Editor’s note: In two subsequent articles, Durning argues that approval voting would start no sooner than ranked choice voting in Seattle, if voters chose it in November, and shares a voting-rights law firm’s evaluation of the two options’ legality.
Seattle voters will decide in November whether to adopt approval voting, ranked choice voting, or no change to their primary election ballots. What do the research literature and practical experience say about these two alternatives to pick-one voting? They say a tremendous amount about ranked choice voting (RCV), and Sightline has summarized the lessons in tens of thousands of mostly encouraging words over years. But they say much less about approval voting (AV). This article summarizes what’s known.
Approval voting has intriguing features. It is simple to implement and easy to tabulate. Its proponents plausibly claim worthy benefits, including more moderate victors and less negative campaigning. It is worth experimenting with and deserving of study. We stand to learn more over time because two communities have recently adopted it. They are, in effect, conducting the clinical trials for this new voting method.
Both intuition and theory say AV will elect people who are, if not necessarily voters’ favorites, at least unobjectionable to most. They would be “consensus-style candidates,” say AV proponents. That tendency to dampen extremism and reward competent, cooperative governance, I surmise, is a major appeal of AV for its proponents and financiers, and it’s a goal Sightline shares.
But adopting AV for primary elections for all city offices in Cascadia’s largest city would be risky. For all its appeal, AV is a novel, unproven, and legally untested system for elections, and it has weaknesses that should give us pause.
For all its appeal, AV is a novel, unproven, and legally untested system for elections, and it has weaknesses that should give us pause.
The most salient fact about approval voting is that we do not actually know much about how it operates in the real world, not just in simulations or in academic papers or in the minds of schemers on Twitter or Reddit but in the pressure cooker of actual government elections, where campaigns rage, TV ads promise and malign, tempers flare, and hopes soar. This is where voters face the unavoidable dilemma into which AV forces them—that is, between helping their favorite and guarding against their least favorite.
We do not know if AV will deliver what it promises (moderation and a dampening of extremes), but we have reasons for doubt. We do not know if AV will lead to fair representation or pass court muster, but we have cause for caution. For these reasons, adopting it in Seattle in November would be risky, especially when Seattle can adopt the well vetted and helpful alternative of ranked choice voting instead. Let’s take a closer look at why.
1. Approval voting is too basic to represent a voter’s preferences
Let’s start with the voter’s experience.
For voters, approval voting, like ranked choice voting, is an alternative to old-fashioned, pick-one voting. In it, you can fill in the bubbles next to as many candidates as you like. Whichever candidate gets the most approval votes wins.
It seems simple, and simplicity is appealing. AV is easy to understand and easy to tabulate. Unfortunately, its simplicity also holds weakness. Indeed, simplicity can make approval voting frustrating for voters.
Approval voting seems simple, and simplicity is appealing. Unfortunately, its simplicity also holds weakness.
The defining feature of approval voting is that it’s binary: approve or not, bubble filled or blank, yes or no. You approve just your favorite. Or your favorite and your second favorite. Or your top three. Or any number you choose. It’s up to you. Simple.
What you cannot do is convey any other preferences. No rankings. No ratings. No way to say that you love Nader but would settle for Gore, that you’d really like Perot but could live with Bush Sr., that you’d be elated with Biden, could get excited about Klobuchar, would be satisfied with Booker, and could tolerate Mayor Pete.
If your preferences are black and white, with no shades of gray, AV may be for you. Otherwise, the more you think about it, the more confounding it becomes. If you actually care who you vote for, you have no good option. Approve Elizabeth Warren and you’ve maximized your help to her but exerted no influence over the rest of the race. Fill Bernie’s bubble too, and you’ve just halved the weight of your Warren vote. You’re now helping them equally. Add Mayor Pete and you’ve cut your weight in thirds. Every candidate you approve dilutes your other votes.
If your preferences are black and white, with no shades of gray, AV may be for you. Otherwise, the more you think about it, the more confounding it becomes.
You have to decide whether to minimize the chances of your least favorite winning (by approving all but Trump, for example) or gamble by voting for your favorite (all in for Ben Carson, for example). The more you care, the more painful the choice. It’s not so simple after all. And that’s why most people just fill the bubble for one candidate, as discussed below.
For now, though, focusing just on voters’ individual experiences, what’s most eye-rolling about approval voting is that, well, AV acts as if no one had ever invented numbers. Its binary nature is not only woefully constraining but also entirely unnecessary.
And from the voter’s perspective, that’s the most basic reason that ranked choice voting is less risky than approval voting. Ranked choice ballots let you vote for candidates in order of preference, the way humans do… pretty much all everyday decisions they make that offer a few options: what to order for lunch when the kitchen has 86’d your favorite, which movie to watch when Netflix doesn’t have the classic you were craving, which household costs to prioritize over others in a tight month.
And ranking your second doesn’t hurt your first choice. RCV ballots differentiate among alternatives and reflect the nuances of your views. Marking Warren second doesn’t hurt Bernie if you’re on the Bernie train. Voting for Bush Sr. second doesn’t hurt Perot if you’re in the tank for Ross. Ranking Biden lower than Pete doesn’t hurt the mayor if he’s the one you like best.
That’s the right kind of simple.
2. Approval voting is new and unproven
Approval voting is new
Approval voting is new to government elections. So far, worldwide, only midsize St. Louis, Missouri (with less than half Seattle’s population), and Fargo, North Dakota (with one-sixth), have used it for elections. Between them, they’ve only used it three times, all since June 2020, with a total of fewer than 100,000 ballots. Three times, ever, anywhere in real-world, government elections.
Some 11 million voters in the United States now live in RCV jurisdictions that in total have completed hundreds of RCV elections, and Americans have cast tens of millions of RCV ballots.
In contrast, ranked choice voting has long histories in Australia, Ireland, New Zealand, and Scotland, and in the United States, its use is growing steadily. Some 55 US jurisdictions are already—or will soon be—employing it. They include the states of Alaska and Maine; blue cities including New York, Minneapolis, and San Francisco; more than a dozen red cities and towns in Utah; and six states where parties use it for presidential primaries. Some 11 million voters in the United States now live in RCV jurisdictions that in total have completed hundreds of RCV elections, and Americans have cast tens of millions of RCV ballots. In Washington state, voters will consider RCV in November not only in Seattle but also in Clark County (which surrounds the city of Vancouver, Washington, in the state’s southwest corner). In Oregon, it is already used in Benton County, which is home to Corvallis, and will be on the ballot in both the city of Portland and surrounding Multnomah County.
Approval voting is unproven
From all of this experience and the exertion of researchers, we can speak with confidence about RCV. A recent compilation that adds to the robust body of research findings, for example, tested some claims of RCV proponents and critics and concluded that RCV is safe and generally helpful, though perhaps not the panacea some proponents imply:
Consistent with previous RCV research, most of the studies in this series found RCV to be either a comparable or modestly better alternative to our standard “first-past-the-post” or plurality method.
RCV, it continues, yields small improvements on measures such as rates of voter error and does no harm on other measures, such as racial polarization. The findings of this new set of papers are quantitative, peer reviewed, and published for all to debate. They add to the terabytes of accumulating PDFs of empirical research. I might quibble with some conclusions or invoke RCV benefits not studied, but the point is that we can debate RCV based on empirical analysis. Almost no analysis of AV’s real-world application exists. It’s too new.
Instead, AV proponents are left to argue based on simulations and math. The Center for Election Science (CES), the world’s main AV promoter (and the largest funder of Seattle’s AV campaign), lays out a series of arguments for AV on its website, and they’re interesting. Almost every one of them, though, is supported only by theory. CES staff write, for example:
Computer simulations using Bayesian regret calculations . . . demonstrate better utility outcomes in elections using approval voting versus RCV even if all approval voters were tactical and all RCV voters were honest.
Um. What?
Seriously, though. Analysis like this does form interesting theory. Developed by a small cadre of smart, sincere mathematicians and thinkers, such as William Poundstone, Clay Shentrup, and Warren Smith, AV theory warrants serious consideration, though it’s far from unarguable. An equally smart and serious set of scholars, including James Green-Armytage, Jack Nagel, and Nicolas Tideman, contest it fiercely. In the end, such theoretical arguments may be unavoidable. All voting systems have theoretical flaws: Nobel laureate Kenneth Arrow proved that point 70 years ago. So choosing one system over another is about choosing which flaws are the least bothersome. And those choices inevitably involve value judgments—judgments best made based on empirical data from actual elections, not just from theory.
Again, AV theory is important work, and it justifies trying AV in places like Fargo and St. Louis, which had barriers to RCV. It justifies studying the results closely in those places. It’s promising. It might even justify testing AV in elections for offices where skills and competence matter a lot, and ideology should not matter at all, like election officials or county auditors. But it provides little basis for thinking AV is ready for choosing the political leaders of Seattle.
AV is like a potential new medicine that performs well in computer simulations and in animal trials and is now in its first clinical trials. Its sponsors are gathering information from the trial in two cities and are hopeful about the outcome. But it’d be dangerous to jump to universal distribution just yet, before anyone has ascertained if it’s safe and effective.
Many—indeed, most—drugs don’t make it through all the phases of clinical trials, because simulations and lab results cannot anticipate the complexities of real life. Elections are similar: RCV has survived its battle testing in the heat of actual campaigns, and it works well. AV is just starting that trial.
AV promises moderation, but . . .
Proponents argue that AV will dampen extremism and elect widely acceptable centrists. Seattle AV leader Logan Bowers told The Stranger, for example, “Right now the camps get pretty Orthodox—you got to be all in or all out, but there’s a lot of voters that are kind of in the middle. I think you would see more candidates adopt a lot of the policies that are just actually popular with voters.”
This argument sounds plausible, but so far, it’s not coming true. In the three AV elections to date, the winners were almost all well-established incumbents or candidates who clearly would have won anyway. And the one controversial, far-right incumbent in office in Fargo, David Piepkorn, who stood for reelection under AV? He won easily. So much for dampening extremism.
