[E-voting] Any statisticians subscribed? Looking for a tipping point.

Catherine Ansbro cansbro at eircom.net
Thu Sep 8 17:04:59 IST 2005


The distinction between DRE and Counting Computer is very important.  It 
will affect any statistical value you might assign to a given voting 

If a Central Counting Computer is in use, it can affect ALL the votes, 
regardless of the accuracy value you might assign any given voting 
methodology or combination of voting methodologies in use for a given 

And if REV is being used, this is automatically also a medium for 
affecting a Central Counting Computer--so once again the use of this 
voting methodology means that it can affect ALL the votes, not just 
those votes that were placed using REV.   This should affect how a 
statistician would approach this issue, if they were trying to show 
"likelihood of accurate reflection of voters' intention."

One would also then have to factor in such things as likelihood of fraud 
occurring at all the possible points of vulnerability.  If DREs or 
Central Counting Computers or REV are used the number of vulnerabilities 
are exponentially greater.  Postal votes (w/o Central Counting Computer) 
also has many additional points of serious vulnerability.

In our current system the major point of vulnerability is during 
transport of the ballots.  It requires trust of the Gardai, which may or 
may not be warranted.  (And in any case, we should never have to take 
ANYTHING relating to elections on trust.  This is an invitation for 
trouble.)  This area would be more trustworthy if all transport of 
ballots were required to include at least 2 candidate reps or 
independent public observers.  As I've posted here before, I know of one 
case where someone was hired to transport a box, and on delivery she 
discovered that it was a ballot box.  There were no Gardai accompanying 
her and she could have done anything she wanted with it.


Brian O'Byrne wrote:

>I accept your point completely, and I'm not suggesting that DRE should 
>be considered for any increase in turnout. It is, as you suggest, too 
>easy to introduce an unprovable wholesale fraud, rendering the entire 
>vote invalid with no way to recover any idea of the voters' intent.
>Given that DRE, whether remote or in the polling booth, is entirely 
>unacceptable maybe my question is off-topic for this list. However 
>since remote voting and other voting channels were mentioned I 
>thought it was an interesting question to add.
>Are there values for those variables (the increased turnout, the 
>reduced accuracy, the skew) that would mean increased turnout with 
>lower accuracy would give a better overall result?
>Without being trained in statistical analysis my gut tells me that as 
>the turnout through the traditional channel decreases the value of 
>each additional vote increases, perhaps to the extent that the 
>election as a whole would benefit from increasing the number of votes 
>even if the accuracy falls.
>On Thursday 08 September 2005 16:12, Catherine Ansbro wrote:
>>Interesting questions.  You make a very good point.  You are right
>>that a relatively small number of skillfully "skewed" votes can
>>affect the election results.
>>What is more serious is--
>>What if you get 70% turnout that gives the honest opinion of 10% of
>>the electorate?  I am exaggerating to make a point, but this is the
>>risk we face when we give up our right to /directly/ monitor
>>counting of all the votes.
>>Once you introduce a system that uses centralised computers to
>>/count/ the vote, you can swing the whole election in undetectable
>>ways.  It becomes irrelevant that not all voters might have used a
>>vulnerble /voting/ method, if a vulnerable /counting/ method is
>>This only serves to emphasize the point that you make below.
>>Brian O'Byrne wrote:
>>>At the risk of being branded a heretic:
>>>When writing that previous message I prepared the paragraph:
>>>I suggest to you that 40% turnout that gives the honest opinion of
>>>40% of the electorate is more useful than 70% turnout that gives
>>>the honest opinion of 40% of the electorate.
>>>Now as stated, that rings obviously true, however what if you play
>>>with the numbers a bit:
>>>I suggest to you that 40% turnout that gives the honest opinion of
>>>39% of the electorate is more useful than 70% turnout that gives
>>>the honest opinion of 60% of the electorate.
>>>If that so obviously true?
>>>Is there a value in increasing turnout at the cost of validity?
>>>Lets say that at any election there is a proportion of the
>>>electorate that will always vote. Because the system is well
>>>designed a high proportion of those votes will be accurate
>>>reflections of the voter intent, meaning the proportion of 'good'
>>>votes is high.
>>>Now add another voting channel with a lower integrity. Some of the
>>>people who always vote use this new channel, but many new voters
>>>also use it. The overall proportion of 'good' votes reduces but
>>>the number of votes increases.
>>>Assume first that the 'bad' votes are randomly distributed. Does
>>>the reduction in accuracy affect the result?
>>>Now assume the 'bad' votes are skewed. How much of a skew is
>>>required to affect the result? What is the relationship between
>>>the integrity of the vote, the skew in the 'bad' votes and the
>>E-voting mailing list
>>E-voting at lists.stdlib.net

More information about the E-voting mailing list