Question re MJ

[Stephen Unger]

Consider the following election, with scores 0-9.

#voters..A..B
1000....9..6
1.......5..6
1000....5..0

I believe that B wins the MJ election: median(A)=5 median(B)=6.

Is that correct?

Steve
............

Stephen H. Unger
Professor Emeritus
Computer Science and Electrical Engineering
Columbia University
............

[Jameson Quinn]

2013/5/25 Stephen Unger <un...@cs.columbia.edu>

Consider the following election, with scores 0-9.

Part of the definition of MJ is that it doesn't use numeric scores. Also, it tends to use 5-7 categories, not 10. But OK; that's not important.

#voters..A..B
 1000....9..6
 1.......5..6
 1000....5..0

I believe that B wins the MJ election: median(A)=5 median(B)=6.

Right. This is also an election in which not a single voter's ballot spans even half of the full range. 2001 voters like that, without a single exception, is literally about as likely as a meteor hitting the ballot counting process.

It is theoretically possible for the MJ winner to not equal the Condorcet winner. But in real elections, that will never happen unless the MJ winner was within a couple of percentage of beating the CW pairwise, AND there is a dissimilar rival threatening both candidates. In other words, the real-world BR penalty for this kind of failure is very small.

You want me to give some implausible yet superficially shocking scenarios for your favorite system? Because no matter what that system is, I guarantee I can do that.

Jameson

[Warren D Smith]

> Part of the definition of MJ is that it doesn't use numeric scores. Also,
> it tends to use 5-7 categories, not 10. But OK; that's not important.
>
>>
>> #voters..A..B
>> 1000....9..6
>> 1.......5..6
>> 1000....5..0
>>
>> I believe that B wins the MJ election: median(A)=5 median(B)=6.
>>
>
> Right.

--yes, and various MJ example elections of this same general nature,
but more realistic, for example with all voters using full range, are
stated in
http://rangevoting.org/MedianVrange.html

> It is theoretically possible for the MJ winner to not equal the Condorcet
> winner.

--which happened in France 2007, apparently:
http://rangevoting.org/France2007.html

> But in real elections, that will never happen unless the MJ winner
> was within a couple of percentage of beating the CW pairwise, AND

--Sorry, in France 2007, the pairwise victory of Bayrou over (the MJ
winner) Sarkozy was apparently 55:45 or 54:46.

> there is
> a dissimilar rival threatening both candidates.

--Segolene Royal?

> In other words, the
> real-world BR penalty for this kind of failure is very small.

--Jameson Quinn is being rather cavalier here. I'm unsure he's right.
Revisit the examples in the first part of

http://rangevoting.org/MedianVrange.html

In my opinion though, the idea that MJ maybe can substantially avoid
the "Burr dilemma" problem gives MJ a lot of new energy, and an
exciting research prospect that JQ might want to undertake (in my
dreams) is to figure out how to model "partly strategic" voters,
modify IEVS while at the same time cleaning it up and making it nicer
and more user-friendly, and then rerun IEVS and see what happens re
Bayesian regret. Conclusions about MJ, or about something else, might
change.

Also, the fact I think we understand why MJ had that problem in France 2007
makes you wonder if there is a way to fix it. [I haven't thought of
any ways to fix MJ except ugly ones that make it even more
disgustingly complicated, but...]

[Jameson Quinn]

2013/5/25 Warren D Smith <warre...@gmail.com>

> Part of the definition of MJ is that it doesn't use numeric scores. Also,
> it tends to use 5-7 categories, not 10. But OK; that's not important.
>
>>
>> #voters..A..B
>>  1000....9..6
>>  1.......5..6
>>  1000....5..0
>>
>> I believe that B wins the MJ election: median(A)=5 median(B)=6.
>>
>
> Right.

--yes, and various MJ example elections of this same general nature,
but more realistic, for example with all voters using full range, are
stated in
   http://rangevoting.org/MedianVrange.html

> It is theoretically possible for the MJ winner to not equal the Condorcet
> winner.

--which happened in France 2007, apparently:
http://rangevoting.org/France2007.html

Using the numbers in your graphs on that page, Bayrou is the winner. I think it's that you're using the Orsay numbers, not the corrected pseudo-national numbers. I'll try to fix that, because honestly without seeing the pseudo-national numbers, I don't really understand what's happening there.

So I guess I'm saying: right now I think I may have been wrong on this point, but I'm reserving the right to keep arguing it if when I look more closely I decide I wasn't wrong.

> But in real elections, that will never happen unless the MJ winner
> was within a couple of percentage of beating the CW pairwise, AND

--Sorry, in France 2007, the pairwise victory of Bayrou over (the MJ
winner) Sarkozy was apparently  55:45 or 54:46.

> there is
> a dissimilar rival threatening both candidates.

--Segolene Royal?

> In other words, the
> real-world BR penalty for this kind of failure is very small.

