The Equal Vote Coalition: dedicated to true equality in the voting franchise. Join us!

Score Runoff Voting

Although there is broad consensus in the election reform community and the electorate as a whole that our election process is significantly flawed, one reason we're still stuck with it is that the reform community is fractured. Various groups advocate for different approaches to reform, measure those offerings according to different criteria, and bash competing reforms.

Score Runoff Voting or SRV is a new voting system designed to address the flaws in our current voting system at the same time it answers opposing critiques from the two leading single election voting system reform camps: those who advocate for Instant Runoff Voting and those who advocate for rated voting methods such as Approval and Score Voting.

SRV allows voters to rate the candidates just like we rate books on Amazon, restaurants on Yelp and songs on iTunes. The winner of the election is determined in two steps. First, all the voters' candidate scores are summed. The top two scoring candidates advance to the second step, where the winner is chosen as the one preferred by the most voters over the other, thus ensuring a majority outcome between the two candidates most supported by the electorate.

The results of this hypothetical election are tallied first by adding up all the scores of all the candidates given by all the voters to determine the top two. Assuming Bianca and Desi were the top overall, the ballot above would count for Desi in the runoff because the voter scored Desi higher than Bianca. A hypothetical overall election outcome might look like this:

The graph demonstrates the simplicity of the SRV tabulation process. The bar chart shows how many total points each candidate received. The percentages on the two highest scoring candidates show the preference percentages of the voters between those two. There are always just two stages: scoring, then instant runoff. By capturing more information from voters, this system provides the advantages of the equal top two without requiring voters to return to the polls.

Advantages versus traditional IRV:

  • SRV so simple that the complete instructions for the voter and the rules for determining the winner can be printed in two short sentences on the ballot itself, whereas the mechanism for tabulating Instant Runoff Voting can remain opaque even for sophisticated voters. Steve Pond wrote the following about IRV after its adoption as the voting system for choosing Best Picture by the Motion Picture Academy: "A year and a half after the Academy went to a different system for counting Best Picture ballots, nominees and voters and campaigners still don't understand how it works. And it's driving me crazy."

  • SRV provides an equal weight vote to every voter. IRV does not.

  • SRV will not penalize the voter for giving her first choice the highest rating by making her second choice more vulnerable to her least-favored candidate. IRV suffers this weakness, which is an example of the criterion known technically as "Favorite Betrayal."

  • SRV is much simpler to tabulate. It is always completed in two rounds with transparent results.

  • SRV shows an accurate measure of voter support for all candidates. The IRV "shadow" effect minimizes support for lesser similar candidates. In the 2012 Manhattan exit poll study, Instant Runoff Voting and Score Voting were included in the alternative voting methods studied. The following bar chart demonstrates quite clearly how the IRV shadow can hide public support for candidates other than the two frontrunners:
  • R-IRV decreases the potential for spoiled ballots. The ballot format is simple and scales to any number of candidates. IRV's complex ballot is much easier to spoil, and many jurisdictions limit the number of rank options given to each voter.

  • R-IRV is precinct-summable. Precinct score sums and pairwise preferences can be tabulated at each precinct and then summed together in a final location. IRV requires all election results to be collected in a single location before the count can begin.

Advantages versus Approval and Score Voting:

  • SRV provides a majority win outcome between the two highest-scoring candidates. The majority of voters in some hypothetical approval and score voting elections may actually prefer a candidate who scores second-highest overall. This feature of these methods runs counter to our particular notion of democracy that we can trace as far back as founder James Madison's declaration in Federalist #57 that representatives "will have been distinguished by the preference of their fellow-citizens."

  • SRV discourages 'bullet voting'. SRV allows the voter to support a second choice without that support counting directly against the voter's first choice. Because two candidates advance to the runoff step, the voter only harms his first choice if both first and second choice were vying for second position. In this case the voter will improve the likelihood of a preferred outcome by supporting the stronger of the two. On the other hand, if both of the voter's choices are in the top two, his full support will go to his actual favorite in the runoff round. Voters are also advantaged by showing token support for candidates other than their worst choice in case no candidate the voter actually prefers makes the runoff.

  • SRV discourages the tactical maximal exaggeration of the score of the more favored frontrunner and all candidates more preferred. The strategic voter is giving up 1/score_max strategic power in the first round in order to gain 100% voting power in the second round. Thus SRV strategically incentivizes more honest scoring, which in turn results in provably more democratic outcomes.

Next Steps

We recommend that leading election reform advocacy organizations and election scientists perform a thorough vetting of Score Runoff Voting. Of particular interest are the evaluations of groups such as FairVote and the Center for Election Science as well as the simulation analysis of SRV versus other comparable systems on the measures of Bayesian Regret and agreement with the utility-based Condorcet Winner.

We also encourage real world election reform campaigns advocating Instant Runoff Voting and Approval Voting for single-winner elections to compare the cost and complexity of implementation of SRV.

Showing 17 reactions

Please check your e-mail for a link to activate your account.
  • commented 2016-08-19 13:52:01 -0700
    This page was recently posted on reddit at

    I’m pasting my comment below:

    I am very attracted to how well score voting can capture the attitudes of each voter but I am against implementing it as its optimum use is not straightforward. Specifically, it is almost never in the voter’s best interest to score a candidate in anything but the top or bottom scores.

    Quick explanation: Anything except the top or bottom scores is functionally similar to a fractional vote. It is foolish for a voter to give their vote less weight even when the person they are voting for is not their top or bottom choice.

    Longer explanation: Consider candidates A, B, and C. Let us say on a 0 to 9 scale you like candidate A at a 9, B at a 6, and C at a 0. All candidates have an equal chance and you rank the candidates earnestly, giving 9 points to candidate A, 6 to B, and 0 to C.

