The problem with recount provision of Voter Confidence Act HR 2239

[Eric J in MN]

HR 2239, the Voter Confidence Act which requires paper ballots for touchscreen machines and is currently in a House commitee, has a mandatory recount provision, section 7.

However section 7 says that .5% of disctricts will be subject to recount without saying that all the ballots even in those districts must be recounted.

It would be better to require 1% of the ballots in EVERY DISTRICT be recounted, after random sorting. Then we'd have real confirmation the touchscreen machines are accurate.

Can people please contact Rush Holt's office about this issue of how to do the best recount?

Here is Section 7 of HR 2239:
---------------------------------------------SEC. 7. REQUIREMENT FOR MANDATORY RECOUNTS.

The Election Assistance Commission shall conduct manual mandatory surprise recounts of the voter-verified records of each election for Federal office (and, at the option of the State or jurisdiction involved, of elections for State and local office) in .5 percent of the jurisdictions in each State and .5 percent of the overseas jurisdictions in which voter-verified records are preserved in accordance with this section immediately following each general election for Federal office, and shall promptly publish the results of those recounts. The treatment of the results of the recount shall be governed by applicable Federal, State, or local law, except that any individual who is a citizen of the jurisdiction involved may file an appeal with the Commission if the individual believes that such law does not provide a fair remedy.

[ramblin_dave]

1. Try posting your concerns here...

under "General Discussion & Politics"

http://www.blackboxvoting.org/blackboxvotingcgi/dcforum/dcboard.cgi

[BevHarris]

4. Well the EFF folks like to pretend we don't exist.

They have an agenda at the moment.

Of course, the audit provision is inadequate, but the issues with it aren't solved at all with the random spot check (halloo, someone call a forensic accountant, not a computer guy who knows nothing about how to audit in fraud-prone activities like banking and gambling, and elections). Computer scientists use math and statistics to give advice on how to audit. That doesn't do it at all.

Now, there are ways to audit properly. There are three ways, and here they are:

1) Do all paper, hand-counted

2) Bar-code a human-readable ballot, use a GENERIC bar code scanner purchased off the shelf by a manufacturer who does this for canned beans and bars of soap, not votes. Must have a readout so you can see that it is reading what your eyes show you is on the ballot. At end of day, while still at polling place, hand-scan 100% and match against machine.

3) Do spot checks, with all of these additions:
- Parties get to choose a small number of discretionary precincts for a 100% hand audit.
- Triggered 100% manual audit for insufficient randomness (ie. 18181 three times, or other anomaly)
- Triggered 100% manual audit for any area where machine tally doesn't match spot check.
- Allow the media to do a 100% audit if they pay for it
- Allow any citizen to do a 100% audit if they pay for it

The problem with doing spot checks properly, being that this is politics, is that they'll try all kinds of insincere shenanigans which will thoroughly muck up the works, and then they'll point to how mucked up it got and say we should not even do it. That's why I favor one of the first two options instead. Both are simple and very robust.

Bev Harris
http://www.blackboxvoting.org

[Eric J in MN]

5. I don't understand your objection

If 100,000 votes are cast at a precinct, and 1,000 ballots are randomly sorted and recounted, why wouldn't that show accuracy?

[BevHarris]

6. Well let's get up to speed on simple things first

A large precinct has perhaps 3000 votes. Not 100,000.

We need to be able to do a 100% audit of any suspicious area. That is how audits are done, my friend. In accounting, and that's what this is, not math, not computer science, but accounting -- just imagine this scenario, and you'll see how silly it is:

Your business is being audited by the IRS. They are allowed to randomly sort and look at only 1 percent of the receipts. (That is not the percentage used for any fraud-prone audit that I know of anyway). But say they do this. So there they are, looking at their one percent, and they see total disbursements like this:
Phone bill: $18,181
Utilities: $18,181
Advertising: $18,181

But after all, they've looked at their one percent, so they don't get to look at 100% of your phone, utility and ad expenses to see how you are cheating.

Part of the problem here is that any spot check audit -- if it's real, like they are with slot machines and money -- means the auditor gets to check any damn thing they want, whenever they want, even just on a hunch.

In politics, you'll have six party hacks standing around saying "we've audited and audited and then re-audited" (sound familiar). And they won't allow additional audits on anomalies and yes, hunches.

This is a human problem, a fraud problem, not a math problem. I continue to be frustrated, as we have expressed here in the past, at the arrogance (not personal to you, Eric J) but the arrogance of computer scientists who just step and and offer "expert" opinions on fraud prevention and then give out a math model for how to audit.

Auditing properly requires human intervention for the all-too-human situation when someone figures out a new way to game the situation. You can't legislate that or put it in a formula, so your auditing has to give leeway to use judgement. Problem is, in politics that will get abused, so a quick 100% scan is better.

Either that or junk the machines and reduce the number of questions on the ballot.

Bev Harris
http://www.blackboxvoting.org

[Eric J in MN]

7. if discrepancies are discovered after looking at 1 percent

If discrepancies are discovered after looking at 1 percent, in my proposal, they could then audit in other ways.

