Most projects start out slowly, and then sort of taper off. – Augustine’s Law #XL
Last week, the University of Connecticut (UConn) released a report on memory card testing covering 2007 – 2010. The results from 2007 until pre-election testing for August 2010 had been previously published, we expected to see the 2010 results much sooner. <report>
From the Conclusion, our comments in brackets [ ]:
Correctness of Card Programming: The audits determined that 100% of the cards actually used in the election [and actually submitted to UConn for testing] showed correct programming in terms of both the election description data and the executable code on the cards. In the case of the pre-election cards, in all cases where small discrepancies in the election description data were discovered, these differences were due to the very late changes, such as candidate name changes, substitutions, and race changes.
Audit Coverage: The number of memory cards submitted for audits fell substantially in 2010. We understand that in some cases districts were advised to not submit cards for audit in an apparent effort to occlude the fact that memory cards were duplicated. It is recommended that the SOTS Office encourages the districts to always submit one out four cards for pre-election audit and all of their used cards for post-election audit. The number of cards examined by the audits needs to be substantially increased in future elections to provide a better statistical basis for the overall election landscape in Connecticut. Not only this will help ensure proper programming of the cards, but it will also help address the reliability problem of the memory cards…
This dramatic drop in card submission renders most of the other statistics in the report unreliable and questionable. As UConn states, officials may be avoiding sending in duplicated cards; they could be choosing to send in more “junk data” cards as they are useless in the election; or avoiding sending in “junk data” cards assuming. incorrectly, it would reflect badly on them . Without public drawings we have no indication that cards are selected randomly, or that officials actually understand that they should be. Without accurate data it is hardly worth reviewing and making decisions based on the statistical analysis of the partial data.
As we have noted in the past, because the cards are not actually and publicly randomly selected, in addition to making it impossible for the reported results to be statistically accurate, it also provids an easy loophole for errors and skulduggery to be covered-up.
An earlier UConn report indicated that the problem was old batteries and that replacing batteries regularly might solve the “junk data” problem. Apparently this is not always so, with some cards quickly draining the batteries:
Continuing with the Conclusions:
This data loss is most likely caused by the weak batteries on the cards (however, as of this writing it is not clear how long a fresh battery lasts in a memory card). We are continuing to examine this issue. Increasing audit coverage will enable us to obtain and evaluate more cards that failed in search for a solution. In particular, we know that some cards drain batteries much faster than most; when we identify such cards it is recommended that they are removed from circulation. Longer term solution may be to develop replacement cards that use non-volatile memory technology…
Memory Card Duplication: In recent elections more then 6% of the cards [selected and submitted by officials] were involved in duplication. We note that the only authorized entity to provide card programming for election in Connecticut is LHS Associates. There is no guarantee that cards duplication done by the districts correctly reproduces data and programming on the copy cards. Additionally, if duplicated cards are not submitted for audits it increases the risk of using incorrect cards in elections. It is recommended that the SOTS Office reinforces its policy that prohibits card duplication…
Adherence to Election Procedures: The technological audits established that the districts do not always adhere to the established pre-election procedures. Most notably, in recent elections over 6% of the memory cards are duplicated by the districts, a practice that is not permitted by the SOTS Office. Additionally, some districts do not prepare all of their cards for elections and/or prepare for elections by running elections instead of running test elections. It is recommended that the SOTS Office reiterates the importance of following the prescribed election procedures. Lastly, some districts send cards for pre-election audit before they test the cards, while other districts send cards after they test the cards. For the pre-election audit to be most effective, it is recommended that districts uniformly send cards after the cards are tested and prepared for elections.
Overall, we applaud the report and the work of the UConn Voter Center. We are disappointed in the data submitted by election officials and the lack of progress in effectively addressing memory card problems. We are sympathetic to officials for the problems bad memory cards cause, yet our sympathy ends when they do not play their part in providing cards needed for UConn to make detailed and accurate assessments. We note that the lack of cooperation happened in the Bysiewicz Administration. We hope that the Merrill Administration will elicit more cooperation and encourage production of more timely reports for both memory cards and post-election audits.
For memory card testing to be useful and reach the potential of the exemplary testing developed by UConn, the program needs to be well defined and mandatory, enforceable, and enforced. The program should be mandated by law and/or all memory cards required to be sent through UConn in both directions from and to registrars, never to and from LHS, the vendor responsible for programming the cards. Or as we have recommended, the cards should be programmed in Connecticut, co-located with an independent testing function using the UConn developed test.
