We Can Do Better than Instant Runoff Voting 8/19/08 Some well motivated progressives have fallen into a trap while pursuing a solution to the problem of how to support a third-party candidate without being called a "spoiler". They advocate "Instant Runoff Voting" (IRV), which does accomplish the desired purpose in cases where the third-party has no chance to win. But it is fundamentally defective in a number of important ways, including failure to eliminate the spoiler dilemma once the third-party becomes stronger. Fortunately another system, "Approval Voting" (AV), does the job without the problems associated with IRV. We are all accustomed to "Plurality Voting" (PV), where we cast a vote for a single candidate and the one receiving the most votes wins. A serious problem with PV is that it effectively pressures many people into NOT voting for the candidate they think best represents their views. Suppose, for example, that polls show that the percentages of voters favoring candidates R, D, and G (you might consider these as representing Republican, Democrat, and Green) are 40, 35, and 25, respectively. Assuming that the great majority of G supporters would favor D over R, a large portion of them might conclude that, since G can't win, the best that can be done is to abandon G and vote for D. In our example, this reasoning would lead to a large number of G supporters voting for D instead of G. This, of course, is what happened in 2000, as the great majority of people who thought Ralph Nader was the best candidate voted instead for Al Gore. (As it turned out, it was not enough to elect Gore as half the registered Democrats in Florida did not vote, and over 200,000 of them voted for Bush. Nevertheless, many people blame the 90,000 Florida Nader voters for Gore's defeat, calling Nader a "spoiler".) A parallel situation exists elsewhere in the political spectrum between Republicans and Libertarians. IRV is an attempt to solve this problem. Under IRV, each voter ranks all the candidate in order of decreasing preference. A candidate receiving a majority of first-place votes is the winner. If nobody receives such a majority, then the candidate with the fewest first-place votes is eliminated from all ballots, and the process is repeated until a winner is found. As an example, consider a 3-candidate election similar to the above example. Suppose the percentages of votes are as shown in Example 1 below, where, e.g., 35 RDG means that 35% of the voters ranked R first, D second, and G third: Example 1 35 RDG 5 RGD 25 DRG 10 DGR 20 GDR 5 GRD Since no candidate has a majority of first-place votes (e.g., R has only 40%), the votes for the third-place candidate, G are deleted, converting the votes to 35 RD 5 RD 25 DR 10 DR 20 DR 5 RD Now D has a majority of first-place votes (55%), and is thereby the winner. So, even tho G had no chance of winning, G-supporters who preferred D to R were able to top-rank their favorite, while still helping their second choice candidate defeat the candidate they most disliked. This sounds great and accounts for support of IRV by some genuine progressives. The bad news is that there are serious problems with the tabulation of IRV ballots, AND that under other equally plausible circumstances, the spoiler effect returns. IRV can also behave strangely under other conditions. The good news is that there is another system, with no tabulation problems, that eliminates the spoiler problem under ALL circumstances. Let's look at some bad news first. Suppose that, at some future time, support for the G candidate grows to make G a real contender. Assume the following predicted IRV ballot set, which, tho optimistic from the G point of view (the Gs have overtaken the Ds), is certainly plausible. Example 2 35 RDG 5 RGD 30 GDR 5 GRD 20 DRG 5 DGR Now, the Ds have the fewest first-place votes and so are eliminated to produce: 35 RG 5 RG 30 GR 5 GR 20 RG 5 GR And this makes R the winner (60-40). Such a scenario returns G-supporters to the "spoiler" role. If more than 25 of the 35 G-supporters abandoned G and made D their first-place choice, then D would win. Back to square-1 with respect to the lesser evil issue! There are many other situations in which IRV motivates strange behavior on the part of voters. Rather than discuss these, I will point out another very serious fault of IRV. When PV, or virtually any other voting system considered by voting theorists, is used, tabulating election results can be decentralized. I.e., ballots can be tabulated in the precincts, and appropriate summaries forwarded to a central point for merging to generate the overall results. For IRV, unless some candidate receives a majority of first-place rankings, all ballots must be tabulated together in one central place, and the process is not simple. In the case of a large state, millions of votes would have to be processed centrally for a presidential, gubernatorial, or US Senate race. This opens the door to all sorts of error and fraud in the course of ballot transmission, offers more opportunities for fraud during tabulation (whether done manually or by machine), and makes manual recounts very difficult, time consuming, and costly. Now for the good news. There is an alternative voting scheme, called "Approval Voting" (AV), which eliminates the spoiler effect in ALL cases, and where vote tabulation can be decentralized and is no more difficult than for PV. Voters simply vote for ALL candidates they approve of; the candidate with the most approval wins. Under this system, one can approve a favorite candidate F while also approving a lesser evil candidate if so desired. In the Example 1 above, the AV votes might have been as below, where the line "10 RD" means that 10% of the voters approved both R and D: 25 R 10 RD 5 RG 20 D 10 DG 5 DR 20 GD 3 GR 2 G So D would have won with 65%, versus 48% for R and 40% for G. AV votes corresponding to Example 2 (where IRV failed) might have been: 25 R 10 RD 5 RG 20 G 10 GD 5 GR 15 DG 6 D 4 DR So G would have won with 55%, versus 49% for R and 45% for D. Under AV, approval of X can only help X win; it cannot affect the scores of other candidates. If F is one's favorite candidate and if B is a possible winner that one strongly opposes, then approving F and NOT approving B is NEVER a mistake. A fundamental weakness of IRV is that a voter cannot express the difference between (1) "I like A best, B a little less, and I hate C" and (2) "I like A best, I hate B, and hate C even more". In either case, the IRV vote would have to be ABC. For the same situations, under AV, we could differentiate between 1 and 2, respectively, by approving BOTH A and B for case 1, or only A for case 2. It s not hard to find examples where such distinctions result in different winners. A generalization of AV is "Range Voting" (RV), where, for each candidate, the voter assigns an integer in some range, such as 0 thru 4. (AV is the special case where the range is 0, 1). This allows voters to express their views more precisely. While a bit more complex than AV, it is not terribly hard to tabulate, and never leads to bizarre scenarios. On balance, it would probably be best to move from PV to AV, which would require only minimal changes in election procedures and solve the major problems of PV. Later, a refinement to RV might be considered. There is no reason at all to adopt IRV as opposed to AV. Further discussion of this issue, with references and more examples, is in "Instant Runoff Voting: Looks Good--But Look Again". Articles on other topics of interest to progressives, including additional election-related issues, can be found at "Ends and Means".

Comments

Click Here to View Comments or Join the Conversation