Review of Extended Runoff Voting

Jarrett Gabriel recently contacted me about an article he posted in his new blog. The article introduces a new Election Method he calls Extended Runoff Voting. You can go and read it here:
http://jarrettgabriel.wordpress.com/2011/09/23/hello-world/ ("The Case for Extended Runoff Voting")

IRV's deceptive simplicity

The article starts by properly summarizing Instant Runoff Voting's (IRV's) main flaws: IRV does not remedy at all to the spoiler effect, which is the current Plurality Voting's most important flaw.

To this day, I still have in mind one of the Third Party Candidate debate during the 2004 US Presidential election campaign. During this debate, both David Cobb, the Green Party candidate, and Michael Badnarik, the Libertarian Party candidate, were advocating IRV. I remember how they described the IRV vote counting process, being "as simple as 1,2,3":

  1. Sort the ballots into piles according to first preference.
  2. Eliminate the lowest pile (implying that we don't even need to count the number of ballots in the pile!!)
  3. Redistribute the ballots from the eliminated pile into the remaining piles according to 2nd or 3rd preference. The candidate represented by the last remaining pile wins.

As we know, the simplicity is deceptive and comes at a cost. I was really sorry to see some 3rd party candidates being deceived and deceiving the public. That's the kind of gimmick we'd expect from candidates from the two-party duopoly.

Jarrett properly highlights the fact that the cause of IRV's failure is due to voters' preferences being largely discarded during the vote counting process. The method he proposes, Extended Runoff Voting, is an ingenious way to reintroduce he voters' preferences where they matter the most.

Ranked Voting methods

The two biggest families of Election Methods are weighted methods and ranked methods. Among the ranked election methods, we have Condorcet, IRV and now Extended Runoff Voting (ERV). IRV has already been thoroughly discredited as a viable alternative by the Election Method community of experts.

Jarrett Grabriel does not stop at the 1,2,3 process described above. I will not reiterate here the process by which second and third level preferences are considered before definitely eliminating a candidate. See the original post.

He offers the following example of the Canadian members of parliament selecting the Prime Minister:

Ballots of Conservative MPs:    Ballots of Liberal MPs:    Ballots of NDP MPs:                              
---------------------------     ----------------------     ------------------
Harper  (Conservative)    1     Rae                  1     Layton           1                                            
Rae (Liberal)             2     Harper               2     May              2                                              
Layton (NDP)              3     Layton               3     Rae              3
May (Green)               4     May                  4     Harper           4
Bloc leader  (Bloc)       5     Bloc leader          5     Bloc leader      5


Ballots of Green MPs:           Ballots of Bloc MPs:       Party Standings:
--------------------------      ----------------------     -------------------------
May                      1      Bloc leader          1     Conservatives   120 seats
Layton                   2      Layton               2     NDP              92 seats                                               
Rae                      3      May                  3     Liberal          56 seats
Harper                   4      Rae                  4     Green            22 seats
Bloc leader              5      Harper               5     Bloc             18 seats

According to ERV, Ray (Liberal) wins the election. What he fails to notice, however, is that Ray is the Condorcet Candidate, i.e. the candidate who wins individually against each of his opponents.

Rae (Liberal) vs. Harper (Conservative): 188 / 120. Rae wins.
Rae (Liberal) vs. Layton (NDP): 176 / 132. Rae wins.
Rae (Liberal) vs. May (Green): 176 / 132. Rae wins.
Rae (Liberal) vs. Bloc leader (Bloc): 290 / 18. Rae wins.

Similarly, in the 2000 US Presidential Election, Al Gore is the Condorcet winner.

Basically, what has been achieved here, is that Jarrett has found an ingenious way of calculating the Condorcet winner of an election without having to calculate the whole Condorcet matrix. I have not verified this mathematically, so if anyone can disprove this statement, let me know ;)

It ensues that ERV has all the qualities of a regular Condorcet election. Indeed, they are one and the same. And Condorcet is the best Ranking EM there is. It also means that ERV shares the same flaws as a Condorcet election.

