When Things Get Worse: The IRV Election in Alaska

Politics aside, the math isn't working out

Ingredients: one winner (with a big smile on her face), two losers from same party slinging mud at each other, and an obscure vote-counting method. Alaska is the second traditionally conservative state, after Utah, to adopt a new (old, actually) preferential voting method called Instant Run-Off Voting (IRV).

The debate whether preferential voting (aka “Ranked Choice Voting” or RCV) is sensible and how to best use it tends to be lengthy, so I skip it, deferring the reader to existing sources (among them, one great book, awesome treasure chest of in-depth information, great talk, and a pro-IRV lobby site, for balance)

However, unlike the RCV concept, IRV is just one of many alternatives to count preferential votes, and the question whether we want to live with it seems to be getting clearer by the day: As we showed here, in Utah most recently, IRV can lead to a plethora of counter-intuitive and pathological outcomes. “Would you like to vote for me? Not so fast—you might make me lose!” a candidate says to a voter. What?! (Well, check out our report and presentation).

What Went Wrong in Alaska: Enter Data

On September 8, 2022, I received an email from Alaska Elections Office announcing full release of “Cast Vote Record” (including all rankings). Kudos! Nothing helps election transparency more than data transparency. Take an example from Alaska, Salt Lake County Clerk!

Working with Warren D. Smith, we converted the unwieldy data files using a notebook (posted here) to a spreadsheet, ran our anomaly detector (described here). We report the following findings (to our knowledge we are first in doing so):

Disclaimer: At this point in time, our counts differ from the official tally (published here) by a few hundred votes. We believe the discrepancy is due the fact that the released data files do not include manually counted ballots (as per this statement).

Problem 1: Begich was the “Consensus Winner” but did not win

Nick Begich (R) defeats Sarah Palin (R) as well as Mary Peltola (D) in head-to-head comparison. But IRV elected Mary Peltola as the winner. This paradox is known in fancy-speak as “Condorcet Failure” (named after an 18th-century French mathematician). Electing the Condorcet (or consensus) winner is a hallmark of a “good” election method. IRV failed this test in Alaska (besides Utah and other places worldwide).

To be specific, Begich defeated Palin with 101229 vs. 63619 preferential votes; he also defeated Peltola with 87883 vs. 79458 votes (and, easily defeated all write-ins).

The presence of a Condorcet Failure is usually a sign of more things going wrong, so read on!

Problem 2: Had more voters voted for Peltola, Peltola would have lost

Yes, the subtitle is correct, your eyesight is not the problem. Had 5163 existing voters changed their vote in favor of the winner Peltola, Peltola’s smile would fade away rapidly as she would have now lost the election! This “More is Less” IRV pathology has been known for decades and it is beyond the pale that it alone hasn’t stopped the push for more IRV adoption. Similar defect(s) showed up in Moab, UT (documented here). When you run as a candidate in an IRV race next time, you may think twice instructing your constituency to vote for you, or maybe just not all of them - who knows?

Problem 3: Palin was the Spoiler!

“Ah, now you are pushing it!” says the skeptic. When someone drops out of a race, it surely can change the outcome - we all know this as the “Spoiler Effect.” But, spoiler avoidance is at the core of the IRV sales pitch:

“IRV Eliminates the ‘Spoiler Effect’” (quote from utahrcv.com)

Having the data, we can now tell that Palin dropping out would have changed the election outcome and Begich would have won. That is the definition of a spoiler. What a fail - thank you, salesman!

Problem 4: Peltola failed to win “True Majority”

One more nail in the IRV coffin. Wide-eyed election officials and legislators newly adopting IRV heard (and repeated) this many times. Repeat after me:

“(IRV ensures) a true majority winner” (quote from utahrcv.com)

The tacit assumption in the above statement is that zero ballots have issues and zero ballots get ever exhausted. That assumption is almost never true and leads many times to failure to reach true majority. In Utah we’ve seen this in 3 out of 16 IRV races and now in Alaska: In the released data, Peltola won with 91276 votes (91266 in the official tabulation, including manual ballots), but the total number of votes cast was 189112 (188582), so she was 3281 (3026) votes short of “true” majority. And we are not even counting the 3177 ballots (1.7% of the total) that were discarded due to voter confusion.

Problem 5: Ballot Issues

Oh, these silly uneducated voters are at it again! 1.7% of blank ballots and additional 3.3% of ballots with ranking gaps - ballot issues are their fault, why do they even show up? Or, could it be that many thought it natural to rank two candidates equal, say, first? Could it just be that election officials failed to explain that that will lead to their ballot being thrown out? Apparently many (3.3%) also thought it natural to rank one candidate first (love her) and another bottom (can’t stand him), leaving the middle empty (don’t care). This leads to the ballot being either rectified by arbitrary rules or truncated, as is done in Utah. We’ve seen some stunningly high percentages of IRV ballots discarded in Utah as well (see Table 1). Good enough for democracy?

What now?

With wider adoption of IRV we will continue seeing issues like the above. And it is only a matter of time when voters will wake up and get upset about how their votes get counted. Poundstone’s book “Gaming the Vote” describes the long history of vote counting, weighing pros and cons of the numerous methods. It is surprising to me that we continue implementing one of the most “broken” among them. It is also clear, that in states with pilot programs like in Utah tying IRV to RCV, a discussion has to be re-started. Voters have questions: is IRV the only viable way to count preferential votes? If not, what alternatives are there and why were they not considered? By the way, my co-author, Warren D. Smith, has an answer: adopt “Range Voting” (aka Score Voting) and you shall never see the “More Votes is Less” paradox and many others. I couldn’t agree more. But it is not about me, it is about an educated decision-making process in the legislature and about voters deserving exactly that. That process also includes proper legislation to guarantee election data release for independent analysis by researchers. Ask me what I was told when requesting data from the Salt Lake County Clerk!

Open Research

We are independent researchers who are interested in collaboration and data sharing. We believe openness and transparency are essential in keeping the technical debate honest and (as much as possible) free of politics. Reach out to us if you are interested in obtaining the Alaska data converted to a csv format. The converter notebook to process the original Alaska JSON-formatted files is available here. We will highly appreciate any feedback or corrections.

Updates

September 17, 2022: We learned about this research posted on arXiv on September 11 that looks into the Alaska special election from same perspective and makes similar conclusions. More researchers looking into the data is very good news!