In last November's election, Maine voters approved a ballot issue that directs future major office elections in Maine to be conducted using ranked voting. Although the campaign surrounding Question-5 rarely (if ever) mentioned the name IRV (Instant-Runoff Voting), it seems clear from the wording of the issue that IRV was the particular system of ranked voting to be adopted. Specifically, the wording of the Question-5 defined ranked-choice voting as "the method of casting and tabulating votes in which voters rank candidates in order of preference, tabulation proceeds in sequential rounds in which last-place candidates are defeated and the candidate with the most votes in the final round is elected".
But in common usage, "ranked-choice voting" refers to any of the many possible systems in which voters are asked to list the candidates in order of preference. Clearly IRV is such a system but IRV defines the vote, counting only one of many possible alternatives. However, IRV has been promoted so widely that it is widely assumed there is only one system of ranked-choice voting.
If we can assume that the objective of a voting system is to accurately assess the wishes of the voters there are a number of things that seem perverse or at least odd about IRV. Some of these have been discussed in earlier articles of this series:
In this article we return to this topic to examine yet another aspect of IRV that is particularly interesting and revealing, not just about IRV but about other systems of voting as well. To describe this peculiarity, imagine that you were unfamiliar with IRV but you wanted to simulate a series of elections in which one (or perhaps more) candidates were to be eliminated with each iteration of counting. How would you decide which candidates to remove (keeping in mind that you want the decision to be a democratic one)? To make this simpler, just consider what the the best way would be to decide which of the candidates to eliminate first. Would it occur to you to ask voters which candidate they like the most?
Maybe instead would you ask yourself what you want to know from the voters and ask them for that. You are going to remove one candidate from contention so why not simply ask the voters how you should proceed? Thinking democratically, you would surely ask each voter which candidate the voter would choose to remove first. With an open-minded approach, would it not seem really odd to instead ask a voter which candidate she likes the most?
This line of thinking leads to something closely resembling IRV, but in a critical way it is the opposite of IRV. Voters are asked to rank the candidates, starting with the one they would most like to have removed from contention and second the one that the voter would most want removed after the first listed is removed -- and so on down at least to include all candidates that the voter views as unacceptable.
The first candidate to be removed will be the one that the most voters want removed. In the subsequent rounds, ballots will be counted as if the eliminated candidates were removed from their ballots and counted as a vote against the first candidate on their ballot that has not yet been eliminated. If ties (or near ties) occur then more than one candidate could be removed at a time, but the recounts would continue until only a single candidate remained as the winner.
It will be convenient to have a name for this alternative system of ranked voting, so let's call it IRRV (Instant-Runoff-Removal Voting) since voters are asked to rank candidates according to their preferred removal order. At first glance, it might seem that a voter would construct exactly the reverse order for IRRV as she would for IRV and that is a useful way to conceptualize it -- but that is an overly simplified view that is apt to be only approximately the case. Pollsters have come to realize that the phrasing of their questions will influence their answers and that lesson probably applies to elections as much as it does in polls. Some examples may be in order.
With ten candidates in a race, if you ask a voter to list the candidates in order of preference you may possibly get a list of ten candidates back, but it is likely that there will be candidates left off of some lists -- particularly candidates that the voter does not know. But the last candidate on the list is very likely to be the candidate that the voter likes least. With IRV such a shortened list will be counted in a way that is at odds to the voter's likely wishes. Unfortunately, if that last-place name on the list is ever counted it will help that candidate to win election. Putting that name last on the list does, however, give the voter the illusion of expressing what great opposition the voter feels about that candidate and that is why a voter is apt to include that name.
However, if instead you ask a voter in what order would he like candidates to be eliminated, the first on the list is apt to be one that the voter particularly dislikes but all of the candidates the voter knows nothing about are apt to make it onto the list. Since the voter does not want his favorite candidate to ever be eliminated, that favorite candidate (and maybe several favorite candidates) will not be on the list at all. Leaving the candidate off the list of those to be removed from consideration is clearly the best way for the voter to express how she feels about that most-favored candidate. But more importantly, with IRRV this intuitive idea of what the voter feels his vote means is in concert with the way the vote is actually counted.