I have to admit, it's hard to admit how any reasonable person could object to a full investigation.
But, actually, I don't care very much. I'm not saying that it's inappropriate to care, just reporting my own psychic state.
Basically all the exit poll data, including the 2004 exit poll data, are archived with the Inter-university Consortium for Political and Social Research. The 2004 exit poll data were also available for unrestricted public download for almost a year. These data contain one record per questionnaire that the interviewers phoned in, with answers (if there were any) to all the questions on the questionnaire, including a bunch of demographic questions. The records have arbitrary precinct IDs, intended not to be matchable to physical precincts. (Nevertheless, some enterprising DUers dove in and apparently matched many of the Ohio exit poll precincts.)
It doesn't really matter whether you personally want to know the demographics. This isn't about you. That's how the data are collected, that's how they're used by many analysts (media and eventually academic), and that's how they're released. And that offers enough information on individual respondents to compromise the privacy of at least some of them.
Now, lots of people are saying that what they really want to know is the precinct totals. Fine, but the individual data have already been released. Some people say that this time, the pollsters should have withheld the data that they always release, so they could release the aggregate results with precinct identifiers instead. Or some think the pollsters should have just let it all hang out because, after all, these data are so important.
Except that no one seems able to explain why these data are so important, and there's reason to think that they actually aren't. One reason is simple logic: indirect evidence on maybe 50 precincts in Ohio is trumped by election data for all the precincts in Ohio. Another is empirical analysis: we know from ESI that in Ohio, there is no significant relationship between red shift and change in Bush vote share 2000-2004, which is a strong indication that the precincts with big red shifts had bad poll results. If that weren't the case, then the change in vote shares would be much smokier than the exit poll result in itself. Mebane and Herron actually looked at the election data. The people who trumpet the urgent need to release precinct identifiers, not so much. Interesting.
First of all, let me mention that in the 1992 presidential election, the exit poll discrepancy was almost as large as in 2004 (the average "within precinct error" was 5.0 points in 1992, 6.5 points in 2004). There were smaller but significant discrepancies in 2000, 1996, 1988, and probably earlier years (it is hard to tell in retrospect). So, that's another reason to be puzzled that you say the exit poll data were "dependable" before 2004. You seem to mean that you think they previously matched the official returns; in fact, they didn't.
Like Febble, I wouldn't say that exit polls "have no validity." If I thought they had no validity, I would ignore them entirely. The first step in analysis is to get past a false binary about whether the exit polls are valid or invalid, Right or Wrong. (Letting go of false binaries is the first step in lots of things, isn't it?)
I don't entirely agree with the way Febble put her last point, but it's really hard to talk about these issues without getting into the specifics. I will leave that alone. I don't think she and I really disagree about it.
It's interesting that you say you found Freeman's comments "reassuring," and also allude to unnamed "many others" who agree with him. Forgive me, but I don't think we should set out to be reassured, one way or another. If you want to defer to expert opinion, then I should think it would make sense to go looking for experts -- not for the subset of experts who say what you may be predisposed to believe on other grounds. If you want to assess evidence, then I should think it would make sense to scrutinize Freeman's arguments and see what evidence of exit poll accuracy he offers.
Following the first path of looking for experts, I invite anyone to look through political science journals for arguments -- based on any combination of evidence -- that Kerry got more votes than Bush in 2004. It may be possible to explain away the dearth -- utter absence, as far as I know -- of such articles. For instance, one might conjecture that political scientists are all Republicans (although this is false), or are afraid of losing their jobs if they tell the truth (although many are tenured), or are reluctant to take such a controversial position (although many argued that Bush should have lost in 2000). But if we're simply deferring to experts, you can probably find more experts to argue that the earth is under 10,000 years old than to argue that the exit polls were right. Both are, at best, minority positions.
But suppose we are assessing evidence, which I think is more useful.
The article to which you linked presents no evidence whatsoever about the accuracy of exit polls. (Maybe we are meant to infer that the exit polls were accurate in Ukraine, Georgia, and Serbia, but there is no way of knowing.) It doesn't even mention the large U.S. discrepancy in 1992, or the smaller significant discrepancies in other presdidential elections.
