Anthony Lorenzo, chair of the Florida-based Coalition for Instant Runoff Voting, in January 2007 produced a critique of range voting (based on his examination of the RangeVoting.org web site) in response to a request from a Florida radio station. His critique consisted of quotes from several pages of the RangeVoting.org web site followed by his responses to those quotes. The radio show in turn asked us to respond to Lorenzo, which we do below in the same style.
(If desired, you can also Read Lorenzo's original comments in full, verbatim.)
Original text: Range voting is simpler than IRV.
Lorenzo: Simpler than 1, 2, 3? That is a matter of opinion.
Our response: Range voting is simpler in these objective senses:
A. Write a range voting computer program and an IRV computer program (preferably with error-checking of the inputted votes). The range voting program will be shorter and will run faster, assuming essentially any reasonable programmer does it. (This, called "Kolmogorov Complexity" is the standard objective metric used by scientists to assess "simplicity.")
B. Range voting runs on all today's voting machines without any modification (including non-computerized machines). IRV does not.
C. Voters experimentally make fewer ballot-invalidating errors when using range than when using IRV.
D. Not simple enough for you? Okay, range is a parameterized class of methods, with the parameter being the number of ratings. The simplest kind of range voting is called "approval voting." It has only two ratings, Yes and No. Approval Voting is absolutely the simplest major voting system reform possible. It requires no changes to ballot forms; all it requires is eliminating the "no-overvote" rule, thus actually simpifying the rules versus now.
Original text: Range voting is more expressive than IRV.
Lorenzo: More expressive than weighing all the candidates you like against each other and determining what order you would prefer them elected? I don't agree. But again, this is one of those opinion things that can't be proven or disproven.
Our response: Range voting also permits expressing ordering, but it further permits expressing strength of preference, a concept Lorenzo seems blissfully ignorant of. E.g. consider an IRV vote A>B>C versus a range vote A=99, B=20, C=0 or a different range vote A=99, B=70, C=0. Range in this case is expressing something IRV cannot express. But, in the other direction, IRV never expresses anything range cannot. So yes, range is strictly more expressive. (Also: range permits "no opinion about candidate X" votes; IRV does not.) And no, this is not an "opinion," and yes it can be "proven" – we just did.
Lorenzo: We have research on exit polling of San Francisco voters who found IRV easy enough to use, the vast majority (over 70%).
Our response: Actually, in San Francisco, see http://www.rangevoting.org/SPRates.html, the voters in IRV races had a 7-times higher ballot-invalidating error rate than those in plurality races by the SF elections office's own figures. (Also, 70% finding something easy to use, is not exactly a strong endorsement... suppose a car manufacturer tried to sell a car for which "70% of the customers knew how to turn on the ignition"! Would that be a positive or a negative in the view of that car company?)
Lorenzo: IRV works without computer programming too. It was used in 1906 in Florida prior to electronic voting. Burlington, Vermont uses paper ballots and hand counts them.
Our response: Sure – but not on voting machines designed for plurality voting! Meanwhile: Range voting does work on them. (Of course, any voting method can be counted by hand, and we have never disputed that obvious fact.)
Also, Lorenzo was factually wrong: Burlington uses computer counting.
Lorenzo: Your assertion that IRV requires software is not accurate. State governments require software and approval of machines, not IRV. Get your facts straight.
Our response: We never made the assertion that "IRV requires software (or hardware, for that matter)"! Manually-counted ballots can handle every commonly-proposed voting method.
In practice, however counting IRV races can be tricky and getting software and hardware certified can be tricky. For example, San Francisco's IRV machines failed to achive certification by the state of CA and last we checked, they were going to be forced to abandon them. For another example, the official IRV rules circulated by the state of North Carolina to its election officials were incredibly complicated and the fact IRV cannot be counted using "precinct subtotals" leads to further immense complexity – discussion of the (further also horribly complicated) rules distributed by North Carolina about that is here.
Lorenzo: Second, any machine could have software developed very easily to conduct IRV elections. The idea it can't be done is totally ridiculous.
Our response: Actually, Lorenzo's sentence is false. First, many machines do not have computers in them. Therefore the claim that "any" machine could have software developed, is false. Second, even computerized machines, no matter how enormous their computer power, cannot count votes that they do not have. That matters because, with range voting, if you have 50 machines, each can count its own votes and then we just combine the 50 subtotals. In contrast, with IRV voting, there is no such thing as a subtotal; you cannot count each machine's votes individually. For example, with IRV, if Bush wins district#1 and Bush wins district #2, that does not imply that Bush wins the combined two-district vote set; Gore might be the IRV winner. That's an example of the fact IRV is a non-additive method. Range voting is additive.
