Blogged by Brad Friedman from somewhere in Texas…
“Publicly observable post-election audits are the single most important safeguard we can have for the integrity of elections in this era of computer-assisted voting,” according to Livermore National Labs computer scientist David Jefferson.
“They allow everyone, winners and losers alike, to be satisfied that the races are correctly called, but without the need to trust any computers or software,” he added in a press release from the California Sec. of State’s office released just moments ago (posted in full at the bottom of this item) on the study he led for the SoS Debra Bowen, which examined the effectiveness of — and made recommendations to improve — the state’s 40-year old 1% manual audit law.
In addition to her stunning announcement on Friday that her team of independent analysts at Univeristy of California, attempting to hack the states electronic voting systems as part of her promised “Top-to-Bottom Review,” were able “to bypass both physical and software security measures in every system tested,” Bowen has also convened a “Post-Election Audit Standards Working Group” in order to “examine whether California’s post-election audit standards should be strengthened.”
Their report has now been released online [PDF].
Among recommendations made by the group (emphasis in original):
- Take a risk-based approach to conducting post-election audits by manually counting a higher percentage of precincts – above the 1% required by law – in close races and in races involving only a small number of precincts.
- Develop an adjustable sample model that fits California’s needs. With an adjustable audit, the size of the random sample for close elections and small races is not a flat percentage, such as California’s current 1%, but is calculated using the margin of victory, number of precincts in the race and other key factors to produce a desired confidence level (for example, 99%) that the winner of the election has been correctly declared.
- Develop a comprehensive approach to verifying election results, including rules for escalating an audit when errors in the machine counts are discovered during the manual count and rules for determining whether to trust the outcome of the election when small discrepancies are found between the manual and machine counts. In California, the law sets no standards when it comes to audit escalation.
“No matter what voting systems California counties use,” Bowen said in the statement, “we have to make sure we’re doing meaningful audits of election results to provide voters with the confidence that every vote is counted as it was cast.”
The complete press statement on the just-released finds of the “Post-Election Audit Standards Working Group” follows in full below…
Contact: Nicole Winger (916) ###-####
SACRAMENTO – In her continuing effort to ensure the security, accuracy, reliability and accessibility of California voting systems and the elections in which they are used, Secretary of State Debra Bowen today received a final report from the Post-Election Audit Standards Working Group, a group created in June to examine whether California’s post-election audit standards should be strengthened.
“California’s 1% audit law is 40 years old, and the charge of this Working Group was to take a fresh look at whether there’s a way to improve the auditing process to increase the chances of catching any errors and improving the public’s confidence in the election results,†said Secretary Bowen, the state’s chief elections officer.
The Post-Election Audit Standards Working Group spent an intensive four weeks reviewing a variety of studies, examining post-election audit models and talking to national experts in the field of election auditing. The Working Group also held a public forum on July 2.
“This report is an initial step toward what we expect to be an extensive study of election audits over the next few months, as part of Secretary Bowen’s comprehensive strategy for securing California elections,†said David Jefferson, a computer scientist with Lawrence Livermore National Laboratory and chairperson of the Working Group. “Publicly observable post-election audits are the single most important safeguard we can have for the integrity of elections in this era of computer-assisted voting. They allow everyone, winners and losers alike, to be satisfied that the races are correctly called, but without the need to trust any computers or software.â€
The Working Group’s report is at http://www.sos.ca.gov/elections/peas/final_peaswg_report.pdf. Highlights of the report include recommendations to:
- Take a risk-based approach to conducting post-election audits by manually counting a higher percentage of precincts – above the 1% required by law – in close races and in races involving only a small number of precincts.
- Develop an adjustable sample model that fits California’s needs. With an adjustable audit, the size of the random sample for close elections and small races is not a flat percentage, such as California’s current 1%, but is calculated using the margin of victory, number of precincts in the race and other key factors to produce a desired confidence level (for example, 99%) that the winner of the election has been correctly declared.
- Develop a comprehensive approach to verifying election results, including rules for escalating an audit when errors in the machine counts are discovered during the manual count and rules for determining whether to trust the outcome of the election when small discrepancies are found between the manual and machine counts. In California, the law sets no standards when it comes to audit escalation.
“How confident should we be that elections name the right winners? This difficult question provides a wonderful opportunity for statisticians and other scientists to help safeguard our democratic process,†said Philip Stark, professor of statistics at the University of California Berkeley and a Working Group member.
The Working Group also suggested several short-term options for improving California’s existing manual count and increasing the chance of detecting error:
- Increase the sample size of the manual count for close races.
- Conduct additional manual count audits of high-risk voting systems.
- Increase the manual count sample size in small counties.
- Consider “targeted audits†where candidates select additional precincts for the manual count in close races.
- Escalate the manual count to investigate discrepancies.
- Select precincts using a transparent method for generating random samples, such as rolling a 10-sided dice.
- Establish uniform procedures for conducting the manual count, including requiring tracking of undervotes, overvotes and cancelled Direct Recording Electronic (DRE) voting machine ballots.
- Increase transparency of the manual count.
- Educate voters who use DRE voting machines about the importance of carefully checking the Voter-Verified Paper Audit Trail for accuracy.
