).

Hand-counted towns (semicolon separated by alphabetical county):
Bridport, Cornwall, Ferrisburgh, Goshen, Hancock, Leicester, Lincoln, Monkton, New Haven, Orwell, Panton, Ripton, Salisbury, Shoreham, Starksboro, Waltham, Weybridge, Whiting; Danby, Landgrove, Mt. Tabor, Peru, Rupert, Sandgate, Sunderland, Woodford; Burke, Cabot, Danville, Hardwick, Peacham, Ryegate, Stannard, Sutton, Walden, Waterford; Bolton, Charlotte, Saint George; Bloomfield, Brighton, Brunswick, Canaan, Concord, East Haven, Granby, Guildhall, Kirby, Lemington, Lunenburg, Maidstone, Newark, Norton, Victory, Westmore; Bakersfield, Berkshire, Enosburgh, Fairfield, Fletcher, Franklin, Highgate, Richford, Sheldon; Alburg, IsleLaMotte, North Hero, South Hero; Belvidere, Eden, Elmore, Hyde Park, Johnson, Waterville, Wolcott, Woodbury, Worcester; Bradford, Braintree, Brookfield, Chelsea, Corinth, Fairlee, Granville, Groton, Newbury, Orange, Topsham, Vershire, Washington, West Fairlee, Williamstown; Albany, Brownington, Charleston, Coventry, Craftsbury, Glover, Holland, Irasburg, Jay, Lowell, Montgomery, Morgan, Newport Town, Sheffield, Troy, Westfield, Wheelock; Bridgewater, Clarendon, Hubbardton, Ira, Mendon, Middletwn Spg, Pawlet, Proctor, Rupert, Shrewsbury, Sudbury, Tinmouth, Wells, West Haven, West Rutland; Calais, Marshfield, Middlesex, Moretown, Plainfield, Roxbury, Waitsfield, Warren; Athens, Brookline, Dover, Dummerston, Grafton, Halifax, Jamaica, Londonderry, Marlboro, Newfane, Putney, Readsboro, Searsburg, Stamford, Stratton, Townshend, Vernon, Wardsboro, Weston, Whitingham, Wilmington, Windham, Winhall; Andover, Baltimore, Barnard, Bethel, Cavendish, Hartland, Mt. Holly, Pittsfield, Plymouth, Pomfret, Reading, Rochester, Royalton, Sharon, Stockbridge, Strafford, Thetford, Tunbridge, Weathersfield, West Windsor, Windsor

Optiscan towns:
Addison, Bristol, Middlebury, Vergennes; Arlington, Bennington, Dorset, Manchester, Pownal, Shaftsbury; Barnet, Lyndon, St. Johnsbury; Burlington, Colchester, Essex, Hinesburg, Jericho, Milton, Richmond, Shelburne, South Burlington, Underhill, Westford, Williston, Winooski; Fairfax, Georgia, St. Albans City, Swanton; Grand Isle; Cambridge, Morristown, Stowe; Randolph; Barton, Derby, Greensboro, Newport City; Brandon, Castleton, Chittenden, Fair Haven, Killington, Pittsford, Poultney, Rutland City, Rutland Town, Wallingford; Barre City, Barre Town, Berlin, Duxbury, East Montpelier, Fayston, Huntington, Montpelier, Northfield; Waterbury; Brattleboro, Guilford, Rockingham, Westminster; Chester, Hartford, Ludlow, Norwich, Springfield, Woodstock

(source: Kathy DeWolfe, Dir. Of Elections, 802-828-2304 http://vermont-elections.org/elections1/voters.html via VerifiedVoting.org)

As it turns out, the Optiscan towns gave Kerry an extra 7% (see posts #66-76). Any thoughts?

[garybeck]

83. but your original statement was

"in a state that's predominately hand-counted paper ballots."

I think it's pretty clear that such a statement would apply to the number of ballots, and not the number of towns.

It should also be noted that when you look at it by precinct, not by town, I think you will find that there are a majority of precincts counted by opscan. I don't think the number of towns that use hand counts has any bearing, and your original statement is not correct.

Regarding the opscans favoring Kerry, my guess is that it's pretty simple:

the opscans tend to be in more populated areas, which have proven to vote Dem/Kerry more. Virtually every significant city and town in VT has switched to opscan. It would follow logically then, that the opscans would favor Kerry. That is not surprising to me at all; I would expect it.

