Skip to main content

There is no doubt that Edison-Mitofsky's exit polls failed to predict successfully proportions of votes cast for the two presidential candidates in 2004. Because the candidates were so close, and because the exit poll deviated from the count in the loser's direction, this meant that a predicted Kerry win contrasted painfully with a Bush victory in the count.

The 64 million (or is it billion?) dollar questions is: which was wrong, the polls or the count?

"Fraudsters" claim it was the count.  Edison-Mitofsky (and the "anti-fraudsters") claim it was the exit polls.  In this diary I have a brand new analysis.  Bear with me, even if you find math icky.  There will be pictures, if not cookies.

First of all, my credentials:  I do not have a PhD, though I am working on it, and if it wasn't for DKos I might have it by now.  But in any case it is not in statistics, although it is a very statistical field, and I use statistics daily.  It is in science, though, so I have a scientific training: in other words I am trained to make and test hypotheses.  I am also trained to treat my own findings (and the findings of others) with scepticism and to be constantly alert for alternative plausible explanations for my findings, and to set up new testable predictions to further disambiguate my results.

Regarding my "fraudster credentials": I am a fraudster.  I believe your election was inexcusably riggable and may well have been rigged.  It was also inexcusably unauditable. I am convinced that there was real and massive voter suppression in Ohio, and that it was probably deliberate.  I think the recount in Ohio was a sham, and the subversion of the recount is in itself suggestive of coverup of fraud.  I think Kenneth Blackwell should be jailed.

However (and I'll come clean now in case you want to read no further) I don't believe the exit polls in themselves are evidence for fraud.  I don't think they are inconsistent with fraud, but I don't think they support it either. Read on for my reasoning.

My analysis here is of data provided in the Edison-Mitofsky report, namely their table of  "within precinct error" for each state (pp32-33).

The exit polls involve two levels of sampling.  The first is their sampling of what the pollsters hope will be "typical" precincts.  The second is their sampling of voters in those precincts.  Both procedures are liable to both error and bias. In their report, EM distinguishes between error/bias arising from each of these two sampling procedures.  By comparing the differences between the real count in their selected precincts and the real count in the state as a whole, they conclude that their precinct sampling was pretty good.  If anything there was a non-significant bias in Bush's favour.  In other words they picked typical precincts.  

The EM report therefore fingers "within precinct error" or "WPE" as the culprit, in other words the difference, in percentage points, between the proportion of sampled voters saying they'd voted for each candidate, and the proportion of votes apparently cast for each candidate in that precinct.  EM have not given us precinct level WPE data, but they have given us "average" WPEs for each state.  

The point about the WPEs is that they are large - there is a lot of room for random error in this kind of sampling.  However, once you average the WPEs for each precinct, the random error should cancel itself out, leaving a mean WPE of near zero.  That is if the error is random.  If there is bias, either because the sampling technique is biased or because the votes counted do not reflect the votes cast (biased counting in other words) the WPEs will not cancel out - the mean WPE will not be zero.  We can test whether the mean WPEs were statistically significantly different from zero, not within each state (because EM have not given us the standard deviations, only the means) but across states.  Again, even if some states had significant WPEs in one direction, and other states in the other direction, unless there was systematic bias across the nation, the state average WPEs should cancel themselves out.  This we can check.

I did so in this diary, and, surprise, surprise, they are massively significantly different from zero.  The mean WPEs are significantly negative, which, in the sign convention used by EM means they over-stated Kerry's share of the counted vote.

However, I do not think this analysis is valid (though it is mine) and what is invalid about it is also invalid about a number of other analyses that have been done.


If you postulate a systematic bias in the polling (whether it is in the count or the poll - I'm remaining neutral at this point) the magnitude of the WPE will mathematically depend on the magnitude of the actual margin.  For the sake of clarity I am going to assume the error was in the polling, but the math is the same if it was in the count.  

Say you have a precinct with 200 voters.  100 of them vote for Kerry and 100 for Bush.  The real margin is 0 (50%-50%).  And if you poll 50% of each set of voters, i.e. you interview 50 of the Kerry voters and 50 of the Bush voters, your prediction will also be 0.  Your WPE for this precinct will be zero. (In real life, as in coin tossing, sometimes you will poll more than 50 Kerry voters, and sometimes more than 50 Bush voters, but over many precincts, the errors will cancel out, provided the coin is not weighted).  However, say there is something wrong with your sampling technique, and you actually poll 56 Kerry voters for every 50 Bush voters (as EM allege they must have done).  You will interview 56 Kerry voters, but only 50 Bush voters, giving you a total of 106 interviews.   56/106 is 53% for Kerry and 50/106 is only 47%.  So your predicted margin is 6% in Kerry's favour.  As the "true" margin is 0, your WPE is -6.

Now, take this same sampling bias, and apply it to a precinct where there are 160 Kerry voters and only 40 Bush voters.  This will give a result of 80% Kerry, 20% Bush.  But you are over sampling.  So you sample 56% of the Kerry voters, giving you 90 interviews, and 50% of the Bush voters giving you 20 interviews, i.e. 110 interviews.  You then compute your proportions, and it gives you 90/110 for Kerry (82%) and 20/110 (18%) for Bush.  So your counted margin is 20%-80% = -60%, but your estimated margin is 82%-18% = -64%.  Your WPE is therefore -4.

The point here is that the bias is identical in both precincts - you are sampling 56% of the Kerry voters and 50% of the Bush voters.  However, in the pro-Kerry precinct this translates into a WPE of -4, and in the evenly split precinct it translates into a WPE of -6.

We'll do it one more time: take a Bush precinct with 200 voters. 160 of them vote for Bush and you sample 50% of them = 80.  40 vote for Kerry and you sample 56% of them = 22 of them.  You have 102 interviews. 80/102 is 78% and 22/102 is 22%. So your ratio in the count is 80% for Bush and 20% for Kerry, but your prediction is 78% for Bush and 22% for Kerry.  "Real" margin is 60% in Bush's favour, predicted margin is 57% (there's a rounding error here).  So the WPE is - 3.

This all means that for a given uniform bias, whether in the polls or the count, a "swing" precinct will give an error of 6 percentage points, while strongly Red or Blue precincts will give an error of only 3 percentage points.  Moreover, you can show that as the bias increases in Kerry's favour, the apparent error will remain low in the red precincts, but increase in the blue precincts.

I have graphed below what the WPEs would look like in precincts with different degrees of partisanship but with an identical,and extreme degree of polling bias (2 Kerry voters for every 1 Bush voter):

An interim conclusion: any observation along the lines of "exits were more strongly in Kerry's favour in the swing states" is simply an artefact of the math.  For a uniform degree of bias, the erroneous margin produced will be greatest where support for each candidate is most evenly matched. What we need to do is to convert the WPE (or any measure of the margin that is expressed as a difference between two percentages) into a parameter that is mathematically independent of the proportion of votes counted.  This I have done (if you want the formula, email me). EM kindly give us the WPE for previous years, so I have also done the same for the previous four presidential elections.

It gives an index of bias that is not only independent of state "colour" but is also a linear scale, which means we can use fairly conventional stats on it.  

Question number 1

Is the mean bias across the nation significantly different from zero - in other words does the bias in different states cancel out when averaged across the nation?

The test here is a single sample t test, and I have graphed the results below.  I omitted DC and Oregon, which for various reason are anomalous.

The error bars are 95% confidence intervals.  Note that none of the error bars crosses the 0 line, which means that all are significantly different from zero.  The vertical axis gives the value of the bias.  Due to a quirk of the math, positive value means Democratic over-sampling (or undercounting, depending on your hypothesis).  You can see that in every single one of these elections, there was a significant over-estimate of the Democratic vote.  However, you can also see that in 2004 it was greater than in any previous year.  Its nearest rival was 1992.  For stats geeks here are the t values:

I therefore conducted a second t test to check whether the bias in 2004 was significantly greater than in 1992. The answer is yes [t(94)=2.100, p<.05].  The variance was also significantly different, but the years remain significantly different, even after appropriate adjustment of the degrees of freedom.

Question number 2:

Is the mean bias greater in swing states than in safe states?  

To do this I tested the relationship between bias and margin twice: once for a quadratic fit (which would imply that the middle was different to the ends - i.e. swing states different) and once for a linear fit). In three years, the linear fit was significant (1992, 1996 and 2004) but in no case did the quadratic fit result in an improvement to the model.  For non-geeks, this means that the swing states were not in any way special.  

The interesting finding, however, is that the linear fit was positive both in 2004 and in 1988 but negative in 1992. Translated, this means that in 2004 and in 1988, both two-horse races (1988 was Bush senior versus Dukakis), the overestimate of the Democratic vote was greatest in Blue states - the bluer the state, the greater the Democratic over-estimate.  However, in 1992, the reverse was true - the redder the state, the greater the Democratic over-estimate.  In 1992, remember, Perot was on the ballot for the first time.

Here is a scatter plot showing the bias for 2004 plotted against the counted margin.  I have identified the three big swing states by different markers:

I also did some other stuff, but this diary is already way too long.

So:  how to interpret all this? Two points:

1.    The initial excitement over the exit polls was simply that they were wrong.  Not only were they wrong, but they made us think Kerry was going to win. They were also significantly wronger than in any of the previous four elections, and even the nearest competitor (1992) simply over-estimated the extent of Clinton's victory, it didn't have him winning when he didn't, so it may have slipped under our radar. My analysis shows that ever single poll has significantly over-estimated the Democratic vote.  The probability of this being due to chance is astronomically low in both 1992 and 2004.  So we can conclude that there is in-built Democratic bias in the exit-polling system that varies from year to year.  It appears (as EM point out in their report) to be greatest where the interest in the election is highest (1992 and 2004).  However, this year the bias was significantly even greater than in 1992 (though only at a probability of p<0.05).  

2.    My analysis shows that the swing states were not in fact more wrong than the safe states.  This evidence shows that the greatest bias was in the safest blue states.  Although the big swing states of Ohio, Florida and Pennsylvania are all on the high side of the line, they are not significantly out of line (I checked).  In fact they are relatively close to the regression line.  So even if there was a legitimate reason (as I suspect there is) for an in-built polling bias that is greatest in Democratic states, these three states are not an exception to this rule.  Moreover, the pattern of polling bias is the same as in the nearest comparable election, 1988, another two-horse race where there was also a large significant over-estimate of the Democratic vote and another losing Democratic candidate (Dukakis).

I therefore conclude that we have a choice between the following hypotheses:

  1. There was legitimate over-sampling of Democratic voters that was greatest in the most Democratic states, that this is a pattern that has been observed before, but for some reason was greater than usual this year.

  2. There was widespread, state-level fraud targeted somewhat inefficiently at the bluest states, not at the swing states.

  3. The first effect, plus targeted small scale fraud in swing states that is lost in the variance due to polling bias.  After all, it remains true that a relative small degree of ballot stuffing in Ohio, Pennsylvania or Florida could have swung the election for Kerry and would not appear in the exit polls.

But of course this would also mean that the magnitude of the exit poll discrepancy is not evidence of fraud.

[Now cross posted at Booman Tribune]

Originally posted to Febble on Wed Apr 06, 2005 at 05:00 AM PDT.

Your Email has been sent.
You must add at least one tag to this diary before publishing it.

Add keywords that describe this diary. Separate multiple keywords with commas.
Tagging tips - Search For Tags - Browse For Tags


More Tagging tips:

A tag is a way to search for this diary. If someone is searching for "Barack Obama," is this a diary they'd be trying to find?

Use a person's full name, without any title. Senator Obama may become President Obama, and Michelle Obama might run for office.

If your diary covers an election or elected official, use election tags, which are generally the state abbreviation followed by the office. CA-01 is the first district House seat. CA-Sen covers both senate races. NY-GOV covers the New York governor's race.

Tags do not compound: that is, "education reform" is a completely different tag from "education". A tag like "reform" alone is probably not meaningful.

Consider if one or more of these tags fits your diary: Civil Rights, Community, Congress, Culture, Economy, Education, Elections, Energy, Environment, Health Care, International, Labor, Law, Media, Meta, National Security, Science, Transportation, or White House. If your diary is specific to a state, consider adding the state (California, Texas, etc). Keep in mind, though, that there are many wonderful and important diaries that don't fit in any of these tags. Don't worry if yours doesn't.

You can add a private note to this diary when hotlisting it:
Are you sure you want to remove this diary from your hotlist?
Are you sure you want to remove your recommendation? You can only recommend a diary once, so you will not be able to re-recommend it afterwards.
Rescue this diary, and add a note:
Are you sure you want to remove this diary from Rescue?
Choose where to republish this diary. The diary will be added to the queue for that group. Publish it from the queue to make it appear.

You must be a member of a group to use this feature.

Add a quick update to your diary without changing the diary itself:
Are you sure you want to remove this diary?
(The diary will be removed from the site and returned to your drafts for further editing.)
(The diary will be removed.)
Are you sure you want to save these changes to the published diary?

Comment Preferences

  •  Blatant request for recommends (4.00)
    There have been a lot of diaries on the UScountsVotes paper, and this is not it - but if you are going to read the UScountsVotes paper, this is relevant reading.

    So it would be nice if it stayed up for a while, so it can be seen by a reasonable range of interested Kossaacks.

    It was also one helluva lot of work.

    •  Febble, I can't see your graphs. (none)
      This might be my security settings, but I don't think so. Someone else needs to say if they can see them.

      Notwithstanding your analysis, which I had to scan as I am running off to work, it still seems to me that there was a very strategic decision taken to make sure Bush won electoral college votes in the big swing states. In any case, your effort and the discussion it engenders are appreciated.

      Also, this was not the only way in which Bush cheated.

      "David Brooks: The poster child for false consciousness"

      by lecsmith on Wed Apr 06, 2005 at 05:06:34 AM PDT

      [ Parent ]

    •  "1992 was Bush senior versus Dukakis" (none)
      I think you mean 1998, no?  

      Working through your post now.  Looks pretty good, but playing the skeptic, I'd like to see the graphs/spreadsheets/formulas/etc.  I'll e-mail.  

      What do you suppose could explain the Presidential race/Senate race WPE discrepancy?  Differential within survey response?  Eager Kerry voters answering only the President question and skipping the other questions?  Reluctant Bush supporters skipping the President question and answering the other questions.  I'm still wondering about that.   I would also like to know if this disparity was present in previous exit polls.  

      I have to assume that the NEP Report was supported by a multi-variate regression analysis.  E/M are professionals.  My guess is that they found WPE is best explained by a host of interviewer (age, education, experience of interviewer, etc.) and precinct characteristics (distance from poll and number of polling stations in same building [not sure the respondents were from the intended polling station], weather, number of opportunities, etc.).  The NEP Report, though, was  not written for data dweebs like US Count Votes or ourselves, but instead for media execs who spent $10 mil on a poll and want to know what went wrong.  If the intended audience started reading about the tests and p-values behind every statement in the report, their eyes would glaze over.  "What's the bottom? Don't confuse us with statistics" sort of thing.

      So, I suspect that they took the basic findings that showed WPE highly correlated with many variables consistent with rBr hypothesis and wrote a narrative based on these findings and included only more simplistic crosstabs on the data in the Report.  Of course, US Count Vote jumps all over these more simplistic crosstabs, but I can't say that I blame them - it's all that is out there.  

      If WPE is correlated and best explained by interviewer and precinct characteristics, I cannot imagine a plausible scenario where these same WPEs can be better explained by fraud, unless of course, the fraud was nearly ubiquitous as you mention in your conclusion #2 above.  Although, #2 goes against the US Count Votes "fraud in Bush strongholds" hypothesis.

      A small portion of these WPEs could be explained by fraud, as you say in #3.  The NEP Report never disproved fraud.  It simply said that the bias in the poll is explained most prominently by differential non-response and non-random within precinct sampling (that lead to differntial response).  It also said that the data do not suggest fraud, which is different than saying the report disproves fraud.  Disproving fraud was not a goal of the NEP Report.  Explaining what most likely went wrong was and they did this.    

      The exit polls are not a smoking gun.  

      •  Whoops... I replied to the wrong entry...n/t (none)
      •  You are right about Dukakis (none)
        I got my years confused.  I think the numbers are OK though.  I'll update the diary.
      •  Date corrected (none)
        Yes, I would assume they did a regression, but they also published a correlation matrix for WPEs in each year.  I checked it and it was the raw WPEs they were correlating.  They will necessarily be correlated if they are dependent on actual margin, given that the margins are highly correlated.

        So if they did do a regression analysis to ascertain which factors contributed most to the WPE, I am not confident they were using the right dependent variable.

        They also show mean WPEs for different types of precincts, including partisanship, and dismiss the ends of the graph with some comment about the partisan precincts having less scope for over sampling or somesuch.  So I have a horrible feeling they don't get it.

        Unless I am wrong with this approach.  But it makes sense to me.

        •  Hmm... (none)
          Regarding the mean WPE by ordinal precinct partisanship see my comments here.  

          I'm not sure what you mean by not using the correct dependent variable.  Which one would you use?  Why is the signed WPE not okay if, as you say, the average should be zero if there was not statistical bias?  

          •  Well, mine. (none)
            I don't think it is valid to use a dependent variable (the signed WPE) that is contaminated by the partisanship of the precinct, particularly when one of your predictors is partisanship of the precinct.

