Voting Benchmark
Hello,
I am implementing a new election method, after initially playing with
Schulze voting. In the process I really want to compare my method to
Schulze and other methods according to several types of quality. Is
there a good benchmark currently, that I can start from?
Marijn
Good question. My opinion is that there are lots of opinions, targets and criteria, but no agreed benchmark that you could use. You mentioned your intention to compare your method to other well known methods. I think that is the correct approach.
You can find many discussions on this mailing list on what methods and properties are good. You can find also many definitions and criteria that people have collected at electorama.com. You can also send a description of the method to this mailing list. I'm sure people will point out possible benefits and problems of the proposed method, and that way you will quickly find out what properties and criteria people consider most relevant in this case.
Juho
On 30 Sep 2015, at 12:12, Marijn Stollenga m.stollenga@gmail.com wrote:
Hello,
I am implementing a new election method, after initially playing with Schulze voting. In the process I really want to compare my method to Schulze and other methods according to several types of quality. Is there a good benchmark currently, that I can start from?
Marijn
Election-Methods mailing list - see http://electorama.com/em for list info
Hi Marijn,
I think you might start with the criteria that motivated Schulze, or criteria that differentiate popular methods.
For instance:Schulze: clone independence, monotonicity (mono-raise), Condorcet, SchwartzIRV: clone independence, later-no-harm, later-no-helpApproval: participation, favorite betrayal (FBC)
Unfortunately it's not necessarily straightforward to demonstrate that a new method satisfies or doesn't satisfy some criterion.
Some of us have simulations to attempt to gauge the incentives to use specific strategies (e.g. compromise, truncation, burial). But since simulations are based on the programmer's assumptions, nothing is standardized.
My hunch (based on my own experiments from 10+ years ago) is that if Schulze is your original inspiration or standard, and you make your own rank method, you may make something that gives similar results to Schulze, but most likely it won't satisfy all of Schulze's technical criteria.
Kevin
De : Marijn Stollenga m.stollenga@gmail.com
À : "election-methods@lists.electorama.com" election-methods@lists.electorama.com
Envoyé le : Mercredi 30 septembre 2015 4h12
Objet : [EM] Voting Benchmark
Hello,
I am implementing a new election method, after initially playing with
Schulze voting. In the process I really want to compare my method to
Schulze and other methods according to several types of quality. Is
there a good benchmark currently, that I can start from?
Marijn
Election-Methods mailing list - see http://electorama.com/em for list info
Yes, actually my main motivation was the strange properties in Schulze
that I didn't like.
Firstly, the propagation of votes through beaten path sounds
interesting, but it is also pretty strange. By voting a dominance of A
over B, you might end up voting for C if some large other group votes C
over A a lot, even if you don't like C at all. This leads to some
complex tactical voting.
Secondly, the votes only flow if A over B has more votes than B over A
say. On first sight this sounds good but introduces a sensitivity in the
method that can completely flip results with a small change of votes,
also leading to possible loss of votes and tactical voting. Why can't
votes simply flow both directions? In my implementation I tried it and
it still led to dominant results in my experiments, but I guess it's
needed for certain properties?
Any thoughts?
Marijn
On 30/09/15 14:40, Kevin Venzke wrote:
Hi Marijn,
I think you might start with the criteria that motivated Schulze, or
criteria that differentiate popular methods.
For instance:
Schulze: clone independence, monotonicity (mono-raise), Condorcet,
Schwartz
IRV: clone independence, later-no-harm, later-no-help
Approval: participation, favorite betrayal (FBC)
Unfortunately it's not necessarily straightforward to demonstrate that
a new method satisfies or doesn't satisfy some criterion.
Some of us have simulations to attempt to gauge the incentives to use
specific strategies (e.g. compromise, truncation, burial). But since
simulations are based on the programmer's assumptions, nothing is
standardized.
My hunch (based on my own experiments from 10+ years ago) is that if
Schulze is your original inspiration or standard, and you make your
own rank method, you may make something that gives similar results to
Schulze, but most likely it won't satisfy all of Schulze's technical
criteria.