Instead, the article basically argues that exit polls should be accurate. Freeman draws an analogy between exit polls and measuring snowfall. The analogy does make a reasonable distinction between predictions and measurements. However, it doesn't take us very far, because voters, unlike snowflakes, decide whether they want to be measured! If Democrats are more willing to participate in exit polls than Republicans are, then changes in method might help or hurt, but there is no reason to assume that even perfect methods will yield unbiased results.
So we still need a way to judge whether the exit poll results or the official results were more accurate. As I suggested elsewhere, one approach is to consider which results made more sense given our other knowledge -- for instance, does it seem more likely that Kerry won New Hampshire by about a point and a half, or by 15 points? We can extend and formalize that question, but first we have to be willing to ask it.
Here is an interesting consideration. I've spent at least 20 minutes trying to make this response reasonably clear, accurate, informative, and possibly thought-provoking. Someone, somewhere is thinking, 'How does that evil OTOH find the time to write such posts? I bet he's paid to do it!' (I'm tempted to say, "If only!", although that isn't a job I want.) I could have dashed off a response in 30 seconds, and then someone would be thinking, 'Why is that evil OTOH so brusque and rude?' (It's not just me, of course; I happen to be the person writing this particular post.) Also, if I had said little about the exit polls, I would be accused of ignoring that evidence; since I've talked about the exit polls, I can be accused of ignoring other evidence; and if I start talking about other issues, I can be accused of being a know-it-all, of trying to poo-poo the possibility of election fraud, of demanding that people start over from scratch.... Even if I agree with folks on half-a-dozen points, I'll probably get clobbered when we move to the 7th. Sometimes it feels like playing a rigged carnival game.
So, I might conclude that there is really no point in talking with people who have made up their minds that the 2004 election was stolen. At most I can offer reminders that the rest of us exist -- in this case, that the OP doesn't represent what most citizens or experts believe.
I think it's fair if people want to believe that 2004 clearly was stolen and don't want to assess the evidence. We can't assess all the evidence for everything we believe, all the time. But I don't think it's fair if such people attack people like me for being interested in questions that they aren't interested in. And I think it could really be a problem if such people start making up stories to explain away the rest of us (they're ignorant, they're scared, they're corrupt...). We all make up stories about each other, too (stereotypes, etc.), but the more we rely on those stories, the more harm we can do.
I don't really know any more why I try to talk with people about 2004, except that I feel someone ought to.
It borders on the absurd to believe that Bush won reelection on the basis of this huge increase in big city votes and vote share from 2000 to 2004. We’re supposed to believe that he went from 2.3 million big city votes in 2000 to 5.4 million in 2004, a 153% increase.
As before, this thing we are supposedly supposed to believe is not borne out in the official returns. It could not be, since as Collins pointed out, the turnout increase in the big cities wasn't nearly large enough.
Meanwhile, Ruy Teixeira pointed out in an article of November 14, 2004 -- an article with which I assume you are already familiar -- that the official returns work out as follows at county level:
Starting with the most urban counties, those that are central counties of large (1 million or more) metro areas, Bush improved his vote margin by 2.4 percentage points (i.e., he narrowed his margin of loss to about 55-44)....
In fringe or exurban counties of these large metro areas, Bush improved his winning margin by 6.7 points (to 62-38)....
More important to Bush's vote gains were medium-sized metro areas (250,000 to a million in population), where he improved his winning margin by 3.5 points (about the national average)....
I recommend the rest, but that is enough. Bush's change in vote margin in the most urban counties was below the national average. This is why the so-called Urban Legend never did catch on: because it bears no resemblance to the official returns. People who are trying to understand the official returns -- right or wrong -- use the official returns whenever they can.
However in terms of the thesis in the "Urban Legend" all it means is that the effect noted in the story has been accentuated in the final numbers.
What "effect"? Once more, let's go to the tape:
The explanation of the Bush victory margin through the 66% big city increase evaporates in view of this data. 16% is not even close to 66%. The 66% did not happen.
