The CES and PR: let's write a white paper.

[Jameson Quinn]

We've talked about PR before, and I think it's safe to say that we're basically positive about it. But there is nothing on our website about it. I think that should change.

I'm proposing we have both a "policy plank" and a "white paper" regarding PR. The policy plank should embrace the following points (expanded and written out)

- Gerrymandering is bad. 

- PR is the best solution. 

- There are many possible implementations of PR, and most are pretty good, as long as they have voter control (that is, no closed lists). 

- The best proposals can have other good characteristics, such as:

    - Simple ballots (by highlighting local candidates) and simple vote-and-done voting process (through MMP and/or delegation)

    - Geographical link between voter and representative (through overlapping territories such that each voter is in the territory of one rep per winning party)

    - Intraparty options for voters; nobody is strategically "stuck with" voting for a corrupt or ideologically non-compatible rep because that's the only option for their party in their district (through allowing cross-district voting through write-ins or other voting/ballot technology)

The white paper should look at who would currently be in congress under a single delegated vote PR system under some reasonable assumptions: 

- System similar to PAL, but without intraparty preferences for delegated votes, and with a single out-of-party preference for each candidate

- Most voters vote as currently. However, small parties get some multiple of their current vote, drawn from the most ideologically similar large party. Also, some small fraction of the remaining votes cross district borders, voting for any same-party candidate statewide in proportion to small-donor fundraising totals. (That makes 2 parameters - the log odds boost for small parties, and the portion of cross-district voters. Once we got the data for this, we could do a sensitivity analysis for those parameters; I expect that the outcome would be pretty stable across reasonable values.)

- We could also try out STV with Fairvote's proposed multimember districts.

I think this kind of white paper research would be a useful thing to have available on the CES website. FairVote has this kind of stuff, and it gets them links. I also think that it would be good to be able to go to state legislators and say "look, this would be fairer, but not too hugely different from the status quo; you don't have to be so terrified of it."

[Clay Shentrup]

I don't know that PR is the "best" solution. Certainly it's the most politically viable.

I generally agree with this. We should advocate for better PR systems than STV.

[Ted Stern]

On Mon, Aug 8, 2016 at 10:24 PM, Clay Shentrup <cl...@electology.org> wrote:

I don't know that PR is the "best" solution. Certainly it's the most politically viable.

I generally agree with this. We should advocate for better PR systems than STV.

Absolutely -- one could do far worse than a method proposed by Jameson Quinn back in 2011 (with some minor simplifications by me).  At the time, Jameson called it Approval Threshold - Transferable Vote, but it is really another variety of reweighted range voting with a quota-based reweighting factor.

• Voter gives a non-zero score to all candidates they approve of
  
• With N voters and M seats to fill, set a Droop quota Q = N / (M+1).
• Each ballot has an initial weight of 1.
• For each seat i = 0 to M-1, the winner W_i is the candidate with the highest weighted total score.
• To reweight ballots, determine the quota-approval threshold as the highest score level S_i at which the total T of voters scoring the seat winner at that level or higher exceeds Q.
• On all ballots that score the winner W_i at S_i or higher, reweight the ballot by multiplying by the factor F_i = 1 - (Q/T_i).  If T_i is less than Q, set F_i to zero.

This method is Droop proportional and satisfies Participation and Immunity from Irrelevant Alternatives (for the single winner case).  Jameson's original method used a different weighting for the score and approval threshold, but I think it is simpler to use a single reweighting factor.

In my opinion, the basic method can be improved by a couple of additional refinements:

• At each step, eliminate any candidates with less than Q/2 approval (where approval = total weighted number of ballots with any non-zero score), unless doing so would leave fewer standing candidates than there are seats to fill.
• Any ballot in round i that would have no more non-eliminated or seated candidates to vote for after selecting a particular candidate j, contributes to a "locked" total for that candidate, L_{i,j}.
• When candidate j is selected as winner, the reweighting factor becomes F = 1 - (Q - L)/(T - L), limited to the range [0,1].  This means that truncated ballots lose their entire weight, but non-truncated ballots don't have to be reweighted down as much.

With these modifications, there is less likelihood that votes will be "lost" due to truncation, as one sees with STV.

