Condorcet.orgIRV is a commonly proposed replacement for plurality. Each voter ranks each candidate (or possibly only some of them) on a ballot. The method starts by finding the plurality loser candidate, that is the candidate who is ranked highest on the fewest number of ballots. This candidate is eliminated. In successive rounds, the new loser is found by finding the candidate who is ranked highest between non-eliminated candidates on the fewest number of votes. This process of elimination proceeds until only one candidate is left. This candidate is declared the winner.
Often opponents of IRV make the claim that it is capricious. That it makes the decision based on quirks of the balloting that can't reasonably be thought to have anything to do with who is the most likely best choice.
Consider the following example:
38 ABC
7 ACB
15 BAC
15 BCA
25 CBA
Round 1: A 45 B 30 C 25 - C is eliminated, the 25 CBA votes are now
considered votes for B
Round 2: A 45 B 55 - A is eliminated, B wins
Here, B is the winner just as in Ranked Pairs. But now let's imagine a similar election. All that will be changed is that some people will rank A lower on their ballots. No one will change their ballots in any other way.
38 ABC
7 CAB - A has been lowered by these voters
15 BCA
15 BAC
25 CBA
Round 1: A 38 B 30 C 32 - B is eliminated. BAC go to A, BCA go
to C
Round 2: A 53 C 47
B is still the winner in Ranked Pairs. However, now in IRV, A is the winner. This is despite the fact that the only difference between these two elections is that some people who preferred A to C in the first, prefer C to A in the second. This isn't reasonable evidence that A should win. This strange property is referred to by the strange name of non-monotonicity.
Here's more evidence that IRV is capricious. Remember that I started from the premise that an election method should pick the candidate most likely to be the best candidate, based on the ballots. Let's imagine that IRV does this. Now, what if we asked voters to instead choose the worst candidate. Obviously, if voters are sincere, their rankings would all be exactly reversed. We could then use IRV to find the most likely answer to the new question, that is, "who is the worst candidate?"
Consider the following example:
40 BCA
25 CAB
35 ABC
IRV proceeds by dropping C, then A wins vs B.
For Ranked Pairs, we get the following table. The margins
are in brackets.
| A | B | C | |
| A | X | 60 (20) | 35 |
| B | 40 | X | 75 (50) |
| C | 65 (30) | 25 | X |
Lock B>C 50
Lock C>A 30
Skip A>B 20, which contradicts B>C, C>A
Final ranking is B>C>A
So, who is right? Let's check what the methods say when voters are asked to pick the worst candidate.
40 ACB (reverse of BCA)
25 BAC (reverse of CAB)
35 CBA (reverse of ABC)
In IRV, B is dropped, and A wins against C. So, if the voters use IRV, they can choose the same candidate both for best and worst. IRV doesn't just contradict Ranked Pairs, it contradicts itself.
Ranked Pairs doesn't have this problem.
| A | B | C | |
| A | X | 40 | 65 (30) |
| B | 60 (20) | X | 25 |
| C | 35 | 75 (50) | X |
We get the mirror image of the previous vote
Lock C>B 50
Lock A>C 30
Skip B>A 20, which contradicts A>C, C>B
A>C>B
So, Ranked Pairs is consistent with itself. Incidentally, notice that in the first example, IRV chose the candidate that Ranked Pairs rates as worst.
Following are an assortment of arguments frequently used to support IRV.
45 A B C
25 B
30 C B A
Now, it seems unrealistic to imagine that such a large number of voters would all vote in so rigid a way. Shouldn't there be at least one person who votes A C B, or C A B, or B A C. Of course. However, if a large majority of people do vote on some kind of left to right scale, it won't make enough difference to change the result. Try some examples for yourself.
Here is an example where people are not voting in any kind of lock-step way, but IRV still fails to select the Condorcet winner.
40 A B C
5 A C B
6 B C A
14 B A C
31 C B A
4 C A B
In Ranked Pairs, if you vote A over B, this decision is given the same weight whether it is found at the top or bottom of the ballot, and independent of how many candidates fall between A and B on the ballot. However, Ranked Pairs clearly does give the greatest preference to higher choices, in the sense that they are scored as preferred to all lower choices.
What Ranked Pairs doesn't do is declare that a voter's choice of first over second candidate should be given more weight than his seventh over eighth, for example. Often it is implied that IRV does this, but in fact it does something very different. For the first round, only first preferences are considered important. As candidates are eliminated, new preferences are revealed, and given full weight, as if they were first preferences. This means that choices very low on a voters ballot can end up weighted just as strongly as first preferences.