--Jameson Quinn is being rather cavalier here.  I'm unsure he's right.
Revisit the examples in the first part of

   http://rangevoting.org/MedianVrange.html

In my opinion though, the idea that MJ maybe can substantially avoid
the "Burr dilemma" problem gives MJ a lot of new energy, and an
exciting research prospect that JQ might want to undertake (in my
dreams) is to figure out how to model "partly strategic" voters,
modify IEVS while at the same time cleaning it up and making it nicer
and more user-friendly, and then rerun IEVS and  see what happens re
Bayesian regret.  Conclusions about MJ, or about something else, might
change.

I also look forward to doing something like this, though frankly I would probably rather rewrite it from scratch in python or coffeescript than fix it in C. If I do it in coffeescript, it could potentially be put on a live graphical web page where people could play with the settings, something vaguely akin to Ka-Ping Yee's Voteline thingy. So I think that would make it worth rewriting. (I also suspect Warren could easily learn enough coffeescript to read and comment on the code. It's quite a readable language overall.)

Also, the fact I think we understand why MJ had that problem in France 2007
makes you wonder if there is a way to fix it.  [I haven't thought of
any ways to fix MJ except ugly ones that make it even more
disgustingly complicated, but...]

Interesting thought, but I'd certainly have to work out the corrected numbers. It's not impossible, actually, that GMJ already fixes the problem. But rather than speculate, I should check.

 

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Warren D Smith]

--sorry; The http://rangevoting.org/France2007.html
page includes the raw Orsay numbers at table on top and with them
Bayrou wins. But later on
the page discusses the correction to all France, causing Sarkozy to win.
Also, later on this page describes 33 different pairwise Bayrou vs
Sarkozy professional
France-wide polls. Bayrou won them in aggregate by 55 or 54 versus 45 or 46.
Bayrou's % varied from about 51 to 60 depending which pairwise poll.

> I also look forward to doing something like this, though frankly I would
> probably rather rewrite it from scratch in python or coffeescript than fix
> it in C. If I do it in coffeescript, it could potentially be put on a live
> graphical web page where people could play with the settings, something
> vaguely akin to Ka-Ping Yee's Voteline

> <http://zesty.ca/voting/voteline/>thingy.

--program needs to be fast. Python is not fast (or wasn't back when I
wrote IEVS).
C is fast. If you want good BR stats and/or good Yee pictures, you
need big speed.

I don't think the problem is the C language, I think the top problem is the
poor user interface stuff because I was lazy and never made that
decent (and it had bugs).
If D becomes popular you could change to D,
which is a better language than C but very similar.
I'd stay away from coffeescript etc, whatever that is, doesn't matter, because
a top goal is to create a program which (a) will still be usable in 15
years after coffeescript evaporates and (b) will be readily modifiable
by others who don't know much
about programming.

I've heard a lot of people saying they were experts and they knew for
sure language X would never die. Mostly they were wrong and their
wrongness wasted my damn time
and prevented me from making a program with lasting value.
The only programs with lasting value & long-life are programs written in
simple and stable languages like C. I admit I'm somewhat sheltered,
but I am unaware of even a single exception.

[Jameson Quinn]

Coffeescript is just a slightly easier way to write javascript. Javascript (js) is installed on more consumer devices now than any other language, including 8086 machine code, and certainly including C. It's not going anywhere. It's also reasonably fast, thanks to Google's optimization work. It's perhaps a factor of 2 or 3 slower than C, but since you wrote IEVS, computers in general have sped up by at least a couple orders of magnitude, so js could easily match the speed you got back then.

and (b) will be readily modifiable
by others who don't know much
about programming.

A coffeescript program is far, far easier to modify than a C one for anyone not specifically versed in either language.

I've heard a lot of people saying they were experts and they knew for
sure language X would never die.  Mostly they were wrong and their
wrongness wasted my damn time
and prevented me from making a program with lasting value.
The only programs with lasting value & long-life are programs written in
simple and stable languages like C.   I admit I'm somewhat sheltered,
but I am unaware of even a single exception. 

A javascript program, written today using coffeescript, could run on the average computer from 15 years ago, without modification, recompiling, or installing a single piece of software that wasn't on that computer at the time. That could possibly extend to 20 years in some cases. That is more than you can say for C, which usually requires recompiling for each different OS or platform it is to run on, and is not immune from bitrot due to incompatible libraries and the like.

Yes, I know, C has been around for far more than 20 years, but since 20 years is already an eternity for computers, I think my point is good.

[Warren D Smith]

A lot of people have told me that kind of stuff, a lot more expert
than either of us, and they were all wrong... also, a fair # people
said they'd redo IEVS better from scratch, and they all didn't. A
fair # said they'd fool with IEVS itself, and about 50% actually did.



--
Warren D. Smith
http://RangeVoting.org <-- add your endorsement (by clicking
"endorse" as 1st step)

[Abd ul-Rahman Lomax]

At 05:25 PM 5/25/2013, Jameson Quinn wrote:

>Right. This is also an election in which not a single voter's ballot
>spans even half of the full range. 2001 voters like that, without a
>single exception, is literally about as likely as a meteor hitting
>the ballot counting process.

I don't know. Meteor strikes are not all that uncommon.