In the three AV elections to date, the winners were almost all well-established incumbents or candidates who clearly would have won anyway…
A reason to doubt the moderation claim is that in this (again, tiny) sample of AV elections, most voters only approved one candidate. In voting-scholar lingo, they “bullet voted.” That is, they just picked one candidate and voted as if AV didn’t exist, as if they were still in a conventional, first-past-the-post system. But the more people bullet vote, the more AV becomes indistinguishable from that system it’s trying to replace.
And it’s not that people are uninformed: bullet voting makes sense in AV. If you prefer a candidate, bullet voting is how to help him or her the most. Candidates understand that fact, and they tell voters. Said Fargo Mayor Tim Mahoney during his reelection campaign in the June AV election, “I would probably bet that every candidate says just vote once because that has more power as a vote.” Likewise, one of his challengers, Republican state representative Shannon Roers Jones, said at the launch of her campaign, “It will be important for me to convey to people . . . that the most effective way to elect the person they care about is to vote only for that person.”
…and the one controversial, far-right incumbent in office in Fargo, David Piepkorn, who stood for reelection under AV? He won easily. So much for dampening extremism.
Fargo and St. Louis voters listened, or they figured out AV’s internal logic for themselves. Most of them seem to have bullet voted. Elections administrators in neither city have released data on approval votes per ballot, but estimates are possible (see appendix). Apparently, between 50 and 90 percent of ballots were bullet voted, depending on the election and the race. In contrast, in ranked choice voting, between 10 and 40 percent of voters tend to bullet vote, ranking only a first choice.
If most people bullet vote, how much moderating impact can AV really have? Over time, AV in Fargo and St. Louis will tell us more, but for now, we cannot know. AV is just too new.
. . . RCV actually delivers moderation.
In Seattle, curiously, the media have characterized RCV as serving the interests of the left and AV as serving the interests of the center. That’s wrong on both counts. AV is a political unknown, and RCV favors neither left nor right but broadly popular candidates who also have a base of strong support. Perhaps that’s why extremists oppose it, from Donald Trump on the right to Kshama Sawant, Seattle’s one socialist city council member, on the left.
RCV favors neither left nor right but broadly popular candidates who also have a base of strong support.
The track record of RCV is one of victories, mostly, for centrists: Virginia’s Governor Glenn Youngkin was the most moderate Republican in his RCV primary. New York City Mayor Eric Adams was the most moderate of Democratic frontrunners in his RCV primary. Maine’s Democratic Congressman Jared Golden won an RCV election in a swing district by hewing hard to the center. Alaska’s moderate Senator Lisa Murkowski is a rare Republican who voted to remove President Trump yet is now positioned to win reelection. She is likely to survive a challenge from a Trump-endorsed right-winger thanks to Alaska’s new open primaries and RCV general election.
AV promises civility in campaigns, but . . .
Proponents suggest AV improves candidates’ civility with one another in political campaigns. Seattle’s AV campaign, for example, claims that AV elections “make politics less divisive. Campaigns won’t fight each other over voters—they’ll fight for wider support.”
Why would AV candidates stay positive, when approvals for others dilute their own support?
So far, though, we have no empirical evidence of this effect. In fact, polling from Fargo showed that voters perceived the June 2022 AV campaign as more negative, not more positive, than normal. It’s just one data point, but it aligns with the incentives of AV: why would AV candidates stay positive, when approvals for others dilute their own support?
. . . RCV actually delivers civility in campaigns.
RCV does change the tone of campaigns and of governing, because candidates (and incumbents) want not just first-place votes but also second- and third-place votes. RCV candidates have less incentive to “go negative” than in the status quo. A lot of research and experience backs up this conclusion.
3. Approval voting is legally untested and might worsen representation
Approval voting may violate a basic principle of US election law: one person, one vote. Courts might throw it out.
What’s more, the AV proposal on Seattle’s November election might underrepresent the city’s political minorities, racial and otherwise. That would be unfair and might lead courts to throw out AV under federal or state voting rights acts.
Either way, the election system might end up in disarray.
AV’s basic method is legally untested
In an approval voting election, I can vote for five candidates while you vote for one, and each of our votes will carry equal weight. Doesn’t that mean I have five times as much say as you do? Isn’t that a glaring violation of “one person, one vote,” the 1964 US Supreme Court principle that “the weight and worth of the citizens’ votes as nearly as practicable must be the same”?
In a similar vein, Washington law says, “Nothing in this chapter may be construed to mean that a voter may cast more than one vote for candidates for a given office.” In a debate about AV in Olympia in 2019, reported by The Olympian, Thurston County Auditor Mary Hall replied to the question of whether it’s legal: “It’s not. There would have to be legislation to allow (it).”
Seattle’s AV proposal might produce worse representation for people of color and violate voting rights laws
Seattle’s AV measure applies only to selecting winners of open, top-two primaries. Converting the general election to AV (or RCV or anything else) would require a change in state law (which Sightline and others have been working toward for years). Such a change would allow localities across Washington to engage in the kind of thorough evaluation of voting methods that Portland, Oregon, has undertaken of late, leading to its charter reform proposal that includes multi-winner RCV—a promising form of proportional representation tailored to Portland’s needs.
In Seattle, though, the only legal option is to change the top-two primary either to AV or to RCV, and in this context, approval voting is not just an intriguing, unproven novelty; it could be a step backward on fairness. Top-two AV in Seattle might worsen representation of the city’s voters. How?
In Seattle’s existing method, everyone gets one vote, and the two most popular candidates move from the low-turnout, less representative August primary to the high-turnout, more representative general election in November. The first-place finisher represents the largest bloc of primary voters. The second-place finisher represents the second-largest bloc.
Example of a RCV ballot for Alaska
With AV, the same bloc of the electorate (perhaps the older, home-owning, educated people who tend to vote in primary elections) could pick both of the winners, dictating the candidates that general election voters choose from. That is, instead of the largest two blocs each picking one candidate, the largest bloc would select both. And that would be representational backsliding, an egregious unfairness.
The root of the issue is that the most informed voters can increase their influence. As RCV proponent Rob Richie of FairVote wrote in a 2011 critique of AV and similar methods, “Once aware of how approval voting works, strategic voters will always earn a significant advantage over less informed voters.”
Using that advantage to sway election outcomes is no certainty, of course; it depends on many people voting the same way. Still, it’s a worrisome prospect, and it’s a worry that justifiably preoccupies some of Seattle’s leaders of color. More reason for concern comes from University of Missouri’s professor David Kimball, who assembled evidence in St. Louis that voters in majority-white wards cast more approval votes on their ballots for mayor than did those in majority-Black wards.
If white areas of Seattle approve more candidates per ballot than do areas home to more people of color, as happened in St. Louis, would that not violate the federal and Washington state voting rights acts? Racially polarized voting plus disproportionate influence for white voters are at the heart of those laws.
Even if, by the vagaries of logic or lawyering, the legal system ultimately permits AV, the court gauntlet might still consume months. It could even confuse the schedule of certifying elections and swearing in winners. That’s yet another reason for caution about approval voting.
RCV represents people better
Ranked choice voting, meanwhile, sidesteps each of these legal perils: each voter gets one vote , and it can transfer in its entirety from disqualified to still-contending candidates, as if in a series of “instant runoff” elections. RCV does not create racially polarized voting or disproportionate influence by white voters, and it’s been tested more than a dozen times in US and state courts. It’s emerged unscathed.
RCV also gets good marks on electoral representation. In a top-two primary, RCV would be a modest improvement on pick-one voting. Eliminating the least popular candidate and reassigning ballots to those voters’ next favorite choices, then repeating the process until only two candidates remain, maximizes the chances that the two most popular candidates proceed to the general: the top vote-getter is chosen by the largest group of voters. The runner-up is chosen by the second-largest group of voters. And unlike in AV, those groups of voters will always be different people.
(9/19/2022 UPDATE: Sightline commissioned a legal analysis from a Seattle-based law firm with experience in election and voting law. Its conclusions include: RCV is legally safe. AV runs a small risk of being thrown out by courts even before implementation because of its apparent violation of the state election law mentioned above. AV also faces a small risk of being thrown after implementation, because it could violate federal or state voting rights acts. Read the legal memo here or Sightline’s summary here.)
4. AV is risky; RCV is tested and delivering the benefits voters want
AV is a novelty: intriguing and worth attention but not yet proven or well understood. Adopting it in Seattle in November, for use in top-two primaries, would put voters on the horns of a dilemma: do they rally to their favorite or spread their votes across all they can tolerate? Adopting it might do little or nothing to dampen extremism or tame negative campaigning; indeed, RCV would be a safer means to those ends. And adopting AV runs the risk of degrading representation of minorities in Seattle, an injustice and an invitation to court challenges.
For all the reasons I’ve laid out in this article—a dilemma for voters, a novel system with unknown characteristics, low confidence it will deliver moderation or a more positive political climate, caution about fairness and court review—AV seems a risky choice for Seattle in 2022. It’s just too novel—a promising new treatment for our democracy that needs to finish clinical trials before we start using it.
Meanwhile, RCV is tested and ready, no panacea but a modest improvement and free of the uncertainties that bedevil AV. From what we know now, therefore, RCV > AV.
Appendix: bullet voting in AV (and RCV)
In Fargo, an opinion poll by the national pro-RCV research and advocacy organization FairVote estimated that 60 percent of voters bullet voted in the seven-candidate AV mayoral election of June 2022. For sake of comparison, in New York City’s contested Democratic RCV primary for mayor in September 2021, which nominated Eric Adams, only 13 percent bullet voted, according to a recent study by the City University of New York. In RCV elections overall, some 29 percent of participants bullet vote, on average, according to a dataset on RCV elections maintained by FairVote, and the more competitive the race, the fewer people bullet voted, dropping toward 10 percent in the highest-stakes, highest-profile races.
Lacking ballot-by-ballot data on approvals, we can glean insight from averaging approvals given over ballots cast. Such figures do not reveal how many people bullet voted, because they do not show whether, for example, many people approved two candidates each or a few people approved ten candidates each. Still, they give a sense.
