Since things have quieted down here, I think it necessary to provide for your daily indignation a thesis of truly volcanic significance: that the 2004 election was stolen. This thesis is current held by only three groups of people: left-wing extremists, conspiracy theorists -- and a large group of completely sober, quite competent statisticians.

I direct your attention to the report at uscountvotes.org. This report, 26 pages long, presents a detailed analysis of the well-known discrepancies between the exit poll data and the vote counts. You may recall this controversy erupting just after the election, and then dying out. It died out because the statisticians went to work to analyse every aspect of that data. The polling group, Edison-Mitofsky, released a study in late January concluding that the discrepancy was due to oversampling of Kerry voters. To put it simply, they argued that Kerry votes disproportionately agreed to be interviewed by the polltakers, while Bush voters disproportionately refused.

The people at uscountvotes went to work on the Edison-Mitofsky report and concluded that their calculations were wrong. They published an initial response in late March, then a revised and improved version in April, and have now released a third draft incorporating every criticism that has been offered. From the abstract:



They also note that their paper is written for the lay person, with the heavy math confined to the appendices.

This paper has been through the mill; it has been publicly criticized for months now, and the authors have taken into account every criticism, closing every loophole, battening every hatch, fixing every possible problem. This thing is close to bulletproof. There remain some unaddressed objections, but they seem pretty far-fetched to me. The only substantial criticism comes from "Febble's Fancy Function", explained in much detail here. I confess, however, that the debate over Febble's Fancy Function is so abstruse as to sorely try my understanding of statistical analysis. For the moment, I think that the USCountVotes people have the upper hand, but they have not yet succeeded in dismissing Febble.

The USCountVotes people have requested that Edison-Mitofsky release their raw data, which could clear up the dispute, but E-M has refused to do so. I am not certain on this point, but I believe that past practice has been for polling organizations to release such data to the academic world six months after the election. Conspiracy theorists claim that this is evidence the E-M has something to hide.

There's a great deal of additional information on this: an affadavit from a programmer at a firm making voting machines testifying that a Republican operative approached his firm requesting software with which to hack their machines -- he claimed this was for evidentiary purposes in a legislative debate. The programmer wrote a hacking program, and demonstrated its effectiveness, but the Republican operative simply took the program and the programmer never heard anything else about it.

It has been demonstrated beyond doubt that most of the election counting computers are easily hacked; I believe one demonstrator pulled it off in under 60 seconds.

The detailed data from Cuyahoga County, Ohio, has been analysed to show serious statistical irregularities in the data highly suggestive of vote tampering. The District Attorney there has initiated an investigation.

So all of this leads to the two questions I offer for debate:

Should Congress subpoena the Edison-Mitosfky raw data so that it can be analysed to (possibly) clear up the exit poll question?

Should a federal-level investigation be initiated into possible electoral fraud? Particularly, how much investigative money does this controversy deserve?

I ask all concerned to address this highly partisan issue with care and sobriety.

Should Congress subpoena the Edison-Mitosfky raw data so that it can be analysed to (possibly) clear up the exit poll question?

There are numerous factors as to why exit poll data can be wrong. First, time of day, proportion of ethnicity and gender chosen, and releasing the info too soon so those on the West Coast find out.

We must keep in mind that there are perhaps dozens of studies and findings out there that contradict eachother in numerous ways. There is no way of finding out who is telling the truth and who isn't.

Should a federal-level investigation be initiated into possible electoral fraud? Particularly, how much investigative money does this controversy deserve?

No. It would be money wasted because if even hypothetically there was some truth...it would be buried under the Republican controlled Congress.

But, lets not make some crazy mistake that the presidnetial race was some discrepancy from the rest of the voting. Republicans won MORE seats in the House. Republicans won MORE seats in the Senate. The fact that they won the presidency should really come as little surprise. If the president elected was a different party than the growing majorities in Congress...then something fishy might have gone on.

The conspiracy of election fraud for this election will continue for the sole reason that people have such a problem with accepting that others disagree with their views. A lot of people were sure that Kerry would win because he thought like they did. When he didnt, its a major psychological blow. Would have been the same for the Conservatives.

I'd sure like to know about any studies other than the Edison-Mitofsky study and the Febble study that challenge the USCV paper. There were a few generalized papers tossed out in January and February, but nothing substantial that I have seen since then. Can you direct me to some of these?


One of the minor data points in all this is the observation that many ballots were reported by the election authorities to contain votes for President Bush and a Democratic senatorial candidate. A quickie statistical analysis of this has shown this phenomenon to be anomalous, but no complete study of the phenomenon has been carried out. This would be one of things that could be studied.


Yes, the conspiracy people and the extreme left-wingers will take this position, but what about the statisticians?

Should Congress subpoena the Edison-Mitosfky raw data so that it can be analysed to (possibly) clear up the exit poll question?

Yes, if it is traditional for this data to be released after an election, it should be released. It is, after all, a matter affecting our whole nation, and all of the data should be available for scrutiny.


Should a federal-level investigation be initiated into possible electoral fraud? Particularly, how much investigative money does this controversy deserve?

Yes it should. Not just because of polling data, but because of the voting machines held back in Ohio while those in poorer areas waited in line in the rain to vote (as wealthier Ohioans skated through the lines in much less time).

And the reason I quoted lederuvdapac's post is this:

Lederuvdapac, you stated that:



This ignores the fact that voters chose these candidates through the voting machines, too! And there were discrepencies in many state and local races that actually indicate fraud, or at very least, serious errors, in the entire process.



I won't vote in 2006 or in 2008 on these machines. While I had no problem (that I know of!), with my voting experience, I was in a precinct with a majority of red voters. There were problems in poorer areas. I don't think Americans should trust their elections to companies with CEOs like Diebolds, who stated he would deliver Ohio to the President (and I believe followed through on his promise!).

Should a federal-level investigation be initiated into possible electoral fraud? Particularly, how much investigative money does this controversy deserve?
Any federal money channeled into election "fraud" investigation would merely being throwing red meat to the Democratic base. It would be politically tainted, suspect, and never yield anything useful. (see 9/11 commission)

While I understand the skepticism of these particular statisticians and this particular study, I think it's compelling that this is seen as something unique to the 2004 presidential election (i.e., "this has never happened before").

For instance, the Washington Post noted that the results for the 2004 exit poll vs. reality were an exact parallel of the 1992 election.

Here is a GIANT PDF file in which CNN analyzes the exit poll vs. reality data for the 2000 election. Highlights (from pg. 45) show some states which were massively overstating Gore votes vs. Bush. Notably some "battleground" states in here.


More questions as to the accuracy of the National Exit Poll people themselves, given well before the 2004 election (from New York Times):


The study's conclusions:

1. No. Doesn't congress have more important issues to deal with than the incompetence of polling institutions?

2. No. The elections are held in the manner that is proscribed by law. There were observers on both sides of the political aisle. In particular, the Kerry campaign had thousands of lawyers poised to pounce if any "hanky panky" was seriously alleged. They didn't.

If polling organizations want to improve their credibility, they should examine their own raw data to determine where their models went off the rails. This is not a job to be funded by taxpayer money.

Three factors need to be taken into account in considering the value of this information. First, the mistakes of past exit polls have formed the basis for continuing improvements in the methodology. The field of exit polling didn't have much money behind it until the 70s, when television networks decided that they were willing to pay for that information. The field has improved dramatically since then. Moreover, they have expanded the size of their polling samples over the years.

