Hacker News Reader

Comment by notriddle | Hacker News Reader

If you're going to talk about vote counting systems, please consult this interactive blog post:

And the HN discussion surrounding it:

https://news.ycombinator.com/item?id=30847133

[Unfortunately, while you can test ranked choice voting systems against each other directly using examples, you can't really test them against cardinal voting systems because the ballots aren't the same. At best, you use statistical models to compare them.]

View on HN

We should really note a key difference in ordinal vs cardinal systems that has a lot of effects surrounding them. That's how the information is embedded. In a ranked system all candidates your preference between any two candidates, in order, is equidistant. In cardinal systems, since you score candidates, your preference doesn't have to be equidistant. You can have a 0 distance between any choices or a large gap. This is pretty powerful and comes at no real cost (plus they are scalable and trivial to calculate winners since it is just argmax(sum of columns))

The problem with cardinal systems is that political parties will tell their followers "just vote 10.0 for me and 0.0 for everyone else". Enough people will do that to obviate the other advantages.

Exhaustive or even partially exhaustive RCV makes that tactic impossible.

You can still do that tactic in RCV by simply voting for a single candidate and leaving the rest blank. I'm not sure why people keep talking about this strategy as if it is unique or a coup d'etat. Additionally, no system is invulnerable to a strategy or candidates manipulating voters. So pointing out one strategy is an ineffective means of argument because you need to compare the results of strategies across different systems and the harm that they can cause (see favorite betrayer as one that causes a lot of harm and what I think the vast majority of voter concerns are with. I'll let you see how cardinal systems vs ordinal systems handle this). The issue I see with voting discussions is people don't realize how complex social choice is but just latch onto a few CGP Gray videos and defend their preference with a resolve well beyond their experience.

That manipulation is also an easy to overcome strategy as you can bullet vote your favorite candidates. There actually isn't any real strategic advantage to not voting your preference (albeit minor ones). There's no concept of tossing your vote away by scoring your preferred candidate high (the most common strategy is to weight your safe candidate the same or just below, but above your actual preference). Yes, you have to teach people how to vote under this new system but that's no difference than teaching them to rank candidates (instead we teach to give them a ?/5 star review. A concept people are highly familiar with).

> The problem with cardinal systems is that political parties will tell their followers "just vote 10.0 for me and 0.0 for everyone else".

Of course parties will try to convince people to do that. The question is whether voters will listen.

It's extremely easy to prove to yourself that, if my ranked preferences are O'TruePreference is better than Doe who is better than McBadGuy, then this ballot:

    Evil McBadGuy: 0
    John Doe 10.0
    Andre O'TruePreference 10.0

... is strictly better than this one, regardless of how other people vote:

    Evil McBadGuy: 0
    John Doe 10.0
    Andre O'TruePreference 0

> Enough people will do that to obviate the other advantages.

I know this happened with Approval Voting, but that doesn't necessarily mean it'll happen to Score Voting. It's likely that voters just didn't read the directions, and thought that if they voted for more than one candidate their ballot would be discarded.

Yeah, this is called "bullet voting" (for those that don't know). It is often brought up as a killer strategy to cardinal systems but it really isn't that bad. No system is immune to strategies and I think people just find one strategy and disregard a system because they found a flaw. Social choice is over constrained. There are trade-offs and local optimizations but no global optimization. We can't talk about voting systems without acknowledging this point.

To be clear, bullet voting in Score is equivalent to Approval Voting. AV still has a lot of advantages over Runoff Voting.

> It's likely that voters just didn't read the directions

This is actually a point (that I don't bring up much because it is minor) that I prefer cardinal over ranked. It is much easier to mess up a ranked ballot than cardinal. Give the same score to multiple candidates? No problem. In a ranked ballot you have issues and no good resolutions (toss ballot, flip a coin on preference and then reorder? Both of which can be called voter manipulation, despite them not really being). These issues arise more as the system scales too.

The big killer to me (besides the information embedding) is the tallying (backend that voters don't see). Cardinal systems are just an argmax reduce-sum (trivially parallelizable). Trivial to calculate and do statistical samples/verification on. This is a lot harder with run-off systems. The multi-elections required actually takes a lot of man power. Just think about how recounts like Arizona would happen under Cardinal vs Ordinal. This is something people really have to consider when accounting for voting systems. There is a front end and back end but most people argue front end + a few cherry picked strategies.

Reply on news.ycombinator.com