If we believe that IRV decides things based on the relative strength of ranking, we would have to believe that in the first round voters are only concerned about their first choice and view all other options as equal. Then, as candidates are eliminated, voters change their minds and decide that some later choice is all they care about, and feel as strongly about as anyone else does their first choice.
Too much is often read into the fact that Ranked Pairs always tabulates lower choices on the ballot while IRV often hides them. It's important to remember that if there is a group of candidates none of whom are majority preferred to a candidate outside their group, then none of them can win. What this means is that if you're worried that your arbitrary ranking of fringe candidates at the bottom of the ballot will somehow affect the outcome of the election, don't be.
It is true that in IRV on average the lower choices are less likely to have an effect on the outcome, since many won't be revealed. However, lower choices are also less likely to affect the outcome in Ranked Pairs. This is because lower choices will tend to lose over all. As a result they won't have victories locked in against the higher ranked candidates.
The argument is that because sometimes in Ranked Pairs, entering a lower choice can help defeat your first choice, voters will be paralyzed with fear and unwilling to vote beyond the first choice. However, the opposite effect can also be shown, that is, where ranking more candidates causes a voters first choice to win. In fact, the effects will on average balance out, so it makes no sense to not enter lower choices, as a general strategy.
Sometimes IRV advocates claim that the procedure described by IRV, of transferring votes only when a candidate drops out of contention, is what the individual voters want. Sometimes it is implied that the rank ballot is not merely a list of candidates in order from best to worst, but a procedure for how the voter wants his or her vote tabulated.
But consider the following result
45 A B C
25 B C A
30 C B A
An IRV advocate would say that what the A voters want is to hold onto their vote until B is defeated, they get to go one-on-one with C, and finally allow their last choice to be elected. In fact, in cases where there is a Condorcet Winner, and where IRV does not choose it, a majority would by definition prefer the Condorcet Winner to the IRV winner. This means that the majority would be better off having voted for the Condorcet Winner in first place instead of backing their candidates until they fell out of contention.
The IRV choice is chosen by a majority over someother candidate. The other candidate is chosen by IRV. Even if you picked two candidates at random, and found the majority preference between them, you could claim this weak kind of majority rule.
Some IRV advocates are rather enamored with the fact that IRV always ends by pitting one candidate up against another and declaring one the majority winner. All this means is that the winner has majority support against at least one other candidate. This is true of Ranked Pairs as well, but Ranked Pairs also says that if a candidate has a majority against each other candidate, that candidate wins. This is not true of IRV.
If you really want, you could pick any one of the candidates beaten by the Ranked Pairs winner, and simulate a one-on-one showdown between these two. Now the Ranked Pairs winner still wins, but you can say the method ended by giving the candidate a majority.
What this refers to is the fact that if a voters first choice is eliminated, their second choice carries full weight. Often it is mistakenly concluded that this means that it is never beneficial to vote for a compromise "lesser evil" over a favorite. But consider the now familiar example.
45 A B C
25 B C A
30 C B A
The A voters would be better served by voting B A C instead of A B C. So, lesser of two evils type thinking is still useful. In general, under IRV, you can either play it safe by voting for whom you think is the Condorcet Winner in first place, or you can adopt the more risky strategy of voting for your first choice, with the danger that this may cause one of your least favorite to be elected.
In fairness, it is worth pointing out that all methods will reward strategic compromise voting in the case where there is no sincere Condorcet winner. However, methods that do not always select the Condorcet winner (unlike Ranked Pairs) will also have this problem in these situations.
Usually when explaining IRV, advocates spend a great deal of time convincing you that the candidate who receives the most votes is not necessarily the best candidate. Then, usually without any explanation, they assume that the candidate with the least number of votes is the worst candidate. They assume that as a former believer in plurality (and that's almost everybody) you started with this assumption.
There's nothing wrong with eliminating candidates. But eliminating them based on the assumption that the lowest plurality candidate is the worst doesn't make sense, unless you actually believe that first preference support is the best indicator of popularity. If you believe that, you can't agree with IRV.
This site is primarily about which method is best. It may well be that in some places IRV reforms have some momentum, and that it is too late to propose a different method. The advantages or disadvantages of IRV with respect to plurality are outside the scope of this site. However, if you agree that Ranked Pairs is superior to IRV, it makes sense to keep this fact in mind.
Strategy is such a large topic in itself that I have an entire page devoted to it.