Other than using the total weighted score sum to select each seat winner, this is basically a generalization of Bucklin to multiwinner elections.  The primary reason to use score to determine winners is to avoid violating Participation and IIAC, and so that the single-winner method is just Score Voting (with total score vs. average score).

My current preferred name for this method would be Quota Reweighted Score Voting (QRSV).

Ted

[Kevin Baas]

Better improvements would be:

* instead of eliminating one candidate at a time, or all candidates under a threshold, eliminate the "weakest" set whose elimination produces a winner. (method for determining formula for weakness can be determined by in-silico experiments, but for approval voting "weakest" would probably just mean resulting in fewest total vote transfers, with ties broken by most candidates eliminated)

* eliminations are _temporary_.  after each round (seat filled), all eliminated candidates are re-introduced for the next round

* hare quota is used instead of droop quota.

I've explained these improvements on STV here:

http://blog.opavote.com/2016/06/guest-post-rethinking-stv-fundamentals.html

The same improvements could be made for any other transferable vote system.   But the ballots have to be normalized to have the same total score, first. (note, this doesn't mean that someone who approved four people will have a vote that counts less than someone who approved 2.  both of their votes will end up counting the same.)

[Brian Olson]

In 10+ years of studying redistricting and gerrymandering, the unexpected thing to find was that "sometimes a district isn't the right answer". People keep trying to do 'gerrymandering for good' to achieve Proportional Representation (PR). So we should just do PR.

People also still want 'their' local rep, and such a person is useful for a variety of constituent services administrative reasons, and because some issues really are locality based. Other issues are ideology or identity based, and PR suits those electoral concerns better.

The best idea I can come up with is to have two houses of the state legislature, let's call them State Senate and State House, and maybe they have 40 and 80 members respectively. The State House is districted (single member) and each of those relatively little districts is elected every 2 or 4 years. The State Senate is at-large proportional representation across the whole state. If they have 4 year terms they might be elected in two waves of 20, or 6 year terms might be elected in 3 sets of one third of the members. The State Senate ballot might be long, I wouldn't be surprised to see 60 names on it competing for 20 seats. A voter can rank as many as they care to. Just 1st choice; 1st, 2nd, 3rd; 1st through 17th; or all 60. Some people who just pick one or two might not get anything, but I can't get too worked up about that because a lot of people don't even show up to the polls. Some people just won't get what they want, but that happens anyway. I think in most forms of N-seat proportional representation there are up to 1/N people who don't get what they want, but that's a lot better than 49%.

--
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 electionscience+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ted Stern]

Brian, I agree in principle to your proposal about two legislative houses (which is essentially the same as Australia's version), but I would reverse their functions:  Senate has single-winner districts, and the House has multi-winner PR seats.  And here's why:  the PR group should provide diversity to get lots of different proposals debated in an analogy with natural genetic diversity, while the single-winner group should contain centrist winners as an analogy with natural selection pressures.

For effective single winners, a gerrymander-proof method such as yours should be used to create districts, and there should be fewer single-winner districts than PR representatives.

Ted

[Ted Stern]

Kevin, we have seen your proposals for modifying STV already.  Personally, I think you should not call your method STV, since it has substantial modifications.  Have you considered giving it a new name?  Regarding my proposal, most of your suggestions don't really apply, since there isn't any vote transfer in QRSV, just reweighting.  You do make some good points about the quota and elimination, however.

I've been looking at the quota-based reweighted score voting method off and on for 5 years now.  There is a good reason to use Droop in QRSV:  it guarantees that a majority faction will win a majority of seats when M is odd, and it is also closely related to median rating, though the winner selection uses score voting.

Elimination is not strictly necessary in QRSV.  I proposed it only because of the (Q-L)/(T-L) modification.  There are some tradeoffs.  A voter might choose a set of candidates who are completely unviable, and non-viable elimination would effectively reduce the weight of that ballot to zero.  The intention would be to encourage voters to select at least one viable candidate so they can be represented.  On the other hand, a voter who chooses several viable candidates who do win seats might also select some non-viable candidates who could never win.  Should other voters for the viable candidates be penalized even when the first voter's ballot is effectively used up?  It is not an easy question to answer.  I'm interested in hearing arguments on all sides.

