The second election held using IRV produced the foulup documented in this page, a year later they repealed IRV.

IRV Failure In The Real World

Well, it has happened. That thing that a few math nerd election theory wankers warned about but nay-sayers said could never happen has happened. Instant Runoff Voting has elected the wrong person.

In the 2009 March 3 Burlington Vermont election for Mayor an IRV election was held and you can see the results on their site. Burlington is very nice in that they publish data of all the votes that I can then feed into my own analysis software.

Here's a histogram of how people voted on the candidates. (Note, I've turned rankings (1st, 2nd, 3rd) around into ratings (higher values better) because that's the native data for my software. "5" == "1st", "4" == "2nd", ...)

Andy Montroll
Rating	Votes
5	2062
4	2630
3	1398
2	497
1	119
votes	6706
average	3.897

Kurt Wright
Rating	Votes
5	2949
4	996
3	712
2	705
1	728
votes	6090
average	3.777

Bob Kiss
Rating	Votes
5	2586
4	1394
3	948
2	717
1	540
votes	6185
average	3.771

Dan Smith
Rating	Votes
5	1305
4	2102
3	1803
2	752
1	132
votes	6094
average	3.606

James Simpson
Rating	Votes
5	35
4	303
3	655
2	1041
1	1357
votes	3391
average	2.003

Write-in
Rating	Votes
5	38
4	46
3	42
2	44
1	73
votes	243
average	2.720

When I look at this table I see that Andy Montroll has pretty good 1st-choice support and even broader 2nd-choice support. Kurt Wright and Bob Kiss however are the favorite of substantial but narrow populations with less support at lower levels. However, because IRV only looks at the highest-ranked slot on a ballot, it misses this and produces the following sequence:

Round 1		Round 2		Round 3		Round 4		Round 5
Name	Count	Name	Count	Name	Count	Name	Count	Name	Count
Bob Kiss	2585.5	Bob Kiss	2599.5	Bob Kiss	2606	Bob Kiss	2982	Bob Kiss	4314
Kurt Wright	2952.5	Kurt Wright	2956.5	Kurt Wright	2963	Kurt Wright	3297	Kurt Wright	4064
Andy Montroll	2063	Andy Montroll	2067	Andy Montroll	2080	Andy Montroll	2554	Andy Montroll	2554
Dan Smith	1306	Dan Smith	1315	Dan Smith	1317	Dan Smith	1317	Dan Smith	1317
Write-in	38	Write-in	39	Write-in	39	Write-in	39	Write-in	39
James Simpson	35	James Simpson	35	James Simpson	35	James Simpson	35	James Simpson	35

That's a few votes off (less than 5, usually) compared to the copy on the Burlington website, but close enough and shows the same result. My software runs more rounds instead of dropping many low count choices in one round as their software does.

Here's what a virtual round robin election (Condorcet's method) looks like:

		1	2	3	4	5	6
1	Andy Montroll		4067	4597	4573	6267	6658
2	Bob Kiss	3477		4314	3946	5517	6149
3	Kurt Wright	3668	4064		3975	5274	6063
4	Dan Smith	2998	3577	3793		5573	6057
5	James Simpson	591	845	1309	721		3338
6	Write-in	104	116	163	117	165

Andy Montroll was preferred over Bob Kiss by 4067 voters. 3477 voters had the reverse preference.

Andy Montroll was preferred over Kurt Wright by 4597 voters. 3668 voters had the reverse preference.

Andy Montroll was preferred over Dan Smith by 4573 voters. 2998 voters had the reverse preference.

Andy Montroll was preferred over James Simpson by 6267 voters. 591 voters had the reverse preference.

Andy Montroll was preferred over Write-in by 6658 voters. 104 voters had the reverse preference.

That first line below the table really says it all. In a head to head election, Andy Montroll should have beaten Bob Kiss by a 7.8% margin. A solid win.

This is an IRV failure. The IRV result is clearly not what people actually wanted. More people liked Montroll over Kiss than the other way around, but IRV elected the loser.

Nail this coffin shut, don't implement IRV anywhere else when there are obviously better methods.

See also my standing page on IRV's shortcomings: http://bolson.org/voting/irv/

Because it's my pet system, here's the Instant Runoff Normalized Ratings result:

Round 1		Round 2		Round 3		Round 4		Round 5
Name	Count	Name	Count	Name	Count	Name	Count	Name	Count
Andy Montroll	3739.74	Andy Montroll	3749.16	Andy Montroll	3826.04	Andy Montroll	4332.18	Andy Montroll	5435.69
Kurt Wright	3499.25	Kurt Wright	3507.25	Kurt Wright	3562.94	Kurt Wright	3970.06	Kurt Wright	4405.94
Bob Kiss	3380.66	Bob Kiss	3388.07	Bob Kiss	3466.98	Bob Kiss	3883.05	Bob Kiss	3883.05
Dan Smith	3116.61	Dan Smith	3122.40	Dan Smith	3191.96	Dan Smith	3191.96	Dan Smith	3191.96
James Simpson	927.82	James Simpson	930.72	James Simpson	930.72	James Simpson	930.72	James Simpson	930.72
Write-in	96.51	Write-in	96.51	Write-in	96.51	Write-in	96.51	Write-in	96.51