[E-voting] Letter in Today's Indo

Patrick O'Beirne mail2 at sysmod.com
Mon Mar 8 09:31:43 GMT 2004


I'll have a go at this but I apologise in advance if I'm going over ground 
that's been well worked before on this list - I'm new here.

>ICTE advocate a dual-vote counting system: electronic and the current 
>manual-paper system. They call for accuracy, yet where disputes arise they 
>want the paper system to have primacy, despite such inherent flaws as the 
>random selection of surplus transfers and simple human error.

That's a difficult point to answer because it depends on what risks you accept.

The politician's mind is to have no choice, so you can't question the 
outcome. That POV is that if there are problems, best not knowing about 
them. The engineer's mind is to have a cross check so you at least know 
when there are problems. That POV is that not knowing how bad something is, 
and therefore accepting an unknown risk, is always worse than knowing just 
how bad it is and accepting that known risk.

If you have parallel systems, how do you check them? Only by reproducing 
the automatic system in the manual one. This has three problems:

1) Human counting is not 100% reliable and cannot produce reproducible 
results. I'm no psephologist but I understand that basically counters just 
accept a "close enough" answer and give up when they get too tired.

2) The extent of checking has two determinants - how defective the system 
is, and how risk averse you are to missing a defect. The lower the error 
rate, the more sampling you have to do to meet a given level of assurance. 
The more risk averse you are, the more you have to sample to gain more and 
more reassurance. The only way of getting 100% assurance is to check 
everything - and that might well be argued as necessary to set a benchmark. 
But given point (1) we can't even get 100% assurance there.  Now, who sets 
acceptable error levels? Do we know what they are at the moment? If we do, 
they can be taken as maxima and the verification process should be designed 
to get less than that - otherwise, why are we doing this?

3) There is an even hairier problem that may well be built into the 
counting legislation and the software. Again, I'm no expert on this, and so 
I should not really even be commenting in a forum of experts that may well 
include tallymen, but I understand that the random selection of surplus 
transfers is not mandated, it is a convenience for the counters and they 
may make a judgement in any specific case that full counting is not needed 
because it would not change the likely outcome. But in close runs (I 
remember the Dick Spring four votes difference) a full count may be done 
(was it in the Spring case?). Now assuming that the software works on a 
deterministic full count basis, we could - eventually - get the manual 
count done enough times that the results scatter within accepted error 
levels around the automatic count.  If the software, however, also has 
random selection built into it, then we are lost - we have no deterministic 
outcome.

With regard to sampling schemes, is there a statistician in the group who 
could comment?

>They want to reject the new, accurate and tested system in favour of the 
>older one. Is this is not a Luddite mentality?

"accurate"?
"tested" - yes, some machines tested some number of times, but we're 
talking about the verifiability of every machine in every use case.
"in favour of" rather than "as well as, as a check on"?


>Perhaps, in the interests of transparency and democratic accountability, 
>they could answer a few questions. Why were they so silent when e-voting 
>was test run in the last general election and the Nice referendum? Why 
>were they not as vocal when the Government clearly signalled its plans to 
>introduce e-voting some time back?

A sore point. It looks like the IT professional was generally asleep at the 
wheel last time. And the losing side (e.g. Nora Owens) were too good 
sports, and accepted the result trustingly. They did not challenge the 
result at the time, maybe a constitutional challenge could have concluded 
that the results were unprovable and therefore unreliable. Still, being lax 
in the past is no reason to be lax now.

>How are they constituted and are people like me, who believe the proposed 
>system is better, welcome to join?

People like Derek need to join and make their points. How else does debate 
get going?



  Patrick O'Beirne,     Systems Modelling Ltd.
  Gorey, Co. Wexford, Ireland. +353 55 22294
  www.sysmod.com  I.S. management consulting
.





More information about the E-voting mailing list