[EM] cardinal pairwise (previously written messages)

[James Green-Armytage]

Hi Mike,
	This is just a review of the recent messages from me to you on the
subject of cardinal pairwise. Probably you have already read them, but I
thought it might be helpful to post them in one place, for the sake of
convenience. The first one of these, I think, was already posted. The
rest, I think, were not.

----------------
      12/2
----------------

Prescript: I recently started calling my method "cardinal pairwise"
instead of "weighted pairwise" for short. The full name is
"cardinal-weighted pairwise comparison," so really, either abbreviation is
appropriate. I just don't want you to be confused when I call it "cardinal
pairwise" in this posting, or later on... Sorry about the switcheroo, I
just thought that maybe "cardinal pairwise" was more descriptive. Feel
free to use the other name if you prefer that.

Mike, you wrote:
>
>Sure, Weighted Pairwise, from what you've said, seems to improve on the 
>criterion compliances of ordinary wv.

	Well, I think that it helps more with qualitative criteria than yes/no
criteria, in particular, resistance to severe strategic manipulation. Also
I think that the extra level of expressiveness (the ratings ballot) is a
good thing, and makes for more-meaningful resolution of sincere cycles. I
suppose that you could make yes/no criteria out of some of these things,
but I'm not sure that they would be able to completely encapsulate the
benefits of cardinal pairwise. 

you wrote:
>
>I haven't thoroughly compared WP to ordinary wv with AERLO & ATLO. WP may 
>not offer the strategy improvements of AERLO & ATLO.

	I doubt that ATLO and AERLO will be as effectively and easily
strategy-resistant as cardinal pairwise, but I'd be open to further
discussion on the matter. I ask that you please read the latest version of
my cardinal pairwise paper before engaging me in discussion about the
method's counter-strategic properties, because the paper goes into great
detail on the topic. Frankly, I didn't really understand cardinal
pairwise's strategic properties very well myself until I started gathering
my thoughts to write that paper; when analyzing it in depth, I found some
things that I did not expect to find. You can read the paper via either
one of these links, which are the "Voting Matters" version, and my own
version, respectively.
http://www.mcdougall.org.uk/VM/ISSUE19/I19P2.PDF
http://fc.antioch.edu/~james_green-armytage/cwp13.pdf
	So, rather than asking whether cardinal pairwise provides the same
resistance to strategy as AERLO and ATLO, I'd like to ask whether AERLO
and ATLO provide the same resistance as cardinal pairwise. Also, I believe
that cardinal pairwise achieves many of its counterstrategic benefits
without any need for conscious counterstrategizing by the voters. Lots of
voters giving Bush a very low rating and Kerry a very high rating makes it
very hard to overrule a Kerry>Bush defeat, but the voters don't all need
to understand majority rule cycles, or anticipate a particular strategic
incursion, in order to guard against an incursion that attempts to
overrule the defeat. The neat thing about cardinal pairwise is that rating
the candidates intuitively is usually a good protection against strategy.
AERLO and ATLO don't have this benefit, because the concept of truncation
lines and equal ranking lines is never going to be as intuitive to voters
as ranking the candidates on a scale from 0 to 100. 

you wrote:
>
>Of course Weighted Pairwise is more complicated than ordinary wv, and so
>an 
>initial Condorcet proposal should probably be ordinary unenhanced wv.
>AERLO 
>& ATLO could then easily be proposed later, as needed, as add-ons.
>If that later proposal were  Weighted Pairwise, a more complicated
>different 
>method, it probably would be more difficult to get acceptance for.

	Yes, I can't avoid the fact that cardinal pairwise is pretty darn
complicated, compared to plurality, compared to IRV, and compared to
regular ranked pairs. I think that I appreciate the absurdity of trying to
convince people to adopt it for presidential elections in the current
political climate, because there is such a strong current of
anti-intellectualism, and it would probably be hard to get most
congresspeople to pay attention for long enough to explain why cardinal
pairwise is a good idea. I guess that it's more of a very-long term
proposal.
	However, in the medium-long term, there is still a question of whether
ordinary winning votes is a good idea for major offices, i.e. president,
governor. I'm not sure that either of us want to go into this issue again
(we've both probably expressed most of our arguments already, and I
remember that we reached a frustrating deadlock on the issue in May of
this year), but my feeling is that the burying strategy might be too big
of a problem in unenhanced WV to use it on this scale. Maybe we can agree
to disagree on that one, and wait until we get more empirical evidence
from Condorcet methods used on a smaller scale. 
	I've been toying with the idea that Condorcet completed by IRV (choosing
from the minimal dominant set) might be slightly more acceptable from this
standpoint, but that has problems too, of course... I guess if that one
turned out to be okay, I would imagine the following progression for
single winner elections:

1. plurality or runoffs 
2. equal-rankings IRV 
3. minimal dominant set completed by IRV.
4. cardinal pairwise??