Second, the proper basis of comparison is not one or two election campaigns, but the entire base of data from all election campaigns. For example, if I express surprise at a 24-pound cat, and you dismiss my suprise with the observation that you once saw a 24-pound cat, you're off the mark. More useful would be a distribution of all the weights of lots of cats; that would permit us to say, "Gosh, only one cat in a thousand weighs as much as 24 pounds. This is certainly a strange cat!" In the same way, our experience with thousands of elections shows that exit polling is generally accurate to well under 1%. A discrepancy of 5%, which is what we obtained in this election, is so far from the mean that it constitutes a statistically significant exception.

Third, the same exit polls that had such a huge discrepancy on the presidential vote were remarkably close on senate and House races. If there were some methodological flaw in the exit polling, why did it express itself only in the presidential race?



Allow me to provide some additional quotes from the same study, these from Mitofsky and Lenski (the quote you provided was from the primary authors of the study):



There follow four specific recommendations for improving the reliability of their predictions. They note:



Actually, the point of the study is that the polling institution was not incompetent, that in fact the error may have occurred in the vote counting, not the exit polling. What the study I quoted demonstrated is that there is no explanation for the discrepancy other than fraud.

A more significant response to your question is to consider the hypothesis that the election was fraudulently modified. At present, we have solid indications that this hypothesis might be correct. If it were correct, then would not the identification of this problem be a matter of paramount concern for the entire nation?



Actually, there are thousands of documented cases of procedural irregularities and outright violations of the law in the 2004 election.

Lastly, I'd like to put a challenge to carlitoswhey and LordHelmet: is your reasoning in this matter completely objective? Would you truly be so dismissive if Senator Kerry had squeaked through on a close vote that had a great many irregularities associated with it? In Washington State, there is even now a controversy regarding the 2004 election, because the Democratic candidate squeaked through on a few hundred votes. The Republican candidate is contesting the result. Do you condemn his efforts?

The results speak for themselves. The polling institutions got it wrong. Exit polls are based on sampling theory. They don't represent the entire population if the assumptions made by the model are incorrect as they clearly were in this case. The real question that I have is were the polling institutions merely incompetent or did they have an agenda to bias the election by releasing incorrect exit polling before the polls closed??



Prove it.



Your premise is false. There were no documented cases of "irregularities" that could erase the Bush margin of victory in ANY case.

And, in the case of Washington, yes I absolutely support the GOP effort there. There is no way the governor of that state should be elected on the basis of votes from convicted felons and dead people.

No, that's false. Strictly speaking, all we know is that one measuring instrument -- the vote count -- yielded one result and another measuring instrument -- the exit polls -- yielded another result. Inasmuch as the vote count is based on a larger and more carefully selected sample, we would normally resolve the discrepancy by dismissing the results of the exit polls. However, when careful examination of the exit poll data demonstrates, as it has in this case, that there is no logical reason to dismiss the exit poll results, then we are left with a logical dilemma: which of the two measuring instruments to believe? This dilemma can only be resolved by further investigation. That is the point of this debate. You are denying that any dilemma exists; your denial rests on a categorical dismissal of the exit poll data. Can you provide any reasoning to support your dismissal? Have you any criticisms of the statistical study I linked to?


Gladly:
The Century Foundation
Conyers Report
National Research Commission on Elections and Voting
Common Cause
People for the American Way
National Voting Rights Institute
Black Box Voting
Voters Unite


Please consult some of the links provided immediately above documenting numerous electoral irregularities.


Then it would appear that your judgment here is based on partisanship rather than principle.

I asked for proof, not links to far left web pages. John Conyers? lol. I live in Michigan and am FULLY aware of that guy. People for the America Way? Why did you leave out moveon.org and michaelmoore.com??

I would accomplish the same thing if you asked me to prove that evolution (which I happen to agree with) was false and I gave you a link to Jerry Falwell's website.

OK, since you're going to base your case on the political leanings of the sources, you must be consistent and explain why you think that Common Cause, The Century Foundation, and The National Research Commission on Elections and Voting are all far left organizations undeserving of serious consideration.

Because some people didn't tell the truth when asked about their vote for President? In all the discussions on this issue, this simple premise doesn't seem to get discussed...probably because it removes the issue. Remember when they used entry polls? Results were way off...even though these people were asked how they were going to vote right on their way into the voting place. Did they have some epiphany on their way through the door? Not likely. How people vote, and how people say they voted, are often two different things. Why the discrepancy then between Presidential polling, and polling for lesser offices? Because people weren't as leery discussing their vote for the latter. Until someone can show this isn't the case...further investigation isn't really necessary, is it? As the saying goes, there are lies, damn lies, and statistics. If there's some proposed mechanism through which this grand conspiracy succeeded in manipulating the votes, then, by all means, lets fix the mechanism. Until then, though, I think we can devote our time to better things than chasing pink elephants...particularly when those spreading the rumor of their existence are curiously reluctant to release any of the photographs.

I am curious just how long the whole 'They stole the election' thing is going to run? Maybe if more time were spent on developing alternative policy, and less time chasing ghosts.....

You are referring to accuracy "to well under 1%" and I can't see where you're getting this. Nor can I see the 5% figure you are quoting.

From the Washington Post article I linked to before:
So, the "final" exit poll data showed Bush ahead by a point. Or was it that Kerry "won" based on the exit poll data. Bush won by what, 2.5 points? So, I guess you're right - the error is only 1 or 2 points...according to the Washington Post anyway.

But, I'm skeptical, so I go to this analysis of the polling data and response to the Mitosky-Edison report, and I see that the exit poll data approximates a "normal" distribution curve, but shifted one standard deviation to Kerry. I see some of the analysis and gymnastics applied, and how some of the "odds" of this to occur were calculated, and they contradict each other. I note that the links from this site, like your sources for election fraud allegations, lead me to a litanny of leftist organizations, and I am skeptical.

To color my view, even though anecdotal, I note that I live in a big, democrat-dominated city. My wife, a (republican) election judge in November 2004, cited dozens and dozens of attempts to subvert the law and allow non-registered-voters to vote (democrat). This is one person in one precinct in one city, and I multiply this out to Philly, Detroit, Cleveland, etc. I also note that Democrats have been actually convicted (not just accused) of election fraud and abuse in Milwaukee and other cities, and my skepticism elevates.

And when I read the conclusions of the original report you cited, I find it incredible. As in not believeable. Because this exit poll data was apparently skewed, and the vaunted statisticians cannot explain the discrepancy (based on admittedly incomplete data), "The US electoral system is vulnerable to massive vote fraud in the future." They made about 3 logocal and rhetorical leaps too many based on their study. The statisticians appear to have forgotten that there are people in between the theory and practice.
I think that leftists questioning this election is completely healthy. I disagree with their results, but sunlight is the best disinfectant. As such, I encourage more analysis of exit poll data, how it's early release may have affected turnout in battleground states (a topic our authors seemed not to find of interest ) and how early we should allow the media to "call" a state.

As for Washington, I think that this election bears re-visiting, which is happening in the current trial. Today's Seattle Times:
Also, just to defend LordHelmet's note on "dead people and felons" this has been verified - from the Seattle P-I link
Also, in the Seattle Times, we read that the mail-in ballots were certified, despite not being properly accounted for, a general pattern of fraud existed in King County.,seattle times

I base my support of the Washington challenge on facts. I suggest that those who have lost a presidential election by more than 3 million votes are basing their challenge on wishful thinking. They certainly seem disinterested in documented cases of election fraud favoring Democrats (Wisconsin, for example).

Or, as one manhattan resident famously put it in 1972, expressing disbelief about election results - "But I don't know anyone who voted for Nixon!"
edit 1 - '72, not '68. I was too young to remember anyway.
edit 2 - found the 5% figure you were quoting, having re-read the Mitofsky data. But I still question the "uscountsvotes" report and note that the most wacko of conspiracy nuts are quite capable of hiring "completely sober, quite competent statisticians."

