[EM] Sortition and the Delegable Proxy system

Bryan Mills bmills at alumni.cmu.edu
Mon Jan 30 22:05:43 PST 2012

On Wed, Jan 25, 2012 at 12:52 PM, Kristofer Munsterhjelm
<km_elmet at lavabit.com> wrote:
> On 01/24/2012 07:28 AM, Bryan Mills wrote:
>> I've been looking at a voting system over the past week or so that I
>> think is really interesting: a combination of the "delegable proxy"
>> system with a sortition procedure to elect a standing legislature.
>> My objective is to find a way to use conventional voting
>> infrastructure to elect a proportional legislature of bounded size by
>> strategy-free means.  I'm not yet 100% certain whether the system
>> actually is strategy-free; I think it is but I haven't yet found a
>> proof.  (It's non-deterministic, so I don't think it runs afoul of
>> Arrow or Gibbard–Satterthwaite but proportionality is only
>> probabilistic.)
>> I can't imagine that I'm the first to examine this system, but I
>> haven't found it in any of the voting literature I've read so far
>> (most of Voting Matters and part of the Electowiki).  My own writeup
>> can be found at
>> (https://docs.google.com/document/pub?id=1XfoGtx2HBYNyZSYwwiQlcXU7mq_WkLhQzdghv8bGF4o).
>> Any direct insight or pointers to relevant documents would be appreciated.
> I think it is strategy-proof, but I wonder if people would irrationally
> reason something like this:
> "My chance of winning is very low, so I shouldn't keep my hopes up. Instead,
> I should delegate my vote so that I can feel I expressed myself if [popular
> candidate] wins."
> Now, that makes no sense, but if people were game-theoretically rational,
> turnout would be very low (and it isn't). So I'm wondering if the people
> would irrationally be more mass-like than you'd want of a sortition-based
> system.

Moderate clustering of votes is desirable, and leads to lower-variance outcomes
(because the count proceeds further on average before hitting the
max-seats limit).

That's among several reasons why you would want a large number of seats;
probably on the order of 400-500 for a practical legislature.  Voters
could hopefully
find a fairly close match among the several-hundred front-runners --
much closer than
would be possible in a single-winner or even a typical STV election,
especially since they
wouldn't have to spend time figuring out a rank-ordering of
less-preferred candidates.

> There might also be another problem: say that people delegate their votes so
> that some famous candidates have a certain chance of winning, but this
> chance isn't absolute. Then any of them could say "if I win, that's great,
> I'll do what I want". In other words, the doubly random nature of the
> process - first, where the candidates don't know who will delegate to them,
> and second, where it's not certain they will win even if lots of people
> delegate to them - could weaken the constituency feedback, making the
> "famous" candidates more likely to do their own thing than to take the
> voters' opinions into account.

Hmm, that's interesting.  Bu it's unlikely that a candidate who
receives a substantial
fraction (e.g. a Droop quota or more) of votes would fail to be
elected in practice, and
any factors that increase that risk proportionally increase a
candidate's prospective
voting power if elected.  I do need to do some simulations to verify
that outlier events
are sufficiently rare with a practical number of seats, though.

In general, I don't think one can solve the "elected candidates ignore their
constituency" problem completely with any long-period election system.
 If you want
to solve that problem you'd need voters to be able to change their
delegations midway
through a term, and while I think that's a very interesting line of
investigation it doesn't
satisfy my initial objective of "conventional infrastructure".  (That
is, any system that
completely solves the ignored-constituency problem presents more substantial
technical barriers to adoption than does DS.)

> Or, the public, knowing this, might delegate to people of which they have a
> general good feeling, e.g. movie stars and the likes. They would then reason
> something like "I know that whoever I delegate won't know I helped him win.
> Thus I can't influence him, so who am I going to choose? Well, I know of
> Harrison Ford (or whoever) and his political position is somewhat close to
> me, so I'll pick him". Now, Ford's popularity as a actor has, in itself, no
> bearing upon his political skills - he would have a greater chance of
> winning for similar reasons to why, when asked for a random number, people
> tend to pick odd numbers.
> Let's try to formalize that a little. Ordinary sortition works by picking a
> representative sample of the people. However, if the people were
> unrepresentative in their delegation - preferring those who were, for some
> reason or other, well known, then the sample picked by delegated sortition
> would not be representative anymore. In ordinary proxy democracy (liquid
> democracy, etc), on the other hand, giving your vote to Ford is next to
> pointless - if lots of people do that, you have no chance of pulling him in
> your direction, so you should vote for someone more local instead.

The idea isn't to pick a representative sample of people, but rather a
representative sample of people's preferences for representation; it's not
necessarily a problem if the representatives are not themselves a perfect
sample of the electorate.  For example, if representatives turn out to put
more-careful-than-average consideration into their political and economic
opinions, everyone might end up better off.

It may be fairly unlikely that your vote would "pull the candidate in
your direction",
but that's kind of the point of using a proportional system instead of
a single-winner
district system.  Rather than attempting to move the position of a
consensus candidate,
voters can instead seek a candidate whose views are already suitably
close to their own.

I think the real concern, then, is that people might vote for
candidates who turn out to be
systematically biased *in a way that the voters didn't anticipate*,
and I think that just
brings us back to the issue of wanting to change delegations midway
through a term.

More information about the Election-Methods mailing list