Responses to attacks on our analysis of Burlington 2009 IRV mayoral election

Quick-access:   (skip to compressed summary)

ballot-validity rates;   "personal attacks";   "biased" author;   monotonicity-I;  
monotonicity-II;   Condorcet;   spoilers;   strategic voting;  
"no-show" pathology;   "later-no-harm (LNH)";   range-voting & LNH;   "mutual-majority";  
range-voting-I;   Tideman-book;   Bucklin;   range voting "unadoptable"
Ramabahama defends rules;   Responses to this page; Compressed summary

The election was held 3 March 2009. A press release immediately appeared from FairVote ("a non-profit, non-partisan organization that advocates for fair elections") signed by Rob Richie ("executive director") declaring "Burlington Holds Second Highly Successful Instant Runoff Voting Election." It was published verbatim by Huffington Post but unfortunately contained the false claim

The Burlington election was a model of clean, open debate without "spoiler" concerns.
which was, of course, also promptly regurgitated verbatim by, e.g. Krist Novoselic in Seattle Weekly, etc. On 6 March, Some Analysis of the 2009 Burlington IRV Election by "Terry Bouricius, a former city councilor and state legislator from Vermont, and senior policy analyst for FairVote" appeared on the FairVote blog, extolling it as having "gone off without a hitch" and lauding its wonderful success without mentioning a single one of its numerous pathologies. Our analysis demonstrating that this election had suffered numerous pathologies appeared on the internet on 14 March. It was also described in Burlington's 2009 Mayoral Election: Did IRV Fail The Voters? by Wes D. Hamilton in the 15 March Green Mountain Daily. That article over time received numerous "comments" by internet denizens, mainly an IRV advocate named Rama Schneider, who also goes by the name "Ramabahama." On 17 March, Response to Faulty Analysis of Burlington IRV Election appeared on the FairVote blog, allegedly demonstrating our analysis "faulty." On 23 March the same venue sported More on Warren Smith's and Anthony Gierzynski's flawed analysis. Still more criticism, mainly from the same players, came in the 13 March Vermont Daily Briefing, in FairVote web pages sometimes referred to by the attackers... and heaven knows where else.

As of the end of March, we feel it is time to try to respond to the attack-attempts. Depressingly, it appears almost all fall into the following categories (or some combination):

  1. State something false – or something which perhaps might be true, but it supported by zero evidence (and often seems dubious), as a "fact."
  2. Redefine some word according to an unusual private definition. Then say we'd misunderstood and it wasn't a problem at all. (Of course, on other occasions, when it suits their purposes, the same attackers employ the conventional definitions, apparently seeing no contradiction.)
  3. Try to create a smokescreen composed of red herrings.
  4. Attack some claim we did not make, acting as though we'd made it (or distort our findings/methods, then attack those distorted versions).

Also, the attackers try to flood with large numbers of these, thus making the task of refuting them all, become long. They hope nobody will want to read all the refutations. We've tried to help the reader by providing a quick access index (above). We now try to deal with many of the attacks on a case by case basis. (We provide handy ABCD codes so the reader can tell which game-plan that attack employs; A is most frequent.)

As usual, quotes by the FairVote head and "senior analysts" containing massively-false propaganda claims will be listed occasionally with green background, as a little humorous interlude, like this:

Choice voting actually works

Choice voting has been used in public elections for over 100 years in the US and around the world, and we have not uncovered even a single example where any of the theoretical defects have occurred. This includes tens of thousands of elections with tens of millions of voters.  If the theoretical problems with choice voting occurred even as frequently as 0.1% of the time, there would be many such examples, but there are none.
Choice voting allows voters to vote sincerely for the candidates they actually support, most voters end up being represented by their preferred candidates, and the outcome is broadly representative of the entire electorate.
– FairVote web page [downloaded 29 March 2009, dated July 2002 and claimed to have last been updated by its authors 5 March 2006].

At an earlier point in his career, Smith used to point these out to FairVote under the naive belief that they might be interested in correcting their errors. However, after it became clear they fully intended to leave such pages up for years (without ever correcting them no matter how much evidence was presented to the contrary – including elections FairVote itself used as examples – and no matter how absurd their claims were) he gave up.


1 Spoilage rate(A) Bouricius(FairVote blog): "[IRV] saw voters handle the system remarkably well, with 99.99% valid ballots." Bouricius (Vermont Daily Briefing): "99.99% of IRV ballots cast were valid." Rob Richie (title of blog post by him): "Instant runoff in Burlington: 99.99% valid ballots in mayoral race."

Response: We greet this claim with mixed emotions. On the one hand the March 2009 Burlington IRV election did display an encouragingly and gratifyingly low ballot-spoilage rate, contrary to IRV's usual historical performance: IRV countries have historically had higher spoilage rates than comparable non-IRV countries, and in elections in which both IRV and non-IRV races were present, the IRV races historically delivered dramatically higher spoilage rates. (Actually, the latter seems still to have been the case here: in the November 2008 Burlington election for US president, 9 out of 20957 ballots were "spoiled," and in the VT governor election 4 out of 20944, for a combined spoilage rate of 0.03% in these two plurality-voting races. Despite the far greater turnout in the Governor and US President contests than in the Mayor contest, this still was 6 times smaller spoilage rate than in the IRV contest reckoned correctly as we do below.)

On the other hand, it is depressing how IRV propagandists like Bouricius, even in an instance like this of IRV exceeding expectations, feel forced to exaggerate their case well beyond the truth. (Or perhaps it was an innocent arithmetic or typographical error, with the exact same error made several times in several different sentences by at least two FairVote authors in several different places. But considering Bouricius was a founder of an IRV-voting company that featured the official counts top center on its corporate home page it is then rather depressing and frightening that he could not produce correct numbers.)

To explain the rudiments of arithmetic: According to Burlington's final official count posted by their elections office and posted by Bouricius's own vote-counting company, "there were 4 invalid ballots" and "8980 valid ballots." We repeat, these are verbatim quotes from Bouricius's own voting company's web page. That'd be a 99.96% valid rate, not 99.99%. Except it is worse than that – in addition to wrong arithmetic, they also employed "creative accounting." Before counting of the "8980 valid ballots" even began, the counters placed 4 of them in the "exhausted pile" and ignored them for the entire election. Those 4 officially "valid" ballots each co-ranked two candidates top, one of them a write-in. Given that, we have some trouble seeing why those 4 ballots were officially described as "valid"; and since they were valid, we have trouble seeing why they were not counted. They would bring the valid rate down to 99.91%. Also (which the official counts failed to mention!) about a dozen voters gave only a 1st and 5th preference. Almost all of these were "Wright (1), Kiss (5)." These were counted as "Wright>Kiss>(all others)" which probably was not the intent of those voters. They probably intended "Wright>(all others)>Kiss." [Later note: Rob Richie contends in email to WDS: they "of course" were counted as "Wright>(all others equally)." But if so, that still was not the intent of the voters, indeed, it seems the least likely of these 3 interpretations to be what was intended. In any event, our original contention happens to be correct because in this election Richie's contention and our contention happen to be, under IRV counting rules, effectively equivalent.] If we regard "wrongly counted" ballots as "invalid" too, then the genuine validity rate was actually about 99.8%.

In short, Bouricius & Richie here exaggerated to make the problems appear to be a factor of 20 smaller than they were. Actually, more examination reveals more problems so the exaggeration was more like a factor of 100.


2 "Personal attacks"(AC) Bouricius/FairVote: "The analysis opens with a personal attack against me [Bouricius/FairVote] as an individual, which is rather odd."

Response. Actually, it is "rather odd" that simply quoting the group, and the individual, most associated with making IRV in Burlington happen, and then refuting those quotes with evidence, is taken as a "personal attack." (We suggest TGB's true intent here was to create a distracting "red herring.")

