Unranked Choice
Error Rates in Maine’s Ranked-Choice Elections
Maine recently became the first state to implement instant-runoff voting, or ranked choice voting (RCV), in U.S. Senate and House elections. Before a court ruling, the state also successfully conducted RCV primary elections for governor in the summer of 2018. With seven candidates vying for the nomination, and as the first time voters in the state would try RCV, this race offered an opportunity to examine voting behavior to see how voters acted in a new electoral system.
This paper explores three areas of RCV voting through the analysis of Maine’s 2018 gubernatorial primary. First, it offers a descriptive analysis of the ballots cast in the primary, examining various voting propensities and patterns. Second, the paper attempts to address questions about the confusing nature of RCV by looking broadly at ballot errors. Third, it addresses the criticism that RCV does not always guarantee majority winners by examining ballot exhaustion in close detail.
Voting Propensities
In the gubernatorial primary, there were 132,250 ballots cast, and 17,513 unique ballot combinations. Voters had the option to rank seven different candidates, as well as any write-in candidate, in one of eight ranking positions. Of these, the most common ballot was a single-shot ballot for Janet Mills, the eventual winner. This means that these voters chose not to rank any other candidate and vote only for this one candidate. In total, roughly 12 percent of voters voter for only one candidate, opting not to rank-order their choices. These rates were less than what has been reported in other studies, however the fact that this was a primary would lead us to expect lower rates of single-shot voting (assuming multiple candidates of the same party would be acceptable to a voter).
In addition to these single-shot votes, 10.3 percent of voters ranked exactly two candidates, and nearly 30 percent ranked all seven candidates. 288 voters ranked the choices in the same order as they are listed on the ballot. In some research on Australian voting, this is called a donkey ballot — comparatively, Maine’s rate was much lower, as this form of voting is thought to be a result of Australia’s compulsory voting law.
Ballot Errors
To address concerns about overcomplication and confusion in RCV, we looked to catalog errors on ballots to determine their frequency, and how they effected election results. We considered ballot errors to include when a voter skipped ranking spots and then subsequently resumed ranking (after two blank rankings, the ballot is exhausted) and when a voter ranked multiple candidates for one ranking (this exhausts the ballot at that particular ranking, and after). One other error that would not invalidate a ballot would be ranking a candidate in multiple non-consecutive positions.
We identified 4.3 percent of ballots with undervotes (a blank ranking) followed by a valid ranking. Only 0.7 percent of ballots contained a double undervote, which would exhaust the ballot (Maine only excludes ballots after two successive blank rankings). Of these, only 111 additional votes would have been added into the first round, if they had not been otherwise excluded. Even if Maine excluded ballots with only a single ballot, the procedure would have only excluded an additional 405 ballots. This seems to indicate that, although there were many ballots that contained undervote errors, the effect of these errors was much smaller.
Looking at overvotes, 1.4 percent of ballots contained these errors, however only 0.4 percent were actually exhausted due to an overvote. This comports with studies of RCV elections in other areas. Of these ballots, 248 had valid rankings at other locations. This small of a number of ballots would not have changed any outcome in the election.
We also identified 2 percent of ballots with multiple non-consecutive rankings. This was higher than what other studies have reported, but this is possibly due to the large number of rankings allowed. It is unlikely these errors had any effect, but they do show a lack of understanding of the voting rules.
Ballot Exhaustion
Finally, we examined how ballots were exhausted, or eliminated from counting. Ballots can become exhausted because they contain multiple blank rankings, a ranking with multiple candidates in one position, or because the ballot’s remaining rankings are for eliminated candidates. Overall, 10.7 percent of ballots were eventually exhausted due to undervotes. 5,535, or 4.2 percent of all ballots cast, were completely blank. This is lower than in some other studies, but enough to offer a good opportunity for further study. Rates for exclusion by overvote were similar to studies of RCV in other areas. Exhaustion by candidate exhaustion is rare with so many ranking positions — only 265 ballots were exhausted this way.
Discussion
Maine’s first effort with RCV seems to have been a success compared to other elections in the United States using this method. Rates of single-shot voting seem to be less than reported in other studies, however, the number of abstentions is still concerning and deserves additional study. While there were a number of errors on ballots, the actual effects of these errors seem to have been limited in this case.
