ISF Logo   IS Forum
Forum Index Register Members List Events Mark Forums Read Help

Go Back   International Skeptics Forum » General Topics » Science, Mathematics, Medicine, and Technology
 


Welcome to the International Skeptics Forum, where we discuss skepticism, critical thinking, the paranormal and science in a friendly but lively way. You are currently viewing the forum as a guest, which means you are missing out on discussing matters that are of interest to you. Please consider registering so you can gain full use of the forum features and interact with other Members. Registration is simple, fast and free! Click here to register today.
Reply
Old 17th July 2017, 03:05 PM   #1
Kestrel
Philosopher
 
Kestrel's Avatar
 
Join Date: Dec 2007
Location: Colorado
Posts: 5,766
Risk limiting election audits?

Came across this rather outlandish claim in a Politico article Colorado launches election safeguards:
Quote:
Risk-limiting audits are cheaper and more efficient than older audit methods because they use statistical processes to select and analyze a smaller sample of ballots than was previously necessary. Stephanie Singer, the project lead at Free & Fair, told Eric that Colorado would only have to check 142 ballots out of the 2.85 million cast statewide in 2016 to determine whether the results were correct, compared to the 32,000 ballots required under a normal audit.
(Highlight added).

Anyone care to explain how this could possibly work?
Kestrel is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 17th July 2017, 04:55 PM   #2
theprestige
Penultimate Amazing
 
Join Date: Aug 2007
Posts: 21,521
Lies, damned lies, and statistics.

This signature is intended to irradiate people.
theprestige is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 17th July 2017, 09:29 PM   #3
shemp
a flimsy character...perfidious and despised
 
shemp's Avatar
 
Join Date: Nov 2002
Location: People's Democratic Republic of Planet X
Posts: 24,580
Why don't we just limit elections to the first 142 voters?
__________________
General Secretary of the Central Committee of the People's Democratic Republic of Planet X

"Shemp, you are the one fixed point in an ever-changing universe." - Beady

"I don't want to live in a world without shemp." - Quarky
shemp is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 17th July 2017, 10:58 PM   #4
GodMark2
Graduate Poster
 
GodMark2's Avatar
 
Join Date: Oct 2005
Location: Oregon, USA
Posts: 1,935
Originally Posted by Kestrel View Post
Came across this rather outlandish claim in a Politico article Colorado launches election safeguards:


(Highlight added).

Anyone care to explain how this could possibly work?
Step 1) Computers count all the ballots, and determine the margin of victory.
Step 2) Depending on the margin of victory, determine the number of samples that will give at least that much 'statistical accuracy'.
Step 3) Count that many randomly selected ballots by hand.
Step 4) compare results. If they match 'well enough', consider the original computer count 'accurate enough'.

The magic in step 3 is that the number of ballots needed for a given 'statistical accuracy' does not depend on the original population size, it's just a constant for that desired accuracy. Also, for races with larger margins of victory, you don't need as much certainly in the count itself to at least know who won. For close races, you will need to manually count more votes to be sure. In a Florida 2000 re-enactment, you'd probably have to count every vote manually, multiple times.

So, in 'normal' races, it lets you count fewer 'verification' votes.
__________________
Knowing that we do not know, it does not necessarily follow that we can not know.

Last edited by GodMark2; 17th July 2017 at 11:00 PM.
GodMark2 is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 17th July 2017, 11:27 PM   #5
McHrozni
Philosopher
 
Join Date: Jun 2009
Posts: 9,956
Originally Posted by GodMark2 View Post
Step 1) Computers count all the ballots, and determine the margin of victory.
Step 2) Depending on the margin of victory, determine the number of samples that will give at least that much 'statistical accuracy'.
Step 3) Count that many randomly selected ballots by hand.
Step 4) compare results. If they match 'well enough', consider the original computer count 'accurate enough'.

The magic in step 3 is that the number of ballots needed for a given 'statistical accuracy' does not depend on the original population size, it's just a constant for that desired accuracy. Also, for races with larger margins of victory, you don't need as much certainly in the count itself to at least know who won. For close races, you will need to manually count more votes to be sure. In a Florida 2000 re-enactment, you'd probably have to count every vote manually, multiple times.

So, in 'normal' races, it lets you count fewer 'verification' votes.
This. Dumbed down version would be that if the election went 70-30, you don't need to know if it was really 75-25 or 65-35 to tell who was the winner.

McHrozni
__________________
لا إله إلا رجل والعلوم والتكنولوجيا وأنبيائه
McHrozni is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th July 2017, 10:36 PM   #6
Kestrel
Philosopher
 
Kestrel's Avatar
 
Join Date: Dec 2007
Location: Colorado
Posts: 5,766
Originally Posted by GodMark2 View Post
Step 1) Computers count all the ballots, and determine the margin of victory.
Step 2) Depending on the margin of victory, determine the number of samples that will give at least that much 'statistical accuracy'.
Step 3) Count that many randomly selected ballots by hand.
Step 4) compare results. If they match 'well enough', consider the original computer count 'accurate enough'.

