Re: DIS: [Draft] Refactoring IRV

2019-05-27 Thread Owen Jacobson

On May 25, 2019, at 5:24 PM, James Cook  wrote:

> Some bugs:
> 
> * R955 specifies invalid options are eliminated before the process
> starts; it's probably good to keep that.
> 
> * The voting strength of each ballot should matter.
> 
> * When determining whether an option has a majority, votes for PRESENT
> or listing only options that have been eliminated shouldn't count.

Thank you. Given how foundational voting is to Agora’s integrity, I am 
deliberately taking this slowly to give time for this kind of feedback. I think 
anyone rushing to pass changes to the voting system should be viewed with 
suspicion, obviously.

I had not intended to remove the language specifying that a ballot of voting 
strength of N is handled as if it were N identical ballots. I suspect that 
there may be differences between that (current) interpretation, and your 
interpretation below that the sum of the voting strengths determines the 
ordering of candidates, but I haven’t done the math to prove it.

Good catch on PRESENT and on eliminating options which have become invalid. 
Thanks also to omd for pointing out must/MUST confusion and the lack of a “by 
announcement” clause or equivalent.

> 2. For an instant runoff decision, the vote collector determines the
>   outcome by the following process. During the process, an option's
>   first-place voting strength is defined to be the sum of the voting
>   strengths of the ballots that list that option before all other
>   options that have not been eliminated, and the remaining voting
>   strength is defined to be the sum of voting strengths of valid
>   ballots in this decision that list at least one option that has not
>   been eliminated.

It might be worth making the modal (or at least modal-ish) nature of these 
predicates explicit by adding a few “at a specific point in time”s or similar 
language, but I like the overall structure here of pulling the definitions out 
front so that they can be applied at various points throughout the ballot 
counting procedure.

>   a) First, all entities that are part of a valid vote, but were not a
>  valid option at the end of the voting period, or are disqualified
>  by the rule providing for the decision, are eliminated.

Probably want to use “ballot” or “vote” consistently, rather than alternating.

>   b) If no ballot lists an option that hasn't been eliminated, the
>  outcome is null.

A useful addition.

>   c) Otherwise, the vote collector successively eliminates options
>  until some option's first-place voting strength is more than half
>  the remaining voting strength, and that remaining option is the
>  outcome of the decision. For an option to be eliminated, its first
>  place voting strength must be less than or equal to the first
>  place voting strengths of all other options, and if it is equal to
>  another's, the vote collector must specify which option was
>  eliminated in the announcement of the decision's resolution.

omd’s must/MUST observation applies here, too. Perhaps the following?

  c) Otherwise, the vote collector MUST successive eliminate options
 until some option's first-place voting strength is more than half
 the remaining voting strength. The outcome of the decision is that
 option.

 When eliminating an option, the vote collector MUST eliminate an
 option whose first place voting strength is less than or equal to
 the first place voting strength of all other options. If the
 eliminated option's first place voting strength is equal to
 another options', then the vote collector MUST additionally
 specify which option they eliminated in the announcement of the
 decision's resolution.

-o



signature.asc
Description: Message signed with OpenPGP


Re: DIS: [Draft] Refactoring IRV

2019-05-26 Thread James Cook
On Sat, 25 May 2019 at 21:33, ais...@alumni.bham.ac.uk
 wrote:
> On Sat, 2019-05-25 at 21:24 +, James Cook wrote:
> > I couldn't resist making my own attempt. It's a lot wordier than
> > yours, unfortunately, but it addresses these points and omd's first
> > point. Maybe there's some middle ground that's less wordy.
> >
> > 2. For an instant runoff decision, the vote collector determines the
> >outcome by the following process. During the process, an option's
> >first-place voting strength is defined to be the sum of the voting
> >strengths of the ballots that list that option before all other
> >options that have not been eliminated, and the remaining voting
> >strength is defined to be the sum of voting strengths of valid
> >ballots in this decision that list at least one option that has not
> >been eliminated.
> >
> >a) First, all entities that are part of a valid vote, but were not a
> >   valid option at the end of the voting period, or are disqualified
> >   by the rule providing for the decision, are eliminated.
> >
> >b) If no ballot lists an option that hasn't been eliminated, the
> >   outcome is null.
> >
> >c) Otherwise, the vote collector successively eliminates options
> >   until some option's first-place voting strength is more than half
> >   the remaining voting strength, and that remaining option is the
> >   outcome of the decision. For an option to be eliminated, its first
> >   place voting strength must be less than or equal to the first
> >   place voting strengths of all other options, and if it is equal to
> >   another's, the vote collector must specify which option was
> >   eliminated in the announcement of the decision's resolution.
>
> There are two potential bugs in part c). One is that the vote collector
> hasn't been given a mechanism to eliminate candidates (they just have
> to do it, and then say what they did, but they don't have any way /to/
> do it); the other is that "For an option to be eliminated," can be
> interpreted as applying to the whole process, in which case it ends up
> overruling part a) due to Cretans.