            Plus, there was statistical bias.  You can use the WPE to find out whether it was significant, but there will still be lots of noise from the contaminat variable (partisanship).  Interestingly, having got rid of that contaminant (I hope) my one sample t tests are significant beyond Ron Baiman's wildest dreams.  Trouble for him is, so are the others (or most of them).  Even the "clean" year (2000) comes out significantly biased, even if only at p<.05

            See my new Exit Poll diary here

            by Febble on Wed Apr 06, 2005 at 07:51:23 AM PDT

            [ Parent ]

            •  Certainly there was statistical bias... (none)
              That's why use of signed WPE as a dependent variable should be valid for testing many independent variables, including interviewer and precicnt characteristics.  Notice that I did not include precinct partisanship in my list of precinct characteristics above.  That is because E/M do not rely on WPE by precinct partisanship to support their differential response/non-response hypothesis.  

              It seems to me that US Count Votes is using the relationship between ordinal precinct partisanship independent and signed mean WPE dependent (and completion rate) as evidence against the E/M hypothesis, when E/M only report those data in cross tabs and do not use it to support their conclusions.

              I'm at work and am only giving this 1/2 second thought, so I could very well be missing your point here.    

              •  yes, and my "bias index" (none)
                amplifies one of the US Counts Votes points.  I converted the mean and median WPEs into my Bias value, and re-graphed the graph showing WPE against precinct partisanship.

                The result was that both the mean and median value for the strongly Republican precincts shot up (although the difference between mean and median gives us pause for thought as the US Counts Votes team rightly point out and explore).

                So this is somewhat counter-intuitive: Republican precincts have a greater over-estimate of the Kerry vote, but Democratic States ditto.  One answer may be "Republican precincts in Democratic states".  But remember its only a handful of precincts (40 in total, and a skewed distribution).

                Maybe this is where fraud occurred.

                But if fraud occurred in a handful of highly Republican precincts, it can't account for the magnitude of the Exit Poll Discrepancy.

                See my new Exit Poll diary here

                by Febble on Wed Apr 06, 2005 at 01:12:57 PM PDT

                [ Parent ]

                •  Interesting... (none)
                  I do think that it is odd that respones rates aren't higher or that refusal rates aren't lower in strong Republican precincts.  But when you throw in the added WPE variable, I think things get tricky.  

                  Are you considering the at least 10% or more margin of error within each precinct, which would place limits on the usefulness of the mean WPE by precinct partisanship?  This was the gut of my previous comments here and here (scroll down), but you didn't address these directly in your response.  

                  Throw geography into the mix as well - wouldn't these Bush strongholds also have to be in urban areas given WPE by precinct area population?  

                  Also, could differential response from non-random within precinct sample selection lead to some of this error, say if a bunch of young former Deaniacs were assigned to urban Bush strongholds, but chose to interview "attractive" (not suggesting purposeful interviewer fraud here) voters that were slightly less than random?  (see my comment here for more on the differential response from non-random precinct sample selection)

                  Lot's of speculation here, I know...  Just trying to talk throw potentially testable hypothesis if we ever had the data...

                  •  Quite - if we ever had the data... (none)
                    We desperately need to know about those regressions.

                    I do not follow the US Counts Votes argument that because any one factor leaves plenty of residual WPE that none are significant. All could be contributors, but tabulating by each doesn't tell us what other factors were present in those precincts as well.

                    Regarding your other questions - I'll have to come back to this later, as I have to help my son with his math homework!

                    Ruddy cricket scores.  And anyway it's nearly bedtime in the UK.  

                    Stick around!

                    See my new Exit Poll diary here

                    by Febble on Wed Apr 06, 2005 at 01:54:27 PM PDT

                    [ Parent ]

                •  Finally had a moment to reflect on this further... (none)
                  I am now curious what results would occur if E/M tested all their variables against your new dependent variable, assuming of course, it is constructed appropriately (looks like Arvin is working on that now).  

                  I have your SPSS data, but am waiting for the variable descriptions and some type of narrative explaining how you derived your columns and the processes you went through to achieve your results.

                  What will you call your new dependent variable?  I think that WPE is the hottest and most widely accepted acronymn in this field since ANOVA.  Perhaps you could simply force a redefinition of WPE, thus appropriating the catchy term and replacing it with a more valid measure.  Or come up with a catchy competing term...

                  Ideas anyone?  

    •  2:1 Simulation Graph description is confusing (none)
      Great work.

      My one criticism is that the 2:1 simulation is improperly described.  Because you are not simulating sampling 2 kerry voters for every bush voter, but instead sampling kerry voters at twice the rate of bush voters -- a small point but might help to clarify your presentation.

    •  Exit polls don't support fraud (none)
      This is a reasonable analysis.  Thanks for the work.

      I guess another thing to take away is that the exit polls are a poor indicator of the quality of the vote-counting system, BUT: they are the only tool we have for testing the unaccountable, unauditable vote measurement system we're using.

      Factoring in the poll bias, can we estimate the standard error of the vote and vote count?  I'd really like to know how bad our counting system is.


      And some categorization for retrieving this from the dKos oblivion:

    •  hooray (none)
      thanks, febble!  It's nice to see reasonable analyses like this one.  I've felt for a while that all the exit poll screaming actually undermined the effort against the vote suppression in Ohio.  

      I've heard before that since part of the interest of the exit pollsters is to get a variety of demographic data, that that by definition benefits the Democrats, since Democrats are more diverse.  But if the precinct choice was a random sample of a state's results, and if they exit pollsters really did pull aside every nth voter, then I am not sure how that bias would be introduced in the data set.

      Politology.US - Politics and Technology in the United States

      by tunesmith on Wed Apr 06, 2005 at 02:14:17 PM PDT

      [ Parent ]

  •  this is far too sensible to get paid attention (3.40)
    but I will forward it to those outside of dKos following this. Feel free to send to USCV, of course.

    Note Febble's position, the only sensible one. Exit polls don't tell us what we need to know, no matter how you slice them.

    "Politics is the art of looking for trouble, finding it everywhere, diagnosing it incorrectly and applying the wrong remedies." - Groucho Marx

    by Greg Dworkin on Wed Apr 06, 2005 at 05:17:51 AM PDT

    •  You Know What I Love About This Diary? (4.00)
      She says she believes there was fraud, but eliminates one possible piece of evidence in support of her suspicions.  True integrity, both in terms of adherence to the scientific method, and simply intellectual honesty.

      Now, the question I would love to see answered--and this is probably not one to be answered by stats people, although their work will have to be applied to test the findings--is why there is the oft-reported (and oft-ignored, at least among most of the fraudsters) built-in Dem bias.  I have suspected all along it's probably something to do with the training and selection of the people actually out at the polling sites asking the questions.  

      Some years ago, when I was involved with the college newspaper, we got a call looking for people to help with exit polling for a Detroit election.  It was about 7:00 PM the night before the election, and they were looking for somebody, anybody who wanted to get paid whatever nominal amount they were offering to stand outside a polling site and ask questions of the departing voters.  Since I have some background in empirical social science, including social psychology (which is one of the most methodologically rigorous--some would say anal--of the social sciences), it shocked me that they weren't paying any heed to the various response biases you get by the race, age and other factors of the questioner.  They weren't training people on how to ask questions, how to approach people, etc.  They were just giving people clipboards and about 5 minutes of instruction.

      To do sound survey or polling research, you have to control for a ton of factors or you'll get really skewed response rates, refusals, unrepresentative response patterns, etc.  Because they're done on the cheap, increasingly so as the networks try to pinch pennies and demand speedier results at lower costs--I suspect the "problem" with exit polls as they're done for U.S. elections isn't with stats and sampling in their construction, but with how the survey takers are chosen, trained and behave at the polling sites, which probably results in some kind of measurable bias along class, age or racial lines that ends up reflected in the partisan skew of exit poll results.  

  •  It is a sad commentary (none)
    About the state of Daily Kos that this will not get recommended.

    For the record:

    THERE IS ANOTHER EXIT POLL: the LA Times did one and had it Bush 51, Kerry 48.  I have never seen this fact referenced in any of the fraud analysis.

    •  I think you are looking at data.. (none)
      that has been weighted to the election result.  That is typical practice.  I don't think anyone knows what the unweighted LA Times data showed.  
    •  But It Has Gotten Recommended (none)
      There seems to be a silent majority (or at least a silent but sizeable minority...though that sounds less sexy 'n' Nixonian) of kossacks who are neither fraudsters (in the sense that they know this election was stolen) nor anti-fraudsters (in the sense of considering fraudsters to be a bunch of tinfoil-hat wearing conspiracy theorists), but are rather concerned citizens who want the investigation of the 2004 elections to continue because, though it's hardly an open and shut case, it remains a real possibility that massive fraud occurred. Count me as one of these folks.

      Because I remained concerned, I try to recommend serious diaries like this one about this subject, whether they argue in favor of, or against, fraud.

      Start doing the things you think should be done, and start being what you think society should become. -Adam Michnik.

      by GreenSooner on Wed Apr 06, 2005 at 07:52:32 AM PDT

      [ Parent ]

      •  Thanks! (none)
        There were some snide remarks on the other Exit Poll diary about the number that had appeared, but I do think it is important to subject these papers that keep appearing to a proper critique.  I was going to critique the US counts Votes paper here, but they diary got way too long.

        They do make some good points, but I disagree with their very definite conclusions.  I know they are working on precinct level analysis (and county level) which is where any bodies are likely to be buried.

        See my new Exit Poll diary here

        by Febble on Wed Apr 06, 2005 at 08:01:25 AM PDT

        [ Parent ]

    •  Why would you assume this? (none)
      As a group we on dailyKos are certainly willing to entertain the idea of Republican fraud.  However we are also basically intellectually honest, within the limits of managing our own cognitive dissonance.

      I for one am a 'fraudster' - but am recommending this diary which I think makes an important and worthwhile argument.

      Apparently I have made the unbelievably naive error of overestimating the intelligence of the American people.

      by Citizen Clark on Wed Apr 06, 2005 at 11:55:48 AM PDT

      [ Parent ]

      •  yeah (4.00)
        I agree with Febble's comments at the start. Blackwell should be jailed immediately for his crimes against the citizens of Ohio in 2004, and the recount was a sham. Whether or not you think there was actual conspiring going on, those two points are most definitely true.

        It's certain that Blackwell's actions resulted in significant penalties against the Democrats, disenfranchising thousands of Ohioans.

        And I'm bothered more than you know by the complete bungling of something as simple as auditing. Duh! People want to know their vote counts. I'm sorry, as a computer engineer I find it really hard to believe that leaving auditing and paper trails out of the e-voting machines was an accident or an oversight. No professional would do that. That kind of design flaw decision is made deliberately and with full foreknowledge of the inevitable results.

        But a level of fraud sufficient to explain the vote here in Ohio is an extraordinary claim, and requires extraordinary evidence. To uncover that sort of evidence, the kind that cannot be dismissed easily by anyone, you have to pursue it with strict intellectual honesty. And so I give a hearty recommendation to this diary.

        The Americans will always do the right thing... after they've exhausted all the alternatives. - Sir Winston Churchill

        by drewthaler on Wed Apr 06, 2005 at 02:15:46 PM PDT

        [ Parent ]

        •  Very nicely put. n/t (none)

          Apparently I have made the unbelievably naive error of overestimating the intelligence of the American people.

          by Citizen Clark on Wed Apr 06, 2005 at 02:22:20 PM PDT

          [ Parent ]

        •  addendum (4.00)
          I said: "It's certain that Blackwell's actions resulted in significant penalties against the Democrats, disenfranchising thousands of Ohioans."

          I should amend that. Not just the Dems... Libertarians as well. I may not agree with all of the Libertarian party platform, but I will stand up and fight for their right to be on the ballot. There are rules, and they followed them. Yet Blackwell rejected over 50,000 signatures on a cooked-up technicality and refused to register the Libs as a party.

          Fifty thousand voices silenced just for the sake of petty political meddling. That really pisses me off.

          His actions really are a disgrace to the office.

          The Americans will always do the right thing... after they've exhausted all the alternatives. - Sir Winston Churchill

          by drewthaler on Wed Apr 06, 2005 at 07:01:08 PM PDT

          [ Parent ]

  •  WHY New England States (none)
    [Your reply in other diary referenced this comment.]

    Anyone a suggestion why especially the New England states were lacking accuracy in election results? Also the few BLUE shifts - zie sorted info below.

    State         RED shift
    Vermont           5,20
    New Hampshire     4,90
    Delaware          4,80
    South Carolina    4,40
    Alabama           4,20
    Alaska            4,00
    Massachusetts     3,70
    Connecticut(1)    3,40
    Pennsylvania      3,40
    Rhode Island      3,40
    Mississippi       3,30
    Nebraska          3,30
    Ohio(1)           3,10
    Minnesota         3,00
    Wyoming           2,70
    National Vote     2,70

    Tennessee        -1,70
    South Dakota     -1,80
    Texas            -2,00
    North Dakota     -2,40
    Kansas           -2,70
    State        BLUE shift

    In 2005 - Be liberal: Support our Allies of Democracy on Human Rights, the Environment, Gay and Minority Rights & EU and UN Third World Development Programs & Our Friends

    •  I don't know (none)
      But it doesn't easily fit with the fraud hypothesis.

      We know, from actual evidence (some of which is reported in the EM report, some is so embarassing they failed to report it but is reported by "Professor M" in Mystery Pollster's blog) that the sampling was done badly this year, and that over-eager voters were more likely to be sampled.  From my own limited experience it is very difficult to approach a scowling 10th person when a smiling 11th person is close behind.

      There is also some evidence in the EM report that the bias was especially big in strongly Bush precincts - in contrast to my finding of greater bias in strongly Democratic states.  This remains true when I applied my decontaminant formuli to EMs WPE - it actually strengthened the effect. We also know that in the very strongly Bush precincts there were some outliers leveraging the WPE upward  (there are only 40 precincts in the very-High-Bush category).  So it is possible that some of these very high-Bush precincts were in strongly Democratic states.

      So here's a hypothesis: Where you have a lagging Democrat (Dukakis, Kerry), Democratic voters in strongly Democratic states are eager to show that they voted for their guy, and especially eager to show that their guy got a vote in a strongly Republican precinct.

      On the other hand in Red states, they are not so eager, knowing they just cast a losing vote. No-one wants to look like a loser.

      But whether it does or not, that kind of process would be consistent with the over-polling of Democrats in Democratic states.

      •  H you read Freeman's forthcoming... (none)
        work?  I think he suggests that these states were rigged to "pad" the popular vote so there would not be a repeat of 2000.
        •  Which shows how asinine all of this is (none)
          Are you suggesting that the biggest conspiracy was in VERMONT????
          •  Dean is a double agent (heh) (none)
          •  "biggest conspiracy" (none)
            The size and location of changed results do not have to depend on a proportionate ground operation.

            I think it may be necessary to establish some shorthand vocabulary to cover the various permutations of hypotheses under which a suspicion can be ruled out or sustained, and then try testing each hypothesis.

            As in:

            The Scalability Hypothesis: That some central capability exists to manage modifications of returns at the local level across the country.

            The Popularity Hypothesis: That an effort was made to distort not just the legal outcome but also the total size and distribution of support.

            The Crowd-Cover Hypothesis: That modification of the results was concentrated in places where large margins could survive the loss.

            The Trend Hypothesis: That subversion of the process has been a growing and perfecting phenomenon over some number of election cycles.

            Otherwise, we introduce these ideas afresh every time, and they don't accumulate resolution.

            •  I am from Vermont (none)
              None of these apply.

              First, the Secretary of State is a Democrat.

              Second, optical scans with audit trails are used in the big cities (like South Burlington, population 15,000).

              And a significant amount of the votes in Vermont are still on paper ballots!

              I would submit that Vermont might be the toughest state to influence the outcome.

              For christ sakes Burlington, the largest city in Vermont, is run by the left and is the home of Bernie Sanders.

              I really think this destroys the fraud myth.

          •  don't have to be in VT to hack VT vote (none)
            and NB - a witticism is not an argument

            hark! hark! the Clark!!

            by Errol on Wed Apr 06, 2005 at 04:00:35 PM PDT

            [ Parent ]

        •  OK (none)
          But why rig in Dem states?

          That was the final nail in the exit poll coffin for me.  If I was going to do a one man, nationwide hack of the count, I'd do it in all states, not just the Dem states. Or if I was going to be selective, I'd choose the swing states.

          If a hack accounts for the bias, it has to be both widespread and more prevalent in Dem states. Or if there was a hack on top of a polling bias, it could be uniform - but then you wouldn't need to invoke a hack to account for the bias.

          People (IMO) have to let go the idea that the bias in the exit polls are evidence of fraud.  Fraud may still have occurred, and it may have contributed to exit poll error, but the exit poll error is not indicative, or even characteristic, of any plausible fraud scenario.

          Which is no worse an argument than the UScountsVotes argument that the observed WPEs could only arise from "implausibly" high levels of Democratic over-sampling.

          What's implausible for the goose is implausible for the gander.

          See my new Exit Poll diary here

          by Febble on Wed Apr 06, 2005 at 08:16:31 AM PDT

          [ Parent ]

          •  why? (4.00)
            to cover your tracks.

            Look how doing it has thrown you off in your estimation of fraud.  By padding in deep blue states it makes the case FOR fraud harder to build since it seems a bit odd to honest people.

            But you are not trying to think like a crook, and that is what you have to do to make sense of why.  It is simply to put in a level of statistical noise to cover the tracks.  

            And your analysis is quite good, but it still as do all the others done trying to show that they polls are NOT a smoking gun, in that it does not account for the senate portion of the poll not being off.

            Ticket splitting is kind of rare, and would tend to be even more so in such a hottly contested race such as this one.  So if the senate numbers are accurate, but the presidential ones are not... there is something wrong.