	I don't know; it seems like something's still missing. What do you think?

you wrote:
>
>Maybe the AERLO & ATLO options, with Weighted Pairwise, would result in
>more 
>improvement than either enhancement by itself.

	I don't know. Maybe. I'll have to think about that one.

-------------------------
           12/10
-------------------------
>
>AERLO & ATLO are transparently purposeful & simple. 

	Simple mathematically, maybe... but I think they will not be as easy to
explain and sell to ordinary voters. Although it is nice for voters to
know how cardinal-weighted pairwise (CWP) works, if they are too lazy to
learn it, they'll still be able to get a good effect by rating the
candidates intuitively. Every numerically literate person is generally
comfortable with rating things on a scale from 0 to 100. With AERLO and
ATLO, it seems that it might be a bit hard to use them without
understanding a whole lot about cycles and the burying strategy, or being
instructed by a party boss. 

>CP
>requires strategic extreme point assignments.

	In what sense does it require them? Don't forget about the maximizing in
scale provision, which automatically rates your favorite candidate in the
minimal dominant set at 100, your least favorite in the minimal dominant
set at 0, and expands the gaps between the remaining candidates in the MDS
in proportion. (e.g. 50, 20, 10 becomes 100, 25, 0.) I'm not convinced
that a non-extreme set of ratings is always a mistake, strategically. I
usually have extreme ratings in my examples because they're supposed to
deal with situations where there is an intense ideological cleft, e.g.
between Dean/Kerry and Bush. If you have a set of candidates where you
don't feel that there is such a hard division, then I don't think that
evenly-spaced ratings would be such a bad bet.

-----------------------
         12/10
-----------------------

>> 	So, rather than asking whether cardinal pairwise
>> provides the same
>> resistance to strategy as AERLO and ATLO, I'd like
>> to ask whether AERLO
>> and ATLO provide the same resistance as cardinal
>> pairwise. 
>
>One could say it either way.
>
>It appears as if they do.
>
	I think the advantage of cardinal pairwise is that it usually prevents
the most severe strategic incursions without any need for conscious
counter-strategizing. I explain the reasoning behind this assertion in the
paper.

	There are two versions, one short and one long. The short is more up to
date than the long, so you will probably want to read the short, although
in some places you can refer to the long if you feel that the arguments in
the short are too abbreviated.
fc.antioch.edu/~james_green-armytage/cwp13.pdf
fc.antioch.edu/~james_green-armytage/cwp21.pdf

-----------------------
         12/13
-----------------------
>
>No, one merely puts the AERLO or ATLO line below the
>candidates to whom one wants to give special
>protection. 

	Okay, fair enough, that could be one way to explain it.

>AERLO, it seems to me is the more useful,
>and maybe, for simplicity, it should be the one
>recommended for general purpose. ATLO gets rid of the
>co-operation/defection dilemma that you described.

	Which one? I thought that was a problem in approval, not WV... perhaps
you can refresh my memory.
>
>> >CP
>> >requires strategic extreme point assignments.
>> 
>> 	In what sense does it require them? 
>
>...in order to gain the criterion compliances that you
>spoke of.

	You mean the 'stronger strong defensive strategy criterion', right? Yes,
that compliance does rely on the possibility of extreme ratings, but I'm
not touting SSDSC as the main counterstrategic advantage of cardinal
pairwise... rather, it's just something that I noted as a "by the way"
point in favor of the method. The main counterstrategic advantages are
summarized in the cardinal pairwise paper. And those advantages do not
require extreme ratings.
>
>As I said, it seems to me that, most likely, CP gives
>protection options like those of AERLO & ATLO. 

	I agree that it offers similar protection. My feeling is that it's more
flexible, fluid, intuitive for voters, etc., plus it allows voters more
detailed expression and resolves sincere cycles in a more meaningful way.
>
>CP, like AERLO & ATLO, is for later, when
>Condorcet(wv) has already been adopted, and when
>voters begin to perceive the desirability of
>strategy-improvement enhancement.
>
>Then, the purposefully transparent AERLO & ATLO
>add-ons seem a simpler proposal than a new count rule.
>
	I'll disagree with you here, for reasons stated above, in previous notes,
and in the paper. In short, I think that cardinal pairwise offers more
advantages and is more elegant than the AERLO and ATLO add-ons.

my best,
James Green-Armytage