Re: Effect of Precinct Size Variation on Audits

From: Kathy Dopp <kathy_dot_dopp_at_gmail_dot_com>
Date: Mon Aug 21 2006 - 15:03:45 CDT

On 8/21/06,
<> wrote:
> Date: Sun, 20 Aug 2006 20:30:51 -0500
> From: Jerry Lobdill <>
> Subject: Re: [OVC-discuss] Effect of Precinct Size Variation on Audits
> To:

> Let's first be clear. The only data that Kathy has from Tarrant
> County TX is that which I have also published
> at This
> data is the result of processing and compressing a lot of the
> original detailed data from a real election. It is very high level
> summary data. If Kathy wants to use it that's fine,

Nice concise paper on audits by Jerry. It is always very good to
bring more attention to this subject.

FYI, Although Jerry's data was very helpful for testing our method of
adjusting audit amounts for precinct-size variations, we are not
including it in our soon to be released paper. We'll be using some
Ohio data instead because we obtained more detailed data so we knew
exactly if we were using total ballots cast versus total votes, etc.
Thank you for providing Jerry as the Tarrant County data was very
helpful for developing our method of adjusting for precinct-size

> Since she seems to be about to publish something different from what
> she's been selling up to now that may involve the data I have

Jerry is selling a mischaracterization of "what she's been selling" all along.

The very first time Jerry, in email, mentioned the possible effects of
precinct-size variation on the ability to subvert audits (by targeting
vote miscount to the largest precincts), I immediately (within
seconds) emailed Jerry to inform him that that is easy to adjust the
calculations of audit size for precinct size variation but is only
needed for audits of precinct optical scan count audits and not for
DRE machine count audits. Then I suggested two ways to possibly do it
that came to mind, but as it turned out I did not use either of those
first thoughts of mine. Nothing has changed about my position, but,
as with any research, it evolves and improves over time.

Jerry apparently disagrees (perhaps still?) with my gut analysis
(above) and that is his right. I also made a temporary minor
programming error in my precinct-size-variation adjustments that had a
big effect which I discovered myself and corrected and let everyone
know about on this list.

> mentioned above, and considering her behavior in the recent past, I
> have decided to publish my paper on mandatory election audit design
> now. I have sent it to a large number of researchers working in the
> field of audit design and election integrity assurance. Some of you
> may have received it directly from me today. If not, and you'd like
> to have your own copy of it you may download a pdf file of it at

I see that Jerry's paper uses the same probability formulas that NEDA
publicly recommended for calculating audits in June, 2005 and that is
great. The more various works confirm each other's methods, the

> corruption. The metrics permit the segregation of polling places into
> two clusters, one of high probability of potential irregularity, and
> one of lower probability.

I see that Jerry has taken the clustering approach to adjusting for
precinct-size variation that I first suggested to him in a quick
sentence in my email response to him re. two possible ways that came
immediately to my mind for ways that precinct-size variation can be
adjusted for. Although I did not end up taking this approach
myself, it seems like a fine approach (although I've not studied it in
depth like Jerry has)

I am glad that Jerry added another alternative to the public domain,
like the recent paper by O'Dell, Simon, Mitteldorf, and Freeman on why
HR550 would not succeed in detecting outcome-altering vote miscount.

It is important for election integrity activists not to nitpick about
details at this point in time.

One reason that NEDA was able to persuade the Carter-Baker Election
Reform Commission Director, Robert Pastor, to recommend audits (the
Carter-Baker was the first group after NEDA to recommend election
audits, and the US GAO was the second, both of whom NEDA had
communication with on this issue of election audits), is because the
paper we released was short, sweet and simple and made it all sound

Jerry's paper adds something valuable to the discussion and is nice
and brief. I have no problem with Jerry's method, although I've not
studied it in depth.

Howard Stanislevic of VoteTrustUSA also publicly released a method
that (although partially trial and error like my July 16, 17 paper)
that is virtually identical to the method that Frank and I have been
working on implementing since July (but I initially made a tiny
programming error that had huge effect). We will be publicly
releasing it soon, and the NEW thing about it will be a direct
solution to the problem of calculating vote count audit sample sizes.

Unfortunately the soon-to-be-released paper by Frank and I is not so
short and sweet, although it suggests a slightly easier (IMO) method
of adjusting for precinct-size variation, plus a direct method to
calculate the audit sample size.

Thank you for your work on this Jerry.

> Clustering the polling places allows a two tiered audit. The first
> tier is small sample of polling places from the small high suspicion
> cluster. The second tier is a larger sample from the larger cluster
> with lower suspicion. There is a 95% probability of detecting
> irregularity for each cluster. If the high suspicion cluster yields
> evidence of corruption the audit need go no further, and a complete
> hand recount would be ordered. In any case, the audit requires
> perhaps less than 5 polling place recounts or as many as 32 depending
> on whether and when an irregularity is discovered. The total number
> of polling places was 211, serving 634 precincts.
> Jerry Lobdill
OVC-discuss mailing list
= The content of this message, with the exception of any external
= quotations under fair use, are released to the Public Domain
Received on Thu Aug 31 23:17:08 2006

This archive was generated by hypermail 2.1.8 : Thu Aug 31 2006 - 23:17:10 CDT