Here's another typical example of FairVote falsehoods/deception (the reader can judge for herself whether by pointing this out we are making a "personal attack"):

More false claims made by Rob Richie (head of FairVote); Feb. 2008 web post responding to Farhad Manjoo [R's spelling errors corrected]
Instant runoff voting has a long, tested history of use in all kinds of important elections – from president of Ireland (Mary Robinson won there after being second on the first count in 1990), mayor of London (Ken Livingstone won as an independent after Labour Party insiders denied him the nomination), house of representatives in Australia (where a minimum of 4 candidates ran in every district in November 2007 because no one has to worry about splitting the vote, and every race was won with a majority) and a growing number of US elections. See www.fairvote.org/irv for more.

Non-monotonicity has no real-world consequences – it never affects how people vote, and almost never happens anyway. The Condorcet winner issue is one where you have to balance priorities. Do we really want a candidate with 6% of first choices to defeat two candidates who split the other 94% of the vote? An IRV backer says no... Certainly I think we're on solid ground with most people who would agree.

We'll see if range voting ever gets adopted for anything.

Let's dissect this quote. (It's going to take a long time because so much is wrong with it. Skip to summary) First of all, while citing Mary Robinson's election in Ireland 1990 as a poster-boy example of IRV success and telling us how "non-monotonicity... almost never happens," Richie "forgets" to point out that that very election exhibited a hybrid of non-monotonicity and no-show failure! (This election is actually quite an indictment of IRV in the following sense: in the entirety of Irish history, this appears to have been the single IRV election in which IRV "made a difference" i.e. elected somebody who would not have won with plain plurality voting – and in that one case, IRV exhibited pathologies!) And while citing the 150 November 2007 Australian IRV elections as another IRV "success" poster-boy, Richie "forgets" to mention that at least 9 of those 150 suffered either a no-show or a thwarted-majority pathology. One shudders to imagine the IRV races not cited by FairVote as "great successes."

"IRV's flaws, if they existed, would have showed up in the decades of IRV elections in Australia and Ireland... But they don't." – FairVote head Rob Richie, letter published in The Nation June 2008.

Also, Ken Livingstone's election was not Instant Runoff Voting according to FairVote's own official definition of IRV: the London rules prescribed only 2 rounds, whereas FairVote ("How Instant Runoff Voting Works" page downloaded 25 March 2009) demands continuing until some candidate "receives a majority"; FairVote IRV rules demand votes be a rank-ordering of the candidates, but London only allowed 1 first and 1 second choice, nothing further permitted. So we don't exactly see why this election was part of the "long tested use" of IRV.

And Richie's claim that the reason for the large number of candidates in these Australian IRV races was that "[with IRV] no one has to worry about splitting the vote" was deceptive. First deception: as the present (Burlington 2009) election illustrates, vote-splitting-like phenomena still happen with IRV. Specifically, Wright's supporters preferred Montroll over Kiss by over a 3:1 ratio. Because they voted for Wright, IRV ignored their stated Montroll>Kiss preferences, causing Kiss to win. This same Burlington election also illustrated vote-splitting in plain-plurality voting: Montroll and Kiss supporters "split the vote" allowing the lose-to-all loser Wright to win. That is, the fact that the Kiss supporters preferred Montroll>Wright by over a 5:1 ratio was ignored by plain-plurality voting (and also by IRV).

IRV propagandists like Richie generally point out phenomena of the latter kind as an argument in favor of IRV and against plain-plurality. Unfortunately they simultaneously pretend (as they did here) that phenomena of the former kind (which make IRV look bad) do not exist! But they are both the same phenomenon, and indeed the IRV version is worse in a few senses:

  1. IRV solicited the Montroll>Kiss preferences of the Wright voters, giving them the deceptive impression that their opinion about Montroll-vs-Kiss mattered. IRV then ignored all those opinions. In contrast, with plain plurality voting, the Kiss voters were unable to express Montroll-vs-Wright preferences and hence there was no deception going on – they knew there was no way their opinions on that would be counted.
  2. IRV refused to count the dramatic Montroll>Kiss preferences of the Wright voters and also refused to count the dramatic Montroll>Wright preferences of the Kiss voters. However it was perfectly happy to count the (less dramatic) Wright-vs-Kiss preferences of the Montroll voters! That, rather arbitrarily, unfairly, and whimsically, gave some voters more say than others. In contrast, with plain plurality voting, all voters had the same say.
  3. IRV also can exhibit pathologies that plurality voting cannot, such as non-monotonicity (altering your vote to favor X makes X lose!) and no-show paradoxes (not voting can yield a better winner than if you honestly vote!) and crazy-reversal paradoxes (voting for the candidate you hate the most, can actually be a better strategic move than voting for the one you like the most!) – all of which occurred in the Burlington 2009 election.

Second deception: the true reason Australian "third parties" are strong enough to run candidates in all or most IRV races is not because IRV allows them to win – in fact in this very 2007 election cycle Richie cites, zero third-party members won IRV seats, same as in the preceding Australian IRV election cycle. The true reason instead is that Australia also has another body called the "senate" elected via a different voting system (proportional representation, multiwinner elections) which does allow third-parties to win a decent number of (senate) seats.

The claim about monotonicity "almost never happening" is of course false, refuted both by the very Burlington 2009 election here, as well as by mathematics which indicates it happens 5-15% of the time in 3-way IRV races and 100% of the time in N-way races when N becomes large. The claim that it "never affects how people vote" is remarkably strong. We might have been willing to accept "usually doesn't affect" but never? Ever? (All with zero evidence presented for it?) In any case, even if it really did never affect how anyone ever voted ever, then non-monotonicity still would remain a significant problem for IRV since it still leads to wrong IRV election outcomes even with 100%-honest votes. And this judgment of "wrongness" is not a subjective one. We know since IRV contradicts itself that at least half of IRV nonmonotone elections deliver wrong winners.

Richie's ploy about "the Condorcet winner issue" has some validity but is not relevant to Burlington 2009. Richie worries that a candidate with 6% of first choices might be a Condorcet ("beats all") winner, and in such a circumstance doubts the legitimacy of electing him. We agree that there can be grounds for questioning the legitimacy of "weak" Condorcet winners. However, in the Burlington 2009 election, Montroll was not a weak Condorcet winner, he was a strong one. Specifically, Montroll beat every rival and did so by pairwise margins in every case over twice as great as the margin (trumpeted by Richie, FairVote, and Bouricius) justifying Kiss's IRV victory over Wright. It is very rare that there are good grounds for questioning the legitimacy of the victory of a "strong" Condorcet winner such as Montroll in Burlington. (In fact, as of 2009 no such case has ever been brought to our attention in all of world history.)

Finally, we'll address Richie's final false sentence below – albeit it is even more false here since Richie uses the extremely strong word "anything."

Summary: When you look at this (a typical web post by Richie/FairVote) it is absolutely astonishing what a large percentage of it consists of lies, misleading half-truths, and attempts to trick the reader. If IRV is so wonderful, wouldn't the truth suffice? Couldn't its proponents achieve, say, at least 80% genuine-truth levels? I mean, even Adolf Hitler and Bernie Madoff probably managed that.


3 Bias/"unethical" nonDisclosures(C) FairVote/Bouricius's next red herring was to accuse professor Anthony Gierzynski of "anti-IRV-bias" because "in 2002 he... ran for state legislature as a Democrat against incumbent state legislator, the Progressive Party's Bob Kiss, who later went on to be the winner in Burlington's two IRV elections for mayor. Professor Gierzynski does not disclose this obvious source of potential bias..."