            Like I have said in the past:  We are not looking for a conviction here, just the indictment.  Is there enough evidence that a crime may have been committed?  And when you answer that question answer it as someone acting as a forensic auditor.  Is what you are seeing in the numbers a series of balancing counter entries in a set of cooked books designed to cover the tracks of the embezzlement of votes?  

            You must start from a base of suspicion, of not only the numbers but the motives.  You should not start from a base of disproving the possibility of fraud, that is the defenses job.  

            And the bias towards dem votes in the previous elections that you note, may be evidence of vote suppression in previous elections, and not something that is innocent.  It could also be nothing, but noone to my knowledge has looked at it in that light so far.  Heck it could be that everyone presumes that dems will vote in larger numbers based upon preelection polls, and the dems just decide not show up.  But it is something that has to be addressed and accounted for if it is to be used as a defense of the anomoly.

            I do wonder though if the two previous dry runs that EM did with the methodology for the '04 Presidential poll are available online?  Since as I recall they tested everything in not only the primaries, but the preceding midterms.  So if the polls were accurate in both previous instances and they same conditions were placed upon pollster placement.... we have a problem.

            But you do touch on, at least in my eyes, the most indicative case for fraud:  the cover up in Ohio.  If there was no fraud then why lock down the information?

            Funny numbers+obfuscation = problem.  

            Do we have enough to indict?  Yes.  Do we have enough to convict based upon the evidence that we currently possess?  No.

            But if we don't indict we can not engage in discovery!  All this round and round that people are doing trying to show that the polls are not a smoking gun; as in enough to  convict.  Is a waste of time.  Since we know they are not, they are just an indicator of probable crime.  It would require additional evidence to prove the crime conclusively.  And the only way that I can think of other then a black bag job on the HQ of the RNC or EM that would not be admisable in a criminal court... would be to get an indictment and engage in criminal discovery!

            And we won't get that indictment as long as our own side continues to give the other guys additional plausible deniability.  When what we need to do is agree that the exit poll in and of itself is not enough to convict, but it is enough to indict, and doubly so when taken with the actions taken post election!

            And that is all that it needs to do!  It needs to get us in the door, and from there if there is something hinky we will find it.  If there was nothing hinky going on we will find that to.  

            And please do not think that I am being harsh or unappreciative of your analysis and attention to this subject, I am not.  I believe  that you are just looking for to much from the numbers.  And as good as your reasoning may be it falls flat in one regard:  If it is the valid reason for the poll and result not jibing... why then were actions taken post election to make investigation moot?  

            •  Fair enough (none)
              All I ask is that any fraud hypothesis takes account of this pattern of bias.

              Up till now people have argued that the bias was greater in the swing states, which it did seem to be on some analyses.  I thought so too.

              But this indicates something else.

              The hackers must be smarter than the statisticians, because they hid it so well we found it in the wrong place....

              maybe, maybe....

              But the psychology of over-eager voters makes more intuitive sense to me.

              See my new Exit Poll diary here

              by Febble on Wed Apr 06, 2005 at 01:58:18 PM PDT

              [ Parent ]

              •  one other (none)
                thing that I thought of and should have included in my first reply...

                As you stated about how you would go about doing a hack of an election.  You forgot to say how you would go about doing it and  HIDING it so that you got away with it.

                Now if you can't hack every state do to what ever circumstance, either you don't have the time or the resources to do so, or you don't have the equipment needed, there are a number of reasons that you may not be able to do every state.

                The best reason and the reason that you would not want to only do the swing states is that it is to be blunt:  Amateurish and ham handedly expected.

                Which is no great insult, it just shows that you are a basicly honest person and would make an underachiever of a criminal.
                And that is a good thing.

                IF you do every state that would be suspicious, even more suspicious than only doing the swing states.  It is literally TO big a theft NOT to notice.  Ambitious?  Yes.  Likely to get your butt in jail?  Oh yeah!

                The key is not to be TO greedy when you are stealing.  Steal only as much as you can get away with, and you will live to steal another day.

                So ask yourself how you would do this so that if you were investigating it, you would be able to get away with it.  How would you hide the crime so that it would be the least likely to show up?

        •  Big problem (none)
          The problem with this scenario is that it requires a much larger, more difficult to conceal conspiracy. Every state has different voting and tabulating equipment, so successfully rigging the vote would require lots of different techniques. You'd also have to have a lot more people involved, so concealing such an effort would be very difficult, espescially considering the number of Democratic and even international observers watching this election. In some states, it's inconcieveable that such a mechanism could be concealed without the cooperation of elected officials, some of whom are Democrats.

          For the hacked-vote hypothesis to fly, it must be confined to a few swing states (and probably only certain precints within those states), otherwise the conspiracy must be implausably large.

          •  And witrhin states ... (none)
            ... most counties have different equipment, and programming.

            And within the same equipment and programming, control tables (coding races and candidates by ballot position) vary down to a few precincts or even single precincts.

            And in most setups, nothing in the equipments' program or control data knows the names of the candidates.

            This presents considerable difficulties for "lone gunman" theories.

            •  Yeah, What Was the Term You Used? (none)
              Exposure points or something like that?  To perpetrate such a conspiracy would require thousands of people, and in many places under the oversight of Dem election administrators (and plenty of scrupulous and honest Repubs as well).  A conspiracy so large would have thousands of points where a mistake, somebody with second thoughts, somebody catching the tampering, whatever, would expose the conspiracy and open up lines to connect that node to the entire conspiracy network.

              Sure, in some isolated places one could rig the votes; it's been going on since the first election was held.  But to rig the results so you would get consistent patters between and within states, from the aggregatated national totals down to the precinct level (with some obvious outliers here and there) would require a conspiracy so huge that it would be impossible to conceal.

              If this election was "stolen," it was stolen only in Ohio, and then the question is why was Ohio such an anamoly in that it would have been one of only three states--VT and SD being the others--where Bush's percentage of the vote fell from 2000, and there it would have been a massive drop compared to VT and SD.  

              •  Perhaps somebody out there has ... (none)
                ... configured and launched a 10,000 point-of-presence distributed processor network, with custom set-up at each point, overnight, single-handed, with no sytems test phase, with no cooperation from users, and had it all come together uneventfully.

                Perhaps he or she can tell us how easy it would be to execute a Lone Gunman hack of the national elections of 1988 thru 2004.

          •  not really (none)
            How many vendors for tabulators are there?  2? 3?  And is there a way to view the source code for the actual voting machines?

            All it takes, is for a sizable number of the central tabulators being online, and those that are going to hack the vote knowing the number for the machine.  That is all.  Hell get me the IP addresses of the machines and some cash and I could have a group of programmers in Bangalore hack the election!

            Or in the case of the source code... you can write the vote swap  into the code itself, and all that takes is ONE guy to code it.

            The assumption of a vast conspiracy to do the job is old style post internet thinking.  One man can in the information age quite literall slash a thousand throats in the manner of the old Arab proverb.

            every precinct may do a count of the votes and then send it to the county...but where does the county send it?  A single tabulator.  And if the votes were changed in that single machine how would you know.

            Well you could check the precinct and county records... except in Ohio where that information is no longer available to the public.

      •  Where WPE was highest... (none)
        You wrote:
        There is also some evidence in the EM report that the bias was especially big in strongly Bush precincts - in contrast to my finding of greater bias in strongly Democratic states.  This remains true when I applied my decontaminant formuli to EMs WPE - it actually strengthened the effect. We also know that in the very strongly Bush precincts there were some outliers leveraging the WPE upward  (there are only 40 precincts in the very-High-Bush category).  So it is possible that some of these very high-Bush precincts were in strongly Democratic states.

        True, the data could be explained in part by the presence of high-Bush precincts in strongly Democratic states. But also consider this: WPE was highest in the most urban areas (presumably Kerry strongholds) and lowest in the least populated areas (presumably Bush strongholds).  Note what E/M said about the margins:

        The analysis is more meaningful if the precincts where Kerry and Bush received more than 80% of the vote are ignored.  In the highest Kerry precincts there is little room for overstatement of his vote.  Similarly the highest Bush precincts have more freedom to only overstate Kerry rather than Bush.  The three middle groups of precincts show a relatively consistent overstatement of Kerry.

        Note also what else I said in this comment I made in another thread regarding the use of "extremely" partisan precincts.  Due to the within precinct sampling error, there are limits to the feasibility of this sampling error at the margins.  In fact, it ensures that the WPE at the margins can only  
        •  According to Professor M (none)
          and Mystery Pollster, over-eager voters was a problem.  The interviewers were very inexperienced.  It is hard enough to interview someone who looks hostile but even harder to reject someone who looks keen.  

          Anecdotes include someone filling out a questionnaire voluntarily and dropping it in the box; couples where one refused but the other volunteered as a replacement.

          This kind of bias would not show up in response rates, but might have a big effect.

          See my new Exit Poll diary here

          by Febble on Wed Apr 06, 2005 at 10:31:13 AM PDT

          [ Parent ]

          •  Very astute! (none)
            In fact, I suggest it might be one of the secrets to the persistent democratic bias since 1988.  But how much of the WPE does it explain?  

            There are two factors to the E/M hypothesis: 1) differential response (influenced by non-random within precinct samples); and 2) differential non-response (reluctant Bush respondents).    

            As Dan Merkle (ABC) and Murray Edelman (VNS) demonstrated in their chapter "Nonresponse in Exit Polls: A Comprehensive Analysis" of Survey Nonresponse by Groves et al (2002), historically, nonresponse patterns have been largely random. (BTW - I HIGHLY recommend this text as it gives a lot of insight into implementation of exit polls).  

            Response rates and survey error were slightly correlated in 1992.  Refusal rates and miss rates did not significantly predict within precinct errors.  The authors write: "Knowing the response rate has almost no predictive power when it comes to explaining exit poll error" (pg. 254).  Yet, in 1992, there was a mean WPE of -5.0, and the WPE has shown democratic bias for each election in the last 20 years.  So, what can account for this persistent bias?  

            Part of the answer is differential response.  As you say, "This kind of bias would not show up in response rates, but might have a big effect."  

            Differential response can refer to not only a difference in random responses (which research shows has not historically explained WPE), but also account for the mostly immeasurable non-random within precinct responses.  

            Notice that mean WPE appears to get larger when the number of opportunities for non-random selection increases (pg. 36 of NEP Report).  In fact, the E/M report state that .  ANOVA or other testing of the hypothesis that the more chances an interviewer had to make non-random selections, they "substituted" their own respondent for the random one, only every so often (doesn't take much), would be nice.

            Also notice that WPE appears to be correlated with distance from the poll (page. 37), ability to reach every boter (pg. 37), number of precincts at each polling place (pg. 41), and when the interviewer was hired (pg. 46).  Again, more rigorous testing would be nice.  

            This suggests to me that these independent variables could explain a portion of the WPE, as stressed by E/M (pg. 35).

            Imagine an interviewer who was hired recently, was assigned to sample every 10th voter from precinct X, but the building that he/she had to stand 50-100 ft away from housed precincts W,X,Y,Z.  Now, voters are exiting the building, some walk away, some toward you.  You are trying to count the "10th" voter, but how do you know if it is the 10th voter from YOUR assigned precinct?  So, you start fudging and approach "approachable" people, people who may look more cooperative to you.

            One thing that has to this point been ignored is the question of interviewer affilialtion or presidential choice in 2004.  Most of these folks were: a) young; and/or b) highly educated.  Sounds like Kerry demographic to me.  Maybe...Just maybe... was there some unconsious bias introduced if some selections were not completely random?  I'm not accusing anyone of purposeful fraud, but if undertrained interviewees like those mentioned in Professor M's anecdote were to self select, then perhaps they would subconsciously gravitate toward voters who might be just a little bit like them?  

            The Merkle and Edelman chapter had this to say about exit poll interviewers:

            All interviewers are asked to fill out a questionnaire after they complete their assignments on Election Day.  The interviewer questionnaire includes a number of background questions, such as basic demographics and prior interviewing experience, as well as questions related to the interviewing experience on Election Day (pg. 245)
            The chapter's focus was not on survey error, but on response rates themselves. But, the E/M report indicated that they followed up with additional interviews of their exit pollsters and I have to wonder if they asked the party affiliation question (pg. 52).  

            I wonder if subconscious non-random self selection of "like minded" voters by interviewers can explain a chunk of the persistent democratic bias since 1988.  

            HOWEVER - it DOES NOT explain a large chunk.

            The E/M report states:

            ...the data show a WPE in the Kerry direction still exists even in precincts where the interviewer was instructed to ask every single voter to participate in the exit poll (and the interviewer had no option in the selection of the respondent).  In precincts with an interviewing rate of "1", there was still a WPE in the Kerry direction of almost 4 points.  Again, this indicates that a portion of the WPE is coming from differential non-response (pg. 35).
            Differential non-response most directly measures the "reluctant Bush respondent" hypothesis.

            Lunch is over, so I cannot get into that...  This is fun, but back to reality - my day job.    

      •  Also a psychological factor (none)
        I honestly think a lot of people were ashamed of their vote for Bush, especially in front of the kind of people who take exit polls. I've been exit-polled twice, and both times it was by earnest but personable, attractive young women. I had the feeling that they were pleased by my "liberal" responses, and I can easily imagine a sneaky Bush voter lying to them just to make for a more comfortable interview.

        Anybody else experience this? Or was this just the case where I vote, on the (ultra-liberal) Upper West Side of Manhattan?

        •  We regularly have "shy Tories" here (none)
          pollsters allow for them.  Embarassing to be a Tory.

          And our count (until this year) was pretty clean.

          So the error was pretty clearly in the polls.

          But now they have postal voting on demand!  And the Labour Party have been stealing sacks of votes!


          See my new Exit Poll diary here

          by Febble on Wed Apr 06, 2005 at 03:45:13 PM PDT

          [ Parent ]

  •  Why? (none)
    Why do you use a 56-50 split?  Why not a 53-47 split?

    Or just assume that for every Bush voter you poll, you will poll 1.12 Kerry voters?

    •  Just a convenience. (none)
      It's the ratio that's important.  I just thought it might make sense written that way.

      In fact doing the math, I expressed it as a ratio with 1.  So the 56:50 ratio is a 1.12: 1 ratio. 53:47  would be a 1.13:1 ratio.

      What I wanted to do was to infer the ratio from the WPE and the count.  I then took a log of the ratio in order to make it both linear and symmetrical around 0 .

      •  Right (none)
        I understand the 1:1.12 ratio.  But it got confusing when you started off by saying you were going to poll half the voters, and then polled over half of them.

        Also, if the ratio is constant, then you'll have no WPE differential based on how strong a precinct leans one way or the other.

        The determining bias would be sample size.  The bigger the precinct, the bigger the bias.

        Unless I'm missing something, which I probably am.

        •  Sorry, I agree that is unclear. (none)
          I was trying to say that if you polled half of each you wouldn't get a bias, but if you polled a few extry Kerry voters you would.

          Trying to describe stuff like that is really tricky, which is why I spend my time on DKos instead of getting down to the tedious task of trying to explain a computational model of an arcane visual illusion to an external examiner who doesn't know anything about vision theory.


          Trying to figure out your last comment.  The bigger the sample, the smaller the sampling error, but for a given sampling bias (ratio) the WPE will be the same regardless of precinct size, no?  Am I missing something?

          But it will be different for different degrees of partisanship.

          See my new Exit Poll diary here

          by Febble on Wed Apr 06, 2005 at 07:38:44 AM PDT

          [ Parent ]

          •  This is why (none)
            I am questioning your assumptions.

            What you are really saying, I think, is:

            Regardless of precinct size or partisanship...we are assuming that for every Bush voter you succeed in polling, you will succeed in polling 1.12 Kerry voters.

            This bias is unintentional and unaccounted for.

            Therefore, in a precinct that voted exactly 50-50 you would get a 56-44 split for Kerry.  

            I'm with you so far.

            But it seems to me that the assumption is flawed if you extend it precincts that voted 80% one way or the other.

            I think the 1:1.12 ratio would be an aggregate average, and not a number that would hold steady.

            •  This is not hypothesis, (none)
              this is math!

              The graph just came out of the math.  Try  it.

              There are no assumptions.  I just modelled the effect of a given ratio on the WPE under different partisanship conditions.  Adjusting the bias slightly skews the graph.  Eliminating the bias makes it flat.

              It's just algebra.  No biggie.

              But it means we can clean up the WPE figures (which are derived in part from the partisanship) and get a "pure" measure of bias that is mathematically independent of partisanship.  The interesting finding is that it varies systematically and linearly with partisanship.

              No assumptions were injured in the making of this diary.

              See my new Exit Poll diary here

              by Febble on Wed Apr 06, 2005 at 01:49:11 PM PDT

              [ Parent ]

              •  Isn't your assumption (none)
                that there was a bias in favor of Kerry voters agreeing to participate?

                And isn't your graph merely an exercise in seeing what happens when that assumed bias is set against differential partisanship?

                If not, then I need a drink.

                •  No, the math works the other way too (none)
                  I just used a Kerry bias in the example.

                  If you look at the scatter plot you will see that the bias index is negative for some states, indicating a Bush bias.

                  But the mean bias was in Kerry's direction

                  The math is symmetrical.

                  And yes, the graph was precisely an exercise in seeing what happens when that assumed bias is set against differential partisanship - I could see that constant bias would generate different WPEs according to partisanship, so I graphed it to make sure (and to show you guys - sometimes it takes a picture to make the point)

                  Then I did the math to get a bias figure that would be independent of partisanship, so that when comparing states (unfortunately we cannot compare precincts as we don't have the data) we are comparing apples with apples, not apples with oranges.

