Every voting system has some flaws (see: Arrow's impossibility theorem), and Instant Runoff Voting definitely saw its worst-case scenario rear its ugly head in Alaska: A Condorcet winner existed (Nick Begich), he lost, and voters whose first choice was Plain and second choice was Begich could theoretically, retrospectivally, all changed their first votes from Palin to Begich to have him win rather than Mary Peltola.
It's important that I say retrospectively. This was a house race with national coverage, tons of high-quality polling from pollsters with nothing better to do, and it was still absolutely not clear from that polling that Begich would do better than Palin in round 2, or that Palin was likely to beat Begich in round 1. If I had the opinions of those hypothetically strategic voters, and I were completely strategic, and completely tuned into that polling, I still would have voted honestly. The strategy was only apparent after the fact.
Thinking about "defense in depth" against this scenario, we'd be remiss not to mention Condorcet ranked choice voting methods, which always elect a Condorcet winner, and only resort to (game-able) elimination procedures if there isn't one. Not only would that have resulted in a better outcome in Alaska, it would have made the possibility of strategic voting much, much more remote, both because a strategic scenario would be much less likely, and because that strategic scenario would be much less visible before the election.
With score voting, strategies are always available and almost always obvious: just exaggerate your score differences between the most likely candidates. The fact that strategic dishonesty would have helped in Alaska is a fair criticism of IRV, but the real commonness of strategic scenarios in STAR and score are a more primary feature.
Gibbard's theorem implies "strategy" in a sense (your optimal strategy depends on distribution of other people's votes) which is much less nefarious than many people's intuition on what "strategic voting" implies.
When people think about strategic voting coming from a FPTP context, the strategy they're thinking of is putting in a dishonest vote for a non-preferred candidate because that provides a better chance at matching their preferences than an honest vote. This is not what Gibbard's theorem is about. Strategic voting in cardinal systems, for example, tends to look basically like dynamic range compression and at an individual level generally only "tiebreaks" between candidates that were otherwise fairly evenly matched, while the FPTP strategy of bullet voting a non-preferred candidate is just straight up bad for your preferences.
>putting in a dishonest vote for a non-preferred candidate because that provides a better chance at matching their preferences
This still somewhat exists. For example, with score voting: preferences A>B>C but you really hate C, and he's leading in polls. It might be reasonable to "dishonestly" score B higher. This minimizes chances that C wins, and while it doesn't harm A it does decrease strength of your preference for A>B. "Dynamic range compression" is good way to put it, or maybe "ballot influence conservation": one ballot has fixed influence, and you can either spend it entirely on A/B>C or A>B/C, or any ratio in between.
Yeah, "dynamic range compression" on a score vote does feel a bit dishonest—kind of independently of whether it's actually "strategic", actually.
Personally, this is a major factor in preferring approval over score vote, which requires someone who intuitively "feels", say, a 0.3/0.6/0.7 about candidates to either vote 0/0.75/1 to maximize their vote spread, or give more influence to people who do. Notice that this is not strategic in Gibbon's sense—it is always better to vote 0/0.75/1 than it is to vote 0.3/0.6/0.7—but it's probably more counterintuitive!
Thank you for pointing out how overused "Arrow's impossibility theorem" is as an argument against all voting systems (and therefore often as an argument against all reforms to the current near-worst possible voting system).
I would go further and be somewhat dismissive even of "Gibbard's theorem", since it only shows that an insincere vote can strategically produce a better outcome for a voter if they can identify what that that vote should be, and if they are one of the voters for whom such a strategy is available.
So it doesn't guarantee (as far as I'm aware) that even a single voter (under every possible cardinal voting system) will necessarily have enough information before or during the election to confidently pick some specific insincere vote that will increase their chances of getting the outcome they want.
> I would go further and be somewhat dismissive even of "Gibbard's theorem", since it only shows that an insincere vote can strategically produce a better outcome for a voter if they can identify what that that vote should be, and if they are one of the voters for whom such a strategy is available.
Gibbard's theorem is even weaker than that. It only assures the constructibility of vote sets for which the right "final" votes to achieve a particular outcome are distinct. It does not guarantee that there are any insincere votes which perform better than a sincere vote!
Optimal strategic approval voting with a strict ordering is always a choice between sincere votes, in fact (since adding a vote for a strictly more preferred candidate never hurts your preferences, you can always find a sincere vote which performs better than an insincere strategic vote by additionally voting for all candidates more preferred than the least preferred candidate on the strategic vote).
It's important that I say retrospectively. This was a house race with national coverage, tons of high-quality polling from pollsters with nothing better to do, and it was still absolutely not clear from that polling that Begich would do better than Palin in round 2, or that Palin was likely to beat Begich in round 1. If I had the opinions of those hypothetically strategic voters, and I were completely strategic, and completely tuned into that polling, I still would have voted honestly. The strategy was only apparent after the fact.
Thinking about "defense in depth" against this scenario, we'd be remiss not to mention Condorcet ranked choice voting methods, which always elect a Condorcet winner, and only resort to (game-able) elimination procedures if there isn't one. Not only would that have resulted in a better outcome in Alaska, it would have made the possibility of strategic voting much, much more remote, both because a strategic scenario would be much less likely, and because that strategic scenario would be much less visible before the election.
With score voting, strategies are always available and almost always obvious: just exaggerate your score differences between the most likely candidates. The fact that strategic dishonesty would have helped in Alaska is a fair criticism of IRV, but the real commonness of strategic scenarios in STAR and score are a more primary feature.