Category: Post-Election Audits

Late last week the University of Connecticut (UConn) VoTeR Center posted three reports from the November election on its web site <Pre-Election Memory Card Tests>, <Post-Election Memory Card Tests>, and <Post-Election Audit Report>.  In Part 1 we discussed the memory card tests and in Part 2 we discuss the Post-Election Audit Report.

Highlights from the official report:

The VoTeR Center’s initial review of audit reports prepared by the towns revealed a number of returns with unexplained differences between hand and machine counts and also revealed discrepancies in cases of cross-party endorsed candidates (i.e., candidates whose names appear twice on the ballot because they are endorsed by two parties). As a result, the SOTS Office performed additional information-gathering and investigation and, in some cases, conducted independent hand-counting of ballots. …Further information gathering was conducted by the SOTS Office to identify the cause of the moderately large discrepancies, and to identify the cause of discrepancies for cross-party endorsed candidates…

This report presents the results in three parts: (i) the analysis of the original audit records that did not involve cross-party endorsed candidates, (ii) the analysis of the audit records for cross-party endorsed candidates, and (iii) the analysis of the records that were revised based on the SOTS Office follow up. The analysis does not include 6 records (0.8%) that were found to be incomplete. ..

The main conclusion in this report is that for all cases where non-trivial discrepancies were originally reported, it was determined that hand counting errors or vote misallocation were the causes. No discrepancies in these cases were reported to be attributable to incorrect machine tabulation. For the original data where no follow up investigation was performed, the discrepancies were small; in particular, the average reported discrepancy is much lower than the number of the votes that were determined to be questionable.

Further on in the report is another conclusion:

The main conclusion of this analysis is that the hand counting remains an error prone activity. In order to enable a more precise analysis, it is recommended that the hand counting precision is substantially improved in future audits. The completeness of the audit reports also need to be addressed. For example, in two of the towns when the second hand count was performed it was determined that the auditors did not count a batch of 25 ballots in one case and the absentee ballots in the second. This initially resulted in apparently unexplained discrepancies. Submitting incomplete audit returns has little value for the auditing process.

We note the details of the investigations to determine the accuracy of human and machine counting includes some counting of ballots and some telephone conversations with election officials:

The first follow up was performed to address substantial number of discrepancies in some precincts (discrepancies over 30 votes). All those Version 1.3 April 20, 2010 UConn VoTeR Center 15 unusual discrepancies were concentrated in four towns. As a result in those towns a second hand count of the actual ballots was performed by the SOTS Office personnel…

We now discuss a batch of records containing 218 (28.1% of 776) records where originally the reported discrepancies were under 30 (these do not include cross-party endorsed candidates). In this case the SOTS Office personnel contacted each registrar of voters and questioned their hand count audit procedures. In all instances, the registrars of voters were able to attribute the discrepancies to hand counting errors. Thus no discrepancies (zero) are reported for these districts. Given the fact that no discrepancies were reported for those records we do not present a detailed analysis.

We have several concerns with these investigations:

1. All counting and review of ballots should be transparent and open to public observation.  Both this year and last year we have asked that such counting be open and publicly announced in advance.
2. Simply accepting the word of election officials that they counted inaccurately is hardly reliable, scientific, or likely to instill trust in the integrity of elections.  How do we know how accurate the machines are without a complete audit, any error or fraud would likely result in a count difference, and would be [or could have been] very likely dismissed.
3. Even if, in every cases officials are correct that they did not count accurately, it cannot be assumed that the associated machines counted accurately.
4. Simply ignoring the initial results in the analysis of the data provides a simple formula to cover-up, or not recognize error and fraud in the future.

As we have said before we do not question the integrity of any individual, yet closed counting of ballots leaves an opening for fraud and error to go undetected and defeats the purpose and integrity of the audit.

We also note that in several cases officials continued to fail perform the audit as required by law or to provide incomplete reports.

On the other hand we note that only 6 records (0.8% of 776) were found to be incomplete. The statistical analysis does not include these records. While some problematic records are clearly due to human error (e.g., errors in addition), in other cases it appears that auditors either did not follow the audit instructions precisely, or found the instructions to be unclear. However, this is a substantial improvement relative to the November 2007 and November 2008 elections, where we reported correspondingly 18% and 3.2% of the records that were unusable.On the other hand we note that only 6 records (0.8% of 776) were found to be incomplete. The statistical analysis does not include these records. While some problematic records are clearly due to human error (e.g., errors in addition), in other cases it appears that auditors either did not follow the audit instructions precisely, or found the instructions to be unclear. However, this is a substantial improvement relative to the November 2007 and November 2008 elections, where we reported correspondingly 18% and 3.2% of the records that were unusable.

Improvement or not, our solution would be to require the towns involved to, correct their errors, comply with the law, and perhaps be subject to a penalty.  Not pursuing such provides a clear formula for covering up errors and fraud.

