[EM] Condorcet strategy and weighted pairwise method

[Dave Ketchum]

On Tue, 20 Jul 2004 03:35:00 -0400 James Green-Armytage wrote:

> James Green-Armytage here, replying to Dave Ketchum.
> 
> 
>>Is this method worth the pain?
>>
> 
> 	There is no pain involved. You are simply giving the voters the option of
> supplementing their ranking info with ratings. If they don't feel like
> doing that, it's fine. If voters fill out the rankings but not the
> ratings, then the system can easily assign default ratings to the ballot,
> by giving the highest candidate(s) 100, the lowest candidate(s) 0, and
> evenly spacing the rest of the ratings gaps. 


The method cannot avoid introducing pain.  While I have the option of 
refusing to do ratings:
      I cannot do that intelligently without understanding the option.
      Since other voters could use the feature, I need to understand what 
they could do to me with it.
      If I get in deep enough, I start wondering how default ratings 
interact with truncation (I see that you do describe one variation - 
bullet voting).


> 	Voters can even bullet vote if they want to, just like they can in
> regular Condorcet. That way, you would be voting your favorite in first
> place with a rating of 100, with everyone else tied for last and with a
> rating of 0.
> 	When you talk about "is this method worth X", I think the main thing is
> just to spring for a computerized voting system / interface (that is
> secure and leaves a paper trail, dammit!), and that supports both ratings
> and rankings in a snazzy, non-baffling way. And my answer to the question
> is, sure, it's a small price to pay.


Let's see:
      "small" requires comparing benefit vs cost, and I remain suspicious 
on both ends.
      Argue paper trail another day.
      That system should be computerized - this is the part I STRONGLY 
support.

> 
> 
>>     Does not matter unless you have more than two candidates evenly 
>>matched enough to produce cycles - often enough to justify the pain 
>>(because cycles can happen they must be attended to - question is how
>>much 
>>complexity to build in).
>>
> 
> 	If you don't have more than two viable candidates, then your election is
> not very exciting. For elections like that you can use IRV or two-round
> runoff and things will turn out fine. But I'd much rather have elections
> with more than two viable candidates. Keep in mind that in close
> multicandidate elections you might get insincere cycles as well as sincere
> cycles... so I'd be relatively cautious as to how evenly-matched
> candidates have to be before I want this method instead of rankings-only
> Condorcet.


You about have to decide what method will be used before you know how 
exciting any particular election will be - many have only one viable 
candidate - on the other side we had three viable candidates for NY 
governor two years ago (out of around ten total).

It is true that more excitement should result from ranked ballots - but 
even this cannot guarantee more than one viable candidate will turn up.

> 
>>     Condorcet precinct results are an array, with the arrays summable 
>>for the whole district or any subdistrict.  IRV is not so simple.
>>
> 
> 	I think it is summable, probably, but to be honest, the summability
> criteria seems like a bit of a red herring to me. Whether you use IRV or
> Condorcet, when people ask for the results, they are going to want the
> full results, that is, not just the matrix, but how many people voted a
> particular preference ranking. So why does it matter that you can produce
> a summable matrix for Condorcet? Anyway, it seems like you are
> underestimating the power of contemporary (and future, since this method
> won't be implemented for awhile) information technology. I think that the
> full rankings & ratings information for any given precinct is really not a
> big deal when you put it in a digital format. For example, you could
> probably fit it all on a DVD or something like that.

With plain Condorcet, each precinct does a matrix that records, for example, 
how many voters rank Green over WF, and how many WF over Green.  If precincts 
report to counties, county simply adds the matrices together, needing nothing 
more.  Likewise for county reporting to state.  This much effort is doable at 
the end of election day, just as it is for Plurality.  When absentee and other 
counts come in later, they are simply added to the totals.

With IRV, voter's first choices can get summed by state but, when a loser is 
recognized, voting patterns are needed to do the substitution.  These patterns 
could have been sent up by the precincts (added load - NY has a few million 
ballots for governor), or could be requested as needed by state (computers 
with this data must stay available).

With your method...?

True, immediate demand is about who won, and how close it was.  Later 
there will be desire to look at details such as which counties liked Green or 
WF best.

Also true that each precinct should do a detailed record on DVD or equivalent.  
BUT, I would expect more desire for county results that are both detailed 
and of manageable size for human analysis.

> 	I do admit that it's going to be pretty hard to report the full results
> of a weighted pairwise election in any kind of succinct way. The full
> results can and must be made available, but they would hardly make for
> nice bedtime reading. Of course, just as with ranked ballot methods, you
> could digest them in different ways to bring forward the kind of numbers
> that people can look at without their heads exploding.
> 	Anyway, I haven't really answered your question. To be honest I'm not a
> big computer science guy, so I'm not sure I know what I'm talking about,
> but my best guess is that it is summable. You can have one summable array
> for the pairwise comparisons, and then maybe a second summable array for
> the ratings differentials. Then, if you put them together you can
> determine the winner? 

This math is no challenge for computers.  Large volumes of data have to demand 
storage space, time to move them around, and time to process - none of these 
are worth noticing for one data item BUT, get enough items and it adds up.

> 	Hmm. I know that Brian Olson implemented a simple version of the weighted
> pairwise method in his election calculator.
> http://bolson.org:8080/v/vote_form.html Plug in some ratings, indicate
> "ratings" in the box up above, and when you get to the results, scroll
> down to where it says "Pairwise Rating Differential Summation". I don't
> really know how he did the code, but the way he has it, it looks summable
> to me.
> 
Brian is almost certainly doing the equivalent of a precinct, where almost 
anything would be doable - and the doing could be useful to his audience.  
I am leaning on the problems of multiple precincts, such as states.
> 
> my best,
> James 

-- 
  davek at clarityconnect.com    people.clarityconnect.com/webpages3/davek
  Dave Ketchum   108 Halstead Ave, Owego, NY  13827-1708   607-687-5026
            Do to no one what you would not want done to you.
                  If you want peace, work for justice.