That still has no bearing on how the opscans relate to the exit polls and the descrepancy with the official count. This could only be studied if we had data on where the exit polls were conducted, i.e., which precincts.

[foo_bar]

84. you say predominant, I say predominate

in a state that's predominately hand-counted paper ballots.

I think it's pretty clear that such a statement would apply to the number of ballots, and not the number of towns.

I said the "state" was predominately PBHC, not the "ballots".

I think you will find that there are a majority of precincts
counted by opscan

Don't towns make that decision? A majority seem to have chosen PBHC (edit for primary source: http://vermont-elections.org/elections1/votingsystems42705.xls from http://vermont-elections.org/elections1/2004_election_info.html):

All cities and towns in Vermont, except for South Burlington, are required by the terms of their charters to hold an annual town meeting, on Town Meeting Day (the first Tuesday in March). However, state law requires that all matters of consequence, including election of town officials and approval of town and district budgets, must be decided by Australian ballot. As a result, many towns actually hold their public meetings the Monday evening preceding Town Meeting Day, and reserve the official day itself for voting.

http://en.wikipedia.org/wiki/Town_meeting

[garybeck]

85. I don't know where you're going with this...

you said

"in a state that's predominately hand-counted paper ballots."

With all due respect, that is not a true statement. I don't see how you can stretch it in any way to make that a true. When you say that more towns are counted by hand, it is beside the point and irrelevant to the discussion.

[foo_bar]

86. I have a similar perception

It should also be noted that when you look at it by precinct, not by town, I think you will find that there are a majority of precincts counted by opscan.

This is also factually incorrect:
http://vermont-elections.org/elections1/2004_election_info.html

[garybeck]

87. what?

first of all, the link does not give any information on voting technology by precinct. if it does, I don't see it there.

second, please note that I said "I THINK you will find that a majority of precincts are counted by opscan."

I said I THINK because I am not sure and I also said that I have the data at home and I will post it here when I get home tonight.

Your statement, however, was categorical and completely false.

[foo_bar]

88. here's the primary source

http://vermont-elections.org/elections1/votingsystems42705.xls

Your statement, however, was categorical and completely false.

You mean 'categorically'?

I wrote:
"184 municipalities (hand count paper ballots); 62 (optical scan)"

You wrote:
"I'm sorry, that is not correct"

Did you mean to reply to a different post?

[OnTheOtherHand]

89. let's be fair

Back up two posts -- he first objected to the "predominately" thing.

He says that's wrong. I'm from New York, where downstaters don't get to say that the state is "predominantly" downstate no matter what percentage of the population lives there, so I call it a wash.

Maybe we could talk about Vermont?

gary, do you really think it's likely that Kerry won Vermont by 36, and the Accuvote precincts by 48, when the last poll (in mid-October) put him up 13?

[garybeck]

90. this is ridiculous...

Your categorical statement was:

"As an ex-Burlington resident, it's hard to imagine a vote siphoning conspiracy encompassing Dean and Bernie Sanders, in a state that's predominately hand-counted paper ballots."

that statement is not true. Vermont is NOT "predominately hand-counted paper ballots". That is the point I've been making from the beginning. There is no way to look at the numbers and think this is true. I even looked up the word "predominately" and it means "greatest in number."

If you want to say the majority of towns or municipalites are hand counted, that is fine, but that is not what you said. You said the STATE is predominately hand counted, and that is not true. The fact about the municipalities is irrelevant to the vote siphoning issue which originally promted this discussion. Everything else is besides the point. Can we stop wasting time on this now, or are you still trying to say that Vermont is "predominately hand counted paper ballots?"

[foo_bar]

91. if I'd written "election" instead of "state", I'd agree 100%

Since you disputed the data ("I believe your information is outdated", "I think you will find that there are a majority of precincts counted by opscan"), I presumed your argument was with my characterization of the local governments that comprise the state.

Now that you've backpedaled on posts #80 and #83, I'll concede that "a state that's predominately hand-counted paper ballots" was a vague assertion.