The "66% big city increase" refers to the increase in big-city turnout (based on the NEP table), not to Bush's big-city vote count. (Go ahead and check me on this.) But since the "66% big city increase" preceded the appearance of a "Bush victory margin" that demanded an "explanation," obviously this argument cannot be right.
Reread that until it sinks in. Then, if you have a response to that, please respond to it. Kindly leave the goalposts where Collins placed them. It is not my fault that I am trying to address what Collins actually wrote.
You can read the post "upside down," but it will be easier like this:
Here is the table at 7:33 PM on election night:
And here is the table at 1:24 PM the next day:
In either order the big-city and rural shares of votes are the same within rounding error -- but seeing them in the right sequence makes it clearer how the pollsters increased the Bush vote shares in each category.
By the way, just using the 2000 size shares -- for instance, assuming that 9% of votes were cast in big cities and 23% in rural areas -- without altering anything else would have shaved Kerry's lead by about 2 points. That underscores the weirdness of arguing that the pollsters inflated the big-city vote in order to make Bush do better.
We don't call that "disagreement" in the United States. I have seen no actual disagreements; as far as I can see, you have tacitly conceded every point. The "Urban Legend" simply doesn't hold up, no matter what else one thinks happened in 2004.
Once more, from the Collins article that started it all:
The only way a Bush victory makes sense, given his failure in rural America, is the addition of millions of votes to the urban centers, the impossible phenomenon.... How did the NEP get it so wrong on the urban vote? Was it simple expediency? They had an election to report. They had that 3% problem to handle, you know, the Kerry 51%-48% victory at the end of the day’s polling. There was very little time to handle it. The urban magic that Charles Cook extolled as a sign of Bush campaign genius was invented. It came to be because it was the only way the poll could match the reported results....
But the evidence says that, in the table Collins refers to, the "addition of millions of votes to the urban centers" occurred before the table was reweighted -- while it still showed Kerry ahead. See this tabulation from just after 7:30 pm Eastern time (see pages 11-12). It looks like this:
Some folks have contested other folks' commitment to democracy, but no one has contested this table, which demonstrates that there is no relationship between the relative big-city vs. rural vote shares in the table and whether it shows Bush or Kerry to be ahead. For comparison, people can refer to the revised tab posted on your website here -- again, pages 11-12.
Both tables show the same proportion of big-city votes, the same proportion of rural votes -- in fact, all the same proportions except for rounding on the suburban share. The reweighting boosted Bush's vote share in all five categories; it didn't detectably alter the purported big-city turnout. (And, as I've noted, the exaggeration of big-city turnout doesn't "match the reported results" at all. It doesn't even appear to match the state exit poll tables. It's just plain wrong.)
Even the premise of Bush's "failure in rural America" depends, as far as I can tell, on the assumptions that (1) the decline in the rural vote is correct although the increase in the big-city vote is wrong (which verges on mathematical impossibility) and (2) somehow the NEP could only reweight the big-city numbers, not the rural numbers (was one of their abacus rods sticky?!). The whole darn argument is as "rhetorical" as one can imagine.
(Oh, the part about Charlie Cook extolling urban magic is "rhetorical," too.)
So, I don't see disagreement. I see refusal to correct demonstrable error. I can't "agree to disagree" in the absence of disagreement -- that would be mere mystification.
You can navigate my detailed response to TIA via the Table of Contents. If you intend to participate in informed debate, you might well spend a few minutes acquainting yourself with the arguments of people who disagree with you. Just a thought. Here's a summary response to the points you copied-and-pasted here:
1. Why not? (I've written a lot about exit poll accuracy, but there's no point in trying to respond to an argument that no one has made.)
2. Zogby's 5 pm calls weren't polls; Zogby's final vote projection showed Bush winning the popular vote (see link in 3). So, again, why not? And why would we cherry-pick one or two pollsters, anyway?
3. Harris's final phone poll also showed Bush winning the popular vote. (See here for final poll results.) TIA doesn't like that, so he uses Internet poll results instead. Most outfits, like PollingReport.com, didn't even report those results.