                  The point is that without converting the WPEs to a bias index as I have done, you could have the illusion that bias was greatest in evenly matched precincts.  But this is just an artefact of the WPE computation.  All the notional precincts graphed have the same bias.  But the WPEs are different.  So we need a statistic for bias.

                  That's what I formulated.

                  Have a drink.

                  (And I like your blog - quiet discussion going on over there too!)

                  See my new Exit Poll diary here

                  by Febble on Wed Apr 06, 2005 at 02:16:13 PM PDT

                  [ Parent ]

                  •  Sample bias v. Sampling error (none)
                    Febble, I think the root of some of the confusion here (and it had me confused too) is that not only does WPE fail to provide a "clean" measure of sample bias (response and non-response bias), it is also contaminated by within precinct sampling error.  I believe this is what I have been trying to communicate, but haven't quite been able to say it just right.  

                    Now, you have a dependent variable (for now it could be WPE or your "bias index") that you are trying to explain.  Assuming a perfectly random within precinct sample, there is still a margin of error of roughly 10% (sample size of ~100).  

                    However, at the extremes of the D-R or R-D vote, this random sampling error has practical limits (that is, tails of the normal distribution are sliced off). I wrote about this here on 4/2 and I am told that it was discussed by pollsters and other data dweebs on AAPORnet in January, but will summarize the effect here.

                    The "apparent" positive correlation could be explained largely by sampling error.  If the vote is already 95-5 in Kerry stronghold, the "random sampling error" could only cause the discrepancy to go up a max. of 5%, but down %10.  If it is 95-5 in a Bush stronghold, the sampling error could only cause the discrepancy to up a max. of 5%, but down 10%.  Since the E/M Report reported WPE by Democratic precinct partisanship, this means that the error is forced to show higher error in Bush strongholds, than Kerry strongholds.

                    Now, add the possibility of non-sampling error (your biased sample selection) and the effect is magnified.  

                    Basically, the table on page 36 of the NEP Report shows an apparent positive correlation, with little error in Heavy Dem precincts (because Kerry can't go higher) and a lot in Heavy Rep precincts  (because Bush has nowhere to go but down).

                    What does this mean for your analysis?  Well, I'm not sure yet... I have to head to class now...

                    Question to anyone: Is Febble's new dependent variable "bias index" any less subject to the practical limits of within precinct random sampling error than WPE?  

                    5 minutes of thought here and there is not the best way to approach this type of stuff...

                    •  Extremely difficult to reconstruct ... (none)
                      ... the underlying distributions.

                      I'm tempted to say "impossible", but that might be stretching it a bit.

                      All we have are means (and medians, IIRC) of arbitrary segment of a distribution of a nonlinear function of the stuff we're trying to study.

                      If we had reason to think the data resulted from some underlying well-behaved smooth distribution, we could try to reconstruct that distribution (best fit to all we know), and then model which points fell into which segments of the sampled and tabulated distributions, and then extract relationships.

                      But the underlyuing distributions are lumpy, we don't always know sample sizes, we know survey takers deviated from protocol, and one of the UNknowns is the smoothness of the underlying phenomena.

                      Precinct data would be useful, individual data would be useful, but there are compelling privacy reasons not to release data at this level.

                      At the other end of the scale, I see no compelling reason to think there's anything very interesting going on here, and most of the people who will argue to the contrary have shown they will argue to the contrary no matter what the evidence.

                      •  Great thoughts, RonK (none)
                        And my sister is graduating from UW in June, so I like Seattle ;-)

                        I raise this question because I'm not yet convinced that febble's formula [Bias Index - log [(R/D) x (1+margin-WPE)/(1-margin+WPE)] does what she think it does.

                        She wrote previously, "I don't think it is valid to use a dependent variable (the signed WPE) that is contaminated by the partisanship of the precinct, particularly when one of your predictors is partisanship of the precinct."

                        I guess I'm not sure that her "bias index" is completely free from "contamination" by the partisanship of the precinct.  I think that precinct partisanship is NOT a predictor of WPE even though the NEP Report table suggests that it is, at least partially because of the practical limits of within precinct sampling error.  

                        E/M clearly do not suggest that precinct partisanship predicts WPE.  In fact, they acknowledge the practical limits in the extreme precincts and say that WPE is relatively constant in the more balanced precincts.  Again, ANOVA or regression would be nice, which would probably require exclusion of the 90/10 or more precincts on both sides (can't imagine there would be too many of these).  

                        feeble, can you provide some more logic/explanation for why your formula accomplishes what you say it accomplishes.  I understand the formula, but I don't understand how it decontaminates the error from precinct partisanship completely.  Remind you, I only have a graduate level stats background and no real solid math or theory, so a decent "dumbed down" explanation of how your formula does what you say it does would be nice ;-).  

                        •  More spreadsheets on their way! (none)
                          Hope they convince you.

                          And, guys, I'm NOT feeble!  

                          It's from my initials, febl.  I'm Febble!

                          I wish I'd chosen something different now.

                          See my new Exit Poll diary here

                          by Febble on Thu Apr 07, 2005 at 03:06:04 AM PDT

                          [ Parent ]

                          •  Sorry febble, I wrote your name too quickly... (none)
                          •  That's OK (none)
                            I do it myself sometimes ;).

                            It's just that some jerk upthread was using it to piss me off.

                            See my new Exit Poll diary here

                            by Febble on Thu Apr 07, 2005 at 07:07:39 AM PDT

                            [ Parent ]

                          •  Does population density predict bias? (none)
                            Most of the extreme Bush states are rural, Kerry states are urban. E/M found low WPE in rural precincts.

                            Maybe there's a "company of strangers" psychology that works differently in R's and D's in urban settings. (I've never thought that "shy" was a good characterization of subject selection bias - and it certainly doesn't touch on active selection bias and protocal error by the survey-taker.)

                          •  I think that would be a good question (none)
                            unfortunately we do not have the breakdown of WPE by anything useful.

                            Agree about "shy".  My guess is that over-eagerness is at least as much of the picture.  Random selection is incredibly hard, which is why social scientists (and cognitive scientists) have to be so careful.  I use the rand function constantly on my computer!

                            And these interviewers simply weren't trained properly.  It is hard enough to get psychology students who are supposed to know some stats to see that their stats are entirely dependent on the assumptions of randomness.  Confounds are everywhere.

                            Honestly, training by telephone! I ask you.

                            See my new Exit Poll diary here

                            by Febble on Thu Apr 07, 2005 at 09:36:29 AM PDT

                            [ Parent ]

                          •  RonK, very true... (none)
                            Ahh you see that as well.  Good.  That is why I question the survey error (whether WPE or febble's bias index") by precinct partisanship relationship.  

                            One way to explain this is the WPE by precinct partisanship is to look at the suburbs, assuming they are Bush strongholds.  But as a City Planner, who does housing and demographics analysis for a living, I know that it is no longer true that suburbs are homogeneous, and certainly not politically homogeneous.  It would not seem that High Rep precincts, while possibly in Dem safe states, would be in highly urban areas.  That means that Bush strongholds are in rural areas (where WPE was lowest) and Dem strongholds are in urban areas (where WPE was highest).

                            So, that leaves us with higher differential response/non-response in Dem urban strongholds.  

                            How do we explain this?  Certainly the US Count Votes "fraud in Bush strongholds" does not explain the WPE by population density.  

                            Perhaps this can be explained more by "embarrassed" and/or "intimidated" Bush voters as well as "reluctant" Bush supporters in the high Dem precincts.  

                            For example, in precincts where Democratic party affiliation was highly clustered, perhaps the environment was so intense that lifelong Democrats, whose family and friends did nothing but harp on how terrible and evil Bush was, were either embarrassed about their vote (maybe they agreed with the war on terrorism or didn't like gay marriage) or felt intimidated enough that when first approached, either rejected to take the poll (which should show up in the refusal rate by party vote, but doesn't seem to), or lied on the poll itself for fear of being outed (shows up in poll error, but not response rates).

                            I read a while back, perhaps it is in the Grove et al text (2002), that refusals to partake in a poll are often visceral reactions.  That is, after given some time to think about it, many respondents who refused the first attempt, will participate if asked again.  In an environment like 2004, if you weren't 100% sure of the anonymity of your response, perhaps when some kid approaches you asking you who you voted for, you might brush them off.

                            Likewise, in these higher Dem precincts, because of their higher populations, there might have been more opportunities for non-random sample selection.  That is by larger gaps between sample intervals, as well as the greater chance that multiple precincts were housed under one roof, making it really hard to determine which "nth" was from your precinct.  Perhaps the younger and highly educated interviewers (sounds like Kerry demographic to me) subconsciously approached people who were more like them, which of course would not show up in response rates.

                            It would be interesting though to see if discrepancies between vote tabulations and voter party registration in these highest Dem precincts were correlated with response rates.  It would also be interesting to see if there was a difference between the vote tally-party reg of highly Dem precincts in other years (1988 to 2000).  But then again, if there was, people would say this was evidence of fraud in these precincts (even though it goes against the fraud in Bush strongholds hypothesis).

                            I think there is a lot of psychology at play here both on the part of reluctant/embarrassed/intimidated respondents and the part of biased non-random within precinct sample selection by interviewers.  

                          •  'you might brush them off' - key point (none)
                            Urbanites become accustomed to brushing people off. All kinds of people, with and without clipboards.

                            I expect there's a whole world of nonverbal interaction between the poll-taker and the prospective respondent.

                            And as I've probably mentioned before, Republicans are conditioned to see the "liberal media" as "the enemy", and E/M poll takers were plastered with media logos. I'm a bit surprised the differential response rate isn't worse.

      •  Just a convenience (none)
        Not to quibble, but that is the exact same split Edison/Mitofsky used, and the apparent reason they used it is because it accounted for virtually 100% of the variance between the exit polls and the actual count.
        •  Yes, that's why I used it in the example (none)
          I say that elsewhere (I wasn't trying to be disingenuous).  I then set up the model, and adjusted the ratio to check the effect of increasing and reducing the ratio.

          But it was fairly easy to re-write the equation to express the bias in terms of the WPE and proportions of votes cast.  The clever bit (though I say it myself) was taking the log so I could do the stats with a linear symmetrical scale.

          See my new Exit Poll diary here

          by Febble on Wed Apr 06, 2005 at 01:43:49 PM PDT

          [ Parent ]

  •  The Argument that Rep/Bush voters were less likely (none)
    to be interviewed by exit pollsters causing an oversample of Dem/Kerry voters to be exit interviewed throwing off the exit polls, has been universally and throughly discredited by a poster named Truth is Absolute (TIA) over on Democratic Underground in the sub-heading 2004 Election Returns - Presidential Politics.  He has taken on all comers with his statistical analysis.  Have you been there and compared your case to his?  Before this happened, I would never have expected that this large of a "fraud" could possible take place undetected.  However, now I see that ANY POSITION can be argued and proven statistical with logic and factual support, so successfuly that NO ONE even is bothered to INVESTIGATE IT.  hahahahahahahahah!

    Fraudsters don't think exit polls were the "smokin' gun" - but they do think that they create a POSSIBILITY THAT REQUIRES A THROUGH EXAMINATION of the underlying facts and proof.
    The likelyhood of 5.5% Margin of Error on a Sample size of greater than 13,600 with a published MOE of 1% will cause most to just cringe.  They know that mathmatically it is almost incalcuable!  Why vote ever again - I'm not going to bother.

    •  'Why vote ever again - I'm not going to bother.' (4.00)
      This is the primary effect of the Fraudster movement: increasing resistance to registration and GOTV among marginal Democratic voters.

      IOW, granting Republicans a persistent Election Day advantage.

      A secondary effect is increasing ballot spoilage among Democratic voters. Experienced voters are successful voters.

      •  In some ways (4.00)
        We are conducting a voter supression effort on ourselves.

        I absolutely think this occured here Florida.

      •  Blaming the victim (4.00)
        Sorry, but I think it's the people railing against those calling for an investigation into such irregularities who are increasing resistance to registration and GOTV among marginal Democratic voters.

        I mean, if Dems can't fight to protect the integrity of the ballot box, what good are they? If an issue can't be made over partisan control over unauditable voting systems what's the point? Seriously.

        Point: We can't know that our votes actually counted. And now when we point out glaring inconsistencies we're told to take our tin-foil hats and go away because were scaring away the marginal Dem supporters? Riiiiight. It's our fault that people are losing faith in the system. If we would just shut the hell up about a too-easily rigged closed-to-oversight system the Dems might get enough supporters to . . . what? Overwhelm the ability of the Republicans in control to rig the count?

        Sure, great plan.

        The polls don't tell us how a candidate is doing, they tell us how the media is doing. And Diebold tells us who won.

        by Thumb on Wed Apr 06, 2005 at 11:06:16 AM PDT

        [ Parent ]

        •  You've never done field work, have you? (none)
          •  And when people ask you if their vote counts? (none)

            The polls don't tell us how a candidate is doing, they tell us how the media is doing. And Diebold tells us who won.

            by Thumb on Wed Apr 06, 2005 at 11:30:53 AM PDT

            [ Parent ]

            •  So you've never done registration or GOTV? (none)
              Where the %$#@ were you in 2004?????

              And what makes you a %$#@ing expert on what real people don't think?

              Yes, tell people their votes will count.

              Tell them the best way to vote successfully is to practice voting in every election ... including the minor elections.

              And tell them anybody who says otherwise is a fool, trying to get you to give up without a fight and -- whether they mean to or not -- helping the other side.

              •  asdf (none)
                I'll tell people their votes are counted when we have an electorial system that's not run outside of any public review by admitted partisan activists. I'll tell people their votes are counted when Republicans stop roadblocking efforts to provide a paper trail. I'll tell people their votes are counted when I see the electorial process move beyond a charade. I'll tell people their votes are counted when we have more than the word of Wally O'Dell, Howard Ahmanson jr. and a handful of dKos regulars that they were indeed counted. Otherwise perpetuating the myth of "of course your vote is counted" does nothing to change the fact that WE DON"T KNOW IF OUR VOTES ARE COUNTED!!!

                You want to fight Dem/swing apathy? Start by finding a way to insure our votes are counted. THAT'S the fight we need to fight, and avoiding that fight is doing a %$#@ of a lot more to help the other side than pointing out that their partisans control the vote counting process.

                Because losing every close election, over and over, to come-from-behind Republicans who all just happen to defy the polls will depress the base more than pointing out that the game is rigged and needs to be repaired.

                The polls don't tell us how a candidate is doing, they tell us how the media is doing. And Diebold tells us who won.

                by Thumb on Wed Apr 06, 2005 at 01:52:23 PM PDT

                [ Parent ]

                •  zxcv (none)
                  You want to fight Dem/swing apathy? Start by finding a way to insure our votes are counted. THAT'S the fight we need to fight, and avoiding that fight is doing a %$#@ of a lot more to help the other side than pointing out that their partisans control the vote counting process.

                  You want to ensure our votes are counted? Start by overcoming Dem/swing apathy and get people voting in every election, even the minor elections, so that we can get state governments and officials in places like Ohio who will actually be responsible and work to ensure proper reforms are made.

                  It's kind of a catch-22.

                  But you know what, that's the wrong way to look at it. It's not an obstacle. It's actually a feedback loop which could be EXTREMELY good for us. Get it started and the rest will follow. We just need to get the ball rolling.

                  How do you get the ball rolling? One of these goals is clearly less difficult than the other and is the obvious candidate for attempting to tackle first. Which do you think it is?

                  • Get reform here in Ohio while Taft is governor, the state legislature is controlled by Republicans, and that criminal bastard Blackwell is Secretary of State.

                  • Get Dems mobilized and voting in every election, including small ones.

                  If you said A, let me know how you plan to do that. If you said B, run, don't walk, over to your next local DFA meetup.

                  The Americans will always do the right thing... after they've exhausted all the alternatives. - Sir Winston Churchill

                  by drewthaler on Wed Apr 06, 2005 at 02:38:59 PM PDT

                  [ Parent ]

                  •  A third option? (none)
                    How about a ballot initiative. I've tried on more than one occasion to suggest that a ballot initiative that called for two simple things. 1) mandatory paper ballots (for both a meaningful audit AND a means to avoid 10 hour lines at three touch screen machines) and 2) just as multiple observers used to observe the vote count when paper and pencil were the order of the day multiple parties should be allowed to provide their own ballot scanners at polling stations. Imagine if once a person filled out their ballot was scanned through a Diebold (or ES&S or Sequoia) machine and also a machine provided by the Dems and/or another provided by a media consortium. If there was any discrepancy between the machines a hand audit would be mandatory.

                    First, no one would ever have to worry that there vote didn't count. Second, the Republican controlled Houses could be bypassed by a ballot initiative and third, considering that this wouldn't cost the tax payers anything it would be hard for the Republicans to fight.

                    Just a thought.

                    The polls don't tell us how a candidate is doing, they tell us how the media is doing. And Diebold tells us who won.

                    by Thumb on Wed Apr 06, 2005 at 03:57:10 PM PDT

                    [ Parent ]

                    •  To Canadians, Australians and Brits (4.00)
                      the answer is blindingly obvious.

                      Hand counted paper ballots.  No problem

                      See my new Exit Poll diary here

                      by Febble on Wed Apr 06, 2005 at 03:59:09 PM PDT

                      [ Parent ]

                      •  Gee, (none)
                        I wonder why Republicans are fighting such common a sense approach? Could it be . . . naaa. I can't imagine they'd take advantage of their proprietary control over the counting of our votes to cheat. That's just to tin-foilish to consider.