Edited for a definition of "predominant":

Having the ascendency over others; superior in strength, influence, or authority; prevailing (Editor's note: as in 3/4 of sovereign town councils); as, a predominant color; predominant excellence.

http://www.dict.org/bin/Dict

[garybeck]

92. Oh please

I have not backpedalled on anything.

when I said 'your information was outdated' it was because you were saying that vermont is predominantly hand counted. it used to be but it is not now.

that is the only point I've been trying to make. vermont is not predominantly hand counted, unless you are simply counting the towns, which has no relevance to anything. certainly does not have any relevance to whether or not a primary election can be rigged, because our primaries are based on the popular vote of the entire state. therefore any discussion of towns and counties is purely irrelevant and I don't see why you would bring it up as an argument for why the primiary couldn't be rigged.

furthermore, even by your dictionary's definition of predominant, I don't see how by any stretch you can say that Vermont is predominantly hand counted. the definition says "superior in strength, influence, or authority;" If you take this definition to say that vermont is predominantly hand counted then you must be saying that every town has equal "strength, influence, or authority" and your assertion is based simply on the number of towns that use the various types of voting systems. This is a completely false premise. The towns do not have equal influence. the towns with more people have more influence on the results of th elections, because each person in our state gets one vote. The only way to look at this is by how many people in the state vote on the various technologies. What town they live in has no affect on the outcome of the election, or the influence of their vote. Each person gets one vote and that is it. there was absolutely no use in bringing up any discussion of the towns, or to say that because more towns use hand counts that hand counted ballots are more predominant in our state. In fact, by posting your definition of the word, you have proven yourself wrong.

I would like to end this discussion and stop wasting both of our time on what appears to be semantics. I don't know your history but I think we're on the same side, aren't we? We both want open and fair elections, right? So let's chalk it up as a misunderstanding. If you want, I will still find the email I got from Kathy where she told me that the majority of our votes are counted on scanners, and she also gave me the list of precincts. I still think you will find that because the larger towns have more precincts that the majority of precicnts are opscan counted, but I don't know for sure. For example Burlington alone has something like 10 precincts and they are all opscan.

but regardless, let's not squabble when we have important work to do.

peace

[foo_bar]

93. look up 'connotation'

It's something you acquire from context, not the first synonym on google.

but regardless, let's not squabble when we have important work to do.

Fair enough. So what's your argument for fraud in VT again?

[garybeck]

94. I don't really have an "argument for fraud in VT", but...

Actually it was someone else who asked the question about it, and you replied that it wasn't likely because we have mostly hand counted paper ballots. I only chimed in after that.

having said that, here is the situation here in Vermont, for those who want to know.

We have 100% paper ballots. most of them are counted by diebold accuvote scanners and central tabulators. the very same scanners that were investigated in Florida and found to have executable files on them. all diebold stuff. Zero audits. Not good.

We also have extensive exit poll deviation, far outside the MOE. We have a relatively small sample size, but that means larger MOE, and we're still well outside of it.

I've had extensive conversations with our (democratic) director of elections and she seems to have no problem with diebold and does not think that random audits are necessary.

I am going to start a fight for random audits here. I could use some help.

[foo_bar]

95. the pic in the OP just broke (for me at least)

But what made it eye-catching was the stalactite hanging over Vermont. Since the "visual" argument for exit poll correctness rests on the precipitousness of the troughs, Vermont seems like a natural test case for the broader argument (as New Hampshire was before the recount).

I am going to start a fight for random audits here. I could use some help.

Say the word.

[foo_bar]

96. the picture's back

Slightly off-topic:

On Wednesday, network company Level 3 Communications cut off its direct "peering" connections to another big network company called Cogent Communications. That technical action means that some customers on each company's network now will find it impossible, or slower, to get to Web sites on the other company's network.

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=102&topic_id=1830332

[OnTheOtherHand]

28. a problem: you are relying on critics of a hypothesis to explain it

-- I think -- but if that is an overreading, I withdraw it. At the risk of just echoing Febble, I will try again to explain the general approach of survey researchers who address this issue.

I've now looked at raw data from eight or nine exit polls going back to 1976, and every one of them overrepresents the Democratic candidate, often dramatically (Carter "beats" Reagan).

This 'effect' can be attributed to any combination of

(1) precinct sampling (for all I know, the 1980 survey deliberately oversampled heavily Democratic precincts);

(2) demographically correlated response bias that the pollsters can try to compensate for (e.g., did blacks respond at higher rates? the interviewers try to keep track of observable demographics of non-respondents);

(3) response or participation bias that doesn't track those observable demographics, so the only way to adjust for it is to reweight based on official returns;

(4) people misreporting their votes; and

(5) election fraud. (I could make that more complicated, but I think it's already bad enough.)

I'm betting on a big role for (3) in 1980, but I can't tell you how big.

