CALL FOR SUBMISSIONS -- GOLD STAR AUDIT PROTOCOL

[Einsteinia]

See the in-progress website for the CALIFORNIA ELECTION PROTECTION NETWORK

at: http://www.califelectprotect.net

Then click on the icon in the left hand column with the gold star on it for details. . . .

[Wilms]

1. A CALL FOR SUBMISSIONS

C a l i f o r n i a ’ s V o t i n g I n t e g r i t y

-snip-

A CALL FOR SUBMISSIONS

We ask everyone who has input on what would constitute a reasonably airtight audit protocol (that is easy enough for the average person to follow) to weigh in on what the parameters of an ideal Gold Star Audit protocol should be.

-snip-

If you have some good ideas in this regard and seek the reward of repairing democracy, we need your input. Please submit your paper (using an easy-to-read outline format as provided above), with your contact information, to: [e-mail] by October 15th.

Thereafter, at a date yet to be determined, we will notify finalists of a panel discussion in a town hall setting where these papers may be explained and debated. The final synthesis of this process will be presented to our California legislators for a possible oversight hearing on these remedies to the inadequacies of the current Election Code. From this process we predict legislation will be drafted for the 2006 legislative cycle.

http://www.califelectprotect.net/GoldStar_091305.pdf

[Einsteinia]

2. Wlms, THANK YOU!

. . . . much better!

[Bill Bored]

3. Oh gee.

Start by reading this paper by Kathy Dopp please. (Don't worry, it's NOT about exit polls...much.) It's hypergeometric though and that's important!

<http://uscountvotes.org/ucvAnalysis/US/paper-audits/Paper_Audits.pdf>

Then look here:
<http://www.votersunite.org/info/auditingissues.asp>
Also hypergeo.

And then find that demo on the web that shows how this stuff works, which Ellen said she was going to link to on Voters Unite with my instructions on how to play, but probably hasn't done so yet, so now I have to post it here again myself! (HARD WORK...workin' HARD! But that's OK. We're problem solvers. We solve problems!) OK, here it is. It is here:

<http://www.math.csusb.edu/faculty/stanton/m262/hypergeometric_distribution/hypergeometric_distribution.html>

Then change N to 51 and hit Clear. Then select 2 squares to simulate corrupted voting machines. Then hit Play and try to find one with that random audit of 4 machines which is a whopping 8%! You almost never do on the first try!

And the moral of the story is: If you have a close election that can be swung on a few machines, you have to audit a lot more than if you have a landslide that can only be swung on lots of corrupted machines.

So I guess the best auditing protocol is (drum roll please!) a HYPERGEOMETRIC one!

Now can you get rid of that robot guy please?

[Einsteinia]

4. I'm not a statistician

I do have Kathy and Ellen's thesis' on file.

Correct me if I'm wrong, but on the face of it, it appears that Ellen's is wrong because she, in her video, demonstrates something like:

If 100 people have a white cork and only 5 have black corks, that the black corks usually don't get selected. (Forgive me if I got this wrong--it's been a long time since I saw the video.)

I thought the whole theory of statistics with the laws of probability, is that:

It isn't that a black cork would be revealed in a selection that happened once or twice, but that, as in the case of California, that it would occur over 3,000 times. In that event, it is likely that an error would be revealed.

Can you explain if this for lay people if this is the right track for my critique or am I just as dumb as I thought I was?

Also, there is a parameter in what a Gold Star Audit protocol might require that says no precinct can exceed 1,000 voters -- so as to eliminate the huge disparity between the small and large groups. So, if we set parameters that would catch error in a sample of 1,000, then wouldn't that cover it? Or?

Good news. I won't be judging any submissions--others who know more will.

[Bill Bored]

5. Don't feel bad, they say Einstein wasn't that good in math either.

Neither am I sometimes, but the video demo used a small sample of 100 precinct machines. The video showed that with 5% of those machines corrupted, which could be various numbers of actual votes but is limited by the total number of votes cast on these 5 machines, it is highly unlikely that a random audit of 5%, 10% or even 15% will find ANY of the corrupted machines.

Now, if you plug these numbers into Kathy's spreadsheet, which is associated with her paper, you get this:

With 5 out of 100 machines corrupted:
5% audit = 23% likelihood of finding one
10% audit = 42% likelihood of finding one
15% audit = 56% likelihood of finding one

Still almost a 50% chance of NOT finding even one corrupt machine with a 15% random audit! That stinks doesn't it?

Now here's the good news: Most jurisdictions have more than 100 machines, and the more machines you have, the greater the likelihood of finding a corrupted one.

Believe it or not, in a 1000-machine jurisdiction, the same 5% audit as above, has a 92.8% chance of finding a corrupted machine. This is because there are 50 of them and the number of times the audit is repeated is 10 times more than with 5% of only 100 machines! You get 50 chances to win the lottery instead of just 5.

Now, the other good news is that we are talking about 5% corrupted machines. If all the machines have the same number of votes, this means that the election would have to be fairly close for the outcome to be reversible by 5% vote corruption or less. So if you have a landslide, you don't have to audit as much. A landslide, if it's fraudulent, would require many more machines to have been hacked, at least one of which will be easily detectable by auditing only a few of them.

So the larger the jurisdiction, the easier it to find significant fraud, and the closer the election, the easier it is have fraud large enough to decide the outcome without detection.

So watch out for close elections in small counties!

There is no inconsistency between the video and Kathy's work that I can observe.

Keep in mind that all these tests may only find one corrupted machine out of many. It's still necessary as part of the auditing protocol to audit more whenever a corrupted machine is found, or whenever the election is close enough to be overturned by fraud on the remaining unaudited machines that might not be detected, according to the spreadsheet.

This is why it's hard to design an auditing LAW for every occasion, but it's not too hard to make a law that says to audit more whenever there is a doubt or a reasonable possibility that the outcome may be incorrect.

Get it?