Thanks. I'm now favouring Bernie's idea to use score voting, partly
because it seems simpler to describe, but if others want to stick with
IRV I might try addressing these bugs.


Re: DIS: [Draft] Refactoring IRV

2019-05-25 Thread ais...@alumni.bham.ac.uk
On Sat, 2019-05-25 at 21:24 +, James Cook wrote:
> I couldn't resist making my own attempt. It's a lot wordier than
> yours, unfortunately, but it addresses these points and omd's first
> point. Maybe there's some middle ground that's less wordy.
> 
> 2. For an instant runoff decision, the vote collector determines the
>outcome by the following process. During the process, an option's
>first-place voting strength is defined to be the sum of the voting
>strengths of the ballots that list that option before all other
>options that have not been eliminated, and the remaining voting
>strength is defined to be the sum of voting strengths of valid
>ballots in this decision that list at least one option that has not
>been eliminated.
> 
>a) First, all entities that are part of a valid vote, but were not a
>   valid option at the end of the voting period, or are disqualified
>   by the rule providing for the decision, are eliminated.
> 
>b) If no ballot lists an option that hasn't been eliminated, the
>   outcome is null.
> 
>c) Otherwise, the vote collector successively eliminates options
>   until some option's first-place voting strength is more than half
>   the remaining voting strength, and that remaining option is the
>   outcome of the decision. For an option to be eliminated, its first
>   place voting strength must be less than or equal to the first
>   place voting strengths of all other options, and if it is equal to
>   another's, the vote collector must specify which option was
>   eliminated in the announcement of the decision's resolution.

There are two potential bugs in part c). One is that the vote collector
hasn't been given a mechanism to eliminate candidates (they just have
to do it, and then say what they did, but they don't have any way /to/
do it); the other is that "For an option to be eliminated," can be
interpreted as applying to the whole process, in which case it ends up
overruling part a) due to Cretans.

-- 
ais523



Re: DIS: [Draft] Refactoring IRV

2019-05-25 Thread James Cook
> I intend to propose the following change to rule 955, in place of the current 
> definition of IRV:
>
> > The outcome of an Instant Runoff decision is:
> >
> > a. If a single option has the absolute majority of valid ballots specifying 
> > it as the first entry on the list, then the outcome is that option; 
> > otherwise
> >
> > b. The option with the fewest valid ballots specifying it as the first 
> > entry on the list is identified, and the outcome is the outcome of an 
> > Instant Runoff decision as if that option had been removed from each valid 
> > ballot that contained it.
> >
> >If there are multiple such options, the vote collector for the decision 
> > can, and must, select one to remove, specifying that they did so in the 
> > message resolving the decision.
>
> Does this contain any obvious scams? Does this accurately capture IRV as 
> performed by Agorans?

Some bugs:

* R955 specifies invalid options are eliminated before the process
starts; it's probably good to keep that.

* The voting strength of each ballot should matter.

* When determining whether an option has a majority, votes for PRESENT
or listing only options that have been eliminated shouldn't count.

I couldn't resist making my own attempt. It's a lot wordier than
yours, unfortunately, but it addresses these points and omd's first
point. Maybe there's some middle ground that's less wordy.

2. For an instant runoff decision, the vote collector determines the
   outcome by the following process. During the process, an option's
   first-place voting strength is defined to be the sum of the voting
   strengths of the ballots that list that option before all other
   options that have not been eliminated, and the remaining voting
   strength is defined to be the sum of voting strengths of valid
   ballots in this decision that list at least one option that has not
   been eliminated.

   a) First, all entities that are part of a valid vote, but were not a
  valid option at the end of the voting period, or are disqualified
  by the rule providing for the decision, are eliminated.

   b) If no ballot lists an option that hasn't been eliminated, the
  outcome is null.

   c) Otherwise, the vote collector successively eliminates options
  until some option's first-place voting strength is more than half
  the remaining voting strength, and that remaining option is the
  outcome of the decision. For an option to be eliminated, its first
  place voting strength must be less than or equal to the first
  place voting strengths of all other options, and if it is equal to
  another's, the vote collector must specify which option was
  eliminated in the announcement of the decision's resolution.