4. So what? There's a whole section of the FAQ on that. I can't make you read and think about it, but it's there. Anyway, in presidential races, the available evidence refutes this "rule."
5. That's the next section in the FAQ, and it simply isn't true. Ford and Reagan (in 1984) both did better than their final polling. As I say in the FAQ, TIA is welcome to argue that his rule is true except when it isn't, but then it isn't much of a rule.
6. That's the next next section in the FAQ. (Gee, what were the odds? Someone might get the impression that I've been rebutting TIA's arguments for years now!) If we have to predict vote shares from approval ratings, well, a best-fit line shows that an incumbent with an approval rating above 45 is likely to win. TIA's "50%" is a cherry-pick, since before W., we don't have any incumbents with ratings between 46 and 55.
7. Why? Palast could be right about that. Bush won in the official returns by 3 million votes.
8. Not a coincidence, misinformation. Measured by mean Within Precinct Error, the exit poll discrepancies were greater in precincts with lever machines (no paper trail, except that some may still print total votes) than in precincts using punch cards or optical scanners. See page 40 of the evaluation report. (If you want to talk about hand-count precincts, see here, pages 12-13.)
9. Again, why would someone have to believe that in order to believe that Bush won?
10. 99-1 odds according to whom?
11. Also an opponent who gained the benefits of incumbency and, at the time, roughly high-50s approval of his handling of terrorism (USA Today/Gallup had it at 60% just after the election, but the data are spotty; see here). If TIA has any evidence that Republican newspaper endorsements weigh more heavily than incumbency or national security, I'm happy to consider it.
12. This is basically a repetition, except even less defensible this time. TIA obviously hasn't supported the claim that "ALL professional pollsters agree that the undecided vote ALWAYS goes to the challenger." And the sources he does cite don't support it. This bizarre assertion alone should lead readers to doubt TIA's reliability as a source.
TIA and his acolytes are, as far as I know, the only people who assert that the national polls showed Kerry and Bush tied. It certainly doesn't look that way on PollingReport.com (see link in 3).
13. Actually, it doesn't matter whether you believe that or not. But Bush obviously didn't win Ohio by stealing votes on machines without paper trails, because relatively few voters used them. If TIA wants to present evidence that paperless DREs produced anomalous returns in 2004, by all means he should. Meanwhile, there is nothing to refute.
Posted by OnTheOtherHand in General Discussion (1/22-2007 thru 12/14/2010)
Wed Aug 22nd 2007, 09:37 PM
This "urban legend" story is remarkably thin. As far as I can tell, no one ever did think that Bush was propelled to victory in the big cities. Teixeira -- one of the few people to mention the table you've now eked maybe 11,000 words from -- actually wrote: "So put a big question mark by those exit poll spatial findings." You're lucky that Charlie Cook spent a few sentences on it as an afterthought, or else you would have even less by way of an "official narrative" to rebut.
I assume you already know, although you have unaccountably failed to emphasize, that the impossible big-city turnout increase you are ranting about appeared in the national exit poll table before the polls closed, while Kerry was still ahead there. See here, or if you don't trust that, here (page 11). If it weren't facially crazy to suppose that the pollsters tried to boost the Bush vote by shoveling people into big cities, here is the smoking-gun evidence that they didn't. That table was screwed up already. But you knew that, right?
So, you've presented strong evidence that this particular national exit poll table didn't even match the recorded vote, before or after the poll was reweighted. Why you think this buttresses your belief that the exit polls prove anything about election fraud is beyond me. I've seen lots of strange twists on 'the exit polls were wrong, which proves that the exit polls were right,' but this just might be the weirdest.
I tried to ask your publisher some questions about this, and he seemed to think that the official counts would show lots of extra Bush votes in the cities, even though your previous article basically demonstrated that that would be mathematically impossible.
People who thought they had 163 reasons to believe that Kerry won will now think they had 164. Even if they can't quite agree what this new reason is.
TIA's table presents four variables as the last four columns. I will reverse the order of the last two.
"Swing actual" (the change in Bush's two-party vote share) ranges from -4.7 to 5.5.