                        The polls don't tell us how a candidate is doing, they tell us how the media is doing. And Diebold tells us who won.

                        by Thumb on Wed Apr 06, 2005 at 04:28:15 PM PDT

                        [ Parent ]

                    •  that's an option (none)
                      Doesn't have to be a third option, it can be your answer to A. :-) Ohio's ballot initiative process is described here (pdf). The requirements are somewhat complicated and would require a statewide org to push it, but it could be done. There are two ways to go about it - as a statute or as a constitutional amendment.

                      A petitioned statute would go to the legislature first, which can simply vote it down. If that happens it never makes it to the ballot. $50 says that's what would happen, with all sorts of justifications and rationalizations made as to why. The current budget problems are likely candidates.

                      A petitioned constitutional amendment goes directly onto the ballot. Of course, the SoS and the AG must approve the wording and you can't appeal their decision if they stonewall. And then the Ohio legislature may freely repeal or amend any initiative which passes. But there's probably a better chance of doing something that way.

                      Still, even with all the ways they could and will try to block it, a sensible and well-written initiative could at least propel it into the state and national news so that they can't attempt to bury it behind closed doors (which seems to be their usual modus operandi).

                      The Americans will always do the right thing... after they've exhausted all the alternatives. - Sir Winston Churchill

                      by drewthaler on Wed Apr 06, 2005 at 10:34:06 PM PDT

                      [ Parent ]

              •  I don't want to lose another election, (none)
                so therefore even if the last one was stolen, I'll just smile and pretend nothing's wrong.

                If the majority really voted our way last time, we wouldn't want them to know about it, because then they would switch to the team that stole their votes and we would lose another election.

                So let's all bury this unseemly topic.

        •  Let's be very clear about this (none)
          I firmly believe that the real issue is voter supression, not fraud.  In 2004 in this state some serious attempts were made to address this.  To wit:
          *Suing to get the felon's list. Ultimately this meant that it was exposed as a fraud, and went almost completely unused.
          *Putting together a team of lawyers to poll watch.  This was an immense accomplishment. They were everywhere (I helped) and credit should go to the Kerry effort for helping to put the effort together. As a result, there were far fewer challenges (near zero)at the polls in 2004 than in 2000.  

          Turnout in this state in 2004 was HUGE.   This was in part a result of the efforts to counter voter supression.

          But we were fighting upstream in one sense.  The disaster of 2000 led some voters to conclude that their vote didn't count.  The continuing drum beat of fraud allegations about the voting machines unquestionably helped to reduce turnout.  And it was OUR turnout that got hurt.

      •  Parallels (2.33)
        I'm beginning to think there's a concerted disinformation effort here.

        It's just like the Kennedy assassination discussions.

        People will CLAIM (falsely) to be "conspiracy theorists" (read "fraudsters" here) and then turn the gun on their own and say but I don't believe "x" supports conspiracy, even though x could be a plain, hard fact.

        In the JFK assassination discussions a group of lemmings will come along and support this person, pretending to be part of the flock. But actually, all of them know each other, communicate outside the forum, and are engaged in keeping the truth from surfacing, rather than promoting it.

        I've seen this pattern for years. I'm not really surprised to see the same pattern on Kos. It's a widely read forum and the same tactics have to be used to keep the truth from the people.

        But Feeble's analysis matches his name. It cannot explain the FACT that the exit polls used correct methodology that correctly predicted Senate races, and, without a change in methodology and having interviewed the very same people, was incorrect on the Presidential race. The only way to explain that would be to say people told the truth about one of their ballot options but lied on another. Sure, that's a theory. But it has no basis in reality.

        Neither does this flimsy hypothesis. I can tell from the posts that people are jumping on a bandwagon because they don't want to do the hard work to read the REAL analysis by REAL PhDs on how the exit poll methodology was without fault, and how the possibility of fraud remains the most likely conclusion. Read the actual report. Then judge. You'll have real questions about this person's true agenda here. Or bury your heads at our collective peril.

        •  It's Febble, not feeble (none)
          and I'm a girl.

          And with luck I'll have a REAL PhD in a couple of months.  I also contributed a little to the UScountsVotes report, which I support.  But declined to sign, for minor reasons given elsewhere.

          I agree the Senate races are interesting, but any fraud hypothesis does have to take account of this data.

          I am not ruling it out.  I repeat: all I am saying is that the Exit Poll Discrepancy per se does not look like evidence for fraud.

          There may well be evidence elsewhere.

          And I think UScountsVotes are doing a good and necessary job.

          See my new Exit Poll diary here

          by Febble on Wed Apr 06, 2005 at 12:35:16 PM PDT

          [ Parent ]

          •  Looking for evidence elsewhere? (none)
            Read this link talking about the NC discrepencies between the 30% of the people that voted absentee vs the 70% that went to the polls.

            The diary points out that 30% is a HUGE sample to measure the accuracy of the voting machines against (touch screen machines without paper trail I might add) and that the 30% is evenly spread across the political left and right.

            Sample: ["Poll" means electronic voting machine on Nov 2]

            LIEUTENANT GOVERNOR (Absentee)
            Beverly Eaves Perdue (DEM): 561,584 (55.7%)
            LIEUTENANT GOVERNOR (Poll Only)
            Beverly Eaves Perdue (DEM): 1,326,798 (55.5%)

            SECRETARY OF STATE (Absentee)
            Elaine F. Marshall (DEM): 575,045 (58.0%)
            SECRETARY OF STATE (Poll Only)
            Elaine F. Marshall (DEM): 1,336,525 (57.0%)

            ATTORNEY GENERAL (Absentee)
            Roy Cooper (DEM): 546,477 (56.7%)
            ATTORNEY GENERAL (Poll only)
            Roy Cooper (DEM): 1,323,222 (55.2%)

            All fairly close, as are the rest of the examples of down ticket races and constitutional amendments. But then lets look at the senate race:

            SENATOR (absentee)
            Richard Burr (REP): 492,166 49.48%
            Erskine Bowles (DEM): 492,536 49.52% .04
            SENATOR (Poll only)
            Richard Burr (REP): 1,299,294 52.4%
            Erskine Bowles (DEM): 1,139,973 46.0%

            Thats a much larger margin between absentees and machine counts compared to the rest of the races, except for the presidential count:

            PRESIDENT (absentee)
            George W. Bush: 529,755 52.9%
            John F. Kerry: 469,522 46.9%
            PRESIDENT (Poll only)
            George W. Bush: 1,431,433 57.3%
            John F. Kerry: 1,056,299 42.3%

            So what do you make of that large of a discrepency? If this doesn't stink I don't know what does.

            The polls don't tell us how a candidate is doing, they tell us how the media is doing. And Diebold tells us who won.

            by Thumb on Wed Apr 06, 2005 at 02:15:49 PM PDT

            [ Parent ]

            •  Look, I'm not saying there wasn't fraud (none)
              I'm just saying that I don't think the evidence is in the Exit Poll discrepancy.

              You are investigating in the right places.  I am concluding that it is unlikely that the nationwide exit poll discrepancy is the right place.

              I think the election stinks.  It's why I am writing diaries about it on DKos instead of writing up my dissertation or out canvassing for the UK Labour Party!

              (Oh, and I think voter suppression in Ohio stinks to high heaven).

              See my new Exit Poll diary here

              by Febble on Wed Apr 06, 2005 at 02:19:54 PM PDT

              [ Parent ]

        •  A few facts (none)
          Contrary to Freeman's unsupported assumption, US voters routinely cross over between Presidential and Senatorial races. A large fraction of Senate D's are from red states, and a significant share of R's are from blue states. Suggested maxim: if you can't count, don't hang your hat on higher statistics.

          Freeman refers (slyly) to correct prediction of Senate "outcomes", meaning "winners" ... but few Senate races are close races. Most races were never in doubt, and few of the contested races were close enough that you'd have to use statistics to call 'em.

          And especially in 2004, in almost every one of the arguably competitive races, the Democratic candidate ran away from his/her party's Presidential candidate. (AK, OK, KY, FL, SC, SD, CO ... did I miss any?)

    •  The bias within the MoE (none)
      But that does not mean the error was in the count.

      The MoE gives you the sampling error, assuming unbiased sampling.  If you tossed a coin 100 times you would get approximately 50 heads and 50 tails. But it would vary each time you did it.  This is called sampling error.  It can be quantified statistically.  But if the coin is weighted, you may tend to get more heads than tails.  This is bias.  It is not sampling error.  The statistical tests enable us to conclude that in the 2004 election the the sampling was biased (or the vote-count was). The bias was outside the MoE. There is simply no argument about that.  My analysis suggests that it was the sampling, so you are safe to vote.

      But you cannot prove statistically whether it was the vote or the count that was wrong.  It is not what this kind of statistics does.  Although some patterns may be more suggestive of one than the other.  I maintain the pattern is much more suggestive of sampling bias.  Can you give me a link to your TIA source?

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 08:39:35 AM PDT

      [ Parent ]

  •  Apologies for earlier lack of pictures (4.00)
    Thanks to Timroff and DemfromCT it seems to have made the Rec list!

    See my new Exit Poll diary here

    by Febble on Wed Apr 06, 2005 at 07:27:42 AM PDT

  •  Forget 'RECOMMENDED'--FRONT PAGE THIS NOW!!! (4.00)
    I, too, am a fraudster in the sense that Febble is. And I am very glad to see this analysis. I have long been troubled by the talking-past-each-other nature of the exit poll debate, though I haven't focused on it much, since I think the observable actions of the principals--most notably Blackwell--is prima facia evidence that demands further investigation.  This diary conclusively closes that door on that debate, IMHO.  

    There is nothing more damaging to critical enterprises, such as election fraud investigation, than pouring enormous effort into blind alleys. This diary should be read by everyone who visits this site.  

    We need to get past the blind alleys and focus our attention on the real and provable issues, which are, of course, part of much more widespread patterns of abuse of power.  The 2004 voter suppression/election fraud matters for the same reason the 2000 stolen election matters--because it is part of a larger ongoing pattern that threatens the very existence of our democracy.  The very seriousness of the threat we face demands the utmost rigor on our part in fighting against it.

    This diary is an invaluable contribution for that reason--even though it is not the news that many of us would want.  It tells us definitively not to continue putting energy into this one direction, and it is extremely important to realize this.  We are, after all, the reality-based community, and we must always rely on that as our strength.

    •  I agree (4.00)
      but the diary first has to pass 'peer review'.

      Works both ways. ;-)  And I'm certain Febble agrees.

      "Politics is the art of looking for trouble, finding it everywhere, diagnosing it incorrectly and applying the wrong remedies." - Groucho Marx

      by Greg Dworkin on Wed Apr 06, 2005 at 08:15:41 AM PDT

      [ Parent ]

    •  Invaluable contribution? (none)
      This diary ignores the facts to promote a theory. If the exit poll methodology was wrong, it would have showed up in the Senate race predictions. But it didn't. The "error" ONLY appeared in the presidential race.

      An analysis that ignores not theories but hard core facts is hardly invaluable. It's hard to find any value, in fact, in such a theory.

  •  Another good one. (none)
    As you may have noted, I didn't mean to reference your diary yesterday when I complained of repetition  in diaries on the report.

    I enjoy good independent analysis, and you present your case well.

    I'll recommend this one, as I did with the previous one.

  •  Inefficiency is not the biggest drawback (none)
    Re "state-level fraud targeted somewhat inefficiently at the bluest states", it may help to look at fraud scheme risk/reward from the would-be perpetrators point of view.

    Every fraudulent intervention entails a risk of direct exposure (via technical inadvertence, human behavior or unexpected audit). Since votes are tabulated at county level, pervasive fraud requires fraudulent intervention at thousands of separate points.

    Each point compounds risk of exposure -- with attendant surrender of the whole reward, and of future rewards, plus additional penalties.

    Very few points of intervention produce incremental proability of reward (which starts out in the neighborhood of .5), and these interventions have rapidly diminishing marginal returns.

    In brief, no conspiracy (even a "lone gunman" conspiracy) would do it this way even if they could (which, as all things, is more difficult in practice than in theory).

  •  Add'l notes 2004 as outlier (4.00)
    2004 as the first cycle (at least the first cycle in a very very long time) that E/M conducted the exit polls ... following the collapse of VNS and son-of-VNS exit polling consortia.

    Anecdotal interviews with individual poll-takers indicates

    1. they didn't get much training
    2. they didn't follow prescribed protocols
    3. they never understood why they should follow protocol, even when it was explained to them after the fact.

    In this light (and also in consideration of Perot effects in 92 and 96), I do not find it surprising that 2004 does not align with bias in previous cycles.
  •  Point is (4.00)
    BUT THE Point is we still don't know!!!

    Sure point 1 of your hypotheses may be more likely than 2 and/or 3 but it doesn't rule them out.  That is what we have been saying.

     Tha canary is dead in the coal mine and you want us to believe it died from natural causes, well maybe it did, but just maybe it didn't. Can we at least investigate?

     Let's look at the other facts

    1. We live in a country that has a rich history of election fraud
    2. We have new electronic balloting that is easily , yes easily, manipulated. One programmer has already testified in court that he was asked to write just such a program
    3. For some reason in 2004, the simple process of getting paper backup for our electronic voting is too much to ask, although the same compnay makes atm that do this everyday.
    4. The makers of the machines are just two companies. Two VERY partisan companies.
    5. Exit polls caught fraud in another Nuclear power just this year and successfully allowed an election to be overthrown.

     What we are asking , and have been asking , is that can we have some paper backup, can we have some investigations to help clean up the fraud we KNOW happens in every election(intimidation, registrations thrown out, polling places moved last minute, lack of machines in districts etc). WHAT we are asking are simple tools to allow to american voter to have SOME confidence that his/her vote count, and are counted correctly.  

    Is that too much to ask?  Personally if the government doesn't take these voting issues seriously, why should we taking voting seriously?

    •  You shouldn't have to argue these things (4.00)
      but it takes two to tango and the left is as fucked on voting fraud and reform however well meaning. Just look at Kerry's concession, congress' silence (with a few notable exceptions) and the left blog's intransegence on the issues. waddya gonna do. They're tanging us into tyranny.

      Darkness washed over the Dude...darker than a black steer's tookus on a moonlight prairie night...there was no bottom

      by moon in the house of moe on Wed Apr 06, 2005 at 09:45:53 AM PDT

      [ Parent ]

  •  The real world model..... (none)
    that your analysis fits is:

    A) Republican voters are liars


    B) Republican voters lie more and more over time.

  •  How can you assume 56/50 for example? (none)
    in all precincts?

    Wouldn't it require a significantly greater oversampling of Kerry voters to sample 56% of them in a precinct that went 80% for Bush, compared to a precinct that was split 50-50? In other words, would the same flawed polling technique get the same 6% oversampling in an 80-20 precinct as in a 50-50 precinct? I think not, based on what happens at the limit, a 100% Bush precinct: as the opportunity to interview Kerry voters (a function of the number of Kerry voters in the precinct) approaches zero, the oversampling percentage of Kerry voters must also approach zero. (If a given Bush voter is shy, there is an increasingly smaller probability that the first shy voter will be replaced with a Kerry voter.) Maybe I'm being too simplistic, but I think there might be a problem in assuming that a greater WPE is the result of a constant oversampling of Kerry voters; in fact, I think that an apparent constant oversampling of Kerry voters across precincts (resulting in an apparently higher WPE in red precincts) is itself indicative of fraud (a constant shift or spoilage of Kerry votes, possibly, as an alternative theory to a constant oversampling of Kerry voters), because a consistently flawed polling technique should produce less effect in red precincts.

    Pipe dreams are not an exit strategy.

    by TrainWreck on Wed Apr 06, 2005 at 09:46:57 AM PDT

    •  I am not assuming anything (none)
      All I am saying is that a uniform sampling bias would give a non-uniform WPE.

      The whole point of this analysis was to generate an index of bias that was independent of partisanship, so that we could assess whether the bias was uniform or not.

      56:50 was just an example, and happens to be the one that EM postulate.  What this analysis shows is that the bias was not uniform, but that it has a significant pattern, being more greater the more Democratic the state.

      It could be fraud, it could be sampling bias. But either way, you have to explain this pattern.  To me, it seems more consistent with sampling bias (over-eager Democratic voters in strongly Democratic states wanting to get their votes into the poll) than with fraud (why would you target fraud in the Blue states?  Why not the swing states where you'd get more bang for your buck?  Why not the red states where there are more Republicans to cover your back?).

      In a 100% Bush precinct I agree, the thing crashes.  However, as long as your samples are big enough, even a 99% Bush precinct will show the bias.  But that's the point.  The bias will tend to look smaller the more partisan the precinct, and also smaller in precincts in which the partisanship is with the undersampled candidate.

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 10:02:51 AM PDT

      [ Parent ]

      •  I should have said (none)
        how can you assume a uniform sampling bias?

        I don't believe that a uniform sampling bias would produce a uniform rate of of oversampling of Kerry voters.

        And I don't think the whole thing suddenly comes crashing down at 100%. I think that the uniform sampling bias, for example, a shyness tendency among Bush voters, becomes more negligible as the number of Kerry voters in a precinct approaches zero. A sampling bias that produces a 6% oversampling in a 50-50 precinct might produce a 2% oversampling in an 80-20 precinct and obviously a 0% oversampling in a 100-0 precinct. As I stated before, the reason for this would be the decreasing probabability that a shy Bush voter would be replaced with a Kerry voter. In a 50-50 precinct a shy Bush voter has a slightly better tha 50% chance of being replaced with a Kerry voter (since there is a slight chance the replacment pollee could also be shy); in an 80-20 precinct the shy Bush voter has an almost 80% probabability of being replaced by a non-shy Bush voter.

