Berkeley + RRV (Article)

Hi folks,

I just posted my article on the Berkeley RRV process for referrals to the City Manager.

I testified in support of RRV on Tuesday before the City Council.
I’m on at 13m and 42m
Not my best - I had about two hours to prepare for this as I had little notice of this issue coming up. Prepared for next time though!

Would love your thoughts and retweets.

3 Likes

No comments? I hate to bump, especially as an admin, but surely there’s something wrong with the article.

Well, describing councilmembers as “spending” a proportion of their vote share suggests that they have a limited number of points to allocate.

Also, how would rating by department make it harder for a large coalition to dominate the top priorities of each department? It seems like that should do the opposite.

Good on Berkeley for allowing each district to have some of their concerns prioritized rather than letting the most powerful coalition dominate the rankings, though.

2 Likes

I know @ClayShentrup has strong opinions on not splitting out prioritization. I’m ambivalent.

I’m sure that the incentives to give out fewer stars than honesty suggests makes more sense in RRV than in single winner, because giving 2 stars (out of 5) to a candidate you dislike makes your vote 0.714x as powerful as if you had given that candidate 0 stars. (In single winner range voting, either that move does nothing or else that particular candidate was that close to another one that your vote could sway the difference.)

The problem with splitting by department is that political power isn’t separated by department. If the majority faction gets, instead of the first spot, the first d spots (where d is the number of departments), then you’re arbitrarily delaying the re-weighting based on the number of departments.

It’s completely fine for a particular faction to dominate a particular department. If the “YIMBY” faction is focused on housing issues, it’s perfectly reasonable for them to dominate the housing department, if that’s where they want to put all their eggs. That’s the whole point of PR.

1 Like

@fsargent Thank you for writing this! Rated proportional methods don’t get enough attention.

It’s important to note that voting for every item does not mean that the district was voting unwisely. Voting for all items but one gives the same relative impact to the weighting as voting for no items except one.

This is true in score voting but not in proportional methods. A vote for all but one proposal does not cancel with a vote for one proposal. This is one of the disadvantages with proportional methods. Almost (there are a few exceptions though they still do not encourage voters to vote for too many candidates) all both rated and ranked proportional methods are vulnerable to some form of free riding where voters can get a better result not voting for their favorite if they think their favorite is going to win anyways (however this isn’t nearly as bad as the reverse case where voters are not voting for their favorite because they don’t think their favorite is going to win). As a result, the more winners you elect under a proportional method, the smaller the percentage of the candidates strategic voters will vote for and the more powerful a vote for one becomes over a vote for all but one. Though what’s interesting is that the city council members gave high scores to large portions of the options despite this.

Also, you linked to the wrong video. The 5pm video contains your speech at the city council. The 6pm one does not.

2 Likes

From the video, one of the ways they wanted to improve the RRV process was to create two categories based on their estimated time of completion: quick priorities and time consuming priorities which would be voted on separately. Under RRV, a long term project and a short term projects would count towards each voter’s weight the same.

RRV was designed for elections where each candidate has the same weight in a legislature, where as some priorities are estimated to take alot longer then others. The estimated time of completion for priorities could also be factored into the RRV equation: 1/(max_rating + rating * time + rating * time + rating * time …). Though the biggest problem with this is it becomes dependent on what time you are using. If you are using years, each voter’s weight will not be reduced very much and if you use seconds, you can pretty much ignore that +5 as it becomes irrelevant. Perhaps the way to fix this would be to change the formula to 1/(max_rating * λ + rating * time + rating * time + rating * time …) where λ is dependent upon the estimated times. Perhaps it should always equal the average time of the priorities chosen thus far in that round of RRV? The common values for K in 1/(K * max_score + sum_scores) are based on lower bound droop quotas and upper bound hare quotas, so what should an equation for λ be based on? Any equation for λ should probably reduce to RRV (with K=1 like in normal RRV) when all the time estimates are the same and the average approach does meets this criteria…

Though despite that flaw, most people at the council seem pretty happy with the RRV and while they don’t think it’s perfect (it wasn’t designed for priorities taking a lot longer then others), it looks like they don’t want to get rid of it and just want to find ways to improve upon it.

1 Like

Ugg, this forum is doing that weird thing again where it maxes everything italicized. Is there an escape key for using the _ (underscore) ?

Edit: I fixed the issue. It wasn’t the underscores, it was the *'s and I fixed it by putting spaces before and after each *.

This sort of voting situation is very far from the ones that I have been studying. This kind of voting is what I would deem to occupy the ‘casual’ end of the spectrum, very unlike the kind of voting that occupies what I would deem the ‘political’ end of the spectrum. Political elections operate on a relatively very large scale, and are always entangled with brutal class interests.

These rather small casual voting situations tend to operate in a single room, and the voters can often negotiate legislation and planning ‘on the floor’. There will still exist class issues, the salary class, the wage class, financial lobbying, and so on. But I presume the decision-making procedures would differ drastically from those of the massive political events.

