On Sun, Feb 1, 2009 at 6:25 AM, brokenladdercalendar
<thebrokenladder@...> wrote:
> --- In RangeVoting@yahoogroups.com, Raph Frank <raphfrk@...> wrote:
>> Looks like
>> election ID is which type of ballot
>
> no, it has nothing to do with "type of ballot". it's actual unique
> election, associated with a given date. recently in san francisco, we
> had an election that included the office of president, as well as
> several ballot initiatives, and some representatives. that would be an
> example of a particular "election" that would have its own "election ID".

Ahh, ok, I should have checked the dates.  However, it is looking that
electionID 1 occurs at more than 1 date.

I had it viewed that electionID 1 meant a list of offices that you
could vote for on the same ballot.

--- On Sun, 1/2/09, Raph Frank <raphfrk@...> wrote:

> > Also he pointed out Holland is districtless, all the
> votes
> > are countrywide. He feels nobody represents his
> geographic district
> > but since Holland is small & pretty homogeneous,
> this defect is not
> > that serious relative to how it would be in some
> larger and more
> > diverse country. I said that Ireland has districts.
> His response
> > was that then you get much less good proportionality
> since
> > only 3 MPs from (say) a 3-MP district.
>
> We have between 3 and 5 seat constituencies.  I would agree
> that it
> gives less accurate PR.  I think a rule where the average
> must be at
> least 5 would be good.

Yes, 5 seats may be enough for a working
PR system.

Finland plans to make a reform since the
number of seats in the smallest districts
(excluding Åland with one seat) has dropped
to 6. There are also larger districts (max
34) and that makes the difficult position
of the smallest parties in the smallest
districts clearly visible.

The reform plan is to count the proportionality
at national level and then distribute those
seats down to the districts. Currently alliances
between parties are supported in the districts
but that system is one step short of perfect.
Unfortunately the reform plan contains also a
proposal to have thresholds (did not exist
before) and some other weird ideas.

Finland uses D'Hondt that favours large parties.
The size of the districts is however a stronger
factor favouring large parties. Duverger's law
thus influences the election process already
with districts of size 6. The plan is to keep
the D'Hondt method partly due to the long
tradition of using it (ok, better get the
district sizes and their bias straight first).

One of the basic questions in many PR
countries/systems is if already one quota of
votes (at national level or per district) will
give you one seat. Most systems set some
limits to this.

(Finland btw has open lists, which means that
the problems of parties deciding who gets a
seat and need of politicians to be loyal to
the party leaders in the hope of getting a
free seat are not that big problems. There is
party internal discipline and party based
block voting on many questions though.)

Juho

On Mon, Feb 2, 2009 at 12:08 AM, Juho Laatu <juho4880@...> wrote:
> The reform plan is to count the proportionality
> at national level and then distribute those
> seats down to the districts. Currently alliances
> between parties are supported in the districts
> but that system is one step short of perfect.
> Unfortunately the reform plan contains also a
> proposal to have thresholds (did not exist
> before) and some other weird ideas.

It would still be party list based, i.e. each voter votes for 1
candidate/party, not separate national and district votes?

There doesn't seem to be an easy way of handling the distribution.  Is
the plan to do it something like:

1) Determine the number of seats each party is entitled to (via
d'Hondt using the national totals)

2) in each district work out the district divisor for each party
i.e.  (party's district vote / (seats assigned to the party in the
district + 1) )

3) Assign a seat to the party with the highest district divisor in any
of the non-full districts.  Parties that have already received their
full allocation of seats are not eligible for more seats.

Thus, the process would work exactly as normal, except that once a
party receives its full allocation of seats, it isn't allowed receive
anymore.

There could still be issues with deadlock.  For example, a party was
entitled to 1 more seat, but all of its non-elected candidates were in
districts where all the seats were already assigned.

As long as all parties run candidates in all the districts, this isn't
an issue.

Also, deadlock wouldn't happen until all of the other parties have
received their full seat allocation.  It would be possible to just
give the remaining seats to the parties that have been locked out,
without reference to their district.  Ofc, that means that some
districts would be under-represented.

Another option would be to assign seats starting from the smallest
party and working upwards.  Large parties are likely to run sufficient
candidates not to have to worry about deadlock.

> Finland uses D'Hondt that favours large parties.
> The size of the districts is however a stronger
> factor favouring large parties. Duverger's law
> thus influences the election process already
> with districts of size 6. The plan is to keep
> the D'Hondt method partly due to the long
> tradition of using it (ok, better get the
> district sizes and their bias straight first).

Duverger's law is generally only applied to single seaters.  But there
is a similar effect towards larger parties/blocs with small
constituencies.

I guess the multi-seat equivalent is that parties smaller than
1/(average constituency size) tend not to win any seats.

> One of the basic questions in many PR
> countries/systems is if already one quota of
> votes (at national level or per district) will
> give you one seat. Most systems set some
> limits to this.

I assume the plan is to guarantee national proportionality based on 1
person/1 vote.  If it is raining in a district, the voters in that
district will have less effect on the national result due to lower
turnout.

This could be compensated by rescaling the vote totals, i.e.

multiply the vote totals by the ratio of the population of the
district divided by the number of voters who cast a vote in that
district.

Also, there is the question of what 'population' should mean.  Does it
mean the voting population, the citizen population or does every
resident count.

core support

35% A > C > B > D
17% B > others
32% C > B > others
16% D > B > others

In this 4-candidate election, Instant Runoff Voting selects candidate
B as the winner, beating A in the final round, 65% to 35%.

But wait!

A huge 67% majority of voters would rather have candidate C than
candidate B.

And candidate C received nearly twice as many first-place votes as
candidate B, 32% to 17%.

So the claim that IRV "elects majority winners" is seriously misleading.

also...

1) candidate A is a spoiler.
2) the first row of voters have an incentive to betray candidate A,
instead of vote sincerely.

but that was obvious of course...

--- In RangeVoting@yahoogroups.com, "brokenladdercalendar"
<thebrokenladder@...> wrote:
> 35% A > C > B > D
> 17% B > others
> 32% C > B > others
> 16% D > B > others
>
> In this 4-candidate election, Instant Runoff Voting selects candidate
> B as the winner, beating A in the final round, 65% to 35%.
>
> But wait!
>
> A huge 67% majority of voters would rather have candidate C than
> candidate B.
>
> And candidate C received nearly twice as many first-place votes as
> candidate B, 32% to 17%.
>
> So the claim that IRV "elects majority winners" is seriously misleading.
>

--- In RangeVoting@yahoogroups.com, "brokenladdercalendar"
<thebrokenladder@...> wrote:
> 35% A > C > B > D
> 17% B > others
> 32% C > B > others
> 16% D > B > others

doh! change around b and d in the first row:

35% A > C > D > B
17% B > others
32% C > B > others
16% D > B > others

now d is also preferred by a majority to the winner b (although by a
narrower 51% majority).

okay, i think i finally figured out how to make this example as
illustrative as possible, tweaking it just a bit:

core support

35% A > C > D > B
17% B > C > D > A
32% C > D > B > A
16% D > B > C > A

In this 4-candidate election, Instant Runoff Voting selects candidate
B as the winner, beating A in the final round, 65% to 35%.