It's an interesting analysis considering there appears to be a solution is search of a problem.

As usual, Republicans gained more seats. So the presidential results would be in line with the rest of the election. And this analysis avoids that issue.

But where is all the variable data and assignable causes in that report? Weather, truthfullness, red/blue quotient, male/female, press in the vicinity, age, and the 1000 other variables that could cause projections to be wrong.

The democrats have been getting their hats handed to them for over 10 years and need to figure out a way to win instead of crying foul everytime something doesn't go their way.

I think the election of 2000 alone merits a more effective revamping of our federal election system.

Voters should be able to receive a paper confirmation of what their vote was.

Congress should err on the conservative side in terms of investigation the results of national elections and investigate is there is the appearance of irregularity.

We need to move to a direction of better faith in our electoral system.

Dismissing democrats as whiners is not the best direction to go. IMO we made too little progress from 2000 to 2004 to attaining a system that cannot be questioned.

I give the results the benefit of the doubt and I am far from convinced or really concerned at this point that the 2004 election was stolen.

But if these things can be tampered with they will be. Let's get federal electoral security up to par with businesses that have to rely on electronic information for $$$$$. Secure the crap out of it and make it seem that corrupting an election is harder than nabbing cash from a Brinks truck.

Let's see...paper ballots as currently used don't meet your standard, since no receipt is feasible with current "butterfly" ballots. I could foresee a two part paper ballot, but remember that you don't just have one race. Each ballot represents a dozen or more races.

Also, there is a privacy issue with receipts. As with all issues, there is a middle ground that needs to be met between privacy and accuracy.

Electronic voting methods, while providing the most accurate counts, are perceived as susceptible to outright fraud. I would think that a bi-partisan audit committee could easily ensure accuracy, but as in all things, perception is stronger than reality.

On to the questions for debate:

Should Congress subpoena the Edison-Mitosfky raw data so that it can be analysed to (possibly) clear up the exit poll question?

No. We have two data sets here: projected votes (exit poll data) and actual votes (ballots cast). Only one set of data are relevant to the election outcome. That would be actual votes cast.

What purpose does the subpoena of data which has no relevance to the outcome serve? While the exit poll data collection methods probably need to be improved, that is not the purview of the government, since the government does not rely on exit polls. It relies on hard data.

Should a federal-level investigation be initiated into possible electoral fraud? Particularly, how much investigative money does this controversy deserve?

Again, the fact that two data sets do not match up is not cause for election fraud inquiries. Carlitoswhey has posted that similar discrepancies have occurred in the past. Your "24 pound cat" example is way off base, since the data set of Presidential elections is more on the order of 10 than the 1000 you posit. Two sets out of 10 (the number is more like 8-10 if the 1970s is your starting point) makes the presence of similar circumstances significant.

From the Rasmussen Report:



Just about every poll had George W. ahead going into election day. The exit poll seems to be one of the few polls that got the results wrong. Maybe Democrats should just admit they lost so they won't be labeled whiners and might even get some bipartisan support at strengthening voter access and security.

But, if you insist on exit polls:

www.cnn.com/ELECTION/2004/pages/results/states/US/P/00/epolls.0.html

Strangely it does not give a final total so you have to extrapolate the winner from the male/ female statistics. A little math:

Males 46% of the electorate, 55% voted for Bush equals 25.3% add females 54% of the electorate voted 48% for Bush equals 25.92%. Total 51.22% for Bush. I don't think there are enough transgender individuals, even though I am betting they were strongly for Kerry, to deny Bush the victory in this poll. It looks like the voting machines may have cheated Bush out of .5% of the vote.

Then show me where they contradict each other. Don't just make the claim -- present your evidence.


If you want to make a claim that the analysis is wrong, by all means do so. Don't waste electrons arguing about the people doing the analysis.


These anecdotes are fraught. There are thousands of them on both sides. They are trees in the forest. We can't draw any conclusions from them because they are the tip of the iceberg and we have no idea how many such incidents went unreported. So we need a way of seeing the forest, not the trees. That's what the exit polling data provides.


OK, so you don't believe it. Is that because you have a logical reason for not believing it or because you don't like the conclusion it points to? What 3 logical and rhetorical leaps are made by the study? Tell me what is wrong with the analysis.




I agree completely that any electoral irregularities in Washington state should be fully investigated. Unlike you, my position here is principled, not partisan. I can happily accept any result, so long as it is arrived by fair and proper methods.



Where do you question the report? What do you question in it? The results, or the methods?



Yes, that point is worthy of consideration. But there are also lots of specifics that show wide discrepancies between votes for president and votes for senator on the same ballots. This particular issue needs to be addressed at some point, but for now we need to focus on the main issue: the discrepancies between the exit polls and the vote counts.


The methodologies take into account a great many factors. You can find details on the methodologies at some of the links I posted earlier.


Let's address the issue here: does that USCV report justify a subpoena of the Edison-Mitofsky raw data, and should there be further investigation?


"Relevant" is an awfully broad term here, and in this context your use of it is incorrect. What you mean to say, I believe, is that the actual vote count is all that legally matters, and that statement is correct. However, there remains the unanswered hypothesis of electoral fraud, and the discrepancy between the vote count and the exit polls supports that hypothesis.


The raw data is relevant to the analysis of the exit polls, which in turn cast a shadow of doubt on the vote count. That's relevant.


You are relying on the assumption that only presidential races need be considered in evaluating the performance of exit polls. To justify that assumption, you must show that there is a substantial difference between an exit poll of any other race and an exit poll of a presidential race, and that the difference produces statistically signficant results. Your task is made particularly difficult by the fact that often exit polls are combined: the exit poll for the senate race is the exact same exit poll for the presidential race. So, how do you justify your assumption?

Hugo, the exit poll results that you report are the massaged results. The exit polls showed Kerry winning by 2 points right up into about midnight Eastern Time, then they were suddenly taken down. They reappeared two hours later with Bush winning by 3 points, even though the actual votes that had been added to the sums had changed by less than 1%. What happened was that the pollsters went back and re-weighted their values to obtain results that matched the reported vote counts. That's why the raw data is so important -- it will show what really happened.

Eeyore, I agree with this sentiment. As I have said in other 'election results' threads...I'm all for anything that materially improves the process. There is no valid reason we shouldn't have the gold standard of election systems...and I think working towards such a standard would both restore voter confidence and create some much needed momentum towards more people voting. This is why I point out that this particular incidence doesn't point towards methodology at all. Nothing in this report indicates what should be fixed...there's basically just a rumour that something fishy went on. So, it's not really doing anything but further degrading faith in the election process. That's not really doing anyone any good.

Just a further note on this...if anyone really wants substantive change in the process, and wants it to attain results more people have faith in...the whole process needs to be arrived at apolitically. Otherwise, the other side will, by default, have reason to be skeptical. Crying foul over the last election results doesn't do this...it just worsens the problem. Whether the intent of this study was political or not...the issue shouldn't be with the results. The issue should be with the process.

From: "With Exit Polls: President Dukakis"



There is a reason they "massage" the results. It's to eliminate biases. Or maybe them darn Republicans robbed Dukakis too. I believe CNN has already analyzed the exit polls and used scientific methods to get the correct result. Just like in the Rasmussen Poll which almost unerringly forecast our last election.