So the idea here is apparently that Gierzynski must have held a 7-year-long grudge against Kiss for having defeated him in a (non-IRV, indeed not even single-winner, indeed IRV had not even been introduced anywhere in Vermont at that time) race (the one time A.G. ran for office), and therefore became biased against IRV (instant runoff voting) in a mayoral race Gierzynski was uninvolved in 7 years later.

Oh.

Here's our suggestion to Bouricius about ethics: if you co-found a IRV-voting company and you received contracts from Burlington – then disclose that when posing as a an objective unaffiliated "voting reformer" from "FairVote" who has no skin in the game.

Oh, and while we are at it, Bouricius "forgot to disclose" that he was a member of the Vermont legislature as a "Progressive" and so was Kiss (both from North Burlington). Bouricius also was involved in a scandal when the Rutland Herald in May 1995 revealed he and other-Burlington-rep Dean Corren were collecting $1600/month in "overnight lodging expenses" so they could stay in Montpelier for their legislative duties. They'd "forgotten to disclose" that they had rented an apartment 1 block from the state house for only $340/month.

Which of these "ethical flaws" do you consider more serious?


4 Monotonicity(ABC) Bouricius/FairVote: The [analysis] incorrectly states that Burlington's IRV election suffered from a monotonicity failure... In fact, no such failure occurred. What his analysis actually shows is that non-monotonicity could have affected the election, but did not. If just over 25% of the supporters of Republican Wright had abandoned their true first choice... But this did not happen, and there is nobody who thinks it was a sensible strategy for any voters or candidates to advocate.

Response: Sadly, Bouricius does not understand the definition of "monotonicity" (and appears to be trying to invent his own private definition just for the present occasion). Indeed, Bouricius misunderstands the whole point of monotonicity. Monotonicity is not about one election, it is about two elections – one before, and one after, votes are altered in a manner favoring X. If those vote-alterations cause X to stop winning, then

  1. That is a monotonicity failure,
  2. We can be 100% certain that at least one of the two elections delivered the wrong winner.
Bouricius desperately plays "3-card Monte" to try to confuse the reader ("That second election didn't really happen!" – of course not) and attempts to throw up a smoke screen of irrelevant red herrings ("That vote-alteration would have seemed to be a dangerous strategy!" – so what? The nonmonotonicity definition does not even mention the words "strategy" or "danger"). This of course will not fool anybody who actually thinks about it – but we doubt Bouricius was focused on that class of readers.

Oddly enough, our definition happens to coincide remarkably precisely with the one on page 3 of Jonathan K. Hodge & Richard E. Klima: The Mathematics of Voting and Elections, American Math'l Society 2005, and also with those in the undergraduate textbook Peter Tannenbaum: Excursions in Modern Mathematics (Fifth Edition), Peason Education 2004, the Handbook of Social Choice and Welfare (Elsevier 2002), etc.

Interestingly, Bouricius actually links to a FairVote page giving the following definition of monotonicity, and then providing an example of a non-monotonic IRV election under FairVote's own definition, which actually does exactly the same thing as the Burlington 2009 election! FairVote had neglected to point out, at the time they wrote that definition and example, that (according to Bouricius) their own example election was (according to FairVote's 2009 claims) really not an example of nonmonotonicity. Oops.

Monotonicity can be defined as follows: A candidate X should not be harmed (i.e, change from being a winner to a loser) if X is raised on some ballots without changing the relative orders of the other candidates. – FairVote's own definition on one of their web pages, downloaded 23 March 2009.

[Note: this is exactly what failed in Burlington 2009 with X=Kiss.]
The very FairVote page Bouricius links to as his source continues by providing this example of a non-monotonic IRV election:

Andrea wins this IRV election:
#ballots 1stChoice 2ndChoice
  39      Andrea    Belinda
  35      Belinda   Cynthia
  26      Cynthia   Andrea
but after a hypothetical change of 10 Belinda votes to now raise Andrea from last to first rank, Cynthia would win.

Note, this example exactly parallels what happened in Burlington 2009 if we rename Andrea→Kiss, Belinda→Wright, Cynthia→Montroll. Specifically, Kiss won, but after a hypothetical change of 753 Wright votes to now raise Kiss to first rank, Montroll would win. (Hodge & Klima also give an example on p.49 also of the same sort.)

We thank Bouricius & FairVote for making our point for us, saving us the trouble of having to point out their errors ourselves.

An even-more-totally confused and false statement (contradicting his own definition in at least two different ways):
This election did not have a non-monotonic result. It could have had a nonmonotonic result, but it didn't – that is, Kiss in fact did win. The only way you have a nonmonotonic result here is if Wright, Montroll or Smith would have won if being raised on ballots in some new combination. – FairVote head Rob Richie, web post 16 March 2009.

5 Monotonicity(D) FairVote/Bouricius: "So it is mathematically impossible for a switch of first choices away from Wright to have made him a winner. Thus, despite the Smith-Gierzynsnki analysis, there was, in fact, no non-monotonicity event in the Burlington election."

Response: We never claimed that "a switch of first choices away from Wright [would] have made him a winner." (This is technique "D" from the above list – attack something we did not claim.) We claimed that a switch of first choices in favor of Kiss would have stopped Kiss from winning. This remarkably resembles FairVote's own example of nonmonotonicity (given above) which the attacker cited (!): a switch of first choices in favor of Andrea would have caused Andrea to stop winning with IRV.


6 Condorcet "beats-all" winners(D) FairVote/Bouricius: "In another point in [their] analysis, Smith and Gierzynski attack IRV for failing to elect the apparent compromise Condorcet-winner. This is disingenuous because Warren Smith himself dismisses the Condorcet-criterion, since his favored method also fails to meet this criterion – range voting can elect the Condorcet-loser."

Response: What actually is "disingenuous" is this attack. "Smith himself" did not "dismiss" the Condorcet-criterion – which states that if a candidate (such as Montroll) exists who defeats every rival pairwise, he should win – Smith said that it was "usually" good but not "always." Read what Smith actually had written (see especially its bottom paragraph).


7 Spoilers(D) FairVote/Bouricius: Smith and Gierzynski assert that the Burlington election suffered from a "spoiler" scenario. In order to make this claim they need to stretch the usual understanding of the term "spoiler" to fit the data. Normally, the term "spoiler" refers to a candidate who receives relatively few votes... I have never seen the term "spoiler" applied to a front-runner "spoiling" the chances of a candidate in third place before. [This stretch] suggests the term could also be stretched so that the actual winner, Bob Kiss "spoiled" the possible election of the third place finisher Andy Montroll, who would have won if Kiss hadn't run... The simple fact is that no candidate was considered a "spoiler" by [either] the media or the voters.

Response: We had given a link to our definition of spoiler:

A "spoiler" is a candidate S, such that, if you vote for S, that could cause both S and your second-favorite candidate Q both to lose, whereas if you had voted for Q (e.g. if S were to drop out of the race, or if you just voted dishonestly to strategically pretend Q was your true favorite), then Q would have won. Spoilers can exist in Plurality, IRV, Borda, and Condorcet voting but do not exist in Approval and Range voting.
Note that under this definition, Kiss was not a spoiler in this IRV election. Bouricius/FairVote here are using technique "D": accuse us of saying the opposite of what we said, then attack that.

