Hello Everyone,
It is a Good Day in our great democracy. :-)
We now have a new scientific method to ensure that election outcomes
in every race are honest and accurate!
The National Election Data Archive (NEDA)'s paper on vote count audits
has been updated and will be updated again soon:
http://electionarchive.org/ucvanalysis/us/paper-audits/paper_audits.pdf
and NEDA's Spreadsheet AuditCalculator.xls for
Determining Audit Percentages that would Detect Outcome-Altering Vote
Miscounts has been updated:
http://electionarchive.org/ucvanalysis/us/paper-audits/
----------------------------------
CHANGES TO AuditCalculator.xls:
1. Added an Estimator to help determine more quickly, the number of
vote counts to audit. This will assist users with the currently
trial-and-error process for determining the exact number of vote
counts that should be audited to be sure to detect any miscounts that
alter outcomes.
(I used the estimated function that the Brennan Center recommended but
I adjusted the calculation to take into account "Candidate Margins" so
that the audit will detect the amount of corruption that would alter
outcomes - not just some arbitrary amount of corruption. When I took
a closer look at the Brennan Center appendix G, I noticed that the
Brennan Center made errors in calculating just a couple of the
probabilities and that, contrary to my first glance, they used a
different function than the function that NEDA recommended for
calculating audits over a year ago. The Brennan Center method
conservatively over-estimates the number of counts that need to be
audited to detect an assumed rate of corruption, especially in small
counties with fewer total counts to audit. The reason the Brennan
Center used an estimation is that it provides an easier equation to
solve so that election officials can have an easy way to calculate the
number of machines to parallel test or audit, given the desired
probability for detecting an assumed rate of corruption. However, I
am sure that taxpayers and election officials alike will appreciate a
more exact calculation of audit percentages required to detect
outcome-altering vote miscounts.)
I've asked one of NEDA's most brilliant volunteer mathematicians,
Frank Stenger, to see if he can solve the exact probability equation
so that we can provide a more exact, easier method for independent
auditors to calculate the correct audit percentages that would detect
any outcome-altering vote miscounts, but the equation looks very
difficult, if not impossible, to solve exactly. We may need instead
to provide an (open source) computer program to do the calculations
more exactly and easily.
2. Corrected the maximum rate of vote switching that is used to
calculate the minimum number of counts that must be corrupted to alter
election outcomes. Now it matches the Brennan Center's threat
evaluation assumption of 15% (not the 30% I mistakenly thought
earlier). To be more conservative, people may want to raise this
constant from the 15% to something higher, or wait to see what future
data shows.
3. Fixed some of the spreadsheet formulas to avoid errors or nonsense
in a few special cases.
4. Added instructions for how to use the Audit_Calculator. xls
spreadsheet to obtain the exact number of vote counts to audit to
obtain any desired probability of detecting outcome-altering vote
miscounts.
----------------------------------------
LESSONS LEARNED:
If you play with the Audit_Calculator.xls spreadsheet you will notice:
1. Audit percentages must be higher when the margins between
candidates is smaller.
It is a Good Day in our great democracy. :-)
We now have a new scientific method to ensure that election outcomes
in every race are honest and accurate!
The National Election Data Archive (NEDA)'s paper on vote count audits
has been updated and will be updated again soon:
http://electionarchive.org/ucvanalysis/us/paper-audits/paper_audits.pdf
Determining Audit Percentages that would Detect Outcome-Altering Vote
Miscounts has been updated:
http://electionarchive.org/ucvanalysis/us/paper-audits/
----------------------------------
CHANGES TO AuditCalculator.xls:
1. Added an Estimator to help determine more quickly, the number of
vote counts to audit. This will assist users with the currently
trial-and-error process for determining the exact number of vote
counts that should be audited to be sure to detect any miscounts that
alter outcomes.
(I used the estimated function that the Brennan Center recommended but
I adjusted the calculation to take into account "Candidate Margins" so
that the audit will detect the amount of corruption that would alter
outcomes - not just some arbitrary amount of corruption. When I took
a closer look at the Brennan Center appendix G, I noticed that the
Brennan Center made errors in calculating just a couple of the
probabilities and that, contrary to my first glance, they used a
different function than the function that NEDA recommended for
calculating audits over a year ago. The Brennan Center method
conservatively over-estimates the number of counts that need to be
audited to detect an assumed rate of corruption, especially in small
counties with fewer total counts to audit. The reason the Brennan
Center used an estimation is that it provides an easier equation to
solve so that election officials can have an easy way to calculate the
number of machines to parallel test or audit, given the desired
probability for detecting an assumed rate of corruption. However, I
am sure that taxpayers and election officials alike will appreciate a
more exact calculation of audit percentages required to detect
outcome-altering vote miscounts.)
I've asked one of NEDA's most brilliant volunteer mathematicians,
Frank Stenger, to see if he can solve the exact probability equation
so that we can provide a more exact, easier method for independent
auditors to calculate the correct audit percentages that would detect
any outcome-altering vote miscounts, but the equation looks very
difficult, if not impossible, to solve exactly. We may need instead
to provide an (open source) computer program to do the calculations
more exactly and easily.
2. Corrected the maximum rate of vote switching that is used to
calculate the minimum number of counts that must be corrupted to alter
election outcomes. Now it matches the Brennan Center's threat
evaluation assumption of 15% (not the 30% I mistakenly thought
earlier). To be more conservative, people may want to raise this
constant from the 15% to something higher, or wait to see what future
data shows.
3. Fixed some of the spreadsheet formulas to avoid errors or nonsense
in a few special cases.
4. Added instructions for how to use the Audit_Calculator. xls
spreadsheet to obtain the exact number of vote counts to audit to
obtain any desired probability of detecting outcome-altering vote
miscounts.
----------------------------------------
LESSONS LEARNED:
If you play with the Audit_Calculator.xls spreadsheet you will notice:
1. Audit percentages must be higher when the margins between
candidates is smaller.
(Note: You can view every article as one long page if you sign up as an Advocate Member, or higher).