For context, in the Fargo mayoral race just discussed, with its 60 percent bullet voting rate, voters approved an average of just 1.5 candidates from a field of seven. The mayoral primary in St. Louis’s one AV election, in March 2021, selected two finalists from a field of four, and in these races, voters approved an average of 1.6 candidates. If the voting pattern in this race was like that in Fargo, that figure might correspond to a bullet voting rate of 50–60 percent. For comparison, I looked at all recent four-candidate RCV races from the FairVote dataset—13 of them—and found that voters ranked 2.3 candidates apiece on average, and about 40 percent of them bullet voted.
In St. Louis’s city council AV primary, two candidates qualified to go on to the general election for each seat, and voters approved an average of just 1.1 candidates per seat, though the races had three candidates each. The low approval rate of 1.1 per ballot corresponds to a bullet voting rate of 90 percent or higher. Mathematically, if 90 percent of voters bullet voted and 10 percent approved two candidates, you’d get 1.1 approvals per ballot. In comparison, in all 17 recent three-candidate RCV races in the FairVote dataset, voters ranked 2.1 candidates, on average, and fewer than 40 percent of them bullet voted.
Overall, therefore, bullet voting rates in Fargo and St. Louis AV elections were likely between 50 and 90 percent, while RCV bullet voting rates in comparable elections have been between 10 and 40 percent.
Clay Shentrup
Virtually everything in this post is incorrect.
For one, the ranked voting method on offer here “bottoms-up” RCV, which has never been used in the USA. Whereas approval voting has been used in Fargo and St Louis, having around 100,000 and 300,000 residents respectively.
“Bottoms-up” ranked voting works differently than the Single Transferable Vote system normally called “ranked choice voting” by its advocates. Rather than eliminating the weakest candidate until a majority winner has been found, it keeps going, eliminating until only two candidates remain. This means you end up with the standard “RCV” (Instant Runoff Voting) “majority winner” running in the general against the candidate favored by those voters who _disliked_ the majority winner.
This is very bad for minorities! Why?
Because bottoms-up is intended as a (semi)proportional _multi-winner_ system, where the explicit goal is to award seats to both the majority and minority factions. But Seattle elections aren’t multi-winner. The primary sends two candidates to the general election, which ultimately results in a single winner. This is a recipe for disaster. The majority winner will in 99% of cases trounce the minority candidate in an uncompetitive landslide.
Whereas approval voting wild tend to find the two most broadly appealing candidates. This means the minority faction actually has influence, not only in tugging the center in their direction in how they influence the two finalists, but also in terms of who wins the general.
This is exactly what St Louis voters saw when they used approval voting in their March 2021 primary. The two by far most progressive candidates advanced to the general election, resulting in a highly competitive race which elected their first black female mayor.
https://electionscience.org/press-releases/st-louis-voters-use-new-approval-voting-system-in-march-primary-election/
Now as for “one person, one vote”…
It’s ranked choice voting, not approval voting, that arguably violates “one person, one vote”. With approval voting, all voters are equal in a very literal mathematical sense. For any way you can vote, I can approve the opposite candidates, and my ballot therefore has an equal but opposite effect that precisely cancels your out. In the political “tug-of-war”, we have the same power.
But ranked choice voting can count some voters’ preferences between candidates X and Y while ignoring other voters’ preferences. This is due to the “later no harm” flaw, described here by a Princeton math PhD and renowned expert named Warren Smith, whose work was the focus of William Poundstone’s book “Gaming the Vote”.
http://scorevoting.net/IrvIgnoreExample.html
This is why, for example, IRV elected the Progressive in the 2009 IRV mayoral race in Burlington, Vermont, even tho a large 54% to 46% majority preferred the Democrat to the Progressive. The IRV tabulation process completely ignored that most Republican voters preferred the Democrat to the Progressive, because it intentionally ignored their 2nd and 3rd place votes (those aren’t looked at until/unless your 1st choice is eliminated).
I could go on for several pages to detail the many other fallacies in this piece, but these alone aptly demonstrate the lack of basic research that typically goes into Sightline’s voting related articles.
Alan Durning
Hi, Clay. Thanks for reading.
Churlishly, perhaps, I wonder if it was incorrect when I complimented you as smart and sincere in the text?
I had read with interest the CES piece on St. Louis that you link. Unfortunately, it is a rather superficial gloss. I encourage you, your CES colleagues, and other AV theorists to dig into the three AV elections completed so far. They’re an opportunity for empirical analysis. Seek access to much more ballot data, so that we can learn how people behaved. Exactly how much bullet voting was there? What was voter behavior in different wards? Which candidates were approved together? Such analysis, in combination with polling, could help resolve some of the theoretical disputes that you rehearse here.
Regarding one person, one vote, I find your logic strained, but my view isn’t terribly important, considering that US courts have considered suits against RCV more than a dozen times and ruled it copacetic. RCV only counts one vote at a time; AV does not. We will see what courts say about AV eventually, I expect.
The Burlington ghost you resurrect is tiresome. For readers, Sightline’s Kristin Eberhard wrote our analysis of the Burlington case five years ago here: https://www.sightline.org/2017/08/28/dont-let-this-voting-theorist-kill-your-electoral-reform-momentum/
And of course, “later no harm” is theorists’ principal complaint with AV, not with RCV. If we’re going to cite mathematicians’ theories — and Warren Smith is smart and sincere! — there’s James Green-Armytage’s analysis showing that AV is subject to strategic manipulation many multiples more often than RCV. But I’d prefer to set aside such arguments and return to my central point: all systems have theoretical flaws; only real world results are of much interest at this point. And we don’t have much empirical evidence yet about approval voting.
Clay Shentrup
Alan,
> RCV only counts one vote at a time; AV does not.
Reynolds v. Sims (377 U.S. 533) stated that no vote should count more than any other so that it has unequal weight. This unequal weight would violate the Equal Protection Clause of the Constitution. This is about voters having equal weight, not about how many votes are counted at a time.
It is mathematically proven that the forms of RCV we’re discussing here violate this principle, and the Burlington result demonstrates this. Your cited article by Kristin Eberhard in no way refutes this—it merely argues that scenario is rare. But with approval voting, it is impossible. Every vote is counted.
Of course judges are not voting experts, and they can and do sometimes rule in ways that make no logical or mathematical sense. So you are at least right in the practical sense, that RCV has more precedent-based safety than approval voting to a particular legal argument. But it also has a lot of vulnerability to arguments that approval voting evades, like complexity and voting machine upgrades. On practical grounds, I think approval voting is far better positions to rapidly scale and replace the status quo.
> “later no harm” is theorists’ principal complaint with AV, not with RCV.
That’s incorrect. RCV advocates complain that approval voting *doesn’t* pass later-no-harm. Experts complain that RCV *does* pass later-no-harm. The key point here being that later-no-harm is a flaw not a benefit, as I explain here.
https://clayshentrup.medium.com/later-no-harm-72c44e145510
In a nutshell, obviously later-no-harm is of little benefit voters like my aunt, who would rank Biden in 1st instead of Warren in order to stop Trump, a strategy called “pushover” that’s impossible with approval voting.
And we’ve meticulously studied the St Lous results. I wrote about them here for instance.
https://medium.com/election-science/st-louis-approval-voting-152357f7a000
Your fixation with bullet voting is a common sign of lack of familiarity with the science of voting methods, and more fundamentally, with statistics in general. Consider this race for instance.
MICHELLE SHEROD 1195 69.04%
TINA PIHL 800 46.22%
DON DE VIVO 106 6.12%
Sherod was, to a statistical near-certainty, the most popular candidate. And yet your framing would characterize this as a poor outcome because the vast majority of voters (rationally) bullet voted. This is basic statistical/reasoning error is rampant among novices poking around in voting reform, and something that experts have had to spend an inordinate amount of time dispelling. It would be helpful if you consulted with experts before writing pieces like this. The gravity of the subjective demands more rigor.
Jason Osgood
*I wonder if it was incorrect when I complimented you as smart and sincere in the text?*
Uncool. Clay is thoughtful, informed, logical, and wellspoken. I’ve constructively debated Clay on other forums. Our minor disagreements are honest and sincere, not because of any emotional attachment to any particular conclusion.
Alan Durning
Jason,
Clay’s comments on this string, from the first one onward, do include reasoned and supported claims, but they are also laced with condescension and insults to the intelligence and knowledge of Sightline and our commenters. These comments arguably violate Sightline’s website guidelines for civility, and the only reason we decided to let them stand is because of Clay’s stature as a well known and energetic AV proponent.
Alan
Parker Friedland
I disagree with Clay’s definition of 1 person 1 vote but he is not saying RCV violates 1 person 1 vote any more then FPTP violates 1 person 1 vote. FPTP also (trivially) passes the later no harm criterion via a technicality (if you told voters to rank all their preferences and only counted the 1st prefs, adding additional ranks does not hurt your favorite). He is simply pointing out that under such systems 2 people with opposite preferences don’t cancel their votes out like they do under approval. For example, if in a FPTP election you have 2 D’s and 1 R, R voters would have twice as much voting power because to completely cancel out an R vote, you need two D votes (1 for the first D, one for the 2nd). RCV can also have similar examples where there is no way to cancel one vote with another but the examples are more complicated. For courts to rule that RCV violates 1 person 1 vote this way they would also have to rule that FPTP also violates 1 person 1 vote (which in this sense it does, but that’s not going to happen). I hope I cleared that up for you. I know Clay isn’t always a great communicator and his rants can be incoherent at times.
Shel Kaphan
Clay Shentrup said:
“For one, the ranked voting method on offer here “bottoms-up” RCV, which has never been used in the USA. Whereas approval voting has been used in Fargo and St Louis, having around 100,000 and 300,000 residents respectively. “Bottoms-up” ranked voting works differently than the Single Transferable Vote system normally called “ranked choice voting” by its advocates.”
Single Transferable Vote (STV) is the name usually used for proportional RCV when used to elect multiple winners, such as a city council, as a version of it is used in Cambridge, MA. It is *not* what is normally called “Ranked Choice Voting” by advocates. In STV, in addition to the usual RCV mechanism for eliminating the candidate with the fewest first place votes, there is another mechanism to redistribute extra votes beyond the threshold to win for candidates that have already been elected. This is what makes the results of STV elections for multi-winner contests more proportional to the interests of the electorate. By far the most common kind of RCV, especially in the US, is for electing a single winner. “Bottoms-up RCV”, as would be used in Seattle to elect the two finalists in primaries, works exactly the same as the usual single-winner RCV, except that it stops one round sooner.
