On 10/10/15 6:17 PM, Juho Laatu wrote:

that means you have to "price check" your paper ballot before
surrendering it to the ballot box.  does not seem to me to be a simple
and direct procedure.

yeah, but we should have a reasonably secure method to tabulate ballots
and get "summable" results at each precinct on the evening of the
election just after polls close.  it should be virtually flawless if
every voter marked their ballot to within, say, 90% of "par" (to use a
golfing term).  this is quite doable for oval or round "bubbles" or
"slots" or other simple binary marking geometries.  OCR is not safe
enough without "human assisted counting" which i think should be totally
avoided unless there is a mandated recount.

would this work for an election in, say, India?  with something like
10^9 ballots?

recounts are one thing.  and a nationwide recount in the U.S. would be
truly a massive and messy thing.  this is one reason why the Electoral
College has some supporters (which i am not).  if, in 2000, there would
have been a knock-down, drag-out recount fight between Bush and Gore, at
least the fight would have been contained in Florida.

but in the normal vote counting and tabulation in a governmental
election in a democracy, it should be routinely done with mindless (and
bias-less) machines, it should be precinct summable (with precinct
results completely transparent to the media and to partisans), and it
should be decided on the evening of the election unless it's very close
or there are other problems calling for a recount.

--

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

Actually I was thinking about the process of checking the votes, and keeping the vote casting process fully manual. If machines would be used in casting the vote, I might prefer (at least in the Finnish tradition) machines that are like simple mechanic typewriters that allow the voter to type some numbers on the ballot paper. This approach could save some votes that would be otherwise rejected. But that may not be necessary or the best approach since people may prefer manual voting (partly because of the tradition), and only very few votes will be rejected anyway. (If some numbers cause too many problems, one could use also Kristofer Munsterhjelm's trick and not use some numbers that might cause confusion.)

I waas thinking about some STV style multi-winner methods where it makes sense to store the votes one by one and not sum them up in a matrix (as would be practical with typical single winner ranked Condorcet methods).

Bubbles and slots would work too. I just ended up on this line because I started from the idea of using as simple ballots as possible. In Finland votes are counted twice manually to make sure that possible alternative interpretations of the ballots will be pointed out.

In Finland the idea is that votes are counted locally right after the election, which means that there is plenty of local workforce available, and the process can be real quick. This approach would scale also to India and 10^9 ballots.

The second count in Finland is called a "checking count". It will be done always (maybe the next day), and usually there is no need to come back to the votes after that (not locally nor centrally).

Bias-lessness is achieved in Finland by inviting representatives of all parties to take part in the vote counting process. I guess the tradition is to not to even start making biased interpretations.

STV is unfortunately not as summable as e.g. Condorcet. One may lose also some privacy and introduce some risk of coercion and vote buying by recording and distributing ranked votes to the central authority (and who knows even publishing them). I have no good foolproof solution for that right now. Risks to be estimated and appropriate protective measures to be taken (or just stay in some simpler methods).

Juho

On 10/11/2015 01:06 AM, Juho Laatu wrote:

all parties to take part in the vote counting process. I guess the
tradition is to not to even start making biased interpretations.

That brings to mind what I'd call a great open question: is the Droop
proportionality criterion compatible with summability? I suspect not,
and I suspect that a proof would make use of a pigeonhole principle. I
don't have much beyond that hunch, though.

Do you mean Droop proportionality with ranked votes? I'm thinking about a voter who votes A>B>C>D>E, where candidates A, B, C and D can not win. To pass the vote to E, the vote probably has to be stored as it is.

Juho

On 10/11/2015 02:23 PM, Juho Laatu wrote:

Yes, I was thinking of ranked ballot DPC. But it's a bit harder than
that because the single-winner analog, mutual majority, is compatible
with summability even though (if I recall correctly) inferring the whole
mutual majority set isn't. That is, summable methods can pass mutual
majority, but they can't let you know the whole minimal mutual majority set.

