Fixing Final Five

17 min readNov 25, 2021

Basic question: What voting method does most of the United States use for electing members of Congress? The immediate answer is “Plurality” (or “First Past the Post”) — but that’s not quite right. What this immediate answer leaves out is the role of primaries. Having every candidate compete in a single election with one winner is very different from having the candidates compete in partisan primaries followed by a general election, regardless of the voting methods being used in these individual elections. However, most people in the voting methods community don’t focus much on the role of primaries.

The people behind Final Five Voting (FFV) are the big exception. FFV would replace existing systems of primary and general elections with this:

First, all candidates compete in a single primary using Plurality; voters can only vote for a single candidate in the primary. All voters can participate regardless of partisan affiliation.
The top five voter-getters in this primary advance to a general election using Instant Runoff Voting (IRV), more commonly (but less precisely) known as single-winner Ranked Choice Voting.

The main goals of FFV are reducing polarization, making politicians in currently non-competitive districts accountable to everyone they represent (instead of just primary voters), and ending gridlock by incentivizing politicians to reach across the aisle and compromise. Here’s the argument for why FFV might achieve these goals:

Most congressional districts are either safe Republican seats or safe Democratic seats, so whichever candidate wins the dominant party’s primary election is essentially assured of winning the general election. In these districts, the primary is everything; the general is a mere formality.
When the primary is everything, candidates need only care about appealing to the primary electorate. The typical voter in a Democratic primary will be less happy with compromising with Republicans than a typical voter in the general election, so stonewalling and brinksmanship are rewarded far more than they would be if candidates were equally incentivized to appeal to all voters.
FFV uses a nonpartisan open primary (where any voter can vote for any candidate, regardless of party), and it makes the primary less important by allowing five candidates to advance instead of just one or two. With Final Five Voting, the primary will never be everything.
Since the general election doesn’t use Plurality, candidates will be incentivized to compete for second and third choice votes by appealing to voters who might never rank them #1.

This argument has both a lot that’s right with it and a lot that’s wrong with it. Let’s start by taking a closer look at primaries in general.

A Primer on Primaries

Here I will summarize some key findings in Lee Drutman’s comprehensive report, What We Know about Congressional Primaries and Congressional Primary Reform. This report is excellent, and I encourage anyone who is interested in learning more about primary reform and how primaries affect polarization to read it.

There are several types of primary elections:

Closed: Only voters registered with a political party can vote in a primary, and only in the primary election of that party.
Open: Each party has a separate primary, but voters can vote in the primary of their choice regardless of their partisan affiliation or lack thereof. (There are also intermediate possibilities between open and closed.)
Top Two: There is only one primary election (using Plurality voting), in which all candidates from all parties participate and all voters can vote. The top two advance to the general election regardless of party, so the general election may have two Democrats, two Republicans, or a Democrat and a Republican.
Approval + Top Two: Just like Plurality + Top Two, but with approval voting instead. Currently, this is only used for municipal elections in St. Louis.
Alaska’s system: Alaska recently adopted FFV, except it’s Final Four instead of Final Five. 2022 will be the first year in which it’s used.

In comparing these methods and looking to improve upon them, there are several facts that need to be understood.

General election voters are about as polarized as primary voters. There is some difference between (say) Republican primary voters and Republican general election-only voters, with the primary voters being further to the right, but it’s tiny next to the difference between Democrats and Republicans. Also, independents generally aren’t moderates, so the idea that open primaries will reduce polarization by bringing in a bunch of moderate independent voters is completely wrong:

In short, if the problem is the voters demanding extreme positions and rejecting compromises, the problem is not limited to registered partisans who vote in a partisan primary. Many independents and moderates are even worse. Notably, Republican primary voters who registered as independents were significantly more likely to support Trump than registered Republican voters in the 2016 primary, and Democratic voters who registered as independents were significantly more likely to support Sen. Bernie Sanders (I-Vt.) than registered Democrats. And according to regression models from an article by Joshua J. Dyck, Shanna Pearson-Merkowitz, and Michael Coates, the effect of registering as an independent was three times greater than a voter’s ideology in predicting support for Trump or Sanders.