But wait!

A huge 67% majority of voters would rather have candidate C than
candidate B. And candidate C received nearly twice as many first-place
votes as candidate B, 32% to 17%. And an even larger 83%
super-majority of voters would rather have candidate D than B (and D
got just a little fewer first-place votes than B.) So the claim that
IRV "elects majority winners" is seriously misleading.

Also...
- A is a spoiler (if he would drop out of the race, C would win
instead of B).
- The first row of voters have an incentive to betray candidate A by
pretending candidate C is their actual favorite - then they get their
second choice instead of their last.
- The third row of voters have an incentive to betray candidate C by
pretending candidate D is their favorite - then they get their their
second choice instead of their third.

Also, C is the Condorcet winner, but doesn't even make it to the final
round:
65% C > A
67% C > B
84% C > D

And A is the Condorcet loser, but makes it to the final round:
65% others > C

--- On Mon, 2/2/09, Raph Frank <raphfrk@...> wrote:

> On Mon, Feb 2, 2009 at 12:08 AM, Juho Laatu
> <juho4880@...> wrote:
> > The reform plan is to count the proportionality
> > at national level and then distribute those
> > seats down to the districts. Currently alliances
> > between parties are supported in the districts
> > but that system is one step short of perfect.
> > Unfortunately the reform plan contains also a
> > proposal to have thresholds (did not exist
> > before) and some other weird ideas.
>
> It would still be party list based, i.e. each voter votes
> for 1
> candidate/party, not separate national and district votes?

Yes, each vote is a vote to one of the
candidates of one of the districts.

>
> There doesn't seem to be an easy way of handling the
> distribution.  Is
> the plan to do it something like:

The proposal is available but unfortunately
so far only in Finnish + a summary in
Swedish. (http://www.om.fi/1208352722412)

>
> 1) Determine the number of seats each party is entitled to
> (via
> d'Hondt using the national totals)
>
> 2) in each district work out the district divisor for each
> party
> i.e.  (party's district vote / (seats assigned to the
> party in the
> district + 1) )

The method uses Hare quota. All parties
in all districts will get some number of
sets at once + a fragment of quota to be
handled later.

The traditional approach of allocating a
fixed number of seats to each district
based on the number of citizens in each
district will be kept. There will thus
be (as before) exact "regional PR".

>
> 3) Assign a seat to the party with the highest district
> divisor in any
> of the non-full districts.  Parties that have already
> received their
> full allocation of seats are not eligible for more seats.

The quota fractions will be handled
starting from the largest fraction. If
the party or district seats have already
been filled there are special rules =>
allocate to the next largest fraction of
another party or (if district full) to
next largest fraction in another district.
There's also a backtracking mechanism for
cases where the algorithm reaches a
dead-end.

>
> Thus, the process would work exactly as normal, except that
> once a
> party receives its full allocation of seats, it isn't
> allowed receive
> anymore.

The proposal seems to be not to skip
entries that can not be allocated but to
allocate that seat right away to some
alternative fraction right away (as
described above).

>
> There could still be issues with deadlock.  For example, a
> party was
> entitled to 1 more seat, but all of its non-elected
> candidates were in
> districts where all the seats were already assigned.

There was a backtrack mechanism.

>
> As long as all parties run candidates in all the districts,
> this isn't
> an issue.
>
> Also, deadlock wouldn't happen until all of the other
> parties have
> received their full seat allocation.  It would be possible
> to just
> give the remaining seats to the parties that have been
> locked out,
> without reference to their district.  Ofc, that means that
> some
> districts would be under-represented.

The method is strict with "regional PR".

>
> Another option would be to assign seats starting from the
> smallest
> party and working upwards.  Large parties are likely to run
> sufficient
> candidates not to have to worry about deadlock.

The method started from the largest
fractions.

>
> > Finland uses D'Hondt that favours large parties.
> > The size of the districts is however a stronger
> > factor favouring large parties. Duverger's law
> > thus influences the election process already
> > with districts of size 6. The plan is to keep
> > the D'Hondt method partly due to the long
> > tradition of using it (ok, better get the
> > district sizes and their bias straight first).
>
> Duverger's law is generally only applied to single
> seaters.  But there
> is a similar effect towards larger parties/blocs with small
> constituencies.

Yes, that's what I meant.

>
> I guess the multi-seat equivalent is that parties smaller
> than
> 1/(average constituency size) tend not to win any seats.

Or in each district parties with less
than 1/size support tend not to win
any seats (assuming D'Hondt). It was
essential in this case that the
smallest parties were unable to get
any seats in the smallest districts
(except as part of election specific
alliances) although they could survive
quite well in the largest districts.

>
> > One of the basic questions in many PR
> > countries/systems is if already one quota of
> > votes (at national level or per district) will
> > give you one seat. Most systems set some
> > limits to this.
>
> I assume the plan is to guarantee national proportionality
> based on 1
> person/1 vote.  If it is raining in a district, the voters
> in that
> district will have less effect on the national result due
> to lower
> turnout.

Yes. They will get all their local seats
though, based on the population of that
district.

>
> This could be compensated by rescaling the vote totals,
> i.e.
>
> multiply the vote totals by the ratio of the population of
> the
> district divided by the number of voters who cast a vote in
> that
> district.

Yes, possible and interesting, but not
done in this case.

One interesting approach would be to
allow each district to get seats in
proportion to the number of active
voters (maybe approximately, allowing
some randomness). Maybe people would
vote more eagerly then :-).

>
> Also, there is the question of what 'population'
> should mean.  Does it
> mean the voting population, the citizen population or does
> every
> resident count.

If I recall right seats of each
district are counted based on the
number of citizens registered in
the towns of that district (=>
largest remainder).


I think this proposed reform is a
step forward in most aspects. The
biggest problems are introduction
of thresholds, and introduction of
a district specific threshold (!)
that allows local parties to get
seats even if they don't meet the
national threshold.

The new proposal gets rid of election
+ district specific party alliances.
It does not fix the problem of the
open list method of not supporting
party internal PR properly
(candidates with highest number of
votes elected). Party internal
alliances / trees would fix this
quite nicely in my opinion.

Juho





>
> ------------------------------------
>
> Yahoo! Groups Links
>
>
>

1) Voters score the candidates as with score voting.

2) W = the (maximum) number of winners.

3) We elect the W candidates which produce the highest sum of each
ballot's highest score for any of those W candidates (e.g. if X and Y
are the proposed set of winners, and I voted X=5,Y=0,Z=9, then 5 is
the value used on my ballot).

