A Clarification by DR Janow-- on HAVA section 301

[FogerRox]

This discussion may help explain the multiplicity of numbers,
admittedly confusing.

The maximum acceptable error rates for 1990 and 2002 were 1
in 100,000
and 1 in 500,000 respectively. These numbers mean that every
time someone records a
single vote, the probability of getting it wrong because of
SW/HW errors (not voter errors)
is no bigger than either 1/100,000 or 1/500,000. This is
similar to saying that every
time you flip a coin the probability of heads is 50%, but
applied to recording a vote.

But the results of recording a vote or flipping a coin are
statistical. Iin the case of
voting, the machine or software can fail to record the choice
or record it incorrectly.
Even if a machine being tested is very, very reliable it could
record a wrong vote early in the
test process (bad luck). If you continue testing, you will
find no more errors for a large
number of tries, and the machine would pass. Alternatively, a
very, very unreliable machine could have
a run of luck and correctly record a long string of votes
without making an error. But that is
unlikely and the longer the test sequence becomes, the lower
the probability that the bad machine passes.

The other numbers you are seeing reflect this statistical
nature of the prescribed test sequences - i.e., how many
tries do you have to make to be 95% sure that the basic error
rates are no higher than
the specifications above (for both 1990 and 2002). If you
want to be 99% sure, you have do
do a much longer and more expensive test sequence. So both
1990 and 2002 specifications pick
the same confidence level: 95% (= 100% - 5%).

First, in order to pass outright, a voting machine has to have
NO errors in the first 297,589 tries (1990) or
1,549,703 tries (for 2002)*. If they get that far with no
errors, you can be 95% certain that on EACH ballot choice
the error rate is no larger than the ones above.

Suppose there is an error early on (before 297,589 or
1,549,703 tries).
Then the test has to continue on. A machine can also pass if
it has
a maximum of ONE error in the first 762,763 tries (for 1990)
or 3,126,404 tries (for 2002)*.

For 2002, if the machine fails very early before 26,997
tries*, game over, go home and don't come back until
you have modified the machine. There is a 95% probability
that the machine's error rate exceeds
the limit. I haven't seen the comparable number for 1990, but
calculate that it should be about 5100
tries.

The 2002 requirements are thus 5 times stronger than the ones
for 1990.

I believe the 1990 testing stopped at either 297589 or 762,763
tries. So even if a machine
can actually pass the 2002 tests, that would not be
demonstrated in the 1990 reports and it
would have to be re-tested to be certified.

I hope this clarifies, rather than obscures, what's going on.

*See the EAC VVS for 2000, Vol. II, Appendix C "National
Certification Criteria", Sec C5, page 7

Rich Janow
South Orange, N. J.
[e-mail]
----------------------
A thread on this same topic--
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x385691

[FogerRox]

1. wake up sheeple--

there are possibly 100's of thousands of voting machines that cant be used in the next federal election--- there is a log jam brewing that may force congress to delay HAVA one year.