There was, however, justification for them claiming that "spoiler" normally refers to a candidate with "little or no chance of winning" [Merriam-Webster dictionary] whereas, our definition does not involve S's winning chances (good or bad) at all. We consider our definition to be superior to Merriam-Webster's because it is mathematically acceptable (i.e, precisely defined, without need to make subjective judgments about the "poorness" of "winning chances" – only this type of definition is acceptable in a mathematical paper and in cases such as this, Webster dictionary definitions are not – e.g, to illustrate the problems, Wright's winning chances under IRV here were actually mathematically zero since he both lost the election and was defeated pairwise by both K and M) and also because it happens to coincide better with what matters to humanity. Specifically, it matters if voting for S causes both S and your second choice both to lose. That's a serious flaw. The fact that S might have seemed to have "good chances of winning" himself does not at all diminish how distressing this kind of pathology is. Indeed, if anything, it seems to us to make it more important.

The difficulty with claiming we invented our definition to "fit the data" is that this definition had been sitting there in our glossary for years prior to the Burlington 2009 election.

The statement that "neither the media nor the voters" considered Wright a spoiler is amazing in its gall. The media also did not mention (that we noticed) the non-monotonicity of this election. Or the fact Montroll beat all rivals by majority preference-counts. (Well, actually, the media did mention all of these things, e.g. in the Green Mountain Daily piece. But we're going along with the FairVote policy of only counting their media pieces.) Essentially, Bouricius/FairVote here are declaring "we managed by virtue of our propaganda campaign that everything about this election was hunky-dory, to stop the media we like from mentioning X. Therefore, that proves X did not happen!"

As a final remark, we note that although historically, those branded "spoilers" in the USA have generally been third-party candidates, Burlington Vermont is a rather exceptionally left-wing place since 64% of its voters were registered as Democrats and it has a long recent history of winners from the Progressive Party, including Kiss himself. (Obama beat McCain by a thumping 17252-to-3086 margin in Burlington 2008.) This caused the Republican Wright arguably to be the "third-party" candidate, while the Democrat Montroll arguably was the "centrist."

8 Strategic voting(A) FairVote/Bouricius: "strategic" calculations played no role in the Burlington [2009] election.

Response: The attacker here is making a very strong claim ("no" role?) but offering no evidence beyond his own vast inner certainty. (There is a saying "extraordinary claims require extraordinary proof." The attackers failed to meet that standard.) Meanwhile there is very strong evidence that IRV voters in Australia vote strategically, and indeed, that voter behavior is the reason Australia's IRV seats are massively 2-party dominated. That's why it's a mistake for US third parties to push for IRV – it will ensure their end! Quite a few have fallen into that trap. Perhaps they were seduced by false claims such as:

"IRV lets voters rank candidates in preference order. A voter's best strategy is to sincerely rank the candidates." – T.G.Bouricius, P.J.Macklin, R.Richie: Science Magazine 294 (May 2001) 303-306.

It would be great if this claim (offered, just like now, without any evidence or proof whatever) had been true, but unfortunately (as was shown by this very Burlington 2009 election in more than one way!) insincere rank ordering can be better IRV strategy than sincere ranking. This false claim was actually in direct contradiction to one of the most famous theorems about rank-order voting, the Gibbard-Satterthwaite theorem, a fact self-proclaimed "political scientist" Bouricius failed to mention. (This error was comparable, in the voting field, to a proclamation that "our perpetual motion machine will continue to provide energy forever" would have been in the field of physics – except that the perpetual motion machine claim would probably have been "backed up" by a genuine crummy machine or bad argument whereas B,M,&R just baldly contradicted one of the biggest theorems in voting with no argument or justification whatever.) This (apparently sole) attempt by these authors to enter the scientific literature was, of course, immediately shot down by outraged genuine political scientists by return mail to the editor of Science, but FairVote continues to cite this article up to the present day (2009), without mentioning this error.


9 "No-show paradox"(CD) FairVote/Bouricius: Another supposed paradox that Warren Smith and Anthony Gierzynski claim occurred in the Burlington IRV election is the "no-show" or participation paradox. This is the claim that some of Kurt Wright's voters would have had a better outcome [in their view] if some of them had simply stayed home and not voted at all... However, using the actual rankings on the ballots shows that this is not the case in the Burlington election... The authors "cooked" the data [to] assert that this paradox did occur. What they do is assign preferences to voters that the voters themselves did not indicate on their ballots. To be clear, they do not use the actual ballot data, but instead made-up ballot ranking data to manufacture the paradox they apparently so wish had occurred in the Burlington election.

Response: Our so-called "cooking" of the data (which we of course had fully explained up-front) consisted in assuming that those Wright voters who had voted for W without expressing a preference between M & K, actually either preferred K slightly over M or preferred M slightly over K (or some mixture), even though they did not say so in their votes.

There were 2008 W voters who did express a preference between M and K in their vote; they preferred M over K by over a 3-to-1 ratio. The problem is that there also were 1289 W voters who did not express any preference between M and K. The Burlington election rules (unlike most Australian IRV rules) allowed voters thus to "truncate" their ballots. Now of course, it is inconceivable that all these 1289 voters really absolutely did not care by even the tiniest jot who won among M or K. If one were asked to state how much K's election was worth to her in dollars, and stated $23.574321... (and so on for an infinite number of decimal places) it is of course ludicrous to assume that M's election would also have been worth exactly $23.574321... to her also (and agreeing exactly for ever and ever no matter how many decimal places they used).

The truth is those voters simply did not care enough to express a preference in their vote, or perhaps were ignorant on the matter of comparing Kiss versus Montroll (but, after gaining knowledge, would have preferred one or the other by some slight amount). It seems reasonable to assume that these 1289 W voters would, if they had been required as in Australia to state their full preferences, have behaved similarly to the other 2008 W-voters and thus favored M by a 3:1 ratio. Or one could instead postulate that these 1289 would instead have had coin-flips as preferences, preferring M and K equally often. Those two hypotheses both would be defensible. One even (if one wanted to take an indefensible and peculiar stance) could postulate that these 1289 would have been extremely unlike the other 2008 W-voters and actually would have preferred K by a 3:1 ratio! We do not care which of these 3 postulates you assume (or anything in between) – whichever one you pick, you find that there existed at least 753 W-voters who preferred M over K.

So we do not consider this 753-voter existence claim to be an outrageous stretch or a "cook." On the contrary, anybody asserting the contrary, is outrageously stretching.

So let us take it as agreed that 753 W-voters who preferred M>K exist.

In that case, those 753 voters were better off staying home and not voting – "no-show paradox." It would have made a candidate they preferred over Kiss win. That is simply, like 2+2=4, a fact.

And note it does not matter whether they expressed these preferences in their vote, nor how tiny some of those preferences might be, nor does it matter whether those voters would need to do further research before becoming sure of these preferences. Regardless of that, 2+2=4 still is true, i.e, those voters were better off staying home and not voting, thus meeting the definition of "no-show paradox." (In fact, meeting the exact wording given by the attacker in his quote above.) Further, even if the 1289 W voters actually truly did regard K & M as exactly equal, then the no-show paradox still happened in the weaker sense that those 753 voters as a group were better off staying home; each one would have got either a better or equal election result.

In short: if FairVote wishes to claim a no-show paradox did not occur then they are assuming that the "silent" Wright voters must have preferred Kiss over Montroll by about a 4:1 (or greater) ratio. It is that which is an outrageous stretch and cook of the data.

We'd classify this attack as a desperate attempt to create a red herring, i.e method C; the idea was to create fear, uncertainty, and doubt that we might be nasty data-cookers, but while keeping the actual details carefully concealed to prevent resolution of that uncertainty, and to prevent the realization that FairVote (or anybody denying the presence of the pathology) were the ones really cooking the data.


10 "Later-no-harm"(A) FairVote/Bouricius: I and many other experts feel the... "later-no-harm" [criterion is] far more important. Unlike IRV, these methods [Range, Approval, and Bucklin voting] violate "later-no-harm" and thus create a clear incentive... for "bullet voting" (ranking only a first choice).

