[EM] Re: Voting Systems Study of the League of Women Voters of Minnesota

[Abd ulRahman Lomax]

At 02:34 PM 6/8/2005, Araucaria Araucana wrote:
>Let me summarize two sides of the main argument.
>
>On one side, we have those who say,
>
>    Let everyone have their say, even if they choose more than one
>    candidate.  Every voice should be heard.
>
>    [with mumbled grumblings about overly stringent voting requirements
>    and inadequate voting facilities]
>
>For now, let me call this the Liberal point of view, in the original
>sense that we should be as free and open as required to allow every
>vote possible.

That's pretty much what I'd say, though the legal standard (in Florida) was 
that if the voter's intention was clear, the vote should be counted. Yes, 
though, a basic argument would be that overvotes should be allowed because 
they do no harm and they evince voter intention.

>On the other side, we have those who mistrust the process:
>
>    How can we be sure that the ballots were not tampered with, between
>    voting and official tally?

Yes. This is the only counterargument I could think of. But it is quite weak.

>    [with mumbled grumblings about stuffing ballot boxes and
>    unqualified voters]
>
>For now, I will call this the Conservative point of view, with the
>original meaning of taking every precaution to avoid counting
>something that is not qualified to be counted.
>
>With Approval voting, marking more than one candidate with no other
>means of verifying the vote means that Conservatives will mistrust the
>process.  The ballots could have been marked with extra votes
>somewhere along the way.

Note that if ballots could have been so marked, they could have been so 
marked with the intention of nullifying the vote of the voter, and the 
present system allows that fraud to be effective. In fact, if ballots can 
be marked, they could probably also be substituted. I see no special reason 
to suspect fraud simply because a ballot is overvoted. In Florida, nobody 
instigated an investigation into how the overvoted ballots could have been 
fraudulently defaced. There were no charges of same. Rather, it was assumed 
that the overvoting was unintentional, which, given the patterns, was 
likely the case.

The result of discarding overvotes is not the prevention of fraud; rather, 
given that most people know that overvoting will void their vote, the 
result is the discarding of overvotes most likely resulting from a voter 
marking or punching the wrong spot, then correcting it. And such a voter 
could easily fully correct it if the voter thought the incorrect vote was 
likely to produce a bad outcome, by simply voting for all candidates as 
desirable or more so than the incorrectly marked candidate; thus the worst 
outcome, if the ballot were counted, would be a nullification of the effect 
of that vote. Discarding these overvotes, if is  not moot, fails to 
consider the expressed intention of the voter.

(The voter, if the voter notices the overvote, could get another ballot, 
but real-world conditions often make this difficult. Not to mention 
possibly embarassing.)

>So simply using standard ballots but relaxing (in a Liberal way) the
>'vote for only one' rule is accessible to fraud, to Conservatives.

Anyway, this *would* be the argument. And the counterarguments would be 
various:

(1) Fraud can take place in many ways, and fraud by adding extra marks 
succeeds if the ballots are discarded. Consider the effect in each pairwise 
election. Voter voted for A. Villain marks B in addition. Suppose in this 
election there were three other voters, one voted for A, one voted for B, 
one voted for C.

Result with ballot discarded: A 1, B 1, C 1.
Result with ballot kept: A 2 B 2 C 1.
Result if ballot not overvoted fradulently: A 2 B 1 C 1.

The result would be a three way tie with the ballot discarded and a two-way 
tie with the ballot kept, and the two-way tie is closer to the true result, 
the victory of A. Keeping the ballot reduces the impact of the fraud. (One 
way to look at this is to realize that overvoting expresses no preference 
between the marked candidates, but does express a preference of the set of 
them over all other candidates. The fraud removes the expressed preferences 
to the extent that extra marks are made, but leaves intact all other 
expressed preferences.

So discarding the ballot would *not* be a true conservative approach. It 
fails to deter fraud. Indeed, it causes an increase in the success of the 
fraud.

Have I missed something?

>But from the Liberal point of view, the current system favors the
>Conservatives because ballots could be tampered with (in order to
>invalidate some votes) by marking an overvote.

And discarding the ballots realizes the intention of the criminal.

>Besides that, single-vote can be implemented by simpletons, of whom we
>seem to have ample supply in our election administrations.  Each
>ballot can be sorted into exactly one candidate stack, and the number
>of single votes sums up to the number of ballots.