4) Each winner gets an amount of "votes" equal to the number of
ballots for which he was the favorite on that ballot, out of the set
of winners (and I suppose we evenly split this if more than one winner
tied on a given ballot).

5) Any "winner" who did not actually get any votes is not actually
elected (addresses the problem Warren brought up).

I think this method may fix the problem with RSV where a voter can
hurt himself by casting an honest vote (no-show paradox). It probably
fixes other common PR problems that I cannot think of.

On Mon, Feb 2, 2009 at 11:22 PM, Juho Laatu <juho4880@...> wrote:
>> There doesn't seem to be an easy way of handling the
>> distribution. Is
>> the plan to do it something like:
>
> The proposal is available but unfortunately
> so far only in Finnish + a summary in
> Swedish. (http://www.om.fi/1208352722412)

How inconsiderate, just publishing in their own language :).

> The proposal seems to be not to skip
> entries that can not be allocated but to
> allocate that seat right away to some
> alternative fraction right away (as
> described above).

Whatever the algorithm, some kind of backtracking does seem necessary.

This is a pity, as where possible a simple algorithm is preferred.

> The method is strict with "regional PR".

However, in terms of parties, it isn't.  Candidates will depend on the
locals to actually get elected, but the number of them elected is
determined based on the national vote.

This means that districts with a smaller turnout would end up with
less effect on the national party totals and it is the national party
totals which decides the coalition negotiations.

For example, if there were 2 equal sized districts and 10 seats each
and the results were

District 1
Party A: 40%
Party B: 60%

District 2:
Party A: 70%
Party B: 30%

Under a pure regional vote, the results are

A: 4+7 = 11 seats
B: 3+6 = 9 seats

However, if the turnout in district 2 was 50% of the turnout in
district 1, then the national totals would be

A = 40 + 0.50*70 = 75 votes
B = 60 + 0.50*30 = 75 votes

Thus the lower turnout caused a clear win for party A to become a draw
with 10 seats each

Having said that, the effect is probably not that massive.

I did see some argue that Bush's loss of the popular vote in 2000
could be explained by lower turnouts in the States that actually voted
for him.

>
> One interesting approach would be to
> allow each district to get seats in
> proportion to the number of active
> voters (maybe approximately, allowing
> some randomness). Maybe people would
> vote more eagerly then :-).

Since influence of a district is proportional to its voter turnout,
that kinda already happens.

A district which tends to have lower voter turnout would be considered
less important by the leadership of the various parties.

> If I recall right seats of each
> district are counted based on the
> number of citizens registered in
> the towns of that district (=>
> largest remainder).

In Ireland, it is pure population.  A person voting in a constituency
with a large child or immigrant population would in effect have a more
powerful vote.  Ofc, a person voting in a constituency with a low
voter turnout has the same benefit.

> I think this proposed reform is a
> step forward in most aspects. The
> biggest problems are introduction
> of thresholds, and introduction of
> a district specific threshold (!)
> that allows local parties to get
> seats even if they don't meet the
> national threshold.

I assume that these are being set so that all current parties will
meet the thresholds (coincidently)?

> The new proposal gets rid of election
> + district specific party alliances.
> It does not fix the problem of the
> open list method of not supporting
> party internal PR properly
> (candidates with highest number of
> votes elected). Party internal
> alliances / trees would fix this
> quite nicely in my opinion.

True.  A tree system for open list would be better than plain open
list which is better than closed lists.

Another option would be Asset voting, but the tree system is at least
completely transparent and open.

Ofc, I still prefer PR-STV.

On Mon, Feb 2, 2009 at 11:34 PM, brokenladdercalendar
<thebrokenladder@...> wrote:
> 1) Voters score the candidates as with score voting.
>
> 2) W = the (maximum) number of winners.
>
> 3) We elect the W candidates which produce the highest sum of each
> ballot's highest score for any of those W candidates (e.g. if X and Y
> are the proposed set of winners, and I voted X=5,Y=0,Z=9, then 5 is
> the value used on my ballot).
>
> 4) Each winner gets an amount of "votes" equal to the number of
> ballots for which he was the favorite on that ballot, out of the set
> of winners (and I suppose we evenly split this if more than one winner
> tied on a given ballot).
>
> 5) Any "winner" who did not actually get any votes is not actually
> elected (addresses the problem Warren brought up).

I assume that you mean that if this condition fails, this winning set
is deemed invalid and pick the next highest set from 3)?

Anyway, this doesn't meet the Droop criterion.

Assume 3 seats, party A (65%) runs 2 candidates and party B (35%) runs
3.  The results are

17:
A1: 0
A2: 0
B1: 99
B2: 100
B3: 99

18:
A1: 0
A2: 0
B1: 100
B2: 99
B3: 98

65:
A1: 100
A2: 99
B1: 0
B2: 0
B3: 0

The best possible result is
A1+B1+B2

This scores perfect.

The 17 will score it at 100 as B2 is in the result, the 18 will score
it at 100 due to B1 and the 65 will also rate it as perfect since A1
is in the result.

All voters are 100% satisfied with one of the candidates.  However,
the A party had 65% of the vote, so should have received 2 of the
seats.

This can be resolved by using the following modification:

1) Voters score the candidates as with score voting.

2) W = the number of winners.

3) Assign a Droop quota of the ballots to each winning candidate.

For each candidate sum up all the scores given to him on the ballots
assigned to him.

The score of the a given result is equal to the sum of the scores for
each candidate.

4) Find the result (i.e. winning set + ballot allocation) which has
the highest score.  The winning candidates in that result are assigned
seats.


The effect is that only 26 of the ballots can be assigned to A1, as
that is a Droop quota

So, testing
A1+B1+B2

A1: 26 rate him as 100
B1: 18 rate 100, rest as 0
B2: 17 rate 100, rest as 0
Total = 6100

testing
A1+A2+B1

A1: 26 rate him at 100
A2: 26 rate him at 99 (still some left out of the 65 for party A)
B1: 18 rate 100, 8 rate at 99
Total: 7766

Thus (A1+A2+B1) wins and Droop proportionality is maintained.

I am not sure if this resolves your original original problem though.

Also, deciding how to allocate the ballots between the candidates in
order to maximise the total for that round could be computer intensive
(maybe there is an efficient algorithm).

--- In RangeVoting@yahoogroups.com, Raph Frank <raphfrk@...> wrote:
> > 5) Any "winner" who did not actually get any votes is not actually
> > elected (addresses the problem Warren brought up).
>
> I assume that you mean that if this condition fails, this winning set
> is deemed invalid and pick the next highest set from 3)?

I have no idea what you are talking about. What "condition"? I am
simply saying that if e.g. there were two seats, and X is everyone's
favorite, then X is the only candidate elected, since X+Y and X+Z
produce an equal sum of scores, but Y/Z receives zero votes.

> Anyway, this doesn't meet the Droop criterion.