“This working group was the first of its kind convened by a Secretary of State. Fifteen states conduct post-election audits, and the number is growing as more people recognize the crucial need to publicly verify election results,†said Kim Alexander, president and founder of the California Voter Foundation and a Working Group member. “Improving the post-election auditing process will benefit California voters while also providing other states with ideas and examples of how they can strengthen their own processes.â€
“The ideas of experts in the fields of statistics and financial auditing have raised the discussion of the post-election manual tally to a higher level. The challenge for elections officials will be to implement additional auditing requirements within the limits of the 28-day canvass period after an election,†said Elaine Ginnold, Marin County Registrar of Voters.
“No matter what voting systems California counties use, we have to make sure we’re doing meaningful audits of election results to provide voters with the confidence that every vote is counted as it was cast,†Secretary Bowen concluded.
Bowen tapped experts in the fields of computer science, financial auditing, statistical analysis, election reform advocacy, and city and county government to form the Post-Election Audit Standards Working Group, including:
- David Jefferson (Chairperson)
Computer Scientist with the Center for Applied Scientific Computing, Lawrence Livermore National Laboratory - Kim Alexander
President and Founder, California Voter Foundation - Elaine Ginnold
Registrar of Voters, Marin County - Amy Lehmkuhl
Certified Public Accountant, Ueltzen & Company, LLP - Kathleen Midstokke
City Clerk, City of Downey - Philip Stark
Professor of Statistics, University of California, Berkeley
In the age of computer fraud we are now in gearing audit size to “victory margin” doesn’t seem like a very good idea. “Victory margin” in a fraudulent election is the difference between the fraud used to alter the result and fraud needed to change the result. The “victory margin” has little if any correlation to how equally divided the voters were on the issue at hand.
Lest we forget another best friend of election fraud detection: Exit Poll Science
Exit poll science provides tools that have been used for decades successfully. Accurate predictions are the legacy of these tools.
The 2000 and 2004 official election results were at odds with the exit poll predictions.
Those doing the polls, Edison/Mitofsky, were of the popular mindset “it can’t happen here“. This mindset is not prepared to entertain any notion of election fraud as the reason for the massive first time discrepancies.
They instead offered the explanation that “republicans were more shy than democrats” and that is why the massive discrepancy. They did not even consider election fraud.
A study done by 8 Phd’s and one MS pointed this out and that election fraud must not be ruled out. They said:
(Phd’s Speak Out, emphasis added). Other experts have said: “Flaws in any of these aspects of a voting system, however, can lead to indecisive or incorrect election results“.
Edison/Mitofsky and the MSM, who used the data, can’t bear to even consider that America could be corrupted in its election processes. A position entirely at odds with common sense and official scientific inquiry.
{Ed note: This is the second time I’ve asked you not to just paste stuff into the comments section here. If you want to draw our attention to something, say so, and provide a LINK to it. –99}
Bob Young #1–exactly! Including the possibility of insider fraud in the statistical model means that the only way to produce a “confidence interval of 99%” is to audit (hand count) 99% of the ballots.
The possibility of fraud, big or small, by insiders or outsiders is quite well covered by the post-election audit process we outlined in the report.
The preliminary results should be public, before the selection of precincts for audit. The random number drawing should be public. The random precincts selected for hand counting should be hand counted in public, and blindly, i.e. without knowledge of the “target” preliminary count. All discrepancies that can be resolved should be documented and resolved in public. Discrepancies that are real, meaning the machine miscounted the ballot, should be documented (and the results adjusted) in public. The calculation of the statistical confidence, indicating whether or not to accept the election results and stop the audit can be calculated by anyone. If the confidence is not high enough the audit should continue with more random precincts audited.
Because all of this is done in public, and observable, the opportunities for insider fraud are drastically reduced.
And the dependence of the sample size on the initial margin of victory (and the other parameters) is exactly what is appropriate mathematically. Failure to do that makes any audit either inadequate or inefficient.
Please don’t make a snap judgment about the auditing process we endorse. Please read the report, and also the documents we cite.
Thanks David!
Sound like you have a much better set up than most I have seen but I still think the size of the audit is much better geared to the results the audit is finding than to how close any fradulent electronic “count” might have been.
David #5
I looked over the report and it is well done and covers the bases IMO. I can’t quite agree with Semperfi #4. The science is more mature than that.
And I don’t think Bob #1 disagrees, because your report does not limit it entirely to any one factor.
It takes several factors to develop an applicable comprehensive sampling theory and apply it to any particular circumstance. One size fits all is a less than skilled approach.
Your inclusion of Philip Stark, a statistician, and Amy Lehmkuhl, a C.P.A. adds comprehensiveness to the equation IMO.
My understanding at this time, after reading some other works too, is that like exit polling, the key is to see the proper statistical sampling pattern to be used in an audit.
And that, as your report details, is a function of the particular context at the particular time. In other words, the “situation on the ground” 🙂 .
Such a well done pattern in an audit, like in an exit poll, can detect irregularities even with what Semperfi would consider to be a small percentage of sampling.
One commentor says that testing a very small sample of votes by way of the voter verification of their paper ballot would virtually guarantee that the precinct has a valid or an invalid tally:
(Adler 2003). Jones cites Adler.
Anyway, it is refresing to see a competent, professional, and non-partisan approach as you have done in the report.
Congrats to all in the whole effort. This is something other states can use, and so can los federales 🙂 .
My short comment above was not meant to be taken mathematically, but figuratively. The point is DRE voting machines do not produce a ballot in the traditional sense, no matter what Holt writes in his bill. VVPT can not be demonstrated to be the same as the votes that are counted. Audits or recounts of such do not provide any “confidence” to me.