![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
Voted today
We voted in person. We brought masks, but there was a pile of free masks on the table at the front door in case we hadn't. There was no line to get in, and no line at the table for our precinct, and no line for "privacy carrels" in which to fill out the ballots, and no line for the optical scanners that read the ballots. In and out in five minutes.
This was my first experience casting a ranked ballot in a civic election (although I'd used them in the Math Department Hiring Committee, and in the local SCA group). The ballot layout was pretty clear. But as this Times article points out, we probably won't know the winner for at least three weeks, due to a quirk of the Instant Runoff Voting system that I hadn't really thought about before.
In a traditional, single-vote, first-past-the-post voting system, if the top two candidates are separated by 10000 votes, you pretty much don't have to worry about anything under 10000 votes. If there are 8000 absentee ballots still uncounted, or 8000 ballots subject to a legal dispute, or something, you can still declare a winner because it's mathematically impossible for those ballots to change the winner. In a Borda-count system, if the top two candidates are separated by 10000 points, you might conceivably need to worry about a pile of ballots as small as 10001/(C-1) where C is the number of candidates, because it's mathematically possible that those ballots could all give C points to the second-place candidate and 1 point to the first-place candidate, thus changing the winner. But it still depends only on the difference between the top two candidates, which is usually at least a percent or so of the total number of votes.
Under IRV, you have to look at the difference between the bottom two candidates, which is likely to be much smaller. Let's suppose an election like today's has 100,000 voters and 10 candidates, of whom half are "realistic", getting roughly 15K first-place votes each, while the other half are long-shots getting roughly 5K first-place votes each. In particular, suppose the two least-popular candidates are tied at around 3K first-place votes. Shifting a handful of votes from one to the other of these unpopular candidates changes which of them gets knocked out in the first round. It seems quite plausible that the supporters of two unpopular candidates -- say, one on the far left and one on the far right -- would have different second choices. So changing which one gets knocked out first could realistically shift as many as 3K votes from one to another of the remaining candidates, either to one of the "realistic" candidates, which could put that candidate over the top, or to another of the long-shots, which could change who gets knocked out in the second round, which could shift another 5K votes, and so on. In short, a single-digit number of votes shifted between two unpopular candidates can be magnified in hard-to-predict ways into many thousands of votes that change the ultimate winner. As a result, unless somebody wins outright at the start, you can't really say anything until you've got all the ballots, and have resolved all the disputes about which ones to count.
In an election with a lot of long-shot candidates, there's even a realistic chance of an exact tie for last place. If the election administrators resolve this by flipping a coin, that coin-flip could have a massively magnified effect and shift thousands of votes between realistic candidates, changing the winner. (If the law allows, they could also resolve it by eliminating both of the tied-for-last candidates at once, which would be less fraught with randomness.)
So in addition to the problem I've mentioned before, that IRV favors divisive candidates over consensus candidates (albeit less than a single-vote FPTP system does), IRV also has a "butterfly effect" problem of extreme sensitivity to small changes in the input. Indeed, the butterfly gets several chances to flap its wings: each round of instant-runoff provides another opportunity for a tiny error or dispute to be magnified into significance.
This was my first experience casting a ranked ballot in a civic election (although I'd used them in the Math Department Hiring Committee, and in the local SCA group). The ballot layout was pretty clear. But as this Times article points out, we probably won't know the winner for at least three weeks, due to a quirk of the Instant Runoff Voting system that I hadn't really thought about before.
In a traditional, single-vote, first-past-the-post voting system, if the top two candidates are separated by 10000 votes, you pretty much don't have to worry about anything under 10000 votes. If there are 8000 absentee ballots still uncounted, or 8000 ballots subject to a legal dispute, or something, you can still declare a winner because it's mathematically impossible for those ballots to change the winner. In a Borda-count system, if the top two candidates are separated by 10000 points, you might conceivably need to worry about a pile of ballots as small as 10001/(C-1) where C is the number of candidates, because it's mathematically possible that those ballots could all give C points to the second-place candidate and 1 point to the first-place candidate, thus changing the winner. But it still depends only on the difference between the top two candidates, which is usually at least a percent or so of the total number of votes.
Under IRV, you have to look at the difference between the bottom two candidates, which is likely to be much smaller. Let's suppose an election like today's has 100,000 voters and 10 candidates, of whom half are "realistic", getting roughly 15K first-place votes each, while the other half are long-shots getting roughly 5K first-place votes each. In particular, suppose the two least-popular candidates are tied at around 3K first-place votes. Shifting a handful of votes from one to the other of these unpopular candidates changes which of them gets knocked out in the first round. It seems quite plausible that the supporters of two unpopular candidates -- say, one on the far left and one on the far right -- would have different second choices. So changing which one gets knocked out first could realistically shift as many as 3K votes from one to another of the remaining candidates, either to one of the "realistic" candidates, which could put that candidate over the top, or to another of the long-shots, which could change who gets knocked out in the second round, which could shift another 5K votes, and so on. In short, a single-digit number of votes shifted between two unpopular candidates can be magnified in hard-to-predict ways into many thousands of votes that change the ultimate winner. As a result, unless somebody wins outright at the start, you can't really say anything until you've got all the ballots, and have resolved all the disputes about which ones to count.
In an election with a lot of long-shot candidates, there's even a realistic chance of an exact tie for last place. If the election administrators resolve this by flipping a coin, that coin-flip could have a massively magnified effect and shift thousands of votes between realistic candidates, changing the winner. (If the law allows, they could also resolve it by eliminating both of the tied-for-last candidates at once, which would be less fraught with randomness.)
So in addition to the problem I've mentioned before, that IRV favors divisive candidates over consensus candidates (albeit less than a single-vote FPTP system does), IRV also has a "butterfly effect" problem of extreme sensitivity to small changes in the input. Indeed, the butterfly gets several chances to flap its wings: each round of instant-runoff provides another opportunity for a tiny error or dispute to be magnified into significance.
no subject
no subject
no subject
no subject