But what about this (note, an AP newswire story):

http://www2.indystar.com/articles/2/192519-7912-168.html



And this (computer tech source):

http://ecoustics-cnet.com.com/Report+Flori..._3-5459186.html



And this (originaly a Boston Globe story):

http://www.truthout.org/docs_04/120204Y.shtml



I'm still looking in vain for credible stories of machine (optiscan included) error that favored Kerry. If the errors were random and innocent, around half of the errors would have favored Kerry, not Bush.

And this is interesting:

http://electionarchive.org/ucvAnalysis/US/...on-Mitofsky.pdf


Note, the next point made is that the Edison/Mitofsky researchers never even considered voter fraud, and so did no analysis based on the possibility. Now, I'm not a scientist by any means, but isn't it erroneous to fail to test a hypothisis that logically would account for the event being examined? Don't all reasonable hypothesis have to be ruled out before one can be decided upon?

We can't fix problems that we refuse to recognize. This study attempts to identify a problem. That's just the first step. Next, we need to investigate it. If it pans out, then we take action.

I'd like to step back and make some basic observations here. Every single conservative (I believe) has rejected the suggestion of further investigation. Every single conservative has denied any validity to the study. How many conservatives actually read the study? More important, how many actually understood it? I have to admit, even with my extensive background in statistics, I found it a tough task ploughing through that thing.

Here's the real issue: it is extremely unlikely that anybody in this discussion fully understands the study, and unlikely that many of those rejecting it have any sound basis for rejecting it. I believe -- but cannot prove -- that most of those rejecting the study do so because of its result, not its validity. That study could be endorsed by 95% of the working statisticians in this country and still they'd reject it. One chap quoted Mark Twain's old saw about "lies, damned lies, and statistics". I don't think that anyone has yet made the cliched observation that you can prove anything with statistics.

But the fact is, this study appears to be solid. It has been shot at for months by a lot of people, and it has been emended, corrected, modified, and steadily improved. It's still not perfect, but there aren't many professionals left who are willing to keep shooting at it. A consensus is emerging that these people are onto something. What are we ignorati to do about it? We can't really understand it, but the experts seem to feel that there's something solid there.

Simply rejecting the experts as biased won't do -- we don't even know the political leanings of the people who did the study. All of the arguments that I have seen presented here attempt to wriggle away from the central truth that we have a solid rational basis for believing that the 2004 election might have been fraudulent.

What I find particularly troubling is the blatant partisanship offered without a blush. The same fellows who flatly reject any investigation of the presidential election eagerly demand an investigation of the Washington state gubernatorial election. That a person could say such things in this forum suggests that we have all come to accept unparalled levels of untruthfulness in our political discourse.

There's one other aspect of this that I haven't brought up yet: what are you afraid of? A few million dollars and a subpoena could get to the bottom of this. We spend that kind of money on all matter of trivia. If you're really right that the election was fair and even-handed, then an investigation will only go to prove you right. There's simply no fiscal reason for opposing an investigation. The only reason that makes sense to me is that conservatives fear that an investigation might just reveal a truth they don't want to acknowledge.

Your comments are off-base.

What one needs to understand is that the exit polls are a subset of the actual election results. On top of that, the exit polls are a random (or semi-random) subset of the election results.

It is much more likely (in my layman opinion) that the polling methods were flawed than any other results. There have been ample examples of how exit polls for Presidential elections are flawed, regardless of who controls Congress, the White House, or any other institution.

The fact that a semi-random subset of data does not always predict the entire set should not surprise a statistician.

CNN has their own experts to interpret their own exit poll which showed Bush winning. The vast majority of final pre-election polls showed Bush winning. The election Bush won. Previous history shows exit polls are biased to the Democrats. How did these so-called experts explain the Dukakis result in '88.

How exactly would I (or you) discount this analysis? We are not statisticians, we do not have the data. We have common sense and a working knowledge of statistics and mathmatics. But nonetheless.

My reasons for skepticism:

Someone named Febble has disproved the disproof of the data. I have no reason to disbelieve Febble nor uscountvotes.org. I've given it a look, and compared to the study you posted, I find it more readable and credible, but the whole exercise is pretty arcane.

I have never seen a normal distribution curve shifted one standard deviation to the left, which leaves me shaking my head. Especially since the authors tell me that "the vote simulation is new and not refined yet. It also assumes a normal distribution of precincts." So, they assume a normal distribution, then go on to analyze and make conclusions given a normal distribution.

In their defense, they "assumed" because they don't have the raw data. This data, is of course proprietary to the pollsters, and other pollsters may have ulterior motives in getting their hands on this data, no? Is this not a lucrative project for a researcher?

They also claim that they will publish data needed to support their claims, but the link is dead. see here

As DaytonRocker notes, they can't explain the uniform response bias theory but they did not perform cluster analysis or overlay any other type of socio- or demo-graphic analysis, (like Claritas clusters or Spectra data). You point us to other links, but I point back - this was not explained by uscountvotes. They only looked at the binary - Kerry v. Bush - dataset within a given precinct, which leaves me lacking for other likely explanations. No multivariate analysis of rich / poor, white / black, rain / sun, media consumption patterns, etc. For all I know, the lack of response to pollsters correlated 100% with a factor that this study did not consider.

They assumed voter fraud because their model could not reconcile the exit poll data with the actual election results. This is bad science. It's as if a meterologist predicted a sunny day. When a thunderstorm showed up, his model cannot explain it. Therefore, Vlad Putin must have seeded the clouds. It's a leap.

Additionally, these impartial, sober statisticians include some highly partisan language in their conclusions, chiding us to do better and clinging to "vote embezzlement" theories which are "highly plausible." I see nothing in this study that demonstrates plausibility for claims of "vote embezzlement" whatever that is.

The electrons wasted are those broadcasting theories about long lines in "minority precincts" and other such conspiracies. You act as if this is some hypothetical exercise, but indeed it is quite real and therefore not a neutral proposition given some set of facts. It is impossible to debate some one particular specific fact without taking into account the environment and motive for this fact coming to light - the hypothesis that there was massive vote fraud in the 2004 presidential election.
1 - The exit polling data were wrong.
2 - Why are these anecdotes irrelevant, the convictions for slashing Republican tires irrelevent, but unsubstantiated claims (anecdotes) of minority disenfranchisement are somehow relevant? Because they come from Conyers?


My skepticism is only partisan in that we are discussing a partisan situation. Your academic dispassion, feigned or otherwise, is not shared by anyone putting forth the same argument as yourself, so it's hard to take on its face. God bless you for supporting the Washington inquiry. I would support more investigations if they weren't turned into sideshows by the politicians involved.


I see nowhere in the Constitution where they have the right to supoena data from a polling firm to help support the theories of other pollsters.


"When did you stop beating your wife" comes to mind. I have a theory, a hypothesis. Congress must therefore supoena data so I may fully investigate. Pay no attention to the partisan rancor behind the curtain.


You're reaching here. We're comparing apples to apples - presidential exit poll results vs. presidential election results. The words "congress" and "senate" did not appear in the uscountvotes document, so I'm not sure which numbers you're referring to here.

And I should trust CNN why?

They changed their exit poll results for Ohio. After the polls were closed! Luckily, someone got a screenshot:

Before:

http://www.democraticunderground.com/image...poll1.gif?IMAGE

After:

http://www.democraticunderground.com/image...poll2.gif?IMAGE

Notice the difference between voters polled is 57. If all 57 votes went to Bush, according to the revised exit poll, it still would not account for the percentage differences when the total number of people polled (with the 57 added) was 2,020.

Bush gained six percentage points thanks to these last 57 people polled, and Kerry lost seven percentage points thanks to the same 57.