Incumbents fear a primary challenge, and this pushes them to the extremes. Primary challenges from extremists are more common than challenges from moderates, and the extremists tend to have an easier time with fundraising. Legislators believe that primary voters are more likely to punish them for compromising than donors or general election voters.

Current reforms don’t appear to help. Reforms will not work by changing the kinds of voters who participate in primaries. Furthermore, it does not appear that the type of primaries has an effect either on who runs or who wins. It is possible that the candidates elected for the first time under Top Two are somewhat more moderate, but the evidence is weak. Lee Drutman quotes what is probably the most promising study:

However, one intriguing data point from Grose’s study is that “among new members of Congress, those elected in top-two primaries are more than 18 percentage points less extreme than closed primary legislators.” (my [lack of] italics) It is possible that reforms will produce more of a moderating effect through member replacement, in which case they will take time. As always, reform is part of a dynamic process.

Nonetheless, this study is an outlier. Another reason for hope comes from the finding that, in general elections with Top Two, the campaign rhetoric is more bipartisan when both candidates are from the same party. But it’s difficult to have both candidates be from the same party outside of districts in which one party is extremely dominant, so this only happens about 1/6th of the time.

Voters have a hard time distinguishing between the other party’s candidates.

Another possibility is that, even within the top-two contests, levels of cross-over voting are lower than expected. Remember, a key expectation of the top-two open primary is that Republican voters will vote for the more moderate Democrat if two Democrats compete in the general election, and vice versa. But this depends on orphaned Republicans bothering to vote in a general election in which the choice is between two Democrats. For many orphaned Republicans, a choice between two Democrats is a choice between two equally bad options; hence, better to abstain. It is the same for Democrats choosing between two Republicans. One study found that almost half of the orphaned voters abstained in a general election. Another found more than 40 percent abstention among orphaned voters, and concluded that voters of the opposing party had a hard time telling the difference between the ideology of opposite party candidates. (That is, Democrats had a hard time telling the difference between moderate and extreme Republicans, and vice versa; to most Democrats, a Republican is just a Republican, and vice versa.)

A closely related point is that many voters rely on partisan cues (i.e. whether a candidate is running as a Republican or a Democrat) in order to assess them. Absent such information (as happens when both finalists in a Top Two election are from the same party), voters have a more difficult time.

Much of the problem has nothing to do with voting methods. In general, extremists tend to have a greater desire to run for office and an easier time fundraising. Furthermore, political elites seem to be somewhat effective at deterring primary challenges that would create a same-party general election under Top Two. Even if primary reform coupled with better voting methods can help, such reforms will still be powerless against other factors.

If you want more than a superficial summary of any of these points, read the report.

Why Final Five will fail to reward bipartisanship

For a first look at whether FFV helps moderates, let’s look at an election that saw an extremist get elected: Lauren Boebert. Here are the results in the general election:

Lauren Boebert (R): 51.4%
Diane Mitsch Bush (D): 45.2%
John Keil (Libertarian): 2.4%
Critter Milton (Unity Party): 1.0%

The Republican primary:

Lauren Boebert: 54.6%
Scott Tipton: 45.4%

And the Democratic primary:

Diane Mitsch Bush: 61.3%
James Iacino: 38.7%

Assuming a substantial majority of Democratic voters would have preferred Tipton to Boebert, it seems he ought to defeat her quite handily head-to-head. (While this may appear at odds with the finding that Democratic voters can have a hard time distinguishing between Republican voters, I’m going to go out on a limb and say that Lauren Boebert is an exception.) How would this election have gone under Final Five? I will make the following assumptions for simplicity:

Both Democrats, both Republicans, and the Libertarian advance to the general election after a rather boring primary.
The fraction of voters who voted for Boebert in the actual general election will equal the fraction of voters who rank either Boebert or Tipton first, and likewise with the Democrats. That is, I will assume the partisan makeup of the general election is unchanged. (This is arguably a bit too favorable for the Democrats since it’s possible that some Tipton voters in the primary voted for Bush in the general.)
The ratio of Boebert voters to Tipton voters will equal the ratio observed in the primary (and similarly with the Democrats). This assumption is supported by the research showing that primary and general electorates are about equally extreme.
All Republican voters will rank both Republicans above the Democrats and vice versa. This assumption is definitely wrong and ignores real-world factors like exhausted ballots, but the deviation from realism shouldn’t be great enough to affect the outcome.
All third-party voters will rank the Libertarian first, Tipton second, and nobody third. (This assumption is designed to be maximally conservative.)

