By Rob Richie, Terry Bouricius, and Philip Macklin
Published October 12th 2001 in Science
Steven Brams and Dudley Herschbach are right about the defects in the
plurality voting system used in most U.S. elections (Editorial, "The
science of elections," 25 May, p. 1449). But, on both theoretical and
practical grounds, they are wrong to tout approval voting (AV) over
instant runoff voting (IRV).
Used for decades in Australia and Ireland and considered in 13 U.S. state legislatures this year, IRV allows voters to rank candidates in order of preference. A voter's best strategy is to sincerely rank the candidates. If no candidate gets a majority of first preferences, candidates at the bottom are sequentially dropped. Each ballot cast for those eliminated candidates is added to the totals of the next choice indicated on that ballot until a candidate achieves a majority. IRV duplicates a series of traditional runoffs, but without the need for additional elections that cost taxpayers and candidates more money and often lead to falloffs in voter participation (1).
In contrast, AV is a binary system, where the voter can only indicate "yes" or "no" for each candidate. The problem is that voters rarely have binary views about a range of candidates. Assume a voter sees Z as most favored, Y as less favored, and X as unacceptable. By voting for "acceptable" candidates Y and Z, the voter could cause Z to lose. But by voting only for Z, the voter makes it easier for X (the unacceptable candidate) to win (2). The voter will be torn between voting defensively against X or strategically for Z because voting for a second choice counts directly against your first choice.
Approval voting has another important real-world flaw. Political behavior has much to do with what is rewarded by the election system, and AV would exacerbate one of the worst aspects of U.S. campaigns: avoidance of substantive policy debate. Because a candidate could lose despite being the first choice of an absolute majority of the electorate (3), smart candidates would avoid controversial issues that alienate any significant number of voters. Smiling more and using policy-empty themes like "I care" will not clarify the important choices leaders must make. Those rewarded by AV could be characterized as "inoffensive" more than "centrist."
IRV strikes a better balance. It rewards candidates who stand out on policy enough to gain first-choice support, yet encourages coalition-building and fewer personal attacks as candidates seek to be the second choice of other candidates' supporters.
These arguments help explain why IRV is used and proposed far more often than AV, and why next year Alaska and San Francisco will hold ballot measures to implement IRV for their major elections (see for details). IRV is the right system for the United States' high-stakes elections with a single winner.
Rob Richie
Terry Bouricious
Phillip Macklin
References and Notes:
1. Ireland's 1990 presidential race provides an example of how IRV works. In the first round, Brian Lenihan won 44% of first choices, Mary Robinson won 38%, and Austin Currie won 17%. After Currie's elimination, Robinson had clear majority support. She won 53% to 47% in the second round of counting. Without IRV, Currie would have been a "spoiler" and handed the presidency to Lenihan.
2. Imagine an AV election with 100 voters. After 98 ballots are counted, the results give 55 approval votes to candidate Z, 60 votes to candidate Y, and 61 votes to candidate X. The two remaining ballots were cast by those voters who really liked Z and intensely dislike X. If knowing these results in advance, they would want to block candidate X by casting votes for both candidates Y and Z. Now suppose the results instead gave 60 approval votes to candidate Z, 61 to Y, and 55 to X. The final two voters in this case would want to elect candidate Z by not voting for Y. But only in imaginary elections can we know the results in advance. Without that advance information, supporters of candidate Z are in a quandary-as, in fact, are supporters of other candidates. IRV's alleged mathematical deficiencies, on the other hand, have almost no strategic impact because they depend on voters making complex calculations with advance knowledge of election results"
3. Here is how AV can allow a candidate with strong majority support to lose in an election with 100 voters. Under plurality voting, Candidate A is the favorite choice of 65 voters, candidate B is preferred by 25, and candidate C is the top choice of only 10. Candidate A is the unambiguous winner, and C is a distant third. Under AV, however, many voters might pick C as a weak, but tolerable alternative. The final count might give 70 approval votes to A, 35 votes to B, and 75 votes to C. Candidate C would win.
Used for decades in Australia and Ireland and considered in 13 U.S. state legislatures this year, IRV allows voters to rank candidates in order of preference. A voter's best strategy is to sincerely rank the candidates. If no candidate gets a majority of first preferences, candidates at the bottom are sequentially dropped. Each ballot cast for those eliminated candidates is added to the totals of the next choice indicated on that ballot until a candidate achieves a majority. IRV duplicates a series of traditional runoffs, but without the need for additional elections that cost taxpayers and candidates more money and often lead to falloffs in voter participation (1).
In contrast, AV is a binary system, where the voter can only indicate "yes" or "no" for each candidate. The problem is that voters rarely have binary views about a range of candidates. Assume a voter sees Z as most favored, Y as less favored, and X as unacceptable. By voting for "acceptable" candidates Y and Z, the voter could cause Z to lose. But by voting only for Z, the voter makes it easier for X (the unacceptable candidate) to win (2). The voter will be torn between voting defensively against X or strategically for Z because voting for a second choice counts directly against your first choice.
Approval voting has another important real-world flaw. Political behavior has much to do with what is rewarded by the election system, and AV would exacerbate one of the worst aspects of U.S. campaigns: avoidance of substantive policy debate. Because a candidate could lose despite being the first choice of an absolute majority of the electorate (3), smart candidates would avoid controversial issues that alienate any significant number of voters. Smiling more and using policy-empty themes like "I care" will not clarify the important choices leaders must make. Those rewarded by AV could be characterized as "inoffensive" more than "centrist."
IRV strikes a better balance. It rewards candidates who stand out on policy enough to gain first-choice support, yet encourages coalition-building and fewer personal attacks as candidates seek to be the second choice of other candidates' supporters.
These arguments help explain why IRV is used and proposed far more often than AV, and why next year Alaska and San Francisco will hold ballot measures to implement IRV for their major elections (see for details). IRV is the right system for the United States' high-stakes elections with a single winner.
Rob Richie
Terry Bouricious
Phillip Macklin
References and Notes:
1. Ireland's 1990 presidential race provides an example of how IRV works. In the first round, Brian Lenihan won 44% of first choices, Mary Robinson won 38%, and Austin Currie won 17%. After Currie's elimination, Robinson had clear majority support. She won 53% to 47% in the second round of counting. Without IRV, Currie would have been a "spoiler" and handed the presidency to Lenihan.
2. Imagine an AV election with 100 voters. After 98 ballots are counted, the results give 55 approval votes to candidate Z, 60 votes to candidate Y, and 61 votes to candidate X. The two remaining ballots were cast by those voters who really liked Z and intensely dislike X. If knowing these results in advance, they would want to block candidate X by casting votes for both candidates Y and Z. Now suppose the results instead gave 60 approval votes to candidate Z, 61 to Y, and 55 to X. The final two voters in this case would want to elect candidate Z by not voting for Y. But only in imaginary elections can we know the results in advance. Without that advance information, supporters of candidate Z are in a quandary-as, in fact, are supporters of other candidates. IRV's alleged mathematical deficiencies, on the other hand, have almost no strategic impact because they depend on voters making complex calculations with advance knowledge of election results"
3. Here is how AV can allow a candidate with strong majority support to lose in an election with 100 voters. Under plurality voting, Candidate A is the favorite choice of 65 voters, candidate B is preferred by 25, and candidate C is the top choice of only 10. Candidate A is the unambiguous winner, and C is a distant third. Under AV, however, many voters might pick C as a weak, but tolerable alternative. The final count might give 70 approval votes to A, 35 votes to B, and 75 votes to C. Candidate C would win.