So though I’ve not begun to give this any decent amount of analysis, I do have a few ideas. Firstly, I am a bug about keeping everything a simple as possible, with very low information traffic. And weighted range voting involves a bit too much math for my taste. Addition does tend to be simpler than division, and I have questions about whether the voting participants will have a strong understanding of what the outcomes of their votes will really be with reweighted votes. So here’s my tentative suggestion.

They meet. The ‘dealer’ lays a set of notes containing various proposals out on the table. Each Councillor has 100 votes that he or she can bid toward the passage of each proposal. Then the dealer adds up the votes that have been bid on each proposal. Some proposals will presumably have more votes than others. The Councillors contemplate the results, and probably some of them add some new ‘compromise’ proposals. Then they do a whole new round, where each Councillor again has 100 votes. And these rounds repeat until a majority of (or more of?) the Councillors agree that the compromises are satisfactory.

[Addendum] They only possess 100 votes in each round, which they can only distribute. They have only the 100 votes to bid. So if a Councillor bids one vote to each of 100 proposals in one round, he or she has finished voting in that round.

Just my two cents.

?

\* gives * and \_ gives _.

But what do the 100 votes actually do?

YIMBY stands for Yes In My Back Yard. It’s the opposite of NIMBY, people who don’t want more housing to be built in their area.

You would need someone impartial to estimate how much time each priority would take, though. Also, should resources besides time be taken into account? Should an expensive project be weighted the same as a cheap one?

=/ But what do the 100 votes actually do? /=

When will people ever stop pestering me with these annoying little details?

I was thinking of just leaving it to people’s imagination. But some want ‘closure’. So (for example):

After the final round (with this Multi-Round Bid Voting (‘MRBV’), when a majority decides to conclude the session, every proposal that has ‘majority approval’ will be passed. This majority approval occurs when a proposal has acquired a pre-established minimum number of votes (an effective majority).

Presumably that minimum would be pre-established as the number of votes alloted to each Counselor divided by the number of Counselors. If each Counselor is allotted 100 votes per round, and there are seven Counselors, then such a majority would require [(100*7)/2 + 1] votes (per proposal, after the final round).

Should the number of Counselors for this formula be the the number of elected Counselors, or the number who appear at the session? The former rule would establish a neat form of quorum policy.

If the passed proposals should later turn out to be mutually incompatible, then perhaps 20% of the Counselors should be able to require a new session. Establishment of a ‘budget’ would presumably be among the first of the proposals, and if the cost of the passed proposals were to exceed that, there would exist a condition of incompatibility. All this seems rather comprehensible and workable to me.

I said:
=/ If each Counselor is allotted 100 votes per round, and there are seven Counselors, then such a majority would require [(100*7)/2 + 1] votes (per proposal, after the final round). /=

This is probably not very good, since a lot of preferred proposals would probably not be able to be passed. Back to the ‘drawing board’.

(I did mention “I’ve not begun to give this any decent amount of analysis.”)

This is very interesting so thank you for posting this.

Regarding RRV specifically though, I consider it to be an obsolete method, since Harmonic Voting http://scorevoting.net/QualityMulti.html#pertrans does a better job. It is still based on the same Proportional Approval Voting, but converts the score ballots to approval ballots first rather than working directly with scores (which can give weird results).

Edit - I posted about RRV’s problems here: https://groups.google.com/forum/#!msg/electionscience/BURkyIxZaBM/GdAcH2z7XkEJ

I doubt that an optimization method would be accepted by the public. I have always thought of such methods as more of an academic exercise than something practical. But if it can get in that would be big news.

He was talking about specifically the way harmonic voting uses the middle scores in the score ballots.

Harmonic voting is simply approval voting + the KP transformation.

You can also get the sequential version of harmonic voting by combining the KP transformation with sequential proportional approval voting. Doing this yeilds a new sequential voting method (of which I’ve occasionally called sequential proportional score voting to be consistent with the SPAV name) that is a huge improvement over RRV, though it is also not as well known and twice as complicated.

Instead of explaining just one voting method (RRV), you now have to first explain how SPAV works and then explain how the KP transformation converts SPAV into SPSV. If you’re going to go that extra mile anyways then it’s worth considering just aiming for the stars with harmonic voting instead.

It’s possible. A lot of people who advocate for IRV don’t even know how it works (they think it works something like Bucklin) so it’s certainly possible.

I’m not sure it is significantly more complicated though. Although you have to do the KP transformation on the original ballots, it’s only done once at the start, and then you’re left with SPAV, which is much simpler than RRV. So I think the complexities cancel out somewhat.

I think you guys are missing the place where people who are bad at math will have the problem. Trying all permutations of winners to find the set which optimizes a metric is very hard for non-math people to get their head around. Sequential systems with a method to choose winners have a much simpler explanation. This was why I made my system that way. I do not think it is a good idea to even get into the KP transform in the explanation. Try to think of explaining it beside somebody who is advocating for STV. Which will the average person go for?