Exhausted Ballots: Did They Affect the Outcome in San Francisco’s Mayoral Election?
By Steven Hill and Pedro Hernandez
In the 2018 mayoral election in San Francisco, 21,000 ballots (8.6% of all ballots cast) became “exhausted” -- is it possible that the number of exhausted ballots impacted the winner in San Francisco's mayoral election? Or, alternatively, if SF voters had more than three rankings (which might have reduced the number of exhausted ballots), might that have affected the outcome? The answer to both questions is: no. By using publicly available ballot-image data to analyze this race, it can be determined that a substantial number of the exhausted ballots came from voters who supported the less progressive-identified candidates in the race. Those voters tended to favor London Breed over Mark Leno by a ratio of 1.36 to 1. Some voters for Jane Kim also saw their ballots exhaust, and those voters tended to strongly favor Mark Leno. Overall, this analysis estimates that, with more rankings for voters leading to fewer exhausted ballots, Leno might have closed the victory margin by an estimated total of 183 votes. This would not have been enough to overcome the 2,600 vote gap between himself and London Breed.
In addition, if Jane Kim had won 700 more first choice rankings and surpassed Mark Leno in the first round, would that have changed the outcome of the election? Definitely not. In fact, London Breed would have won by an even greater margin, because Kim would have picked up fewer second and third rankings from Leno supporters than Leno picked up from Kim supporters.
Analysis of exhausted ballots in San Francisco’s mayoral election
In the recent June 5 election for mayor, voter turnout was extremely high by local standards, with over 250,000 voters participating, approximately 52% of registered voters. That represents the second highest vote total ever in San Francisco’s mayoral history, and one of the highest rates of participation.
Nevertheless, some concerns have been raised about the number of “exhausted ballots,” which are ballots that don’t select one of the “top two” finishers. While more than 9 in 10 voters (91.4%) supported a candidate who made it to the final round of the election, that still left about 8.6% – 21,510 voters – whose ballot did not count in the final round.
Given that this very close election was decided by a margin of 1.1%, is it possible that, if San Francisco had allowed more than three rankings, and fewer of those ballots became exhausted, that it could have changed the outcome of this election?
The answer is: NO.
To answer this question, we took a look at anonymous ballot image data that the Department of Elections makes public each election cycle. The department releases this information so researchers and observers can monitor our elections process and analyze the data.
From the data, we can see that only 8,181 of the 21,510 exhausted ballots were what is called “involuntary” exhausted – meaning they used all three of their rankings yet did not vote for either of the two finalists, Mark Leno or London Breed. So their ballot exhausted after the third ranking was counted for a non-finalist candidate.
The rest of the exhausted ballots, 13,329, were “voluntary exhausted,” which means that the voters did not use all three of their rankings but instead used only one or two of their rankings. These voters could have selected either of the two finalists, either Mark Leno or London Breed, but chose not to do so, presumably because they did not support either of them.
So if San Francisco voters were allowed more than three rankings, only the smaller subset of 8,181 involuntary exhausted ballots could have potentially changed the outcome. Assuming that voters with involuntary exhausted ballots would have actually used more rankings and voted in similar patterns as with their first three rankings, how might this election have turned out?
No difference, it turns out. Mark Leno might have gained a few more votes, but not enough to change the outcome of the election. Here’s what the ballot image data shows.
A high percentage of the involuntary exhausted ballots was dominated by candidates Angela Alioto, Richie Greenberg and Ellen Lee Zhou. These three candidates were named as the first ranking on 5598 (68%) of the involuntary exhausted ballots. Jane Kim was named first on only 1937 (24%) of these ballots.
Transfers from Kim voters. If voters had more than three rankings, let’s assume the involuntary exhausted ballots that ranked Kim first would have transferred to Mark Leno and London Breed at the same rate as the non-exhausted ballots (77% went to Leno and 23% to Breed, 3.35 to 1). That would have yielded approximately 1,491 more ballots for Leno and 446 for Breed, so a gain of 1,045 ballots for Leno. But Leno lost by over 2,600 votes. He would need to pick up another 1,600 votes from supporters of the more conservative candidates, Alioto, Greenberg and Zhou.
Transfers from Alioto, Greenberg, Zhou voters. In examining the non-exhausted ballots of voters who ranked Alioto, Greenberg or Zhou first, we can see that those ballots tended to favor London Breed over Mark Leno by an overall ratio of 1.36 to 1 (57.5% Breed to 42.5% Leno). If we extrapolate that trend to the 5,598 involuntary exhausted ballots from the voters who ranked Alioto, Greenberg or Zhou first, we find that Breed likely would have gained approximately 3,230 votes and Leno 2,368 votes. That’s an increase of 862 votes for Breed.
Altogether, it would appear that allowing voters more than three rankings might possibly have resulted in a net gain for Mark Leno of approximately 183 votes. This would not have been enough to overcome the 2,600 vote gap between Leno and Breed. Breed still would have won the election and become San Francisco’s next mayor.