I don't know why that should matter. I care about proportionality, not
some Droop criterion. Warren proposed the only sensible
proportionality theorem I ever heard, that a party gets the same ratio
of seats as voters who maximally supported all its members, and
minimally supported all others (paraphrasing).

> Assume 3 seats, party A (65%) runs 2 candidates and party B (35%) runs
> 3.  The results are
>
> 17:
> A1: 0
> A2: 0
> B1: 99
> B2: 100
> B3: 99
>
> 18:
> A1: 0
> A2: 0
> B1: 100
> B2: 99
> B3: 98
>
> 65:
> A1: 100
> A2: 99
> B1: 0
> B2: 0
> B3: 0
>
> The best possible result is
> A1+B1+B2

That would be: 17*100 + 18*100 + 65*100 = 1700+1800+6500 =10,000 points.

A1 gets 65 assets though, so can essentially act as a dictator in all
legislative elections. Which is fine. It simply means no candidates
ran sufficiently close to his ideology space.

> This scores perfect.

Ah, good point.

> All voters are 100% satisfied with one of the candidates.  However,
> the A party had 65% of the vote, so should have received 2 of the
> seats.

No. It got 65% of the vote, so it gets 65% of the vote power, as I said.

> This can be resolved by using the following modification:
>
> 1) Voters score the candidates as with score voting.
>
> 2) W = the number of winners.
>
> 3) Assign a Droop quota of the ballots to each winning candidate.

no. i don't care about this droop quota thing. the issue is already
resolved by giving candidates the right amount of "vote power".

> Also, deciding how to allocate the ballots between the candidates in
> order to maximise the total for that round could be computer intensive
> (maybe there is an efficient algorithm).

yes! that is the big problem i wish i could solve. more of a task for
warren. :)

On Tue, Feb 3, 2009 at 5:22 AM, brokenladdercalendar
<thebrokenladder@...> wrote:
> I have no idea what you are talking about. What "condition"? I am
> simply saying that if e.g. there were two seats, and X is everyone's
> favorite, then X is the only candidate elected, since X+Y and X+Z
> produce an equal sum of scores, but Y/Z receives zero votes.

I didn't realise about that you meant there would be a non-equal assembly.

I doubt in a real election that would actually result in anyone being
eliminated.  As long as the candidates themselves rate themselves as
unique max, this rule will have no effect as they will all get at
least 1 vote.

>> Anyway, this doesn't meet the Droop criterion.
>
> I don't know why that should matter. I care about proportionality, not
> some Droop criterion. Warren proposed the only sensible
> proportionality theorem I ever heard, that a party gets the same ratio
> of seats as voters who maximally supported all its members, and
> minimally supported all others (paraphrasing).

In the context of a rating ballot, that would be how I interpret the
Droop criterion too.  Ofc, Droop does assume that the assembly is
going to be a peer assembly.

> A1 gets 65 assets though, so can essentially act as a dictator in all
> legislative elections. Which is fine. It simply means no candidates
> ran sufficiently close to his ideology space.

Btw, are you planning for there to be districts, or just a single
State wide election?

>> All voters are 100% satisfied with one of the candidates. However,
>> the A party had 65% of the vote, so should have received 2 of the
>> seats.
>
> No. It got 65% of the vote, so it gets 65% of the vote power, as I said.

I misinterpreted that section.  I though you were just assigning those
votes in order to decide who to eliminate.

>> This can be resolved by using the following modification:
>>
>> 1) Voters score the candidates as with score voting.
>>
>> 2) W = the number of winners.
>>
>> 3) Assign a Droop quota of the ballots to each winning candidate.
>
> no. i don't care about this droop quota thing. the issue is already
> resolved by giving candidates the right amount of "vote power".

Right.  However, most current PR systems are designed to result in a
peer assembly.

You could use Asset to convert the 'first-stage' election into a peer assembly.

>> Also, deciding how to allocate the ballots between the candidates in
>> order to maximise the total for that round could be computer intensive
>> (maybe there is an efficient algorithm).
>
> yes! that is the big problem i wish i could solve. more of a task for
> warren. :)

It might not be that difficult.

Another option is to just allow each candidate (or other interested
party) to submit a result.

The result with the highest score is declared the winner.

This gives a potential slight advantage to those with better computing
power, but it is unlikely to actually make any difference, in
practice, as one result will be a clear winner.

--- In RangeVoting@yahoogroups.com, Raph Frank <raphfrk@...> wrote:
> I doubt in a real election that would actually result in anyone being
> eliminated.  As long as the candidates themselves rate themselves as
> unique max, this rule will have no effect as they will all get at
> least 1 vote.

so what? that doesn't guarantee them a seat. and if they do get a
seat, they get to represent...themselves. hah! a lot of good that will
do. it will give them a chance to argue about proposed legislation
before it goes to a vote perhaps.

> Btw, are you planning for there to be districts, or just a single
> State wide election?

if people think that it is beneficial for a candidate to be from their
region, they can increase their score for him. so, no districts as far
as i'm concerned. but you can have districts if you want to, where
each district gets e.g. 3 representatives or something.

> most current PR systems are designed to result in a
> peer assembly.

what does that mean?

> Another option is to just allow each candidate (or other interested
> party) to submit a result.
>
> The result with the highest score is declared the winner.

you're a genius. "you want power? then do some math for us...thank
you." hah!

> This gives a potential slight advantage to those with better computing
> power, but it is unlikely to actually make any difference, in
> practice, as one result will be a clear winner.

yup.

________________________________


--- In RangeVoting@yahoogroups.com, Raph Frank <raphfrk@...> wrote:

On Mon, Feb 2, 2009 at 12:08 AM, Juho Laatu <juho4880@...> wrote:
> The reform plan is to count the proportionality
> at national level and then distribute those
> seats down to the districts. Currently alliances
> between parties are supported in the districts
> but that system is one step short of perfect.
> Unfortunately the reform plan contains also a
> proposal to have thresholds (did not exist
> before) and some other weird ideas.


<snip>


I assume the plan is to guarantee national proportionality based on 1
person/1 vote.  If it is raining in a district, the voters in that
district will have less effect on the national result due to lower
turnout.

This could be compensated by rescaling the vote totals, i.e.
multiply the vote totals by the ratio of the population of the
district divided by the number of voters who cast a vote in that
district.

<snip>


Raph,
What a crazy idea. Why would anyone particularly care about the "problem"
you describe, to the extent of tampering with  '1 vote-1 value'  to address
it??

I think the supposed problem is actually a good thing because to the extent
that voters are parochially motivated (i.e. wish their own district to elect as
many
MPs as possible) it will encourage higher turnout.

The luck factor of different weather in different districts is one of the
arguments in
favour of compulsory polling-booth attendance, or of just making it more
convenient
for people to vote.

Chris Benham


       Make Yahoo!7 your homepage and win a trip to the Quiksilver Pro. Find out
more

[Non-text portions of this message have been removed]

On Tue, Feb 3, 2009 at 6:19 PM, brokenladdercalendar
<thebrokenladder@...> wrote:
> so what?