The consequence of this is that the whole system the USA has used for generations – "vote counting in precincts" – no longer is possible with IRV, no matter how many computers and how much software you have. Many people regard that as a big disadvantage, although it is not necessarily a fatal one.
Example to make above claim clear: http://rangevoting.org/IrvNonAdd.html.
Original text: Adopting IRV will cause voter errors ("spoiled" ballots) to become 7 times more frequent (based on San Francisco numbers http://rangevoting.org/SPRates.html). But adopting range voting appears to decrease errors.
Lorenzo: Wrong. San Francisco limits voter rankings to 3 choices.
Our response: We nowhere said nor implied SF "did not limit voter rankings to 3 choices." Therefore, since we did not say that, we were not "wrong." Incidentally, "top 3 choices IRV" has serious further problems beyond those regular IRV suffers.
Lorenzo: We are not following that model in Florida, as it is not as good and has a high number of "exhausted" ballots (ballots where all the choices have been eliminated, and thus a vote can not be recorded in that round from that specific ballot. What do you base your assertion that Range Voting will cause less errors on? Please show some OBJECTIVE RESEARCH DATA to support this claim.
Our response: We already gave, and Lorenzo already stated himself (apparently he never followed the hyperlink) the URL to the objective research data (from the SF election office): http://rangevoting.org/SPRates.html.
Original text: Range voting is monotonic, i.e. increasing your vote for somebody can help but cannot hurt them. IRV is not monotonic.
Lorenzo: IRV has VERY LIMITTED EXAMPLES of where it fails the monotonicity criterion. What that means is that if you rank your most prefered choice highest, they are more likely to lose. This is IRV's main flaw, though the chances of it emerging and actually happening are so slim. It has NEVER happened in reality.
Our response: Examples of where "favorite betrayal" (voting your favorite top, was a strategically bad decision under IRV) has happened in reality include the 2003 Debian election, the Peru 2006 presidential election, the Chile 1970 presidential election, and the 1970 NY State Senatorial election (according to analyses on the CRV web site and in books cited there). There were also non-monotonocity examples in Louisiana governor elections such as 1991 and in Burlington 2009 (this last case is especially clear since we have the full list of all ballots) and Frome 2009.
Original text: In range voting, scoring your favorite candidate top cannot hurt either you or him. In IRV, it can hurt both. (Example, another, & another.)
Lorenzo: Didn't we just discuss this?
Our response: Actually, favorite betrayal is not exactly the same thing as non-monotonicity. But yes, we just did discuss it; and in Peru 2006, voters who foolishly ranked their honest favorite top, thereby hurt themselves. If Peru had been using range voting, they would not have been hurt by deciding to rank their favorite top.
Original text: With IRV the "Nader spoiler" and "wasted vote" problems are not solved, contrary to pro-IRV-propaganda. Indeed, because of fear of these very effects, IRV voters tend to rank third-party candidates below top (even if favorite) and hence prevent their election, which presumably is why every IRV country is and always has been 2-party-dominated. For this reason IRV cannot attract support from intelligent third-party members.
Lorenzo: Again, show some data that supports this assertion, as IRV is not used nationally or statewide anywhere currently. Please provide some data to support this claim.
Our response: IRV is used nationally and statewide in Australia (house members), Ireland (president), and was used in Fiji before the 2006 coup. And again, Lorenzo appears not to have followed the very hyperlinks he quotes! Extensive data concerning the 2-party domination of Australia, Fiji, Ireland in IRV seats, Malta, are available on the CRV website:
etc. Lorenzo should note that a search box is available on the CRV front page http://rangevoting.org. This often enables you to rapidly find such data rather than asking futily where it might be.
Lorenzo continues: In fact, IRV is immune to tactical voting in theory
Our response: Utterly false. We should emphasize this. Lorenzo is here flatly contradicting the most important theorem in all of voting theory, the Gibbard-Satterthwaite theorem. That is an illustrative example of the unfortunate ignorance (albeit expressed in tones of ringing certainty) that many IRV supporters have about the basics of their own subject. We must disparage the unfortunate tendency to push "reforms" first and think about their properties second (if ever). Many counterexamples to Lorenzo are in some of the URLs Lorenzo has already cited, for example
Original text: IRV makes ties and other nightmare-scenarios much more likely; Range voting makes them much less likely.
Lorenzo: Wrong. We have a built in mechanism in our language to do a pairwise comparison (Condorcet) between the two who tie to resolve ties. Ties are highly unlikely, but surely possible. I am not sure where you say IRV is more prone to ties. What are you basing that on?
