A STAR-voting spoiler effect
-
It's definitely a problem with STAR that you are limited in which candidates you can distinguish between despite being billed as a method where you can be honest about both your cardinal and ordinal preferences.
One way to solve this would be to have scores out of e.g. 99. But then only the 10s column is used to average the scores (so the available scores are essentially 0, 10, 20, ..., 90) but then the whole score can be used for the run-off to indicate ranks more finely.
This way someone can give 99, 98, 97 to their three favourite candidates knowing that they are each getting the max score, but then if two of them happen to make the run-off, they have a preference indicated.
-
@lime This is an example of where no system is perfect, and it's also an example of why we say STAR eliminates the "Spoiler Effect or vote-splitting caused by the voting method itself" when we're getting technical.
This is a problem that could happen with real human behavior under any voting method. There will always be a psychological and practical disadvantage for a faction that runs more candidates. There is also the fact that these three candidates are also competing with each other for volunteers and donors.
Simulations show STAR topping the charts at ~99% VSE. This is an example of that unsightly 1%.
But is it fair to only blame the voting method? In this case, the Liberal voters could have rated their faction higher and won. Voters who's real utility on their candidates (measured 0-100) would have been (0, 1, 98, 99, 100) should have translated that in to a 5 star ballot as 0, 0, 5, 5, 5. Even if their real utility was 0, 10, 80, 90, 100 that should have rounded to 0, 1, 4, 5, 5. If indeed this was voter error then hopefully these voters would learn to be more generous with their coalition next time and the voter data would get better over more election cycles.
Liberal voters also could have come together around a single candidate or a more unified preference order and been a stronger coalition and that would have also allowed them to win. Ultimately that's a failure of the candidates and the party as much as the voting method.
Here's the crux of it. In order to modify STAR to add more granularity and mitigate this potential issue you end up introducing other potential issues. Cognitive Load limits the ideal ballot range to around 6 or 7 tops. If you go up to 0 - 9 or 0 - 100 then you start to get serious distortions caused by inconsistent voter behavior and overwhelm. We know that voters (and people taking polls) tend to use the ends, the middle, and the numbers next to the middle; 0, 1, 50, 99, 100. The 0-5 ballot is optimized to eliminate this distortion since that would be 0, 1, 2, 3, 4, 5 for a 5 star ballot. It also keeps it simple for both cognitive load and ballot design. A long ballot with a sea of bubbles is a huge risk factor for voter error, and it also costs a lot more to print. Going to written numbers also would increase cost, decrease accuracy, and increase the time to process the ballot.
One solution could be to move away from the STAR tabulation entirely. Smith//Score is a cool proposal that basically switches the two steps of the STAR process (scores, then preferences). In Smith//Score the first step looks at the voter preferences (Condorcet style) and then the second step uses the scores. It tests about the same as STAR, (fantastic) but it's more complicated to explain. It also has its own edge cases that are unsightly because it devalues the scores to tiebreaker status even when that data was super relevant. The same downsides exist for all the Condorcet methods, which are the only real competition with STAR in terms of accuracy.
So, let's say we stick with STAR Voting. Are there other ways to prevent this type of scenario? Party primaries are one option that may be worth it for competitive high stakes elections. Multi-winner elections can also mitigate these kinds of issues. Having candidates petition onto the ballot with volunteer collected signatures only is another option. Having factions give formal endorsements to just one candidate is another. Voting reform is a puzzle and the voting method is just one piece.
-
Hi Sara, thank you for your thoughtful and informative comment. I agree with you that this is unlikely to be a major problem in practice, and STAR is a great voting method. However, I also think that we should always be looking for ways to improve it and make it more appealing to the public. Here are some of my thoughts on how we can do that:
- Cognitive overhead: You're right that too many options can cause a paradox of choice. Too few options can also cause analysis paralysis, though. I think the ideal number of options depends on the voter, and we should try to accommodate both types of voters. Some studies have suggested that 5 is the optimal number of choices for most people, while others have found that 10 is the upper limit .