Response: The "later-no-harm" criterion is a silly one left unmentioned in most books about voting methods because of its lack of importance. Meanwhile every book on voting methods mentions the "far less important" Condorcet criterion. "Later-no-harm" (LNH) demands that, if a voter ranks some candidate X as a less-favored choice (as opposed to not ranking X at all) that should not be able to hurt any of her more-favored choices. (IRV satisfies LNH.) The point of LNH is supposed to be that a voter will thus have no incentive to leave candidates unranked – it encourages voters to provide more information.

That sounds initially like a good idea, until you realize that:

  1. LNH is actually the wrong criterion for addressing that point! The right test is that a voter, by ranking X, should not be able to hurt herself (i.e. cause the winner of the election to, in her view, worsen). IRV does not satisfy that criterion and hence IRV voters unfortunately can be incentivized to fail to rank candidates (since ranking them would make the winner worse). Why should we care about IRV satisfying the fake test when it fails the correct test?
  2. And it should be noted that this fake test was carefully worded to ensure that IRV was almost the only nontrivial voting method satisfying it. There are many possible fake tests resembling, but not precisely the same as, the correct test. IRV propagandists generally choose this precise wrong choice of the test, from among all the possible wrong-wording choices, to make IRV look good.
  3. The whole point behind LNH anyhow is irrelevant and of no importance whatever if (as in Australia, the most experienced IRV-using country) the rules simply demand that voters must provide full rank-orderings.

Oops.

Here's another deceptive FairVote quote about LNH:

"For example, under Range and Approval voting, giving any support to a second choice may cause that voter's first choice to lose (violation of the Later-No-Harm Criterion), causing some voters to strategically truncate their true preferences. IRV complies with the Later-No-Harm Criterion, and is thus immune to such strategic calculation." – FairVote web page "IRV Myths and Facts (FAQ)" downloaded 25 March 2009.

FairVote actually (very carefully) phrased the second quote to be technically true, but maximally misleading. (It is kind of a work of art.) With IRV, truncating your preferences can be strategically better for you (i.e. result in a better winner in your view) than honestly providing a full rank ordering. Indeed, as we saw in this very Burlington 2009 election, full "truncation" – i.e. not voting at all – can be better for you than providing a full honest-ordering vote! The intent of this FairVote quote was to mislead readers into believing that since IRV satisfies the so-called "later-no-harm" condition, voters never have incentive to truncate their IRV vote. That's false. Indeed, the American Mathematical Society (perhaps fooled by FairVote propaganda on this point!?) used to hold its own elections using IRV and issued a statement on their ballots saying that a voter could not obtain strategic benefit by ranking fewer than all the candidates. They then were embarrassed by an article

S.J.Brams & P.C.Fishburn: The AMS nomination procedure is vulnerable to truncation of preferences, Notices of the American Mathematical Society 29 (1982) 136-138.
giving a simple election counterexample! Soon afterwards the AMS dropped Instant Runoff Voting and switched to Approval Voting.

With range voting, giving "support" (i.e. a score above 0) to your 2nd choice could indeed cause that 2nd choice to defeat your first choice. That's bad, and that's why range voting violates LNH. But on the other hand, it also could cause your 2nd choice to defeat your 3rd or 4th or 5th (etc) choices. That's good. In view of this, would you be incentivized to give your second choice a score above zero? Bouricius thinks you'd refuse, but actually the correct answer is "usually." You the (range)voter have to make that judgment on a case-by-case basis each time you vote.

Example:

  1. In the USA 2000 presidential election, suppose your honest preferences were "Nader>Gore>Browne>Bush." Would it have been strategic for you to "bullet vote," scoring Nader, and Nader alone, above zero? Or would it have been more strategic also to award Gore a positive score (maybe even the same score as Nader)? ...Clearly, the latter. The former kind of vote would have (and did) make both Gore & Nader lose; only 600 votes of the latter kind would have made Gore win.
  2. Suppose instead you preferred "Gore>Nader>Browne>Bush." Would you have refused to give Nader a positive score because of the horrible fear that Nader might defeat Gore?
  3. Suppose instead you preferred "Bush>Gore>Browne>Nader." Would you have, with IRV, given this vote (because you realized that, thanks to IRV's later-no-harm property, ranking Gore 2nd in this way would not hurt your favorite Bush) – or would you have "played it safe" and dishonestly voted Gore bottommost?
    In Australian IRV elections, it is fact that about 95% of IRV voters simultaneously rank one major party top and the other bottom or 2nd-bottom in a 7-party field, even though (based on seat-count) they are by far the two most popular parties in Australia. Indeed, the major parties themselves advise this dishonest strategy (even though they undoubtably have voting-methods experts advising them, computers, etc)!
Many studies have been conducted of real-world range voters. Did they usually refuse to score anybody besides their first choice? Well, no. That wasn't what happened. Indeed, in every range voting and approval voting study there, they scored other choices a lot more frequently than these Burlington 2009 IRV voters did!

So it would seem a clear known fact, based on absolutely massive real-world evidence, that LNH makes little or no difference to real-world IRV voter behavior, and also that with range voting (despite its violation of LNH) voters usually score lots of candidates.

In view of this real-world evidence undermining LNH's whole point, the Bouricius stance that "later-no-harm" is "far more important" is revealed to be completely ludicrous. It has very little importance either practically or theoretically. The whole later-no-harm criterion seems to have been invented specifically to make IRV look good (IRV is almost the only method satisfying it) and it's exceedingly rarely (or never) mentioned by authors who are just comparing non-IRV voting methods.


11 Mutual-Majority(A) FairVote/Bouricius: I and many other experts feel the "mutual-majority" and "later-no-harm" criteria are far more important than the Condorcet criterion, for example.

Response: We addressed "later-no-harm" last response, now we address FairVote's other "far more important" criterion, "mutual-majority" (MM). We can easily explain why Bouricius is wrong that MM is "far more important" than Condorcet by using the present Burlington 2009 election as an exhibit:

In view of that, IRV, which obeys the MM criterion, elected KISS. But any method obeying either the Condorcet or "Smith set" criteria would have elected MONTROLL because of his majority-defeat of every rival. Note that, although majorities favored Montroll over every opponent, they were not always the same majority set of voters. The voter-sets who did it may have differed from rival to rival, but for any particular rival X, in net, more voters preferred Montroll>X, than X>Montroll.

The MM criterion says that if a candidate (e.g. Montroll) is preferred over every rival not just by a voter-majority, but by the exact same voter-majority-set for each rival, then Montroll should win. Plainly, that is going to happen much less often. So this MM criterion is strictly weaker than the Condorcet criterion (and its set-version is strictly weaker than the Smith-set criterion). That makes MM less important, not "far more important." Indeed, it could be argued that it makes it so much less important, that it is almost silly.

Also, one has to ask (philosophically speaking) why it is that we should devalue majorities with diverse opinions, versus majorities composed of mindless cloned robots. Some would argue that consensus – agreement among diverse factions – is good and our voting system should not be artificially biased against it. In that sense obeying MM (but not Condorcet) arguably is undesirable. A probably-related effect is that IRV is strongly artificially biased against "centrists" (e.g. Montroll, here) and in favor of "extremists."

All that is presumably why most books on voting do not even mention MM, but all of them discuss the "far less important" Condorcet criterion.

Sanity check: Although the "political scientist" Bouricius and his unnamed "many experts" think MM and LNH "far more important" than Condorcet... strangely, a search of databases of scientific/academic literature finds something a little bit different.