Our response: Again, had Lorenzo bothered to follow the hyperlink he himself just quoted, http://rangevoting.org/TieRisk.html, he would have been led to extensive evidence from logic, actual elections, and computer simulations, all backing this claim up. Lorenzo's idea of resolving ties in IRV rounds by use of pairwise comparisons might be a good one, but it is a "red herring" since it does not affect our contention about the commonness of ties and near ties, in the slightest.
Original text: IRV is historically more likely than range voting merely to lead to a backslide to plurality voting.
Lorenzo: Again, what are you basing this on?
Our response: we're basing it on the large number of times in US history it has backslid, versus the fact no such backslide has ever occurred with range voting, and only once we are aware of with approval voting. A large number of US cities used to use STV and no longer do. [STV reduces to IRV in the single-winner case.] Again, there is something to be said for actually following the hyperlink you yourself quote.
Original text: IRV will (in plausible scenarios) elect candidate X in preference to candidate Y, even though based on the IRV votes, Y is pairwise-preferred over X (and over everybody else too) by an arbitrarily-huge supermajority of the voters. This appears to have happened in both the Peru 2006
Lorenzo: This is possible, but highly improbably and extremely unlikely to occur.
Our response: But it just did occur, as you yourself quoted, in Chile! (Also Burlington 2009.) Also, consult the pictures in http://rangevoting.org/IrvExtreme.html for an interesting graphical look at voting methods. Every single pixel in every single picture there represents an election with about 5000 voters. Every pixel in which the "IRV" picture has a different color than the "Voronoi diagram" picture above it, is an election is which IRV refused to elect a beats-all winner. As you can see, such pixels are quite frequent. It looks to me like the entire middle of the first four IRV pictures (14 candidates each) qualifies, which is about 1/4 of the pixels roughly and constitutes about 40000 different elections all of which are examples of this "extremely rare" phenomenon. Note that these pictures were constructed not at all adversarially by me – I just used 14 random candidate points. These points were not chosen in any way to make IRV look bad.
Similar pictures produced by a different person using different software are at this URL http://zesty.ca/voting/sim/ (he's a grad student at Berkeley, Ka-Ping Yee).
Finally, it may be mathematically proven as well as experimentally verified that in the random election scenario, IRV fails to elect Condorcet winners (even if we restrict attention to elections in which they exist) asymptotically 100% of the time if the number of candidates is made large.
Lorenzo continues: Arrow's theorem says no voting system is perfect, including Range Voting and IRV.
Our response: Actually Arrow's theorem does not apply to range voting, see http://rangevoting.org/ArrowThm.html (However, this is not to say we claim range voting is "perfect.")
Original text: Raising a candidate in your IRV vote from bottom to top-ranked can actually cause him to lose!
Lorenzo: Yes, you covered this in monotonicity criterion. It is POSSIBLE, but improbable to ever occur.
Our response: The probability of "favorite betrayal" scenario (where ranking your true favorite top, is not your best vote) is (if all 3-candidate N-voter elections are regarded as equally likely and N is large) 19.6%. The probability of non-monotonicity in 3-candidate IRV elections is 15% in the same model, see analysis here: http://www.rangevoting.org/Monotone.html; also computer simulations have yielded the same numbers in case you cannot follow or do not want to trust the theoretical math. (And these probabilities presumably rise if we consider elections with more than three candidates.) Thus favorite betrayal seems a more important problem for IRV than non-monotonicity. Is 19.6% "improbable"? Not terribly. Also, it can be (and is) argued there that it can effectively happen 100% of the time, in the sense that voters can 100% of the time be strategically wise to act as though they think it'll happen...
Original text: Contrary to pro-IRV-propaganda, pathological IRV elections seem unpleasantly common in practice. Two of the last five Debian (http://rangevoting.org/Debian2003.html) elections would have exhibited pathologies had they been held using IRV.
Lorenzo: There are 4 cities using IRV currently. I am not sure where you get that they are common practice, first and foremost, or where you see that IRV is a disease-causing system, as it has no ability to transmit diseases. I don't know what Debian means, or what this references.
Our response: "Pathological" does not mean "cause disease" in this usage. We use it the way mathematicians employ the word, i.e. "causes problems." (E.g. "the spoiler pathology.") Sorry for any confusion!
Debian: if Lorenzo had followed the URL he himself just gave, he would have been brought to a page about the Debian elections. These are elections "in practice," i.e. conducted annually by real humans with debates, hundreds of voters, etc. They are a useful dataset for elections researchers since they publish all ballots in a sensible format. (Also, google searching finds Debian and their elections in about 1 minute...)
Return to main page