Well, it means that the rule is pointless, you could just leave it out
and the method is simplified.

> what does that mean?

It means an assemble where everyone is equal .. i.e a peer.

Your proposal is a non-peer assembly.

>> This gives a potential slight advantage to those with better computing
>> power, but it is unlikely to actually make any difference, in
>> practice, as one result will be a clear winner.
>
> yup.

It also allows some weird criteria to be used.  Rather than defining
how to achieve the result, you just define the measure.

Shortest split-line defines a method for determining the result rather
than a criterion for measuring a result.  This leads to maps that
aren't aesthetically acceptable to people.

It might be possible to define a simple measure that would give better
looking districts, but which is no possible to exactly calculate.

On Tue, Feb 3, 2009 at 7:14 PM, Chris Benham <cbenhamau@...> wrote:
> Raph,
> What a crazy idea. Why would anyone particularly care about the "problem"
> you describe, to the extent of tampering with  '1 vote-1 value'  to address
> it??

Would you support assigning seats between the districts based on the
number of people who vote?

A district with a larger turnout would get more seats than its
population would merit.

> I think the supposed problem is actually a good thing because to the extent
> that voters are parochially motivated (i.e. wish their own district to elect
> as many
> MPs as possible) it will encourage higher turnout.

So you agree, people would care.  It is possible (at least in theory),
that not including the rule could result in some districts objecting.

Some people complain about losing 'local' politicians as an objection
to reducing the number of 3 seat constituencies in Ireland.

> The luck factor of different weather in different districts is one of the
> arguments in
> favour of compulsory polling-booth attendance, or of just making it more
> convenient
> for people to vote.

I agree with making it convenient, but don't support compulsory
voting.  I don't see the value in forcing someone to give you their
opinion.

Btw, rescaling effectively achieves compulsory voting ... you just
assume that they vote the same way as the rest of the district did on
average :).

A decade from now, people may look back with amazement at how the US
renewed, and enlarged its Afghan incursion, with virtually no public
discussion or protest. Long after having avenged 9/11 by driving bin
Laden and al-Quaida out of Afghanistan, what could possibly justify
sending more young Americans to be killed or maimed in this remote,
primitive country? Are we going to take on the endless task of
stabilizing all the failed governments of the world? Have we done such
a wonderful job eliminating social, political, and economic injustice
in the US that we are now in position to do the same worldwide? If so,
is the tool of choice the helicopter gunship?

Perhaps the answers to the above questions are no, and our real
objective is simply to root out terrorists, wherever they may hide. If
so, then war with Pakistan may be just around the corner. For further
discussion of this matter, see my article at
http://www1.cs.columbia.edu/~unger/myBlog/endsandmeansblog.html

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

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

--- In RangeVoting@yahoogroups.com, Raph Frank <raphfrk@...> wrote:
> On Tue, Feb 3, 2009 at 6:19 PM, brokenladdercalendar
> <thebrokenladder@...> wrote:
> > so what?
>
> Well, it means that the rule is pointless, you could just leave it out
> and the method is simplified.
>
> > what does that mean?

who _cares_ if some guy wants to show up each day and cast one vote?
most likely he won't want to do that. the elimination is there as a
formality, so that we don't say "okay, you're elected, but you have
zero vote power". that would just be incredibly stupid.

> It means an assemble where everyone is equal .. i.e a peer.
> Your proposal is a non-peer assembly.

why should representatives be equal if they represent an unequal
number of voters?

> It also allows some weird criteria to be used.  Rather than defining
> how to achieve the result, you just define the measure.

yeah, that is interesting.

> Shortest split-line defines a method for determining the result rather
> than a criterion for measuring a result.  This leads to maps that
> aren't aesthetically acceptable to people.

really? how so? they look fine to me. you think people are going to
reject them because they don't look "pretty"?

i'd like to befriend any of you who have facebook accounts. you can
look me up under my name, clay shentrup.

On Wed, Feb 4, 2009 at 3:15 AM, brokenladdercalendar
<thebrokenladder@...> wrote:
> --- In RangeVoting@yahoogroups.com, Raph Frank <raphfrk@...> wrote:
>> Shortest split-line defines a method for determining the result rather
>> than a criterion for measuring a result. This leads to maps that
>> aren't aesthetically acceptable to people.
>
> really? how so? they look fine to me. you think people are going to
> reject them because they don't look "pretty"?

Yes.

People want boundaries that follow rivers/mountain ranges/county boundaries etc.

They also would like that the boundaries don't split 'communities of
interest' in two.

Shortest splitline generally places its boundaries right though towns.
  It also ignores all the other criteria people want.

The simplicity of the algorithm means that it is hard to manipulate
and is clearly unbiased.  However, that is traded off against
producing maps that are not the normal way this maps are created.

One measure I think would be good would be

1) A district map which has a lower population imbalance is better
2) Subject to 1), the best distict is the one with the shortest sum of
the boundaries of all the districts
- All boundary lengths are scaled by population density at each point
i.e the boundary length is the line integral of the population density
around the boundary.

This means that running a boundary through a low density area would
'cost' less.  The effect should be that the boundaries would stay away
from high populate areas.

Density is hard to define in sparsely populated areas.  There would be
a need to have a minimum size of area (say 100 people) for the average
to be taken over.

A decade from now, people may look back with amazement at how the US
renewed, and enlarged its Afghan incursion, with virtually no public
discussion or protest. Long after having avenged 9/11 by driving bin
Laden and al-Quaida out of Afghanistan, what could possibly justify
sending more young Americans to be killed or maimed in this remote,
primitive country? Are we going to take on the endless task of
stabilizing all the failed governments of the world? Have we done such
a wonderful job eliminating social, political, and economic injustice
in the US that we are now in position to do the same worldwide? If so,
is the tool of choice the helicopter gunship?

Perhaps the answers to the above questions are no, and our real
objective is simply to root out terrorists, wherever they may hide. If
so, then war with Pakistan may be just around the corner. For further
discussion of this matter, see my article at
http://www1.cs.columbia.edu/~unger/myBlog/endsandmeansblog.html

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

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

On Wed, 4 Feb 2009, Raph Frank wrote:

> On Wed, Feb 4, 2009 at 3:15 AM, brokenladdercalendar
> <thebrokenladder@...> wrote:
>> --- In RangeVoting@yahoogroups.com, Raph Frank <raphfrk@...> wrote:
>>> Shortest split-line defines a method for determining the result rather
>>> than a criterion for measuring a result. This leads to maps that
>>> aren't aesthetically acceptable to people.
>>
>> really? how so? they look fine to me. you think people are going to
>> reject them because they don't look "pretty"?
>
> Yes.
>
> People want boundaries that follow rivers/mountain ranges/county boundaries
etc.
>
> They also would like that the boundaries don't split 'communities of
> interest' in two.
>
> Shortest splitline generally places its boundaries right though towns.
> It also ignores all the other criteria people want.