Since you said there are only about 3,000 ballots per precinct, maybe 10% per precinct would make more sense than 1%.

[BevHarris]

8. I'll show you what's wrong with it.

Again, the problem is this: When you audit, you get to look at any damn thing you want, even if it's just a hunch. Any rule that only allows you to look IF will result in a cheater figuring out a way to game the system.

Talk to a banker.
Talk to a casino.
Or at least, talk to a bookkeeper.

Proper auditing is never premised on "you get to look at x% and that's all unless you spot an anomaly in your x%." Think of it this way: an audit is more like a search warrant than a pat-down at the airport.

In a pat-down, you have legal limits as to how much patting you can do, unless you discover a metallic bulge.

In a search warrant, you get to throw the clothes out of the drawers and, if you get a hunch, pull out the plumbing pipes and look in there too. You are expected to do a minimum amount, but if you decide to grab the fern out of a fish tank and have it analyzed, you have the right to do that too.

This is a human problem, not a math problem.

To refer to my earlier example:
$18,181 phone bill
$18,181 utility bill
$18,181 advertising bill

So you check 10% and don't find anything. Is that good enough? It is? Well then what about this:

$18,181 phone bill
$18,181 utility bill
$18,181 advertising bill
$18,181 rent
$18,181 salaries
$18,181 office supplies

If you check 10% on the above and don't find anything, is that good enough? Not in an accounting audit.

Bev

[Eric J in MN]

9. how could they cheat

If there are 3,000 ballots in a precinct, and recounting 300 shows the same proportions as for the 3,000, how can they be cheating?

I studied statistics in college.

[Eric J in MN]

10. I suppose that if

I suppose that if all the real ballots were swapped for fake ballots before the auditors arrived, then all the fake results would be proportional, but there should be poll watchers to guard against that.

[BevHarris]

11. This is not a statistics problem

How can they cheat? Any way they figure out. I notice that banks don't use statistics to audit, they use real people, eyes on the docs, hunches welcome.

Here's an obvious way to cheat: You have your statistical percentage checked over here. So I hack the central tabulator and put the cheat over there. Woops, 6,300 votes disappeared overnight after the polls closed and everyone went home? "What exactly happened I don't have the intelligence to say."

But votes disappeared! We want to recount! "Sorry, we checked our statistical percentage. Now you want to audit and re-audit and then audit again."

Just politics, folks.

Bev

[Eric J in MN]

12. Wouldn't the auditors be told the machine totals first?

Step 1:
The auditors are told the machine totals.
Step 2:
They randomly sort the paper ballots and then recount 10%.

As long as the paper ballots are the same ones the voters checked, the proportions would be the same if the machine total is accurate.

[Eric J in MN]

13. You make a good point that they would have to count total paper ballots

You make a good point that they would have to count the total number of paper ballots to make sure that matches the total votes in the machine first.

Otherwise, matching proportions would be meaningless.

[lysergik]

14. If you need to count the number of paper ballots..

why not just do an audit at that time.

Personally I think using the results from a DRE (a good DRE, open source, etc) could be used for projecting an election.

The paper ballots would then be put through an optical scan machine and counted. If the numbers were within a certain margin of error then the election could be certified. If any races were within a close margin of whatever percentage then a hand count of the paper ballots would be ordered.

But.. thats just me :)

[Eric J in MN]

15. if you have about 3000 paper ballots

If you have about 3000 paper ballots, counting the total number of ballots by hand wouldn't take long.

Counting votes for each candidate would take a lot longer.

[BevHarris]

16. Can't do optical scan machine. Needs to be generic bar code scan

optical scans are proprietary to the voting industry, not an adequate independent source. Need to use something generic and entirely separate from voting machine vendor.

Also, you can visually confirm a bar reader is reading ballot correctly. You cannot do that with an optical scan.

[creativelcro]

17. obviously,

a 100% recount is preferable, if feasible... Generally, many statistical methods were devised because testing the entire population of interest is not possible. If it is possible to test the entire population (sample size = entire population), then it should be done. It seems that, if the size of a precint is less than a few thousand votes, then a manual check ought to be possible. Say, 3000 votes, split into 3 groups of observers (1000 each). Each ballot takes ~ 10 sec (no hanging chads here, just reading out). So, it would take just under 3 hours...
-CV

[creativelcro]

2. what percentage do you need to recount ?

You cannot compare the paper ballots one by one with the corresponding record in the EVM. The pairing is not there, of course. What you can do is recount a random x% of the paper ballots in each district. Then, you need some stats to decide whether what you find in this sample is consistent with the overall result provided by the EVM for that district. The size of x is based on statistical theory. If you want to detect really small deviations between the paper ballots and the EVM result, then you need a relatively large x. If you do a total recount, then you can detect arbitrary deviations (obviously, since your x becomes the entire set of votes for that district). Experts in this field should decide on the size of x as a function of the discrepancy size that one wants to detect.
-CV

[Eric J in MN]

3. I suggest 1 percent of all ballots

I suggest 1 percent of all ballots, every precinct, be randomly sorted
and recounted.

If there is a significant discrepancy at that point in the proportions
at a particular precinct, then more should be recounted.