This analysis is assuming that, given more rankings, those voters with involuntary exhausted ballots would have actually used those additional rankings, and voted in similar patterns as all voters did with their first three rankings. Those are reasonable assumptions, given what the publicly available data shows about voter behavior in this election.
Other observers of this election have wondered what might have happened if Jane Kim had won 700 more first choice rankings and surpassed Mark Leno in the first round. Is it possible that she might have vaulted past Leno into second place, and then beaten London Breed by being the recipient of more second and third rankings when Leno was eliminated from the race?
The answer: definitely not. While Mark Leno was ahead of Jane Kim in the first round by only 678 votes, by the next-to-last round (Round 8), he was ahead of her by over 2600 votes. Leno picked up more support from the voters of eliminated candidates than did Jane Kim, showing his broader support base.
In addition, while Leno picked up support from 77% of Kim voters, Jane Kim picked up support from about 65% of Leno voters, while London Breed would have gained about 35% from Leno supporters. So if Jane Kim had somehow managed to pass Mark Leno, Breed would have won by an even greater margin than the 1.1 % she beat Leno by.
Exhausted ballots versus Exhausted voters
This issue raises other important points about exhausted ballots, and how those compare to “exhausted voters” during the previous December runoff system. Between the first election in November and the second runoff election in December, San Francisco usually saw enormous declines in voter turnout.
In any runoff system, whether an “instant runoff” or a “separate runoff” held many weeks later, differentials in voter turnout between the first and last rounds is common. For example, in the 2001 election that elected city attorney Dennis Herrera, voter turnout plummeted by 44% between the first election in November and the second runoff election in December, when less than 17% of registered voters participated. Herrera actually won in December with fewer votes than the initial leading candidate, Jim Lazarus, had in November.
In the mayoral election of 1995, when Willie Brown was first elected, the number of voters declined by over 10 percentage points from November to December. This was a very common story during San Francisco’s use of December runoff elections. In the seven December runoff elections for mayor since 1975, voter turnout declined in four of them (the average change in turnout across all seven races was a slight increase of 0.067% -- essentially no change). In the eleven December runoffs held for Board of Supervisors races, voter turnout plummeted from November by an average of 47 percent. Many times the winners in December had fewer votes than the leader had in November (see chart).
Essentially there were two different electorates, one in November and another in December. The substantial number of voters who often did not return for the December runoff is called “exhausted voters.”
And that’s not all: the exhausted voters were skewed toward certain demographics and neighborhoods of San Francisco. Those who returned to vote in December were overwhelmingly whiter, older and wealthier than the city as a whole, and tended to be residents of the high-turnout parts of the city.
With RCV, on the other hand, not only has voter turnout usually been higher than in the old December runoff system, but far more voters on average participate in the decisive final round of the election. That means a lot more voters are having a say in who their elected officials are. Also with RCV, it’s not possible for winners to be elected with fewer votes than the leader had in the initial round. Yes, there are exhausted ballots – but the number of “exhausted ballots” on average has been far less than the number of “exhausted voters” in the previous December runoff system, in which voter turnout usually plummeted.
For example, in a previous study FairVote found that the median percentage of exhausted ballots in the 24 RCV races needing multiple rounds between 2004 and 2016 was 13.2%, meaning that 86.8% of ballots that were validly cast in the first round also counted in the final, decisive round. In the 2018 mayoral election, the number was even higher, with 91.4% of voters having their vote count in the final decisive round. With voter turnout for Board of Supervisors races dropping in the December runoffs by an average of 47%, that means there were nearly three times as many “exhausted voters” in December as “exhausted ballots” in RCV races.
Another way to analyze “exhausted ballots” versus “exhausted voters” is to consider the winner’s votes as a percentage of first round ballots. In RCV elections, the winners received an average of 46% of first round ballots. But with the December runoff elections, the winner’s final vote tally was only 32% of “first round” votes (i.e. the November election). One study found that in all the December runoffs for the Board of Supervisors, winners had 8,500 votes on average; but in all the instant runoff contests, winner's on average won over 11,900 votes, a 40 percent increase.
This shows that RCV’s “instant runoff” results in the election of a candidate by a much larger proportion of the voting population than do December’s separate runoff elections. More voters now are participating in the final decisive rounds of the RCV elections than under the previous December runoff elections. Generally, the use of RCV to eliminate low turnout runoff elections has allowed more voters to have a say in who their elected officials are. That is good for democracy, and good for governance.
Still, having more than three rankings would be an improvement, and has been a policy goal of the San Francisco Elections Commission since the first RCV election in 2004. Fortunately, starting in 2019, San Francisco will begin using new voting equipment that allows voters the option of ranking up to ten candidates. The new equipment also will have a simpler ballot design, where the names of candidates will appear only one time, instead of the current ballot where candidates’ names appear three times, in three separate columns.