Compared to BAV, ranked-choice voting asks much more of voters. And it puts much more effort into evaluating the responses, so it is not unreasonable to ponder whether all that extra effort is
justified.
These two voting systems have very little in common. They ask voters fundamentally different questions, and they interpret the answers in dramatically different ways. As with comparing apples to oranges, it is right to wonder whether a comparison could ever be meaningful.
The title adopts the term, ranked choice voting, because that terminology has come into widespread use in recent years. Unfortunately, this phrase is easily confused with the more general notion of ranked voting, so in the remainder of this article, I will abbreviate ranked-choice voting as IRV, an acronym from only a few years ago when the voting system was still generally called, instant runoff voting.
With IRV, a voter asked for a list of the candidates in what the voter determines is a decreasing order of preference. This ballot data provides a quite fine-grained description of voter preferences, at least for candidate on a voter's list, but the voter can abstain from listing all the candidates. The candidates not on the list are ignored, effectively lumped together at the same position, below any on the voter's list. IRV tallying involves multiple counts of ballots, with each count removing just a single candidate. With nine candidates for example, the ballots must be counted as many as eight times to determine a single winner.
Instead, BAV voters are asked to indicate which candidates they support and which candidates they oppose. As with IRV, abstentions are allowed but with they are treated in a neutral manner by ignoring them. Abstentions neither increase nor decrease the BAV candidate's chance of winning election. Tallying a BAV election requires only as single count of the ballots to determine the number of opposition votes and the number of support votes for each candidate. The net vote is then computed by simply subtracting the opposition count from the support count and the candidate with the largest net vote wins the election. The net vote for a candidate can be negative, zero or positive, reflecting clearly whether voter sentiment is negative neutral or positive. The number of abstentions would likely be counted as well. That count would be of academic interest, but for BAV elections, abstentions play no role in determining winners.
BAV requires little effort from voters. Even with several candidates such as we often have in our primary elections, with little difficulty, most voters have a clear understanding of which candidates they prefer, just as they know which candidates they oppose; there may also be a few candidates that a voter neither supports nor opposes. Voters simply fill out the ballot to put each candidate in one of these three categories. Since voters can easily say explicitly how they feel about each candidate they should not be tempted to do otherwise.
With two candidates, ranking them for an IRV ballot is not especially difficult but with only three, possible difficulties arise, such as when a voter finds two of the three to seem so similar that choosing between them stressful; but choose they must, perhaps in several instances while reflecting on how to rank the candidates. Sometimes a voter may rank a candidate better based on some issues, but worse based on other considerations; weighing the alternatives can prove difficult, figuratively only to be resolved with the toss of a coin. As illustrated in an earlier article, it is possible for a voter's honest preferences to form a loop. Nevertheless, the voter required to break that loop and somehow rank the candidates in decreasing order. As with driving a square
peg into a round hole, compromise and some force may be required. An IRV election does collect fine details about voter preferences, but it is a mistake to trust its accuracy.
Would it be possible to predict outcomes from one of these two systems if given the ballots for the other? That might provide a way to compare the two. From BAV ballots, there seems no possibility of determining what the IRV would have been; the very fine details from an IRV election could not possibly be determined from the much coarser data collected in a BAV election, and this would be clear even aside from the many arbitrary decisions IRV voters would probably make in constructing their ordered lists.
But IRV does ask much more from the voters so a derivation in the opposite direction might be feasible. The major difficulty is that with IRV the voter gives no indication of which candidates are opposed or supported and this is the very information that BAV requires.
However, only minor changes to the IRV ballot could provide that information. Voters could each be asked to provide two markers, one to identify both the last candidate (in that voter's list) that the voter supports and the second to mark the first in the list that the voter opposes. These two markers divide a voter's list into three pieces containing in turn the supported candidates, the abstentions and the opposed candidates. Instrumenting an IRV election in this way could provide a way to compare IRV and BAV based on outcomes.
But there would remain a problem with the handling of abstentions on the IRV ballots. One approach would be to add the count of abstentions to the opposition total for the BAV count. That would conform to the treatment of abstentions in an IRV election, but it likely would differ from how the voter would vote in a BAV election. So instead, one might add the number of IRV abstentions to the number candidates in the middle range of IRV votes. Probably that would reflect the intent of most voters, but unfortunately not all of them.
A few voters would understand that IRV abstentions are treated as if these candidates were the very least favored by the voter; knowing this, these voters probably rank candidates differently, with abstentions intended more as indications of opposition. For these IRV ballots, it would be more accurate to add abstentions to the opposition count. But there is no good way to determine which ballots this applies to; adding a question to the IRV ballot to determine that that would likely distort the IRV voting. However, after the election, IRV voters could be polled to determine the percentage of ballots that would require this correction. So, comparing this approach for comparing IRV outcomes with BAV outcomes does seem to be a possibility.
Voter data from these instrumented IRV elections could also serve to evaluate a hybrid voting system that combines BAV with IRV. Such a hybrid system would first tally the votes using BAV in order to select three finalists for competition in a subsequent IRV runoff; at most, only two rounds of IRV ballot counting would then be needed. The hope for this hybrid system is that the fine detail gathered in by the IRV ballots might prove valuable mostly in the last few ballot counts.
