Better rules for EM (was Re: FAQ Rules Explained)
Steve Eppley
seppley at alumni.caltech.edu
Fri Mar 7 11:40:56 PST 1997
Mike Saari wrote:
> We are trying out a new system of democratic parliamentary
> procedure, to generate a FAQ for this group and other related
> issues. This system is NOT consensus, nor is it majority rule, nor
> is it Robert's Rules - although it has elements borrowed from all
> of these traditions. This message provides a brief overview. This
> method will allow this group to generate FAQ content and make other
> decisions in a rational manner, without giving any member
> (including myself) excess influence.
But it's NOT rational. Combining a supermajority passing ratio with
plain Yes/No voting irrationally privileges one rival alternative
over others.
For instance, there recently was a choice to be made in
priorities at deliberate.com, which operates under Saari's rules using
a 3:1 passing ratio. There were three alternatives proposed for the
name for a new maillist:
constitution
bylaws
bylaws-theory
Saari proposed making "constitution" the default name if no other
name was chosen, and this passed with two votes of support (his and
someone who preferred a different name even more--keep reading):
10 --> Saari at aol.com
10 --> jjjacq at ozemail.com.au
(I was busy preparing the Pasadena League of Women Voters program
on election reforms, so I didn't cast an opposing vote before Saari
declared his proposal passed and closed his poll. His proposal
gave "constitution" 3:1 protection, and though "bylaws-theory"
turned out to be more preferred by the group, Saari's vote to
oppose "bylaws-theory" was enough to keep it from reaching the
3:1 threshold.
-4 --> Saari at aol.com
10 --> jjjacq at ozemail.com.au
-1 --> williams at linux1.elision.com
Rational? No, the group prefers bylaws-theory more than
constitution. John Jacq's vote was counted for constitution, instead
of against it: his full strength preference for bylaws-theory over
constitution wasn't respected. (I cast a vote of +5 for
bylaws-theory, but Saari closed the poll before my vote registered.
I would have voted +10 instead of +5, but I didn't like the idea of
making any alternative privileged by a bad set of rules.)
-snip-
> But sometimes discussion IS important. Won't we get a bad result
> if there are "premature" calls for votes?
>
> If someone makes a premature proposal, you (and anyone else who
> feels similarly) can just vote OPPOSED.
This process wastes the labor of those who spent time voting on the
flawed proposal. And since votes of others are not available,
everyone who sees a flaw needs to spend time voting No, since flawed
decisions are BINDING and hard to undo because of the 2:1 privilege.
But with a little bit of discussion first, better proposals are made,
labor is saved, and members won't always be "under the gun"--they can
anticipate upcoming proposals.
> What if someone's proposal is "pretty good" but needs to be fixed
> or has a fatal flaw? Why can't I amend someone else's proposal?
>
> Amending someone's proposal means the group never gets to vote
> directly on the original concept.
True, but a better alternative system is to append another "fixed"
proposal to a multiple choice (ranked) ballot, rather than hold a
vote to amend. This is the system I recommend.
For instance, suppose we vote to put a section X in the FAQ. (I.e.,
X receives at least 2:1 support and at least 4 votes of support.)
That's really a preference vote between X and "don't include X",
where "don't include X" is a supermajority-privileged choice.
Then later someone proposes replacing X with Y. It's irrational
for X to be privileged--the only alternative which might sensibly be
privileged is "include neither X nor Y" in the FAQ. The rational
procedure would be for us to then use a preference vote where X, Y
and "include none of these" are choices to be ranked, and the
"include none of these" alternative is privileged in the tally.
See below for more info on how to tally with a Condorcet modified
for supermajority.
> In most decision systems with "amendments" there arise situations
> which encourage "tactical" or dishonest voting
But Saari's system has such problems too. And the 2:1 passing ratio
opens additional options for manipulation, as the "constitution"
example above illustrates.
> Under this new system, there is no need to settle for
> second-best however. You can always post your own
> proposal, similar to the other one but with the defect
> corrected. Each variation then gets judged on its own
> merits.
Not if the first variation passed; then it gets special "status quo"
privilege--the 2:1 incumbency protection. Better if we append the
"better" proposal to a ballot which includes the "second-best" and
"none of these" so the members can rank the alternatives.
> Why 2:1 for the passing ratio?
>
> Using a "passing ratio" is a balance between "majority
> rule" and "consensus", preserving many of the
> advantages of both. The actual value may move up
> or down (by group decision) but I believe that nearly
> every group, regardless of size, would have good
> results with a passing ratio somewhere between about
> 3:1 and 10:1. (We are starting a bit low with 2:1 but
> can correct later if we want.)
In my opinion, only certain proposals (e.g., repealing individual
rights, amending a long-standing law, appointing a person to an
office like judge which will exercise broad power for a long term)
should require a supermajority. Saari's view is that all decisions
should be equivalent because he wants to keep things simple, but I
don't consider simplicity to be that important here.
However, when a group wants to privilege a certain alternative
like "none of the above", it can do so more rationally by using the
supermajority variation of Condorcet which was briefly discussed in
the EM list:
1. The privileged alternative(s) would be choices on the preference
ballot.
2. The preference orders are counted pairwise into a pairwise matrix.
3. If the group has a supermajority threshold ratio (R:1):
In the pairwise matrix, preferences for a privileged choice over a
nonprivileged choice are multiplied by the "supermajority" factor R.
Example: 2:1 threshold (R=2)
46: X > Y > None
20: Y > X > None
34: Y > None > X
prefer X less prefer Y less prefer None less
X vs Y 54 46
X vs None 34 66
Y vs None 0 100
The matrix cells which would be multiplied by the supermajority
factor are "prefer X less than None" and "prefer Y less than
None":
prefer X less prefer Y less prefer None less
X vs Y 54 46
X vs None 34*2=68 66
Y vs None 0*2=0 100
4. If the group has a "support quorum" (SQ) requirement, as Rob
has imposed on us ("at least 4 votes of support"--meaning SQ=4):
In the pairwise matrix, any preferences for a privileged choice over
a nonprivileged choice which total less than SQ are bumped up to SQ.
Example: (continuing the example from step 3)
prefer X less prefer Y less prefer None less
X vs Y 54 46
X vs None MAX(4,34*2) 66
Y vs None MAX(4,0*2) 100
5. The modified pairwise matrix is tallied by Condorcet's method.
Example: (continuing the example from step 4)
prefer X less prefer Y less prefer None less
X vs Y 54L 46
X vs None 68L 66
Y vs None 4 100L
---- ---- ----
Largest Loss: 68 0 100
Y is undefeated "supermajority-pairwise", so Y passes.
Here's the old, more compact matrix format, which may better
illustrate this method. Choices N1, N2, and N3 are nonprivileged,
while choices P1 and P2 are privileged. The value in a cell [i,j],
before the matrix is modified by the supermajority functions, is the
number of voters who ranked the choice i (listed at the left of the
row) ahead of the choice j (listed atop the column).
N1 N2 N3 P1 P2
N1 |
N2 |
N3 |
------------------------+--------------
P1 * * * |
P2 * * * |
The value p[i,j] in each cell [i,j] in the lower left quadrant of
the matrix (marked by an asterisk) is modified according to this
formula:
for all privileged i and nonprivileged j
p[i,j] := max(SQ, R*p[i,j])
where
SQ is the support quorum threshold
and
R is the support/opposition ratio.
---Steve (Steve Eppley seppley at alumni.caltech.edu)
More information about the Election-Methods
mailing list