Database#papers with "Condorcet" in titleWith "later," "no," and "harm"With "mutual" and "majority" "monoton" in title & topic "voting"
ISI "web of knowledge+science" 2750 [in title]1 [in title] but this 1 paper does not include mutual majority criterion about voting, instead concerns a group-psychological notion they call "mutual majority and minority influence." 16
AMS "math-sci-net" (Mathematical Reviews)112 0 [in title]0 [in title] 10
AMS "math-sci-net" again 11214 hits if "later" & "harm" allowed to appear anywhere, but none of the 14 papers are about voting.20 hits if "mutual" & "majority" allowed to appear anywhere, but only 4 voting-related and none of those involve "mutual" & "majority" as adjacent words. 91 if monoton either in review or title
Total3870 0107

A cynic would argue that the sole reason for FairVote & Bouricius's emphasis on MM is because it, like "later-no-harm," is one of the few criteria that IRV obeys while a lot of other voting systems do not. Also, Bouricius gains the benefit of sounding mysterious and impressively-expert-like by intoning "mutual-majority criterion" (whether he knows what he is talking about, or not).


12 Range Voting(A) FairVote/Bouricius: "Since Range Voting has never been used for any governmental elections it is hard to be certain how often various paradoxes and pathological outcomes would occur. But it certainly is subject to genuine spoiler scenarios..." (See also Richie on this.)

Response: Range voting has been used for many governmental elections, apparently highly successfully, dating back to the dawn of democracy. Indeed, it had already been used (for much longer than IRV has ever been used) even back at the time IRV got invented!

At least with our definition (above) of spoiler, range voting never exhibits a spoiler (unlike, as the Burlington election showed, IRV). That is, it is impossible in range voting, by giving S the maximum score, to cause both S and your 2nd choice Q both to lose. That's because the decision about how to score Q (or anybody else) is an independent decision, unaffected by how you score S; and raising their score can never hurt a candidate's winning chances with range voting (unlike IRV, as the Burlington election showed). Theorems like this enable us to be "certain" that range voting never exhibits certain kinds of pathologies.

Notice the contrasting meanings of the word "never." When mathematicians employ "never" it means never. When Bouricius and FairVote employ "never" it apparently means "sometimes" or "frequently" or "hugely frequently"; it is hard to predict which meaning they will have in mind at any particular time.

It is, however, certainly true that range voting can exhibit various kinds of pathologies. There are an infinite number of different voting systems and different possible pathologies that they might fail. The student can easily get lost in that thicket. The right way out of it is called Bayesian Regret. It is a systematic, objective way of comparing voting systems which effectively examines every possible pathology (of which there are an infinite number), weighting each by how serious and how frequent it is. When that measurement was done, Range Voting was found to be superior both to IRV, and to every other commonly-mentioned-in-books voting system – with either honest voters or strategic voters (or mixtures) under all 720 different possible combinations of modeling assumptions tried.


13 Tideman's book(A) FairVote/Bouricius: Collective Decisions and Voting by Prof. Nicolaus Tideman states (p.238) that Range voting is one of six voting methods that "have defects that are so serious as to disqualify them from consideration."

Response: We had reviewed Tideman's book and explained there why Tideman's "strategy vulnerability" score was measuring the wrong thing and yielded plainly-wrong results. FairVote here conveniently forgets to mention that Tideman also finds instant runoff voting to be "unsupportable" (this is on page 240 for elections in which it is "feasible" to construct a pairwise-table – which it evidently was in the Burlington 2009 election, since we constructed it). The difference is that Tideman's analysis showing Range Voting "unsupportable" was (as we demonstrated in the review) flawed, while his arguments showing IRV unsupportable were valid. As far as we can see (as of March 2009) they have not been disputed by FairVote or anybody else.

Here's a helpful hint. If the sole voting-methods book you cite to support your case, says the very voting method you are trying to push, is "unsupportable," that's a little tiny clue your argument might be a teeny tiny bit in trouble.


14 Bucklin, Approval, & Range voting(AD) FairVote/Bouricius: "In fact, Bucklin, Approval and Range voting quite possibly would have elected Kurt Wright (the Condorcet-loser among the top three), although this cannot be known for certain – it is true that Wright would have won if second choice rankings had been randomly reduced in half."

Response: First of all, we ran Bucklin on the exact same ballots input by IRV. It elected Montroll. Not Wright. We "know this for certain." (Of course, with Bucklin, the voters might have voted differently because of making different dishonest-strategic lying decisions than with IRV. But remember, FairVote itself has assured us that "strategic voting played no role" in this election, so they themselves think we don't have to worry about that. In any event: with the official ballots, Bucklin elects Montroll.)

Second, as our analysis (also) already said, there indeed is no way to know for certain who range and approval voting would have elected, because they do not employ rank-order ballots (while IRV and Bucklin do). Therefore the only way to guess what Range and Approval would have done is to somehow infer/guess what Burlington's Range/Approval ballots would have been. Nevertheless, we have a high degree of confidence they would have elected Montroll for two separate reasons. First, by theorem, range (and approval) voting always elects the honest-voter Condorcet winner (CW) provided enough of the approval/range voters are strategic and the CW would with honest voting have been either the top or second-top range-voting finisher. (In the "random elections model" in 3-way races, the chances a CW finishes top or second in range voting are 99%.) Second, in the present election, the IRV voters in a sense provided "approval threshold" data by ranking either "W only" (presumably meaning M,K not approved) or "W and M only" (presumably meaning K not approved) or "W>M>K" full ordering (presumably meaning either "W & M approved, K not" or "W approved but K & M not.") It does not matter which of those two "either" possibilities was intended – no matter which way all those choices were made, the votes were such that we always get Montroll as the approval-voting winner. That's a lot of room for variation in our assumptions – but no matter where we run within that room, Montroll always wins with the official Burlington ballots. So while we agree we "cannot be sure" Montroll would have been the approval-winner, it is the guess best-justified by the evidence. FairVote and Bouricius did not discuss this whole model and that theorem at all, acting as though we had no basis whatever for our claims, nor did they mention our analysis had already said we had no way to be sure. That's approach D – distort what we said, then attack that.


15 Range voting "unadoptable"(A) FairVote (web page "Single-winner Voting Method Comparison Chart" cited by Bouricius, downloaded 26 March 2009): "Most people (and certainly most elected officials) are unlikely to be willing to abandon the principle of majority rule in favor of Bayesian regret (the core concept of Range Voting). If 55% of voters prefer candidate A and 45% prefer candidate B in a two candidate race, Range Voting promotes the concept that perhaps B should win if the 45% feel very strongly about their choice, while the 55% are only lukewarm about their choice. This characteristic of Range Voting, the fact that it expressly rejects the notion of majority rule, means that it will never be adopted for government elections."

Response: Let's see how many bogus uses of the (ever-popular with FairVote) word "never" occur here. Range voting will "never" be adopted for government elections – except for the fact it already was and used for over 1000 years. Government elections in which a non-majority wins (or might) could "never" be adopted? Gee. We're surprised to learn that the US constitution (2/3 and 3/4 supermajorities required to amend) could never have been adopted. And the US senate's cloture-vote rule (60%). And US veto-override votes (2/3). And the USA's procedure for enacting laws (pass by 3-0 majority among house, senate, & president; if even one objects, not a law). And juries in US criminal cases (12-0 required). And the USA's original (and present) presidential-election methods. And golly, FairVote's own favored voting method IRV, as we just saw in Burlington 2009, refused to elect Montroll despite majorities favoring him over every rival.

As usual, FairVote offers absolutely no evidence for their "unadoptable" and "never" claims. They simply arise from their own immense inner certainty, which is so great as to surmount both the need for evidence, and the facts that directly contradict them.

(Oh, and we've never particularly advocated range voting for use in two-candidate elections, at all. Voting only gets interesting when there are more than 2 candidates.)