Ted

--

[Jameson Quinn]

I'm always happy to discuss theoretical PR systems. But for talking to legislators, I think we (that is, the CES) should have a single proposal. And that proposal shouldn't be optimized for ideal performance on some well-crafted proportionality metric; that's fun to talk about, but generally leads to impractical systems. No, the system should be optimized for "no objections"; that is, for being as close as possible to being Pareto dominant over FPTP, better from any possible angle, even if the improvement is minimal.

Here's the system I think meets that criterion. It's what I tried to summarize in the initial message here, but with all the details. I'll use "state" to mean the entire area of the election, though this system would also work fine at other levels such as city council.

"Districts" are left intact. Gerrymandered districts are fine, though in the long run probably nobody would bother, as their role is largely cosmetic.

Before the election, each candidate declares their party, and also may declare one candidate from outside their party who will get their votes if their entire party is eliminated. Parties may have membership criteria if they so desire, to prevent "false flag" candidates.

Ballots are "choose one". They differ cosmetically by district. Each ballot lists the candidates residing in the local district first; then, any incumbent candidates whose territory includes the local district; and finally, the rest of the candidates, sorted by party, and perhaps in a smaller font (or, if ballots are marked by touchscreen and printed out, in some kind of subsidiary menu). Thus, most "lazy" voters will vote for the local candidate from the party of their choice; while more "engaged" voters may find the same-party candidate whom they like best.

It takes a Droop quota to win. That would be the majority of votes from any one district, so if most voters are "lazy", it is unlikely that any candidate will have a quota to start out with.

Candidates are eliminated in ascending order of the number of direct votes they got. When a candidate is eliminated, their votes are split equally among any remaining candidates in the same party. If this takes any candidate(s) above a Droop quota, that candidate is elected, and any overvotes are transferred as with elimination. When the last candidate from a given party is eliminated, those votes are transferred according to the predeclared preference of the original candidate they were cast for. If that "target" has already been eliminated, the votes continue on one step further, to the predeclared preference of the target. If that second target has also been eliminated, the votes are exhausted, and the quota is adjusted accordingly. If the original candidate did not declare a preference, the votes are also exhausted.

This procedure is enough to elect a full slate of candidates. The process of voting is quite simple. It allows cross-party transfers of votes, but would not lead to wins by micro-parties like the Australian "Motor Enthusiasts Party" because such candidates would be eliminated early on. (In fact, even sizeable third parties would need to either have a concentrated geographic power-base, or to have voters engaged enough to avoid splitting their votes, if they wanted to win.)

Once all seats are filled, each party would assign each of its elected representatives a "territory" of one or more districts, so that each district would be in the territory of one representative from each party. The "territories" within a given party would have to be roughly equal in terms of numbers of that party's voters. (That is, it should not be possible to make them more equal by changing one district from one rep to another.)

Why do I say this is "Pareto dominant" over FPTP? Because it avoids disadvantages common to other PR systems:

-Keeps simple ballot format, accessible to both "lazy" and "engaged" voters.

-Avoids any closed lists; choices rest on the ballots of the voters.

-Keeps geographical link between voters and candidates/representatives.

-Keeps clear chain of responsibility; any voter can point to "their" representative (the rep from their favorite party whose territory includes their district).

-Avoids "mixed members" winning in two different ways.

-Allows engaged voters to help determine which subfaction of their party gets more seats.

-In a state which happens to already have a proportional result, where all voters are "lazy", and where third parties have no more than 1/3 of the voters in any one district, this would give exactly the same results as FPTP. That is, it's as unthreatening to incumbents as possible for a PR system.

This system is similar to PAL, but it simplifies the predeclarations and the within-party transfers. I'm tentatively calling it Delegated Regional voting, DR voting ("Just what the doctor ordered to fix gerrymandering.")

[William Waugh]

There's also proxy assignment.

