2012-03-24

Should the Pirate Party use Score Voting?

UPDATE: You can participate in a mock Score Voting election for chairman here.

The German Pirate Party (Piratenpartei) has primarily been using Approval Voting, both in single-winner and multi-winner elections. Approval Voting is primarily promoted as a single-winner voting system. For multi-winner elections, many people prefer proportional representation. But it makes sense that a political party would want its party list to be composed of relatively moderate representatives. Proportionality within a political party could produce leadership which would not work well together. Their views could be too diverse for them to work effectively. So we can see why, for the purpose of creating a party list, it makes sense to use a single-winner voting system for a multi-winner election.

That being said, there have been some complaints about Approval Voting, some seen today on Twitter:


As a single-winner voting method, Approval Voting is excellent. But that fact is often misunderstood due to complex and counterintuitive aspects of voting theory. For instance, many people naively believe that Condorcet systems such as Schulze are better than Approval Voting. But extensive Bayesian Regret calculations show that Approval Voting tends to actually produce more satisfying outcomes. You can see this in the sample graph below.


Further, tactical voting can cause Approval Voting to be more likely to elect a Condorcet winner than real Condorcet methods. And there can be severe problems when voters use tactics with ranked systems, such as Condorcet methods.

Nevertheless, the frustrations expressed by these Pirate Party members can't be completely the result of logical fallacies. These voters seem to feel like Approval Voting wasn't sufficiently representative. It didn't allow them enough ability to discriminate between different candidates. So how can we fix that, without resorting to a ranked voting system?

The answer is to use Score Voting! Score Voting (aka Range Voting) allows the voters to rate the candidates independently, on a scale such as 0-10. Here is an example Score Voting ballot with a simplified 0-2 scale.


Unlike ranked systems, Score Voting can never punish voters for supporting their favorite candidate. For tactical voters, it behaves just like Approval Voting, because they will want to use maximum and minimum scores only. But for sincere voters, Score Voting allows much more expressiveness. And it turns out that sincere/expressive voting is a pretty good strategy. So good, in fact, that many voters who want to be tactical will be better off just voting honestly, because it will be so much easier than trying to predict the outcome and calculate the ideal strategy.

Their use of advanced voting methods, as well as online organizing tools like LiquidFeedback, make the Piratenpartei one of the world's most advanced democratic institutions. My hope is that these problems with Approval Voting will not cause a step back to the problematic ranked systems, but instead will only encourage them to upgrade to Score Voting. While it is not quite as simple as Approval Voting, it is still simpler than any ranked system, and it preserves the important property that voters can never be harmed for supporting their favorite candidate. Lastly, it produces amazing results according to the Bayesian Regret metric, leading to objectively more satisfied voters.

Clay Shentrup
The Center for Election Science
San Francisco, CA

5 comments:

Demon said...

I second your opinion that using an approval voting system pushed us to the edge of an epic fail, but this was in fact due to the 50%+ approvement rule.

The election system by itself was okay, but the surrounding elements in the serry of rules were a bit problematic.

I could think of a system like yours with a range of -3 to +3 with the extra rule of "you need at least a positive overall score to be the winner".This would enable the ability of kicking every contestant out of the voting if not one person is (in the eyes of 400 members) capable of doing the job.

--
Michele Marsching
President
Pirate Party NRW, Germany

Clay Shentrup said...

Michele,

I'm truly honored that you responded to my blog post. I presume you yourself were elected via Approval Voting?

Over the years, a lot of folks have suggested a symmetric rating scale, with negative and positive values. While I don't think this would cause any catastrophic problems, we have outlined a number of reasons that we believe a zero-based scale is better.
http://ScoreVoting.net/Why99.html

You could still use your rule though. E.g. a -3 to +3 scale would become a 0 to 6 scale, where you'd need at least a 3 point average to be viable.

But what about another approach? What if you just gave voters a "virtual candidate" that they could rate? This would mean "you need at least as high a score as that candidate to be the winner". The benefit is that a voter who prefers X to Y by a large margin can say that Y is (to use your scale) a "-1", but only candidates rated "-2" or worse should be kicked out in favor of a new election. It puts that threshold in the hands of the voter, instead of arbitrarily using a middle score.

Also, I wonder what you think of the idea of Asset Voting.
http://ScoreVoting.net/Asset.html

Asset Voting is helpful in an election with a great number of candidates. Voters only have to pick the candidate or candidates who get their votes. If you like Joachim, you can give him all your votes. If he doesn't have enough votes to get a spot on the list, he can give his votes to another candidate who he knows to be ideologically similar and competent. This takes away some control from voters, but the benefit is that the candidates probably know more about the other candidates than the voters do, so it can allow more informed decisions. Asset Voting is a proportional system, so this may not be ideal for party lists. But it is certainly an interesting system that may be useful for some part of the party process.