- Double bubble: I'd like to propose a compromise that I call "double bubble". It allows voters to use half-scores by filling in two bubbles. For example, filling in 2 and 3 would assign 2.5 points to a candidate. This has several advantages:
- It gives voters more choice, but only if they want it. Most people can stick to the 0-5 scale; discerning voters can also use half-scores.
- It lets voters assign scores of 2.5/5 (exactly in the middle) to "average" candidates. This lets voters express a neutral opinion for candidates they feel unsure about, or don't have enough information to vote on.
- It doesn't require any change to the ballots, only a small adjustment to the tabulation algorithm.
- It reduces the number of spoiled ballots, by accepting ballots from voters who include 2 marks. This can help avoid voter frustration and confusion, especially for those who are not familiar with the voting method.
What do you think of this idea? Do you have any other suggestions or feedback? I'd love to hear from you and other forum members. Thank you for your time!
-
@lime For me, the point of voting theory is voting reform. There are a million variations, including the ones you suggest that are totally valid and fine in the right context, say an on-screen voting experience where voter errors are detected upfront and not allowed by the platform, and where you know that the voting method is a good fit for the voters and the question being voted on.
Unfortunately, our experience with even highly educated voters voting on our platforms (for example, a team of environmental lawyers who voted on policy priorities for their org) shows that voter error and failure to follow incentives without clear instructions is a real concern.
The solution is to have very, very clear, simple, and consistent instructions on the ballot, and to do very clear, simple, and consistent voter education at large so voters get better over time. This is fundamentally incompatible with having a bunch of similar cardinal/ordinal methods out in the world being used differently for different elections.
I don't see an issue with a voter using Approval for some elections and something else (STAR) for others, but if we have people using STAR for some elections, STAR 0-9 for some elections, Score for some elections, and Smith//Score for others that gets problematic for voter behavior, even though all of those are mathematically good voting methods. Even worse would be if voters use RCV for some elections and STAR for others.
When we look at elections and cognitive load we REALLY need to remember that even if a voters cognitive capacity is 0-10, we can't use all that up on the ballot design. They need some cognitive load for dealing with a new idea, more for comparing each candidate, and more for other stressors or complications that might be making their life harder than average in that given moment.
We made STAR Voting as expressive as possible given those limitations, and it remains the most expressive serious proposal on the table.
-
@sarawolk Disclaimer. I was the one who convinced Mark Frohnmayer and Equal Vote to change it from a 0-9 ballot to a 0-5 ballot and to rename it from Score Runoff Voting (SRV) to STAR Voting in 2017. https://www.starvoting.org/why_5_stars
-
@sarawolk said in A STAR-voting spoiler effect:
Unfortunately, our experience with even highly educated voters voting on our platforms (for example, a team of environmental lawyers who voted on policy priorities for their org) shows that voter error and failure to follow incentives without clear instructions is a real concern.
Yeah, one reason I'm concerned is because the naive strategy of "cast a Borda vote for your top-5" is likely to be a common one, even if it's not optimal.
I'm happy to give up the half-star if it really does increase support in polling or experiments. On the other hand, if it makes everyone less likely to support STAR (and IME that's everyone's first objection), it's definitely not worth giving up that support. If nothing else, it might be the ballot equivalent of "I'm feeling lucky"--Google doesn't keep that option there because people use it, they keep it there because they have research showing that people like having the option there, even if they ignore it.
In particular, working memory constraints about 7+/-2 just aren't that relevant here, because of chunking and pattern-recognition. There's a lot of research showing 5-integer sequences aren't the limit of human memory. The magic number 7 is specifically for sequences of random integers, not sequences with obvious patterns or the ability to be grouped into chunks.
It also doesn't apply to sequential decisions. People asked to make two decisions in sequence--e.g. "Choose a rating from 0-5; now, are you a 'low 3' or a 'high 3?'"