With these assumptions, here are the percentages in the first round:

Lauren Boebert (R): 28.1%
Scott Tipton (R): 23.3%
Diane Mitsch Bush (D): 27.7%
James Iacino (D): 17.5%
John Keil (Libertarian): 3.4%

Keil is eliminated, and we assume his support goes to Tipton in the second round (though, realistically, a lot would also go to Boebert and the Democrats).

Lauren Boebert (R): 28.1%
Scott Tipton (R): 26.7%
Diane Mitsch Bush (D): 27.7%
James Iacino (D): 17.5%

Iacino is eliminated, and his support goes to Bush.

Lauren Boebert (R): 28.1%
Scott Tipton (R): 26.7%
Diane Mitsch Bush (D): 45.2%

Tipton is eliminated, leaving Boebert to win the final round:

Lauren Boebert (R): 51.4%
Diane Mitsch Bush (D): 45.2%
Exhausted: 3.4%

This…isn’t any different from what happened in the actual election using partisan primaries. Under the assumptions given (excluding the one about Independents all preferring Tipton), voters who voted for a Democratic or third-party candidate would have only needed to prefer Tipton over Boebert by 10 percentage points for him to defeat her in a hypothetical head-to-head matchup — but this doesn’t matter. What went wrong? Basically, IRV mimics partisan primaries. One Democrat and one Republican made it to the final round, so at no point did the Democratic voters who preferred Tipton to Boebert get their votes transferred to him. Hypothetically, Tipton could have made it to the final round instead of Boebert if a large fraction of the Democrats who voted for Iacino preferred him to Bush, but the second choices of the Bush voters would be irrelevant in any case.

We can see a similar scenario playing out in Alaska. From FiveThirtyEight:

Even in a high-profile contest like Alaska’s 2022 Senate race, the top-four system won’t necessarily help an incumbent like Sen. Lisa Murkowski, a moderate Republican who has actively defied Trump in a state he won by 10 points in 2020. While other major contenders could still enter the race and Murkowski hasn’t officially announced her reelection bid, it looks increasingly like she will face one other notable Republican running to her right: former commissioner of Alaska’s Department of Administration Kelly Tshibaka. And Murkowski’s moderation could actually hurt her because it has significantly eroded her standing in the Alaska GOP in what is, remember, a fairly red state. The state party, for instance, has censured her for voting to convict Trump in his second impeachment trial and then endorsed Tshibaka, who also earned Trump’s coveted endorsement. Murkowski is no stranger to hard-fought races, though. After losing renomination in the GOP primary in 2010, she won reelection as a write-in candidate in the general election, thanks to her ability to appeal to broad swaths of the state’s electorate, such as Alaskan Natives and some Democrats. Yet her anti-Trump bona fides could make it more difficult for her to win this time around, as she could struggle to hold on to a significant chunk of the GOP base, which may be necessary to win.
Murkowski should still be able to advance to November in the top-four primary, but she could be in trouble if something like the following scenario plays out in the ranked-choice voting process: In the first-choice vote, Tshibaka wins a majority of Republican voters and Democrats have a high-profile candidate who they largely back instead of Murkowski. In this situation, Murkowski could easily find herself in third place among the first-choice votes. So even if she’s the preferred second-choice candidate for most of the voters who backed the fourth-place candidate, she might still be in third after those votes are reallocated, which would mean game over. In other words, even if Murkowski were the preferred option for the state’s electorate in a head-to-head matchup with Tshibaka, that wouldn’t matter if she never got into a position to find out. So contrary to reformers’ expectations, a top-four primary might not be the ticket to victory for more moderate candidates either.

These examples are hardly exceptional. The common theme holds for a wide range of elections: should a Republican win, voters who prefer every Democrat to the Republicans will have no say in which Republican will win, and, should a Democrat win, voters who prefer every Republican to the Democrats will have no say in which Democrat will win. This means that Republican candidates will have no incentive to appeal to Democratic voters who will rank the Democratic candidates first no matter what and vice-versa: Final Five reproduces the incentives of partisan primaries.