My Congressional district starts out in the Bronx, spills into
Westchester County, and then crosses the Hudson River into Rockland
County.And it is by no means unique in its exciting borders. Of
course  these result from big-time gerrymandering. But, apart from the
political effects, who cares if an election district is split by a
river or other geographic obstacle? There are SOME effects if the
geographic spread is large. For example my congressman has local
offices in 3 counties. Of course big spread is inherent in sparsely
populated areas. But it isn't as tho the people of such a district are
members of a social club that meets in different people's houses.

The key point is that the procedure for defining the district
boundaries not have any slack in it that can be exploited by
politicians.

Steve

On Wed, Feb 4, 2009 at 3:48 PM, Stephen Unger <unger@...> wrote:
> My Congressional district starts out in the Bronx, spills into
> Westchester County, and then crosses the Hudson River into Rockland
> County.And it is by no means unique in its exciting borders. Of
> course these result from big-time gerrymandering. But, apart from the
> political effects, who cares if an election district is split by a
> river or other geographic obstacle?

People care.

In any case, politicians want to maintain control of the districting
process and thus they care alot about any excuse they can use to argue
that an automatic system is not feasible.

> The key point is that the procedure for defining the district
> boundaries not have any slack in it that can be exploited by
> politicians.

What about a rule that 'features' can only be updated 10 years in advance.

That would make it hard to abuse the system.

In any case, I think there is a grey area between absolutely no
parameters and giving the legislature total control of the process.

--- On Tue, 3/2/09, Raph Frank <raphfrk@...> wrote:

> Whatever the algorithm, some kind of backtracking does seem
> necessary.
>
> This is a pity, as where possible a simple algorithm is
> preferred.

Alternatives to this could be e.g. 1) to
give up exact regional PR and allocate
the seats to regions separately within
each party or 2) Michel Balinski's
allocation method that adjusts some
divisors to make exact regional and
political allocation at the same time
(the committee studied this too) (the
procedure doesn't have explicit
backtracking but it is a trial and error
based method though) or 3) just leave
some party without a seat if they don't
have many enough candidates in some
region (note that some party can in
principle run out of candidates in all
districts).

> > The method is strict with "regional PR".
>
> However, in terms of parties, it isn't.  Candidates
> will depend on the
> locals to actually get elected, but the number of them
> elected is
> determined based on the national vote.

Each district (within a party) gets however
seats (approximately) in proportion to the
number of votes given in that region. The
national result has usually only small
(balancing, fractions of seats) impact on
the district internal (old style) results.

> This means that districts with a smaller turnout would end
> up with
> less effect on the national party totals and it is the
> national party
> totals which decides the coalition negotiations.
>
> For example, if there were 2 equal sized districts and 10
> seats each
> and the results were
>
> District 1
> Party A: 40%
> Party B: 60%
>
> District 2:
> Party A: 70%
> Party B: 30%
>
> Under a pure regional vote, the results are
>
> A: 4+7 = 11 seats
> B: 3+6 = 9 seats
>
> However, if the turnout in district 2 was 50% of the
> turnout in
> district 1, then the national totals would be
>
> A = 40 + 0.50*70 = 75 votes
> B = 60 + 0.50*30 = 75 votes
>
> Thus the lower turnout caused a clear win for party A to
> become a draw
> with 10 seats each
>
> Having said that, the effect is probably not that massive.

Yes, the differences in turnout and district
population vs. people with right to vote are
probably not too big.

Both districts will get 10 seats although
there were much less voters in region B.
One can not allocate the seats within the
districts in proportions 40:60 and 70:30
(= 35:15) but must seek new proportions
that sum up right at the national level.

The formula is btw

party seats in district y = (party votes in
district y / party votes in election area)
* party seats in the election area

The election area will be the whole country
(they considered also other alternatives).

I guess this could lead to having too many
seats (integer parts) in districts that
have lots of votes when compared to the
number of seats.

One could see this question also from the
point of view of the sleeping voters (or
people without voting right). They are
fully counted when the number of seats per
district is determined. On the political
PR side we could either think that the
sleeping voters of some district must have
the same opinions as the other voters in
that district, or we may assume that their
opinions follow the national opinion shares.

I think the current approach of simply
counting all given votes at national level
to determine political PR is very straight
forward and accepted as a fair approach.

> In Ireland, it is pure population.  A person voting in a
> constituency
> with a large child or immigrant population would in effect
> have a more
> powerful vote.

I'm not quite sure how the Finnish seat
allocation to districts votes but I believe
children are counted. The differences are
anyway quite small.

> > I think this proposed reform is a
> > step forward in most aspects. The
> > biggest problems are introduction
> > of thresholds, and introduction of
> > a district specific threshold (!)
> > that allows local parties to get
> > seats even if they don't meet the
> > national threshold.
>
> I assume that these are being set so that all current
> parties will
> meet the thresholds (coincidently)?

Yes, that is the reason. I'm a bit
surprised that most parties are happy with
the proposal because the threshold is not
too far from the support level of the
current parties.

The target of avoiding excessive
fragmentation of the party field was
mentioned in some papers. I don't think
having one or two parties with one
representative would destabilize the
country. Those parties could rather enrich
the discussion (the old parties and their
positions are already well known and very
stable and without any turbulence and
major conflicts between parties).

> > The new proposal gets rid of election
> > + district specific party alliances.
> > It does not fix the problem of the
> > open list method of not supporting
> > party internal PR properly
> > (candidates with highest number of
> > votes elected). Party internal
> > alliances / trees would fix this
> > quite nicely in my opinion.
>
> True.  A tree system for open list would be better than
> plain open
> list which is better than closed lists.
>
> Another option would be Asset voting, but the tree system
> is at least
> completely transparent and open.
>
> Ofc, I still prefer PR-STV.

PR-STV is a good method. If I could decide
I'd use parts from both (and trees too),
with different balance for different needs.

In Finland PR-STV would have also some
problems. The number of candidates per
district may be e.g. 150. Filling one's
ballot so that one can be certain that
one's vote is not lost could become
tedious. The traditional ballots are also
very simple, just a small circle where the
voter writes the number of her favourite
candidate.

Having a clear structure (e.g. a tree) of
the candidates may also help voters when
choosing the right candidate from the
large set of candidates. Of course also
PR-STV may use such structures informally.

In some environments it is good if the
candidates clearly position themselves in
the political map. In some other
environments we may want to keep the
candidates as independent individuals only
(e.g. in some purely non-political
elections like picking a chairman for some
open source community).

Juho


<http://en.wikipedia.org/wiki/Finnish_parliamentary_election,_2007#Election_dist\
ricts>
<http://fi.wikipedia.org/wiki/Suomen_vaalipiirit#Eduskuntavaalien_2007_vaalipiir\
it>

--- On Wed, 4/2/09, Raph Frank <raphfrk@...> wrote:

> People want boundaries that follow rivers/mountain
> ranges/county boundaries etc.
>
> They also would like that the boundaries don't split
> 'communities of
> interest' in two.
>
> Shortest splitline generally places its boundaries right
> though towns.
>  It also ignores all the other criteria people want.

