US count votes on Random Audits

[FogerRox]

Here is the new pdf on how we can ensure Accurate Election results
http://uscountvotes.org/ucvAnalysis/US/election_officials/Audits_Monitoring.pdf


"How can we audit vote count accuracy? To have scientific validity, precincts must be selected for audit through a truly random selection process that gives every machine or precinct an equal chance of being selected."

I emailed Kathy about this above quote and it -- to me -- is somewhat contradictory. A truely RANDOM Audit--- like a roll of the dice -- where 6 die all come up sixs-- it does happen--not to often but is does happen would be unacceptable in a VVPB audit

This would not give- in my mind-- every voting machine or precinct an equal chance-

A NJ Lawyer told me that she likes Rush Holts definition of Random- best. ANd guess what --its not a pure mathematical randomness. Its a sample if you will,

What are your thoughts?




.

[napi21]

1. It really depends on how they really do their selection.

Random tests are done all the time in the accounting business when they do an annual business sudit. If there are no material errors found, the test is done. If even one material error is found, the selection is wodened and the test is done again.

I could accept this same kind of test in voting, IF it was really done that way.

[FogerRox]

2. Right- but what is the definition of Random?

if you have 6 dice-- and roll them a bunch of times-- eventually they will come up 6, sixs.
ANd from thhe standpoint of Voter Verified paper record laws -- this would be very bad---

What i'm saying is true mathematical randomness is not desirable. Those dice that come up 6, sixs would represent -- for expample-- all the precincts in the corner of the state get audited--- ignoring all other areas-- THIS IS BAD. If you audit a minimum of one voting machine per polling place-- that is a different version of Randomness, and maybe more desirable---

This is where a statistician could describe why --- in more detail, and better than me-- maybe someone like Febble, or On the other hand would do this issue justice.

[OnTheOtherHand]

3. not really my field, but

Foger, I agree that if you think fraud might be concentrated in certain regions (which you may not know in advance), you might want a geographically stratified sample -- perhaps sampling some percentage of precincts from each region of the state, or each type of precinct (e.g., making sure that heavily Republican and heavily Democratic precincts are proportionately represented). The best approach would depend a lot on your notions about the likely fraud tactics.

However, if we're just trying to spot at least one corrupted precinct, then Kathy's analysis should be pretty much worst-case. Concentrating the fraud in a region, or a type of precinct, should not reduce the probability of detecting fraud below what Kathy's analysis indicates. Rather, stratifying the sample should make it even harder to avoid detection.

And, just based on a little Excel doodling, it doesn't necessarily matter much. Suppose there are 1000 precincts, 30 of them are corrupted, and 50 of them are sampled: the probability of sampling at least one of the corrupted precincts is about 79%. Now suppose that there are five regions (or whatever) with 200 precincts apiece, 30 of the precincts in one of those regions are corrupted, and 10 precincts are sampled in each region. The probability of sampling at least one of the corrupted precincts in that region is about 81%. Not a big difference.

(Let me know if there are other sorts of scenarios we should consider -- I sure haven't thought deeply about this.)

All these percentages assume that if a precinct is corrupted, a recount will reliably identify it as corrupted. That's probably a fair assumption for concentrated fraud.

(EDIT -- I didn't see the suggestion of one machine per polling place -- umm, isn't that going to be an awful lot of auditing? Surely more than a 3% or 5% sample.)

[FogerRox]

9. OTOH-- on your edit-- SB55 VVPB law in Conn. audit rate is basically just

that. 1 Machine per polling place-- and it might be an Audit rate of as high 14%--depending-------------- on voting machine disbursment.

My larger point is that Rush Holts bill has what I think -- and have told is the "better" definition of "Random". The Executive Director of NJ Applessed Renee Stienhagen expressed her opinion that Rushs was tops---

AS to your 1st point-- We cant really write VVPB law based on "where" - one might think is to take place--- dont get me wrong-- but a true random selection of precincts is not--
I think--what is needed---- odds are, that selection -- at one point will be as the dice roll, 6 die-- all coming up sixes--- OTOH isnt that unacceptable?

[FogerRox]

12. the lawyer who s opinion Im parroting is the Executive director of NJ

Appleseed-- of the national Appleseed Foundation-- Renee Steinhagen-- she liked HOlts random selection of precincts to audit--- over the random selection of precincts to audit in SB55 Bill in Conn. Though SB55 would do more "auditing". Truckin said 8% to 12 % IIRC.