There are exceptions where Final Five can outperform partisan primaries: districts that are so lopsided that all the candidates in the minority party will be eliminated before the final round. (This requires a district to be at least 2/3 Republican or 2/3 Democratic.) An example is the district that elected Marjorie Taylor Greene. Had Final Five been used there she would have faced a Republican in the final round, so there would have been some incentive to appeal to Democratic voters.

Optimizing Final Five’s general election

Partisan primaries are a problem, and the currently proposed version of FFV effectively mimics them in the general election. To fix these problems and create a system that properly rewards politicians for bipartisanship and compromise, let’s look at what we have to overcome.

First, Independents and voters who only participate in the general election are about as polarized as the registered Democrats/Republicans who participate in closed primaries. Currently-excluded groups aren’t going to save us, no matter how good of a job we do at bringing them into the democratic process. If you ask a hundred registered Democrats whether Democratic legislators should be more willing to compromise with Republicans, you’ll probably get a lot of people saying yes and a lot of people saying no, and if you ask the same question of Independents who vote for Democrats the responses will be quantitatively similar. However, you’ll find near-unanimity if you instead ask these Democratic voters whether Republicans should be more willing to compromise. If we want to reward compromise, we need to give Democrats the power to favor some Republicans over others and vice versa.

Second, we need to help Democrats distinguish between Republicans and Republicans distinguish between Democrats. Lee Drutman suggests fusion balloting as a possible solution, and I agree that this is a good idea. I also propose an additional option: incentivizing candidates and other partisan actors to provide information that helps ordinary votes determine the lesser evils in the opposing party.

In Australia’s RCV elections, political parties commonly hand out “how to vote” cards telling voters who to rank second, third, etc., and these recommendations strongly influence voter behavior. They do this because it is in their self-interest: If you’re running for office and you tell your supporters to rank a certain candidate second, that candidate’s supporters are also more likely to rank you second. This is especially true if you and that candidate make an explicit agreement to “endorse” one another, and since such agreements are mutually beneficial we can expect them to become common. This isn’t only a factor in RCV — increasing the level of support for both yourself and one of your rivals by a comparable amount will increase your odds of victory under basically any voting method (though this obviously doesn’t work in Plurality). While a “rank this candidate second-to-last” quasi-endorsement tends to be worthless in IRV Final Five races, they can make an enormous difference when other voting methods are used.

How about Approval Voting? In order to convince a Democrat to vote for a Republican, he needs to be convinced that either (a) the Republican in question is almost as good as the Democrats or (b) the Democrats have hardly any chance of winning, so voting for the most tolerable Republican as well as the Democrats will give him the most leverage. (a) is typically a tough sell, and (b) only works in districts in which the Republicans are strongly favored. This is still a big step up from IRV since (b) will hold in plenty of districts where less than 2/3 of the electorate votes Republican, but it still leaves a lot to be desired. Also, since all votes are equal there is a major cost to endorsing a rival by saying, “Voting for them too”, and this is especially unlikely to occur with a candidate from another party. Directions from candidates are unlikely to help the rank and file distinguish between candidates of the opposing party under Approval Voting.

STAR does much better on both counts. Giving all the candidates a different score is strongly incentivized since it guarantees you an equal voice in the runoff, so a savvy Democrat would want to give a 1 to the more acceptable Republican. It would be strategically sound to cast a ballot like this:

Lauren Boebert (R): 0 stars
Scott Tipton (R): 1 star
Diane Mitsch Bush (D): 5 stars
James Iacino (D): 4 stars
John Keil (L): 2 stars

Moreover, there is little cost for a candidate to tell their supporters to give an opposing party candidate a 1 (especially when that other candidate might reciprocate), so such endorsements should still be common.

A final option is to use a Condorcet method. With Condorcet methods, there is almost never a downside to ranking additional candidates; for such a downside to exist would require a Condorcet cycle. Similarly, there is almost never a downside for candidates to give a “less-of-two-evils” endorsement.

Both STAR and Condorcet perform well here, but I believe Condorcet would do a better job at meeting the goals of Final Five. The cost of giving an opposing party candidate 1 star is small and generally worth paying, but it’s still more than the cost of ranking that candidate right after your party’s candidates under Condorcet.