        If I am right about this, and I may not be, you may still be right as well. The larger effect in red precincts does seem to imply either a relatively linear rate of oversampling of Kerry voters or a relatively linear rate of undercounting of their votes. Since the Bush voter shyness effect should be reduced in red precincts, the fact that it wasn't would seem to indicate that Kerry votes were undercounted.

        Pipe dreams are not an exit strategy.

        by TrainWreck on Wed Apr 06, 2005 at 11:43:41 AM PDT

        [ Parent ]

  •  Thanks. (none)
    Absolutely fantastic diary, and I'm a complete anti-fraudster. I'm always very very happy when actual professionals and experts (I understand this isn't your speciality, but you're obviously no dilletante) weigh in on Dkos. Professionals with rigorous academic and experiential knowledge are the spine of this site. Thanks for taking the time to share.

    I've always thought (as a red stater in a red precinct) that Kerry voters living in red territory were far more likely to respond to exit pollsters than were Bush supporters in that same area. Since they knew, based on our winner take all system, that our vote wasn't going to count, they were more likely to have something count, even a meaningless exit poll response. That's just a guess.

    I know it'd only be a correllation, but maybe check African-American or other interest group turnout against the graph of exit poll bias toward the Democrats. Blacks or other interest groups might be more likely to respond, and their increased turnout/participation in the various elections might explain the fluctuating poll bias. That or something as simple as training or geographical placement of pollsters or methodology.

    •  the one problem is (none)
      that for someone that values the take of professionals on a subject... it doesn't jibe with the conclusions made by people who this IS their specialty.

      It is an intrigueing argument, I will give it that.  And one that should be peer reviewed.

  •  I'm not an expert (4.00)
    nor do I have to become one to have an opinion about exit polls.

    Like in any field, I rely on experts. So in this case I rely on these guys:

    * Authors and Endorsers of the US Count Votes Analysis:

    Josh Mittendorf, Ph.D. Temple University Statistics Department
    Steven F. Freeman, Ph.D. Visiting Scholar & Affiliated Faculty, Center for Operational Dynamics, University of Pennsylvania
    Brian Joiner, Ph.D. Professor of Statistics and Director of Statistical Consulting (ret), University of Wisconsin
    Frank Stenger, Ph.D. Professor of Numerical Analysis, School of Computing, University of Utah
    Richard G. Sheehan, Ph.D. Professor, Department of Finance, University of Notre Dame
    Paul F. Velleman, Ph.D. Associate Professor, Department of Statistical Sciences, Cornell University
    Victoria Lovegren, Ph.D. Lecturer, Department of Mathematics, Case Western Reserve University
    Campbell B. Read, Ph.D. Professor Emeritus, Dept. Statistical Science, Southern Methodist University
    Jonathon Simon, J.D. Alliance for Democracy
    Ron Baiman, Ph.D. Institute of Government and Public Affairs, University of Illinois at Chica

    over this post which I find confusing. If this one is helpful to you fine, for me it's confusing and the opinion of a single individual who doesn't have a PHD in statistics.

    Darkness washed over the Dude...darker than a black steer's tookus on a moonlight prairie night...there was no bottom

    by moon in the house of moe on Wed Apr 06, 2005 at 10:46:53 AM PDT

    •  Have you looked into 'these guys' credentials? (none)
      I have.

      A couple of them are solid, albeit dated, and not necessarily on-point as to the critical subject areas of the purported analysis.

      And if I'm not mistaken, somebody we know was originally listed in this roster, and is no longer listed.

      •  You can argue it (none)
        but from a layman's perspective it looks official enough for me.

        I just spent a week listening to a parade of people opine on the state of Terri Schiavo's brain. She was everything from semi-conscious to talking to the nurses at the station.  In contrast, Dr. Cranford and Dr. Kaplan told us she was in a persistant vegetative state. Between them they had degrees in medicine and over 50 years experience with people in comas and vegitative states. Yet you wouldn't know that their opinion was worth any more than whack job Randal Terry, or a woman who emerged from a double brain stem stroke after 60 days and now had a cause to protect anyone in a coma, or a right to life doctor with credentials but a huge agenda. Facts didn't matter. Expertise in a field didn't count. Dr. Cranford lost it with some of these people on the Larry King Show and then later on Scarborough. And he was right. What he knows on the subject is worth more than what they know. He's devoted his life to learning it and dealing with it.

        And when he gave his conclusion he didn't attempt to walk us through all the details which he gained over many years of medical training and practice. He just told us and if we had any sense we believed him. I did not think I had to struggle through a couple of medical books to accept his conclusion.

        Which is why this paper doesn't help me. It's pitched to someone with a background in statistics. I don't understand it. One becomes lost in a thicket of detail that is beyond the non-specialist. The inference is the subject is complicated. We can't be sure. It's muddy. It's for the experts to hash out.

        This does not help me.  I just want to know what happened. For that as I said, I recommend the report linked in the earlier post.

        Darkness washed over the Dude...darker than a black steer's tookus on a moonlight prairie night...there was no bottom

        by moon in the house of moe on Wed Apr 06, 2005 at 11:28:51 AM PDT

        [ Parent ]

    •  I actually contributed (none)
      a little to the UScountsVotes report, but declined to be a signatory, mostly because my contribution had been fairly slight.  But also because some of the conclusions still bothered me.

      I needed time to think. This diary is the result of those thoughts.

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 12:15:01 PM PDT

      [ Parent ]

  •  unfortunately... (none)
    ...your analysis in no way explains why the exit poll methodology worked for the Senate race but not the presidential race with exactly the same voters. THAT is the smoking gun, IMO.

    See my diary entry today re this.

  •  one error (none)
    did you mean to put p<.05 instead of p<.5? if you did mean .5, then this finding is almost perfectly random and you can't include this hypothesis in your findings and in 2004 the democrat bias was far less than in 1992.

    your conclusion: your conclusions are not in line with the ones out of they conclude that it is statistically impossible for the exit polls to be off that much.

    also, your stats assume that the measures are comparable across different election years. i would argue that they are not. because of diebold shinnanigans, the last 3 (2000, 2002, and 2004) elections are chock full of fraud, and exit polls are just one piece of evidence.

    you caution that exit polls alone are not evidence of fraud, which is true. but the main argument is that if there is fraud, is there evidence? of which, exit polls is one. another is fixed recounts in ohio. another is the destruction of voting machines/evidence in florida in 2000. another is obstruction of justice in florida in 2004. another is voter complaints in 2004. another is the utter, complete, and total lack of errors in favor of democrats anywhere. the list goes on and on.

    •  Yikes! (none)
      will change!  Thanks

      And yes, these analyses are consistent with fraud going back a long way.

      All I'm saying is that if you are going to invoke fraud to account for the discrepancy it has to be widespread, greater in Democratic states in two years, but in Republican states in a third year.  It also has to have a fairly linear relationship with state partisanship - the correlations are fairly strong.

      At this point, sampling quirks start to look more attractive as an explanation, to my mind.

      I'm not concluding this, just saying that the fraud hypothesis has to account for the shape of the bias, over both time and space.

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 11:33:00 AM PDT

      [ Parent ]

    •  Changed .5 to .05 (4.00)
      Thanks for spotting it.  It was right further up the page.

      And I completely agree with you about the other evidence.  Someone here had a diary that showed that all the new votes (collected since the election) consistently reduced Bush's margin.  As these are presumably postal votes, and unhackable, this looks very fishy.  As does a lot of other stuff.

      My point is that the exit poll discrepancy is not good evidence.  It was only slightly greater than in 1992, and its distribution is odd, for the fraud hypothesis.

      But there is plenty of slack in the stats for small targetted fraud.

      In fact, the less we cry "Exit polls!" and the more we cry "shenanigans!" especially regarding voter suppression in Ohio, the more convincing (IMO) the  case will be.

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 11:40:22 AM PDT

      [ Parent ]

  •  As requested I recommended this diary, (4.00)
    Not because I necessarily agree or disagree with it, I really have no opinion on it one way or another, but the analysis is good and offers up great points and stirs debate on this important issue.

    In addition, I am in no way qualified to challenge your methods or really to understand them deeply, for I am not as educated as you are in mathematics.


    Over the years, I have noticed a tendency of folks to analyze things to death when there does not seem to be a need for doing so, and in that light, whether or not my perception is right or wrong, I'll make these comments.

    1. The usacountsvotes study does not really appear on its surface to challenge anything in your findings, it was largely done to determine 1. the plausibility of the reasoning Edison-Mitofsky's statement concerning WPE and under-sampling of Bush voters being the reason for the shift, and 2. To determine that if indeed it did occur how plausible it was that it would all occur in the favor of one man. While your study does show that bias exists and that WPE percentages on the face of it given certain occurrences could account for it, the data they analyzed seemed to prove just the opposite, thus leading them to conclude that Edison-Mitofsky's reasoning was flawed. In addition then they ran a probability analysis of it occurring in the benefit of one man it was statistically unlikely that such would occur.

      Additionally they are stating that it is absurd that "inaccurate election results" are not even considered, when anyone who even makes a real attempt to solve a problem of this nature would have to assign the probability of that being one of the reasons as equal to WPE. In addition because of the significance of it in a democracy, it would have to be weighted so heavily that is defies reasoning that it is not even considered.

    2. In my limited understanding of analysis of this sort, I cannot see how aggregated data at the state level tells you anything at all. Most of the states have 25-50 sampled precincts. Meaning that an analysis of this sort and the one done by usacountscotes is working with aggregated data from 1500 or so precincts sampled and 81K plus respondents. The issue is not really at the state level is it? The issue is where did this shift occur specifically and why. In this the usacountsvotes authors and endorsers specifically offers up their analysis as a hypotheses that should be investigated, nothing more and nothing less.

    3. I think people are not recognizing all the truly accurate data that is obtainable for verification of the exit polls themselves, these are,

      • Demographic data.
      • Voter registration data.
      • Party registration data
      • Consumer profile data.
      • Current phone polling data.
      • Historical election data.
      • Live trial results from the primaries and other trials

      There is probably more there, however, in looking at a sample questionnaire from Georgia for example, we see several questions on the survey that could easily allow us to compare to the above data in real time to see how our sampling is going. Georgia Questionaire

      Example Questions

      • Are you:

        1 ? Male  2 ? Female

      • Are you:

        1 ? White 4 ?? Asian
        2 ?? Black 5 ?? Other
        3 ?? Hispanic/Latino

      • To which age group do you belong?

        1 ?? 18-24 4 ?? 40-44 7 ?? 60-64
        2 ?? 25-29 5 ?? 45-49 8 ?? 65-74
        3 ?? 30-39 6 ?? 50-59 9 ?? 75 or over

      • No matter how you voted today, do you usually think of yourself as a:

        1 ?? Democrat 3 ?? Independent
        2 ?? Republican 4 ?? Something else

      • 2003 total family income:

        1 ?? Under $15,000 5 ?? $75,000-$99,999
        2 ?? $15,000-$29,999 6 ?? $100,000-$149,999
        3 ?? $30,000-$49,999 7 ?? $150,000-$199,999
        4 ?? $50,000-$74,999 8 ?? $200,000 or more

      Now, right there are 5 questions that I can track back to the other data to verify how my sample is going, whether or not I've got more women, what my percentages of Democratic, Republican, or otherwise party affiliation are, Age groups, Consumer profiles, and Race.

      If all of that is coming in to existing data, then I can state pretty clearly that my sampling is good and matches the data in my model.


      • In today's election for U.S. senator, did you just vote for:

        1 ?? Denise Majette (Dem)
        2 ?? Johnny Isakson (Rep)
        9 ?? Other: Who?_________
        0 ?? Did not vote for U.S. senator

      • How did you vote on Amendment 1, defining marriage only as a union between a man and a woman?

        1 ?? Yes
        2 ?? No
        0 ?? Did not vote on Amendment 1

      There are two more questions that I can verify my results to the actual election results,

      The first set of questions allow me to verify my model and the second allow me to compare some of the other exit poll selections to that of the election results.

      Using all of the above allows me to see pretty clearly what is going on in a precinct and to see specifically what is wrong.

      I also believe that all of the above would have been accounted for before the percentages were released that day, i.e. if a large number of Democratic and Independent respondents are filling in the questionnaires,and there is a large lack of Republicans respondents, its easy to tell in real time if there are problems with the sampling in that precinct, is it not? I doubt that if this were to show up as significantly off of the model then it would be immediately noted.

      Of course, that's not to say its so, just that its plausible.

      So in the first cut, we ought to be able to see how the data corresponds to the precinct being sampled and then how that precinct corresponds to the precincts it represents.

      So we're able to check the data against all of that, if demographics match up, if voter registration numbers match up, if party registration numbers match up and if the votes for other politicians in the ballot match up and if the votes for other items on the ballot match up, if all of that is within the margin of error and the presidential section is not, then we have a problem, do we not?

    4. Your analysis, not withstanding, I think the real issue here is that as long as we are told we must vote on black boxes with no paper trail, we allege fraud, loud and often until our politicians give in.

      Period, we cannot accept anything else and we really need to try to keep out of the smokescreen of statistics where anything can be prov-en and or disproved just by changing your view of the data.

    Now, I personally agree with your original statement

    I don't believe the exit polls in themselves are evidence for fraud.  I don't think they are inconsistent with fraud, but I don't think they support it either. Read on for my reasoning.

    I could not agree with you more, however I think in this case, the usacountsvotes is asking for nothing more than a better look at the data, and personally I do see that there is a basis for doing so, and really requires a mandate to do so in a democracy.

    We always suppose our votes have been stolen, and we suppose it no matter which political party is in power currently, and we always make our politicians prove to us that it has not been. I really don't care what the outcome is, in the end I want our political parties held accountable and I want our citizens to have trust in the accuracy of the exit polls as the last line of defense against fraud. In addition I want our politicians to fear them.

    I really think that we all, in an attempt to try and figure out how they work, why they could be accurate/inaccurate, underestimate the science in our attempts to reinvent the wheel and peel back the onion layers. This should be one of the most researched, well founded, scientifically sound approach possible with nearly 50 years of experience under the belt through 3000+ elections.

    Additionally, I think that this whole thing is nothing more than an attempt to discredit the exit polls. We cannot allow that to happen, once that is done, the door to fraud will be wide open with no means for verification of the voting results possible. Also, we cannot afford to let the debate descend into the realm of "sore loser", there is no greater cover for voting fraud than the "sore loser" allegation for wanting an investigation.

    I also want to state that if we are not allowed to look at the models and the data for independent verification, then we need to assume that "Edison-Mitofsky's" methods for conducting exit polls are so inaccurate and problematic that they should be called the "Hacks" that they are. One cannot have these types of problems and assume that these people are anything more than con artists, additionally we do not know if 10 million or more will buy their silence if indeed the data shows election fraud.

    So we either completely discredit them so that the networks never hire them again or we make them explain exactly where things went wrong and why, I can be totally black/white on that issue as there is no in-between.

    Thank you for your analysis and keeping the debate alive!

    •  I completely (none)
      agree with many of the points made by the UScountsVotes team.  I was actually part of the team, and contributed a little, but decided not to sign the final version, mostly because I had not contributed that much, but also because I had some reservations with some of the conclusions (mostly with the conclusiveness of the conclusions).

      I do think this all ought to have been properly investigated, and EM have been extremely stingy with the data needed to investigate properly.

      However I also think, in retrospect, that the insistence that the Exit Poll Discrepancy is the key to the fraud hypothesis has been a great distraction.  It was always tenuous, whereas the voter suppression story in Ohio was really real.  You might like to read one of my analyses of voter suppression evidence in Franklin County, Ohio, here, on the UScountsVotes site.

      They are also doing a fantastic job of logging data, especially precinct level data, which should  discourage fraud attempts in future, and enable us to do a rapid audit earlier.

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 12:05:11 PM PDT

      [ Parent ]

      •  I don't disagree with that... (none)
        I have been discussing this a lot with some conservative friends who are pretty smart and certainly make the same points you have made.

        I always walk away with the belief that in a "black box voting world", I have to assume that my vote was stolen, to ever assume that it has not been when there are no paper trails is to invite fraud.

        If our politicians thought we always assume they are lying to us and will raise holy hell at even the slightest hint of fraud or "tom foolery" then they are much less likely to engage in it IMHO.

        Additionally, in my discussions, I purposefully assumed the role of defender of the exit polls, instead of making the assumption that they are flawed, I made the assumption that they are accurate and need to be prov-en to be flawed, therefore disproving my assumptions, which is a basic principle in math and especially statistics I believe.

        I also, thank the folks who did the study and yourself, you are all citizens of this country who's most likely priority is the democracy itself and the generations that come after us, and you spent a considerable amount of time on this and everyone who did sign or endorse the study did put a lot on the line of behalf of "we the people" which I think is deserving of significant praise and gratitude.

        In the end, I think we err on the side of fraud on this one and keep it alive until they prove "us" wrong and proper auditing and controls are put in place. I'd bet that those thoughts were not far from the minds of those who performed this study and yourself included.

