Skip to main content

View Diary: Armando's Challenge, Or The Informed Citizen's Guide To The 2004 Election (389 comments)

Comment Preferences

  •  wiki and fraud (4.00)
    I'm not sure what to suggest in terms of putting it on a wiki.  Some wikis allow source material to create a pdf, but this is going the other way around.

    I think part of the confusion about this whole issue is what fraud is, so I agree on the need for a fraud definition.  We are talking about so many things here.  In an attempt to integrate all of it, we are talking about:

    1. Whether there was opportunity for irregularities
    2. Whether there was evidence of irregularities
    3. Whether there was proof of irregularities
    4. Whether said irregularities were enough to change the state result
    5. Whether said irregularities were deliberate, IOW "fraud".

    Furthermore, there's a huge difference between irregularities concerning existing ballots, and irregularities concerning opportunity to vote: an honest recount could overturn the former.

    To me, "fraud" is the same thing as a "deliberate irregularity".  But an irregularity can exist without it being fraud.

    I currently think there is no proof of ballot-level result-changing fraud.  I also think there is not proof of result-changing irregularities in general.

    I think there is plenty of everything else, though.  There was opportunity for result-changing fraud.  There is evidence of fraud.  There is possible evidence of result-changing irregularities, but this is very much up in the air.

    What I would really like to see:

    1. One clearly described case of evidence of result-changing, ballot-level fraud.  (I really don't think we have that; I think it is more likely to be a possible result-change when you add in all fraud and irregularities together.)  Arnebeck has alluded to this, but he really hasn't shown anything yet.

    2. A mathematical proof that deliberate and nondeliberate irregularities were enough to change the result - this would include "opportunity to vote" matters. It would not be overturnable through a recount, but it would help vastly with credibility.  I think we do have this, but no one has bothered to pull it together in an airtight paper yet.

    I disagree with Armando about exit polls.  I think that whole thing should just be left alone, at least by us.  There are way too many unknowns that even the PhD's have just been skipping past, only to make asses out of themselves.  I am not an expert at statistics, but I know enough to know that clustered samples don't follow the same margin-of-error rules as random representative samples.

    Exit pollsters don't interview at all polling places.  They pre-select polling places.  They do not know how much turnout each polling place will receive, they cannot project.  Each polling place has a different Bush/Kerry ratio.  Since they cannot project turnout at each polling place ahead of time, then that means that higher than expected turnout in some polling places can warp the results of the exit poll raw samples.  That's just logic.  It's also exactly what happened in this election - unexpectedly high turnout in small rural precincts, nationwide.  (That's my impression of the effect of having many pre-registered evangelicals turn out that didn't vote in 2000, anyway.)  I have not yet seen any explanation that disproves that as a reasonable explanation for much of the exit polling "problem".  Instead I see a lot of people say,

    1. Why was the exit polling wrong?  When actually, the exit polling is supposed to be wrong in the way they are looking at it.

    2. Why was the exit polling not accurate when it's been accurate in the past?  When actually, exit polling is used in an entirely different way to predict election results than how we are choosing to judge its accuracy.

    I've seen both Ruy and Mystery Pollster explain that exit polls routinely overrepresent the Democratic side, because they are seeking to interview diverse populations for greater demographic variety in their datasets.  And when you have more black, asian, and female respondents, you've got more Democrats.  Diversity means Democrats.  The very nature of exit polling means that it is supposed to be corrected later by the actual vote results.

    Some of us do have a more advanced, yet still partial understanding of how exit polls work.  Weighted but not yet corrected.  Crosstabs, partial returns.  Etc.  But no one has really pulled it all together and explained it all.  We've got many attempts to pull together the exit polling argument, but many of them don't even know what clustering or stratifying means.  I don't fully, either.  But I sure know when an author knows less than I do.  None have even attempted to reconcile their arguments with the fact that the pre-weighted-yet-final exit poll numbers in 1988 showed Dukakis beating Bush.

    So leave the exit polling stuff alone.  There are too many unknowns.  Too many unknown unknowns.  Wait for the team of blue-ribbon statisticians to analyze it after the data is released.  Otherwise we are just going to be plastering five layers of intelligance on top of ten layers of ignorance.  It will only be a distraction from the kind of thing that a community like this WILL be able to do well - assemble together known bits of proven information into a compelling presentation.

    •  To be sure (none)
      I don't know from exit polling, but I actually know a great deal about polling generally.

      I think your point on adjusting the raw data to the actual VOTING population is the central problem in all polling - drawing a representative sample from the relevant universe.

      Here the problems are twofold - the representative sample cannot be drawn BEFORE the Election because the relevant universe - who actually went to vote - won't be known until, well people go out to vote.

      It is my central problem with the pre-election likely voter models - the pollsters pretend to know much more than they ACTUALLY do as to who will come out and vote.  The EARLY exit polling -by definition, is given too much credence - the relevant universe - who voted - is simply not fully formed at that time.

      The problem of voter suppression (in the broad sense, including undervotes, etc.) significantly exacerbates this problem.

      And this is the entire rub on the fraud issue - it seems to me the strongest evidence - circumstantial of course - has been the theory that the actual results are widely divergent from the early exit poll results.  But, as you say, the exit polling is shaky at best.

      I often said that "All Polls Suck" and I largely stand by that.

      I'm thinkin', I'm thinkin'

      by Armando on Mon Jan 03, 2005 at 07:28:51 AM PST

      [ Parent ]

    •  As to stratifying and cluster sampling (none)
      I think you make too much of that issue - those are merely techniques to insure a representative sample is drawn. I doubt hte problems stem from that.

      The problem is in NOT knowing the relevant universe.

      I'm thinkin', I'm thinkin'

      by Armando on Mon Jan 03, 2005 at 07:30:18 AM PST

      [ Parent ]

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site