The September/October 2008 issue of Technology Review magazine carried an article by us about range voting. The online edition also had additional online material about range voting vis-a-vis the 2008 US presidential race. We respond to reader questions and comments below, and correct infelicitous wording in the article and/or provide additional notes on it, at the end.
[You can also read our local copy of article and local copy of the additional online material; we've edited these to get rid of the "infelicitous wordings" but these local copies have been stripped of Tech Review's additional features like reader commenting.]
Comment on "strategic voting" (Anthony Lorenzo)
"What Mr. Smith, Mr. Sherman, and Mr. Carback III fail to not[e] about the inherent flaw in Range voting is that it is seriously susceptible to insincere or tactical voting... The authors of this article seem to imagine people won't work to game the system and will vote entirely honestly."
Response: To quote from the article:
"Range voting also outperforms all common alternative systems on average--no matter how many honest, strategic, and uninformed voters cast their votes, and no matter how many candidates run. (See William Poundstone's Gaming the Vote for a good summary of this analysis.)"
The question is not whether range voting permits strategic voting, but how range voting performs in comparison with other election methods if strategic voters are or are not present. We did not "fail to note" the issue of strategic voting.
Comment -- more on strategic voting (Anthony Lorenzo)
For instance, If i like Ralph Nader or Bob Barr as my choice for president, and I want my candidate to win, then in the Range Voting system, i can further their interests by rating every other candidate a 0 out of 9, or whatever the arbitrary top end number is determined to be in this entirely theoretical system. So i would vote the following: Nader 9, Barr 9, Obama 0, McCain 0, McKinney 9.
Response: This vote would be allowed by the rules of 0-to-9 range voting, but, contrary to Mr. Lorenzo's notion, it would not be a brilliant strategic move by that voter, but rather a strategic mistake. Why? Because, assuming the voter is highly confident that Obama or McCain will win (since, e.g. polls at the moment indicate they are vastly ahead of all the others) this voter would be wasting her vote by giving both McCain and Obama the same score of zero -- intentionally choosing (with high probability) to have zero impact on the election. Why bother to vote at all if you are going to have zero impact? So a voter who really wanted to further her own interests, instead would want to give Obama and McCain highly different scores. [She still could give Nader, Barr, and/or McKinney 9s each.]
If this voter went with (say):
Nader=McKinney=Barr=
that would make more sense. It would still be a strategic exaggeration, but it now would be a strategically sensible exaggeration.
Smith's computer simulations indeed allowed voters to do precisely that (we also did other simulations with mixtures of strategic and honest voters, honest-only, etc) and the result was that range voting outperformed all other commonly-proposed voting systems, no matter what the voter mix, as measured by Bayesian Regret.
We also reject the claim range voting is "entirely theoretical." As the article said, it has been around in some form for a long time; and those were not the only examples. Venetians successfully used range voting with a three-point scale between 1268 and 1797, i.e. far longer than the USA has existed. A system that was in essence approval voting worked well in the U.S. Electoral College between 1788 and 1800 and was also used in the Soviet Union in June 1987. Numerous organizations use forms of approval or range voting, there was even a board room product in the 1960s where members turned dials to indicate their strength of preference on each issue.
Comment -- how often such strategic voting would occur (Anthony Lorenzo)
"If i can change my vote, as in the example above, to increase the likelihood that a candidate that matches my values will win, I and 99% of the other people in America will do so."
Response: Fewer than 25% of voters in a 2004 exit poll study cast an all-max-or-min style range vote (see paper #82) and fewer than half even employed both endpoints of the numerical range! Therefore, Lorenzo's "99%" estimate for the number of strategic range voters is drastically refuted by actual data. A French 2007 range voting study similarly refutes it.
Why might this be the case? Perhaps the extra expressivity of range voting inspires voters to be more honest.
Alternative hypotheses: Or perhaps they are being partially honest and partially strategic when casting their vote, thus sacrificing some but not all of their ability to "change their vote... to increase the likelihood." We also note, the strategic incentive to exaggerate about candidates such as McKinney whose winning chances are very tiny, can easily be so tiny it is outweighed by other incentives. Suppose the voter thinks it is 99.999...9% probable that McKinney cannot come close to winning (with 100 nines). Numbers such as that can easily be a realistic assessment. Suppose this voter votes 9 and 0 for Obama and McCain and 3 for McKinney. Now the voter asks herself "should I dishonestly alter my McKinney score to 0 to increase my vote's impact?" The probability that exaggeration will have an effect is tiny. In this way, RV encourages honest participation.
