[EM] STV and Trees

[Juho]

On Aug 22, 2008, at 12:47 , Raph Frank wrote:

> On 8/22/08, Juho <juho4880 at yahoo.co.uk> wrote:
>> On Aug 22, 2008, at 2:18 , Raph Frank wrote:
>>> Thus, I don't see them as massively different ... the trees just  
>>> add more
>>> structure and reduce the freedom.
>>>
>>  The intention was not to reduce freedom. If a voter wants to  
>> bypass the
>> default inheritance order as given in the tree he just ranks all the
>> candidates (also other tools like names of some groups could be  
>> available if
>> we want to go for maximum flexibility, e.g. "women", "Dublin",  
>> "under 30",
>> "those who voted yes for X", "candidates that support Y"). Also  
>> declaring a
>> vote to be just as written is ok (no inheritance if candidates on  
>> the ballot
>> run out).
>
> Well, I meant compared to candidate lists.  The candidate can set
> his list up any way he likes under a candidate list system, however,
> under a tree system, there are fewer valid choices.  This means
> freedom is reduced.

If the inheritance tree and the candidate determined preference/ 
inheritance lists are used just as a default inheritance rules then  
the level of freedom is the same in the sense that voters can in any  
case override those rules by giving a full ranking of the candidates.  
The inheritance rules are more flexible with the candidate given  
lists but the voter van override those anyway and from this point of  
view it is possible to plan them so that they offer solutions to the  
most common cases (voter opinions).

> However, it is likely that this freedom is not much use as a  
> candidates
> probably would just fill out his list semi-randomly as he gets to  
> the lower
> rankings.  The tree system helps here as it fills out all the  
> rankings in a
> standard manner.
>
>>  Yes, this leads to STV votes that support equalities. Bullet vote  
>> to C1
>> could become in effect C1>G1=G2=G3>P1=P2=P3=...
>
> Yeah.
>
>>  Yes, except that I think no flexibility is lost, so on the  
>> "negative" side
>> there is only the added information in the poster listing all the  
>> candidates
>> and that the voters are expected to roughly understand (well, it  
>> is maybe
>> much easier to understand that than trying to dig out the opinions  
>> and more
>> detailed affiliations of all the potential candidates oneself).
>
> Well, the trees/lists might be used for national vote transfers.   
> In that
> case, the voter cannot rank all the candidates.
>
>>  Bullet voting could be common if the tree is informative enough.  
>> Also, in
>> elections with multiple candidates the lost fractions could be  
>> relatively
>> large.
>
> Not sure that is a good thing, but if that is how the voters vote,  
> then
> that is their choice.

In a bullet vote with tree inheritance the first thing that one loses  
is that the voter does not determine the ranking between the  
candidates of the nearest grouping in any more detail than saying  
which candidate within that group in the best.

>   However, even with 2-3 choices, most of the
> vote power would be assigned based on the voters direct choice.
>
> Few candidates end up with multiple quotas, so most of your vote
> goes to your first choice (assuming they get elected).

If you don't know which one or two of the numerous candidates of your  
favourite party/grouping will get elected you need to rank them all  
(to make sure that your vote is not lost after all the candidates on  
the ballot have been eliminated).

>>> (Though, I would
>>> allow a candidate to opt out).
>>>
>>
>>  It probably is possible for a candidate to establish one's own  
>> party and
>> not contribute and not benefit of the votes given to the party (or  
>> group).
>> In most cases I'd expect the candidates to benefit of being a  
>> member of a
>> party/group (even if the algorithm wouldn't favour large groupings  
>> like e.g.
>> d'Hondt).
>
> I meant opt out while not having to leave the party, though I
> guess that is down to party rules.
>
>> Candidates that are not members of any subgroups (at some level in  
>> the
>> tree) could be grouped together to form a default list (unless  
>> they object
>> this too).

I used term "list" in the meaning of "subgroup" / "branch" here.

> They way I would work it is that the list they submit is constrained
> by what groups they are members of.

I guess we are now talking about a combination of trees and candidate  
determined preference lists. Limiting those lists to the tree  
structure sounds natural (avoids the total confusion that could be a  
result of e.g. some candidate forwarding votes to a competing party  
although and being listed in the tree too).

> If you are part of a party, then you must rank party members first,
> but otherwise there is no restriction.
>
> If you are part of a sub-group, then you must rank members of that
> subgroup first and then the other party members.
>
>> Well, maybe we just automatically give the "non
>> left wingers" an option to form a corresponding grouping when we  
>> learn that
>> a left wing grouping has been formed (and candidates may opt in  
>> and opt
>> out).

Btw, the reason why I'm interested in allowing a corresponding group  
to be formed in the right wing (or "remaining parts") is that in a  
situation where one party has ten candidates, and each of them gets  
about the same number of votes, but five of them form a subgroup, and  
the party would get one seat only, many seat allocation algorithms  
would allocate that single seat to the group of five, and that might  
be considered unfair.

> Under the above rule, a member of the left-wing might rank as
>
> L1>L2>L3>C1>C2>C3>R1>R2>R3
>
> whereas someone who is unaligned might rank as
>
> L1>R1>C1>R2>C2>L2>R3>C3>L3
>
> (i.e. it actually means not aligned, rather than not left wing)
>
> Someone in the right wing could rank
>
> R1>R2>R3>C1>C2>C3>L1>L2>L3
>
> This also raises an interest point, this would be a valid
> right wing ranking
>
> R1>R2>R3>C1>L1>C2>L2>C2>L3
>
> It would meet the condition of ranking all members of
> the right wing first, however, the left and centre wings
> are not ranked as would be expected by a ring-winger.

If the R-party supporters generally feel that C-party is much closer  
to them than the L-party then there is also the option that R-party  
and C-party form a coalition, and as a result the candidate given  
inheritance rules should reflect this change in the structure of the  
tree.

> This is probably not an issue as it is the early votes on
> a list that actually matter.
>
>>> The only rule would be that they must rank all party members  
>>> before any
>>> other candidate.
>>  I didn't quite understand what the alternative to this would be.
>
> Well, the alternative would be to allow party members complete
> freedom on how to set up their lists.  This was talking about the
> case where a party member doesn't take part in the tree structure.
>
> However, I doubt parties would give candidates that much freedom,
> so a specific rule is not required.

Having a mixture of tree inheritance and candidate defined  
inheritance could be quite complex to the voters. When limited to the  
tree structure the candidate defined inheritance lists could however  
still fit in the overall picture that the tree tries to give to the  
voters (e.g. in posters). I however expect that there would be many  
voters that do not fully agree with the order that the candidate has  
given. For this reason the candidate given order would maybe more  
often be overridden by the voter. If one expects the full agreement  
with the candidate given list to be rare (say, the preference order  
of >50% of the voters is different), then one could see those lists  
just as (unofficial) recommendations that the candidates would be  
free to advertise but that would not be used as default inheritance  
orders (since that could mean that the candidate decides on behalf of  
the (unaware bullet) voter (maybe more often wrong than right)).

Juho





		
___________________________________________________________ 
Try the all-new Yahoo! Mail. "The New Version is radically easier to use" – The Wall Street Journal 
http://uk.docs.yahoo.com/nowyoucan.html