Why Ranked Choice Voting Isn’t Great

Link post

Sev­eral friends are col­lect­ing sig­na­tures to put In­stant-runoff Vot­ing, branded as Ranked Choice Vot­ing, on the bal­lot in Mas­sachusetts ( Bal­lot­pe­dia, full text). I’m glad that an at­tempt to try a differ­ent vot­ing method is get­ting trac­tion, but I’m frus­trated that they’ve cho­sen IRV. While ev­ery vot­ing method has down­sides, IRV is sub­stan­tially worse than some other de­cent op­tions.

Imag­ine that some­how the 2016 pres­i­den­tial elec­tion had been be­tween Trump, Clin­ton, and Ka­sich, and prefer­ences had looked like:

  • 35% of peo­ple: Trump, Ka­sich, Clinton

  • 14% of peo­ple: Ka­sich, Trump, Clinton

  • 17% of peo­ple: Ka­sich, Clin­ton, Trump

  • 34% of peo­ple: Clin­ton, Ka­sich, Trump

(This is very hy­po­thet­i­cal; Ka­sich wasn’t this pop­u­lar.)

In a tra­di­tional elec­tion we’d have 35% for Trump, 34% for Clin­ton, 31% for Ka­sich, and de­clare Trump the win­ner. Un­der IRV we would elimi­nate the last-place can­di­date, Ka­sich, and di­vide their votes based on who was put down in sec­ond place:

  • 49% of peo­ple: Trump, Clinton

  • 51% of peo­ple: Clin­ton, Trump

Since Clin­ton has passed 50% we de­clare her the win­ner.

But let’s take a look at how peo­ple would have voted in one-on-one elec­tions:

  • Trump vs Ka­sich: Ka­sich wins 65% to 35%

  • Clin­ton vs Ka­sich: Ka­sich wins 66% to 34%

  • Trump vs Clin­ton: Clin­ton wins 51% to 49%

About 23 of peo­ple pre­fer Ka­sich to any other can­di­date on offer, but he doesn’t win with a tra­di­tional elec­tion, and he also doesn’t win with IRV. This elec­tion has a ” Con­dorcet win­ner”, some­one who would win a one-on-one elec­tion against ev­ery other can­di­date, and IRV isn’t de­signed in a way that it will always choose that per­son. [1]

Ad­di­tion­ally, some pro­po­nents of IRV will say that vot­ing your true prefer­ences will never hurt you: that un­der IRV you never need to hold your nose and put some­one first when re­ally you pre­fer some­one else. But there are cases where you can get an out­come you like more by putting a higher rank­ing on a can­di­date you like less. Con­sider an elec­tion like:

  • 32% of peo­ple: Clin­ton, San­ders, Trump

  • 33% of peo­ple: San­ders, Trump, Clinton

  • 35% of peo­ple: Trump, Clin­ton, Sanders

Now, this is a ter­rible elec­tion and there’s no clear win­ner be­cause peo­ple’s prefer­ences point around in a cir­cle. Un­der tra­di­tional vot­ing Trump wins, while un­der IRV Clin­ton is elimi­nated and San­ders wins. But if a few peo­ple prefer­ring Trump > Clin­ton > San­ders had in­stead voted just Clin­ton > Trump > San­ders we could have had:

  • 32% of peo­ple: Clin­ton, San­ders, Trump

  • 33% of peo­ple: San­ders, Trump, Clinton

  • 32% of peo­ple: Trump, Clin­ton, Sanders

  • 3% of peo­ple: Clin­ton, Trump, Sanders

Then Trump would have been elimi­nated first, with only 32% of the first place votes, and his sec­ond place votes would have gone to Clin­ton, mak­ing her win. So by vot­ing for a can­di­date they liked less, these vot­ers got an out­come they liked more. Stay­ing home and not vot­ing would similarly have helped them.

I do want to be clear: no vot­ing sys­tem han­dles situ­a­tions like this well, and all rea­son­able vot­ing sys­tems have situ­a­tions where you can come out ahead by vot­ing in a way that doesn’t re­flect your true prefer­ences. But “IRV means you can always vote your heart” isn’t true.

So what do I think we should do? Since ev­ery vot­ing sys­tem has down­sides it’s not enough to say that a vot­ing sys­tem is flawed and so we shouldn’t use it. But at the same time some vot­ing sys­tems are still bet­ter than oth­ers.

For ex­am­ple, a clear im­prove­ment over IRV would be to say that if at any point there’s a can­di­date who would win a one-on-one elec­tion against ev­ery other can­di­date (a Con­dorcet win­ner) then that can­di­date should win. While it does make IRV a lit­tle bit more com­pli­cated, it means IRV will no longer elimi­nate can­di­dates who are the clos­est there is to a con­sen­sus choice. This is called Con­dorcet-IRV or WoodSIRV, and turns IRV into a vot­ing method that I would be very strongly in fa­vor of.

(I think ap­proval vot­ing is prob­a­bly a bet­ter choice, but it would be good to see more ex­am­ples of it in prac­tice in com­pet­i­tive elec­tions first.)

[1] A bit of his­tor­i­cal irony: in 2008 Greg Den­nis, an IRV sup­porter, wrote in Why I Pre­fer IRV to Con­dorcet:

In this sce­nario, the pres­ence of a can­di­date with strong core sup­port causes a Con­dorcet win­ner with lit­tle core sup­port to lose. For­tu­nately, de­spite the the­o­ret­i­cal pos­si­bil­ity of this sce­nario, the em­piri­cal ev­i­dence sug­gests that it is van­ish­ingly rare in prac­tice. In fact, if you look at the bal­lot data that is pub­li­cly available for the IRV elec­tions held in the US (San Fran­cisco, CA, Burling­ton, VT, and Pierce County, WA), for in­stance, you’ll see that IRV and Con­dorcet agree on the win­ner ev­ery sin­gle time.
This was at a time when IRV had been used for only a few elec­tions. The very next Burling­ton VT elec­tion, in 2009, ran into ex­actly this situ­a­tion. This led to enough con­tro­versy that Burling­ton re­pealed IRV in 2010.

Com­ment via: facebook