Now, the part about "democrats who voted for Bush (or Reagan) but were afraid to admit it" is total speculation, and it's not part of the basic hypothesis. The basic hypothesis is that Bush voters and Kerry voters participated at different rates, which could happen for various reasons having to do with the voters, the interviewers, and their interactions. That basic idea is part of Survey Research 101.

Professional inquiry here starts from the observation that the exit polls have diverged from the official returns over years and years, and there is considerable evidence that (1) and (2) don't entirely account for the differences. (5) is possible in all years, but the typical line on ERD has been that the exit polls were phenomenally accurate until 2004 and that electronic voting machines made the difference. That line is somewhere between a stretcher and a whopper.

Thus, (3) and conceivably (4) are very much in play.

One basic difference in perspective between many ERDers and most (I would say almost all) survey researchers is this:

* Many ERDers seem to think that exit polls are generally accurate, and that the idea of response bias was a desperate invention.

* Survey researchers know that exit polls have often been inaccurate, and response bias is one of the first things they learn about as a potential pitfall in every study. This is why Mitofsky's "ludicrous" hypothesis is widely accepted, and certainly not ridiculed, by survey researchers.

Whether or not fraud -- even massive fraud -- occurred in 2004, asserting that Mitofsky's explanation is "ludicrous" immediately signals survey researchers not to take you (generic "you") seriously. It's a little hard to win the battle of ideas when a relevant group of experts is laughing at you.

I don't know what would be an accurate and neutral term for activists who refuse to pay attention to relevant expertise. By any name, it seems like a bad idea to me.

[Febble]

29. It is therefore possible

that the exit pollsters didn't hear of a single democrat who voted for Bush either.

That may have been why the poll underestimated Bush's vote.

[OnTheOtherHand]

30. actually, they heard of eight

(i.e., eight people in Vermont who reported that they usually thought of themselves as Democrats, and that they had voted for Bush) -- out of 242 Democrats, and a total of 674 respondents who answered both questions.

I betcha 8 out of 242 is low, and in this particular context it is tempting to conclude that Bush voters felt a bit cowed (in the "raw" exit poll results they are outnumbered by over 40 percentage points, about 69-27; in the official returns they lost by about 20 points). But then, in Utah, Kerry voters were outnumbered almost 3:1 in the official returns (72-26), and Kerry did better in the exit poll there, too (but not nearly as much better -- the average WPE is -6.2 in Utah, -16.5 in Vermont). Also, the exit poll discrepancies are pretty similar in Vermont (anything but a swing state) and New Hampshire (very much so).

I sort of envy the people who think they know exactly what all this means....

[Febble]

25. Few things

1. I can think of some "logical" reasons, but science is about observing data - and past data indicates that exit polls over-estimate the conservative vote, both in the US, and, interestingly in the UK. It may not seem logical, but it appears to be a real phenomenon. But no surveys are immune from sampling bias, unfortunately, and certainly not exit polls.

2. I don't think anyone (well maybe one) is suggesting the polls were skewed on purpose. Non-random does not mean intentional. It just means non-random.

3. Yes, it is possible that the bias was in the count. All I'm saying is that there is no reason to suppose it COULDN'T have been in the poll. All we can say for sure is that there was bias

4. Yes I did ask why there was more bias in the blue states. Your answer is plausible. But the reason I raised it is that you asked why there was more bias in the swing states - and there wasn't! We need to decide what the data is here, before we attempt to account for it.

But starting from the assumption that a hypothesis is "ludicrous" when it is actually supported by quite a lot of past evidence seems to me to be stacking the deck. It is certainly not considered "ludicrous" by professional pollsters whose job it is to try and deal with non-response bias on a daily basis. Interestingly, it looks as though the San Diego parellel election ran into a comparable problem.

Look, I'm not from the dark side - I think your interesting NM story may be drawing from some of my own work. I just think we need to distinguish between facts, inferences and assumptions, and most of all to check the facts - the swing state meme, for example.

[BeFree]

38. Hogwash!

1... past data indicates that exit polls over-estimate the conservative vote

Unsupported and unsourced. Prove what you say.

2. I don't think anyone .. is suggesting the polls were skewed on purpose.

Hogwash. There are millions who say the polls were skewed. The last 600 skewed it.

3. .... no reason to suppose it COULDN'T have been in the poll.

