Worse than The Worst IRV Nightmare Election In Existence


#1

305 voters, 5 candidates.

1 A
49 ABCDE
26 BACDE
26 BCADE
51 CBDAE
49 CDBEA
26 DCEBA
26 DECBA
50 EDCBA
1 E

Counts: A = 50, B = 52, C = 100, D = 52, E = 51. A is eliminated by a margin of one vote. B gains 49 votes. B = 101, C = 100, D = 52, E = 51.
Then E is eliminated by a margin of one vote. 50 votes transfer to D.
B = 101, C = 100, D = 102. C is eliminated by the margin of one vote.
B = 152, D = 151. B wins by one vote.
In EVERY ROUND, the election was within one vote. That means expensive recounts and lawsuits.

D is a spoiler. D drew away votes away from the centrist C. If two members of the DE party had instead artificially ranked C top, then D would get eliminated and then C would win over B by about 200-100. This tactical voting would have given the two voters a better outcome.

C was preferred over every candidate by about a 2:1 ratio. And the D party, which has been branded as a SPOILER, may drop out. Meanwhile A and E were too extreme to have any chance, so we end up with two party domination between B (and maybe A/E, until they get too greedy and become a spoiler again) and C.

The D voters were tricked in another way. The second place votes of the DCEAB voters were not counted because their first vote was still on D at the critical moment that eliminated C. IRV ignored their second place votes while giving notice to other voters’ second choice candidates.

The election is equivalent to this:
1 A
49 AB
52 B
51 CB
49 CD
52 D
50 ED
1 E
IRV never examined more than two layers in. Annoyingly, C was even ranked first or second by a majority of voters, but still lost.

Yep, just about every flaw that real world voters care about, happened.


#2

Your example implies that Condorcet would be better than IRV in this case.

I agree.


#3

I mean, C has by far the most support and is ranked third or above by every voter. If you think of AB as the leftists and DE as the rightists and C as the centrist OF COURSE C should win.

Let me make one quick change to make this even more nasty…


#4

This is what your example looks like on my webpage.
candidates:A;B;C;D;E
b 1 1
b 49 1 2 3 4 5
b 26 2 1 3 4 5
b 26 2 3 1 4 5
b 51 3 2 4 1 5
b 49 3 4 2 5 1
b 26 4 3 5 2 1
b 26 4 5 3 2 1
b 50 5 4 3 2 1
b 1 5

My page can do Condorcet (click radio button) but not at the same time it runs the election.


#5

Hmmm… now I shifted my hate to this election:
42 Bush>Gore>Nader

7 Gore>Bush>Nader
3 Gore
20 Gore>Nader>Bush
(24 total)

28 Nader>Gore>Bush

(Gore falls and Bush wins 49-48 over Nader)

The reason is that it knocks down all four of IRV’s main selling points.
It does not elect a majority winner, because (1) Bush won with 49 of 100 votes and (2) Gore was preferred over Bush (and Nader) by a majority.

It does not always transfer your vote to your second choice, because sometimes your second choice is eliminated first (and your vote has no impact on that decision!!) and then your first choice transfers to a bad candidate. The Nader votes were stuck on Nader so Gore had to drop, and then Nader lost to Bush. Meanwhile, those 7 G>B>N centrists conveniently had their second place votes counted…

It does not prevent spoilers or make honest voting the correct strategy. If Nader had not run, or at least 2 Nader voters dishonestly put Gore first, then Bush would NOT have won and the lesser evil Gore would win!