But the percentage increase between 1963 and 2020 is just under 3 percentage points. I'll be generous and call it three. So Bush could not possibly have gotten a jump of six points, with Kerry losing seven! It doesn't add up!

Interestingly, Rush Limbaugh gave a little pep talk to his discouraged listeners about the exit polls. He told his election day audience (I was listening because it's good to know your enemy), that the exit polls would favor Kerry during the day, but toward the end, the exit pollsters would alter their results so as not to look silly like they did in 2000!

Now how did Rush know this would happen?

(I hope someone else on this board heard him say this, I haven't been able to find a trace of his comments online anywhere!)

Well, except on the Freeper board:

http://www.freerepublic.com/focus/f-news/1...s?q=1&&page=101




Supposedly CNN switched its final tally in Florida, too, with a similar mathematical error, but I haven't seen screen shots for that one.

carlitoswhey, I am pleased to see that you are now admitting the obvious: neither you nor I is in any position to determine the overall reliability of this study. We both look it over, find some things that are interesting, but the bottom line is, this is pretty heavy statistical analysis and we can't decide. You did take some shots at it:

No, their central piont is that there exists no good explanation for the discrepancies. They point to electoral fraud as one explanation that has not yet been discounted. It could also have been the aurora borealis, but the electoral fraud hypothesis seems more plausible than the aurora borealis hypothesis. They do not assume the electoral fraud theory, they conclude that it's left standing after they've shot everything else down.


You're right on this point: they have not demonstrated the "high plausibility" of the electoral fraud hypothesis. They have knocked just about every other hypothesis out of the picture, and so the electoral fraud hypothesis starts to look more interesting, but their study itself does nothing to advance the electoral fraud hypothesis.


That's a conclusion, not an analysis.


I have made no claims with respect to the relevance of these anecdotes; my observation is that it is impossible to draw useful conclusions about the election as a whole from these events.


Um, so you're only partisan when considering partisan issues, and you're completely nonpartisan when discussion nonpartisan issues? While I am relieved to learn that you do not consider the words of Tom DeLay while choosing a brand of table sugar, I would hope that partisanship would not color your judgment on any issue.


Well, tell that to the others when they speak up. But for now, let's stick with the issue here.


On this point, I agree that there is a basis for reservations about investigations. If an investigation could be carried out far away from Washington by nonpartisan experts, we might be able to accomplish something. But doing it the same way that the Clinton investigation was carried out is simply too divisive to accomplish anything worthwhile.


If they're investigating a possible crime committed on a national level, they certainly do.


If the hypothesis had not already been demonstrated to have some substance by a great deal of effort, your point would be well taken. But the fact is, the data have been squeezed really hard, they do show something suspicious, and releasing the raw data would clear matters up. Moreover, this raw data has no proprietary value to Edison-Mitofsky at this time. In the past, they have released such information to the public domain after six months. (I still can't find the documentation on this point, though). The only reason I can imagine for E/M to continue holding onto it is to preserve their reputation from the stains it might suffer should an analysis of the raw data reveal major bias in their treatment of the data.


Those numbers appeared in some of the many other studies. I'll try to find them. If I can't, I'll retract the point.

Final polls from RealClearPolitics.




History clearly shows the late pre-election polls are superior to the exit polls. The election results were no surprise to anyone who followed the pre-election polling. CNN also declared Bush the winner. Once again, pre-election polls gave the election to Bush and so did the, properly interpreted by the Ted Turner Network, exit polls. Yes, you can find a lot of bloggers on the internet arguing differently.

I'd like to see the statistics on that. Are there any?

There is an empty wine bottle on my desk that I think we purchased in February, so it is questionable if I am as sober as you're asking me to be...

If we Democrats could manage to enact a swing to say a 2/3 - 3/4 majority in both houses next year, perhaps we could persuade Congress to look for fraud in last years election.

The problem I see with using exit polls to determine how people voted though, is that we would have to factor in the people interviewed in an exit poll who were too embarassed to admit they voted for George W. Bush. I know from personal experience that when my parents left the voting booth, they would only say that they had cast a secret ballot, and I never heard them once state how they voted.

Should Congress subpoena the Edison-Mitosfky raw data so that it can be analysed to (possibly) clear up the exit poll question?

Isn't the data available for any organization to analyze? If not, then subpoena the data and put it into the public domain. Then get an army of university stat profs and students to poor over it, worldwide, using grid computing to crunch the numbers (if needed). The stat profs would have fun (might even publish!), and the students would learn with a real-world, meaningful problem.

I don't think Congress is sane enough to possibly select a non-partisan team for the analysis. This needs to go up into the ivory tower, where tough intellects can roll around with it.

And what if the conclusion becomes that fraud happened? Fire the Bush people and install Kerry's? I don't think that would do anything positive for the country. No, we want to carry on with business as usual, only more aware of what to look for in future elections.

On the other hand, the studies might show no significant correlation between exit polls and election results. This looks to me to be intuitively obvious, but oh what the heck, run the math.

I personally would be tempted to fib to exit pollsters. Hey, what business of yours is this, and how much is it worth? If nothing, then that's what you get in return.

Should a federal-level investigation be initiated into possible electoral fraud? Particularly, how much investigative money does this controversy deserve?

In general, fraud election investigation is worth the money just to help keep things as honest as they can get in a democracy (oh, put away the republic club). Voting fraud in democracies is an inherent problem that needs to be constantly investigated. I'm not sure how much money this would cost, but I like the idea of non-voting professionals taking up the cause. I'd pay the pros well for giving up their right to vote in favor of better democracy. Where do I send my resume? Do I get a badge and black suit?

Here's one obvious test of fraud: Did any Democrats win in the last election? They sure did, taking over Colorado's legislature and sending one Demo senator and one Demo rep (brothers at that) to Washington. The tyranny of Republican control over the past 40 years was broken in Colorado, a so-called red state. That indicates to me that the voting was at least honest enough to allow for this surprise outcome.

However, I am now worried that the Repubs will be trying to influence future state and local elections in illegal ways. We need to be vigilant, as this has happened before -- from both parties if you look back far enough. Fraud is built into our system just by the fact that human nature allows for it.

Heh, I suppose a Republican might think that the Demos stole the elections for the state legislature. Ah well, walking a mile in someone else's moccasins, priceless.


Sidebar: The state Demos are performing well, using mature and professional approaches to real problems, such as a busted state budget. There's value in a new majority party having to prove itself.

I ask all concerned to address this highly partisan issue with care and sobriety.

Okay, I'm carefully avoiding the statistical badlands and soberly sitting on a verdant hill, overlooking the vast plains, strumming my guitar and thinking, "Bottle of wine, fruit of the vine . . . ." Deer and elk munch the spring grasses down below. It's a happy place

Yes, there are. . Of course the final pre-election polls and the exit polls are available. According to the "unmassaged" exit polls Dukakis won in 1988. Of course, the final pre-election polls in 1988, just like in 2004, showed the true winner.

The only stolen Presidential election, since 1876, was in 1960.

Would you mind actually presenting those statistics? What I'd like to see is a list of the last N elections, and the representative values of late pre-election polls, along with the exit poll numbers. Do you really have such statistics?

I have already presented statistics, since 1988, that show according to "unmassaged" exit poll numbers that the Democratic candidates vote percentage was overstated by exit polls. Being 47 myseld I quite well remember that the final pre-election polls showed a victory for Bush in "88 and were much closer, to the final result, than the exit polls in '92 and '96. In 2000 the final pre-election polls were off in Bush's favor, which of course indicates massive voter fraud favoring Gore.

The statistics you presented were exit poll statistics for those elections. They say nothing to support your claim that "history shows" that pre-election polls are more accurate than exit polls. Do you have any evidence to support that claim?

