3 major criticisms of the Score Voting (SV, RV, STAR, ...)

These are the top 3 reasons why I’m against Score Voting, and I don’t hear much about it.
Anyone know if there is something wrong?

Dependence on minority candidates (IWA fails)
Given 2 candidates A and B from opposing factions, the results of the election would be:
A[55%] B[45%]
If an Ae (minority) candidate with extreme A ideologies had been added in the same context, some voters who supported A would now have voted like this:
Ae[10] A[7] B[0]
reducing the points assigned to A. This reduction of points in A could make B win even if the voters are always the same.

High dependence on results forecasts
Given 3 candidates A,B,C. One voter loves C and despises A and B, but prefers B between A and B.
3 cases:
(1) A and B are the favorites, while C is a minority. The voter will vote like this:
A[0] B[10] C[10]
In this way, even if C loses the vote doesn’t become null in the clash between A and B.
(2) C is one of the favorite candidates. The voter will vote like this:
A[0] B[0] C[10]
(3) There is insufficient information about expected results. The voter will vote like this:
A[0] B[?] C[10]
False predictions about the results could be disseminated, with the aim of making certain candidates receive more or less points (B in the example).

How do you vote?
Given 10 candidates, I consider 5 of them positive (I’m happy that they win) while I consider 5 of them negative (I’m sad that they win). All 10 however have an order of preference.

  1. I vote in the following way:

  2. But if however all my favorite candidates lose, my vote becomes useless regarding the negative candidates, so I vote like this:

  3. But negative candidates are too little away from the positive ones, therefore to favor my favorites, I vote as follows:

  4. But if I really want to favor my positives to the maximum and negatives to the minimum, then I should vote like this:

  5. But to provide a good representation of my interests, I also want to indicate the order of preference, so I should vote like this:
    I went back to step 1.

These aren’t tactical votes; it’s only the voter who doesn’t know exactly how to vote.

In the Score Voting:
) it’s not known exactly how to vote.
) high dependence on results forecasts (very exploitable).
) results that depend on minority candidates (very exploitable).
Problems related to almost all Score Voting-style voting systems, both single and multiple winner.

(IRV and DV don’t have these big problems)

1 Like

I think the biggest issue with these systems is that they don’t allow you to maximally support multiple candidates at once. In other words, vote-splitting is happening on the voter’s ballot, so how could it not happen in the overall election?

That being said, it is good to point out that while some systems mathematically solve vote-splitting, this may open them up to even greater problems when looking at how voters change their preferences depending on who is in the race and so on.

1 Like

I don’t think that is whatever happens on a ballot is what “vote splitting” means. Vote splitting is in how the ballots are interpreted.

If a ballot allows you to say, for instance, “I like Nader best but I still like Gore better than Bush”, the fact that you put Nader above Gore is not in itself vote splitting. If the tabulation says “Since Nader can’t win, I’m now going to interpret this ballot as simply saying that this voter prefers Gore to Bush”, then vote splitting hasn’t occurred – Gore got just as much support from you in the end, as he would have if you put him above everyone.

Condorcet works this way the vast majority of the time, as does STAR and even IRV most of the time.

The post is not about vote-splitting, so I prefer not to change the subject.

What is this if not vote splitting?

Ok, it’s vote splitting, but what you said doesn’t negate the problem described in the post regarding the Score Voting.
I’m only interested in knowing if there is any mistake in the 3 criticisms I made.

Having said that, IRV and DV do not have the Score Voting problem described because the vote would take this form:
Ae[1st] A[2nd] B[3rd] and when Ae loses (because it’s a minority) the vote becomes A[1st] B[2nd] as if Ae had not been there.

What I said, or what AssetVotingAdvocacy said?

I was simply attempting to correct his misapplication of the term “vote splitting”, where he seemed to suggest that a ballot that allows or encourages you to differentiate between two candidates will necessarily split the vote between them. A Condorcet election seems to be as close to the “holy grail” of avoiding this as you can come: “please rank the candidates honestly, and we’ll make sure your preference gets applied full strength depending on who can actually win.”