Optimizing Final Five’s Primary

Final Five uses Plurality for its primary. Given that we never want vote splitting and want to let voters support non-viable candidates without fear, this can’t possibly be right.

The biggest question for the primary is whether we want the candidates to be similar to one another or very different. Clay Shentrup argues for having them be similar on the grounds that ideology is very important and there’s an ideology that would best reflect the preferences of the electorate, so all of the best candidates must be similar to this ideal and therefore to each other. And, if we believe the purpose of elections is to elect the best possible candidate, advancing candidates who don’t look like they could possibly be ideal is pointless. Because of these considerations, Clay argues for using St. Louis’ Approval + Runoff system instead of Final Five.

I am broadly sympathetic to this perspective, but there are a couple points that argue for something closer to Final Five. First, let’s look at St. Louis’ most recent election for ward 5.

The primary:

James Page: 406 votes, 40.76%
Tammika Hubbard: 621 votes, 62.35%

(There were only two candidates, but the numbers don’t add up to 100% since it’s Approval.)

And the general:

James Page: 717 votes, 52.37%
Tammika Hubbard: 646 votes, 47.19%

While I don’t know how much of this swing is the result of general election-only voters preferring Page and how much is Hubbard voters changing their minds, we want our electoral system to be able to handle such swings and leave voters with meaningful options in the general election that aren’t too ideologically similar.

Another issue is that many voters without an ideologically similar candidate will choose not to vote. To be fair, a large fraction of these voters probably wouldn’t rank or score the other party’s candidates anyway, but if their preferred candidates are telling them to do so then a significant number should.

So, how should we determine which candidates advance to the general election? Even before we get to what the ballots look like, there are two main options:

Block voting: Elect the strongest candidate. Then elect the second-strongest candidate, and so on, generally by using a single-winner voting method several times in a row. There are block versions of Approval, STAR, RCV, and Condorcet methods.
Proportional representation: Have the general election candidates quantitatively reflect the preferences of the electorate. So, if there are 60 Republican voters and 40 Democrats, send 3 Republicans and 2 Democrats to the general election.

I think block voting is liable to leave too many voters without any general election candidate they like, but with proportional representation the second-best candidate is liable to fail to make it to the general election at all. Therefore, I lean toward a hybrid approach: use block voting to select the first three primary winners, and a proportional method to select the last two. With Approval Voting, for example, this could mean advancing the three candidates who get the most votes in the primary and using Proportional Approval Voting to select the last two, taking the first three winners into account when determining the last ones.

However, these optimizations to the primary are unlikely to matter much; the best few candidates will almost certainly advance regardless of the voting method being used. Moving away from IRV in the general election is at least an order of magnitude more important.

Conclusion

The core idea behind Final Five is a good one. It’s a reform that has the potential to reshape candidate incentives and lead to far more compromises and bipartisanship, along with a lot less political division — but only if it’s modified to use STAR or a Condorcet method instead of IRV. If it continues to be implemented with IRV it will basically be doomed. Partisan primaries followed by a general election with an essentially predetermined outcome are a big problem — and it’s a problem that IRV mimics. In its current form, FFV replaces an all-important primary with an all-important penultimate round of IRV tabulation; either way, the minority party doesn’t get a real say in the outcome.

To be fair, FFV with IRV should still be effective in districts that skew so heavily in one direction or the other that two candidates from the same party can be expected to make it to the final round. Perhaps this will lead to less polarizing rhetoric from a handful of particularly extreme members of Congress, but it won’t end gridlock or do anything to reduce the levels of political division that exist in most districts. We should not expect significant increases in bipartisanship from a reform that only works in the states and districts that are fundamentally the least inclined towards bipartisanship.

While FFV is fatally flawed, it still gets a lot of things right. If we use the best possible voting method in partisan primaries and the best possible single-winner voting method in the general election, it’s quite possible that nothing would change. It doesn’t matter what voting method you use when there are only two real candidates, and if third parties remain unpopular (as they have in Australia) then everything will be for naught, aside from letting a few maverick incumbents run as Independents. Proportional representation would be a game-changer, but if we stick with single-winner districts we need to both reform the primary system and fix the flaws of Final Five. Doing only one or the other is nowhere near enough.