[EM] IRV vs Range on totalizing machines

[Abd ul-Rahman Lomax]

At 10:26 AM 8/19/2005, Warren Smith wrote:
>Second, it is *false* that you can do IRV on totalizing machines
>such as New York's lever mechanical-counter machines (which involve
>a lot of binary levers on the front, and there are counters you can read on
>the back).  There could indeed be a way for a voter to INDICATE a ranked 
>ballot
>to those machines via a range<-->ranked transformation.  But so what?
>The machine cannot do anything useful with any such indication.  It is sort
>of like me talking to you in Ancient Babylonian.  Of course, you are capable
>of listening to me, but it does not do either of us any good.

I have elsewhere argued differently, including, I think, today, but now I 
think I realize my error. The ballots individually express the rankings, 
but these machines are not ballots except for a few minutes. They do not 
store the rankings. Instead, they would total each position marked, but not 
the relative positions on each ballot, which is what is used in IRV. (I.e., 
IRV, absent special equipment, involves recounting the ballots, such that 
each ballot with a loser at top preference(s) is recounted to be a vote for 
the remaining first preference. Silly, I should have known better.... There 
are no ballots to recount with that type of voting machine. Which is a very 
good argument for them being quite dangerous, by the way. Given that 
optical scanning could be used at very low cost with total verifiability, 
dumping the remaining lever machines could be a good idea, thus eliminating 
this whole argument.

>On the other hand, with range voting, it is not only possible for the 
>voter to INDICATE
>the range vote to a New York style machine, it also is possible for the 
>machine to DIGEST
>those votes and for the results of the range election to be easily 
>computed from the
>readouts on the back of the machine.  Exactly how this is done, is discussed
>on the CRV site
>    http://math.temple.edu/~wds/crv/RangeVoting.html
>and click "VotingMachines" on the left.

Yes, we knew that it could be done with Range. Range uses simple 
summations; HOWEVER if it is going to average only those who did not 
abstain, though (as Mr. Smith proposes and advocates), it would have to 
have some method of distinguishing between (three voters, for a single 
candidate):

10
10
0
and
10
10
(blank)

These two votes produce the same sums and they involve the same number of 
voters. But they would produce a different average vote, 6.67 for the first 
and 10 for the second.

Voting methods should not produce different results depending on the type 
of voting machine in use in a precinct....