The magic in step 3 is that the number of ballots needed for a given 'statistical accuracy' does not depend on the original population size, it's just a constant for that desired accuracy. Also, for races with larger margins of victory, you don't need as much certainly in the count itself to at least know who won. For close races, you will need to manually count more votes to be sure. In a Florida 2000 re-enactment, you'd probably have to count every vote manually, multiple times.

So, in 'normal' races, it lets you count fewer 'verification' votes.
In other words this form of audit only proves that the count was accurate enough to assure the correct person won. It will not detect systematic errors or tampering that miscounts only a few percent of the votes when the election is a landslide.

Knowing that a poll with a 3% margin of error of error requires over 1,000 samples the 142 figure didn't make sense.
Kestrel is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th July 2017, 11:29 PM   #7
Brainster
Penultimate Amazing
 
Brainster's Avatar
 
Join Date: May 2006
Posts: 14,123
Originally Posted by Kestrel View Post
In other words this form of audit only proves that the count was accurate enough to assure the correct person won. It will not detect systematic errors or tampering that miscounts only a few percent of the votes when the election is a landslide.

Knowing that a poll with a 3% margin of error of error requires over 1,000 samples the 142 figure didn't make sense.
I think what they are doing is checking 142 ballots for what they show visibly versus what the voting machine said they show. If the ballots are chosen to represent a reasonable cross-section of the state, and if they all check out 100%, then the likelihood that the reported vote is in error becomes vanishingly small. On the other hand, if even one or two ballots turn out to have been read by the machine in error, then it would be a red flag that something might be wrong, and that quite a few more ballots should be checked.

Note that this does not mean that there's no need for a recount in a close election, just that the machines seem to be counting fairly.
__________________
My new blog: Recent Reads.
1960s Comic Book Nostalgia
Visit the Screw Loose Change blog.

Last edited by Brainster; 19th July 2017 at 11:30 PM.
Brainster is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th July 2017, 12:36 AM   #8
Planigale
Master Poster
 
Planigale's Avatar
 
Join Date: Jul 2013
Posts: 2,234
Originally Posted by Kestrel View Post
Came across this rather outlandish claim in a Politico article Colorado launches election safeguards:


(Highlight added).

Anyone care to explain how this could possibly work?
So in any sampling you need to ask how big an 'error' do I want to detect and with what certainty. Say that you have two candidates, and the winner had a 100,000 majority, one needs to detect a difference of 50,001 votes for it to make a difference. This will require a much smaller sample than if the majority was only 100 and you need to detect an error in only 51 votes. Then there is the certainty of detecting an error of the pre-defined size. Someone may have taken the view that the legal term 'on the balance of the probabilities' means that a 50% certainty is sufficient - this would explain a small sample size particularly if a large difference is looked for. To calculate this one would need to know the total number of votes cast and the votes cast for the candidates.

If you adopted a Bayesina approach an even smaller sample could be taken. E.g. this is a republican voting area, the opinion polls showed that republicans were in the lead so the prior probability that democrats won is low. So the number of ballots to be checked to provide a 50% certainty that there was not sufficient fraud for the democrats to have actually won would be low.
Planigale is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th July 2017, 01:21 AM   #9
Kestrel
Philosopher
 
Kestrel's Avatar
 
Join Date: Dec 2007
Location: Colorado
Posts: 5,766
Originally Posted by Brainster View Post
I think what they are doing is checking 142 ballots for what they show visibly versus what the voting machine said they show. If the ballots are chosen to represent a reasonable cross-section of the state, and if they all check out 100%, then the likelihood that the reported vote is in error becomes vanishingly small. On the other hand, if even one or two ballots turn out to have been read by the machine in error, then it would be a red flag that something might be wrong, and that quite a few more ballots should be checked.

Note that this does not mean that there's no need for a recount in a close election, just that the machines seem to be counting fairly.
Some voting machines print out a record of the votes cast. But they don't keep a copy of that internally. The voting machines just accumulate the votes for each candidate or issue. The same is true for optical scanners used to count paper ballots.

What can be done is to hand count all the ballots for a precinct and compare the results to the figures compiled after the polls closed.
Kestrel is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th July 2017, 01:38 AM   #10
quadraginta
Becoming Beth
 
quadraginta's Avatar
 
Join Date: May 2009
Location: Central Vale of Humility
Posts: 17,936
Originally Posted by Kestrel View Post
Some voting machines print out a record of the votes cast. But they don't keep a copy of that internally. The voting machines just accumulate the votes for each candidate or issue. The same is true for optical scanners used to count paper ballots.