See 2 above. Mitofsky skewed it to match the machine vote. Anybody with any sense knows the machines were skewed.

4.... why there was more bias in the swing states - and there wasn't!

Hogwash. Every thing was biased to bush in the swing states. Exception: the exit polls until the last additions.

.... most of all to check the facts ...

Yes, show us some facts. All you have shown is words. Blah. Now TIA, he has facts. That's my take. You are all blah, and no facts with sources.

[OnTheOtherHand]

40. ooh! ooh! can I start #1?

OK, TIA has facts, so try these:

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x341940

Notice the error favoring Democrats in each year.

[BeFree]

42. Tell the truth, you always remember.

From TIA's link you posted:
Surprise: Preliminary Exit Poll was right on for Gore in 2000

Actually, the errors are taking away from dems, not favoring. In 2000 the exit polls selected Gore as winner. He won. The exit pollers nailed it. They nailed it 2004, but it was skewed at the last second... you can't remember? Hmmmm

First you state:

1... past data indicates that exit polls over-estimate the conservative vote
Well, over-estimating is favoring.
Then, here you say:

Notice the error favoring Democrats in each year.

Having trouble keeping track, eh?

[OnTheOtherHand]

45. well, something we can clear up, other things maybe not

The 2000 poll was pretty close overall, yes. It still overstated Gore's share, but slightly.

It's interesting that you are convinced that the exit pollers nailed the correct result in 2004, and then immediately embarked on a massive conspiracy to conceal their success. But I would not say that I "remember" that, no.

Febble did state that "exit polls over-estimate the conservative vote"; in context, it seems clear that she meant "under-estimate." I didn't bother to correct that, since you seemed to understand her meaning the first time through.

[Name removed]

47. Deleted message

Message removed by moderator. Click here to review the message board rules.

[Febble]

50. As OTOH says

yes, I meant under-estimate the conservative vote.


Source for 1:
Look at the Edison Mitofsky report here:

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

(pages 32-33)

2) You misunderstand. My point was that I don't think anyone is suggesting that people deliberately refused to participate in the poll apart from one DU poster, who does think so. Of course the poll was matched to the count - it had nothing to do with the last 600, it had to do with the availability of vote returns. See here:

http://www.exit-poll.net/faq.html#a10

3) "Any one with sense" is not an argument or a source. Many people with sense do not think it is likely. These include people who have demonstrated convincingly that Gore lost Florida to overvotes

http://macht.arts.cornell.edu/wrm1/overvotes.pdf

and the butterfly ballot

http://macht.arts.cornell.edu/wrm1/butterfly.pdf

4) See #31 and #34

You can prefer TIA's unsourced facts if you like. But your tone is uncivil, and my facts have sources.

[garybeck]

62. swing states

I've been looking at this stuff long enough to know that data can appear any way someone wants it to. In regards to your denial that the swing states were more skewed than the others, have you seen Dr. Webb Mealy's report?

http://www.selftest.net/redshift.htm

do you refute this study also? it is pretty extensive.

[Febble]

64. Well you put your finger on the problem

when you say that "data can appear any way someone wants it to". This is the key to understanding statistical analysis. You have to have an a priori hypothesis, then test it against the data, not look at the data then form a hypothesis that fits. Well, you sometimes do do that latter thing, but you then have to make a prediction from that hypothesis and test it on some other aspect of the data.

However, I think you are rather over generalizing my point. Regarding my "denial" that swing states were more skewed - I think they may have been, but I am not at present sure. My point was simply that the graph you posted does not show that - it shows that it was the bluest states that had the greatest red shift.

But as Mealy points out, that doesn't take into account weighting by size.

I think (though I may be wrong) that Mealy's analysis pre-dates the E-M report, which told us a lot more. The state exit poll projections (before reweighting to the count) have at least three sources of error: the demographic weighting; the precinct selection; and the "within precinct error" - which could be due to voter sampling bias or to vote-count corruption. The E-M report tells us something about each source of error, but very clearly tells us that by far the greatest source of error was at the level of the precinct - but, of course, not whether it was in count or poll.

Now, when E-M compared precincts in swing states with precincts in non-swing states they did indeed find that the bias was greater in swing state precincts - which is somewhat odd at first glance, given that the state average WPEs were NOT higher in swing states - like the projection figures in your plot (I assume that's what they are, though I don't know how the blood drips are calculated) they are highest in the blue states. In other words, if you plot state average WPE against state vote margin you get a slope that is low at the red end and high at the blue end, not a hump in the middle where the blue states are.