Renee described it as a better deifintion of random, And my guess is that Michelle Mulder would agree.

[FogerRox]

14. late edit--

Correction-- We cant really write VVPB law based on "where" - one might think fraud is to take place---

Kathy says truely random, but latter uses a qualifier: "Sample" which is where I agree.

My understanding -- not being strong in math as --Kathy said--- is that a sample might be weighted--- so its no longer random in a pure --(my word) & True sense.
SO can a sample be more effective in an audit, rather than a Random audit?

With my math impaired brain-guessing- I would say yes-- a sample might be more effective.


"precincts must be selected for audit through a truly random selection process"

truly random vs a sample--- ?

Any thoughts?

[sunshinekathy]

4. That's not an audit

First Roger, Please read our detailed suggestions for auditing precincts so that you understand one method of auditing:

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

Second, you are confusing an audit with a hand recount of every precinct, i.e. of the entire election results in the case of precinct-based opti-scans since there is only one machine per precinct.

Third, Selecting precincts (or machines) is not remotely like throwing a die because once you select a precinct, it is not still available for selection again. i.e. it is not possible to roll the same precinct twice, as it would be to roll six twice.

Forth, As it says in the handout, audit procedures would need to be determined separately for each COUNTY and state, so if each county were being audited, it would not be possible to miss counties. Obviously one would audit every county's election results and not skip any. In addition, some county's audit procedures may want to ensure auditing at least one precinct from every district for each race, as well as doing a county-wide audit.

Fifth, If you don't select precincts (or machines if you prefer to audit machines - which I think is not as practical), randomly, then are you going to let:

1. The election officials select them like Blackwell did in OH?

2. The candidates select them like Gore did in FL in 2000?

In both cases the selection processes above failed to select precincts that had voting errors, although in FL 2000, Gore would have won if he'd used a random audit procedure for selecting counties, and it is very possible, in fact highly likely that there were many incorrectly counted precincts in OH in 2004.

Sixth, If Holt's bill does not use random methods to select his sample for auditing, then it is not going to be an effective audit procedure. Holt is a physicist, so he's probably smart enough to know that truly scientific procedures to ensure that samples for audits are chosen randomly must be used. I believe that you are mischaracterizing his bill.

Seventh, Do you always immediately assume that anything you don't yet understand is incorrect, or assume that your own illogical ideas are shared by others, even despite the evidence? i.e. Do you always immediately try to put the kabosh on anything that you don't understand?

Audits and election data monitoring happen to be the only practical solutions to fixing our election problems by 2006 because it does not depend on fixing the voting systems first, or passing legislation which would be difficult to pass, given that many of those in Congress today got there by miscounted elections.

http://uscountvotes.org/ucvAnalysis/US/election_officials/Audits_Monitoring.pdf

You could have replied to my thread, rather than attempting to make it sound as though these doable practical solutions are not a good idea and mislead people with your illogical nonsense. What is your motivation?

Please make some attempt to understand something before you attempt to wrongly convince others that it is wrong.

Why would you want someone like Febble, who understands math, but does not consistently apply logic correctly and misled the press and others very wrongly, to help you?

Please open your mind to the possibility that it would be possible to detect and correct all vote count errors with a combination of independent randomly selected audits and detailed election data reporting if one does not assume such illogical, unlikely procedures as you suggest here.

I suppose, if one were doing a state-wide audit, that there would be a miniscule probability of selecting all precincts in one county in a state using the selection method that you suggest. That is true, but that probability would be calculable for any particular state. If you're that worried about it, why don't you calculate its probability. However, no one said to ONLY do a state-wide audit and conduct no county-level audits anyway, so your discussion is not relevant to our proposal. However, if you think that calculating such a probability is like rolling a dice, then you probably have no idea of how to begin to calculate it.

Would you prefer that no independent audits were ever done to protect your banking institutions from inside embezzlement because the audit might in one out of ten million or more tries select a lopsided sample to audit?

What is your real motivation in saying such obviously nonsensical things to tear down the only practical ways to ensure vote counts are accurate in America? Do you want to leave our elections entirely open to insider embezzlement and errors rather than apply the solutions that would obviously ensure their accuracy? Are you eager for another round of the wrong candidates being sworn into office following the Nov 2006 election so that the probability of ever fixing our elections is drastically reduced once more?