Incidentally, you can read about the Los Angeles 2001 mayoral election, which is the only election known to us (as of 2009) in which the range voting (perhaps) would have "overruled a voter majority." You may judge for yourself whether that was a good or bad thing, and compare with the behavior of IRV voting (which made the same 'overrule').

This FairVote web page is quite interesting. It presents a list of voting methods and properties which FairVote "believes are most important to U.S. voters," and then in a chart states that the satisfaction of each property by each voting method is either "low," "medium," "high," "yes," "no," or "yes/no." You may be interested in our own such chart, or Tideman's criteria.

Incidentally, some errors we spotted in the FairVote chart are (a) the Condorcet-loser criterion entry for Condorcet methods should have said "yes/no" since Condorcet's own "least reversal" variant fails it, and (b) range voting should have been yes for cloneproof. FairVote claims since range voting "can degrade to approval which can degrade to vote-for-one plurality, Range may not satisfy this criterion." They fail to mention that under the Burlington 2009 IRV rules they themselves proposed, vote-for-one is allowed, but oddly enough conclude IRV satisfies cloneproofness... but anyhow not only is their argument bogus, their conclusion is too. It is entirely irrelevant whether a range-voting election happens to be the same as a plurality election in the sense every voter scores exactly one candidate max and all rivals min – range is still cloneproof.

Anyhow, rather oddly, by the exact list of properties given by FairVote, a voting method which satisfies more of them to a greater degree than IRV (by FairVote's own reckoning) is Schulze beatpath Condorcet voting ("high" resistance to spoilers and strategic voting, satisfies majority and mutual majority, satisfies Condorcet-loser and -winner unlike IRV, cloneproof, and monotonic unlike IRV). Schulze is also recommended by Tideman in the very book Bouricius cited above which had branded IRV as "unsupportable."

Another method that satisfies all the criteria they list for IRV, plus also Condorcet-Winner, is WBS-IRV (but it fails "later no harm").

Golly. So why is FairVote pushing IRV and not Schulze beatpath (and not WBS-IRV) voting? (As of March 2009, Schulze is not even mentioned anywhere in the entire "Fair"Vote website!) The only two reasons we can see in their chart are

  1. Schulze fails later no harm.
  2. Schulze "prospects for US adoption" are "low" (just like range voting, but unlike IRV which has "high" prospects).

Oho. So why it is that Schulze is "unadoptable"? Supposedly because it is complicated and harder to count than IRV. Only trouble with that is that Schulze actually is easier to count than IRV in the sense it is "precinct-summable," that is, with Schulze, but not with IRV, each precinct can publish a concise "subtotal" and the full election results are deducible from them.


16 Ramabahama defends rules – "Ramabahama" (Rama Schneider) argued, rather repetitively, that Kiss's election was valid because he won according to the rules of the IRV system. Essentially, R would simply repeat the rules, then state "Kiss won." If anyone pointed out a logical pathology present in the election, he would say that was a "what if" scenario, but the actual scenario was: Kiss won.

Response: In some sense, Ramabahama is entirely right. (And we actually appreciate his argument rather more than most preceding ones since we think he was actually being honest about it, not trying to deceive!)

And if the election rules had instead been "we will kill a goat, and if the entrails end up pointing South, then Simpson wins" we daresay Ramabahama would have argued that this was a fully legitimate Simpson victory and repeated those rules. Any attempt by us to say that the election rules themselves were illogical would have been met (we presume) by simply claiming those were "what if" scenarios. The actual scenario was: the entrails pointed South, and those were the rules – so Simpson won!

But we think even a staunch defender of rules like Ramabahama must feel a little disquiet. IRV said "Kiss won." Using the exact same rules and same votes except that some Wright voters switched their vote to Kiss, Kiss would not have won (Montroll would have won). We daresay Rama#1, in election#1 would have defended Kiss's legitimacy; he was elected by the rules. And in in election#2, Rama#2 would have defended Montroll's legitimacy; he was elected by the rules.

We're just a little worried about what would happen if Rama#1 ever encountered Rama#2. We also are a little worried about what would have happened inside his mind if the crux Wright↔Kiss switching voter who made the winner change the "wrong way" had, in fact, been Rama Schneider.


Update – responses to the responses to the responses

After this page appeared, Richie & Bouricius sent us some emails about alleged errors herein. In most cases their arguments, even if correct, fail to help the case for IRV and make them look even more irresponsible. But they did make one good point:

Mea culpa: Bouricius pointed out that the company Election Solutions Inc. (ESI) he co-founded with Caleb Kleppner and Kenny Mostern, is not the same company as Voting Solutions Inc. (VSI) founded by Jim Lindsay, Bill Gram-Reefer, and Steve Willett; and neither is the same as Premier Election Solutions Inc. formerly known as Diebold Corp. This Burlington 2009 election was counted using VSI software, not ESI software, in spite of the fact that ESI featured the output of VSI's software (i.e. the Burlington vote count) front center on their company web page (downloaded 30 March 2009) while VSI's web page did not mention it. Also, ESI's subpage about Burlington 2009 claims that "Burlington is going to use Choice Plus Pro, a program for tallying IRV elections..." and then that software written by VSI is freely downloadable from the ESI web site even though that "software may not be copied or used for commercial purposes without their (VSI's) permission." Got all that? We have now corrected our misunderstandings of this.

Hypothetical Approval/Range victor: Richie says "I think the odds are high that [Montroll] would have lost with approval and, especially, with range." Why? He felt that the Wright & Kiss backers were "passionate" and hence would have "bullet voted" voted for their favorite alone, depriving Montroll of, e.g. intermediate range-scores, and thus preventing him from winning.

Of course, the official IRV voting system rules allowed voters to "bullet vote," and indeed many did so, but despite that

  1. Montroll defeated all rivals pairwise,
  2. The votes led to an approval-voting victory for Montroll under the assumption the IRV bullet voters continued to bullet-vote with Approval.

But supposedly with range or approval the voters would have acted differently, bullet voting more-often, under the influence of their "passion." (I have no idea why Richie feels Montroll would have "especially" lost with range voting.) He presented no evidence for this conjecture.

Recountification and more about invalid ballots: Concerning their exaggerated claim, which we'd refuted, of "99.99% valid" ballots, R and/or B point out:

To respond to this: When assessing the number of invalid ballots and errors in an election, we have to use the final official tallies. If the officials and/or voters were confused and imperfect, that is sad, but their confusion and imperfections count. If Richie and Bouricius feel that they would have been (or actually were) better counters than the officials, or better voters than the voters, that in no way affects our conclusions about the actual error/invalid rates in the official counts. The plain fact is, after doing whatever counting, re-counting, and re-re-counting they were ever going to do, they came out with final official tallies and data files, and according to those, the error/invalid rates were 20 times larger than Richie & Bouricius said.

Notice that these final official tallies posted online by Burlington are date-stamped 3 March 2009 at 8:15pm, i.e. before all the misleading posts from FairVote came out about "99.99% valid."

Thus FairVote's "defense" of IRV's supposed "success" in this election which (they'd stated before) "went off without a hitch" consists in now claiming, basically, that "The official tallies were wrong. The officials were confused and unable/unwilling to incorporate recount-based corrections into their data. We (B & R) would have been way smarter than them."!

Oh. Sorry, we didn't realize this was what Richie & Bouricius had meant by "went off without a hitch."