If natural boundaries are the target then the
natural approach could be to first do the
background work and split the country in
small enough natural areas and then use some
algorithm to build the actual voting
districts from these atomic areas.

Each border line segment could also have
additional parameters like type of segment
to allow the algorithm to favour the most
natural boundaries where possible.

The atomic areas need not be updated often.
If some area grows too large it can be split
in two. Areas that lose population need not
be updated as often.

The districting algorithm (initiation) could
contain some randomness so that the atomic
areas can not be tested with the algorithm
when they are formed.

Juho

On Wed, Feb 4, 2009 at 9:00 PM, Juho Laatu <juho4880@...> wrote:
> If natural boundaries are the target then the
> natural approach could be to first do the
> background work and split the country in
> small enough natural areas and then use some
> algorithm to build the actual voting
> districts from these atomic areas.

Right, in the US census, everything is based on census blocks.

The problem is that the more dials there are the easier it is to manipulate.

I think a reasonable compromise here is that the algorithm is decided
10+ years in advance.  If you don't have precise census data, then it
is much harder to manipulate the maps indirectly via algorithm
selection.

> The districting algorithm (initiation) could
> contain some randomness so that the atomic
> areas can not be tested with the algorithm
> when they are formed.

That would work too.  However, people might object that it makes it
harder to confirm the result.

Perhaps, a few trusted officials would throw dice to generate the random seed.

This can be set up so that as long as 1 of them actually is
trustworthy, the result is random.

--- On Thu, 5/2/09, Raph Frank <raphfrk@...> wrote:

> On Wed, Feb 4, 2009 at 9:00 PM, Juho Laatu
> <juho4880@...> wrote:
> > If natural boundaries are the target then the
> > natural approach could be to first do the
> > background work and split the country in
> > small enough natural areas and then use some
> > algorithm to build the actual voting
> > districts from these atomic areas.
>
> Right, in the US census, everything is based on census
> blocks.
>
> The problem is that the more dials there are the easier it
> is to manipulate.
>
> I think a reasonable compromise here is that the algorithm
> is decided
> 10+ years in advance.  If you don't have precise census
> data, then it
> is much harder to manipulate the maps indirectly via
> algorithm
> selection.
>
> > The districting algorithm (initiation) could
> > contain some randomness so that the atomic
> > areas can not be tested with the algorithm
> > when they are formed.
>
> That would work too.  However, people might object that it
> makes it
> harder to confirm the result.
>
> Perhaps, a few trusted officials would throw dice to
> generate the random seed.
>
> This can be set up so that as long as 1 of them actually is
> trustworthy, the result is random.

Here's one simple approach.

The basic algorithm could be such that it
works by always splitting the remaining
areas in two same size ore near same size
(n districts, n+1 districts) areas. One
can achieve randomness by making the very
first split random (r districts, m-r
districts). That should be enough to make
atomic area planning as well as border
segment type/weight/naturalness planning
quite useless.

One approach to getting that random number
is to throw dice once (in the presence of
media, TV cameras and party
representatives) and then use n-r to split
the country (or state or other smaller
natural region). This leaves some marginal
space for planning since someone could
test all six alternatives when planning
the new atomic areas and segment types.
This sounds still like a quite hopeless
attempt to forge the results.

The correctness of the result can be
checked easily if the algorithm and atomic
areas and weighted border segments are
publicly available. The algorithm could be
also changed/improved often if there is a
need, but the algorithm could be also very
simple and only the atomic areas and
border segment types would change as
needed.

You mentioned also the changing census
data. That alone, with a deterministic
algorithm to do the division (instead
of the ruling party), could be enough to
stop gerrymandering. And even the
changing opinions of the voters (with
the algorithm) could do that.

Juho

http://scorevoting.net/CoreSuppPocket.html

thanks warren.

-http://online.wsj.com/article/SB123388752673155403.html

And the Oscar Goes to...Not Its Voting System
Selection of Academy Award Nominees and Winners is Flawed, but
Reformers Can't Seem to Elect a Better Candidate

     *
       By CARL BIALIK

Columnist's name

     * Article
     * Comments

more in US »

     * Email
     * Printer Friendly
     * Share:
           o Yahoo Buzz more
           o facebook
           o MySpace
           o LinkedIn
           o Digg
           o del.icio.us
           o NewsVine
           o StumbleUpon
           o Mixx
     * smaller Text Size larger
     *

Academy Award nominees and winners are selected using two different
voting systems that are, according to some political mathematicians,
the worst way to convert voters' preferences into an election outcome.

The nominees are selected using a system called instant runoff, which
has been adopted in some municipal and state elections. Out of last
year's 281 eligible films, each voter selects five nominees in order
of preference for, say, best picture. All movies without any
first-place votes are eliminated. The votes for those films with the
least first-place votes are re-assigned until five nominees have enough.

One problem with that system is a kind of squeaky-wheel phenomenon: A
movie that is second place on every ballot will lose out to one that
ranks first on only 20% of ballots but is hated by everyone else.
Then, in another upside-down outcome, a movie can win for best picture
even if 79% of voters hated it so long as they split their votes
evenly among the losing films. This isn't as unfamiliar as it sounds:
Some people think Al Gore would have won the Electoral College in 2000
if Ralph Nader hadn't diverted more votes from him than he took from
former President George W. Bush.
[Oscar] Getty Images

"It's crazy," says Michel Balinski, professor of research at École
Polytechnique in Palaiseau, France. The nomination system's properties
are "truly perverse and antithetical to the idea of democracy," says
Steven Brams, professor of politics at New York University. He thinks
the final vote for the Oscar winner may be even worse than the
selection of nominees.

The big problem: If voting systems themselves were put to a vote,
prominent scholars would each produce a different ballot, then
disagree about which system should be used to select the winner. So
it's no surprise that advocates of alternate voting systems, which
range from simple yes/no approval ratings to assigning numerical
scores to each candidate, have had little more luck reforming
political elections than they have with entertainment awards.

Consider two systems that, on the surface, seem similar. Prof.
Balinski and mathematician Rida Laraki have devised a system they call
majority judgment that requires voters to rank each candidate on a
scale from 1 to 6. The votes are lined up in order, and each candidate
is assigned the middle, or median, score. The highest median score
wins. Another system, range voting, isn't that different: The
candidate with the highest average, or mean, score wins.

Yet the second system's leading advocate, Temple University
mathematician Warren D. Smith, has devoted a Web page to the
Balinksi-Laraki system's "numerous disadvantages."