I think I agree with your main points, which by my interpretation complain of Score’s susceptibility to A) vote splitting and B) having to know polling data to be able to vote effectively. (you say you have 3 critiques, but to me they boil down to 2)

I think STAR is better than most methods on both of these issues, while also staying relatively simple to use and simple to tabulate. Approval is very subject to B – having to know polling data to vote most effectively. Score is kind of subject to both, but certainly in much lesser degrees than the status quo.

So I’m not sure why your title seems to lump STAR in with others having these problems. It is an imperfect method but it certainly attempts to address these things. That’s actually kind of the whole point.

Because IRV doesn’t usually allow equal ranking, this is actually more accurately “forces you to differentiate” except when looking at who is ranked equal last.

Perhaps these 3 practical problems are related to only 2 theoretical problems, but they remain.

Rereading the post assuming that STAR is used instead of SV, it seems to me that there are still the same problems.
How does STAR avoid the problems described in the examples?

I mentioned IRV and DV, to point out that these methods do not have these specific problems (they have other problems, but those listed in the post are too big for me to be accepted).

This seems roughly analogous to center squeeze.

Tactical voting exists in every system.

Is the problem here just that a 10 point scale is too small?

Any system with favorite betrayal as IRV and DV have will definitely have something similar in principle to 1 & 2.

I’m not sure I know what your level of understanding of voting methods is, so it is hard to explain in a short summary. STAR isn’t perfect, but for starters, think about how it would have worked in two well known presidential elections that had 3 candidates that were significant. 1992 (Bush Sr, Clinton, Perot) and 2000 (Bush Jr, Gore, Nader). Let’s ignore the electoral college, and assume these three are the only candidates running.

In '92, a voter that liked Perot best and Bush second best could honestly express their views in STAR. If Bush and Clinton ended up being the front runners, their preference for Bush over Clinton would count just as much (in the runoff) as if they had given Bush a 5. If Perot and Bush ended up being front runners – not all that unlikely under STAR, since Perot got 19% of the total vote and tended to liked by people on both right and left – their preference for Perot over Bush would count just as much as if they had given Bush a 0.

In 2000, a voter that liked Nader best and Gore second best could also honestly express their views in STAR. If Gore and Bush ended up being the front runners, their preference for Gore over Bush would count just as much as if they had given Gore a 5.

With plain old score voting, you would be better off carefully following the polls, and exaggerating your preferences accordingly. This would gain you very little with STAR.

That said, if there are more than three candidates, and/or it is a very tight race among the three, STAR’s imperfections may result in providing some incentive for people to closely follow polls and carefully strategize. But it won’t be a straightforward strategy, and it may well backfire.

That statement strikes me as misleading.

There is a point where the benefits of tactical voting don’t outweigh the costs. For instance in a Condorcet election, how accurately would you have to know exactly how others will vote in order to gain a strategic advantage by voting with anything other than your honest preferences? And how much advantage will it actually give you?

And remember, this isn’t just that you have to accurately know what other voters’ preferences are. You also have to predict how they are going to apply strategy. And they have to know, in turn, how you will. It’s a great big feedback loop.

This is very different than, say, plurality or approval voting where it gives you a very strong advantage to know who the front runners will be and vote accordingly (in approval, giving your approval to only one of the two front runners).

That’s not to say approval is bad, I think the effects of tactical voting will generally tend to cancel out, which is not the case with plurality. But still…saying “all methods are imperfect” is far less meaningful or useful than saying “some methods are far better than others.”


It’s true that

However I don’t think that DV and IRV are far better than Score and especially STAR with respect to the specific problem that @Essenzia presented: that when a voter prefers A>B>C, polling will be greatly advantageous in deciding how much to score B. In the case of IRV, polling may suggest that such a voter should vote B>A>C. Since DV is so similar to IRV, it has essentially the same concern.

1 Like

Right, I’m not defending Essenzia’s preference for DV and IRV. I do agree with him (or at least my interpretation of his complaints) that we should seek methods that minimize the incentive to be strategic and follow polls, and ones that minimize vote splitting effects. I think STAR does well at both of these, and approval does well at the latter one. In the grand scheme of things, vote splitting is the real enemy and the cause of polarization… so I’d be happy with either approval or STAR.

