STAR CW/CL stragedy

[Clay Shentrup]

Imagine you prefer CL > CW > Other

E.g. HRC > Bernie > Trump

Your best strategy is to bullet vote for CW/Bernie.

Why?

Because the only scenario where you get your third choice is if your second doesn't make it to the "runoff".

Discuss.

[Clay Shentrup]

Now imagine you prefer:

Other > CW > CL

You want to vote for Other and CL — your 1st and 3rd but not your 2nd.

[Clay Shentrup]

All other combos:

CW > Other > CL = vote for CW and CL

CW > CL > Other = bullet vote

CL > Other > CW = Vote for your two favorite

Other > CL > CW = Vote for your two favorite.

[Jameson Quinn]

You're suggesting that STAR is vulnerable to a turkey-raising betrayal strategy. I believe that people's desire to vote expressively, as well as their uncertainty/risk aversion, will stop this from being a significant problem.

I do think that turkey-raising exaggeration is a minor problem with STAR. 

Note that 3-2-1 is clearly better than STAR in both regards, though not entirely immune to these issues.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscience+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ciaran Dougherty]

If it's true that uncertainty/risk aversion, in conjunction with the desire to vote an honest/expressive ballot will stop that from being a significant problem... What is the benefit of adding the (majoritarian) step of the runoff?

[Jameson Quinn]

For the purposes of the runoff step, more-strategic and less-strategic ballots are on an equal footing. That's a direct good, though of course it goes hand-in-hand with the direct bad effect (from a utilitarian perspective) that honestly-strong preferences are on an equal footing with honestly-weak ones.

My own prior is that if two people's "honest" ballots (by which I mean, purely-expressive; I don't think that necessarily equates to "true normalized prospective utility" well, but it certainly does so better than a purely-strategic ballot) show different preference strengths, it is as likely to be an effect of normalization, rhetorical inflation, and/or prospect theory/behavioral economics as it is to be a true utilitarian difference. I therefore think that even at the level of direct effects, a runoff stage is a good thing.

Aside from the direct effect, there's also the effect of reducing strategic incentives to exaggerate. That strategic effect is nearly an unalloyed good. It does potentially open the door to turkey-raising strategies, but I'm not concerned about those (unless the incentives are extremely strong, as in Borda). 

[Ciaran Dougherty]

...reducing incentives to exaggerate?  I've always seen the runoff step as reducing disincentives to exaggerate, by satisfying Later No Harm (where I assert that not satisfying it is a feature, not a bug, in Score).

With straight score, exaggerating means that you increase the probability that a Later candidate can beat a Favorite (if exaggerating upwards), or that a Later candidate would be beat by a Worse candidate (if exaggerating downwards).  With STAR, however, if you exaggerate a Later to Max-1, you get the benefit of increasing their score (even beyond that of your favorite, potentially), without having to worry about your vote helping them beat your favorite.

And because the benefit of doing so is much more obvious to the average person, and the problem is less obvious to the average person (eg, 3-way near-tie scenario, replacing your Favorite in the Top Two), I expect that to be a lot more common.  This is especially true given that the probability of 3 way ties is incredibly unlikely, and that the polling accuracy required to make the Min/Max voting strategy viable appears to be functionally impossible at this point.

...which brings me back to Clay's original question: The proposed strategy is, in fact, a pretty good one... assuming you have knowledge that no one actually does.  A majority of the population, virtually all polls, consistently classified Trump as the Condorcet Loser in a Sanders/Clinton/Trump election, and yet he won.  The world thought that Clinton was the Condorcet Winner (she beat Sanders, and was expected to beat Trump), and that Trump was the Condorcet Loser, but we found that none of those classifications was accurate.

As with Warren's theorems about Min/Max voting, Bullet STAR voting in a known outcome scenario crucially relies on information nobody actually has.   And it gets worse if we end up with more candidates (as we should, given the drastically reduced instances of Favorite Betrayal and Impactful Irrelevant Alternatives)

[Mark Frohnmayer]

Your support of CL in this scenario makes it more likely Other will not make the runoff at all. This approach would be arguably valid if there were two discrete elections (score + top two), but in STAR the most common distortion this will cause is pushing your favorite out of the top two and the casting your vote for the CL.

