# How does STAR Voting compare with plain Score Voting?

I've been reading a lot about voting systems, and Score Voting seems like a pretty good system, but Equal Vote Coalition advocates STAR Voting (= "Score Then Automatic Runoff", previously SRV = "Score Runoff Voting") which adds an additional runoff stage, and I don't understand why, or whether it's actually an improvement or not.

They claim that STAR is only a "partial solution" and their addition of a runoff "answers the strategic voting criticisms" of Score, by reducing strategic bullet voting, because "you have an incentive to differentiate candidate scores in order to have a voice in the second step", but is this actually a problem that needs solving? Why would voters intentionally harm candidates that they actually like?

They claim that STAR "provides a majority win outcome", which sounds a lot like the majority criterion, but from what I've read:

1. this criterion doesn't really apply to cardinal systems at all
2. this is a bad criterion to meet, since it leads to "tyranny of the majority", where polarizing/controversial candidates win (who are liked by a slight majority, regardless of whether everyone else hates them) instead of consensus/compromise candidates (who are generally liked by everyone, even if not their first choice).

Does the runoff stage mean that STAR leads to majority tyranny, too?

Now that I understand strategy in Score better, here's an example:

• A: Love
• B: Like
• C: Neutral
• D: Dislike
• E: Hate

So if the Score ballot goes from 0-9, an honest (normalized) ballot might be:

``````A: █████████ 9
B: ██████    6
C: ████      4
D: ██        2
E: ▏         0
``````

Now suppose you watch the news and see that polls show C and D are definitely in the lead compared to everyone else:

``````A: ███
B: ██
C: ██████████████
D: █████████████
E: ███
``````

In this case, your optimal strategic vote would be to exaggerate relative to the frontrunners, maximizing C and minimizing D:

``````A: █████████ 9
B: █████████ 9
C: █████████ 9
D: ▏         0
E: ▏         0
``````

So you're expressing your true support for ABC, relative to the average voter, but exaggerating C and D to increase C's chances, considering how everyone else will vote.

However, if the polls show that B and C are in the lead:

``````A: ███
B: █████████████
C: ████████████
D: ████
E: ███
``````

your optimal strategic vote would be to exaggerate your like/dislike for B and C:

``````A: █████████ 9
B: █████████ 9
C: ▏         0
D: ▏         0
E: ▏         0
``````

As another example, maybe the polls are not very certain, and more than 2 candidates are competitive:

``````A: ██
B: █████████████
C: ███████████
D: ████████████
E: █
``````

In this case, your best strategy is to "hedge your bets", by giving honest ratings for the competitors and min/max to those who don't have a chance:

``````A: █████████ 9
B: █████████ 9
C: █████     5
D: ▏         0
E: ▏         0
``````

(Note that these strategies are still "semi-honest", in the sense that there's no incentive to reverse the rank order of any candidates, like there is under ranked-choice voting systems.)

STAR voting is supposed to reduce the incentive to maximize or minimize all your votes, because if you maximize two candidates and they both make it the runoff, your vote no longer distinguishes between them. So in this example:

``````A: ███
B: ██
C: ██████████████
D: █████████████
E: ███
``````

Instead of maximizing everyone, you might show a small amount of preference between them, in the off-chance that B makes it to the run-off, and likewise for the off-chance of E making it to the run-off:

``````A: █████████ 9
B: ████████  8
C: ███████   7
D: █         1
E: ▏         0
``````

So you're exaggerating, but not as much. Still, if B and E really don't have a chance, you might still want to exaggerate as in Score.

Why would voters intentionally harm candidates that they actually like?

Because they like other candidates better.

Consider the following case: three candidates named Sanders, Clinton, and Trump. You want Sanders to win, but you prefer Clinton to Trump. If you vote your actual preferences, you might vote

``````Sanders 100
Clinton  80
Trump     0
``````

But what if you are the last voter and Clinton is ahead of Sanders by 50 points. Both are ahead of Trump by more than 100 points. If you vote as above, Clinton wins by 30 points over Sanders. But what if you vote

``````Sanders 100
Clinton  10
Trump     0
``````

Then Sanders wins by 40 points.

One could argue that the latter ballot better expresses your real preferences, but things change if Sanders is more than 100 points behind Clinton and Trump is 99 points ahead. Then you might vote

``````Sanders 100
Clinton 100
Trump     0
``````

You vote differently depending how your vote interacts with those of everyone else.