However, at precinct level, there appears to be a hump - maybe. And Mealy also argues for a hump (though I do have issues with his analysis - but I'm not, actually disputing his conclusion as such). This is almost certainly because of the fact that many swing states were large, and therefore had more precincts.

HOWEVER - and this stuff isn't simple, much as we'd like it to be - WPE as a measure naturally humps in the middle. So it is not clear, without doing precinct level analysis on a better measure than WPE, whether or not the effect is real. It may well be - but it is likely to be smaller than it looks.

As I said, Gary, - check my posts elsewhere if it wasn't a post to you - I'm not in the business of denying anything. I want to know what happened. And I want to see radical election reform, not only regarding security of the ballots (which is vital) but of the voter suppression issues that disenfranchise whole swathes of Democratic voters, particularly African Americans, and probably cost Kerry NM andn possibly NH. But I do think that if the theft that took place in 2004 is going to be investigated (as opposed to preventing future theft) we need to look at the evidence properly. Which means being very specific about hypotheses we want to test - and also being prepared to accept negatives if that's how the cookie crumbles.

To quote the Cat in the Hat:

To find a missing something you must find out Where It's Not.

That's what I'm trying to do.

[Time for change]

32. Please see post # 31 with regard to swing states n/t

[autorank]

17. Gary, Tom Hartman did a presentation on this in Portland.

He argued that things are so bad, exit polls are an essential tool to monitor elections. He talked about Serbia, Georgia, and, of roucse, Ukraine. He did a great job and was well received.

People were also most grateful to get the CD from solarbus.org. I was there when they were passed out.

[garybeck]

23. glad the CDROMs made it

you helped.

[Hissyspit]

20. If YOU or any of us tried to "assure" a Presidential election, this is...

what it would look like.

[jen4clark]

56. Please check out

the Exit Poll Chapter of "WHO'S COUNTING".


Good Poll, Bad Poll



The Zogby Polls, which were known for accuracy, had Kerry winning a clear majority and sweeping the Electoral College. At 5 PM on election night, Zogby predicted Kerry with 311 electoral votes, beating out Bush by nearly 100 electoral votes. Likewise, Princeton Professor Sam Wang’s Meta-Analysis on November 1st called for Kerry with 108 electoral votes over Bush.



Zogby states that he first noticed significant polling differentials beginning in 2002. Substantial differences between polls and final results were recorded, often staggering and historic in scope. In 2002, Zogby says he was wrong in Illinois, Georgia, Colorado, and for the first time ever – wrong even in New Hampshire. This was the first national election year after HAVA was enacted, and electronic voting more widely installed.



Exit Polls, usually even more accurate - had Kerry winning as well (51 to 48), and especially strong in the critical battleground states of Ohio and Florida. Ten of the eleven swing states that had been favoring Kerry in the polls unexpectedly shifted to Bush. According to a report released by Mitofsky/Edison polling firm in January 2005, the greatest discrepancies between the polls’ predictions and the official count were found in the battleground or swing states where Bush shifts exceeded the accepted Margin of Error.

-snip-


One exit pollster, Ken Warren – St. Louis University Professor of Political Science- who has been working in the field now for over 20 years, says he has only on ONE occasion seen any error greater than 2% between the exit polls and the final tally. That was in a 1982 Primary election in which massive vote fraud was subsequently uncovered.



Here’s an amusing snip from an article that appeared in the Washington Post, on November 4, 2004, offering one rationale for the polling ‘mix-ups’ : “A server at Edison/Mitofsky ‘malfunctioned’ shortly before 11 PM. The ‘glitch’ prevented access to any Exit Poll results until the technicians got a backup system operational at 1:33 AM yesterday. The crash occurred barely minutes before the consortium was to update its exit polling, with the results of later interviewing that found Bush with a one-point lead.” Those midnight technicians keep stepping in to help out in our new electronic era elections....and just in the nick of time!



This time we have the vote totals and the exit polls both changing their minds in the middle of the night. Both flip flopping in the same direction – deciding they preferred Bush after all… After the perfunctory moonlight visit from the “technicians”. It could happen.



As the unruly racket about the Exit Poll discrepancies grew even louder from the keyboards of disgruntled Democrats... the preponderance of ‘official theories’ explaining just WHY they were so inaccurate grew proportionately. At the end of the day, we ended up with nearly as many rationales for Exit Poll discrepancies as we had reasons for invading Iraq.