What can be done is to hand count all the ballots for a precinct and compare the results to the figures compiled after the polls closed.

Which would verify the results for that precinct.

Extending beyond that requires the assumption that all the other precincts have equally dependable machines and precinct staff.

Speaking with North Carolina's last election as a basis for comparison, I don't know that this can be confidently assumed.
__________________
"It never does just what I want, but only what I tell it."
quadraginta is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th July 2017, 10:53 AM   #11
Kestrel
Philosopher
 
Kestrel's Avatar
 
Join Date: Dec 2007
Location: Colorado
Posts: 5,766
Originally Posted by quadraginta View Post
Which would verify the results for that precinct.

Extending beyond that requires the assumption that all the other precincts have equally dependable machines and precinct staff.

Speaking with North Carolina's last election as a basis for comparison, I don't know that this can be confidently assumed.
A precinct should be audited when there are indictions of problems. For example the number of ballots counted doesn't match the number of voters that signed the poll book. Or a high number of ballots that did not register a vote for any of the candidates at the top of the ticket.

It is also good practice to audit some precincts selected at random. This gives you some chance of detecting systematic errors or tampering that does not trigger any of the normal warning signals.
Kestrel is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th July 2017, 03:21 PM   #12
turingtest
Mistral, mistral wind...
 
turingtest's Avatar
 
Join Date: Aug 2013
Posts: 3,629
Originally Posted by shemp View Post
Why don't we just limit elections to the first 142 voters?
Or just one?
__________________
I'm tired of the bombs, tired of the bullets, tired of the crazies on TV;
I'm the aviator, a dream's a dream whatever it seems
Deep Purple- "The Aviator"

Life was a short shelf that came with bookends- Stephen King
turingtest is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th July 2017, 05:27 PM   #13
Brainster
Penultimate Amazing
 
Brainster's Avatar
 
Join Date: May 2006
Posts: 14,123
Originally Posted by Kestrel View Post
Some voting machines print out a record of the votes cast. But they don't keep a copy of that internally. The voting machines just accumulate the votes for each candidate or issue. The same is true for optical scanners used to count paper ballots.

What can be done is to hand count all the ballots for a precinct and compare the results to the figures compiled after the polls closed.
We have the optical scanners here, but the point is that you don't need to do the hand count. Let's assume that immediately after each election, all the machines are returned to a central location (as would seem a rudimentary security procedure). Then you take 142 ballots, trying to be sure that they are reasonably distributed throughout the state, and examine them by hand, record the votes, then you run them through each machine. If all tallies agree, you can be sure (to a reasonably high degree) that the votes were correctly recorded.

ETA: And if you find a problem with one particular machine, no big deal--just run all the ballots that were originally tabulated on it through a couple of the others until you get a good count.
__________________
My new blog: Recent Reads.
1960s Comic Book Nostalgia
Visit the Screw Loose Change blog.

Last edited by Brainster; 20th July 2017 at 06:13 PM.
Brainster is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th July 2017, 06:17 PM   #14
DevilsAdvocate
Illuminator
 
DevilsAdvocate's Avatar
 
Join Date: Nov 2004
Posts: 4,605
This doesn’t seem right, but I haven’t worked out the math. With these low ballot checks they talk about something like a 5% risk. But they aren’t talking about a 5% risk that the vote count is wrong, but a 5% risk that the vote count is so wrong that the wrong person was declared the winner. Wouldn’t that mean that statistically there could have been the wrong winner declared in 5% of the votes? With 435 members of the U.S. House that would mean there could be a couple dozen people sitting in Congress who didn’t actually win the election. Even with only a 0.1% risk that could mean 1 wrong election out of 1000. That’s only a couple years of House votes.

One way to put it would be that the elections were verified within the very, very tiny risk of only 0.1% risk. Another way to put it would be that there is almost a 50% chance that someone sitting in the House did not actually win the election due to a machine counting error.

Am I missing something?
__________________
Heaven forbid someone reads these words and claims to be adversely affected by them, thus ensuring a barrage of lawsuits filed under the guise of protecting the unknowing victims who were stupid enough to read this and believe it! - Kevin Trudeau
DevilsAdvocate is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Reply

International Skeptics Forum » General Topics » Science, Mathematics, Medicine, and Technology

Bookmarks

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


All times are GMT -7. The time now is 04:03 PM.
Powered by vBulletin. Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.
2014, TribeTech AB. All Rights Reserved.
This forum began as part of the James Randi Education Foundation (JREF). However, the forum now exists as
an independent entity with no affiliation with or endorsement by the JREF, including the section in reference to "JREF" topics.

Disclaimer: Messages posted in the Forum are solely the opinion of their authors.