Obviously you don't get to be the last voter after everyone else has been counted. So you have to vote before this. But you could see it in the eventual results. After a couple times voting the first way, many would vote the second way.

Another issue is that real elections aren't that close. A presidential election never comes down to just one vote. Sure, it could happen, but it almost certainly won't in any particular election. But you aren't the only one facing this particular decision. Every other person who prefers Sanders to Clinton has the same decision.

The fundamental problem with Range Voting (the name I know for what you call Score Voting) is that it encourages people to care how others are voting. This is what is called tactical voting. Instead of voting your preferences, you vote as best supports your goals. But that ruins the whole point of casting a ballot, which is to express your preferences. The voting system is then supposed to balance your preferences with those of everyone else. Unfortunately, if you don't express your actual preferences in the ballot, then there's no voting system that can balance your actual preferences.

Note that Range Voting is not alone in that. Most voting systems have some circumstance in which it is advantageous to vote a candidate lower or higher so as to gain a tactical advantage for your most preferred candidate. Range Voting suffers from the situation being easy to describe and demonstrate.

Another issue is that Score Runoff Voting doesn't fix the problem that I described initially. It fixes the problem of you marking Sanders and Clinton the same so as to beat out Trump. It does not fix the problem of you marking Clinton close to Trump because you want Sanders to make it to the two person runoff over Clinton.

• But it would be stupid to vote Clinton as 10 if you actually think of her as 80, since you are increasing the chances of an outcome that is drastically bad for you (Trump winning) in the hope of getting an outcome that is marginally better for you (Sanders winning). – endolith Jan 5 '17 at 5:15
• Its actually been proven that all voting systems with more than 2 candidates are susceptible to tactical voting. Its called Gibbard's Theorem: en.wikipedia.org/wiki/Gibbard%27s_theorem – B T Jan 2 '18 at 8:49
• @BT Yes, so "The fundamental problem with Range Voting is that it encourages people to care how others are voting" is misleading, since this is a problem with all voting systems. – endolith Jan 9 '18 at 19:29
• "Every voting system has flaws" does not mean "every voting system is equally flawed". A system might reduce a very important flaw by increasing a less important one, or might reduce one flaw by a large amount while increasing an equally important flaw by a smaller amount (or vice versa). – Foo Bar Jan 18 '20 at 14:11

Brythan wrote, "Another issue is that Score Runoff Voting doesn't fix the problem that I described initially. It fixes the problem of you marking Sanders and Clinton the same so as to beat out Trump. It does not fix the problem of you marking Clinton close to Trump because you want Sanders to make it to the two person runoff over Clinton."

There is zero value in having a candidate you like make the runoff and then lose to a candidate you really don't like. You're better off propelling at least two candidates to the runoff and differentiating scores so that if it's down to two candidates you like, that the candidate you like more gets your full support.

Endolith writes, "They claim that SRV 'provides a majority win outcome', which sounds a lot like the majority criterion," -- not quite. The Majority Criterion is really most applicable to rank-order systems. We have provided a new criterion (yay! One more!) called the Relaxed Majority Criterion - http://www.equal.vote/rmc - that is more useful for score and score-like systems.

• Do you agree with the optimal strategies shown in my answer, or did I miss something? – endolith Mar 15 '17 at 21:01
• this seems more comments to other answers rather than an answer in itself. – Federico Mar 15 '17 at 21:35
• But you don't always have the opportunity of propelling two candidates that you like to the runoff. If Trump is the leader regardless of what you score, your choice is only about who the second candidate is. I agree that there is no reason to make a candidate you like lose the runoff. But there is quite a reason to make a candidate that you like win the runoff. Unfortunately, it is difficult to discern between those two circumstances when voting. – Brythan Mar 15 '17 at 23:49
• If you think your strong opponent is guaranteed to make the runoff, you then want to have your strongest candidate face that opponent. Propelling only a weak candidate forward and failing to give support to your consensus choice will give you a worse outcome. The honest scoring is a strong vote. – Mark Frohnmayer Mar 16 '17 at 17:34
• Endolith, your strategy analysis looks good. One of the reasons we dropped the default range to 0...5 is to further decrease min/max distortion. – Mark Frohnmayer Mar 16 '17 at 17:39