From the Wikipedia page on STV: https://en.wikipedia.org/wiki/History_and_use_of_the_single_transferable_vote#History_in_the_United_States
As of 2016, the only official governing bodies that use STV to elect representatives are the City Council and School Committee of Cambridge, Massachusetts, and the Board of Estimate and Taxation (2 members) and the Park and Recreation Board (3 members) of Minneapolis, Minnesota.[73] However, STV was more widely used in the United States in the first half of the 20th century.
“Rather than eliminating the weakest candidate until a majority winner has been found, it keeps going, eliminating until only two candidates remain. This means you end up with the standard “RCV” (Instant Runoff Voting) “majority winner” running in the general against the candidate favored by those voters who _disliked_ the majority winner.”
No. At each round of the tabulation, the weakest remaining candidate will be eliminated until two are left, the two finalists. At that stage, while one of the two will (except for a tie) have the majority of the votes, the other one is not necessarily disliked by the finalist with more votes. Some ballots for the first finalist may have the second finalist as their second (or lower) choice, and some ballots for the second finalist might have the first finalist as their second (or lower) choice.
“This is very bad for minorities! Why?
Because bottoms-up is intended as a (semi)proportional _multi-winner_ system, where the explicit goal is to award seats to both the majority and minority factions. But Seattle elections aren’t multi-winner. The primary sends two candidates to the general election, which ultimately results in a single winner. This is a recipe for disaster. The majority winner will in 99% of cases trounce the minority candidate in an uncompetitive landslide.”
First of all, I think it’s fair to say that we do want a system where in a single-winner election, the majority wins. Secondly, the claim about a landslide is pure speculation. With this form of RCV, at least the two finalists will represent the two largest consolidated blocs of voters, because any vote splitting will have been taken care of by the elimination rounds which reassign ballots to the next choices. If the majority wins in the general election, by a little or by a lot, that’s pretty much the desired outcome. At least the most consolidated of the minority interests will have a chance.
“Whereas approval voting wild tend to find the two most broadly appealing candidates. This means the minority faction actually has influence, not only in tugging the center in their direction in how they influence the two finalists, but also in terms of who wins the general.
This is exactly what St Louis voters saw when they used approval voting in their March 2021 primary. The two by far most progressive candidates advanced to the general election, resulting in a highly competitive race which elected their first black female mayor.”
It’s not at all clear that is a desirable outcome. If two identical clones were running in a two-winner election, in approval voting the same voters would presumably approve both, and if one won, so would the other, and they’d both advance to the general election. This is not what we need. We would like the two finalists to represent different groups of voters, otherwise if there are similar candidates running, the largest plurality will completely dominate the outcome, and won’t give even the next largest group a chance for representation.
“Now as for “one person, one vote”…
“It’s ranked choice voting, not approval voting, that arguably violates “one person, one vote”. With approval voting, all voters are equal in a very literal mathematical sense.”
… but not in a practical, pragmatic sense. In fact the very literal sense is questionable as well.
“For any way you can vote, I can approve the opposite candidates, and my ballot therefore has an equal but opposite effect that precisely cancels your out. In the political “tug-of-war”, we have the same power.”
But a person who approves 5 candidates has potentially 5 times the influence of a person who only votes for 1 candidate.
“But ranked choice voting can count some voters’ preferences between candidates X and Y while ignoring other voters’ preferences. This is due to the “later no harm” flaw
…
This is why, for example, IRV elected the Progressive in the 2009 IRV mayoral race in Burlington, Vermont, even tho a large 54% to 46% majority preferred the Democrat to the Progressive. The IRV tabulation process completely ignored that most Republican voters preferred the Democrat to the Progressive, because it intentionally ignored their 2nd and 3rd place votes (those aren’t looked at until/unless your 1st choice is eliminated).”
Alan was correct that no election system is without artifacts. At least with RCV we are pretty familiar with what the artifacts are. RCV ignores all preferences other than the voters’ first remaining choice on their ballots. This is a feature, not a bug. It’s why RCV adheres to “one person, one vote”. Only your first remaining choice counts, until that candidate is eliminated. And then your next choice counts. “Later no harm” is not a “flaw”, it’s an essential feature. The definition of “later no harm” is that a vote for a lower choice candidate cannot harm a higher choice candidate. Approval voting does not have this property. Without this property, voters will soon learn that if they have any preference at all between candidates, they are better off just voting for their favorite candidate that has a chance to win. Otherwise they risk pushing that candidate into third (or lower) place.
Clay Shentrup
> “Bottoms-up RCV”, as would be used in Seattle to elect the two finalists in primaries, works exactly the same as the usual single-winner RCV, except that it stops one round sooner.
Completely false. Instant Runoff Voting aka IRV (what people normally call “ranked choice voting” or “RCV”) could stop as early as the 1st round, if a candidate has a majority. Whereas “bottoms-up” will *always* continue until only two candidates remain. So you could have a case where there are 20 candidates, and IRV/RCV halts immediately with a first-round majority winner, whereas bottoms-up eliminates 18 candidates until only 2 remain.
It’s pretty concerning that advocates willing to take the time to read a Sightline article can’t even understand how the Seattle proposal works.
> Single Transferable Vote (STV) is the name usually used for proportional RCV when used to elect multiple winners, such as a city council, as a version of it is used in Cambridge, MA. It is *not* what is normally called “Ranked Choice Voting” by advocates.
Incorrect. Here’s FairVote:
“In 2019, the City of Eastpointe, Michigan adopted proportional ranked choice voting (RCV) as part of a settlement with the U.S. Department of Justice (DOJ) under the Federal Voting Rights Act.”
Source: https://www.fairvote.org/ranked_choice_voting_and_proportional_representation_successes_from_eastpointe_michigan
Or this: “Cambridge, Massachusetts has used the at-large form of ranked choice voting, an American form of proportional representation, to elect its City Council and School Committee since 1941.”
Source: https://www.fairvote.org/history_rcv_cambridge
And here’s the city of Albany, California, which adopted STV: “Ranked Choice Voting
An information session to learn more about Ranked Choice Voting (RCV) in Albany and for anyone who is interested in running for local office in Albany., open to all, was held on Monday, July 25, 2022, 6 PM – 8 PM.”
Source: https://www.albanyca.org/departments/city-clerk/election-information/ranked-choice-voting
But you’re missing my point in the first place. My point was that the Seattle proposal has never been used in the USA, or virtually anywhere, despite its advocates calling it “Ranked Choice Voting” so as to imply it has precedent.
> No. At each round of the tabulation, the weakest remaining candidate will be eliminated until two are left, the two finalists.
This is exactly what I said. But note that “weakest” here means “having the fewest first-place votes”, not “least popular”. Here’s a hypothetical example to show you what I mean:
35% Bruce > Jessyn
33% Lorena
32% Jessyn > Bruce
Jessyn is eliminated and then Bruce trounces Lorena in the general. Progressive are steamrolled.
But Jessyn is preferred to Lorena by a massive 67% landslide, and makes a much better choice to advance to the general against Bruce. That would be a much more competitive election, and it would allow progressive Lorena supporters to at least tip the balance of the general election for a compromise choice.
> At that stage, while one of the two will (except for a tie) have the majority of the votes, the other one is not necessarily disliked by the finalist with more votes.
“Necessarily” is an overly strong word, that obscures the huge statistical bias here. As you can see, the 2nd/3rd/etc. choices of Bruce supporters are completely invisible to the tabulation process, and thus it is those voters having Bruce further down their list who disproportionately decide who advances to the general to be his opponent.
There’s simply nothing to argue here. This semi-proportionality is the *entire point* of the bottoms-up system. It’s described here by FairVote, the nations leading activist for RCV and proportional representation.
https://www.fairvote.org/types_of_rcv_for_multi_seat_bodies
> Some ballots for the first finalist may have the second finalist as their second (or lower) choice, and some ballots for the second finalist might have the first finalist as their second (or lower) choice.
Yes, and some Trump voters have Bernie as their second choice, and vice versa. But this is a statistical anomaly, and you cannot just ignore it because there are a small percentage of exceptions. In general terms, my point is 100% correct.
> First of all, I think it’s fair to say that we do want a system where in a single-winner election, the majority wins.
No. It’s well established that there is not even a mathematically sound *definition* of “majority winner”, and it’s mathematically proven that even when there is an undisputed majority winner, the electorate as a whole may prefer someone else.
https://web.archive.org/web/20190219005032/https://sites.google.com/a/electology.org/www/utilitarian-majoritarian
> Secondly, the claim about a landslide is pure speculation.
No, it’s not. It’s simple objective statistics.
> With this form of RCV, at least the two finalists will represent the two largest consolidated blocs of voters
This is, again, simply wrong. You’re essentially describing here what proportional STV would do. With bottoms-up, the normal RCV winner will generally be the strongest candidate *overall*, and the runner up will be the strongest candidate *among those voters who didn’t rank the strongest candidate above the that candidate*.
> any vote splitting will have been taken care of by the elimination rounds which reassign ballots to the next choices.
I just demonstrated that this is false. It’s already well known that RCV is vulnerable to vote splitting, but this bottoms-up form is *especially* vulnerable, as we can see that there’s vote splitting between Bruce and Jessyn in my example.
Indeed, even FairVote says, “Vote-splitting is posible under bottoms-up RCV compared to other forms of RCV.”
> If the majority wins in the general election, by a little or by a lot, that’s pretty much the desired outcome. At least the most consolidated of the minority interests will have a chance.
You’re making a well known mathematical fallacy here. The best way to maximize the odds of electing the most popular candidate in the general is to advance the two strongest candidates from the primary. While it’s true that there’s a disparity between primary and general electorates, the direction of that disparity is not in the same direction as the disparity between those who favor the majority winner and those who don’t.