Best,

Kathy Dopp
http://electionarchive.org

[FogerRox]

10. Well now-----------

Kathy-- I think you need to clam down -- and behave. Then apologize---

for this-----


What is your real motivation
saying such obviously nonsensical things
Do you want to leave our elections entirely open to insider embezzlement
Are you eager for another round of the wrong candidates being sworn into office
Would you prefer that no independent audits were ever done
then you probably have no idea
illogical
Please make some attempt to understand something before you attempt to wrongly convince others that it is wrong.
You could have replied to my thread, rather than attempting to make it sound as though these doable practical solutions are not a good idea and mislead people with your illogical nonsense. What is your motivation?
Do you always immediately try to put the kabosh on anything that you don't understand?
Do you always immediately assume that anything you don't yet understand is incorrect, or assume that your own illogical ideas are shared by others, even despite the evidence?
Do you always immediately assume that anything you don't yet understand is incorrect, or assume that your own illogical ideas are shared by others, even despite the evidence?

[FogerRox]

11. Additionally Kathy the reason I didnt

respond to your thread is it didnt exist. I posted in the AM you posted in PM

[Febble]

17. ahem....

Kathy, you are quite correct. I am not the one to help in this case. I don't know nearly enough about the way US elections are run.

I do absolutely agree with you that random precinct selection is vital, although I can see that some kind of stratified random sample might be useful. The vital thing, as the San Diego study exemplifies, is that the selection needs to be unpredictable, which also means being done immediately before the recount, so that no "preparation" of ballots can be done. It also, I would say, needs to be independently scrutinised.

It is hard for a mere Brit to understand these things as the UK system works so well (apart from in increase in voter fraud this year due to the expansion of postal ballots). We vote on paper, ballot boxes are sealed in front of witnesses at the polling station, transported to the count, and unsealed in front of witnesses. Then the count is done, in public, under the scrutiny of bipartisan volunteers, and not concluded until all candidates are happy with the count. If they are not, a full recount is done immediately. At the very minimum, US election officials need to tighten up their procedures for ballot security (where there are ballots).

We are on the same side here, as I am sure is Roger.

(edited for grammar)

[FogerRox]

19. same side - fighting like hell- yes of course (Homer Simpson -DOH!)

starting this thread off I used this quote-- the bold is mine--

"How can we audit vote count accuracy? To have scientific validity, precincts must be selected for audit through a truly random selection process that gives every machine or precinct an equal chance of being selected."

Is there a better selection of words-- I have read some VVPB law and find differences in the way RANDOM SELECTION IS mandated, as I compared HOlts bil and S55 in Conn in a previous post-- Folks like Michele Mulder (from Rushs office) and Renne Stienhagen of NJ Appleseed, Both lawyers--- have told me that there is random and then there is random.

I would think election officials could digest sliding scale as easy as truly random.
What is the writers intent?
Is it important enough to not use truly random?
Or another term?

A very simple thing is being put forward here--- certainly not worth............

I think you all get my drift here.

[Febble]

20. Randomness is a critical quality

However, as you suggest, there are different ways of applying the quality of randomness. "Stratified" random sampling means that you decide in advance on categories you want to see represented, and then select randomly from within those categories. I think there are lots of issues here.

For example, in the Ohio recount, IIRC the law required that a certain percentage of ballots should be recounted. In some counties it appeared (I am going from memory) that election officials chose large precincts for the recount so that the percentage of ballots required would come from the minimum number of precincts. This is clearly absurd.

In fact "percentage of ballots" would seem to me to an odd requirement, as fraud is going to be detected at precinct level, not vote level. "Percentage of precincts" seems to me more sensible. Maybe this is how it is in some places. I'm just thinking about it theoretically.

So you might want some kind of stratified random selection for audit whereby precincts were "stratified" by size, and a certain percentage randomly selected within each stratum. Then if fraud is targetted at, say large precincts, large precincts will definitely be represented in the recount. Similarly with precincts in shared polling places - again you might expect fraud to be targetted in particular polling places rather than precincts. So you could randomly select polling places rather than precincts, then randomly select precincts within polling places. And machines within precincts.