[Andy Jennings]

My opinion on PR and CES:

I look at FairVote's "innovations" page:
http://www.fairvote.org/innovations

and I see two PR reforms at the bottom to which I have no objection:

http://www.fairvote.org/districts_plus
http://www.fairvote.org/open_ticket_voting

On the other hand, clicking on "Proportional Representation" at the top you see a page with lots of sections and different ideas: http://www.fairvote.org/proportional_representation

The video highlights one idea they've been pushing recently, redistricting the states into fewer, larger multi-winner districts with PR.  It doesn't really say what the actual election method is.  Is this their preferred reform?  It seems to be on their 2020 roadmap.  I'm not super excited about it, but I suppose I would support it as better than what we have now.  I'd prefer more winners per district so even smaller parties get some representation.  And I don't see how it's wise to start with the US House of Representatives.

Anyways, is there a reason to forge our own path on this issue?  Can we just sign on to one or more of their reforms?  For the sake of momentum.

Of course, once there are more than two parties in the legislature, if none have a majority, it seems like they will find a majority coalition to govern.  Will it be any better than what we have now?  Will the minority get the cold shoulder?  At that point, I think it's important to solve the "caucus-resistant speaker" problem:
https://groups.google.com/d/topic/electionscience/DJJQ2a0q9GY/discussion

[Jameson Quinn]

FairVote  wants multi-member districts with STV. I'd support that idea if it were politically feasible, but I think that it isn't. 

Here are the various objections, in no particular order. Note that I say "the" objections, not "my" objections; none of these are deal-breakers for me, but I think the combination of them adds up to a deal-breaker in terms of feasibility.

• Complex STV ballots
  
• Hard for voters
• Many voters would blindly follow party list recommendations, making the list-writers into unaccountable power-brokers (issues with money in politics here)
• Leftover disproportionality (rounding error) is still notable.
• No direct "chain of representation". If I want to write to "my representative", it's unclear who that is.
• STV weirdnesses (premature elimination, a la IRV)
• Voters have if anything even less power than today to pull their party in a given ideological direction. Frequently, the two marginal candidates (that is, the weakest winner and the strongest loser) from a given party in a given district will both be from the same ideological faction, meaning marginal ballots in that district carry no effective ideological valence. Obviously, the preceding sentence is a technical explanation of the problem which most people wouldn't understand, but I think people would still get the feeling of "why bother".
• For several of the above reasons, I don't think it would be as good for turnout as other PR systems.
• There would still be issues of gerrymandering around candidate residency. And furthermore, it would be impossible to "smooth the transition" fully; unless it coincided with redistricting, implementing this system would inevitably put more incumbents out of a job than simply the number of seats which switch parties. To me, any churn that is not a result of better proportionality or more accountability is a dead weight loss, in terms of being able to get incumbents to support the reform.

So what does that mean for us? I think we should definitely "sign on" as supporting their effort. But I think we could have a useful role in bringing new ideas to the table as well, and I think it's worth doing that insofar as our ideas are more politically feasible.

I totally agree on the "caucus-resistant speaker" thing. One offhand idea there: elect the speaker by graded Bucklin (that is, simplified MJ, where the tiebreaker is simply the above-cutoff count), with the ballots remaining secret and only the overall tallies announced.

[Drew Spencer]

At the time that video was made, we were supporting an approach that would allow states to choose any proportional or semi-proportional voting method, which is why we were vague on the method. Since then we have become more focused on STV, though our proposal would allow what we call the "open ticket method" (unordered open list PR) in states where it wouldn't be feasible to administer STV.

This page describes the proposal in more detail: http://www.fairvote.org/fair_rep_in_congress#fair_rep_act

And this memo attempts to capture it in a two-page summary: https://fairvote.app.box.com/v/fair-representation-act-memo

Andrew Spencer

[Jameson Quinn]

Drew:

Thanks for commenting. (For the rest of you, here's the link I found for FairVote's Open Ticket Voting page. It's basically multimember districts, open list, vote-for-one, where a vote counts for both candidate and party, and candidates can designate a secondary party in case their party is eliminated.)

What would it take to get FairVote to take a serious look at the system I've called DR voting? This is basically the same as what you've called Open Ticket, with a few additions/changes:

-The entire state would be counted as a single election.

-However, the state would still be divided into single-member-sized districts. The ballots from each district would list that district's candidates first, in a larger font size, to simplify voting for low-engagement voters.