I mean, this is obvious. Lorena was trounced. She had no hope of winning. But a slightly more progressive candidate than Bruce would have had a much more plausible case that would have made for a more meaningful contest. (I’m hypothetically using Jessyn in my example, because she’s well known, but you can of course think of some candidates who ran who were politically somewhere between Bruce and Lorena.)
> It’s not at all clear that is a desirable outcome.
Yes it is clear. To elaborate even further, it was at least conceivable in St Louis that:
1. The general electorate would have a slightly different preference on Tishaura vs Cara as compared to the primary electorate, and/or
2. That closer inspection of just those two candidates (e.g. watching their one-on-one debates) would shift preferences somewhat, even if the primary electorate was already statistically/demographically comparable to the general electorate.
Thus by pitting these two candidates against each other, voters had the best possible chance of electing the most popular candidate. Indeed, it was about a 4% difference between them in the general.
Whereas if they had sent Tishaura Jones against Lewis Reed, odds-makers could have predicted with great confidence that Jones would defeat Reed by 10% or more. It was simply statistically inconceivable, based on the primary results, that Reed would end up being the more popular candidate—either because of demographic differences in the general, or by making his case to voters during the general election campaign.
What I’m saying isn’t remotely controversial among anyone with basic familiarity with statistics.
> If two identical clones were running in a two-winner election, in approval voting the same voters would presumably approve both, and if one won, so would the other, and they’d both advance to the general election. This is not what we need.
I just showed in painstaking detail that you’re wrong.
> We would like the two finalists to represent different groups of voters, otherwise if there are similar candidates running, the largest plurality will completely dominate the outcome, and won’t give even the next largest group a chance for representation.
You have this precisely backwards. Sending an unpopular candidate, with no hope of winning, to the general, just to “have diversity”, actually *harms* the minority voters, by giving them a candidate with no hope of winning.
> … but not in a practical, pragmatic sense. In fact the very literal sense is questionable as well.
You’ve presented no evidence of this. Whereas we can show in precise provable mathematical terms that our “equal vote criterion” supports approval voting and *not* RCV.
https://www.equal.vote/equalvote
> But a person who approves 5 candidates has potentially 5 times the influence of a person who only votes for 1 candidate.
This is mathematically false, because you can only vote once *per candidate*. If the frontrunners are Biden and Trump, the most power you can possibly have is to either vote for Biden (and not Trump) or vice versa. In *any* voting method, your vote can have more power if you strategically put more distance between the frontrunners. But approval voting (and almost every other voting method) uses all of the available information on your ballot in the tabulated result. RCV doesn’t. This is extremely rare and weird among voting methods.
> Alan was correct that no election system is without artifacts. At least with RCV we are pretty familiar with what the artifacts are.
As someone who’s spent 16 years in this field, working with people like Andy Jennings (who did his math PhD thesis on voting methods), I can tell you that many of us understand the “artifacts” of the different voting methods extremely well. In 2008, I was in a book called “Gaming the Vote”, in which the author painstakingly analyzed the five commonly discussed alternatives, interviewing their proponents, such as NYU professor of game theory and politics, Steve Brams.
> RCV ignores all preferences other than the voters’ first remaining choice on their ballots. This is a feature, not a bug.
No, it’s a bug. Several experts have shown why, in great detail.
https://clayshentrup.medium.com/later-no-harm-72c44e145510
> It’s why RCV adheres to “one person, one vote”.
No, it’s why RCV *doesn’t* adhere to “one person, one vote”. I just showed an example that proves this. I can refer you to several math PhD’s on this if you want to argue the point. Or I could refer you to several well known books written by experts. See “Mathematics and Democracy” by Steve Brams, for starters.
> “Later no harm” is not a “flaw”, it’s an essential feature.
You mistakenly believe this because you’re not familiar with the science. Here’s a Princeton math PhD demonstrating how later-no-harm doesn’t even mean what most IRV proponents think it means. They confuse it with HSF (honest second favorites) and/or SFC (sincere favorite criterion), which appears to be the error you’re making here.
https://www.rangevoting.org/LNH
> The definition of “later no harm” is that a vote for a lower choice candidate cannot harm a higher choice candidate. Approval voting does not have this property. Without this property, voters will soon learn that if they have any preference at all between candidates, they are better off just voting for their favorite candidate that has a chance to win. Otherwise they risk pushing that candidate into third (or lower) place.
You’re just repeating the same common myth that I debunked at length here, proving that you still don’t understand what later-no-harm means.
https://clayshentrup.medium.com/later-no-harm-72c44e145510
tl;dr
Obviously a voter who prefers the Green but normally votes Democrat *also* approves the Green with approval voting. He does not bullet vote for Green out of fear that the Democrat will win.
Whereas a voter who favored Thomas-Kennedy in the Seattle mayoral race would want to insincerely rank Holmes in 1st place to stop Davison. RCV fails the Sincere Favorite Criterion, which you’re apparently confusing with LNH, as so many people do.
Felix Ling
This is an amazing and well-researched piece. I was very heartened to see that many of the points that I have made myself over the years about the two systems echoed here, as I enjoy reading about political science. I agree 100% that RCV would be a much safer bet because it has real-world history, as social science models in particular are themselves very unproven in that they have not demonstrated an ability to forecast accurately.
I have an intense interest in both methods and had read the Nagel paper on the “Burr-Chicken” dilemma several times until I understood it, so it’s rare that I come across an argument I haven’t seen before, but you make an excellent and insightful point here:
“[T]he top vote-getter is chosen by the largest group of voters. The runner-up is chosen by the second-largest group of voters. And unlike in AV, those groups of voters will always be different people.”
I’ll definitely have to remember that one!
I also applaud you for staying positive about both methods, as it’s too often that I run across something that tears down one or the other, rather much like you’d expect a Democrat or Republican to do with each other. We need more positivity in all political campaigns, whether it be for candidates or electoral systems!
I do have a question. Early in the piece, you argue that bullet-voting is often the most effective way to tactically vote in an Approval election because adding approval to other candidates beyond your favorite can dilute the voter’s support.
Later on, you say, “If white areas of Seattle approve more candidates per ballot than do areas home to more people of color…, would that not violate… voting rights acts? Racially polarized voting plus disproportionate influence for white voters are at the heart of those laws.”
Which is it? Do you get more influence in Approval voting for just one, or for more than one?
Alan Durning
Thanks, Felix. Different tactics might work in different circumstances, and one risk in Seattle in the context of a top-two primary is of a dominant voting bloc all approving the same two candidates. Other scenarios are also possible. It sure would be great to have in-depth empirical analysis of the elections in St. Louis and Fargo, but I’ve not seen anything that’s more than superficial. Unfortunate!
Felix Ling
Ah, right! Thanks for the clarification.
Clay Shentrup
Alan,
This argument doesn’t make sense. This novel “bottoms-up” ranked-choice voting system pits the majority winner against the “minority winner” so to speak— who is then virtually guaranteed to be trounced in the general. (It’s intended as a semi-proportional multi-winner voting method.)
This means that majority voting block effectively picks the winner in the primary.
With approval voting, you instead advance the two strongest candidates, who both have to fight for every vote. Thus at least the minority has a chance to tip the balance in that competitive election.
Shel Kaphan
Clay is ignoring the fact that many more voters vote in general elections than in primaries, and also, many things can come to light between a primary and a general election, so it is not a given that the finalist with fewer votes in the primary would be “trounced” in the general.
Clay Shentrup
Shel,
Nothing is a “given” in terms of absolute certainty, but with very high statistical confidence, the bottoms-up RCV proposal will yield uncompetitive general elections. Again, the *entire point* of this voting method is to be semi-proportional, and FairVote themselves make this point.
https://www.fairvote.org/types_of_rcv_for_multi_seat_bodies
Advancing the two highest approved candidates is obviously better at maximizing the chance that the most popular candidate advances to the general. There is no serious dispute about this.
Shel Kaphan
It’s not at all clear why there would be high statistical confidence that RCV would produce an uncompetitive general election. It could easily be 50.1% vs 49.9% between the two finalists. The assertion is unconvincing.
And I agree that approval voting could yield two popular candidates as finalists … popular with the same subgroup of voters anyway, and therefore not representative and not giving the voters much of a choice. The fact that it could be competitive is of little consequence if the two candidates are very similar and appeal mainly to the same plurality (quite possibly less than a majority) of voters.
John Whitmer
I’ve yet to read a more complete, well-written, perceptive comparison of approval voting and ranked-choice voting than this article (perhaps because none exit). This article leans over backward to be upfront, unbiased, and honest about both methods. I fully agree with Felix Ling’s comments here.
I read Clay Shentrup’s comment in support of approval voting several times and as carefully as I could. Honestly, I really tried to understand his points but could not. Perhaps I’m simply not as smart, sincere or mathematical a thinker as he is. Anything more confusing than RCV may be beyond my cognitive abilities – I’m not a rocket scientist.
Clay Shentrup
Let me suggest another piece then, because I believe that virtually everything in this one is wrong. And a lot of PhD experts do too.
https://electionscience.org/library/approval-voting-versus-irv/
Clay Shentrup
> It’s not at all clear why there would be high statistical confidence that RCV would produce an uncompetitive general election.
You seem to be unaware that this proposal is not for “RCV” (as proponents normally use the term, i.e. STV/IRV), but rather the “bottoms-up” system, which has never been used in the USA.
Bottoms-up does not attempt to find the two most popular candidates, but instead attempts to find diversity. FairVote describes it here as a “semi-proportional” method.
https://www.fairvote.org/types_of_rcv_for_multi_seat_bodies
Diversity is a worthy goal if you’re actually electing multiple candidates to sit on a council, but very bad if you’re ultimately going to choose a single winner.
Here’s a simplified example to demonstrate that.
35% Bruce > Jessyn
33% Lorena
32% Jessyn > Bruce
With these preferences, Bruce and Jessyn are far and away the most popular candidates. Both of them are preferred to Lorena by a crushing 67% landslide majority! But the bottoms-up method eliminates Jessyn with only 32% of the vote, and advances Bruce and Lorena to the general.