That may mean that summable methods that meet mutual majority do some sort of an overkill. I mean that having mutual majority depends on the actual votes in the same way as vote transfer did in my A>B>D>C>E example. You can get the same matrix with votes where some mutual majority exists or it doesn't exist. If this is the case, some vote sets that do not have mutual majority will be handled as if they had mutual majority since the (summed up) matrix can not tell us if there was a mutual majority or not.

Juho

On 10/11/2015 03:59 PM, Juho Laatu wrote:

Right. My point is that one could imagine this to be the case for Droop
proportionality as well. A person who thinks that you can have both DPC
and summability could say that there might exist methods out there that
elects a candidate from a Droop solid coalition and also elects that
candidate in a case with the same matrix but no Droop coalition. If
knowing the full mutual majority set is incompatible with summability
yet electing from it is compatible with summability, then that might be
the case for a Droop set too.

I don't think so; but because of the equivalence, the proof can't just
be that you don't have enough space to encode that the vote has to be
transferred to E. If it were, it would also prove that electing from the
mutual majority set is impossible because IRV is non-summable.

Ok, I think I got the point, or at least some part of it.

That led me to playing with the following votes
50: A>B>C
50: C>B>A
In the (typical) matrix all three candidates are identical, but A and C should be elected if there are two seats. The summing process can thus hide quite a lot of information.

What should we learn from this?

Juho

On 10/10/15 7:06 PM, Juho Laatu wrote:

we do that, too, in the U.S.  both for election night and also for
recounts.  but machines are more immune to bias (unless seriptitiously
programmed to be such, but open non-proprietary code should guard
against that).

and the other (non-machine) issue regarding "biasedness" is that because
of Duverger's law favors the two major parties, so the playing field is
not level for third parties and independents.

well, i think that is a permanent disadvantage to STV.  and IRV
opponents used that as an issue, alluding to the possibility of
something nefarious happening during transporting the voting data (like
in a thumb drive or whatever physical instrument with data from all of
the ballots) from the precinct to the central ballot-counting venue or
something else nefarious happening at a single obscure point (in the
code) at the central counting location (that some inside person could
slip in).  this is why precinct-summability is a desirable property of a
voting system.

i suppose that with STV, at each precinct, instead of posting vote
totals as you would for FPTP or Condorcet, you would post totals for
each possible way to mark the ballot.  but i don't think that candidate
organization nor the media would want to use those results.

i think Condorcet is simpler than STV.  because it's precinct-summable
and there isn't this kabuki dance of transferred votes.

and Condorcet is even simpler than FPTP with regard to burdening voters
in multi-candidate elections with tactical voting (because of the
ranked-choice ballot). normally the tactic ends up the "compromising"
tactic, but voters should not have to put up with that.  this was the
main reason we adopted STV in Burlington Vermont in the first place.
now we're stuck with it again.

L8r,

--

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

One can reduce the risk of something happening to the votes during transport by introducing means to check that the information is the same at both ends. Summability helps, since you can see the sums at both ends, and the local sums can be directly summed further to the end results. If the votes are not summable, the simplest approach is to publish the votes. Then the local records can be compared with the records at the central counting location, and final results checked against all the published local votes. One could use also some other compressed results like publishing the results counted from the local ballots (as if the election was local) at both ends. This is of course not fool proof in the sense that the even if this can guarantee the safety of the transport quite well, counters at the central location could still modify the votes after publishing the local results and before counting the final results. But that's another story, and could be fought against by other means (like allowing the presence of representatives of all parties).

This is a working alternative way to sum up the votes if the number of candidates is small enough. I note that I joined this conversation by presenting ways to handle elections with numerous candidates. Having 100 candidates would make it unpractical to sum up the number of all possible candidate orderings, but with 5 candidates it would be easy.

Yes. I just note here that STV (in multi-winner elections) and IRV (in single-winner elections), although technically similar, have different benefits and problems in practical elections. Some of the problems of IRV get diluted (influencing only the last seats in some rather random way) when applied to electing multiple representatives.

Juho