People have no trouble working with even very long sequences of integers (try remembering "0704177609112001")--as long as they follow a pattern or can be broken down into chunks that can be considered one at a time (July 4th, 1776 and September 11th, 2001).
In general, people don't have trouble working with half-integers running from 0 to 5, especially when talking about issues they are familiar with or have strong opinions on (e.g. political questions). This is why polling organizations like Pew, Gallup, or ANES use 5- or 7-point scales for policy questions, but 100-point feeling thermometers for politicians or political parties: many respondents' opinions cluster at multiples of 10 or even 5 (a 21-point scale!), not multiples of 20.
If half-stars weren't available, I agree 9 options would be too many. However, the ability to use half-stars gives voters who want more expressivity a lot more choice at the cost of little-to-no increase in complexity. (And those voters might be your most important; the kinds of voters with very fine-grained political opinions are exactly the kind of people you'll want to win over, since they have a disproportionate amount of influence on public opinion.)
-
@lime said in A STAR-voting spoiler effect:
Yeah, one reason I'm concerned is because the naive strategy of "cast a Borda vote for your top-5" is likely to be a common one, even if it's not optimal.
In STAR Voting showing your preference order is optimal. If the strength of your preference is negligible then you should vote no preference, but generally the incentive is to show your preferences.
-
@lime said in A STAR-voting spoiler effect:
There's a lot of research showing 5-integer sequences aren't the limit of human memory.
For me, ethically, elections have to be as accessible as possible. Some people have less education, less time, have to vote in their 2nd languages... A good system shouldn't be pushing voters cognitive capabilities to -or even near- the limits.
One of the biggest critiques we hear about RCV and STAR from status quo fans is that a more complex ballot will hurt voter turnout. While I don't buy it (I think a more representative system will inspire turnout) the point is that it's a real concern that people worry about and we have to accept that some people will feel that way.
I really do think that more complexity than 0-5 is too much, if not for actual voters, than for a reform's political viability.
-
@sarawolk said in A STAR-voting spoiler effect:
I really do think that more complexity than 0-5 is too much, if not for actual voters, than for a reform's political viability.
Maybe. Either way, I'd definitely like to see a decision like this made by evaluating real polling or research, instead of just being based on gut intuitions, or on "7+/-2" being when it's not clear if that really does apply.
I think listing out too many options would be a bad idea for the reasons you raise. The main question is whether adding half-stars actually creates additional complexity for voters, or if they feel like it would make their job easier by not forcing them to think strategically about their vote in these kinds of situations.
-
@sarawolk said in A STAR-voting spoiler effect:
In STAR Voting showing your preference order is optimal. If the strength of your preference is negligible then you should vote no preference, but generally the incentive is to show your preferences.
That's the issue--there's two competing heuristics ("Showing your preference order is optimal" and "If the strength of your preference is negligible, vote no preference.") When these heuristics conflict, it makes people feel uncomfortable and they have trouble making a decision (i.e. casting a ballot is harder and takes more cognitive effort). I think there's a good chance this played a role in the narrow defeat for STAR voting a few years back.
-
@sarawolk said in A STAR-voting spoiler effect:
We made STAR Voting as expressive as possible given those limitations, and it remains the most expressive serious proposal on the table.
Is there an objective measure of expressiveness that it comes out top in? Compare it with a ranked-ballot method. Under STAR you can indicate degree of preference. But under a ranked method you can indicate more preference distinctions (you can rank all the candidates in order), rather than just the 6 in STAR. So how do you combine these two things together for a single measure of expressiveness?
-
@toby-pereira On a ranked ballot you can't show if a 2nd choice is as good as a 1st choice or as bad as a last choice. In practice, ranked methods have to constrain how many candidates can be ranked. There are a few different ways you can look at it, but here's why I think it's the most expressive on the table (excluding cardinal methods with a bigger range)
- the ability to vote on every candidate no matter how many there are,
- The ability to show preference order
- The ability to show degree of support on a 6 rating scale.
