[EM] STV+AV

[Bryan Mills]

>>> > Why STV? The original poster wanted elected representatives to have
>>> votes
>>> > proportional to their electoral support yes? There's no need for
>>> fractional
>>> > transfers from elected candidates then.
>>> > >
>>> >
>>> > IRV is a form of STV, but it's not my favorite.  Some of the other STV
>>> > methods (e.g. Schulze-STV and CPO-STV) tend to produce better
>>> eliminations.
>>> >
>>> > But the question of why not STV is a good one.  Several reasons.
>>> >
>>> > STV requires much more work on the part of the voter - ranking all the
>>> way
>>> > down to a candidate likely to be elected, instead of just one.  That
>>> > probably means a much larger ballot and/or an arbitrary cutoff between
>>> > ballot-candidates and write-in candidates.
>>> >
>>> dlw: If the number of possible rankings is the number of seats + 2 then
>>> it's not too bad.  And nobody would be forced to rank umpteen candidates,
>>> so the low-info voters could just vote for their favorite candidate.
>>>
>>
>> The number of possible rankings is quite a lot larger than S+2.  Even if
>> you don't transfer votes from elected candidates, there are still C-S
>> candidates eliminated -- so you'd have (C choose C-S-1)*(C-S-1)!
>> distinguishable rankings, and even more if you allow equal rankings.  The
>> only way out seems to be to pre-filter the set of candidates, so you
>> basically have to drop to approval voting at some point --
>> candidate-registration petitions and the like -- and then we're back to an
>> arbitrary cutoff.
>>
>
> dlw: You misunderstood me.  If voters are only permitted to rank S+2
> candidates then it's not as bad for voters.

I don't see how arbitrarily reducing voter options is "not as bad for
voters"; especially if S is small, only allowing one to rank S+2
candidates seems like it would result in either frequent undervotes or
an increase in the need for strategic compromise-ballots.


>> Partial rankings might be workable in a weighted-seat STV variant, though.
>>  If a vote only transfers in case of elimination (and not in case of
>> surplus), one would only need to rank candidates down to the first
>> candidate sufficiently likely to be elected, and you could split the ballot
>> into manageable chunks by party.  Determining a suitable cutoff candidate
>> still has a cognitive cost, but it probably wouldn't be that bad in
>> practice.
>>
>
> If there are 3-5 seats STV then the number of candidates won't proliferate
> too much and there'd be 5-7 places to vote.  This would keep things
> reasonable.

To get reasonable proportionality with only 3-5 seats per district
you'd probably need to go to an MMP system, with all its added
complexity.  Otherwise Droop proportionality doesn't buy you much over
FPTP; with 5 seats the Droop quota measures to a precision of ~17%,
and the remaining 17% in each district is still susceptible to
gerrymandering.

If you assume two major parties with ~40% of the electorate each, that
means that the 5th seat in each district is noisy -- but it's not
random noise, it's systematically biased by the parties' voting
strategies and the choice of district boundaries.  Larger districts
allow finer-grained Droop quotas and thereby reduce that noise.


>> But if we assume that partial rankings are effective, there's still the
>> strategy/computation tradeoff to deal with: allowing truncated ballots
>> still doesn't help with favorite-betrayal, and STV variants less
>> susceptible to favorite-betrayal are also less susceptible to efficient
>> counting.
>>
>
> dlw: Truncated ballots may not end favorite betrayal, but it'll help with
> it.

I don't see how; please elaborate.


>> > The STV variants that are less strategy-prone are computationally
>>> > inefficient, and even those are not strategy-free.
>>> >
>>> > And perhaps most importantly, the more resistant an STV method is to
>>> > strategy, the more complicated it is to explain and understand.
>>> >
>>> > As deterministic methods go, I do like STV methods; but DS fixes a lot
>>> of
>>> > the worries I have about them.
>>> >
>>> dlw: One could also apply the same sort of approach to simplifying STV
>>> with the
>>> initial treatment of all of the rankings as approval votes to get the
>>> number of candidates down to N+2, where N is the number of seats.
>>> As with IRV, it's easier to explain STV when there's relatively few
>>> candidates to eliminate.  And, it'll mitigate the strategy effects, which
>>> have to be examined more closely.
>>
>>
>> The initial treatment of rankings as approval votes introduces some other
>> problems, though.
>>
>> With an explicit "approval threshold" in the ranking, it induces a
>> substantial cognitive cost on the voter (determining the approval threshold
>> strategically).
>>
>
> dlw: Once again, if the no. of seats isn't that great then they'd not have
> to sweat it too much.  Do I rank 1, 2, 3 or 5 candidates?  Who can I live
> with?I'd say it's an empirical question whether such would be a reasonable
> demand on voters.  And I'd reckon they'd get the hang of it with some
> practice...

The question of how many candidates to rank isn't a function of the
number of seats, it's a function of the number of candidates.  Even
with only 5 seats, you'd still need to either compromise your vote or
rank all of the candidates between your most-preferred and a
front-runner.

That is, the problem is in reducing the candidate pool, not reducing
the number of seats.


>> With an implicit "first-preference" approval, it has the same problem as
>> traditional STV (i.e. IRV), namely of unduly rewarding favorite-betrayal.
>> With an implicit "all-ranked" approval, the overall system would likely
>> violate later-no-harm with much higher frequency; by expressing a
>> preference between two dispreferred candidates one might unintentionally
>> put the higher of the two in contention.
>>
>
> dlw: I'd say empirically we'd see just how high of a frequency LNH would be
> violated.  Jameson Quinn had a hard time coming up with a pathological
> example for IRV3/AV3 and I imagine it'd be similar for the above.  The 1st
> stage would reduce the number of candidates to N+2 and it seems likely that
> the N+2nd and N+3rd candidates in terms of "all-ranked" approval are less
> likely to be among the N winners.

Hmm, ok.  I'm operating on the assumption that voters will vote
strategically if doing so is easy, and will vote approximately
honestly if strategic voting is difficult.

We're taking the top S+k winners and running some ideal STV method on
them; let's try to find an "easy" strategy.  Here's my idea:
1) Gather a set of related parties to form a majority-coalition.
2) Have the coalition propose exactly S+k candidates.
3) Ask coalition voters to vote for all of the coalition candidates in
any order they choose.

Since a majority of candidates approve of every coalition candidate
and disapprove of every competing candidate, the coalition candidates
win the approval vote.
By adding the "approval" phase to the STV election, I'm able to turn a
simple majority into a 100% supermajority.

Is there a flaw in my strategy?  (I don't think there is, but I may be
missing something.)  If not, we'll either need to abandon a fixed
limit on the number of candidates or we'll need something more
sophisticated than a simple approval-vote to filter them.


>> It may well be that these issues are all less severe than in the
>> deterministic alternatives to STV, but I still think they're enough to
>> merit consideration of nondeterministic alternatives.
>>
>
> In terms of the US's political culture, nondeterministic alternatives are
> not going to happen anytime in the near future and we need electoral reform
> ASAP!!!!

Sadly, I think both nondeterminism and STV share the "not going to
happen in the near future given political culture in the US"
classification, given that US law requires single-winner FPTP
elections for federal representation and the major parties (who
control the legislature and benefit greatly from FPTP) have no
incentive to change that law.

So as far as I can tell the only option for meaningful reform is a
constitutional amendment, and that means reforming 75% of the states
as a first step.  This is not a short-term process.

That being the case, I'd rather look for systems that are likely to be
unambiguously successful at the state and local level rather than
compromise-reforms with unclear benefits.