[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Per-item "quorum" and truncated ballots

moth@magenta.com wrote:
On Fri, May 23, 2003 at 05:24:59PM -0400, Buddha Buck wrote:

Imagine a vote along the lines of:
100 ballots of the form:
  [1] Red,    [ ] Blue,    [ ] Default

100 ballots of the form:
  [1] Red,    [ ] Blue,    [1] Default

25 ballots of the form:
  [ ] Red,    [1] Blue,    [ ] Default

with an R of 105.

I presume you mean with a quorum of 105.  Which would mean that we have
something like 44100 debian developers.

The defeats matrix looks like
     Red   Blue   Default
Red   ---   200      100
Blue  25    ---       25
Def    0    100       ---

In this example, Red is the IDW, and absolutely no one thought that Red was worse than the default option. Yet Red is rejected because fewer than 105 people thought it was better than the default option, so default wins.

Is this the "expected" behavor?

Out of more than fourty thousand debian developers, less than one hundred
stated that they preferred red over the vote defaulting?

Yes, I'd say that this is the expected behavior.  Those other votes,
which rank red and default equally above blue, constitute votes against
blue, not votes for red.  Those people would have accepted red as being
no worse than the current situation, but they didn't actively think it
was a good idea.

So you are saying it is acceptable and desirable for there to be no way to express truely equal preference for "Further Discussion" and some other option? Using my example, voting red equal to blue doesn't hurt or hinder either options chances of winning in relation to each other. But voting red equal to the default hinders red's ability to get accepted in favor of defaulting.

We've made three changes to SSD:

1) We've instituted a per-option quorum, requiring a minimum number of votes
for a particular option over the default in order to be considered.

2) We've instituted the ability to rank options as equal on a ballot.

3) We are using the ranking relative to the default option as proxy
for an approval ballot, and only considering options that are approved by the majority.

Our analysis of 1 and 3 have been based on the law of the exclused middle:
for any ballot, either A>D or D>A.  We haven't considered the effects of 2.

I think that the combination of all three changes has unforseen effects.

In particular, I think change #2 sets up the expectation with developers that they can vote "I don't care which of these two options win", while the other two changes do not treat a ballot equating an option and the default as neutral with regard to that option and the default -- they favor the default.

I think change #1 is the most troubling for me, and I would not mind seeing its removal. It's designed to eliminate options that don't have many supporters. But surely if there are enough ballots cast, an option that doesn't have many supporters will be unlikely to win anyway. The only time an option that doesn't have many (active) supporters has a chance of winning is when it also doesn't have many (active) detractors -- and when everyone else is expressing no preference one way or the other.

Why shouldn't the will of a small number of supporters win out over the will of a smaller number of detractors when the majority opinion is "I don't care"?

Reply to: