How competitive are ballot measures in San Francisco?

It’s infamously common for San Francisco supervisor races to come down to razor-thin margins. It would have taken just 93 voters changing their final-round choice to flip Dean Preston’s 2019 win, and 63 to flip Connie Chan’s in 2020 (less than 1% of each district’s turnout). Ballot measures don’t share this competitiveness.

Prop D on the upcoming Nov’26 ballot limits your right to vote on issues by making it harder to get on the ballot in the first place. One argument in favor is that shortening the ballot could reduce the number of people who don’t finish it or improve voters’ attention to the remaining measures. This is unlikely to have an effect worth the cost.

This table shows the tightest results of all 429 measures between Nov 1995 and Jun 2026.¹ Sorted by the smallest shares of voters required to change sides for the result to flip. There are very few of them, many are quite old, and many are close to the top of the local section of the ballot. 96% of all measures have substantially wider margins.

Substantial persuasion required to flip local measures

Campaigns have to both turn out supporters and persuade people who would otherwise vote against them. Even after that work, the losing side on most measures was still thousands of voters away from changing the result. That’s a substantial gap for a field campaign to close.

Reducing roll-off doesn’t change outcomes

A small percentage of voters don’t complete their entire ballot. But most results wouldn’t change if they did. Looking at all past measures, 74% of measures mathematically couldn’t flip, even if every incomplete ballot² voted against the winning result. The remaining 112 measures could mathematically flip, but most would require very unlikely voting patterns.

I’ve split the possibilities by how differently those new votes would need to lean from the votes already cast. For example, June 2022’s Prop A (a bond for Muni reliability) received 65% support but needed two-thirds to pass (😭). If all 16,680 ballots without a valid Yes or No vote had included one, at least 14,444 of those new votes would have needed to be Yes to pass the bond. That’s about 87% support among the new votes, a difference of about 21.5 percentage points from the votes actually cast.

For the median measure among those 112, the new votes would need to favor the losing side by 31 percentage points more than the votes already cast. Only three measures fit within a three-point difference.³⁴ And realistically, some “undervoting” will occur no matter how long or short the ballot is.

Augenblick and Nicholson’s study compared the same contests at different positions on San Diego ballots. Each additional preceding contest increased undervoting on local propositions by 0.117 percentage points. If that estimate carried over to San Francisco, removing one preceding contest would produce about 117 additional votes per 100,000 voters on each later measure. Most San Francisco measures aren’t close enough for an effect this small to change the result.

The full distribution

Appendix: data and assumptions

Vote totals and precinct results come from the Department of Elections’ certified election results and Statements of Vote. Ballot totals come from its election data archive. Passage requirements were checked against the original voter information pamphlets and official results.

For each measure, the calculations find the fewest voters who would need to change sides to cross the passage threshold. The scenarios with additional votes keep existing votes fixed and increase the total.

The additional-votes pool is ballots cast minus Yes minus No. It includes overvotes and accounting differences as well as blanks, so it gives an optimistic limit rather than the exact undervote count from the election data.