The Alaska Story investigation: How ‘ranking the red’ really works when the Republicans don’t consolidate

By SUZANNE DOWNING

Sept. 8, 2026 – A computer simulation of a hypothetical Alaska governor’s race shows how three Republicans splitting the vote against one Democrat could produce sharply different outcomes, depending on voters’ willingness to rank additional candidates.

The simulation, conducted with Google’s Gemini artificial intelligence model, demonstrates how particular assumptions about voter behavior — including “ballot exhaustion” and crossover voting — can affect the ranked-choice tabulation.

To keep the exercise focused on the mathematics rather than the personalities, the four candidates in our simulation are identified as Candidates A, B, C, and D.

Alaska’s general elections are conducted through ranked-choice voting, and have been since 2022. Since then, only three races have ever been actually decided through ranking. It’s poorly understood and difficult to explain, due to the variable of human nature.

The way ranked-choice voting works is that if no candidate receives a majority in the first round, the candidate with the fewest votes is eliminated. Ballots assigned to that candidate move to the next-ranked candidate who remains in the race. Thus if your first choice was Candidate C, and Candidate C was eliminated, your ballot would go to your next choice, such as Candidate B, The process continues until two candidates remain, with the candidate receiving the most active votes declared the winner. The most votes is 50+1.

Voters are not required to rank every candidate. They may rank one candidate or several. A second choice is counted only after the voter’s first choice has been eliminated, according to the Alaska Division of Elections. 

Thus was born the phrase, “ranking is for losers.” Ranking only matters for your ballot if you did not choose the most chosen candidate first.

A ballot becomes exhausted (or inactive for purposes of later rounds) when it no longer contains a valid ranking for a candidate who remains in the contest. That can happen because the voter ranked only one candidate, or two candidates.

The first simulation we ran began with 300,000 votes.

Candidate D received 43%, or 129,000 votes. The three Republicans divided the remaining 57% evenly, giving Candidates A, B and C 57,000 votes apiece, or 19% each.

The model assumed that when a Republican candidate was eliminated, 20% of that candidate’s original supporters did not rank another active candidate. Another 10% transferred to Candidate D, while the remaining 70% moved to one of the remaining Republicans.

Of Candidate C’s 57,000 votes, the simulation sent 19,950 to Candidate A, 19,950 to Candidate B and 5,700 to Candidate D. The remaining 11,400 ballots became inactive. That may be an underestimate of how people actually vote and choose not to rank. A20% ballot exhaustion rate is a historically conservative estimate.

After those transfers, Candidate D had 134,700 votes. Candidates A and B each had 76,950. The active vote pool fell from 300,000 to 288,600.

The model then eliminated Candidate B. After transferring the eligible ballots, Candidate A finished with 134,306 votes, while Candidate D finished with 142,894.

Candidate D therefore won the hypothetical contest with 51.55% of the 277,200 active votes remaining in the final round. Candidate A had 48.45%.

A total of 22,800 ballots became inactive by the end of our simulation. Candidate D also received 13,894 votes transferred from voters who initially supported Republican candidates.

The second simulation used the same 300,000-voter electorate and the same transfer assumptions, but it began with Republican support much more heavily concentrated behind Candidate A.

Candidate D again started with 43%, or 129,000 votes. Candidate A received 41%, or 123,000 votes. Candidates B and C received 8% each, or 24,000 votes apiece.

Candidate C was eliminated first. Its votes moved according to the same assumed pattern: 8,400 went to Candidate A, 8,400 went to Candidate B, 2,400 went to Candidate D and 4,800 became inactive.

That left Candidates A and D tied at 131,400 votes, while Candidate B had 32,400.

When Candidate B was eliminated, Candidate A received an additional 24,150 votes under the model. Candidate D received 3,450 more.

Candidate A won the simulated final round with 155,550 votes, or 53.56% of the remaining active vote. Candidate D finished with 134,850, or 46.44%. Only 9,600 ballots became inactive before the final result.

The two scenarios illustrate the mathematical advantage of entering the ranked-choice count with a large first-choice base. In the divided scenario, where Republicans are not consolidated behind one candidate, Republican voters must pass through more elimination rounds, creating more opportunities for ballots to exhaust, mistakes, or transfer to Candidate D. In the consolidated scenario, Candidate A began close enough to Candidate D to benefit quickly from the elimination of the two smaller Republicans.

The exercise also illustrates the political dispute surrounding the phrase “rank the red.”

Some Republican organizers use the slogan to encourage voters to rank every Republican on the ballot, even when their preferred candidate is eliminated. Mathematically, doing so reduces the number of Republican ballots that become inactive before the final round.

But ranking additional candidates remains optional. Some voters may determine that another candidate carrying the same party label does not share their values sufficiently to earn a second or third ranking. In the old days, it was “vote your values.” Now, some have morphed that into, “rank whoever has a R behind their name.”

But campaign appeals, party pressure or slogans cannot guarantee that conservative voters will transfer their support to candidates they do not favor or who do not reflect their beliefs. In House District 23, Ruby Jones just registered as a Republican and filed to challenge Rep. Jamie Allard. Both have Rs behind their names but no one would confuse “Decoy Jones” for an actual conservative.

The Division of Elections says voters may rank “as many or as few candidates as you would like.” Ranking another candidate does not weaken a voter’s first choice, because the later ranking is not considered unless the earlier choice has been eliminated. If you vote for the winner, your ballot is done.

The simulation is based on certain assumptions. Its 43% starting share for Candidate D, its 20% Republican exhaustion rate and its 10% Republican crossover rate are one set of assumptions. Different starting percentages or transfer patterns would produce different results.

This simulation also simplifies the order in which transferred ballots move after multiple eliminations. A real tabulation follows every voter’s individual rankings rather than applying one uniform transfer formula to an entire group. This is one reason ranked-choice voting is so difficult to explain to people. Each ballot is its own universe. And each round of counting is like its own election.

The model demonstrates a structural challenge and a cognitive challenge in ranked-choice voting for the three Republicans now running for governor.

Under ranked-choice voting, first-choice totals matter, but they are only the beginning. Whether voters rank additional candidates, and whom they are willing to rank, may determine how many ballots remain active when the race reaches its final two candidates.

‘Rank the red’ is a slogan, but not a strategy. Alaska’s election history proves it

Suzanne Downing: ‘Rank the Red’ message makes no sense to nearly two-thirds of Alaska voters

Suzanne Downing: Republicans are being told to ‘rank the red.’ Democrats know better than to ‘rank the blue.’