VVPB, precinct totals posted, hand counted precinct level random....

[GentryLange]

Audits. Most of the country is Optical Scan... Right? So reflashing the Diebold machines using Open Sourced Publicly Owned Code... and then instituting procedures for public counting, and random recounting at the precinct level. No central tabulators, and simple precinct level totals posted publicly, and then transmitted by phone. You could randomly recount ballot boxes based on two rolls of dices added, from different party reps or something...

Ok there's my thougts for tonight.

[Boredtodeath]

1. Random recounts don't work

Quicktime movie demonstrating why (thanks to VotersUnite.org)

http://www.votersunite.org/info/RandomSample.mov

[yowzayowzayowza]

2. Imprecise.

A random audit of PRECINCTS won't work, but a random audit WITHIN each precinct will. Ie, rather than auditing x% of precincts you audit x% of ballots in each precinct.

[Boredtodeath]

3. Simply not true

Watch the video again - presume the 100 dowels are 100 ballots; pull 20% you still get NO ERRORS.

In a random sample - let's take 100 apples - 20 of the 100 are rotten. You randomly pull out 20 apples. You have NO GUARANTEE you'll get even 1 of the ROTTEN apples. You can't even SAFELY assume you'll find 1 rotten apple.

Not imprecise at all.

[yowzayowzayowza]

4. Umm..

What would 20 rotten ballots look like? What is it our sample is testing?

[merwin]

6. Let's say you test 5% of each precinct.

If the precinct had 100 ballots cast, that would only be 5 ballots. To actually semi-reliably detect a discrepancy, you'd have to be hand recounting at least 25%. And even that has a good possibility of coming up with nothing.

Random recounts of any sort are equavilent to putting a lock on a door. It'll keep out people who aren't intent on stealing, but might steal if the door was unlocked. It won't, however, keep out someone who is intent on stealing.

[yowzayowzayowza]

7. Umm... again.

I've actually studied statistics and implemented them in processes. Plz answer these questions:

What is a rotten ballot?

What is our sample testing for?

[merwin]

8. A rotton ballot is one with a different result than originally tabulated.

The sample is testing for outright ballot stuffing and tabulator accuracy errors.

If you count a small portion of each precinct, you could very well miss the stuffed ballots and won't be able to test for undervotes since you're not going through each ballot in the precinct.

For example, at the airport, would it be better to search one or two pockets in everybody's carry-on's, or the whole bag of a random number of people.

[yowzayowzayowza]

9. A stuffed ballot looks ...

no different than a real one. Sampling will never identify ballot stuffing. Tabulator errors are another matter.

If each precinct has itz own tabulator, her example is correct. If, however, I test EACH tabulator it will very likely identify all the counting errors. The key is matching your test set to the machines not the whole population.

On edit: for more clarity...

Her example: If each precinct has itz own counting device, then testing a random few precincts is unlikely to find the few which are miscounting.

OTOH, if each counting device is tested with a few random ballots, then the counting errors are VERY likely to be identified.

Make sense?

[TruthIsAll]

12. If 20 of 100 are fraudulent, prob=98% that 15 at random will show it.

Probability of NOT detecting fraud if x of 100 total precicnct ballots are chosen at random and 20 are fraudulent:

x Prob (NO fraud detection)
5 25.2% (1 in 4)
10 7.4% (1 in 14)
15 2.0% (1 in 50)
20 .5% (1 in 200)

Probability of detecting fraud = 1- Prob (NO fraud)
Therefore the probability = 1 - 0.5% = 99.5% that fraud will be detected if 20 of 100 ballots are tested.

[merwin]

5. Both methods would end up with the same results.

Both are completely random (in theory), and the chances of coming up with potential fraud (or missing fraudulent precincts) is pretty much the same in both cases.

[TruthIsAll]

11. The probability is 1 in 200 that ALL 20 apples would be good

Rem Good	Rem	P(Good)Joint Prob
0	80	100	0.800	0.800
1	79	99	0.798	0.638
2	78	98	0.796	0.508
3	77	97	0.794	0.403
4	76	96	0.792	0.319
5	75	95	0.789	0.252
6	74	94	0.787	0.198
7	73	93	0.785	0.156
8	72	92	0.783	0.122
9	71	91	0.780	0.095
10	70	90	0.778	0.074
11	69	89	0.775	0.057
12	68	88	0.773	0.044
13	67	87	0.770	0.034
14	66	86	0.767	0.026
15	65	85	0.765	0.020
16	64	84	0.762	0.015
17	63	83	0.759	0.012
18	62	82	0.756	0.009
19	61	81	0.753	0.007
20	60	80	0.750	0.00495

Prob = 1 in		1/.00495 =	202.144

[yowzayowzayowza]

10. Now ya went and dunit.

I gota looken at yer source site a lil more:

http://www.votersunite.org/info/auditingissues.asp

Their analysis is flawed by the same erroneous assumption: that the sample must be from the ENTIRE ELECTION.

However, the continuity required for a statistically significant sampling does not exist in this context.

Each election has a unique set of contests and candidates, and each county prepares many different, unique ballots since different districts and even different precincts can offer a different set of choices to voters. Each ballot with a unique set of choices is called a "ballot style."

They correctly state that continuity does not exist over the entire election but a statistically significant sample taken from EACH "ballot style" origin (precinct, district, ward, whatever) would be well defined. This is a simple population granularity issue. Each "ballot style" origin would provide a stack of audited ballots for validation. The total ballots audited would be the same. Very poor analysis, IMHO... IT CERTAINLY DOES NOT SHOW THAT RANDOM RECOUNTS DO NOT WORK.

[GentryLange]

13. Precint Level Handcounts and Publicly posted totals

Are just as important to me as VVPB. The random hand recounts, combined with Publicly owned open-source code, triggered audits in the case of anomolies, etc.

Mine is not an exhaustive list. And random hand counts are not the only part of the package. I will put together a more exhaustive list in the coming month. But anyone who wants to help, feel free.

Gentry

[FogerRox]

14. The problem we face is the definition of Random

A state wide race--
Random audit of 3% of precincts-- truly random -- like a roll of the dice--could end up auditing all 3% in one area. AS in the rotten apple scenario.

YOu commit fraud you pick 5 numbers between 1 & 100, ASK a friend to pick the Audit precincts-- 3 numbers between 1 & 100. Chances are you get away with fruad, try it again and again. Most times I do this the bad guy gets away with it at 3 times. Which to me suggests that we need to have an audit that is minimum

I've played a lot of Miniatures war games, and every once in awhile somebody rolls 10 dice and 9 are say all the same number. That would be potentially very bad for us.

Any way what Im leading to is Rush Holts definition of what a Random audit should be, which may be a very good definition. It may be that an audit should not be truly random.

[yowzayowzayowza]

15. You do not audit ...

random precincts, but rather random ballots in EVERY precinct which must post their total ballot count and random audit totals BEFORE forwarding the ballots to central tabulation who must in turn post matching numbers or the precinct faces further manual audits.

[FogerRox]

16. kick for Yowza cause he gets it

[Carolab]

17. Yes! Exactly!!!! n/t

[Stevepol]

18. Another idea: have a random 5% audit but add to the random audit

some chosen precincts, 3-5 chosen by one side, 3-5 chosen by the other side, some number so you have random audits plus counting of a couple chosen precincts which have high indications of fraud (of course each candidate should be able to choose a couple of these precincts).