Voter Error in Top Two Primary Can Be Far Higher than in RCV Races

by Eli Hanson-Metayer, Rob Richie // Published August 12, 2014

Interested in this topic? Sign up to receive our newsletter and other updates on elections and electoral reform. 

You might think that nothing could be easier than voting for one candidate in an election, but analysis of voter error in California’s Top Two primary system shows surprisingly high rates of invalidated ballots in such vote-for-one elections. Our analysis of overvote rates shows that voters in Bay Area cities have made up to nineteen times more errors in high-profile Top Two primary elections than in their local ranked choice voting (RCV) elections with similar numbers of candidates.

In addition, “undervotes” – meaning ballots where the voter skips voting for an election – are on average far lower in RCV elections than in previous elections for these offices held without RCV. This data suggests that, contrary to conventional wisdom, RCV is not “confusing” and the modestly higher voter error rates with RCV derive primarily from the number of ballot choices rather than use of RCV.

 Overvoting occurs when a voter selects more candidates than the voter is eligible to vote for, thereby invalidating the ballot. For example, suppose a citizen is voting in a race where each voter can cast one vote for any candidate. If the citizen casts a vote for two candidates instead of one, that vote will be counted as an overvote. An overvote can occur due to voter error, like when a voter accidentally marks off the wrong candidate and tries to fix the error by crossing off the bubble and fills in another bubble, or due to machine error, like when a punch-machine brushes off a perforated box for a candidate not selected.

 Similarly, if a citizen does not cast a vote for a race or does not cast all the votes that each citizen can distribute in a race, that is considered an undervote. Undervoting can occur due to error, when a voter checks a bubble to indicate their choice rather than fills in the bubble, but most often undervotes occur when the voter chooses not to vote for any of the listed candidates due to indifference.

 bay area overvotes2

In California, the primary election is exceptionally important. It narrows the field to two candidates, with voters not even allowed to write in a candidate in the higher turnout election in November. Furthermore, the great majority of top two elections effectively determine the outcome, as one candidate usually will enter the general election as the overwhelming favorite – that is, that candidate is associated with the state or district’s majority party, and his or her opponent is not also associated with that party.

 The 2012 election cycle showed a problematically high level of overvotes in the June primary for the top-ballot race for U.S. Senate race with incumbent Diane Feinstein facing 23 other candidates. We focused on four jurisdictions that we have studied closely because they all use ranked choice voting for electing their mayor and other city offices. The overvote rates in the U.S. Senate primary were remarkably high in San Francisco (1.56%), Berkeley (2.46%), Oakland (4.52%), and San Leandro (4.6%). In San Leandro, almost one in twenty voters did not have their vote counted for Senate in the June 2012 primary.

 The June 2014 primary races also indicate an overvoting problem, although a reduction in candidates and potential improvements in ballot design reduced error. In the 2014 top-ballot gubernatorial race with incumbent Jerry Brown facing 14 opponents, overvotes were 0.5% in San Francisco, 0.24% in Berkeley, and 0.51% in Oakland and San Leandro. Although this is down sharply from 2012, it is still higher than ranked choice voting mayoral elections that allegedly are “confusing.”

 San Francisco, Berkeley, Oakland, and San Leandro all elect candidates to local office through RCV, using ballots that allow them to rank their first, second and third choice candidates. San Francisco elects 18 offices by RCV, including the 11-member Board of Supervisors elected from single-member districts and seven citywide offices, including mayor. At least one of those offices has been up for election every November since San Francisco started using RCV in 2004. The three East Bay cities all started using RCV in November 2010, and also used it in November 2012. They all use RCV to elect their mayor and city council, with Oakland electing nine other offices by RCV as well.

 People rank their options when making choices every day, and so it should be no surprise that they have little trouble doing so when voting in elections using RCV. When surveyed in San Francisco’s first RCV elections in 2004 and 2005, between 80 and 90% of voters in RCV elections report understanding the system well or fairly well. Far higher percentages of voters cast valid ballots. Consider these numbers of overvotes from the first mayoral elections with RCV in San Leandro and Oakland (in 2010) and in Berkeley (in 2012) and the first contested RCV elections for mayor in San Francisco (in 2011).