-After the election, each winner would be assigned a "territory" of one or more districts, such that each district would have one representative from each winning party; equivalently, each party would have one winner covering each district. 

I'm not attached to the DR name. If you want to call it "open ticket territorial" or whatever, that's fine with me.

If FairVote would be interested in cooperating with us at CES in writing a white paper suggesting who would have won the latest election under this system... well, I can't speak for CES definitively, but I at least would want to pursue this possibility. 

[Drew Spencer]

Hi Jameson,

First, I should say that we're very hesitant to advocate for something that hasn't been used before, so I wouldn't be optimistic that we'd be on board.

But also, I'd like to at least understand the system you're describing, and I don't think I get it from this description. Is there a website where it's written up?

Thanks,
Drew.

[Jameson Quinn]

Here's an infographic I made about DR voting. (This was actually originally about PAL voting, a similar but slightly-more-complex system. I've updated it to refer to DR.)

Since it's based on Open Ticket, it shares the same advantages. But the add-ons give it some extra advantages:

-Simpler voting for low-engagement voters, but wider options for high-engagement voters.

-Doesn't require redistricting.

-None of the edge-effects, rounding errors, or possible gerrymandering that come from carving the state into multi-member districts.

-Clear chain of responsibility; each voter can know who their representative is.

-Locality is maintained; neighbors from the same party will share a representative.

-Minimal distruption: only changes the FPTP result by minimum needed to maintain proportionality, so if it was already proportional, there would be no change. This is helpful in convincing incumbents to support it.

As for your comment about it being "something that hasn't been used before": that's true, but then, I don't think that Open Ticket has been used in exactly the form described on FairVote's website either. They're both clearly based on well-understood ideas that have been used before — basically, open list single-vote systems — but both have attempted to combine the best details of various systems and proposals.

If you have any further questions, I'd be happy to answer.

Jameson

[Brian Olson]

Responding to Jameson's proposal:

A pick-one ballot is weak, that's what we have now, it's a huge part of the problem, hacked up machinations on the back end aren't a fix for that.

The hacked up back end enshrines party power because transfers are done based on parties. I want less party power, which is why I'd rather have a candidate based PR instead of a party based PR, so I'd rather have STV where I directly vote for candidates.

---

STV has IRV style non-monotonic/discontinuous zones, but based on a few election space graphs (Ka Ping Yee diagrams) I've done they're smaller than IRV. I think the multi-winner small-quota nature of STV "takes the pressure off" the mechanisms in IRV that cause discontinuities. I have another candidate-proportional method I'm working up based on ratings ballots that still has runoff-discontinuities but they're smaller still. (busy busy busy, I'm working up a new simulator for multiseat elections that probably won't be ready for a while)

[Toby Pereira]

When we were discussing the Canada reforms, we discussed mixed-member proportional, and I think that, implemented properly, it works as a good compromise.

Each constituency/riding (or whatever you call it) has one representative, so this is the local representative of everyone living there, regardless of who they actually supported. But then there is a wider region as well that rebalances things in a more proportional way. I'm not sure that I personally need the idea of a local representative (who could be from a party I hate), but from having spoken to people and read various things, it does seem to be quite a popular idea. That's why I think that mixed-member is quite a good compromise.

The proportion of representatives that are local representatives, as opposed to belonging tot he whole region, would be up for discussion - it could be a minority, majority or exactly half. The fewer local representative there are, the larger the area of their local representation.

Basically what I put here - https://groups.google.com/forum/#!msg/electionscience/aP7ybKMb1zs/giaYAh6wAwAJ (reproduced below):

"I'm going to clarify my latest method, based on Warren's 13-riding district system with 5 top-up MPs. The numbers 13 and 5 aren't essential parts of the system, but it wouldn't work with too many ridings as the ballot paper would get too large.

1. Each riding's ballot paper would list the candidates standing in that riding, along with any party affiliation, with a way to give the candidate a score. This could be a box to enter the score, or a list of potential scores in a line with the voter to circle one.

2. Next to or underneath each party candidate, there would be a separate mention of their party, which can be separately scored.