Further, even if this 99% claim had been true, then as we remarked, our computer simulations still indicate range outperforms all other commonly proposed alternative systems, including instant runoff.
By contrast, this sort of strategic voting does happen frequently with other voting systems. For example, in Australia (which has used instant runoff voting [IRV] for about 80 years) about 95% of the voters rank A top and B bottom-or-second-to-bottom (or the reverse) where A and B are Australia's two major parties, each election.
The rules of IRV then prevent a third-party candidate from winning, which is why, of the 150 federal IRV seats, zero are occupied (as of 2008) by a third party member. Zero also were occupied last election cycle. This is despite the fact every such seat was contested by at least one third party and 7.03 candidates ran (on average) for each seat. Australia's third parties are trying to get rid of IRV in Australia.
Comment -- Instant Runoff and strategic voting (Anthony Lorenzo)
"Unlike Range Voting--and even worse Approval Voting--Instant Runoff Voting provides very insignificant incentives to rank candidates different than a voter's sincere preferences. There are highly improbable (never happened thusfar in a real world election in over 90 years of use in Australia, 50 years plus in Ireland, and quite a few other places in recent times) mathematical possibilities which if a voter changed the order of their rankings, the outcome would change."
Response: As mentioned above, data show that an overwhelming percentage of IRV voters in Australia vote strategically.
Mathematical analysis and computer simulations both indicate that "favorite betrayal" in instant runoff voting (where a voter finds it strategically better not to rank her true favorite topmost) happens about 20% of the time in 3-candidate "random elections" assuming voter perfect knowledge. This is not "very insignificant." Further, with realistic imperfect knowledge, favorite betrayal for third-party favorites seems advisable 100% of the time in this same model, essentially because the 20% chance favorite betrayal will "work" far exceeds the historical chances that a third party candidate would win (in which case favorite betrayal strategy would instead "backfire").
Lorenzo's second sentence claims that no vote changes have ever affected an election outcome under instant runoff voting. There are two responses to that claim (a) it is obviously false if we speak of vote changes for multivoter blocs [obviously, in every election, if a large enough bloc of voters changes their vote, it will alter the outcome]; (b) it might be true if we speak of only a single voter changing a single vote [i.e. Lorenzo then basically would be making the difficult-to-verify claim that no "tie" or "1-vote-off-a-tie" scenario has ever occurred in an IRV election, albeit it is known that in the 135-candidate non-IRV California Gubernatorial recall election of 2003 alone, several occurred]. In either case, we fail to see what possible relevance to anything Lorenzo's sentence had. Perhaps therefore what Lorenzo was trying to (but failed to) say in this sentence, was that no example has ever occurred, of an IRV election in which voters were motivated to misorder. If so, there are a vast number of counterexamples to him, e.g. Ireland 1990 and numerous Louisiana governor elections.
Comment -- FairVote's mission (Anthony Lorenzo)
"...to categorize Fairvote's mission as purely Instant Runoff Voting, when in fact they also promote Single Transferrable Voting and proportional representation systems for legislative, multi-member bodies, is just completely innacurate. Fairvote also promotes DC voting rights (in congress), a national popular vote for President, and many other issues you can read for yourself on their Web Site: www.fairvote.org . Fairvote is far from a single-issue advocacy group as they are painted in this article."
Response: The article's sole statement was this:
"Groups such as the FairVote Center for Voting and Democracy in Takoma Park, MD, advocate the Australian practice of instant-runoff voting (IRV). "
This is a true statement. The article did not "categorize Fairvote's mission as purely Instant Runoff Voting" and never claimed FairVote was a "single-issue advocacy group." We are happy about FairVote's work in many directions but their advocacy of Instant Runoff Voting seems ill-advised and often has been done in a misleading way.
There is evidence that when third parties start to get closer to mainstream candidates spoiler effects would be common. After the initial round of plurality voting in the 2007 French presidential election, the top three candidates were Sarkozy (31.2%), Royal (25.9%), and Bayrou (18.6%). In the runoff among top two, Sarkozy beat Royal. IRV presumably would have produced the same result. Yet polls indicated that Bayrou would have won face-to-face races versus either Sarkozy or Royal. An exit-poll study by M. Balinski and R. Laraki of the École Polytechnique concluded that Bayrou would have won with range voting.