Brace yourselves for "Ishtar" defeating "The Godfather." Suppose 49
voters award "The Godfather" six points and "Ishtar" only four. One
voter grants the desert debacle four points and the mafia masterpiece
three, and the remaining 49 award "The Godfather" three points and
"Ishtar" only one point. "Ishtar" actually wins with a median score of
four points compared to "The Godfather's" three points. Prof.
Balinski, in turn, calls range voting a "ridiculous method," because
it can be manipulated by strategic voters.

Despite the flaws in Oscars voting, the system remains as it has since
1936. Every 15 years or so, the Academy re-examines its voting and has
decided to stick with it, says the Academy of Motion Picture Arts and
Sciences' executive director, Bruce Davis. "It is a very effective
method of reflecting the will of the entire electorate," Mr. Davis says.
[Oscar voting system chart]

But many voting theorists aren't so keen on the system. It's called
instant runoff because it is used in political elections in lieu of a
two-stage vote in which top candidates compete again if none receives
a majority of the vote. Among the potential problems, showing up to
vote for your favorite candidate may create a worse outcome than not
showing up at all. For example, your vote could change the order in
which candidates are eliminated, and the next-in-line candidate on the
ballot for the newly eliminated film may be a film you loathe.

To choose Oscar winners, voters simply choose their favorite from the
nominees, and the contender with the most votes wins. That could favor
a film that has a devoted faction of fans, and sink films with
overlapping followings who split their vote. Even most critics of
instant runoff say it beats this plurality system that led to the
Gore-Nader-Bush result. In the film realm, Prof. Brams of NYU blames
the current system for the best-picture victory of "Rocky" over films
such as "Network" and "Taxi Driver" that he speculates would have won
head to head.

How this works out in reality is hard to know, because the Academy
doesn't release any details about the balloting, even after the
telecast, in part to avoid shaming fifth-place films. Mr. Davis says
even he never learns the numbers from his accountants: "Are there
years when I'm curious as to what the order of finish was? Absolutely.
But I recognize it as a vulgar curiosity in myself."
More

The Oscars involve two stages of voting, for nominees and for winners.
Delve into the math of elections in the Numbers Guy blog.

     * Complete Coverage: Academy Awards

Such secrecy frustrates voting theorists who are anxious for
experimental data about voter behavior that may help them choose from
among different voting systems. Without such evidence, they are left
to devise their own studies, to dream up examples that sink rival
systems or to create computer simulations to study how easily
different systems can be manipulated.

Sports fans cry manipulation when votes don't go as they'd hoped. Many
sports awards and rankings are derived from what is known as Borda
count, which asks voters to rank candidates and then assigns points on
a sliding scale, with the most for first-place votes and the least for
last-place ones.

Critics of these systems fear that strategic voters will assign their
top choice the highest possible score, and everyone else zero, thereby
seizing more power than voters who approach the system earnestly; or,
in the case of rankings, bury or omit a preferred candidate's top
rival. Boston Red Sox fans will tell you to this day that such
strategic voting by a New York beat writer cost Pedro Martinez the
American League Most Valuable Player award a decade ago.

Says Prof. Balinksi, "Not everyone will do it, but enough will do it
to manipulate the results."

There is a philosophical question obscured by that criticism: Should
voters with stronger feelings have more influence? A voter may support
Candidate A strongly and loathe all the rest; two other voters may
like Candidate A but slightly prefer B. Should B beat A even though
all voters would have been fine with A?

Some scholars back the Condorcet winner, the candidate that would beat
all others in head-to-head matchups. Trouble is, there isn't always
one. As an alternative, Prof. Brams advocates approval voting, which
tallies the number of voters who approve of each candidate and chooses
the one with the most votes.

Rob Richie, executive director of FairVote, which has had success
pushing the adoption of instant runoff for elections, says that
approval voting doesn't fly with politicians: They're uncomfortable
with the idea that voters who prefer them might throw equal support to
a rival. For advocates of alternate systems, it's crucial to get
support from politicians because voters aren't likely to get excited
about such issues unless the country is hanging on a chad.

Mr. Richie argues that, in practice, instant runoff hasn't displayed
the feared paradoxes. He says his critics should go get their
preferred systems adopted so they can offer their own proofs of
concept. He adds that mathematicians haven't made much headway
changing voting laws "so they hound reformers who are being
successful, and that's just irritating."

Vanderbilt University mathematician Paul H. Edelman, who has consulted
with the Country Music Association on its annual awards, says his
colleagues should tone down the dogma and embrace a range of voting
systems for different situations. "The mistake that mathematicians
make is to assume that all elections are the same," Prof. Edelman
says. "That's a terrible thing to do."
Get Me a Recount
[Vote]

While Academy Award nominees and winners are selected using two
different voting systems, there are at least six other major ones that
have been proposed and studied by scholars. And each one can produce
different outcomes from the same ballots.

In a hypothetical 11-voter election, in which voters score eight
candidates from 0 to 20, each candidate would win under one of eight
major voting systems. Bolds mean that voter approves that candidate --
roughly equivalent to a yes/no vote.

Number of Ballots Candidate
A B C D E F G H
4 18 4 5 17 15 0 13 14
3 0 14 5 11 12 10 8 9
2 0 12 20 10 11 9 18 19
1 2 0 12 17 1 11 16 3
1 0 1 4 2 3 16 15 5
Wins in Plurality Runoff Instant runoff Borda count Condorcet
Approval voting Mean range voting Median range voting

See how each candidate wins in each system:

A wins in plurality: A has four first-place votes, more than any other
candidate.

B wins in runoff: All but the top two first-place vote getters, A and
B, are eliminated. B is preferred by three of the four voters who
ranked other candidates first, and beats A, 6-5.

C wins in instant runoff: Under this system, each voter selects five
nominees, in order, in a given category. E, G and H have no
first-place votes and are eliminated first. Then come D and F, which
each have one first-place vote. Among remaining candidates, C ranks
second on those ballots, so C picks up two more first-place votes and
is now tied with A, with four. B, with three, is eliminated next, and
C ranks above A on the ballots that belonged to B, so C beats A, 7-4.

D wins in Borda count: Borda count asks voters to rank candidates and
then assigns points on a sliding scale, with the most for first-place
votes and the least for last-place ones. On each ballot, give seven
votes to the first-place contender, six to second, and so on, down to
zero for the last-place candidate. D edges E, 52-48.

E wins in Condorcet: The Condorcet winner is the candidate which beats
all others in head-to-head matchups. E beats every other candidate
head to head, by ranking higher than each on a majority of ballots. E
beats A, 6-5; B, 6-5; C, 6-5; D, 6-5; F, 9-2; G, 7-4; H, 7-4.

F wins in approval voting: This system tallies the number of voters
who approve of each candidate and chooses the one with the most votes.
F is approved by seven voters, edging D, approved by 6.

G wins in mean range voting: The mean vote for G is 13, edging D, with
12.7.

H wins in median range voting: The median vote for H is 14, beating G,
which has 13.

Sources: Center for Range Voting; WSJ Research

Write to Carl Bialik at numbersguy@...