"Swing adjust" (what TIA thinks the swing ought to be) ranges from -1.1 to 8.6.
"Redshift actual" (what TIA thinks the redshift actually was?) ranges from -2.5 to 5.4. (Presumably this figure is based on someone's calculations from screen shots, which means that it incorporates pre-election polls.)
"Redshift adjust" (of which more below) ranges from 3.3 to 6.3.
TIA's two swing measures are highly correlated, but TIA's redshift measures are barely correlated. His "adjusted" redshift measure has a standard deviation one quarter that of the original measure. As far as anyone else in the world who has ever analyzed exit poll discrepancies is concerned, he might as well be using random numbers in his "adjusted" red shift measure. How did this happen?
TIA hasn't explained this "adjusted" red shift measure very clearly, but it is derived from Kerry's recorded two-party vote share. The assumption is, the more Democratic the state (in the returns), the larger the "adjusted" red shift: Kerry's vote share and "adjusted" red shift rise in lockstep. Ergo, "adjusted" red shift was largest in DC and smallest in Utah. If a fixed percentage of Kerry votes was changed to Bush votes in every state, this "measure" would indicate what a perfect exit poll "should have" found. The very low correlation between "adjusted" red shift and "actual" red shift should serve as a warning that something has gone wrong. The exit polls don't support the inference that vote shifting affected a fixed percentage of Kerry's votes, and neither does any other line of evidence I've ever seen.
So, "adjusted" red shift doesn't seem to have much to do with red shift, but in correlational analysis, it will give basically the same results as Kerry (or Bush) vote share.
Your response to my query in #36 appears to be that indeed I should just ignore the adjusted red shift, but maybe I am misreading. At any rate, it turns out that I can't just ignore adjusted red shift.
Anyone who wants can look along at the scatterplot to which you linked, <Image>. The plot purports to present "Adjusted 2000 Swing" on the X axis and "2004 Red-shift" on the Y axis. Which brings us to the questions that you refused to answer.
"glengarry, what is the range of 'adjusted swing' in the table in the text to which you link?"
Answer: adjusted swing in the table ranges from -1.1 to 8.6, as I noted above.
"What is the range of swing in the scatterplot?"
Answer: "adjusted 2000 swing" in the scatterplot ranges from 3.3 to 6.3.
In other words, "swing" in the scatterplot is actually TIA's "adjusted red shift" measure.
So, what TIA presents as a refutation of the non-correlation between swing and red-shift actually doesn't use swing at all.
Note, however, that TIA fails to report a correlation, R-squared, or P value for the relationship between "swing" in the scatterplot (which actually isn't swing) and red-shift. That relationship is not statistically significant, so even if TIA were using the right measure, his finding would be a non-finding.
Note further, however, that if one correlates either of TIA's swing measures with his actual red shift measure (as these appear in his table), the correlation is very slightly negative. Thus, TIA has replicated Febble's basic result -- an insignificant negative correlation between swing and shift -- using state-level data. This finding will be a useful confirmation of Febble's work, if he ever gets around to reporting it correctly.
Once we abandon your unfounded confidence about whom Gore 2000 and Bush 2000 voters voted for, we can't extrapolate vote shares from turnout.
But it's worse than that. We have evidence from all those exit polls that people tend to overstate having voted for the incumbent four years earlier -- and the 2000-04 NES panel indicates (not surprisingly) that the people who misremember having voted for Bush in 2000 tend to have voted for him in 2004. It is fairly straightforward to extrapolate that the Gore->Bush defection rate is higher than it appears, and that the Bush->Kerry defection rate is lower. I walk through these points in the FAQ. (And, again, my example here shows how misreport of past vote can influence the percentages.)
We might as well cut to TIA's so-called rebuttal of this point:
"Wow! That can only be described as a convoluted, faith-based set of hyper-hypotheticals."
OK, so far we have adjectival handwaving. No substantive criticism whatsoever. Actually, no evidence that he ever read what I wrote. I have no way of knowing. I only know that he has offered no rebuttal.
"And this is where we part company."