The main flaw of ERV as presented here, is that it does not address the problem of a circular dependency, i.e. an election where A wins against B, B wins against C and C wins against A (i.e. A > B; B > C; C > A).

Condorcet experts offer many variants to resolve such circular dependencies. The problems is that experts do not agree on which variant is the best, so how can we expect the regular voter to understand the differences and the implications of those variants?

Also, both ERV and Condorcet (and IRV, for that matter) fail the Paper and Pencil Criterion. Such Election Methods are difficult enough to tabulate to increase the demand for electronic voting machines, which are a danger to an open democracy.

Weighted voting methods

The best of the Ranked voting methods, Condorcet of which ERV is an alias, being in my view discarded as a viable alternative to the current system, we are left with the family of Weighted voting methods.

In this family, there are several variants, such as: Approval voting, Emocracy and Score voting (Range voting). Weighted voting methods a simple methods, but unlike IRV they are not deceptively so. They are simple to understand even by a low IQ voter. They are easy to tabulate: a fourth grader could calculate the winner of such an election. And they give good results, in terms of fairness.

You can try one of the existing polls. Here is a sample:

  1. Approval voting: What is your take on Global warming?
  2. Score voting about the penitentiary system.
  3. Score voting about the Electoral College
  4. Emocracy voting: What issue will the most influence your vote for the 2008 elections? (the 2012 version is coming soon)
  5. Emocracy voting: 2008 U.S. Presidential Poll (the 2012 version is coming soon)
  6. Approval voting: How would you best describe Sarah Palin?

You can also create your own poll.

Among the various Weighted voting variants, my favourite one is Emocracy (although the other variants might be more appropriate for some of the opinion polls on social topics, like some of the above) because it is simple; it allows for negative scores as well as positive ones. More importantly, Emocracy allows the pollsters to gauge the level of notability of the candidates, something that the other variants do not offer. There is an important, much too often neglected, difference between voting down a candidate because you don't like him/her or because you don't know enough about him/her. In the second case, more media coverage would be appropriate, to allow voters to form their opinion.

Comments

Extended Runoff Voting

I appreciate your review and intend to respond to some of your criticisms. First, I was wondering if you could provide an example of an ERV election which would result in a circular dependancy, so that I would be better able to come up with a solution. Regarding other kinds of Condorcet methods, I noticed that at least with the simplest system called 'minimax,' that the results could be converted into a table where the winning candidate would get the highest number of points, the second ranked candidate would get a lower point total, and so on right down the list. As a very simple example, we could take a one vote election between Bush, Gore and Nader. Bush is ranked 1, Gore 2, Nader 3. Since Bush is ranked ahead of Gore and Nader, and is defeated by no one, he gets a plus 2. Since Gore defeats Nader but loses to Bush, he gets a 0. Since Nader is defeated by both Bush and Gore, he gets a -2. If there was a fifty candidate race, and one voter casted their first place vote for Bush, he would get a plus 49 since that is the total number of candidates he would have defeated, the second ranked candidate would get a plus 47 (48 wins to 1 defeat), a 45 to the third ranked candidate, and so on until the last ranked candidate who gets a -49. The problem is that this amounts to a point based system, which has the same flaws of a Borda Count. The same problem arose with another Condorcet method I examined and so I was hoping that the method I proposed would overcome this. If the Condorcet winner produces the same result as a point based system, then I don't see it being superior at all, just more complicated. I understand that you have some familiarity with a number of different Condorcet systems. Do most Condorcet methods amount to a point based system, and are there any Condorcet methods that lead to different results?

Unfortunately, the systems you are proposing, Score and Approval Voting, are a flawed alternative. Suppose we have an election between Gore, Bush and Nader. I don't want Bush to win, so I give Gore five points. I have a much stronger preference for Nader, so I also give him five points. Gore is essentially, the lesser of two evils, while Nader is the person I want to win. The problem is that Nader can not advance on Gore, unless we give Gore fewer than five points. However, if I don't give five points to Gore, we decrease the likelihood of him defeating Bush. This is the problem with Score voting and other related systems. The only way Score Voting would be acceptable is if it is the least worst of the alternatives, which could still be the case if ERV doesn't lead to fair elections.