    •  Relevant Comment (none)
      on Blaming Mitofsky here
  •  still looks fishy to me (none)
    I think maybe you don't see the wood for the trees:

    1) There is clear evidence that the discrepancy
       between exit polls and official results was
       very large this year (larger than in previous
       elections).  Nobody has given a convincing
       explanation of that, so the "fraud" explanation
       seems at least as plausible as the "differential
       response rate" explanation.

    2) The USCountVotes report, and Edison-Mitofsky,
       indicated that the WPE was very low for precincts
       with paper ballots (and presumably therefore with
       manual tabulation); and rather high for all the
       automated-count systems (punch-card,
       optical-scan, touch-screen).  That certainly
       seems consistent with fraud in the automatic
       tabulation.  There could be an alternative
       explanation - maybe the paper-ballot precincts
       have very different demographics.  But it should
     be investigated.

    At the least, this all seems fishy enough that we
    should demand that independent statisticians get
    access to the raw precinct-level NEP data.  If
    there is a smoking gun, that would probably be
    sufficient to show it.

    •  I am all for investigating (none)
      I just don't think it is likely that fraud was the major cause of the nationwide Exit Poll Discrepancy, for the reasons given.

      We do need precinct level data, but not raw data, we need the weights they used for their "call three" predictions.  We also need to be able to identify the precincts at least by county.  But we can't.

      I have always thought that it would be precinct level data that would open the lock.

      Bastards. (EM that is).

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 12:10:45 PM PDT

      [ Parent ]

    •  Manually tabulated paper ballots (none)
      ... were a statistical rarity, confined almost entirely to rural precincts (where WPE -- regardless of equipment -- was minimal, pehaps due to the "everybody knows everybody" ethos, or perhaps due to red state locus, or both).
  •  Yeah (none)
    Let's hope they have auditable results in future elections, so we don't have to go through this again.

    Bonus points for using a word I've never seen before: "disambiguate".

    "I feel your scorn and I accept it." - Jon Stewart

    by starkness on Wed Apr 06, 2005 at 12:04:00 PM PDT

  •  Snark-Free Exit Poll Analysis (none)
    Keep in mind that US Count Votes did not charge fraud.  What they said was that the variance between the exit polls and the actual count was too great to be a function of random chance.  Even E/M conceded that.  You are presenting a different hypothesis than US Count Votes arrived at, but youy also agree that random chance can be ruled out as an explanation for the discrepance.  You have only underscored the thesis of the US Count Votes analysis.  Something is very wrong, and an investigation needs to be conducted and the raw data released for study.  All US Count Votes did was to take issue with the January E/M analysis and say that fraud cannot be ruled out as an explanation (although they never actually used the "f" word.  They actually were very careful never to offer a preferred alternate explanation.  They did point outr, accurately I think, that E/M never even considered the possibility that the count was wrong.  Since they, like US Votes Counts, agreed that random error had to be ruled out, that left only WPE as an acceptable hypothesis.  Let's stop quibbling about what went wrong and demand that someone investigate to determine exactly what went wrong.  Failing that, let's see to it that legislation is passed that assures that the voting process is honest and secure from tampering.
    •  See (none)
      my response to this comment on your diary.

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 12:53:55 PM PDT

      [ Parent ]

      •  I have a question... (none)

        I just can't think of one, whereas I can think of one that is related to sampling bias (over-eager Democrats in Democratic states).

        I stated above that you have a model of the precinct, and one of the data in this model is voter party affiliation, either from the publicaly available registration data, or perhaps from past elections.

        So if you have data that indicates a 53/47 split of party affiliation, and your sampling is coming in at 47/53 would you not know immediately that you are over-sampling one party.

        Given the fact that party affiliation is on the questionaire, it would seem to provide an immediate check on over-under sampling.

        And unless they are using mainframes from the 60's to analyze the data in real time then, they would have this data live and in real time and would likely spit out these precincts as outliers and unreliable indicators would they not?

        The same goes for the other information obtained on the questionaires.

        Unless I'm mistaken I just don't see how advance knowledge of over/under sampling is not known and further built into or excluded from the predicted results if it occurs.

        Do you have any insight?  

        •  We are not talking about party affiliation (none)
          we are talking about votes.

          Sorry if that wasn't clear.  We don't have party affiliation in the UK, so I forget it is thing to consider in the US.

          I am simply talking about the estimated margin (from the poll) and the actual margin (from the count).  I don't know how the weighting is done, and how party affiliation comes in.  I should, but I don't.  I just play with the numbers.

          And yes, the party affiliation would give us a check - but we don't have that data.

          And they don't know whether they've under or over sampled until afterwards and people complain they got it wrong.  They didn't mean to over-sample Democrats, but they seem to find it hard not to!

          See my new Exit Poll diary here

          by Febble on Wed Apr 06, 2005 at 03:00:15 PM PDT

          [ Parent ]

          •  Well, Hmmmm (none)

            And they don't know whether they've under or over sampled until afterwards and people complain they got it wrong.  They didn't mean to over-sample Democrats, but they seem to find it hard not to!

            Ok, I can pull a list of registered voters in Gwinnett County Georgia, I can get a list of registered voters who are registered as a republician or a democrat, these are maintained in order to insure the primaries are done correctly and Republicians vote for their candidates and Dems theirs.

            So if I have registered voters, and that number is equal to 25,000 and I have 7,000 registered dems and 13,000 registered repubs, then I know my profile for the voter affiliation in Gwinnett is 28% dems, 52% republicans and 20% other.

            On the questionaire, I have a question that asks me

            1. No matter how you voted today, do you
              usually think of yourself as a:
              1 ?? Democrat 3 ?? Independent
              2 ?? Republican 4 ?? Something else

            All of questionaires are tallied up 3 times uring the day ans sent into the main computer,

            Once there, I compare the data from the sample precinct to the model above, so if I have a situation of 40%, 40% 20%, dems, repubs and other

            I have an immedidate comparison of

            Source dems repubs other
            Registration 28 52 20
            Questionaire 40 40 20
            over/under 12 12 0

            And this applies to many of the other questions on the questionaire also.

            And let me say this, if they don't have this data in a model and are not comparing it on an ongoing basis then they are nothing but con artists and I most certainly agree that the exit polls are crap.

            Am I missing something?


            •  I think you're way ahead of me.... (none)
              All I'm talking about is whether the selection process was random or not.

              They may well weight by party affiliation.  And that should alleviate the problem.  But they don't seem to know how to fix it.

              It happens every year, as I show.

              Unless fraud happens every year.  Maybe it does.

              But I know that random sampling is hard to do, and if you got it wrong one year, it's correcting your model will be difficult the next year.

              See my new Exit Poll diary here

              by Febble on Wed Apr 06, 2005 at 04:12:45 PM PDT

              [ Parent ]

              •  That's why I mentioned existing data in my.... (none)
                Other post. People are focusing on the sample data and seem to be forgetting the data fed into the model used to do the sampling.

                This is probably highly propietary, but I know a little bit about what type of information is available on the people of this country and trust me it would not be very hard to build a very accurate profile of any 25 square mile area in this country.

                The model sophistication of this is far more advanced than people realize, none of it is random to say the least.

                It would be very easy for me to get information on every Republician in my town who makes over 100k, has a daughter over 16, what banks they bank with, their credit rating and even their buying habits. I would know what magazines they subscribe to, how much they donate to political parties, if they are registered voters, what party they are registered to, information about wife and all of the same data.

                I could build detailed profiles of every person in any area at will.

                I also have historical data from past elections, historical exit poll data, primary data, demographic data, etc.

                All of this data would I assume be used to build the profile to select the precinct's that will be sampled and are selected to be the most representitive of the precincts thney represent in the area.

                I am assuming the last part. I know that all of the data is available to build the model, and I assume that if I was a polling company I would use it, and I believe that I could probably come within 2% of predicting the outcome in any given area using a model from the data.

                The sampling should serve to verify the model and vice versa, and any anomolies should show up right away.

                It seems to me that there is just no other way to conduct the poll.

                You must have a model and it must be damn accurate and representthe people in the area to a tee...

        •  Answer: you would not know (none)
          We know turnout varies by party. We know turnout by party varies depending on other races and initiatives on the ballot. We know it varies between poll voters and absentees. We know many voters don't think much about "party", and don't necessarily remember how they are registered (in states that have party registration). And we know stated party identification will vary from poll to poll for the same voter.

          None of this lends itself to meaningful detection of a few % oversampling.

  •  Truth Is All (TIA) over at democraticunderground.. (none)
    ...has been working for months on the analysis of the exit polls and there is much more there (in the exits) than you have presented above.

    ...for example...are you using the "corrected" last exit polls that now conveniently reflect the reported vote tabulations, the exit polls up until the final "mystrious Bush surge" that showed Kerry winning all night long, or the raw data that is still being sat upon by the pollsters?

    ...the numbers broken down by demographic don't add up to a BushCo win at all.

    Check it all out here
    They are doing some impressive work.

    As TIA explains:

    It comes down to these simple facts:

    1. The Final Exit Poll (13660) promulgates an impossibility
    in using the 43% Bush/37% Gore "Voted in 2000"
    demographic weighting.

    2. Even if ALL 50.456 million Bush 2000 voters were still
    alive and returned to vote, his maximum 2000 voter share is
    just 41%, as stated in the "pristine", unadjusted
    Preliminary Exit Poll of 13047.

    3. Approximately 3.5% voters have died since 2000, so the
    Bush 2000 voter share is now reduced to a MAXIMUM OF 39.8%,
    assuming 100% turnout. Of course, it MUST BE EVEN LESS THAN
    39.8%, because an unknown number of Bush 2000 voters stayed
    home this time. But we will assume 100% turned out - to be

    4. We KNOW that at least 17% were NEW voters, because we KNOW
    the change in total voter turnout from 2000 (104.777) to 2004

    5. We KNOW that 3.17% voted for Nader/Other.

    Therefore, it follows that if the Bush 2000 share is 40%, we
    can calculate the Gore 2000 share of 40%:

        Gore = 100% - Bush - NEW - Nader/other
         40% = 100% - 40% - 17% - 3%


    Using the Final Election Poll (13660) stats and the 40/40
    split, Kerry is an easy winner by 3.5 million votes. The
    Final Exit Poll weights ADJUSTED the Preliminary Exit Poll
    weights in order to MATCH the Recorded Vote.

    MIX CHANGED TO 39.8/40.3%.

    All other Final Exit Poll numbers are unchanged:

    2000    Mix    Bush    Kerry    Nader
    No    17.3%    45%    54%    1%
    Gore    40.3%    10%    90%    0%
    Bush    39.8%    90%    9%    1%
    Other    2.6%    21%    71%    8%

        100%    48.18%    51.04%    0.78%
        122.26    58.91    62.40    0.95

    Using the Preliminary ("pristine") Election Poll
    (13047) stats and the 40/40 split, Kerry is a landslide
    winner by 7 million votes.

    2000    Mix    Bush    Kerry    Nader
    No    17.3%    41%    57%    2%
    Gore    40.3%    8%    92%    0%
    Bush    39.8%    90%    9%    1%
    Other    2.6%    21%    71%    8%
        100%    46.68%    52.37%    0.95%
        122.26    57.07    64.02    1.16

    The detailed mathematical analysis is explained here:'

    "Those who cast the votes decide nothing. Those who count the votes decide everything" - Joseph Stalin

    by Blue Shark on Wed Apr 06, 2005 at 01:07:12 PM PDT

    •  Just posted by adolpho at DU... (none)
      ...This from Florida where machines seem to have reported the exact vote count margins in the tabulations as their neighboring counties.

      ...There are so many instances of fraud it is almost impossible to log them all.

      Has anyone here looked into the odd instances of precincts that share the same ratio of votes with their neighbor? Are there other states that display the odd results?

      I stumbled upon these numbers while examining precinct reports a while ago. Here is the list compiled so far:

      ---------------------------------------------------------------------------------------------------- --

      =listed by: 1.)precinct - 2.)Kerry vote total - 3.)ratio of registered Democrats in precinct - 4.)Bush vote total - 5.)ratio of registered Republicans in precinct))=


      PCT 28 - WESTSIDE MENS CLUB 167 (19.6%) 584 (44.4%)
      PCT 29 - GRACE PRESBY. 198 (19.6%) 672 (52.3%)


      4C 92 (13.3%) 411 (131.3%)
      4D 68 (13.3%) 360 (159.3%)


      65 Holy Name of Jesus Catholic Church 517 (74.8%) 644 (63.9%)
      66 First Christian Church of Merritt Isla 577 (62.0%) 1154 (63.9%)


      143 FIRST CHURCH OF NORTH MIAMI 1200 NE 135 ST 627 (69.8%) 124 (74.7%)
      144 FIRST CHURCH OF NORTH MIAMI 1200 NE 135 ST 236 (69.8%) 83 (98.8%)

      777 MIAMI ELK'S LODGE #948 10301 SW 72 ST 126 (75.0%) 190 (65.5%)
      778 MIAMI ELK'S LODGE #948 10301 SW 72 ST 6 (75.0%) 21 (60.0%)

      429 IGLESIA ADVENTISTA EBENEZER 6566 SW 33 ST 219 (61.9%) 392 (58.0%)
      430 IGLESIA ADVENTISTA EBENEZER 6566 SW 33 ST 267 (71.8%) 598 (58.0%)


      36 - Hillside CC 258 (50.3%) 264 (67.5%)
      37 - Ridge Manor W Club House 281 (50.3%) 408 (69.7%)


      519 438 (61.6%) 624 (80.6%)
      520 688 (61.6%) 939 (77.5%)

      119 506 (48.6%) 861 (77.2%)
      121 654 (66.1%) 754 (77.2%)

      433 175 (81.8%) 288 (74.4%)
      434 466 (65.6%) 660 (74.4%)

      658 141 (95.3%) 113 (76.4%)
      659 451 (62.6%) 541 (76.4%)


      26 - ARIPEKA 201 (71.3%) 174 (73.1%)
      27 - SHADY HILLS 559 (50.6%) 732 (73.1%)


      412 1214 (60.6%) 793 (82.6%)
      413 479 (51.5%) 618 (82.6%)


      302 - Robert Ensslin Armory 296 (49.8%) 614 (59.2%)
      303 - Memorial Lutheran Church 226 (44.1%) 427 (59.2%)


      "Those who cast the votes decide nothing. Those who count the votes decide everything" - Joseph Stalin

      by Blue Shark on Wed Apr 06, 2005 at 01:14:24 PM PDT

      [ Parent ]

      •  Good thinking (none)
        I am still relatively convinced there may have been fraud in Florida.  Some things stink.

        See my new Exit Poll diary here

        by Febble on Wed Apr 06, 2005 at 01:17:18 PM PDT

        [ Parent ]

        •  I am absolutely (none)
          convinced there may have been fraud in Florida.... but I am a philosopher so....

          The problem of power is how to achieve its responsible use rather than its irresponsible and indulgent use JFK

          by Responsible on Wed Apr 06, 2005 at 02:09:02 PM PDT

          [ Parent ]

          •  Yes... (none)
   code piggybacked on the tabulators to record pre-determined (read fraudulent)results.  But they should have spread it out a little further by a couple of counties so it wasn't quite so obvious.  Then again... they got away with it and subtlety was never a strong suit at BushCo.

            "Those who cast the votes decide nothing. Those who count the votes decide everything" - Joseph Stalin

            by Blue Shark on Wed Apr 06, 2005 at 02:26:36 PM PDT

            [ Parent ]

      •  Could you (none)
        Explain the fraud mechanism that would result in these numbers?

        The problem of power is how to achieve its responsible use rather than its irresponsible and indulgent use JFK

        by Responsible on Wed Apr 06, 2005 at 02:13:51 PM PDT

        [ Parent ]

        •  And BTW... (none)
          ...the above results happening in neighboring counties by chance?????

          ...I need to buy a clue here Pat...

          ...OK...I guess a gazillion to one?

          ...Ding Ding Ding...we have a winner!

          "Those who cast the votes decide nothing. Those who count the votes decide everything" - Joseph Stalin

          by Blue Shark on Wed Apr 06, 2005 at 02:30:11 PM PDT

          [ Parent ]

          •  I understand (none)
            the likelihood is low.  I was wondering what particular method of stealing the vote would result in this particular kind of unlikelihood?  For instance, if I were in control of the Diebold code, I would just switch (something like) every tenth Kerry vote to Bush.  This would not result in these unlikely numbers.

            This is not a criticism but a requet for information.

            The problem of power is how to achieve its responsible use rather than its irresponsible and indulgent use JFK

            by Responsible on Wed Apr 06, 2005 at 03:20:43 PM PDT

            [ Parent ]

            •  I am no expert... (none)
              ...regardless of my best efforts to convey that message here on dKos...there are experts out there though and this  is one of the best.  Familiarize yourself with Mr. Herrin's work and the important work of others (Conyers et al.) and the cumulative weight of evidence of fraud in Nov. 2004 is undeniable and overwhelming. (2000 and 2002 also)

              ...there was motive, opporunity, a pattern of cheating, and the unlikely outcome that was reported by the in-curious media after the Black Boxes said Bush Won!...Hell there was even a body...the rank corpse of American Democracy.

              ...had our guy not QUIT for Godsake, all of this would have been investigated until the light of day and the power of the truth melted the witches and demons.

              "Those who cast the votes decide nothing. Those who count the votes decide everything" - Joseph Stalin

              by Blue Shark on Wed Apr 06, 2005 at 05:24:02 PM PDT

              [ Parent ]

      •  Could you (none)
        Explain what you find remarkable in these data?