There were the college-aged pollsters that didn’t do the job, the early over sampling of women, the chatty Democrat theory, the ‘polls aren’t perfect’ theory, there was the glitch in the malfunctioning server, the danged Dixiecrats theory who- even though they vote Republican- stay registered as Democrats, and then, my pet theory-- the Shy Republican Theory.


Nothing beats the Shy Republicans theory...those voters who purportedly, in record numbers, recoiled from pollsters in shame – unwilling to discuss their Bushian vote aloud with anyone.

Two problems with that theory - First; the polls are conducted anonymously on forms filled out in private, and have been for years...No names or identification requested. And secondly; in Mitofsky-Edison’s own report of January 2005, data showed the Bush voters were slightly more likely to complete the exit poll surveys. Six months later - and they’re still coming up with new theories.


And they call us “conspiracy theorists”...

With all the annoying exit poll discrepancy reports whisking across the internet at lightning speed, it should come as no surprise that the Republican National Committee Chair, Ed Gillespie, waited only 2 days following the 2004 Presidential election to publicly call for the elimination of all exit polls before the National Press Club.... Thereby, entirely eliminating from American elections, the traditional stop-gap device and primary election fraud detector historically relied upon by the rest of the world.

Bad polls.


-much more at WHO'S COUNTING

[OnTheOtherHand]

57. hi, jen4clark

It isn't possible to respond, point by point, to every uneven article about the 2004 exit polls that has ever been written. This one is pretty calm. Let me make three points, and then if you are interested, we can keep talking.

(1) I suspect that Ken Warren is being subtly but decisively misparaphrased here: "he has only on ONE occasion seen any error greater than 2% between the exit polls and the final tally...." I haven't consulted the original source, but compare this from AlterNet:

http://www.alternet.org/story/21036/

"Ken Warren (2003) has never had an error greater than 2 percent...."

The difference between "seen" and "had" is crucial here, because it is a matter of record -- no one has denied, as far as I know -- that U.S. presidential exit polls have had errors greater than 2%. It's nice to know that Ken Warren hasn't had one, but it isn't really pertinent.

(2) The argument about the E/M computer glitch actually works in the wrong direction. At 11 PM, everyone in the world with an Internet connection had access to exit poll results that showed Kerry ahead in Ohio. The computer glitch, according to the story, delayed the posting of revised results -- based in part on official returns -- that gave Bush the edge. As has been explained here before, exit poll results are updated to reflect official returns; it neither proves nor disproves fraud.

(3) The whole thing about "official theories" is basically made up. Febble has written a lot about the general topic of survey bias; I won't try to summarize it all here. Frankly, it really depends on whether you are willing to learn why a lot of the article you posted is wrong (WHETHER OR NOT Bush stole the 2004 election), or whether you have a prior investment in it being right.

[jen4clark]

58. Hi OTOH

Thanks for the reply.

I'll send the author over to address your points, as she's got the inside knowledge regarding the information presented in Who's Counting. She invested close to a year gathering information, investigating it, writing and re-writing before the launch of the website last week and knows this stuff inside out!

[OnTheOtherHand]

59. OK, and just so we understand each other --

I've poked around the Who's Counting web site for a while, it has lots of good information, and as far as I can tell I generally agree with the policy recommendations (actually, I might go a bit further on paper ballots versus paper trail, although I'm not a hand-count purist). I don't agree with her take on the exit poll arguments, but I'm not trying to throw out all the good stuff.

[jen4clark]

60. Cool.

She's actually very open to suggestions, criticism, tips, etc. and there is a "contact us" link set up just for that purpose! But as we're trying to garner interest, get people informed, and spread the word far and wide, I think it's beneficial for issues like this to be discussed where others can see it too.

[jen4clark]

68. OTOH: reply from M. England

author of Who's Counting. I sent her an email with your comments and got this reply. Posting her reply so you know we're not ignoring you! Thanks, jen

M England writes:

I have been so busy today I just saw your emails and do not even have a minute to go see the thread (and I am not registered at DU)

I know there are lots of people out there that hate the exit poll argument which is why I placed it so low in the chapter list and why I tried to simply present OTHER experts points of views and not form conclusions myself....or NOT make it any kind of rock solid BASIS or foundation upon which to base the theory of election fraud. It's just one of the many shall we say "anomalies" present in the entire bad-smelling scenario.