Why does this happen? Because, by design, RCV doesn’t look at the 2nd choice preferences of Bruce voters, and thus it doesn’t accurately measure the overall preference of the electorate. Voters feel like they’re casting this precise statement of their ranked preferences, but then the algorithm ignores a lot of that information. It’s very misleading.
More precisely, the lower a voter ranks Bruce, the more influence that voter has over who gets to face Bruce in the general. And this means that the challenger will, by definition, be statistically less preferred and thus more likely to lose the general by a big margin.
> It could easily be 50.1% vs 49.9% between the two finalists. The assertion is unconvincing.
“Could” means it’s possible, not probable.
> And I agree that approval voting could yield two popular candidates as finalists … popular with the same subgroup of voters anyway, and therefore not representative and not giving the voters much of a choice.
You’re apparently making the common fallacy here that the point is to give voters a “choice” (as in, two different kinds of candidates). But that’s wrong. The point is to give voters the most popular and satisfying winner possible. The best way to do that is to put the two strongest candidates in the general, who will of course tend to be highly similar. That’s a benefit, not a flaw. I’ve clearly explained why, using straightforward math.
If you push for giving voters a “choice”, then you’re effectively pushing for uncompetitive elections which deprive the minority of having any chance of tilting the balance of power slightly in their favor.
Bruce vs Lorena doesn’t have to fight very hard for progressive votes, because polling shows he’s going to absolutely crush here.
Bruce vs Jessyn puts some pressure on him. I just picked Jessyn because she’s well known and a good example of someone walking the line between mainstream and progressive, but you could surely think of several other similar candidates from the last Seattle mayoral race.
> The fact that it could be competitive is of little consequence if the two candidates are very similar and appeal mainly to the same plurality (quite possibly less than a majority) of voters.
Nothing could be further from the truth. Statistically, my argument is ironclad.
Shel Kaphan
> You seem to be unaware that this proposal is not for “RCV” (as proponents normally use the term, i.e. STV/IRV), but rather the “bottoms-up” system, which has never been used in the USA.
I described bottoms-up RCV accurately in another comment. I also explained that it’s not STV, which is rarely used in this country. You’re making it sound like bottoms-up RCV is something radically different from standard single winner RCV, when all it is is the exact same process except stopping when there are still two candidates left (for a top two primary) instead of running it down to one.
> Bottoms-up does not attempt to find the two most popular candidates, but instead attempts to find diversity. FairVote describes it here as a “semi-proportional” method.
https://www.fairvote.org/types_of_rcv_for_multi_seat_bodies
It doesn’t “attempt to find diversity”, it attempts to find the candidates with the largest distinct consolidated blocs of support. It’s the “distinct” that differs from what AV will produce, and I think that’s very important.
Full STV would do even better at that, but that’s not on the table.
> Diversity is a worthy goal if you’re actually electing multiple candidates to sit on a council, but very bad if you’re ultimately going to choose a single winner.
Why? The general election voters should have the choice among the two most popular choices among ALL primary voters, not just among the largest plurality. This may be a philosophical point we will just have to disagree on. I don’t believe that the largest plurality, especially if it is only a minority, should be able to completely determine who runs in the general election. That’s a problem that already exists with our top two primaries, but even worse with approval voting.
> […] by design, RCV doesn’t look at the 2nd choice preferences of Bruce voters, and thus it doesn’t accurately measure the overall preference of the electorate. Voters feel like they’re casting this precise statement of their ranked preferences, but then the algorithm ignores a lot of that information. It’s very misleading.
Accurately measuring the overall preferences of voters, i.e., more than one preference per voter — a variable number of preferences depending on how a voter voted, violates one-person-one-vote. You have to count one preference per voter at a time.
This does point out an important piece of voter education that’s necessary: to make sure voters understand that their lower preferences are *back-up* choices in case their higher preferences are eliminated.
> More precisely, the lower a voter ranks Bruce, the more influence that voter has over who gets to face Bruce in the general.
More precisely, if a voter ranks Bruce low, then whoever they gave their first preference to is more likely to be a finalist (and challenge Bruce, assuming Bruce is also a finalist).
> And this means that the challenger will, by definition, be statistically less preferred and thus more likely to lose the general by a big margin.
Clearly the person who ends up with fewer votes is “statistically less preferred” among primary voters, but there’s nothing that says in advance whether and by how much they are less preferred among general election voters. We don’t know until we count votes.
>> It could easily be 50.1% vs 49.9% between the two finalists. The assertion is unconvincing.
> “Could” means it’s possible, not probable.
Right, I don’t want to overstate the case. We don’t know the magnitude (or sign!) of majorities until ballots are actually counted.
>> And I agree that approval voting could yield two popular candidates as finalists … popular with the same subgroup of voters anyway, and therefore not representative and not giving the voters much of a choice.
> You’re apparently making the common fallacy here that the point is to give voters a “choice” (as in, two different kinds of candidates).
Yes, a choice, not between two different “kinds” of candidates, but between the two candidates with the most support among ALL primary voters, not just among the largest plurality.
> But that’s wrong. The point is to give voters the most popular and satisfying winner possible. The best way to do that is to put the two strongest candidates in the general, who will of course tend to be highly similar. That’s a benefit, not a flaw. I’ve clearly explained why, using straightforward math.
Readers can decide for themselves if they agree with this!
> If you push for giving voters a “choice”, then you’re effectively pushing for uncompetitive elections which deprive the minority of having any chance of tilting the balance of power slightly in their favor.
Yes, I think voters should have a choice in the general election between the two candidates with the largest amount of support among ALL primary voters, not just among the largest plurality.
>> The fact that it could be competitive is of little consequence if the two candidates are very similar and appeal mainly to the same plurality (quite possibly less than a majority) of voters.
> Nothing could be further from the truth. Statistically, my argument is ironclad.
Readers can (and hopefully will) decide that for themselves.
TTFN
Clay Shentrup
> Accurately measuring the overall preferences of voters, i.e., more than one preference per voter — a variable number of preferences depending on how a voter voted, violates one-person-one-vote. You have to count one preference per voter at a time.
This is a false claim I’ve already debunked in detail. The term “one person one vote” refers to the weight of votes, not to how votes are expressed. This is obvious when you consider that scores of US cities already use at-large plurality voting elections, where some voters might vote for 3 candidates, while other voters only vote for 1 or 2 candidates. These votes are all counted at the same time.
The U.S. Supreme Court made the “one person one vote” rule explicit in Reynolds v. Sims (377 U.S. 533). The rule stated that no vote should count more than any other so that it has unequal weight. This unequal weight would violate the Equal Protection Clause of the Constitution. And it was Baker v. Carr (369 U.S. 186) that extended the Equal Protection Clause to districting issues. In Reynolds, the state of Alabama set up its districts so that they varied wildly in population. The districting was so bad that it gave some voters’ ballots as much as 41 times more weight than others. Because the weights of the ballots were different between districts, that violated the “one person one vote” rule.
A common misconception is that approval voting gives more weight to voters who vote for more candidates. To see why this isn’t the case, imagine a tied election between a liberal and two conservatives. Bob casts a vote for the liberal, while Alice casts an opposing vote for the two conservatives. After Bob and Alice have voted, the election is still tied. Bob and Alice have an opposite but equal effect on the election. Another way to think of it is that if you vote for all candidates, that has the same effect as not voting at all. The key here is that no voter has an unfair advantage. Effectively, every voter casts a “yes” or “no” vote for every candidate.
Finally, consider that voters are already allowed to vote for multiple candidates in “at large” races. For instance, a city council may simultaneously elect three representatives. Some voters may vote for three candidates, while others may vote for only one or two candidates.
> I described bottoms-up RCV accurately in another comment.
But *inaccurately* in this comment:
>> “Bottoms-up RCV”, as would be used in Seattle to elect the two finalists in primaries, works exactly the same as the usual single-winner RCV, except that it stops one round sooner.
> I also explained that it’s not STV, which is rarely used in this country.
Incorrect. Single-winner STV (aka IRV) is used in dozens of US cities. The *proportional multi-winner* form of STV is rare in the US. All IRV is STV, just like all squares are rectangles.
> You’re making it sound like bottoms-up RCV is something radically different from standard single winner RCV, when all it is is the exact same process except stopping when there are still two candidates left (for a top two primary) instead of running it down to one.
I.e. “it’s the same process, except continuing to eliminate candidates even after a majority winner has been found, until there are only two left”. I’ve explained in great detail why this is so problematic. Even more examples here:
https://clayshentrup.medium.com/ranked-choice-voting-wrong-for-seattle-e7e08349717e
If you wanted to make this process reasonable, you could use “preferential block voting”, where you first find the normal RCV winner, then eliminate them, then repeat a second time. Then you’d at least have the two generally strongest candidates in the general, and the minority would have a hope of having meaningful votes that count for something.
https://en.wikipedia.org/wiki/Preferential_block_voting
Bottoms-up is a semi-proportional method, thus completely inappropriate for this purpose.
> It doesn’t “attempt to find diversity”, it attempts to find the candidates with the largest distinct consolidated blocs of support.
Finding candidates who represent distinct blocs of support is literally the same thing as “finding diversity”.
Except you’ve made an error here, because bottoms-up *does not* attempt to find the “two largest” distinct consolidated blocs of support. That would be what two-winner PR-STV does. Bottoms-up instead finds the candidate (generally) most representative of *the entire electorate* (not a 50% bloc), and *then* finds the generally strongest candidate among those who didn’t like the strongest candidate. You’ve mischaracterized the effect of bottoms-up multiple times now.
> It’s the “distinct” that differs from what AV will produce, and I think that’s very important.
On the contrary, it’s very problematic, as I meticulously explained. Because it produces uncompetitive general elections, by design. Bottoms-up is intended to be a semi-proportional *multi-winner* method, not a single-winner method.
> Full STV would do even better at that, but that’s not on the table.
By “better”, you mean “better for multi-winner elections”, which means *worse* for this specific purpose of advancing the two candidates most likely to be the most popular single winner.
You can go on all day pretending this is a multi-winner election, but it’s not. There’s only a single winner in the end.
> Diversity is a worthy goal if you’re actually electing multiple candidates to sit on a council, but very bad if you’re ultimately going to choose a single winner.
