Practical IRV with Double-Win Final

This article was written with a specific legislative bill in mind, so you might find some of these items irrelevant. But please check out the double-win concept.

IRV = instant runoff voting,
popularly referred to as “ranked choice,”
voters rank candidates,
the whole field of candidates is compared at once,
each claims a vote by being ranked highest on a ballot,
the candidate with the fewest votes is eliminated,
the process repeats until one candidate remains.

For anyone considering an IRV proposal,
knowing that IRV is hot right now,
and that it has a major flaw,
and that complex proposals are more likely to fail,
I offer these simple suggestions,
with plenty of explanation.

I’ll put 4 asterisks on each of my 5 core suggestions so you could skip ahead to them if you like. The first explanation is long, so skip to the bottom for more.


Vote-for-one is a poor election method, because it’s prone to vote-splitting, and it discourages candidates from running. So vote-for-one produces the right winner, who knows, maybe only half the time.

IRV is much better, but it has an occasional, annoying way of kicking out the strongest candidate in 3rd place.
(But not 2nd place, because they win whenever they can get into the top 2!)
A well-known case was a Burlington, VT mayoral election, that made people mad enough to repeal ranked choice.

A ballpark guess as to how often IRV works right would be around 95%. I don’t have data to back that up, but occasional misfires are certain with IRV because it violates the Condorcet criterion, which is one of the good ones. (Some “criteria” contradict others because they are just someone’s opinion.)

We could greatly improve the accuracy of IRV with a modification that uses a better way to compare the top 3.


Condorcet’s evaluation method for ranking elections is to examine every possible combination of two candidates. Examining two, head-to-head, is the perfect use of ranking ballots. The one ranked higher on more ballots is preferred over the other, it’s indisputable.

This pairwise comparison of candidates is the best insurance against the spoiler effect. But a 3rd candidate in an IRV comparison can split enough support (spoiler effect) to undo the most popular candidate before the top 2.

One candidate who has a “win” (winning a two-candidate matchup) over every one of the other candidates should always win the election.
(This undefeated candidate is called a Condorcet candidate.)

The average person dislikes the word “Condorcet.” No one knows how to say it because it’s French (my best guess is Cone-DOOR-say), and they don’t want to learn another vocabulary word.

One could call it “round robin,” after a tournament where each competitor goes head-to-head with each of the others. But since the word “round” is already used in reference to IRV rounds, maybe call it a double check, a pairwise check, condorcet check, or double-win final. (Double-elimination would be nice, but it’s not accurate.)


The evaluation of these pairings gets complicated with many candidates, but it’s simple when there are only three.

There are three possible combinations of three candidates called A, B, and C:
A vs B,
A vs C,
B vs C

And the two possible relevant outcomes are:
1. One candidate wins over both of the others.
2. Some kind of tie.

* * * * When three candidates remain, one candidate who wins over both of the others in a double-win final is the strongest candidate, and should be declared the winner.
If one candidate does not defeat both of the others, the double-win final is inconclusive, so resume the IRV evaluation to determine the winner.

IRV with Double-win Final won’t fully comply with Condorcet criterion, but at least the top 3 will, which I regard as necessary for both accuracy and lasting voter satisfaction. If one state uses this plan, another might then be inspired to use a top 4 or top 5 final. Onward and upward.

This part is really about semantics. But it might clarify some things if it’s still hazy to you.

I see two ways to describe the same double-win final procedure. I think the 1st way sounds better to people, and the 2nd way is more like what the computer would do:

1. Pause the IRV process when 3 candidates remain.
Use a double-win final, in which the three possible combinations of two candidates are compared.
The one that wins head-to-head against both of the others is the election winner.
If there is no such candidate, resume IRV to determine a winner.

Or, 2. Use IRV to determine the 1st seed.
The candidate eliminated from the top 3 becomes the 2nd seed. (The one eliminated from the top 2 has already lost a matchup, and can’t win.)
Use a 2-way runoff between the 1st seed and the 2nd seed.
If the 1st seed wins (double-win), the IRV result is confirmed, and the 1st seed is the election winner.
If the 2nd seed wins the runoff, a double-win is required.
Use a 2-way runoff between the 2nd seed and the other candidate from the top 3.
If the 2nd seed gets the double-win, the 2nd seed is the election winner.
If the 2nd seed does not win both of its runoffs, the 1st seed is the election winner.
(Because IRV is the tiebreaker for the double-win runoff)


Because this would use IRV both to narrow the field and as a tiebreaker, these 3 most common criticisms of Condorcet method will NOT be a problem:
1. Frequently inconclusive (ties)
2. Burying (Pure Condorcet could produce an unpopular winner when enough voters strategically, or insincerely, rank the same unpopular candidate. IRV provides no incentive for this, so it could backfire. And with only 3 candidates, “burying” is indistinguishable from consensus, one and the same.)
3. A candidate could win with few 1st-choice votes (True, but very unlikely to be a problem, especially with IRV early rounds depending heavily on 1st-choice votes.)


* * * * A tie between last-place candidates should be determined by a one-on-one comparison, rather than by lot. One might expect last-place candidates to be unelectable anyway, but that is not always the case.

Again, a comparison of two candidates is the perfect application of ranking ballots, so let the people decide which of the two they prefer.

Both possibilities could be tested, and if they result in the same election winner, there’s no grounds for a hand recount on the tiebreaker. But if the election winner differs, we had better get it right.

The people who write software for IRV elections should have no trouble giving us different options such as using one-on-one comparisons as needed, especially since the last round of IRV is also one-on-one.

But if it’s two candidates tied for last in a field of 7 or more, they are tremendously unlikely to be relevant, and could both be eliminated at once. Clearing out the riffraff clears up the picture of the true contenders.


[2-5-2022 Note: Instant-Runoff Ranked Choice does not require the following instruction; it is only needed for the Condorcet portion.]

* * * * Put an instruction on the ballots that reads
“Only rank candidates that you would want to win.”
This will clear up a lot of confusion, and is fair warning that ranking a candidate can cause them to win.
(Someone might think they need to rank 5, even if they only approve of 2. They should rank only 2.)


Nebraska’s constitutional requirement for nonpartisan primaries must be addressed.

Condorcet method and IRV are designed for producing one winner, but neither is good for producing two winners in a nonpartisan primary. Here are three possible fixes.

* * * * Option 1: For simplicity’s sake, don’t change the legislature election at this time. Have ranking only apply only to governor, congress, and might as well add state officers such as auditor.

Option 2: Let the legislature primary produce one winner, and skip the fall election. I haven’t checked if the constitution would allow for this. But there’s potential for saving money with one-phase elections. Legislature and state officers might be right for no-primary.

Option 3: The best nonpartisan ranking primary could be this: Use the IRV procedure (or the IRV / Condorcet combo, it doesn’t matter which) to determine the 1st seed. And one head-to-head comparison of the 2nd and 3rd place candidates will determine the 2nd seed for the fall election.


Just putting it out there.

There’s a lot of support for more term limits. I’d say it’s a shame to fire someone who does good work.

An intelligent term limit would be to allow people to vote NO on one candidate.

* * * * A candidate receiving over 50% of NO votes is disqualified. No matter how many positive votes they get, they cannot win.

That’s the will of the people. Majority rules. We shouldn’t have to agree on someone else. We should be able to agree on blocking a bad guy.

The parties won’t like it.

The best place for it would be in a nonpartisan primary, so maybe we need more of those.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s