Re: DIS: [Draft] Refactoring IRV

2019-05-24 Thread omd
A few issues with the wording:

> > b. The option with the fewest valid ballots specifying it as the first 
> > entry on the list is identified, and the outcome is the outcome of an 
> > Instant Runoff decision as if that option had been removed from each valid 
> > ballot that contained it.

It's odd to create a new, hypothetical decision to base the outcome
on.  If there's a tie in a round after this step, the vote collector
would arguably have to identify that e's removing an option from a
hypothetical decision rather than the actual decision; e may not even
be able to do so, since that decision is never actually resolved.

> >If there are multiple such options, the vote collector for the decision 
> > can, and must, select one to remove, specifying that they did so in the 
> > message resolving the decision.

Could be clarified; "must" is presumably supposed to mean that the
vote collector cannot resolve the decision without doing so, but it
could be interpreted as MUST.


DIS: [Draft] Refactoring IRV

2019-05-22 Thread Owen Jacobson
Hi folks.

The ruleset for Agora handwaves the definition of instant runoff voting. From 
rule 955 (“Determining the Will of Agora”):

> For an instant runoff decision, the outcome is whichever option wins 
> according to the standard definition of instant runoff.

I assume this was originally done because fully specifying IRV takes a fair bit 
of text, as the method is not simple. There have been blessedly few disputes 
over this, but IRV is used in some sensitve places - in particular, for the 
election of officers, which are, in turn, essential to conducting business. I 
think it’s worth fixing.

The closest I can find to a “standard definition” of IRV is from Robert’s 
Rules, where it’s called “preferential voting”. From 
:

> One method is described here by way of illustration. On the preferential 
> ballot—for each office to be filled or multiple-choice question to be 
> decided—the voter is asked to indicate the order in which he prefers all the 
> candidates or propositions, placing the numeral 1 beside his first 
> preference, the numeral 2 beside his second preference, and so on for every 
> possible choice. In counting the votes for a given office or question, the 
> ballots are arranged in piles according to the indicated first 
> preferences—one pile for each candidate or proposition. The number of ballots 
> in each pile is then recorded for the tellers’ report. These piles remain 
> identified with the names of the same candidates or propositions throughout 
> the counting procedure until all but one are eliminated as described below. 
> If more than half of the ballots show one candidate or proposition indicated 
> as first choice, that choice has a majority in the ordinary sense and the 
> candidate is elected or the proposition is decided upon. But if there is no 
> such majority, candidates or propositions are eliminated one by one, 
> beginning with the least popular, until one prevails, as follows: The ballots 
> in the thinnest pile—that is, those containing the name designated as first 
> choice by the fewest number of voters—are redistributed into the other piles 
> according to the names marked as second choice on these ballots. The number 
> of ballots in each remaining pile after this distribution is again recorded. 
> If more than half of the ballots are now in one pile, that candidate or 
> proposition is elected or decided upon. If not, the next least popular 
> candidate or proposition is similarly eliminated, by taking the thinnest 
> remaining pile and redistributing its ballots according to their second 
> choices into the other piles, except that, if the name eliminated in the last 
> distribution is indicated as second choice on a ballot, that ballot is placed 
> according to its third choice. Again the number of ballots in each existing 
> pile is recorded, and, if necessary, the process is repeated—by 
> redistributing each time the ballots in the thinnest remaining pile, 
> according to the marked second choice or most-preferred choice among those 
> not yet eliminated—until one pile contains more than half of the ballots, the 
> result being thereby determined. The tellers’ report consists of a table 
> listing all candidates or propositions, with the number of ballots that were 
> in each pile after each successive distribution.

(Line breaks thus, unfortunately.)

This is a bit wordy, but provides a good starting point.

I intend to propose the following change to rule 955, in place of the current 
definition of IRV:

> The outcome of an Instant Runoff decision is:
> 
> a. If a single option has the absolute majority of valid ballots specifying 
> it as the first entry on the list, then the outcome is that option; otherwise
> 
> b. The option with the fewest valid ballots specifying it as the first entry 
> on the list is identified, and the outcome is the outcome of an Instant 
> Runoff decision as if that option had been removed from each valid ballot 
> that contained it.
> 
>If there are multiple such options, the vote collector for the decision 
> can, and must, select one to remove, specifying that they did so in the 
> message resolving the decision.

Does this contain any obvious scams? Does this accurately capture IRV as 
performed by Agorans?

-o