Why?
I just explained why, in great detail. Here’s a blog post I wrote on it.
https://clayshentrup.medium.com/ranked-choice-voting-wrong-for-seattle-e7e08349717e
> The general election voters should have the choice among the two most popular choices among ALL primary voters, not just among the largest plurality.
Incorrect. They should have a choice between the two candidates most likely to be the most popular among *the entire electorate*. There is only going to be one winner, not two.
And it’s deeply misleading to use the term “largest plurality” when talking about approval voting, because the winner(s) tends to be the most broadly appealing “majority winner”, regardless of whether the approval votes total to a majority.
https://www.rangevoting.org/AppCW
> This may be a philosophical point we will just have to disagree on.
Objective facts don’t allow for legitimate disagreement. The best way to ensure the most popular candidate wins is to ensure the finalists are the two candidates most likely to be the most popular *among the entire electorate*, not among specific blocs of voters. You’re fundamentally applying a principle that makes sense in multi-winner elections, to a single-winner election. This is a mistake, pure and simple.
> I don’t believe that the largest plurality, especially if it is only a minority, should be able to completely determine who runs in the general election.
Again, it’s misleading to say “plurality”. Approval voting tends to find the most popular candidate(s) among *the entire electorate*, not among a “plurality”.
> That’s a problem that already exists with our top two primaries, but even worse with approval voting.
No, the problem with the existing primaries is vote splitting, which prevents the two most popular candidates from advancing. Approval voting fixes that.
> More precisely, if a voter ranks Bruce low, then whoever they gave their first preference to is more likely to be a finalist (and challenge Bruce, assuming Bruce is also a finalist).
This is no way refutes or even contradicts my completely correct point that, the lower a voter ranks Bruce, the more influence they have on who Bruce’s opponent is. Or, to say it in reverse, the higher a voter ranks Bruce, the less influence they have on who his opponent is. And this is what gives you the semi-proportional outcome problem, leading to uncompetitive elections that offer the minority virtually no influence on the final outcome.
> Clearly the person who ends up with fewer votes is “statistically less preferred” among primary voters, but there’s nothing that says in advance whether and by how much they are less preferred among general election voters. We don’t know until we count votes.
Yes there is something that says this in advance. I’ve thoroughly described it. It’s “the semi-proportional design of bottoms-up, caused by intentional vote-splitting”. You’re just ignoring the very clear and thorough explanation I gave.
> Right, I don’t want to overstate the case. We don’t know the magnitude (or sign!) of majorities until ballots are actually counted.
But we know general PROBABILITIES that result from the design of the electoral system, well in advance of actually using the system in a real election. We KNOW that bottoms-up is semi-proportional, and attempts to put a weaker candidate in the runner-up spot. And I don’t just mean weaker in the sense that you get the 1st and 2nd strongest candidates in the general. I mean weaker in the sense that you can easily get the 1st and 5th most popular. It can be arbitrarily bad depending on how many candidates run and split the vote.
You don’t use a proportional or even semi-proportional voting method to choose finalists in a *single-winner election*.
>> You’re apparently making the common fallacy here that the point is to give voters a “choice” (as in, two different kinds of candidates).
> Yes, a choice, not between two different “kinds” of candidates, but between the two candidates with the most support among ALL primary voters, not just among the largest plurality.
You’re misusing the term “plurality” here. If we instead use the more appropriate term, “voting bloc”, then this sentence makes sense. And then it’s approval voting that sends the two most popular candidates among the entire electorate, while bottoms-up chooses as the runner-up the most popular candidate among the minority voting bloc (or “the plurality of voters who don’t like the stronger candidate”).
So you’re stating a fundamentally correct criterion, but you’ve gotten mixed (reversed) up about which system better satisfies it.
>> But that’s wrong. The point is to give voters the most popular and satisfying winner possible. The best way to do that is to put the two strongest candidates in the general, who will of course tend to be highly similar. That’s a benefit, not a flaw. I’ve clearly explained why, using straightforward math.
> Readers can decide for themselves if they agree with this!
It’s just an objective statistical fact.
> Yes, I think voters should have a choice in the general election between the two candidates with the largest amount of support among ALL primary voters, not just among the largest plurality.
Then you should support approval voting, because the *entire purpose* of bottoms-up is to make the runner-up be the candidate with the “largest plurality” (bloc) of support from among those voters not happy with the frontrunner.
You could easily write a computer simulation over a weekend to run mock elections and show this, using any kind of generally realistic utility distribution function.
Clay Shentrup
John Whitmer,
I’m sorry you found my comments complicated. Approval voting is much simpler than ranked choice voting. But the science of *evaluating* voting methods is complicated, yes. This is often the case in scientific fields.
For instance, there may be two different diets that are very simple to follow, but the science of evaluating which is better for overall health could involve all kinds of complex chemistry.
Approval voting keeps everything the same, except you can vote for more than one candidate. Vote for as many as you like. Voters in Fargo and St Louis have used it with no signs of confusion.
Angry
“Unproven” Approval is better than the proven failure that is Ranked Choice Voting, which:
1. Doesn’t make it safe to vote honestly for your favorite, making third parties non-viable.
2. Is strongly biased in favor of polarizing candidates, and against those who actually represent the will of the voters, perpetuating a two-party system.
3. Doesn’t guarantee that the candidate preferred by the majority is elected.
RCV has had over a century to live up to its promises, and still hasn’t delivered. Why anyone would continue to advocate it when there are dozens of better alternatives boggles my mind.
For it to be slapped onto the same ballot as Approval, without having to go through any of the scrutiny or vetting, and increasing voter confusion in the process, is infuriating.
Clay Shentrup
>> Computer simulations using Bayesian regret calculations . . . demonstrate better utility outcomes in elections using approval voting versus RCV even if all approval voters were tactical and all RCV voters were honest.
> Um. What?
The author cites a brief reference to Bayesian regret, one of the most important concepts in “Voting Theory 101”, and then responds, “Um. What?”
Holy cow. This would be like writing a Sightline article on housing and not knowing what “setbacks” or “FAR” or “zoning” means. This is just incredibly inappropriate lack of basic familiarity with the subject matter.
Jason Osgood
https://ncase.me/ballot/ is a nice interactive simulator for getting some intuition for how various systems of voting, especially edge cases.
TLDR: Approval Voting is the best balance between simplicity and fairness. https://ncase.me/ballot/img/comparison.png
Complexity is the enemy of election integrity.
As for minority representation, the root problem is single seat contests, not the voting system used.
The correct solution is proportional representation. Probably using STV as the voting system.
RossB
I agree Jason.
Debra J Morrison
Thanks to all of you for a detailed conversation about this important topic.
Clearly there is much still to learn about how AV works in real-life city elections, but to label it “risky” in a city that is fairly homogenous in political perspectives seems unnecessarily provocative. Perhaps Seattle primary voting would instead be an excellent opportunity to try it out so we can learn more, as Alan Durning suggests we need to, while we all work together to keep trying to change the laws that would allow more election methods to be tested in our state and to be able to use RCV as it is intended, as well as a variety of methods for proportional representation.
Strong points, Clay Shentrup, and fascinating back and forth with Alan and Shel. Quite a bit above my comprehension level, and I’m probably just a hair above average in understanding this stuff, so I appreciate the rigor, but can’t argue any of those points while talking to voters. And there’s the biggest selling point for AV that I can think of. Voters get how to use it, even if they don’t take advantage of it, and the counting part is EASY to explain and transparent to display after the election. These things count A LOT these days, as Jason Osgood mentions, so I will just reiterate his point.
>>Approval Voting is the best balance between simplicity and fairness.
Let’s hope one of these much better, albeit imperfect, methods gets adopted so we don’t suffer a setback to all of our electoral reform efforts in November.
RossB
With all due respect Alan, I think you have it wrong. But lets start with the basics:
1) There is no perfect electoral system. Experts who study the issue have different favorites. In one poll, approval voting was the favorite of the experts for their town, but there wasn’t a majority who supported it. https://en.wikipedia.org/wiki/Comparison_of_electoral_systems.
2) Ranked choice voting comes in many forms. Instant-runoff voting (IRV) is only one type. It only effects those who choose outlier candidates.
3) We have a non-partisan blanket primary. This means that the top two candidates advance, regardless of party.
4) The new system will only be applied in the primary.
This leads to several interesting results:
5) IRV is meaningless in a three person race.
6) In contrast, approval voting could change who advances in a three person race, ultimately leading to a consensus candidate.
To show why, consider a real-world example. The last election for Seattle City Attorney was a closely fought race — https://ballotpedia.org/City_attorney_election_in_Seattle,_Washington_(August_3,_2021,_top-two_primary). Consider what would have happened with the two systems:
With IRV, nothing would have changed. The top three got 99.7% of the vote. Applying votes for the other candidates wouldn’t have changed the outcome.
In contrast, it is quite easy to see how approval voting would. By Seattle’s standards, Nicole Thomas-Kennedy was a far left candidate while Ann Davison was a far right candidate. It would have been pointless to vote for all three in the primary. It is also highly unlikely that someone would approve of both of the extreme candidates. Thus it is quite likely that Holmes would have advanced. It is also quite likely that Holmes would have beaten either candidate. He was, essentially, the consensus candidate — the “B” in this diagram: https://en.wikipedia.org/wiki/Comparison_of_electoral_systems#/media/File:IRVCopeland.png.
There are other systems where Holmes would have advanced. These include systems that have ranked choice voting (e. g. Copeland). But instant-runoff ranked-choice voting is meaningless in this case.
In this scenario, it is instant-runoff that “is too basic to represent a voter’s preferences”. Sure, they put down their preferences, but that preference is ignored. It doesn’t matter if Holmes was every voter’s second choice. In contrast, with approval voting, it clearly matters. Unless you marked only one choice, your second choice is very important.
There are other recent examples where the same thing occurred. For example, in city council position number 9 (https://ballotpedia.org/City_council_elections_in_Seattle,_Washington_(2021)). Instant-runoff would not have changed anything, but it is quite possible the consensus candidate (Brianna Thomas) would have advanced, and beat either candidate in a head-to-head race.