        If % of D-registered voters for Kerry,or % of R-registered voters for Bush, expressed to one decimal place were uniformly random (equiprobably from 0.0% to 100.0%), you'd find a coincidental match between one number or the other for about 1 in 500 adjoining precincts.

        Considering that many numerically-adjoining precincts are similar samples (often adjoining blocks in the same voting site), and that a lot of D's vote for kerry and R's for Bush, you'd expect a much tighter pattern ... and coincidental matches for perhaps 1 in 50 adjoining precincts.

        Wouldn't you?

    •  I am using the WPEs (none)
      published by EM in their report.

      I am not using the screenshot data.

      I got the results for previous elections from President Elect

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 03:20:07 PM PDT

      [ Parent ]

  •  Febble, (none)
    Thanks for doing all this work.  I desperately want to believe this election was stolen, and your analysis doesn't necessarily convince me it wasn't (nor was it meant to) -- but I truly appreciate your acting as a voice of reason and concern for truth over ideology.  What you have presented here makes sense-- at least to me, a non-statistics geek.
  •  nitpick (none)
    Is this correct:?

    In the second example, you wrote "20/100 (18%) for Bush" instead of 20/110 (18%) for Bush.

    In the second example, you wrote "82% - 18% = -64%" instead of 18% - 82% = -64%

    In the third example, you wrote "40/102 is 22%" instead of 22/102 is 22%.

    The recommend is for your clear analysis. The 4 is for "inexcusably riggable" and "disambiguate."

    •  Thanks, I'll fix (none)
      But it's nearly midnight here, so I'll have to do it tomorrow or I'll make more errors.

      The answers are right though.

      I hate typing stuff in that little box!

      Thanks for the 4!

      See my new Exit Poll diary here

      by Febble on Wed Apr 06, 2005 at 03:30:39 PM PDT

      [ Parent ]

  •  I'll try this again... (none)
    Febble, please know that I am not attacking you, just raising some questions that my dense-self can't understand on my own.  

    If I understand your bottom line correctly, you are saying that WPE is not the correct (or best) dependent variable.  If so, the E/M Report AND US Count Votes used an inferior and perhaps innapropriate dependent variable.

    The theory behind WPE is that when averaged over the all the precincts with a characteristic (state, geo, vote machine type, weather, interviewer characteristics, etc), the random sampling error was SUPPOSED to leave only the statistical bias (random error averages out).  As you have demonstrated, this is NOT the case in the most extremely partisan precincts.

    So, you created your own "Bias Index" that is supposed to control for the precinct partisanship effect compromising use of WPE as a dependent variable.  Per the name of your new variable, your index should leave ONLY the true statistical "uncontaminated" bias and is therefore a better dependent variable.  

    However, you seem to be assuming that the precinct partisanship effect ONLY affects the bias.  I am suggesting that your variable does not control for all the "contamination" from precinct partisanship, because there is a portion of that contamination resulting from the practical limits of the sampling error at the extremes of the political stratum.  That is, even your variable is still not a measure of the pure exit poll error - it is still impacted by the practical limits of random error in the most extremely partisan precincts - the very effect you seem to claim to have controlled for.

    Now, I'm thinking about this practically and not with a mathematical background.  I am a social science guy with a graduate level understanding of statistics, but no solid math background (that is, I know how to apply the stats, but don't know all the theory behind it).  So, I understand your formula, but I'm not sure it does exactly what you think it does and would like more clarification on why/how you arrived at the formula.  

    •  Just emailed you the derivation (none)
      of the formula. Let me know if it makes more sense.

      I'm still thinking about your question, not sure if I've answered it.

      I really appreciate your interest.

      See my new Exit Poll diary here

      by Febble on Thu Apr 07, 2005 at 09:39:36 AM PDT

      [ Parent ]

  •  One big assumption you are making... (none)
    is that the interviewers systematically (across the nation and historically) did not do their basic job as interviewers.  That is, select respondents at random.

    The procedure with the EM exit poll survey was NOT for the respondents to voluntarily come to the exit poll booth and participate.  The interviewers were to select every nth person leaving the election polling area to take the survey.

    A systematic failure of this procedure across the country -- with the error always favouring Democrats -- still seems highly unlikely to me.

    Otherwise, interesting work.  However, as you say, without the raw data, we still cannot be certain (as yet) whether there was counting error or oversampling error.

    When is the MSM going to respect democracy and release these numbers???!!!

    •  It is horribly easy (none)
      to depart from random sampling, and there is evidence that the interviewers were seriously under trained. Once you depart from randomness all sorts of biases can creep in, as people who work in the social sciences know only too well.

      And the other possible way Kerry voters could have been "over-sampled" is that some of the "Kerry voters" were in fact lying "Bush" voters.  This migh explain the senate race discrepancy, if it was Bush they were embarassed about (and so they should be).  It would also account for greater bias in Democratic states (more embarassing to admit to voting for Bush were the state is overwhelmingly pro-Kerry).

      People do lie.  They lied in Britain in 1992 about voting Tory, making us all think Labour had won, or at least that the Tories had failed to get an absolute majority.  Being Tory in the John Major days was such an embarassment.  But the Tories won by a whisker.

      See my new Exit Poll diary here

      by Febble on Thu Apr 07, 2005 at 11:36:39 PM PDT

      [ Parent ]

      •  As far as I understand it... (none)
        This survey was to be written anonymously and privately, was it not?  That is, noone was looking over the shoulders of these respondents when they were filling out the survey.

        Did anyone look at the accuracy of other "personal & private" responses -- such as personal income or religious affiliation, etc.?  Are the stats for those particular responses out of wack?  This would help show the "truthfulness" factor, would it not?

        Sorry, I just don't buy the notion that only Republicans were "shy" about filling out an anonymous and privately conducted survey.  In my experience, conservatives that I know are much more up front about getting what they have to say off their chests -- in a more blunt, straight-forward way. Quantitatively, I'm not convinced with the numbers presented so far to overcome my qualitative bias that the "shy-Republican" theory makes no intuitive sense.  

        •  Good point. (none)
          Although if my correlations are correct, the avoiding (or lying)occurred far more in Democratic states, where admitting to voting for Bush might be more of an embarassment.

          The more I look at the pattern of correlation over the last five years the more it makes sense to me as a behavioural issue.  In 1988 and 2004, the correlation was positive - more bias in Dem states.  Let's just say that this represents more Bush avoiders (lying, avoiding pollsters) in Dem states.  Little bias in 2000, when everybody said there wasn't much to choose between Bush and Gore (how wrong they were....) and turnout was low - and what bias there is isn't significantly correlated with state colour.  But in 1992 and 1996, bias is correlated negatively with state correlation - more REP lying/avoiding in red states than in blue (though plenty in both).  But in these years, the republicans were behind, and Perot was on the ballot.  I suspect that Rep voters didn't admit to voting for Clinton, but did pretend to have voted for Perot.  No-one likes to back a loser.

          I know this isn't your point - you say, why would anyone lie on a form?  Well, they do!  I think they lie on forms more easily than they lie in speech.  I know I put tiddlysquat on some forms when I'm feeling antsy.  But my point is, that lying (in the way I've hypothesised) would fit the pattern of the data rather well.

          And we do have to account for the very great bias in 1992.


          See my new Exit Poll diary here

          by Febble on Fri Apr 08, 2005 at 08:46:55 AM PDT

          [ Parent ]

          •  Where's your proof (or precedent)? (none)
            "I know this isn't your point - you say, why would anyone lie on a form?  Well, they do!  I think they lie on forms more easily than they lie in speech."

            Do you have any proof (other than it loosley conforms to non-specific data)?  Can you give us any other conclusive precedent for this happening, given the circumstances?

            With all due respect, I think the "onus of proof" is  on your side of the court for this particular hypothesis.

            Keep in mind:

            1.  What may look like historic similarities might not be similarities at all.  That is, the reason there was a overestimate one year may be very different from the reason there was an overestimate in another year.  (assuming that there was, in fact, an exit poll overestimate at all).

            2.  There were widespread voting irregularities accross the nation.  Moreso than in any other year.  The 2004 voting system was also very much more prone to fraud and manipulation than in any other year.  Couple that with documented affadavits from programmers who have testified that a certain high-profile Republican asked to have machines rigged in Florida earlier this decade.  Etc., etc., etc....
            •  No, I don't have proof (none)
              only anecdote.  I have lied on forms myself.  I know people who have lied to pollsters.

              And I am not on one side of the court or the other.  I would have loved the answer to be fraud.

              And I agree the reasons for overestimates may be different in different years.  I suspect they were. The distributions are certainly different.

              But I am trained to look for patterns, and the patterns to me don't look like fraud.  They look like people embarassed about their vote.  Republicans with a weak candidate in Dem states.  Republicans in Republican states with a weak candidate who would have liked to have voted for Perot.

              If someone can come up with a coherent account of fraud that sits with the pattern I would be the first to welcome it.  I would love to think America really voted for Kerry.  But at this stage I think the pattern looks more like biased sampling, possibly for a number of different reasons, the preponderance of each of which varies from year to year.  

              But always in the Democratic direction.

              See my new Exit Poll diary here

              by Febble on Fri Apr 08, 2005 at 02:02:03 PM PDT

              [ Parent ]

              •  We also see patterns (none)
                in grilled cheese sandwiches.

                Please don't get me wrong.  I'm not dismissing what you say.  I'm just not convinced that you've provided enough proof.  Numbers can be interpreted a number of ways -- especially when there is important and significant missing information.

                Which brings me to my point.

                Time and time again, the American public is prevented from being told the whole truth by people in higher power.  The MSM will not release the raw data.  SOS Kenneth Blackwell intentionally prevented the recount process from being conducted properly.  Diebold would not provide a paper trail.  Black Box Voting found original ballot counting tapes in the garbage.  Congressman Freeman continually denies that he asked to have vote-rigging software developed -- contradictory to the evidence and testimony.  Etc., etc., etc.

                •  Yes, seeing pictures in (none)
                  grilled cheese sandwiches is precisely what statisticians have to guard against, and is what many statistical procedures (post hoc tests for instance) are there for.

                  What you must not do, is see pattern, then test for its significance. Or if you do, you have to factor in the billions of grilled cheese sandwiches that didn't have the picture.

                  What we try to do instead is to have an a priori hypothesis - a pattern that will be there if a hypothesis is true.  I had two - I hypothesised that a fraud pattern would have a quadratic relationhips with state partisanship while a behavioural effect, such as over-eager Dems or reluctant Republicans would have a linear relatinship or none.  I found no quadratic relationship but linear relationships in four out of the five years, the only exception being 2000 which had relatively little bias.

                  Moreover, it was the years that had Perot on the ballot where the linear slope went the opposite direction (but now we are getting into grilled cheese sandwich territory - I did not predict this - however I did do a post hoc test, and the difference in slopes between years is significant after correcting for the grilled cheese sandwich factor.

                  But these kind of stats can't indicate causality.   But they can tell us what kind of hypothesis is required to explain the data, and I struggle to explain this pattern by fraud, whereas a reluctant/over-eager voter explains it rather well.

                  As for your last paragraph, I agree utterly.

                  And even though I'm a Brit I emailed everyone I could, including the Kerry-Edwards legal team in Ohio, with my evidence of voter suppression in Ohio, and with support for a proper recount.

                  Ohio was a disgrace to democracy!

                  See my new Exit Poll diary here

                  by Febble on Sat Apr 09, 2005 at 12:57:06 AM PDT

                  [ Parent ]

                •  Congressman Freeman? Hmmm... (none)
    •  Also, look at the data (none)
      for WPE by sample frequency.  As the sampling frequency decreased (every "nth" got larger), the WPE increased.  That suggests to me that the more opportunities that an interviewer had to "self select" (non-random selection), the larger the error.

      But, of course, this cannot explain ALL of the WPE as explained in my comment here.

    •  It's not an assumption (none)
      "Interviewer" debriefs almost universally indicate thay did not follow protocol.

      And the details should never be released. Surveys were taken under promises on confidentiality, and identification at precinct level would enable public identification of most/all individual voters and how they voted. This is not acceptable under any circumstances.

      A generally satisfactory alternative is for skeptical statisticians to submit specific tests to be run against the data, and for E/M to run them. Inconvenient on both sides, and a bit slow, but should wring out anything that can be wrung out of the data.

      •  Do you have (none)
        documentation to back this up?

        Do these surveys have the names (or any other identification information) of the respondents?   I'm pretty sure they don't.

        •  Try looking at the E-M Report itself... (none)
          I'm not sure what "survey" you are asking about, but I assume you are questioning RonK's statement: ""Interviewer" debriefs almost universally indicate thay did not follow protocol."

          If they didn't conduct post-election surveys of the interviewers, how do you suspect that E-M got data regarding:

          1. "interviewer distance from exit"  (pg. 37)
          2. "possible to approach every voter?" (pg. 37)
          3. "precinct official cooperative?" (pg. 38)
          4. "voters cooperative?" (pg. 38)
          5. "any other people interfere with interviewing?" (pg. 39)

          Or, this from page 44:

          "This information confirms what we found in our post-election survey of our exit poll interviewers."

          Or, the kicker from page 52:

          "...we conducted a follow-up telephone interview with our exit poll interviewers in the three weeks following election day.  In total we were able to contact 1,350 interviewers (92%)."

          Now, look at the data: When interviewers were asked to sample EVERY voter, the mean WPE was -3.9.   Mean WPE increased as the sampling interval increased.  

          page 35: "The increased WPE in these precincts could suggest that some interviewers do not follow the interviewing rate exactly.  As the interviewing rate increases so does the potential for interviewers to exercise more of their own judgment on whom they will approach in order to participate in the exit poll."  

          Also, consider that the data show that WPE was correlated with interviewer age and education.  The younger and more educated the pollster was, the more negative the WPE.  Highly educated young folks are a very reliable Democratic voting demographic.  

          This suggests that younger more educated interviewers were more likely to: 1) fudge the rules from time to time and not select randomly; and 2) when they did, subconsciously select voters that they didn't think would reject them.

          However, this doesn't explain all of the bias in WPE, only a portion.  Some portion must still be coming from reluctant Bush responders, or "something" else - yes - including fraud.

          However, the patterns in the data, as febble so eloquently demonstrated with this diary, do not suggest that fraud could account for a large portion of the bias.  

          Fraud in a precinct here and there? Possibly.  Widespread and massive fraud leading to the magnitude and geographic dispersion of the exit poll discrepancies?  Come on...

          •  The data you've just listed... (none)
            is hardly conclusive that the interviewers were not following protocol (to favour Kerry).  There just isn't enough data to make it so.

            What's more, The data that you have specifically used to support your argument about "interviewer corruption" contradicts the hypothesis that Feble is arguing for.

            If interviewers are more likely to select democratic voters (as the hypothesis goes), then we should see a decrease in the refusal rate as the interviewing rate increases-- since democrats are more likely to respond.  This is not the case.

            It seems to me that there is data that supports Feble's hypothesis and data that doesn't.

            In the end, we will only have convincing evidence when all of the numbers are released.  I wouldn't dismiss the massive fraud hypothesis so easily, given the factual evidence of voting irregularities and the significant increased opportunity for fraud in 2004.  Stats -- especially vague stats -- is not everything.  Remember, Freeman et al. are only arguing that significant vote-counting error is still a viable possibility -- something Mitofsky et al. did not even consider (which is curious in itself).

  •  Febble Analysis of EE/MES 2004 (none)
    The demographic profile of the interviewers was predominantly female (63%) and predominantly younger (18-24 35%, 25-34 15%), and the report acknowledges a need to try and recruit a broader based demographic for poll takers next time, and  with a more standardized scripting see page 53.

    Are there any statistical comparisons made of the 2004 interviewers with previous years exit poll interviewers?  EE/MES2004 doesn't say (at least that I see) other than stating that they note the demographic tilt to college and younger adults.  It seems unlikely (but not impossible) the poll takers could be such a contributing factor.  I would think that conditions would be similar for most polls-that college aged and young adults would be more predisposed to take part in the process.  If it can be shown that the interviewer profile is reasonably consistent, it would eliminate what appears to be a variable being used as qualified wiggle room.

    If I find any stats I'll pipe back in.

    Great job!

  •  I am in no way a statistician (none)
    However I have followed closely the practical issues   in the accounts of voting problems.
    Is this a correct rephrasing of your conclusions:

    • Dem vote is always over-estimated by exit polls(or under-counted by actual count)

    • Dem vote was significantly more over-estimated (or under-counted ) this year than in past

    • There was greater under-counting (or over-polling) of Dem votes in Dem states than in "Red" States

    • The 3 critical swing states were slightly skewed towards greater vote count of Bush votes than polls predicted- but not much.

    I am curious as to how biased "spoilage" for urban/poorer/dem precincts might figure into this equation as one part of the story. - That is - it is well known that higher non-counting of ballots is more typical in urban areas, and poorer rural (usually dem) areas due to lousy old equipment (punch cards, lever machines, etc)- wouldnt this be consistent with the finding of typical undercounting of Dem votes in Dem states?  - And if I understand correctly, that would not affect the WPE - that is since spoilage affects all voters equally - the percentages Bush/Kerry would not differ from the precinct report - However since this effect would be more frequently found in Dem-leaning counties - It would affect the state vote-count totals.

    Could this not be a partial explanation for point 1) above???

    •  I am just saying (none)
      That I would like to see a state-by-state precinct-level analysis of spoilage rates - and if they show that a Dem vote is more likely to spoil than a repub one - but especially more so in Dem states that would be interesting.

      The cause might not be because the state is Dem as much as it is more likely for Dem states to have spoilage

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site