Ranked Choice Voting Elections for Mayor




Number of Candidates

% Overvotes









San Francisco




San Leandro




 This accords with the very high percent of voters who are able to rank candidates without spoiling their ballots in other cities using RCV. For example, in the hotly contested election with candidate for mayor in Burlington (VT) in 2009, only one vote out of 10,000 could not be counted after a recount. In Minneapolis the same year only one vote out of 46,000 mayoral votes could not be counted, and last year’s RCV election for mayor in Minneapolis has further demonstrated how well voters understand (and prefer) being able to rank their choices. In Portland (ME), its first use of RCV in an open seat election for mayor in 2011, only 0.18% of ballots were invalid in an election with 15 candidates and 15 available rankings.

 There are no primaries in RCV elections in the four Bay Area cites using it. Voters arrive at the polls once to cast a ballot in the high turnout November election, ranking up to three preferences among candidates. (When new equipment is in place, voters will have additional rankings.) In these Bay Area RCV elections, the average overvote percentages overall are much lower than the 2012 Senate primary overvote percentages in the same cities and lower than in 2014 gubernatorial primary elections. The overvote rate in the 2010 and 2012 RCV elections ranges from 0.14% to 0.28%. In San Francisco, the average RCV overvote rate is five times less than the overvote rate in the 2012 Senate primary and almost half the overvote rate in the 2014 gubernatorial primary. In Berkeley, the average overvote rate in RCV elections is seventeen times lower and 60% lower, respectively. The same story holds for San Leandro, where the RCV overvote rate is seventeen times lower, and almost half the overvote rate, respectively. In Oakland, the average RCV overvote rate is nineteen times lower than the overvote rate in the 2012 Senate primary, and over half the overvote rate as the 2014 gubernatorial primary.

 There also is compelling information about undervotes in RCV elections. San Francisco’s undervote rates suggest that voters are less likely to skip races or make mistakes, flatly contradicting the claim that RCV imposes “information costs” that drive voters away. San Francisco formerly had a December runoff if no candidate won a majority of the vote in the November election. Comparing, in an apples-to-apples way, the 20 RCV races that went to multiple rounds to determine a winner in 2004-2012 with the November vote in the14 non-RCV races that that ended up having a runoff 2000-2003 renders some surprising results. Due to the more limited information about undervotes and overvotes in some of the elections, we simply look at the “non-vote” – that is, the percentage of voters at the poll who did not vote in the city election being studied.

RCV Races Compared to Non-RCV Races


RCV Races
(20 multi-round races, 2004-2012)

Non-RCV Races
(14 general election races, 2000-2003)

Median Non-votes



 The RCV races had a median non-vote rate of 9.48%, while the non-RCV races had a median non-vote rate of 13.01%. Looking only at races for the Board of Supervisors gives similar numbers: 9.74% non-votes in the 16 RCV races and 14.44% non-votes in the 11 non-RCV races. 

RCV Races Compared to Non-RCV Races: Board of Supervisors Only


RCV Races
(16 multi-round races, 2004-2012)

Non-RCV Races
(11 general election races, 2000-2003)

Median Non-votes



The fact that both non-vote and overvote rates are consistently and significantly lower for RCV races than some comparable non-RCV races in California elections suggests that ranked ballots are easy to use and that voters get more interested in the election than they did in the pre-RCV era.

Higher overvote rates in elections where more than two candidates compete, such as primary elections and RCV elections, are almost certainly due to the simple fact of having a race with more candidates. It is the important step of deciding among many candidates which can be a greater challenge for voters, not asking voters to rank them. Choice among a variety of candidates is so vital for democracy that we should strive to use systems that use the highest level of voter participation.  

With that greater choice in November, RCV is giving voters more power, as indicated by such statistics as 16 of Oakland’s 18 RCV winners securing more votes than the winner of the preceding non-RCV election. RCV provides a more representative government, increases voter choice, boosts participation in decisive elections, encourages candidates to connect with voters, and results in less voter error than the “simple” Top Two primary in races with similar numbers of candidates.