Actually, when I buy table sugar, I think of the cozy relationship our Congress has with the Corn Lobby and Florida sugar growers and its effect on the price I pay (and the taste of Coke Classic). A topic for another day.


Yes. A crime committed by democratic/republican/independent precinct captains and democratic/republican/independent county officials and democratic/republican/independent secretaries-of-state and democratic/republican/independent election judges and democratic/republican/independent judges and ... the entire argument is too ludicrous to make, really. For vote tallies to have gone from Kerry to Bush across every precinct in the country is impossible. You may as well say Bush caused the Tsunami. Oh, wait...


As Amlord noted, the exit poll is a subset of the data. The election results are the complete set, and they are all counted and we have the results. To say that a subset of the data was wrong, therefore the election results must be wrong, is, well, wrong. Is voting perfect - no. Were there errors with machines, ballots, people - yes. Was there a national conspiracy, including the hacking of Diebolt machines and a sinister plot to suppress the black vote - no. There is, however, a psychology of victimhood and denial on the losing side of the past 2 presidential elections that is feeding on these silos of suggestive data. There is no credible explanation to link all of these little silos together other than the "grand conspiracy" theory of Bush somehow rigging the election, therefore there was a grand conspiracy.

There is a hard core activist constituency (note - not naming anyone on this board!) who just cannot believe that this monkey, this buffoon, this Hitler, this (to use the words of Janeanne Garofolo) "unelectable candidate" actually won. The fact that he won is a prima facie case that something sinister is afoot. Therefore, let's Lexis-Nexis what the Diebolt guy said, let's take Rush Limbaugh's partisan cheerleading as some sign of inside knowledge, let's pretend that heavy turnout and mediocre county government didn't cause long lines in inner-city Cleveland, but rather that Karl Rove did.

You can choose to look at releasing this data as some individual, discrete, theoretical debate topic, and that's fine for this board. I choose to look at the big picture, and think that we are doing a disservice in chasing shadows.

Indeed so. However, the hypothesis under consideration does not require electoral fraud in every precinct; indeed, it does not require electoral fraud in any precinct. One possible mechanism is vote tampering at the county level -- and even then, only a few score counties need have been tampered with to obtain the counts that were reported.


I didn't.


Indeed it is wrong. However, that's not what study says. What the study says is that there are statistically significant discrepancies between the exit poll data and the count data, and that at this time there is no plausible hypothesis other than electoral fraud that can explain the discrepancy.


The study does not claim that there was a national conspiracy. It claims that there are statistically significant discrepancies between the exit poll data and the count data, and that at this time there is no plausible hypothesis other than electoral fraud that can explain the discrepancy.



This is a classic ad hominem argument, and must be rejected as such. Let us focus on the logically derived results of the study: that there are statistically significant discrepancies between the exit poll data and the count data, and that at this time there is no plausible hypothesis other than electoral fraud that can explain the discrepancy.

You are inferring motive here. Does it? If so...is the problem with the process, or is their issue with the results? If it is indeed process related....why not point to any flaws in the process? No, this study only has an issue with the results. Bush was elected, and they didn't like it. Pretty much ditto for all the other studies cited here...UCB? Come on! Do I even need to begin to discuss their motivation? You want some consensus on their being a problem? Start by coming up with a non-partisan study that indicates there is one. Second, have it be process focused. Otherwise, we just have chicken little running around screaming that the sky is falling. I have three letters that solve that problem...B...B....Q



OK...are you misreading what I've said, or am I not a conservative?

+-->

I'll admit I haven't delved into the study. As you can tell from my posts, I see no reason to. I have fundamental issues with any polling used to predict/explain election results...regardless of the outcome. So, I don't think statistics of poll results have any reliability to begin with. As I said previously:



Why waste time analyzing data from known biased sources that have a readily available explanation that is neither incriminating nor has been disproved...particularly when the known biased study doesn't offer anything in the way of process flaws that could have led to their supposed discrepancies? Frankly, I'm tired of this whole 'they stole the election' thing. Particularly considering the history of the party making the accusations. In short, they've spent decades destroying their own credibility on the issue...so, to suddenly demand to have it, without even pointing to any mechanisms through which this great conspiracy was supposed to have enacted its evil plan....well, I'm sorry. I'm not going to waste my time with it.

You want me to buy into this study....provide me with any reason not to suspect its data. Fundamental rule of programming...garbage in, garbage out. People don't vote the way they say they will/did to pollsters. Therefore, polling data is pretty meaningless. End of story. Better yet...show me something in the process that could/should be modified to eliminate any concerns. I'll be behind that, and we can skip the whole statistical thing completely.

I'm agreeing with Hobbes above. I stipulate that the discrepancy from this 2004 exit poll data may never be reconciled with the election results. I then ask you - So what? It's an exit poll. It's usefulness is over. An exit poll is to predict the results and look at demographic and motivational data of the voters. That's all it's good for. We have the results. The election is over. I'd rather that everyone channel their efforts towards world peace or literacy or something useful.

That's because the data does not address mechanisms for cheating. It addresses only the results -- and the results indicate a discrepancy.


Might I suggest that you are guilty of precisely the same sin: inasmuch as you are happy with the result, God Himself couldn't convince you that the election was fraudulent.


The UCB study is not part of this topic. This topic is about the USCV study. You bring in the USB study only to deliberately cloud the issue.


The only evidence you have that this study is not non-partisan is that you disagree with its results. You are offering an ad hominen argument. Surely you know that ad hominem arguments are beneath consideration.


I cannot see how I misread your statement. You have certainly "rejected the suggestion of further investigation" into the possibility of electoral fraud. You have not read the study. It would seem that my observation fits you to a "T".


You don't even know what the study says. You don't know what methodology it used. You don't know how it addresses the problems you cite. You don't know anything about this study other than the fact that it produced results that you find unacceptable. Have you no respect for rationalism? Have you no sense of duty to the truth? Are you willing to dismiss any fact, reject any data point, deny any evidence that contradicts your political beliefs? Have you abandoned all sense of intellectual integrity? I ask you, is it possible that any evidence could convince you in this regard? If I were to produce a High Definition TV recording in stereo sound of Karl Rove convening a meeting of Republican operatives and describing in detail his plans to steal the election, would you not dismiss it as an obvious fake?



First, "bias" is not the correct term. The exit poll data constitutes a measurement. Like every measurement, it has its flaws. That's why we apply a large number of correcting factors and signficance tests to the measurement. The end result of all these calculations is a statistically significant If you have an objection to the study's methodology, by all means make it to the study's authors. However, I'm certain that, if you tell them that "all exit polls are crap", they won't take you very seriously.


The party making the accusations is USCV and it is less than a year old. If you wish to discuss something other than their work, please start a topic on that. This topic is about the USCV study. You are deliberately obscuring the issues raised by this study by bringing in outside issues.


This factor is addressed in exit polling methodology. The data is not garbage, it is data. Every bit of data has some degree of information content. Some has more, some has less. It is the task of statistical analysis to extract the useful information content from a mass of data. That's what has been done here. To dismiss data because it is imperfect is to deny the entire body of science, because all scientific data is imperfect.

Therein lies the flaw. When results indicate something that has no basis for coming about...then the results themselves are what is in question.



You can suggest it all you want...but that doesn't make it true. I have indicated numerous times I am very open to discussing improvements in the process.



Review the thread...I'm not the one who brought it up.



I do believe we have a Fruedian slip, here.... . As for being beneath consideration...ditto for studies using questionable data, performed by biased sources, with no attempts made to explain how the results came about, and with the data withheld from further scrutiny.