Looking a little deeper, one may ask: why, after partially completing a recount, were the Burlington officials unwilling/unable to use whatever data they'd collected during the partial recount, to update their official tallies? We don't know. (In fact, we don't even know that they did, or did not; we're just relaying Richie & Bouricius's stories.) But we do know this: IRV, unlike a lot of voting methods (range, approval, plurality, Borda, Schulze-beatpaths Condorcet...) is not additive. That is, with IRV there is no such thing as a "subtotal." With additive methods, if recounts in Kalamazoo Precinct produce some corrected data, then Kalamazoo Precinct can publish its new corrected subtotal, and then the central officials trivially can combine the precinct subtotals to get the (new corrected) overall total. With nonadditive methods like IRV, trying to incorporate corrections can be much trickier. That may have been why Burlington did not (if they did not) incorporate corrections from the partial recount. So there are two possibilities here:

  1. Burlington incorporated corrections from the partial recount into their official totals, in which case, FairVote's whole story about "99.99%" is complete garbage.
  2. Burlington did not incorporate them, in which case, this is an indictment of IRV because IRV made it too hard to incorporate corrections from partial recounts, thus getting much higher error-rates and problem-rates than they should have. With nonadditive methods like IRV, it is inherently a lot harder to do either full or partial recounts, causing recounting to be both more of a nightmare, and more of a risk of a nightmare.
  3. In view of this, we find it interesting that FairVote is currently trying to push IRV for use in Minnesota. As we write this, Minnesota is currently still, 5 months after the elections, in the midst of a giant legal battle about elections+recounts between senator-candidates Coleman and Franken. Neither has been seated in the senate (3 months after the rest of the US senate was seated) so the US is currently operating with only 99 senators. Now imagine how this nightmarish battle would have been amplified if the MN senate election had instead been conducted with a nonadditive method like IRV. If IRV made it too hard for even a town like Burlington to incorporate data from a partial recount of its ballots, well... note that MN is not a town, but rather a (quite large) state...

More on ballots and their invalidity: Bouricius (still-later response) now claims the 4 "invalid" ballots were not the same as the 4 "exhausted" ballots but were in fact blank, i.e. expressed no opinion in the mayoral race. That appears to be true. Juho Laatu (even-later response) has now summarized the problematic ballots. Laatu found 113 ballots "skipping" at least one rank, including 46 which skipped 2-or-more ranks; 6 ballots with tied-rankings; and 4 blank ballots. If all these are counted as "invalid" or at least "problematic" then the problem-rate would be (113+6+4)/8984=1.37%. In other words, the Bouricius/Richie/FairVote "99.99% valid" claim exaggerated to make the problems appear about a factor of 100 smaller than they were. Even so, we repeat that the performance of the Burlington voters and counters remains impressive.

Update – latest ploy from Rob Richie! After a period of apparent indecision, FairVote appears to have adopted a clever new strategy to use to respond to anybody on the internet who brings up this Burlington 2009 election:

"Come again? Montroll was in last place when the Burlington was reduced to three candidates. There in fact is no election method used in any public election in the world or any office that would have elected him."
– FairVote head Rob Richie, web post 5 June 2009, responding to our summarizing sentence "pretty much every voting mankind ever invented would elect Montroll."

Richie's gimmick here is that, while very many election methods have been invented by mankind, very few are currently in use by governments. The ratio is over 100. (One reason is that only very old and bad methods are presently used by governments.) Richie cleverly changes the question from (what our quote was about) "which voting methods have mankind invented?" to "which are currently in use in public governmental elections?" hoping the reader will not notice he changed the question. Then, voila, the new question has the answer Richie (in this case) wants – because the only two election methods (among those most commonly attributed to mankind) that don't elect Montroll happen to exactly coincide with the two bad ones most popular in governments today. Hence our original quote was "clearly" in error! But of course, actually, our whole point in making this Burlington analysis was precisely that the top-two-most popular election methods used in governments – plurality & IRV – both performed badly. Plus that ploy about "last place" is so cute; it makes us sound so clearly nuts on the surface, that there is no reason for the reader to actually examine the numbers to comprehend that actually, the IRV voting method is doing the stuff that is nutty. Richie's idea is to discourage anybody from examining the facts.

But in other cases, in which, say, Richie were unsatisfied with election methods in use by some government, he magically would reject this "argument."

(If we were on Mars dying from lack of oxygen, and somebody said "Let's get some oxygen tanks!", perhaps Richie then would say "What?! Nobody on this entire planet is presently breathing oxygen, so this suggestion is clearly ludicrous.") We have to congratulate Richie – he went beyond the IRV propagandists' usual bogus-argumentation techniques ABCD introducing a wholy-new bogus-logic technique. Perhaps it should be called "proof by government"? While we're inspired to see this great intellectual progress from Richie, one has to wonder (as usual) "why is it, that IRV propagandists have such difficulty making a case when restricted to actually logically-valid arguments?"

The man devotes such great effort to devising bogus, yet naively-impressive-sounding, arguments to promote a poor voting method (IRV), carefully polishing every word. We wish he'd devote a decent fraction of that effort to promoting a good voting method, and then he could actually use valid arguments, and actually encourage readers to examine the facts.

There is also another problem with this. Remember, Montroll would have defeated each and every rival in a head-to-head contest. Are there any head-to-head public elections for any office anywhere in the world? Well, yes, actually. They're quite common.

Finally, Bouricius states that he has "more important work to do, than refute all of your claims, but I assure readers that nearly all of them are simply false, or debatable at best" but he "stands by" all his old analysis, which he notes can be accessed at the following URLs:

http://www.fairvote.org/blog/2009/03/response-to-faulty-analysis-of-burlington-irv-election/
http://www.fairvote.org/blog/2009/03/more-on-warren-smiths-and-anthony-gierzynskis-flawed-analysis/
Many quotes from these are already present here, with their refutations. Meanwhile for his part Richie tells us he "did get very far into" reading all this since "the tone is so nasty." (Apparently Richie has a policy of only reading news he finds pleasant.)


Compressed Summary

  1. The IRV propagandists erroneously claimed invalid ballots were about 20× fewer than they really were.
  2. The IRV propagandist who made a ridiculous accusation of "bias" against Gierzinski failed to divulge his own far more serious biases and conflicts of interest, such as contracts, founding IRV company with financial interest in having IRV in Burlington, etc, instead posing as "non-partisan advocate for fair elections." (He also calls himself a "political scientist.")
  3. FairVote IRV propagandists pretended this election suffered no monotonicity failure by redefining monotonicity. They failed to note, however, that their own FairVote "monotonicity" web page's definition of the term agreed with our definition, and according to it there was a failure. Not only that, it was exactly analagous to the numerical hypothetical-failure-example that page had itself given. Oops.
  4. IRV propagandists pretended this election was free of "no-show failure" by pretending we'd "cooked the data," but actually it is they who, in order to deny this failure, need to cook the data tremendously.
  5. IRV propagandists falsely claimed the Condorcet failure did not matter since we ourselves had "dismissed" Condorcet. Well, no.
  6. They then tried for an aura of expertise by making the easily-refuted claims the obscure "later-no-harm" and "mutual-majority" criteria were "far more important."
  7. They then made the ludicrously-false claims range voting had "never been used for any governmental elections" nor indeed for "anything."
  8. The IRV propagandists cited only one voting-methods book to support their arguments – by Tideman. Unfortunately they forgot to point out that 2 pages further on in Tideman's book, he concludes it was "unsupportable" to use IRV in Burlington. Oops.
  9. The IRV propagandists erroneously claimed there was no way to be certain whom Bucklin voting would have elected. (It would have elected Montroll.) Then they pretended we had no basis for our claim Approval and Range voting would probably have elected Montroll.
  10. We've often wondered why IRV propagandists have often contended (always presenting no evidence) that they could just magically tell that range voting was "unadoptable" hence should not be supported in spite of its excellent performance. We found two actual reasons on a FairVote web page referred to by one of the attackers! – but unfortunately they both were factually incorrect statements.

Return to main page