Finally, since only “good” records were fully analyzed we question the value of some the reported statistics based only on those results. We do agree with the reports recommendations:

The main conclusion of this analysis is that the hand counting remains an error prone activity. In order to enable a more precise analysis, it is recommended that the hand counting precision is substantially improved in future audits. The completeness of the audit reports also need to be addressed. For example, in two of the towns when the second hand count was performed it was determined that the auditors did not count a batch of 25 ballots in one case and the absentee ballots in the second. This initially resulted in apparently unexplained discrepancies. Submitting incomplete audit returns has little value for the auditing process.

For the cross party endorsement, it is important for the auditors to perform hand counting of the votes that precisely documents for which party endorsement the votes were cast, and to note all cases where more than one bubble was marked for the same candidate. The auditors should be better trained to follow the correct process of hand count audit…

We also believe that our reporting of the analysis, and the analysis itself needs to be improved. A major change planned for future analysis is to assess the impact of the perceived discrepancies on the election outcomes (in addition to analyzing individual audit return records). This is going to be exceedingly important for the cases where a race may be very close, but where the difference between candidates is over 0.5% (thus not triggering an automatic recount)[*]

* CTVotersCount Note: Connecticut has an automatic ‘recanvass’, triggered at a difference of less than 20 votes or 0.5% up to  a maimum difference of 2000 votes.

In January, the Connecticut Citizen Election Audit Coalition Report analyzed the November 2009 Post-Election Audit data and the observations of citizen volunteers:

In this report, we conclude that the November post-election audits still do not inspire confidence because of the continued lack of

• standards for determining need for further investigation of discrepancies,
• detailed guidance for counting procedures, and
• consistency, reliability, and transparency in the conduct of the audit.
  

Compared with previous reports of November post-election audits:

• The bulk of our general observations and concerns remain.
• The accuracy of counting has improved. There was a significant reduction in the number of extreme discrepancies reported. However, there remains a need formuch more improvement.
• There was a significant improvement in counting cross-endorsed candidate votes
• The number of incomplete reports from municipalities has significantly decreased.

We find no reason to attribute all errors to either humans or machines.

There is no reason to modify the Coalition’s conclusion based on the official report. Many of the same concerns and conclusions we discussed last year still apply.  See last year’s post for more details, here is a summary:

• The investigations prove that Election Officials in many Connecticut municipalities are not yet able to count votes accurately
• The audit and the audit report are incomplete
• Even with all the investigations and adjustments we have many unexplained discrepancies [Unless we accept the belief of officials that they counted inaccurately, and in all those cases the machine counted accurately]
• The Chain-of-Custody is critical to credibility
• Either “questionable ballot” classification is inaccurate in many towns or we have a “system problem”
• Accuracy and the appearance of objectivity are important
• Timeliness is important
• The problem is not that there were machine problems. We have no evidence there were any. The problem is that when there are or ever were, dismissing all errors as human counting errors, we are unlikely to find a problem
• We stand by our recommendations and the recommendations of other groups
• The current Audit Process in Connecticut demonstrates the need for audits to be Independent and focused on election integrity, not just machine certification reliability

As we said last year;

We recognize and appreciate that everyone works hard on these programs, performing the audits, and creating these reports including the Registrars, Secretary of the State’s staff, and UConn.   We also welcome Secretary Bysiewicz’s commitment to solve the problems identified.  Yet, we have serious concerns with the credibility of the audits as conduced and their value, as conducted, to provide confidence to the public in the election process.

Late last week the University of Connecticut (UConn) VoTeR Center posted three reports from the November election on its web site <Pre-Election Memory Card Tests>, <Post-Election Memory Card Tests>, and <Post-Election Audit Report>.  In Part 1 we will discuss the memory card tests and in Part 2 the Post-Election Audit Report.

As we said last year:  We should all applaud the unique memory card testing program, yet we must also act aggressively to close the gaps it continues to expose.

We note the following from this year’s reports:

• An increase in the percentage of memory cards in the pre-election test.

[pre-election 2009]  The VoTeR Center received in total 491 memory cards from 481 districts before the elections. This document reports on the findings obtained during the audit. The 491 cards represent over 80.6% of all districts, thus the audit is broad enough to draw meaningful conclusions.

[pre-election 2008] the VoTeR Center received and examined 620 memory cards [about 74% of districts] as of November 3, 2008. These cards correspond to 620 distinct districts in Connecticut. About 2/3 of these memory cards were randomly chosen by the VoTeR Center personnel during the visits to LHS and before the cards were packed and shipped to the towns. Another 1/3 of the memory cards came from the towns directly, where the cards were randomly chosen for preelection audit (this procedure applied to the town for which the cards were not selected at LHS).