So I won't be able to respond today. Am doing major contracts on my business side this morning that will keep me pretty occupied these next two weeks. I will try to respond when my schedule clears. Please explain that I am a mere citizen voter who compiled research from many available sources who in NO WAY was posing as an authority, an expert or proferring concrete conclusions. Posing the questions, seeking the answers-- and when so much has been covert, hidden and distorted....that is no easy task.


OTOH made an error in assuming the Exit Poll chapter was MY CONCLUSION on the issue as opposed to my presenting various other experts and statisticians conclusions for the readers consideration. On exit polls I wouldn't even pretend to have the final word knowledge -- but I can say, as with all other issues surrounding these past two national elections-- something is so rotten in the state of Denmark/ USA! I can confidently say that now after all this research.

[OnTheOtherHand]

71. thanks for relaying the response

To clarify what I said earlier, I didn't mean that I had previously checked out the site and was sulking (grin) about the exit poll entry -- I only looked at it after you recommended it.

The web site does present as fact some things that just aren't factual. For instance, I'm not quite sure what was the source for this:

"They (Mitofsky International) boast an average 1% margin of error in their work. For 38 years, they offer this acute accuracy level in thousands of elections both here and abroad. EXCEPT in the 2000, 2002, and 2004 US elections."

I won't generalize across thousands of MI exit polls -- I really have no clue -- but it's on the record that their 1992 presidential poll had much more than a 1% margin of error, and a much larger error than the 2000 presidential poll. I dunno whether she had something else in mind.

Ohmigosh here's another one, about the German exit polls: "Because for generations the polls' projections have never been any more than one-tenth of a percent off from the final tally!" For generations?? No more than one-tenth of a percent?? Sorry, it just isn't true. She can check Freeman's paper. She should also note that the recent German exit polls had -- I've forgotten now, we went through all this right here -- several-point projection errors.

I know most of the experts she cites, and I like them (well, things are a bit iffy between Ron and me right now, but I am told he is a doll). But she has missed all the experts who are so sure there is no there there that they haven't bothered to go on the record rebutting it -- and, actually, most of the experts who have gone on the record. I'm not saying which experts are right, although I know what I believe. But since she is making the decisions about which experts to quote, she is responsible for her conclusions even if she disclaims having the "final word."

Cheers,
OTOH

[jen4clark]

74. I'll do the best I can

in addressing your claims. I didn't do the research or writing but I did work on this project from beginning to end as proofreader/editor and I can assure you everything was checked, referenced, doubled checked and double referenced.

We appreciate the critic nature of your comments - and were aware in the writing that there exists out there a group of people who thrive on just that-- they seem to exist to criticize the efforts of others and in doing so find their own sense of satisfaction in their manufactured esteem...

The role of the critic is necessary and we can even be thankful they exist to alert others to possible errors or shortcomings.

The role of the messenger is also vital when the message has been stifled and squelched as it has -- as it never has been in history as it is today-- on this subject.


The Mitofsky 1% figure was from Mitofsky - (Author thinks it came from their website, but may have been an article in which they were quoted. When she has time she will sort through her documents and we'll provide the exact link.) Until then I found this:

http://www.inthesetimes.com/site/main/article/1970/




In reference to Germany - that info was from Thom Hartmann's recollections of his time living there.

http://www.truthout.org/docs_04/110604Z.shtml

From the BLIND TRUST Chapter:

Best-selling author Thom Hartmann fondly remembers the voting process when he lived in Germany, not dissimilar to the one we enjoyed here for 200 years before HAVA.

The people fill in hand-marked ballots which are then hand-counted by civil servants who are supervised by volunteer representatives from each political party. The time it takes is irrelevant according to Hartmann, because the people rely exclusively on the news media exit polls that night for results....because, for two generations of reporting....they have never been more than one tenth of a percent off...

-cont.

[OnTheOtherHand]

75. well, that's a disappointing response

I thought I was doing what you wanted by making the observations public.

I don't know what to say about Thom Hartmann's fond memories, so I will leave well enough alone, I guess!

[kster]

65. Voting methods

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=201x5446#top

[Febble]

67. Thanks, that's helpful

and good to have it by population served rather than by county, which makes a huge difference to some states, e.g. Ohio.

I don't know if anyone has graphed exit poll discrepancies by state by proportion of population using each machine type - the only plots I have seen were based on some very strange data regarding voting method.