So the random (and therefore unpredictable) element is present, but by defining your categories in advance you can ensure that particular categories of precinct are represented in your sample. It's the same as any quality control procedure. If I were editing that sentence, I would simply insert the words "within any given category" after the words "machine or precinct". But I agree with Kathy, that within each category being considered, the selection has to be "truly random" - as in tossing coins - or using random number generators. Which means that within a category every precinct or machine has an equal chance of being selected.

And I was trying to make the further point that the critical factor is unpredictability - randomness is only part of that; so is minimising the time interval between random selection and count - and keeping the ballots secure.

But what can be practically achieved before 2006 is another issue, of great importance.

[Bill Bored]

5. Personally, I like this one:

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

But as far as Holt's vs. other definitions of random, AFAIK, Holt and a number of others state that the percentage (e.g. 2%) must apply to each jurisdiction or county in a state, thus ensuring that at least one machine in each county will be audited. So it's random within each jurisdiction but there is a minimum of one machine per jurisdiction.

What else specifically about Holt's definition, in HR550 I assume, are you referring to?

As far as what random means, I'd say it refers to a uniform distribution such that every machine in a jurisdiction has an equal chance of being selected.

Having said that, I think we must be prepared to assume that there could be fraud sufficient to swing the results of a close election, that would NOT be detected by a random audit. Such fraud is likely to be concentrated on a small percentage of machines. E.g., 3% corrupted machines would most likely NOT be detected by a 3% random audit.

So we need a way to get additional audits -- either random, or cherry picked by the losing candidate, or both -- until we reach a point where the margin of the election can no longer be overturned because the odds that there is a large enough discrepancy to do so are very small. This is why I like the paper I've cited above, also by Kathy, which spells out exactly how to arrive at these probabilities, assuming the selections are random. What is needed however are laws to allow such audits in the absence of any discrepancy in the initial random audit.

Now I will read the rest of this thread, which I hope is as at least as civil and well considered as our other debates (except of course for those about exit polls!).

[sunshinekathy]

6. But it is only half the solution to ensure accurate vote counts

Bill said:

Having said that, I think we must be prepared to assume that there could be fraud sufficient to swing the results of a close election, that would NOT be detected by a random audit. Such fraud is likely to be concentrated on a small percentage of machines. E.g., 3% corrupted machines would most likely NOT be detected by a 3% random audit.

Bill the percentage that must be audited in order to detect at least one corrupt precinct if 3% of precincts are corrupted depends on several factors as outlined in

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

and so to determine one audit percentage for every state and county in America, if that's what Holt's bill does, is not the best solution. However, if it passes, it would be a wonderful improvement, but then every county will have a different probability of detecting various percentages of corrupted precincts.

PLEASE read this overview of how the COMBINATION of audits and election data monitoring would be likely to detect ALL types of miscounts that might cause a wrong candidate to be sworn in. It is not enough to do audits, we must also collect and analyze the detailed election data. Please read this handout and give it to your local election officials:

http://uscountvotes.org/ucvAnalysis/US/election_officials/Audits_Monitoring.pdf

It's only two pages and non-technical and a very commonsense approach to ensure election accuracy that could be implemented by the Nov 06 election if it gets funding soon enough.

Best,

Kathy

[Bill Bored]

8. I have read it. See post #7.

We do not disagree in principle.
I choose to emphasize the legal aspect because in many states, it seems that elections are conducted lawlessly! This must change.

[FogerRox]

13. yes -- so far election reform in the guise of VVPB is state by state

so at his point looking at it as a state problem might be appropo. Bill you know a bit more about the particulars of HOlts Bill VS. say the bill in COnneticut SB 55

Which definition of Random do you like better? I'm guessing you would say HOlts bill --

[Bill Bored]

15. Well, at least I understand Holt's bill.

The other one, I'm not sure about and can't remember the particulars.
I don't know how they define election districts, or whatever they're called in CT, but they do seem to go out of their way to define randomness in a mathematical way, which may not be a bad idea.

But Kathy's and my point is that the likelihood of detecting corrupted machines depends on the number of machines in the jurisdiction and therefore ideally, a sliding scale might be better. I.e., audit a higher percentage in counties with fewer machines.

[FogerRox]

18. Bill understood-- sliding scale

Which is why I dont go for the language of "truly random", which in my math fuzzy brain is this----Sliding scale is not a truly random selection is it?

Let me put it this way-- VVpaper trial or vvpaper ballot? I like VVPballot---
part of that problem is that trail might be confusing to legislators, if you use the word trial enough it becomes acceptable-- and trail is not acceptable.

