Electronic Vote Counting: Where Did We Go Wrong?

By Howard Stanislevic, VoteTrustUSA E-Voter Education Project
November 21, 2006

http://www.votetrustusa.org/index.php?option=com_content&task=view&id=2067&I
temid=26

Recent discussions with other researchers in the election auditing field led
me to a paper published over 30 years ago that shows just how "far" we've
come in ensuring the integrity of our elections.

In 1975 while working for the National Bureau of Standards (NBS), Roy G.
Saltman, authored a paper entitled, "Effective Use of Computing Technology
in Vote-Tallying." In Appendix B of that seminal work, "Mathematical Considerations and
Implications in the Selection of Recount Quantities", Saltman described a
methodology for determining sample sizes to be used in independent audits of
electronic vote counting systems which include optical scanners, punch card
readers and touchscreen or pushbutton direct recording electronic (DRE)
voting machines with voter-verified paper audit trails.

Among other things, Saltman reported that regulations covering recounts
in different states varied, as they do today. Some typical recount regulations
were:
(1) a manual recount could be demanded by any candidate willing to pay
for it;
(2) a full manual recount was automatic if the candidates'
totals differed by a very small percentage of the vote and;
(3) a fixed percentage of precincts were manually recounted regardless
of the apparent vote separation (margin) between the candidates.

He also pointed out that the then current law in the state of California
which called for a fixed audit percentage of only 1% of the state's precincts
was inadequate, stating that, "recount percentages should increase as the
opposing vote totals approach equality." In other words, narrower reported
margins of victory require larger audits. Yet the 1% law is still on the
books today and until recently, did not even include absentee ballots!

Federal legislation proposing a 2% audit has also been introduced, and some states
have other fixed percentages in their auditing laws and regulations. None of
these audits may be adequate to confirm the outcome of all races on the
ballot, and some of them may actually be larger than necessary. The idea
that they are better than nothing, which seems to be their only
justification, is not particularly reassuring when the integrity of our
democracy is at stake. Such laws should specify a probability of miscount
detection or confidence level -- not a fixed auditing percentage. For
example, a confidence level of 99% means there would be a 99% probability of
detecting any miscount large enough to alter the outcome of a race.

Saltman was not only concerned with the mathematics of the auditing
problem in 1975. He went on to say, "The selection of some precincts for
recounting should be granted to candidates." This is a far cry from the
"sore loser" status accorded candidates who question election results today
despite the lack of meaningful audits (even where possible) and the absence
of observable independent verification processes inherent in many electronic
vote counting systems. According to Saltman, "Candidates' supporters and
precinct workers are those persons most likely to have the keenest sense
that a possible discrepancy exists." Few would disagree, but there still has
to be something to audit to prove that such discrepancies exist; a hunch is
not much to go on.

Consistent with the above, Saltman also introduced a notion called the
"maximum level of undetectability by observation." I called this same parameter
"within-precinct miscount" in a recently published paper, "Random Auditing of
E-Voting Systems: How Much is Enough?" But like some others working on the auditing
problem today, I had no idea that Saltman had previously come up with the same
essential concept so long ago. What it means in English is that there is a limit
to the amount of vote shifting that could take place in a precinct without arousing
suspicion by the folks on the ground. This actually limits the necessary
size of the sample to be used for the audit without compromising its
effectiveness, as long as anomalous precincts can be recounted in addition
to those selected by the random audit, as suggested above. For example, an
assumption that no more than 20% of the total votes in one machine or
precinct could be switched from one candidate to another without anyone
noticing (or another value based on historical election data) could be used
in determining the sample size of the audit.

One thing that was not thoroughly explored back in 1975 is that varying
precinct sizes could conceal miscounts in such a way as to render them
undetectable by an audit that implicitly assumed all precincts had about
the same number of votes. One way of addressing this as explained in my
paper is to adjust the sample size, based on ranking the precincts' vote
counts and determining the absolute minimum number of corrupt precincts that
could change the outcome. Saltman briefly mentioned a more complex method
involving proportional sampling.

Clearly precinct size needs to be taken into account somehow, but this was not
overlooked by Saltman 31 years ago. Those responsible for our election systems
have certainly had ample opportunity over the last three decades to decide on
the best way to audit them, so why has it become necessary for election integrity
activists to reinvent the wheel and to also rediscover such solutions in archival
documents?

Over three decades have passed since the electronic vote count
auditing problem was essentially solved by Roy Saltman. During this period
the NBS has been renamed the National Institute of Standards and Technology
(NIST); three versions of national voting system standards have been
produced (none of which has an independent auditing or verification
requirement), Congress has allocated billions of dollars to the States to
buy voting systems of poor reliability and in many cases, no auditability;
electronic vote counting systems have become almost ubiquitous -- and we are
still auditing them improperly, if at all. Some are attempting to make new
laws using the same flawed methodology of fixed-percentage auditing shown by
Saltman to be inadequate back in 1975. And others seem to prefer to do
absolutely nothing.

So after all these years, how did we get it so wrong? Only those responsible
can answer this question, but it would probably be better to just get about
the business of fixing this problem so we don't have to have this discussion
again 30 years from now.

_______________________________________________