Only to someone with a clear bias towards the study. I have done nothing of the sort, as your own quote of mine shows.

Consider the following. There was a time (don't know if this still holds) in which the height of the Secretary of Defense was perfectly inversely correlated with our country being at war, with short SoD's always leading us into war. Statistically perfectly correlated. Now, would you say that this would call for a thorough investigation into why these short people are driving us to war? Or would you say, 'hey, wait a minute, that doesn't really make any sense, unless you can offer any mechanism through which these results are obtained." I would hope the latter...if the former, then, fine, by all means, investigate away. But that doesn't mean the rest of us should also waste our time.



Interesting that, even though you claim to have read the study, nowhere in this rant do you offer anything to refute what I said. I am also curious as to your objective here...are you hoping to taunt me into analyzing the study? If so....[claxon sound], thanks for playing, but no luck. Perhaps, as someone who says they have analyzed the study, if you gave me a little mopre info on how they addressed my issues, and a little less info on your take on my intellectual integrity? Just a thought....

It might also behoove you to skip inferring something about someone's political beliefs when you really have no information about what those beliefs are.



Oh, I see...'large number of correcting factors.' Yes, I feel muuucchhhh better now

I saw a mathematical analysis which seeked to show conclusively that Earth was the only populated planet. In the analysis, they calculated all the planets that could potentially hold life...and divided that by the same number (their 'correcting factor')....magicallly producing the number one! So forgive me if I am skeptical regarding 'correcting factors.'



How is it addressed? Without knowing how people 'really' voted, there isn't any way to isolate any factors indicating the authenticity of their responses. If the authors of this study claim to have such knowledge, then, of course, they're violating election laws, and should go directly to jail, do not pass Go, do not collect $200. If they don't have such knowledge...they have no basis for any causal 'correcting factors'...the best they could do is compose more statistical correlations with other suspect studies based on the same flawed data. Answer me this simple question. If you ask people a set of questions, and they reserve the right to basically lie to any of them they choose...as do all the other participants in the study...and you have no way of knowing if they lied or not on any particular question...and further no mechanism through which the supposed results came about is offered, and the data for the study is kept safely locked away to prevent any corroborating studies....why then would I assign any validity to the study?

You are assuming that there is "no basis for [the results] to come about". That assumption is false. There are many mechanisms for the results: electoral fraud is one. You begin with the assumption that electoral fraud is impossible; you therefore conclude that any study suggesting otherwise must by your definition be incorrect. This is not logical.


You are changing the subject, which is the study itself. You have adamantly refused to consider the study, using your presuppositions as justification for closing your mind to the results.


As I said before, all data is imperfect. The claim that the imperfection of data nullifies the result arises from a fundamental failure to understand the nature of statistical analysis.


You have done nothing to demonstrate the claim of bias other than your objection to the result.


These are lies. Alternative hypotheses have been offered and the data is publicly available.


This is an elementary error whose flaw is obvious to the most casual observer: the confusion of correlation with causation in the absence of causal mechanisms. A freshman stat student would be flunked for making such an error. This study was carried out by a group of statistics and poli sci professors. Your suggestion that statistics professors would make an error that they would not tolerate in a freshman defies belief.


I have read every release of the study since its first draft. I have read all the published objections to it, and I have followed its evolution as it addressed each and every objection.


Your only substantial objection that I have not yet addressed is your claim that voters deliberately lied to exit pollsters in order to skew the results. In the first place, there is no reason to believe that Kerry voters would be more or less inclined to lie than Bush voters. Thus, even if lying were a signficant factor, the positive lies would cancel out the negative lies. In the second place, if lying were a significant factor, then precincts that reported high percentages of votes for Bush would show lower percentages of exit poll results for Bush, and vice versa for Kerry. This discrepancy can be used to produce a correction factor, which, it turns out, is microscopic. In other words, lying does not happen often enough to affect the results of the study.


Please see preceding paragraph and retract your claim about lying voters.


You're right, I have nothing to go on other than your own assertion that you are conservative.


Here are some of the correcting factors that went into the E/M exit polls. To save space and to remain within copyright limits, I have eliminated the explanatory text for some of the factors:




If you wish to debate that calcuation, please feel free to start a topic on it. I suggest that you remain on topic and discuss these calculations, not somebody else's.


This belies a misunderstanding of statistical methods. When you have huge masses of data, you can cross-connect that data to extract pertinent information. Here is a simple example:

Previous polls have established that, nationwide, women prefer candidate A to candidate B by 54% to 46%. Men prefer candidate B to candidate A by 58% to 42%. Now you carry out your exit polls and you find that in precinct #873, 27 women were polled by only 21 men were polled. This means that your results are biased in favor of candidate A. You therefore adjust your exit poll results by scaling down the women's votes and scaling up the men's votes in double proportion to the ratio of their sample population and the ratio of established candidate preferences. This is a highly simplified example of how this kind of thing is done -- and it is done for all manner of factors: age, gender, race, and so forth.


Please see preceding paragraph explaining how the issue of deception is resolved.


Please see preceding paragraph referring to electoral fraud as a possible mechanism.


That is a lie. The data for the study was published in the original Edison/Mitofsky report

Why are we discussing the validity of a poll over an official election result?

There were exit pollers outside of my voting place and I was asked and refused to answer, and yes I voted Bush.

Here is my biggest take on the whole exit polling flop.

1. Exit polling normally takes place earlier in the day(likely before 5pm in a given time zone). Who are the majority of people voting before 5pm. Retired elderly, spouses that don't work(mostly women), unemployed. That segment of the population was much more for Kerry than for Bush. Would you agree with this?

2. Like the study says. Bush voters did not participate in the poll as often as Kerry voters did, that should tell you right there the data is skewed.

3. It was a sampling of 120 precincts... How many precincts are there nation wide??

Here is a good example I can come up with. In 1955 the Philadelphia Mint produced what is one of the most popular doubled die cents in history. The story behind it was mint officials caught the doubling effect during a midnight shift before they were supposed to be shipped out. But there was a shortage of coins because of recent Hurricane Dianne. Rather than scrap and melting an entire shifts production they released the doubled die cents into circulation for the area that needed them. Now if you were to take a "coin poll" of the bags of cents from that time when they were released. Would that give you an accurate view of the entire quality control of the Philadelphia mint?

See, this is where your bias (as a statistician, not as a partisan) is shown.

You have yet to address how Data set A (the exit poll) data is perfectly representative of Data Set B (the actual vote results). Massage Data Set A all you want, it still does not (and will never) perfectly represent Data Set B.

Data Set A's usefulness is in providing a quick, probable outcome for Data Set B. It is not infallible.

So when Data Set A fails to predict Data Set B, somehow Data Set B is questionable.

Remember that Data Set A is much smaller (by orders of magnitude) than Data Set B and the data point are not randomly selected, but statistically selected (X proportion of males, Y proportion of females, etc.) The assumptions of those setting up the poll, the subsequent "correction factors", and the subsequent results are all questionable. And all of that has nothing to do Data Set B.

The logic here is questionable at best.

From the cited study......



Doing a quick statistical analysis, the word "estimate" or a variation thereof appears in this citation fully SIXTEEN times. Former Senator Hollings might comment on this by saying, "Why we got a whole lot a estimatin' goin' on here." And now we want Congress to investigate? You have to be kidding.......

Should Congress subpoena the Edison-Mitosfky raw data so that it can be analysed to (possibly) clear up the exit poll question?

No, not unless Congress paid for the poll and I don't think it did. Otherwise, it's none of Congress' business why this polling firm screwed up. That's their problem.