--

You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.

[Jameson Quinn]

2018-05-06 11:09 GMT-04:00 Ciaran Dougherty <lona...@gmail.com>:

...reducing incentives to exaggerate?  I've always seen the runoff step as reducing disincentives to exaggerate, by satisfying Later No Harm (where I assert that not satisfying it is a feature, not a bug, in Score).

I think that the chicken dilemma is the main situation where exaggeration might be effective. Imagine the following honest ballots:

35: A5 B4 C0

25: A4 B5 C0

40: A0 B0 C5

Totals: A275 B265 C200; A>B 35 to 25.

In Score, 4 B voters could swing the outcome by "rank-honest exaggeration" (RHE: lowering their A votes to 1). In STAR, it would take 24 RHE and 1 going all the way to SHE (semi-honest exaggeration; lowering their A vote to 0), OR 10 of them casting maximal betrayal ballots (lowering A to 0 and raising C to 4) in order to swing the outcome. But note that STAR still has disincentives for SHE or MBB.

So while you are right that STAR reduces disincentives to exaggerate, because it makes RHE safe (C loses even if all A and B voters RHE), it also reduces incentives to do so (in most cases takes a lot more exaggeration to swing the election). I think that the reduction in incentives is the dominant effect in practice.

Note again that 3-2-1 is as good or better than STAR in all of the above respects.

[Clay Shentrup]

> Your support of CL in this scenario makes it more likely Other will not make the runoff at all.

other > CW > CL

Changing it from CW+other to CW+CL makes no difference.

[Clay Shentrup]

Another way to phrase it Mark, is that the ONLY thing that can help you is keeping CW out of the runoff. So of course you vote for your #1 and #3.

[Mark Frohnmayer]

The point is that if you are afraid your #1 is weaker than the CW, your support of #3 is more likely to knock your #1 out of the runoff, at which point your full vote would go to your least favorite candidate. Not a smart strategy by any measure.

On Sun, May 6, 2018 at 11:01 AM Clay Shentrup <cshe...@gmail.com> wrote:

Another way to phrase it Mark, is that the ONLY thing that can help you is keeping CW out of the runoff. So of course you vote for your #1 and #3.

[Clay Shentrup]

> The point is that if you are afraid your #1 is weaker than the CW, your support of #3 is more likely to knock your #1 out of the runoff, at which point your full vote would go to your least favorite candidate. Not a smart strategy by any measure.

I repeat: knocking your #1 (other) out of the runoff by supporting CL... CANNOT possibly hurt you. Simple proof:

CW+O => CW+CL = same winner either way

[Clay Shentrup]

I suspect Mark is thinking of the scenario where you don't have much certainly about head to head runoff prospects.

[Clay Shentrup]

So talking to mark out of band, I think he makes some good points.

For one, there's sort of a chicken dilemma where if people use this strategy, then the CW and CL will actually change. It's indeterminate.

Also this could be interpreted as supporting certain rather specious FairVote criticisms. Not my intent. Someone just asked me for an example where you want to mis-order candidates and I was proposing some.

[Mark Frohnmayer]

The key here is that a perceived CW in STAR is very likely to also be the top scorer. If CW is your #2 preference, burying the CW under your #3 ends up being a very risky strategy -- because CW is likely the top-scorer, elevating the perceived CL runs the risk of elevating CL over your #1, knocking your #1 out of the runoff, and then casting your vote for CL instead of CW. If enough people use this strategy to make it effective in elevating CL, CL could beat CW in the runoff.

Fundamentally, you would necessarily be expressing a greater or equal level of increased dishonest support in favor of CL than your dishonest decreased level of support for CW.

On the balance it's a dumb move in STAR. We know that voters in the real world overwhelmingly vote _defensively_ in order to avoid electing the "greater evil." There is no evidence to support they would start adopting risky strategies under STAR.

Where this strategy makes plenty of sense is if there are two discrete elections (i.e. Score + Top Two). If you know you'll get a 100% honest vote in the runoff, the supporters of Other would have absolutely no reason not to maximize support for Other and CL. This is one of the reasons we believe STAR is superior to the two-election Score + Top Two.

Our response to the FairVote piece that first raised the "burying" strategy can be read here: https://www.equal.vote/fv