3. Underneath the list of the ridings candidates (possibly separated by a line) would be a list of all independent candidates standing in the whole region as well as a list of all parties fielding candidates elsewhere in the region but not in this particular riding. These would also be scorable by voters.

4. Voters can give scores to as many of the listed candidates or parties as they like, and can ignore as many as they like. The score for a candidate and their party do not have to be the same.  If a voter gives a score to a candidate but not their party or vice versa (so one is left blank as opposed to being given an explicit zero), then by default the same score is applied to both. Other than this, blanks are taken to be zeros.

5. The scores given to the ridings candidates on the ballots from that particular riding are added up and the highest scoring candidate is elected as MP for that riding.

6. The top-up phase commences. In addition to the scores given to the candidates from voters in their ridings, all scores are now considered. Any score given to a party counts for all candidates in that party from the region, apart from the scores explicitly given to ridings candidates if they are different. The scores explicitly given remain as they are. Scores given to independent candidates from outside their riding are also now considered.

7. The rest of the seats are now allocated using a proportional score system. The seats that have already been allocated are taken into account, so that the result is the most proportional it can be given those already elected.

END

Just a note on scoring candidates and their parties separately, I'm not sure what the cleanest look would be. Such as:

Candidate name BOX Party name BOX

or

Candidate name 0, 1, 2, 3, 4, 5, 6, 7, 8, 9

Party name         0, 1, 2, 3, 4, 5, 6, 7, 8, 9

or even

Candidate name 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 Party name  0, 1, 2, 3, 4, 5, 6, 7, 8, 9

It makes sense to try and get them on the same line if possible because the ballot would have quite a few names on it. I still think the size would be within acceptable limits without having to go to delegated votes/asset voting or only allowing voters to rate candidates in their own riding." END QUOTE

I've suggested score voting, not just because I generally prefer it for proportional elections, but ranks wouldn't really work in this mixed member scenario, where you have to mix in two separate sections of the ballot for the proportional phase.

Also, other than your single local representative, voters don't have one single representative from the larger region. I actually prefer this. They all belong to everyone.

[Toby Pereira]

The problem with STV is that you have to list all the candidates on the ballot paper, which limits the size of the regions and therefore the level of proportionality.

Having some sort of shortcut compromise might not be such a bad thing. People can still vote for independent candidates. But party candidates are still considered together to some extent, and the x candidates elected from a party should still be the most popular x candidates from that party. So it doesn't necessarily give parties an advantage over independent candidates.

Where Jameson's proposal differs from mine is that my proposal still gives a single local representative, as well as several more from a wider region. Jameson's system attempts to get national proportionality with no set number from any particular region, and voters work out for themselves who "their" representative is - based on which representatives their vote would have been transferred to and the locations of these representatives.

[Toby Pereira]

How specific should this paper be about proportional representation? Should it specify a particular system of PR deemed to be the "best" or narrow it down a little bit or eave it fairly open? Jameson has outlined a specific system, but I think it will be hard to get overall agreement on one system with that much detail.

[Jameson Quinn]

Part of the reason I proposed the system I did is that, under such a system, ballots and votes would look similar to what they do now. Thus, with relatively light assumptions, we could actually infer specifically who would have won under such a system. This would make a white paper meatier, and give natural headline hooks so that it would draw more media attention. Furthermore, we could approach actual incumbents and, in many cases, argue "under the system we're proposing, you'd be able to keep your seat, as long as you remain as popular as you were last election".

It's true, there would be some incumbents whom our white paper would show as vulnerable / proportionally illegitimate, and we wouldn't be making friends of them. But I think it's better to find a way to get that out in the open. If we simply propose an entirely different system, every incumbent will feel it endangers them. It's better to focus those fears down to a relatively few specific cases, then allow them to remain a hazy and ubiquitous barrier.

2016-08-30 0:14 GMT-04:00 'Toby Pereira' via The Center for Election Science <electio...@googlegroups.com>:

How specific should this paper be about proportional representation? Should it specify a particular system of PR deemed to be the "best" or narrow it down a little bit or eave it fairly open? Jameson has outlined a specific system, but I think it will be hard to get overall agreement on one system with that much detail.

--