There are basically three situations to consider:
1) Both systems lead to the same outcome.
2) Approval voting leads to different candidates advancing.
3) Instant-Runoff leads to different candidates advancing.
In both 2 and 3 I assume that this is a good thing. This then begs the question: Which is more likely to happen, *under our system*. The answer, clearly, is 2. For *our system*, instant-runoff will rarely make a difference, and when it does, we would likely get exactly the same outcome with approval voting. In contrast, with approval voting, there are races that would change, and change for the better.
Alan Durning
Thanks for your comment, RossB.
If you narrow the scope down enough, you can find a hypothetical case where your preferred voting system will do something than another alternative does not. And you’re right that all else being equal, in a three-person race with two winners, the particular type of RCV proposed in Seattle will behave as does pick-one voting. Approval Voting might behave differently, precisely because it allows people to vote for more than one candidate at a time: that’s its strength and that’s the idiosyncrasy that makes it interesting, paradoxical, and—some would insist–problematic.
That said, are three-candidate races the problem Seattle is trying to solve? There were 15 candidates for mayor in Seattle last year and 18 candidates for the two at-large city council seats.
Moreover, the thing about changed voting systems is that they aim to change the dynamic of the races, so arguing about past cases and trying to guess what would have happened is a fool’s errand. In the city attorney’s race you cite, maybe Pete Holmes would have campaigned harder or differently. Maybe other candidates would have run. Maybe voters would have behaved differently. Maybe people would have approved both of Pete’s challengers: remember, he was extremely unpopular on both the left (BLM) and the right (homeless camps) during the campaign. We cannot know.
Indeed—and here’s the real point—the entire thrust of my article was to argue for wariness of theoretical and hypothetical arguments. We should seek empirical evidence to see what happens in the heat of actual campaigns with real voters and real candidates, not speculate about what might happen or might have happened.
And in the two places that have used AV, we’ve seen (as I cited above) more than 90 percent of voters only approve one candidate in three-candidate races, even in two-winner races. A careful study of those races, complete with independent polling, would be a great thing to have. The only such study I have found (linked in the piece) showed a lot of people voting strategically, aka, disingenuously. There was no evidence presented there that AV yielded moderates or consensus candidates. Maybe it did, but we just don’t know.
To repeat: we need a lot more study to really know how AV actually works. We need ballot-by-ballot, precinct-by-precinct studies of voter behavior. We need interviews with candidates and campaign experts, with potential candidates and election administrators. We need, in short, the kinds of studies that scholars have been conducting about RCV for years now.
Finally, implicit in your argument, as in many arguments for approval voting and other so-called utilitarian voting systems, is the idea that all of a voter’s preferences ought to count, not just their top live choice. If a whole lot of people like Jessyn almost as much as Bruce, AV holds that their second choice preferences for Jessyn should, morally speaking, weigh in the result, along with their first-choice preference for Bruce. This is a value judgment, not a mathematical syllogism: it’s a moral claim that all your preferences count. In conventional voting (including RCV), though, you only can vote for one candidate at a time. If you’ve cast your vote, your second choice is irrelevant. That’s a different value judgment, and it’s the one most people—and US courts—have so far proved comfortable with: one person, one vote. No more and no less.
From this divergence in value judgments comes the deep concern that critics of AV have when they point out that using it in a top-two primary might let the same bloc of voters pick both the finalists.
RossB
My previous comment was long, and didn’t really address the arguments here. This will be more of a rebuttal:
1) “Approval voting is too basic to represent a voter’s preferences.” You could say the same thing about instant-runoff. In a three person race, it doesn’t matter if every single voter has the same second-choice. All of that is ignored. You might as well have the current system. In contrast, with approval voting, the second choice matters a lot.
2) “Approval voting is new and unproven.” Again you could say the same thing about what is being proposed. Ranked choice instant-runoff (RV-IRV) is most commonly used in general elections, or party nominations. I believe Alaska is the only place in the country where they use it as part of a nonpartisan blanket primary. Even Alaska’s system is different than ours, as ours leads to two candidates facing a head-to-head contest, whereas Alaska’s system winnows to four candidates, who then face each other again in a RV-IRV. It is quite possible we would be the only city in the nation with the proposed system.
3) “Approval voting is legally untested and might worsen representation.” The arguments against the legality of approval voting also apply to RV-IRV. As for representation, I can’t think of a single race that would have been different with RV-IRV. Yeah, sure, you get to rank candidates. But at the end of the day, the result is the same. The problem is not with ranked voting, it is with instant-runoff. We already have a nonpartisan blanket primary, which serves as a runoff. At best there is a runoff for a runoff. Most often you get the exact same results as you do now, and on the rare occasion where there is a difference (e. g. the vote being split among four candidates) you are likely to get the same results with approval voting. Meanwhile, there are many cases — including two from the last election — where the results would have been different, and more in line with the way people feel in this city. Twice we nominated an extreme left candidate, and twice they lost to a moderate right wing candidate. In both cases, the moderate progressive would have easily beat either candidate in a head-to-head race.
4) “AV is risky; RCV is tested and delivering the benefits voters want.” It is not “risky” to try and create a better electoral system. RCV, meanwhile, would rarely effect the outcome. It is a worthwhile system, but not with a nonpartisan blanket primary. Most of the time, it doesn’t make any difference at all. Thus it become political theater. People get to pick their favorites, but most of the time, nothing changes. This also makes it *more* difficult to enact meaningful change.
Ken Dammand
Voting is the most common way that citizens express their opinions about the direction they want government to go. Therefore, the more information a voting system allows the voter to put into the system, the better the results will reflect the will of the public. Under the current “first past the post” voting system, the existence of polling data has resulted in voters often choosing a candidate other than their true preference in order to avoid weakening support for a candidate they would reluctantly approve and electing a candidate of which they disapprove, i.e. the spoiler effect. This is what so commonly results in voters “holding their nose” while voting. Voters shouldn’t have to do this.
Both “approval voting” and RCV are responses to this flaw in our current voting system. An important difference between the two proposals is the amount of information they allow a voter to impart to the decision process. Approval voting is blind to the order of preference the voter holds for the array of candidates the voter “approves.” The voter’s least preferred but still acceptable candidate is indistinguishable from the voter’s most preferred candidate. RCV allows the voter to tell the system not only which candidates would be acceptable but also to convey their actual preference among those candidates. The more information the voter is able to express and which actually becomes part of choosing the ultimate winner, the better the system will do at reflecting the sentiments of the electorate. It’s that simple. RCV is a better choice.
Ben Chapman
Really really appreciate this article! For me, the crux of why I don’t view AV as a very positive change is that voters just don’t seem to use it. A voting reform that doesn’t change incentives or voter behavior isn’t one that’s going to reform much. This paragraph explained it well.
——-
Fargo and St. Louis voters listened, or they figured out AV’s internal logic for themselves. Most of them seem to have bullet voted. Elections administrators in neither city have released data on approval votes per ballot, but estimates are possible (see appendix). Apparently, between 50 and 90 percent of ballots were bullet voted, depending on the election and the race. In contrast, in ranked choice voting, between 10 and 40 percent of voters tend to bullet vote, ranking only a first choice.
—–
Thanks for writing, Alan!
Eric Bidstrup
The article was a thoughtful and well researched read that tried to provide some level of balance between the two new options. Probably took me longer to read the comments than the full article. A few thoughts to offer in response:
1) The lack of AV allowing voters to express different levels of support for different candidates seems to not reflect the way most people think in the real world. In any election I’ve ever voted in, I have a (usually strong) point of view on which candidate I like the most, 2nd most, etc. I think most people have a similar POV in how they think, voting methods notwithstanding. To me, this is the dealbreaker.
2) The arguments from Shentrup that “bottoms up IRV/RCV” is unproven in the context of top two primaries is literally correct strictly speaking but isn’t convincing to me that AV (only used in 3 actual elections so far) would produce different outcomes in real world primaries for city offices. The bottom line is that with either method, the two candidates with the most popularity will advance. (yes, I did read the links provided arguing against that but didn’t find them convincing). To me, that “bottoms up” line of thought pales compared against the larger number of cities using IRV/RCV vs. AV in terms of being unproven.
3) The comment from “Angry” on “scrutiny or vetting” was entertaining. Let’s be real here: Seattle Approves (to their credit) took mainly out of state money to fund paid signature gathering to get their initiative on the ballot. Full respect to them for making that happen, but not really what most people would call “scrutiny or vetting”. Especially considering in the dozen or so friends and family I know who were approached to sign the petition found the signature gatherers claimed AV was “just like RCV” (literally said that).
I’ll take either AV or RCV over the current system, but for me I think RCV is the stronger choice.
John Whitmer
Wonderful discussion here. Wonderful comments. Getting folks to think about (perhaps even rethink their position on?) an issue is the hallmark of a good article.
After rereading the article as well as the many excellent comments here from knowledgeable people, it is clear there are a number of honestly arguable pros and cons concerning ranked-choice voting, approval voting, and currently used plurality voting. Leaving aside the arguable points, there are several points – most mentioned either in the article or by commenters here – that IMHO are undeniable.
First, all three methods are straightforward. None are difficult to understand.
Second, plurality and ranked-choice voting have – for better or worse – long well-documented track records. Neither are new or untested. Approval voting is not as good by this metric.
Third, as a means of reflecting the collective will of voters reliably and consistently plurality voting is a distant third among the three. Both RCV and AV increase the influence of voters and reduce the influence of political parties especially their extremes.
Finally, (and this point is more about RCV than the other two) the value of RCV is not about changing who is elected. (I think in well over 90% of RCV elections the candidate with the most first-round, first-place votes eventually wins.) The value of RCV is in changing incentives: changing perhaps who is temped to run, who actually decides to run, how candidates behave on the campaign trail, how campaigns are conducted, how elected candidates behave in office. RCV, in my opinion, has a better chance of encouraging these behaviors in the right direction than does approval voting. Alas, plurality voting largely encourages these behaviors in the wrong direction.
No election method is flawless but some are better than others. My bottom line now (always subject to change) would rank-order these methods: RCV, then AV, and then plurality a distant third.
.