It's also sometimes called alternative vote, and the Australians call it preferential voting.
I've wondered if IRV couldn't be substantially improved by using Borda count for the elimination step rather than eliminating the candidate with the least first-choice options. (I think there's an actual name for that system, but I don't remember what it is.)
> I've wondered if IRV couldn't be substantially improved by using ... rather than eliminating the candidate with the least first-choice options.
This is exactly the family of RCV systems -- IRV merely defines one implementation by which a winner is selected from a ranked ballot.
Borda count has its own pathologies, too -- the usual cited problem is that it fails independence of irrelevant alternatives.
https://en.wikipedia.org/wiki/Independence_of_irrelevant_alt...
(Further, Borda count fails the majority criterion, namely there are situations where a candidate is the first pick of a majority of the electorate, yet fails to win. Some proponents of Borda count argue this is a feature, not a bug.)
The wiki article for Borda count has a solid table of various voting systems and whether they satisfy various criteria:
https://en.wikipedia.org/wiki/Nanson%27s_method
and the related Baldwin's method, which, while being Condorcet methods, both fail monotonicity.
Note that all RCV implementations have limitations, per the oft-cited https://en.wikipedia.org/wiki/Arrow%27s_impossibility_theore... which outlines a number of ostensibly desirable attributes then demonstrates that no RCV implementation can satisfy all of them.
But that's not to say that approval voting methods are flawless, either. One often cited desirable property is the Condorcet criterion, namely, if one candidate beats all others in head-to-head matchups, that candidate should be the winner. (There's a related Condorcet Loser criterion which is the inverse: if one candidate loses to all others in head-to-head matchups, that candidate should not be elected.)
Suppose we applied approval voting to a simplified view of the 2016 election. (Y to indicate approval, N to indicate disapproval)
- 48%: Partisan R. Trump (Y) > Stein (N) > Clinton (N)
- 48%: Partisan D. Clinton (Y) > Stein (N) > Trump (N)
- 3%: 3rd-party, very slight D lean. Stein (Y) > Clinton (Y) > Trump (N)
- 1%: 3rd-party, very slight R lean. Stein (Y) > Trump (Y) > Clinton (N)
Stein beats both Clinton and Trump in head-to-head matchups, 52% to 48%. But with an approval voting system (selecting Clinton with 51% approval), this fails the Condorcet criterion! (IRV is no better here, to be clear. And Condorcet methods have their own pathologies.)