Well, I hope so. I do try to present actual arguments.
"For one to believe that over 7 million Democratic Gore voters would defect..."
Actually, TIA made that up. I never asserted that over 7 million Democratic Gore voters voted for Bush. I never even asserted that over 7 million Gore voters, period, voted for Bush, although I suppose it's possible.
According to the 2000 exit poll, about 30% of Gore's votes came from self-identified Republicans and Independents. It's no use pretending that they were all Democrats.
"...and vote for the most incompetent, dishonest pRresident (sic) in history (with a 48.5% approval rating)..."
Still with the approval rating argument?! Hey, TIA, please report the results of your own regression analysis: what approval rating is associated with a 50% two-party vote share?
"...who STOLE the 2000 election from them..."
At this point an empiricist might start looking for evidence that Gore voters were almost universally livid with Bush after the 2000 election (or whatever TIA intends to insinuate -- I can only guess at the argument, really). It may not have been apparent from the 57% approval / 25% disapproval Bush had in the Gallup Poll in early February 2001, but maybe they were just being polite at that point. (Obviously approval is not the same thing as belief that the outcome was fair, but then, that isn't the point. Presumably TIA is suggesting that Gore voters were unlikely to vote for Bush in 2004 because they were embittered by 2000; evidently some were more bitter than others.)
It's possible to inquire honestly into what Gore voters might have done in 2004, but not if we pretend that they were all Democratic loyalists.
"...is really a stretch -- a perfect example of faith-based naysayer fundamentalism."
In other words, he just doesn't believe me. So, who is being "faith-based" here? Does he have a criticism of my arguments, or doesn't he?
I had missed or forgotten Alvarez et al.'s study just before the 2006 election (fielded in late October). http://www.vote.caltech.edu/reports/2006-V...
I've been wanting to read a credible voter confidence study with more than one or two questions, although this one is certainly not very extensive.
One question read, "How confident are you that your ballot in the November of 2004 presidential contest between George Bush and John Kerry was counted as you intended? Would you say that you are very confident, somewhat confident, not too confident, or not at all confident?"
Among registered voters who are likely to vote, 75% responded that they were confident their votes were counted as intended during the 2004 election (55% were “very confident” and 20% were “somewhat confident”) and 22% responded as possessing a lack of confidence (12% were “not too confident” and 10% were “not at all confident”).
The percentage who are confident that other people's ballots were counted as intended could be lower. (Lest anyone be confused, I'm obviously not offering this as a measure of confidence in e-voting. It is what it is: a question about vote counting in 2004.) They note that in responding to a similar question in March 2005, only 11% expressed a lack of confidence (6% "not too," 5% "not at all").
Then there was a series of questions:
You may have heard discussion about the use of electronic “touchscreen or “direct recording electronic” voting machines that will be used in the November 2006 elections. I am going to read you some statements about electronic voting and want to know whether you agree or disagree with each statement or if you have no opinion. Some people say that electronic voting systems are:
• More accurate
• Increase the potential for fraud
• Prone to unintentional failures
• Make voting easier for people with disabilities.
Here is a summary of the responses, in the order presented by the analysts (the order in the survey was rotated) :
Pretty remarkably high "no opinion" figures, even considering that no opinion was explicitly offered as a choice. No 90%s here in any direction, obviously. I'm not drawing any strong inferences about voting attitudes, just passing along the info.
There is considerable misinformation in this post. I will touch on a few points.
Critics of TIA's arguments have offered extensive statistical evidence to explain why the exit polls are not convincing evidence of fraud. TIA mostly ignores it, but that doesn't mean that it doesn't exist. Maybe TIA actually thinks that things he disagrees with do not exist?
TIA's repeated assertions that his critics "never consider that the discrepancies could be due to fraud" undermine his credibility. I do not know why TIA is committed to repeating this canard. I am happy to have people read my work and raise questions about it; I am not at all happy to have people invent or distort my positions. Of course I and others have considered that the discrepancies could be due to fraud. Could we please keep it real around here?
Similarly, TIA alleges that critics "dismiss" the pre-election polls. In fact, I and others have discussed these polls extensively; TIA just doesn't like what we have to say.