SO if a sliding scale is not truly random--- then dont say truly random--- say sliding scale.

So a sliding scale -- it might be said -- changes the sample rate of the audit?????

SO that the audit will be more efficient-- ? Efficacious.

[Bill Bored]

22. Look at Kathy's spreadsheet

And plug in some numbers.

Sliding scale means it's truly random, but the percentage needed to detect fraud is larger in smaller jurisdictions because in smaller jurisdictions, a single corrupted machine could be enough to change the outcome. It's harder to find one corrupted machine (in smaller jurisdictions) than to find one of several corrupted machines (in larger jurisdictions).

E.g., Randomly audit 10% of machines in 100-machine jurisdictions;
Randomly audit 3% of machines in 1,000-machine jurisdictions.

That's what I mean by sliding scale. But it's still random within each jurisdiction.

[MadeinOhio]

21. Kathy, thanks for those two excellent points.

What would you suggest would make a proper, uniform auditing procedure for counting ballots in Ohio?

There is alot of conflicts of interest there, with certain people taking bribes and demos or repubs also being paid off. So what kind of process would be mandatory to restore honesty and the fairness doctrine?

:hi:

[Bill Bored]

23. Welcome to DU MadeinOH!

The first thing you need to do is put your recount procedures, and everything else that's not codified, into state law. If I'm not mistaken, I believe the 3% random hand recount requirement, which is precincts totaling at least 3% of the vote in each county, is only a regulation issued by the Sec. of State -- not a law. If that's not true, and it is actually law, then Blackwell should be prosecuted for violating that law. If it's not law, it should become law, or a better procedure should become law.

That's your manual recount rule. Not sure about random audits. There may not be any!

Also, see this site if you're not already familiar with it:

http://www.reformohionow.org/

Peace.

[Bill Bored]

7. Here's a random audit simulator:

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

1. Go the page above. Scroll down to the game.
2. Change N to = 51 which simulates a small county with 51 machines.
3. Change Repetitions to = 1.
4. Change k to = 2, which simulates almost a 4% random audit.
5. Hit the Clear button.
6. Select the machines you wish to corrupt by clicking on the orange squares. They will turn blue as you corrupt the vote count on each one. Try a small number such as 2 or 3 which is about 4-6% of the machines.
7. Click Play repeatedly to see how often you will actually find one of the corrupted machines on the first try (blue squares turn pink when detected). Almost never!

This shows the futility of a 4% random audit in a small jurisdiction. Those 3 corrupted machines could be enough to swing a close elections.

Now, if you had more machines, you'd have to corrupt more and you'd have a greater probability of finding one with a 4% audit.

The problem is this: We need a LAW to say when you have to keep auditing when the first audit is clean, as can happen very frequently in a small jurisdiction. Fortunately there are larger ones and this is where most of the votes are cast. It's easier to find corrupted machines there.

But this gives you an idea of the problem. The spreadsheet on Kathy's site can give you more ideas.

Bottom line: we need recount/auditing laws and we need candidates who are willing to use them.

[WillYourVoteBCounted]

16. Voters Unite article on audits, how much is enough, video

Voters Unite took a look at this, and I recommend the video they have
demonstrating audits using marbles (I think it was).
Very eye opening.

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

Excerpt of write-up:

"August 2002. Clay County, Kansas. The tabulation machine showed that one candidate for commissioner had won, but a hand recount showed that his opponent had won by a landslide. In one ward, the computer had mistakenly reversed the totals.


Note that the reversal occurred in one ward. Hand counting every other ward could not have revealed the vote-reversal in that ward."...



"So, in a county with 640 different ballot styles, if 5% of those ballot styles have tabulation programming errors, there is a very real possibility that the outcome could be affected. If 2% of the ballot styles are audited, there is only a 49% chance that ONE of the 32 problems will be found.

That's less than a 50/50 chance that ONE of the badly programmed ballot styles will be picked for the audit! Only a 13% chance that TWO of the bad styles will be picked. And even if they are picked, that leaves 30 bad ballot styles completely undetected by the audit. Programming errors in 30 ballot styles could reverse the outcome of an election."



However, any audit is better than none, and I consider it a starting point if nothing else.

Things are really bad in some states, and it is a real accomplishment to have a ballot to audit.