Should a federal-level investigation be initiated into possible electoral fraud? Particularly, how much investigative money does this controversy deserve?

If there was true electoral fraud then of course it should be investigated. A messed up exit poll does not fraud make. People don't get elected to office with estimates.

We are discussing an analysis of the discrepancies between exit polls and official election results, a considerably more sophisticated concept.


That's one datum among tens of thousands.


Yes, this factor was taken into account and corrected for.


Yes, this factor was taken into account and corrected for.


And what error margin does your calculation show this to generate?


Here we go again with another ridiculous example offered to attack a study without actually addressing the study itself. Gee, let me tell you a story about a conservative whom I knew once who was wrong. Doesn't that mean that you must be wrong?


You got me. I confess it for all the world to know: I am biased towards statistics. I have been known to consort with card-carrying statistics. Oh, the shame of it all!


It isn't perfectly representative of Data Set B and nobody has ever claimed that it is. Statistical analysis has all manner of techniques for dealing with sample sets that are not perfectly representative of other sets. One of these is called "factor analysis". These techniques make it possible to correct for misrepresentation.


The fact that you don't understand the mathematics at work here does not justify your questioning it.


Yes, they are estimates. They are estimates of complex functions that behave like linear functions over short ranges. They also provide estimates of how much in error the estimates are, which in turn allows us to say how reliable the overall calculation is.

Let me step back and lecture you folks on some basic principles.

1. All data has errors in it. Every last measurement we can ever take has some uncertainty associated with it, at the very least because of the Uncertainty Principle, but more often because of limitations in our measuring devices. Scientists have been aware of the existence of such error from the very beginning. Over the centuries, increasingly sophisticated ways have been developed to account for this error, and to determine just how much it affects any particular calculation. The mathematics of statistical analysis is the result. It permits us to extract information about reality using many data points, each with its own error.

2. Statistics is a reliable field. Done badly, it fails; done properly, it works. The best way to determine whether any given statistical analysis is done properly is to have it done by experienced statisticians and then submit the work to public review. This has been done with this report. To brush off statistics as some sort of black magic is the act of an uneducated person.

3. It seems that everybody here wants to play amateur statistician, offering all manner of sophomoric criticisms of the study. C'mon, guys, you are out of your league. You sound like a 12-year old interrupting a neurosurgeon to ask, "Didja remember to take his temperature?" or "Are you sure he won't bleed to death -- that sure is a lot of blood!"

I have explained the many errors in your criticisms, but you just aren't paying attention; you keep denying the basic issue: that a careful study demonstrates serious discrepancies between the exit polls and the vote counts. To re-use an image from earlier, the clouds could part, and God Himself could descend from Heaven surrounded by brilliant light and singing angels, and He could say, "There's a significant discrepancy between the exit poll data and the vote counts", and you'd still refuse to accept it.

That said, let me now remind you that there are some valid criticisms of the study; I pointed them out in the very first post because I have the intellectual integrity to recognize the complexity of such analyses. This study does not constitute proof of electoral fraud. It does not contain a smoking gun -- metaphorically speaking, it has only the silhouette of a smoking gun. But this study raises serious questions. I would like to answer those questions. I would be genuinely relieved if a proper investigation put to rest any fears of electoral fraud, and I think that our system would be better to have that result. I reject the notion that this is an issue best swept under the rug. If you folks are mad about all those conspiracy theories, then let's do something about them rather than idly complain about them. Let's take the strongest of the cases and investigate it. But your refusal to contemplate an investigation under these circumstances strongly suggests that you are afraid of what such an investigation might reveal.

I can only conclude from the nature of the attacks presented here that no amount of logic can possibly convince some of you. Those who are open-minded enough to entertain the hypothesis might derive benefit from further discussions. The close-minded ones I shall henceforth ignore.

Erasmussimo,

Sigh...you win. In fact, I'll let the secret out of the bag (sorry, Aquilla and Amlord). Yes, the Republican party did indeed carry out the world's largest conspiracy, and stole the previous election. I attended some of the planning meetings. We did it so well, in fact, that despite intense scrutiny, no one has been able to uncover the means through which we carried out this great deed. Imagine the skill this took...to convince tens or hundreds of thousands of people throughout the country to carry out electoral fraud, and leaving essentially no trace. I don't mind letting the cat out of the bag...because there's no reason not to. Our methods are infallible. We are already putting in place plans to undermine the upcoming elections. In fact, the voting doesn't even need to take place, as we can tell you the results ahead of time...we just haven't decided yet what the final margin will be. I'm pushing for something over-the-top, to avoid the confusion of the past election...but I'm not the one in charge (no, I won't reveal his name), so, we'll just have to wait and see. I'm sure there will be plenty of statistical anomolies (we can't cover up everything, after all)...but as long as our mechanisms are undiscovered, nothing can be done...so we're pretty safe.

So, while I would love to carry on this discourse, I can't...I have meetings to attend, discussing plans for the next great electoral heist. Besides, I see no need for my participation here...you have done a wonderful job of inferring all my thoughts and beliefs for me anyway (a habit, by the way, which does nothing to lend credibility to either your arguments, or the methodology you seek to defend).

I would leave you with one question, to continue Amlord's excellent point...at which point in your statistical studies did they ever teach you to prefer the sample over the entire data set?

Carry on, then...after my meetings, I might rejoin this conversation if I could be so blessed as to be able to inject my own thoughts, and not have them espoused for me. Oh, that's right....I'm out of my league. Never mind....I'll just catch the rest of the lecture off-line. I'm off to go play catch with the other 12 year olds.

Sarcasm aside, Erasmussimo...this is clearly a topic you are passionate about...but belittling the responders isn't getting you where you want to go, IMHO. Clearly you would like to discuss the statistics of the study....if I weren't currently so busy, I would indulge you. I do in fact have a fairly extensive background in statistics myself (although its been a while), including the pros and cons of applying such methods (and others) in complex social or business situations. But being called a sophmoric 12 year old, and lectured on-line isn't doing anything to create any incentive for me. So, if you're interested, we should probably cool off for a day or so, and try this again later.

I will echo what Amlord and Hobbes have both now said...

At which point in your statistical studies did they ever teach you to prefer the sample over the entire data set?

A statistician doesn't "prefer" one set of data over another. One simply takes each one for what it's worth. Your statement presumes that the vote count constitutes the ultimate reality, but that assumption is not valid, because it denies any possibility of electoral fraud, counting errors, lost ballots, etc. We know beyond all doubt that errors occur in the official count. We seldom worry so long as those errors appear to have had no effect on the outcome of the election. But where they might have had an effect, concern on our part is warranted.

Both exit polls and vote counts are measuring devices which purport to measure the will of the people. The vote count is almost always the more reliable, but the greater reliability of the vote count does not make it definitive -- it could still be in error.

Here's a simple analogy to make the point. Let us compare the accuracy of the speedometer in your car with the accuracy of a policeman's radar gun. We all know that the speedometer is only good to plus or minus 5 mph, maybe even 10 mph. We also know that the policeman's radar gun is good to better than plus or minus 1 mph. So if your speedometer says 35 mph and the policeman gives you a ticket for going 45 mph, well, that's probably error on the part of your speedometer.

But what if your speedometer said 35 mph and the policeman's radar gun said 85 mph? That's a really serious discrepancy! That doesn't mean that your speedometer is right or wrong, nor does it mean that the policeman's radar gun is right or wrong. It means that there's a serious discrepancy between the two. If you were a judge, and the speeder has proof that his speedometer read 35 mph, and the policeman has proof that his radar gun read 85 mph, wouldn't you decide that some further investigation was called for?