(Nor do I "dismiss" Bush's 48.5% approval rating. Actually, I demonstrate that extrapolating from previous elections, the break-even point is around 45%: on average, an incumbent with an approval rating over 45% would be expected to win. That isn't to say that a victory was foreordained -- I don't recommend predicting election results from approval ratings alone.)
TIA asks, "Why would Gore (or Bush) voters lie in a confidential survey? Why would Gore voters forget who they voted for just four years later?" Here, TIA fails to engage the evidence to which he is supposedly responding. As I have explained repeatedly, we know that in the 2000-2004 NES panel, some respondents who said in 2000 that they had voted for Gore, said in 2004 that they had voted for Bush in 2000. I have no way of knowing whether they were lying, or forgetting, or what. But whatever exists, is possible. No number of rhetorical questions can wish those respondents away.
TIA asserts as a "fact" that "the 2004 pre-election polls matched exit polls." This claim is, charitably, tortured, for reasons that I have explained in detail. Even on average, the pre-election polls showed Bush ahead, nationally and in the state of Ohio. Even if one believes that these polls should be interpreted to give Kerry an advantage, it misrepresents the work of the researchers to claim that their polls "matched" the exit poll results. Repeated and deliberate misrepresentations of other researchers' work violate scientific ethics.
TIA's comments on New York err in several respects. The New York exit poll gave Kerry a 31.3-point lead, not a 28-point lead. There were four pre-election polls in the last week, each showing Kerry ahead by 15 to 18 points; the Siena and SurveyUSA polls alone had a combined N of almost 1700, not 600. These results do not "match." Worse, while TIA strains to make these results "match," he sets aside all the other state results that likewise infirm his assumptions. TIA did not predict, and has not accounted for, the lack of correlation between exit poll "red shift" and deviations from pre-election polls at the state level. This is one of the many lines of statistical evidence that, for whatever reason, he apparently prefers to claim don't exist.
All these points and more are made in my FAQ, in considerable detail. It is not feasible to rebut repeated misstatements each time they are made. I recognize that some people regard it as uncharitable, unhelpful, and/or unpatriotic to subject TIA's arguments to scientific scrutiny. If such people read carefully the criticisms of TIA in the GD thread that begins here, they will find much food for thought, or for rage, as they prefer. I associate myself with the remarks of RoyGBiv regarding TIA's 2004 analysis in post #216, here:
My friend then forwarded the presentation to a colleague whose life is statistical analysis. His conclusion? I'm paraphrasing here, but it boiled down to something similar to what Skinner said regarding this presentation, "The underlying assumptions are false. This is a case of a conclusion being sought and the methods and assumptions fixed to surround what is sought. This is the kind of thing that makes people distrust statistical analysis." And all this took place well before the election.
Since then I have paid little attention to any of this, except to offer my own little bit here, which I'm sure will draw the ire of many. The point for me is simply this. The worthy goal of exposing election fraud is not aided by bad statistical analysis and in fact works exactly in the opposite direction.
I likewise agree with Skinner's judgment of TIA's analysis of 2006, offered in post #126 in that thread: "This analysis is an embarrassment." Again, I have presented my reasons in detail; again, I have been told that I just don't consider the possibility of fraud; again, TIA has disregarded or cherry-picked the data and prior research. Arguing with TIA is like battling the Black Knight, except that TIA never even offers a draw.
The ten most recent threads posted on the Democratic Underground Discussion Forums.
Obamacare offers more goodies for insurers
By No Elephants
NYC train derailment airs queries about technology
By No Elephants
Paging Leopold's Ghost and NYC_SKP
By No Elephants
Amazon wants to deliver by drones.
By No Elephants
Maintaining nationalism and capitalism for dear life.
By No Elephants
A city this cool really deserves a much better mayor (no, I am not talking about Boston).
By No Elephants
I will never understand this kind of thing.
By No Elephants
The ten most recommended threads posted on the Democratic Underground Discussion Forums in the last 24 hours.
Use the tools below to keep track of updates to this Journal.