Re: TAR--audits. urgently need help to get law.

From: Charlie Strauss <cems_at_earthlink_dot_net>
Date: Wed Nov 22 2006 - 00:28:52 CST

On Nov 21, 2006, at 8:43 PM, Kathy Dopp wrote:

> On 11/21/06,
>> Date: Tue, 21 Nov 2006 16:49:43 -0500 (EST)
>> From: charlie strauss <>
>> Subject: Re: [OVC-discuss] TAR--audits. urgently need help to get
>> law.
>> We also liked the idea of counting whole machines rather than say
>> sampling 10% of the ballots in every machine. Our feeling was
>> that most error modalities and most suspciious events are machine
>> or precint specific. By recounting a machine in it's entrirety you
>> get defintive and actionable information about the machine's
>> electronic total accuracy.
> Charlie,
> I agree with most of what you say here.

but of course, since it's not unlike your own approach. Just
streamlined for practicality while retaining the essential protection.

> Your approach of counting all
> the races on sufficient machine counts to detect outcome-altering
> miscount in the closest race is interesting, and may result in routine
> 100% hand counts. I agree with you that ballot sampling is very
> impractical and not as informative as audits that count 100% of all
> ballots associated with particular machine COUNTS.

The primary practical advantages of the other uniform sub sampling
approach are a uniformly distributed and predictable workload,
possibly getting it over with on election night when one is well
staffed, and because it's procedurally easier to check every single
race rather than subsets when the number of ballots to count is small
(like 10% per precinct). Statistically it's biggest advantage is
that it avoids the essential assumption of "what is the largest
undetectable vote swing in a given machine" which plagues the machine-
centric sample.

it's weakness is that it's somewhat harder to take action from the
results since you can't specifically focus on specific machines and
analyze the source of errors. If one is doing counts post election,
with a reduced staff, then I think it's better to concentrate efforts
on a few machines rather than try to open up and sample all of them.

> You should change your language though to audit "machine COUNTS", not
> "machines". As you are aware, auditing "machines" could be easily
> misinterpreted as post-election day machine testing which would never
> detect deliberate vote fraud.

I think the legislation is clear on this. My colloquial discussion
is brushing over the details. But hand counts of machine totals of
actual ballots are what consists of the check.

> Is NM using Precinct-based op-scan (PCOS) machines or Central count

Both. M-100s are the precinct machines. Most but not all counties
will uses banks of M-100s for early voting. But absentee and some
provisional ballots will go into the model 650 central scans.

Our plan calls for batching the absentee ballots into sets that are
roughly the size of a precinct before counting them. Then later
these batches can be considered virtual machines for the purposes of
selection. This essentially normalizes the terminlogy and avoids the
ugly, non-transparent, complexity of trying to weight the statistics
for machines with grossly different numbers pf counts. If all the
machines have roughly the same numbers of votes and one is sampling
more than a handful of machines, the possibility differences induced
by having different numbers of ballots per machine mostly averages
out. Or if you like ergodic arguments it averages out even better
over repeated trials in different elections. THis makes the
calculations simple.

> If PCOS, then you are aware that you are really selecting polling
> locations COUNTS, rather than precinct COUNTS, if you specify the
> machine rather than the machine COUNTS to audit.

I'm completely puzzle by this terminology. We have a machine. it has
some ballots in the hopper and an electronic readout of the totals.
The recount sample consists of picking a machine (or absentee batch)
and seeing if and by how much the total from a hand count of the
ballots differs from the electronic total.

in all cases we are machine centric not precinct centric. But in
most cases there will be one machine per precinct, so there is
commonly a one-to-one correspondence. (However if one only looked at
a canvass total for the precinct then this would also sum in the
early, absentee, and provisional ballots. In that case the apparent
total for the precinct would not be the precinct total for the
election day by itself.)

> If you are selecting
> polling location counts to audit, then you should adjust for variation
> in the total number of ballots cast in each polling location as it
> explains and shows how to do in our paper here:

no we are selecting machines since that's where the correspondence is
made at the finest grain that is still exactly verifiable.
> Best,
> Kathy
> _______________________________________________
> OVC-discuss mailing list

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 Nov 30 23:17:10 2006

This archive was generated by hypermail 2.1.8 : Thu Nov 30 2006 - 23:17:19 CST