Given the fact that many elections are generally being run simultaneously, 
if this is the way that they count ballots, it is about as inefficient as 
can be imagined. Our town counts ballots by hand, and I think the official 
holds up the ballots and reads off the votes, which are tallied by someone 
else. Allowing overvotes would not increase the time necessary, unless 
there were a lot of overvotes.... But the time increase, even in that case, 
would not be large.

And most jurisdictions now use machine-counting devices, the best, in my 
opinion, use ordinary paper ballots, which are scanned and the votes 
extracted. This leaves a paper trail. I think that there are some 
experiments going on with the ballots being scannable by the voter to 
confirm votes.

And where electronic voting machines are used, it would be just as easy for 
a criminal to change a vote as to void it with an overvote. And that would 
have twice the effect. Voiding lowers the vote of the "victim" candidate 
but leaves the vote of the "aided" candidate the same. Changing the vote 
lowers the vote of the victim and increases the vote of the aided.

Counting the overvoted ballot causes the effect of the fraud to be exactly 
what fraud was committed: it adds a vote to the aided candidate. Discarding 
the ballot reverses that fraud, but also eliminates the legitimate vote.

(I also did some simple analyses where the fraudulent extra vote did shift 
the election in the direction of the fraudulent intent. This happens when A 
would not be a winner, for example, and without the fraud, B and C would 
tie. The fraudulent vote, if counted, leaves B as the illegitimate winner. 
So my first analysis above certainly does not tell the whole story. The 
bottom line is that if someone can come in and mark ballots, they can shift 
an election regardless of whether or not ballots are discarded. Discarding 
ballots does not appear to minimize the impact of fraud.

>A similar issue exists with ERIRV.  Under standard IRV, there is only
>one vote for each ordinal rank.  So at any round, a ballot would be
>sorted into exactly one stack.

Again, ballots are not usually counted this way, where they are counted by 
hand. Only single-election ballots would be counted this way, I'd expect.

(Stacking the ballots might seem to be a quick way to get a count and 
confirm it, but it does not actually do so, since recounting the stacked 
ballots will simply repeat a sorting error. Rather, to recount ballots, it 
is necessary to repeat the process. Indeed, I think this should be normal 
procedure. Ballots should be counted and totals reported in blocks, by two 
independent groups. If the totals match, the counts are accepted, if they 
do not match, they are recounted a third time.

(There are observers to counting processes, there to prevent just the kind 
of fraud described.)

>You can have Approval or Equal-Rank IRV and still satisfy the
>Conservatives by having the voter explicitly Approve or Disapprove
>each candidate.  Then the total number of Approval and Disapproval
>votes should add up to number-of-ballots times number-of-voters.
>Ballots cannot be tampered with simply by checking extra boxes.  But
>this is *much* more complicated for the voter.

Yes. And voters will make mistakes. By complicating the ballot procedure, 
in an effort to prevent a fraud that can be prevented in other ways, the 
result is increased failure of the tallies to reflect the real intention of 
the voters.

>A less-onerous alternative (in many-candidate races) would be to have
>the voter explicitly state the number of candidates being approved or
>equal-ranked.  This might be reasonable with machine-aided ballots.
>[Note that I advocate only machine-ASSISTED ballots, with only a
>voter-readable paper ballot used for the count.  I mistrust electronic
>voting.]  The voting form could automatically keep track of the number
>of candidates at each rank.

I just don't think the risk is high enough to justify the complexity. If 
ballots are vulnerable to fraudulently adding overvotes, they are, quite 
simply, vulnerable to manipulation of election results, whether or not 
overvotes are discarded. It's the wrong solution. The solution is to guard 
the ballots (as is done); ballots are available to be handled only in the 
presence of multiple observers in addition to the election officials, 
otherwise they are kept in locked boxes.

>Abd is correct that the more Liberal procedure is used in face-to-face
>meetings, but in such a setting there are many ways to prevent the
>count from being falsified.

Yes. But the crucial argument here is that discarding the overvotes 
partially realizes the intention of the criminal. Wrong solution to the 
problem. Is there any other cogent objection to counting overvotes?

How about this: for hand-marked ballots, NCR paper is used so that there is 
a second copy of the ballot. These second copies are collected securely at 
the same time but in separate containers from the first copies. The second 
copy container is sealed and locked away. If there is suspicion of fraud, 
they would be available for examination on the order of and under the 
supervision of a court. This would indeed detect fraudulent overvotes and 
would indeed allow correction of them.

There are good reasons not to allow voters to take away records of how they 
voted. But with machine voting, a paper record *could* be created on the 
spot, to be sequestered as described above. Again, this would detect and 
allow correction for subsequent fraud.