The argument against this is basically that B supporters, wishing to avoid Ae winning, would give A enough points to make up the difference. In fact, this may go so far that candidate A actually encourages inferior clones of themselves to run.

Center squeeze concerns the strongest three candidates, while in my example Ae is a minority (even only 10%) which however takes away points from A (55% originally).

Not with the same intensity.

No, the problem is that I don’t know what is the right way to vote (the example that I did also applies with a range [0,1000]).

Similar but not of the same intensity.
Later I explain why DV isn’t subject to these 3 problems.

Reevaluating, it is true that STAR better resists problems 1 and 2, in the case of 2 majority candidates (which is no small feat, and that actually makes it better than the others). After this reevaluating, STAR moved to my 2nd place for single winner methods.

Read this System Comparison - IRV

Problem 1
(I added a candidate to better show that it works)
Original vote: A[75] B[0] C[25]
I add Ae (minority), the vote becomes: Ae[56] A[33] B[0] C[11]
When Ae is eliminated because it’s a minority, the vote becomes the same as the original thanks to the normalization (which proportionally redistributes the 100 points among the remaining candidates).
DV is therefore completely immune to the addition of minority candidates (meet IWA).

Problem 2
That C (and D) is minority or majority, my vote will in any case be like this:
A[0] B[1] C[60] D[40]
In practice, I assign 99 points to my favorites (as I think) and 1 point to one of the two majority candidates. False predictions about the results will only guarantee 1 point out of 100 to certain candidates (B), which is a very small quantity (makes the use of fake polls useless, in this case).

Problem 3
Points are limited so a voter does not waste them for negative or unknown candidates when he can assign them to positive candidates to increase the probability of winning as much as possible.
In the DV the honest vote has a well-defined form, in which the points are given only to the positive candidates and proportionally to what they are favored.
Ex: [50,30,15,5,0,0,0,0,0]
Tactical votes favor the accumulation of points but this is another matter (also present in SV-style methods).

You may be interested in this property:
if in the DV all the voters separate the points equally between the favorite candidates, then the single winner will be the same as the AV.
Conversion example:
DV: A[25] B[25] C[25] D[25] E[0]
AV: A[X] B[X] C[X] D[X] E[0]
Normalization of the DV reduces the vote splitting every time it is used during the count.

Furthermore, satisfying the IWA makes the DV independent of the addition of minority candidates and this certainly favors the reduction of the vote splitting.

To be sure they should give A 10 points, but there would be no difference between Ae anymore (that is, Ae couldn’t win). This problem actually relates to the one after which, it is the predictions of the results that determine how many points to give to A (hiding the real interests of the voter).
My main complaint is not the presence of tactical votes but the fact that they are easily exploitable by political factions that can distribute fake polls.

You talk a lot about DV but I haven’t heard of it previously. Is this something you came up with?

I don’t understand how a ballot would look, unless I’m misunderstanding something it doesn’t seem designed for real world elections. How does a voter distribute their 100 votes without having to do it on an electronic device or do a bunch of arithmetic or what have you?

To be honest I’d think that a STAR voting interface (give each candidate between 0 and 5 stars) would do the same thing. Aside from not having the same granularity, but you can easily add up the stars and divide them into 100. That’s a much more intuitive interface than anything that has you actually deal with 100 votes and allocate them, in my opinion.

1 Like

This is the DV, and is a recent voting method.
Anyway, my aim was to understand if the problems exposed in the post were such, or if I had misunderstood some things.

This is the solution for the paper format: Vote writing
This is an example of an electronic solution: Poll example (Beta version)

In doing so, there is a risk that the average voter will find himself in the condition described in problem 3 (he would not know exactly how to vote).
One of the fundamental objectives of the DV is that negative and unknown candidates must receive the same score (0 points); this is what eliminates the ambiguity of problem 3.
Giving limited points to distribute, ensures that the voter uses them all on the favorite candidates (to favor them to the maximum).
I’m extremely opposed to any voting method that contains the concept of negative voting (like range [-5,+5] ).

It sounds like you interpreted my comment as “Ae-top supporters should give A a 10”, but I actually said

So in other words, if Ae voters feel Ae has such a good chance of winning that they can afford to downscore A, then this same thing would make B voters upscore A.