I agree that a likelihood of higher-than-normal percentage of spoiled ballots is a potential problem with IRV, at least with mail-in or in-person paper ballots. Voting machines shouldn't have a problem with it.
Approval is about as simple as it gets. In order to spoil the ballot you'd have to do something pretty strange like tear the ballot in half. Or with mail-in ballots, the usual things like not signing the envelope (assuming that's required in your state) or mailing it too late to arrive on time.
I don't see ballot spoiling as necessarily a show-stopper for using RCV on its own, but there are other compelling reasons not to use it (such as failing monotonicity).
And you've always been able to produce invalid ballots, like voting for two candidates when you're only allowed to pick one.
You're describing a UX problem which has UX solutions. It's not even remotely a justification for selecting a democratic procedure to express the will of the people.
Agreed but that's far from my only gripe with FPTP and IRV.
> Unlike many single-winner methods, instant-runoff cannot accept equal rankings, and must discard ballots with multiple first-preferred remaining alternatives: such ballots would be equivalent to casting multiple ballots in a plurality election
With all-the hubbub on the last election about vote validity, and hanging chads before that, the last thing we need is screw-up-able ballots.
IMHO we should not even consider any methods in which you can violate invariants on a paper ballot. That rules out FPTP, IRV, and any others where the user with pen-and-paper can enter an invalid state.
Err, I guess that rules out 321, which is my new favorite. Simple fix: just count multi Bad/Ok/Good entries toward that total. In the first round, you only look at "good" scores, ignoring ok/bad. Then in the bad round, same thing. This ensures a) you can't produce a bad ballot b) ballot counts stay orthogonal for each category c) multi-entries kinda average out.
E2: well I guess you can manage to screw up ANY style of paper ballot and violate some invariant. I guess it matters the difficulty of screwing it up and whether it's recoverable.