• And a significant drop in the percentage of memory cards in the post-election test:

[post-election 2009] The VoTeR Center received in total 120 memory cards from 49 districts [approximately 8.0% of all districts] after the elections. The cards were received during the period from December 12, 2009 to February 12, 2010. Among the received cards, 49 were used in the elections,

[post-election 2008] The VoTeR Center received in total 462 memory cards from a number of districts after the elections… Among these cards, 279 were used in the elections… The 279 cards represent over 30% of all districts,

As we understand it, the Secretary of the State’s Office asks all towns to send in memory cards for each district, they are not randomly selected.  This means that we cannot be sure the percentages of  “junk” data or procedural lapses reported actually represent a reliable measure of all memory cards and official actions, yet it seems reasonable to conclude that:

• “Junk” data continues at an unacceptable rate:

[pre-election]The audit identified forty two (42) cards, or 9%, that contained “junk” data; these cards are unreadable by the tabulators, and easily detected as such. This is a high percentage of faulty/unusable cards. We note that this is consistent with the percentage reported for the pre-election audit of November 2008 elections. The percentage is lower than detected in the post-election audit for the August 2008 primary (15%), but higher than detected in the pre-election audit for the August 2008 primary (5%), post-election audit for the February 2008

[post-election] Concerning the remaining cards, 14 (12% of the total number of cards) were found to contain junk data, that is, they were unreadable, which is easily detected by the tabulators; had a card contained junk data at the time of the election,

So the problem of “junk” data continues at a likely rate toward the middle of past testing results.  As we have said before 5%, 9%, 15%, even 1% is a huge failure rate for relatively simple technology such as memory cards.

• Very good news on the “Junk” data cards:

We have determined that weak batteries are the primary cause of junk data on cards; a separate report will document this in more detail. It is recommended that batteries are replaced before each election.

It seems that UConn has identified a likely cause of the “junk” data cards.  Perhaps a solution is near.  We look forward to reading that separate report.

• Officials continue to fail to follow procedures at a significant rate

[pre-election] The audit identified twenty-three (23) cards where the audit log indicates card duplication events. Card duplication is not authorized per SOTS Office instructions. Otherwise the cards were properly programmed for elections…There are 76 cards (15%) that were properly programmed, but were found in unexpected states or contained unexpected timing of events. This does not necessarily present an immediate security concern, however the findings indicate that the established procedures are not strictly followed in some cases.

[post-election] 14 contained junk data
2 were not programmed (formatted, but blank)
3 were involved in duplication
4 were non-standard cards (32KB instead of 128KB) [LHS not election official error] 4 were programmed for different elections

The main concern with such failures to follow procedures is that they are symptomatic that other procedures are frequently not being followed, yet each failure represents a possible lapse in security and election integrity.

Comments from our post on last year’s report still apply:

• A non-random partial post-election audit of memory cards is useful, but it is insufficient
• How many more tests, reports, and elections will it take before the junk data problem is significantly reduced? [Thanks to UConn, based on the 2009 report, we may have an answer soon]
• Almost every failure to follow procedures is an opportunity to cause problems, cover up errors, or cover up fraud. [including not sending in cards for testing]. We can only hope that the Registrars of Voters will join in the commitment to meet a much higher standard.

For more details behind these comments please read our post on last year’s report.

Press Release, Full Report etc: <click>

Summary, from the Press Release and Report:

Coalition Finds Unsatisfactory Improvement
In Election Audits Across The State

Citizen observation and analysis show the need for more attention to detail by officials, improvement in counting methods, and ballot chain-of-custody

The Coalition noted significant differences between results reported by optical scanners and the hand count of ballots by election officials across Connecticut. Compared to previous audits, the Coalition noted small incremental improvements in the attention to detail, following procedures, and in the chain-of-custody.

Coalition spokesperson Luther Weeks noted, “We acknowledge some improvement, yet there is still a long way to go to provide confidence in our election system that the voters of Connecticut deserve.”

From the report:

In this report, we conclude that the November post-election audits still do not inspire confidence because of the continued lack of

• standards for determining need for further investigation of discrepancies,
• detailed guidance for counting procedures, and
• consistency, reliability, and transparency in the conduct of the audit. .

We find no reason to attribute all errors to either humans or machines.

Cheryl Dunson, League of Women Voters of Connecticut’s Vice President of Public Issues, stated, “We continue to support our past recommendations to the Secretary of the State and the Legislature for improvement in the post-election audit laws, counting procedures, and chain-of-custody.”

Tom Swan, Executive Director, Connecticut Citizen Action Group, said, “Among our greatest concerns are the discrepancies between machine counts and hand-counts reported to the Secretary of the State by municipalities When differences are dismissed as human counting errors, it is unlikely that an audit would identify an election error or fraud should that occur”

Cheri Quickmire, Executive Director, Connecticut Common Cause said “There needs to be training and accountability.  Election officials need to be familiar with the procedures, follow the procedures, and the procedures must be enforceable.”

Press Release, Full Report etc: <click>