Thanks again for sharing your thoughts. Please feel free to share any other thoughts you may have, at our discussion list.
groups.google.com/forum/#!forum/electionscience

Clay Shentrup
The Center for Election Science
San Francisco, CA, USA

Anonymous said...

Hello

First, i'd like to say, that approval voting is a kind of score voting, just with the scale of 0-1...^^

In my opinion score voting has some disadvantages:

If you vote strategically, you would either choose minimum or maximum points. Such voters are much powerful than voters who vote honestly.

Even if everybody tries to vote honestly, it is still hard to decide how much points you give a candidate, who is good but not perfect.

Different people will use their options on the scale very different. Some people might think, that 60% of the points would be for a good, but not perfect person, whereas others would give 90% with the same opinion of this person.

I hope you understand the problems, but i suppose, that every system has its disadvantages.

I would be ok with a system with a scale of 3 options:
Agree -- No opinion -- Disagree

Then you calculate a "value of agreement" for each candidate by substracting (not dividing!) Agrees with disagrees, and than you have your candidate(s) or list...

(I hope my english was ok)
--
@teilersumme twitter

Clay Shentrup said...

@teilersumme,

> First, i'd like to say, that approval voting is a kind of score voting, just with the scale of 0-1…^^

Yes. I often point this out myself.

> In my opinion score voting has some disadvantages:

I don't see how the things you cite are disadvantages. And I reiterate that the Bayesian Regret figures I cited (in the graph) show Score Voting ("Range Voting") doing better than Approval Voting with any mixture of tactical or sincere voters.

> If you vote strategically, you would either choose minimum or maximum points. Such voters are much powerful than voters who vote honestly.

This is an extremely common criticism, which I made myself when I first got into voting theory in 2006.

It's helpful to think about it like this:

1) Score Voting is better for expressive ("sincere") voters, like the people who voted for Green Party candidate Ralph Nader in our 2000 U.S. presidential election, even though they knew he had little chance. They, by definition, care more about expressing their views than about getting the most power, and Score Voting let's them express a lot more than Approval Voting. I have even had Nader supporters complain to me that with Approval Voting, they'd have to approve Nader and Gore, which wouldn't allow them to say that they liked Nader better than Gore. Score Voting would fix that.

2) Score Voting is better for tactical voters. For them, Score Voting and Approval Voting are the same. But the expressive/sincere voters voluntarily give them more power, which makes them happier too.

3) Score Voting is better for the whole electorate, on average. This is because of game theory issues. The expressive voters give up less utility than the tactical voters gain. This is an interesting property of tactical voting, which you can see in the Bayesian Regret graph in my post. Tactical voting makes tactical voters better off, but it harms other voters by more than it helps each tactical voter, which causes a net decrease in utility. By giving forcing expressive voters to only use a max and min score, Approval Voting causes more harm than good. This explains why Score Voting has lower Bayesian Regret.

4) It turns out that with enough sincere voting, Score Voting can actually be better even for sincere voters! Not just because it lets them express themselves, but I mean it actually can make them happier with election outcomes. See the results of this experiment.
ScoreVoting.net/ShExpRes.html

> Even if everybody tries to vote honestly, it is still hard to decide how much points you give a candidate, who is good but not perfect.

This seems to be a much bigger problem for Approval Voting. If a candidate is right on the line between approval or disapproval, it can be hard to decide which way to go. But people tend to find it very easy to give such a candidate a "4" or "5" or "6" out of 10, for example. In fact, this is why HotOrNot.com chose Score Voting. They wanted voters to provide as much input as possible, instead of sitting there trying to decide how to vote.

And I've done exit polling with Score Voting, on people in a small town in Texas. NO ONE spent an appreciable amount of time trying to decide how to score the five candidates. I don't think the problem you're pondering really exists to a significant degree.
ScoreVoting.net/Beaumont.html

Clay Shentrup said...

@teilersumme (continued)

> Different people will use their options on the scale very different. Some people might think, that 60% of the points would be for a good, but not perfect person, whereas others would give 90% with the same opinion of this person.

Our Bayesian Regret figures already take this into consideration, since voters are normalizing their sincere utilities in order to get their scores. That means the scores will inherently mean different things for different voters.

You can see some sample Score Voting data that we recently got from an election in a local club here in San Francisco:
www.electology.org/sf-frontrunners-2012

We're well aware of the types of voting behavior that exist.

> I would be ok with a system with a scale of 3 options:
Agree -- No opinion -- Disagree

That would be an improvement, but 0-9 or 0-10 leads to much better results, our studies find.

> (I hope my english was ok)

It was great, and certainly much better than my German.