Kevin
De : Marijn Stollenga m.stollenga@gmail.com
À : "election-methods@lists.electorama.com"
election-methods@lists.electorama.com
Envoyé le : Mercredi 30 septembre 2015 4h12
Objet : [EM] Voting Benchmark
Hello,
I am implementing a new election method, after initially playing with
Schulze voting. In the process I really want to compare my method to
Schulze and other methods according to several types of quality. Is
there a good benchmark currently, that I can start from?
Marijn
Election-Methods mailing list - see http://electorama.com/em
http://electorama.com/emfor list info
Hi Marijn,
Yes, a few thoughts:1. The path tracing (propagation of votes through paths) may seem odd, and it does create problems if you want to create guarantees about the effect of a given ballot (like LNHarm or FBC). But path tracing can provide a useful framework for resolving clone situations or Condorcet cycles. I'd note also that the problem of "voting A>B can help C" is a problem common to many or most election methods.2. Regarding sensitivity to a small change in votes: I would suggest that this is true even with a two-way FPP election: a contest's win vs. loss is the most important thing, and considering the strength of the win at all is a novelty. But it's an interesting question, whether the single-direction flow of votes creates more tactical voting opportunities.3. I call Schulze with votes flowing in both directions "Schulze(pairwise opposition)." The results can be less decisive. For example, suppose that candidate A has 45% of the vote (bullet votes, no lower preferences), and the rest of the votes are some combination of B, C, B>C, and C>B votes. Say there is an A>C>B>A cycle. It's likely that B and C will tie for the win because each received their greatest opposition from A. Whatever tiebreaker rule is used, if it elects C, and C has fewer preferences in total than A's 45% in first preferences, that would violate the Plurality criterion. Some of us would find that to be a bad outcome (i.e. hard to defend), so you may want to consider it.
Kevin
De : Marijn Stollenga <m.stollenga@gmail.com>
À : Kevin Venzke stepjak@yahoo.fr; "election-methods@lists.electorama.com" election-methods@lists.electorama.com
Envoyé le : Jeudi 1 octobre 2015 4h21
Objet : Re: [EM] Voting Benchmark
Yes, actually my main motivation was the strange properties in Schulze that I didn't like.
Firstly, the propagation of votes through beaten path sounds interesting, but it is also pretty strange. By voting a dominance of A over B, you might end up voting for C if some large other group votes C over A a lot, even if you don't like C at all. This leads to some complex tactical voting.
Secondly, the votes only flow if A over B has more votes than B over A say. On first sight this sounds good but introduces a sensitivity in the method that can completely flip results with a small change of votes, also leading to possible loss of votes and tactical voting. Why can't votes simply flow both directions? In my implementation I tried it and it still led to dominant results in my experiments, but I guess it's needed for certain properties?
Any thoughts?
Marijn
On 30/09/15 14:40, Kevin Venzke wrote:
Hi Marijn,
I think you might start with the criteria that motivated Schulze, or criteria that differentiate popular methods.
For instance: Schulze: clone independence, monotonicity (mono-raise), Condorcet, Schwartz IRV: clone independence, later-no-harm, later-no-help Approval: participation, favorite betrayal (FBC)
Unfortunately it's not necessarily straightforward to demonstrate that a new method satisfies or doesn't satisfy some criterion.
Some of us have simulations to attempt to gauge the incentives to use specific strategies (e.g. compromise, truncation, burial). But since simulations are based on the programmer's assumptions, nothing is standardized.
My hunch (based on my own experiments from 10+ years ago) is that if Schulze is your original inspiration or standard, and you make your own rank method, you may make something that gives similar results to Schulze, but most likely it won't satisfy all of Schulze's technical criteria.
Kevin
De : Marijn Stollenga m.stollenga@gmail.com
À : "election-methods@lists.electorama.com" election-methods@lists.electorama.com
Envoyé le : Mercredi 30 septembre 2015 4h12
Objet : [EM] Voting Benchmark
Hello,
I am implementing a new election method, after initially playing with
Schulze voting. In the process I really want to compare my method to
Schulze and other methods according to several types of quality. Is
there a good benchmark currently, that I can start from?
Marijn
Election-Methods mailing list - see http://electorama.com/em for list info