Methods — Could an Alternative Voting System Have Stopped Trump?

In the polls that I used, second-choice percentages were presented for each first-choice candidate, so I was able to use each candidate’s breakdown as the set of probabilities that their supporters would pick each of the other contenders as their second choice.

Though no poll that I could find bothered to ask respondents about their third choice, I extrapolated from the second choice data by creating weighted averages of the preferences of first- and second-choice candidates. I used similar weighted averages to get probabilities for fourth, fifth, etc. choices as well.

With these probabilities, random ballots could be generated by assigning a first-choice candidate to the ballot, and randomly (with the probability weights described above) drawing a second-choice candidate from the pool of remaining candidates, and a third-choice candidate, and so on until all of the candidates were ranked on the ballot.

Election simulations were then carried out by using the number of votes each candidate received in the primary to generate that number of fully-ranked ballots with the given candidate as the first choice, then carrying out the elimination and redistribution procedure described in the main article until one candidate had obtained at least 50% of the votes.

Each state was simulated 1000 times, with random noise added to each candidate’s baseline second-choice probabilities in order to allow for uncertainty in the polling data. The degree of random noise was approximated by the amount of discrepancy between the two polls I used.