    What happens if you move candidate B from 6 points to 7? You create an additional point for your favored candidate to overcome but you also add a point that your despised candidate must surpass as well. Let’s call the probability of a point affecting any candidate P. The utility of these two cancel because while you have hurt your top candidate (-P) you have also hurt your bottom by the same amount (+P).

    However, the odds of candidate B winning has increased by 2*P and since you like candidate B (6 out of 9), you’ve increased your expected utility from the vote by 2 * P * (6 / 9 – .5). (Subtract .5 because we want the change in utility from average. If we scored candidate B at a 3 or 4 then we should lose utility.)

    This logic is the same for 7 to 8 and so on. To maximize your utility of a score voting system you should give maximum points to any candidate you like and minimum points to any candidate you dislike. (Unless none of your favored or disfavored candidates can win, then more tactical voting is required.)

    A couple of other thoughts:

    1) Which is better, candidate A that 50% of the people love and 50% of the people hate or candidate B that everyone thinks is average? This system goes with the former. I’m not sure I agree.

    2) The page says, “SRV will not penalize the voter for giving her first choice the highest rating by making her second choice more vulnerable to her least-favored candidate.” I believe this is incorrect.
  • commented 2016-08-08 09:53:11 -0700
    Paul, the notion of balanced voting expressed in the article linked does indeed meet the test of balance that the equal weight vote requires, but as you have written it, it is the least expressive equal voting system possible — only being able to support or oppose a single candidate, and therefore still has the same scaling problems of plurality voting – i.e. the more candidates you have, the less you can say with the ballot. Score Runoff Voting does not have this problem.
  • commented 2015-08-24 18:34:13 -0700
    Tien Shang Chang,

    Another terminology for this concept is the concept of balanced voting – each voter has as much an opportunity to vote for as to vote against a candidate. As noted in the video you reference, such a system will decrease polarization and it opens a path for the of multiple parties to compete on issues.

    There is not just a single balanced voting system, indeed there are many of them and some are discussed in <a href=“”">"> of short articles – most of them under a thousand words.

    One consideration in choosing between voting schemes is how well a voter can express his or her opinions in an election. The simple system described in the video meets this criterion well if there are only three or perhaps four candidates, but another system would be more appropriate when there are more candidates involved. </a>
  • commented 2015-08-24 17:31:44 -0700
    Negative Vote is even better. Each person still has only one vote, but has the option to vote against a candidate instead of for. The vote is counted as minus one. Winner is the person who gets the higher net positive votes. This will increase voter participation and reduce the influence of extremists. Check this video link: and
    join our discussion group on Facebook:
  • commented 2015-05-19 13:35:37 -0700
    Find this concept interesting. Had a top-three with RCV proposal considered by the NV legislature this session. Proposal was revised due to state constitutional issues but was given a hearing as SB 499; died in committee. Will be working to bring it back next session, 2017 and will look more closely at R-IRV as a possibility. @nvelectreform
  • commented 2015-01-09 13:38:09 -0800
    “Plus, I don’t really feel it makes much sense for a voter to give a non-zero rating to a less preferred candidate.”

    That should read “Plus, I don’t really feel it makes much sense for a voter to give a non-zero rating to a less preferred candidate when only 2 candidates are on the ballot.”
  • commented 2015-01-09 13:34:24 -0800
    I think also that Warren discredits himself with his statement that “… range voting is superior to plurality voting even in 2-candidate elections…” when you look at the very simple math of 2 candidate elections. First, in a 2 candidate election, the candidate that is the first choice of the voter would inevitably get a 99. So let’s say that 51% of voters prefer A to B. This means that if the 49% of B voters put A at zero, the score vote of A would be 50.49. B voters putting A at zero is the situation that Warren himself promotes when he refers to the “pizza example” to try to discredit the majority criterion: there might be a subset of the population that absolutely cannot condone a particular choice (in this case, a Jewish person with a dietary restriction against pepperoni) and will give it a zero. It is then up to the majority who would prefer pepperoni to offer a compromise choice through their score vote of cheese or mushroom to “elect” one of those rather than pepperoni.

    But if you look at the math, the 51% of voters who prefer A to B would only need to average a score vote of 3.89 to put B over the top of A. Put another way, if all the A voters gave B a score of 3, A would win 50.49 to 50.04, but if all the A voters gave B a score of 4, B would win 50.55 to 50.49. Now, we are talking about one low score near zero causing A to win and another causing B to win, yet could anyone really tell me the difference between a score vote of 3 and 4? What would it mean to the voter for candidate A if he chose a 3 instead of a 4 (or vice versa) in terms of how that voter really felt about candidate B? There is no real difference as far as I can tell. It seems that A’s voters find candidate B to be quite unsupportable in both cases, but one of the two practically identical scores that are an indictment on B’s support can end up causing B to win. I can think of no voter who would say to themselves “I prefer A to B, but since B voters are completely inflexible, it would be better for society if I gave B a nonzero score because I felt B had one redeeming quality and handed the election to B instead of my preferred candidate.”

    Plus, I don’t really feel it makes much sense for a voter to give a non-zero rating to a less preferred candidate. Let’s say the 51% A voters are trying to evaluate B to give a rating. They look at 10 issues that are all of the same importance. Let’s say that on 9 of 10 issues, A voters are opposed to B’s stance, but B’s stance on issue 10 is the same as A’s, who agrees with every A voter on every issue. It would then seem that a rating of 10 (out of 99 or 100) might be advisable, since A voters agree with B 10% of the time. But A voters would be better off fully choosing A, since A will also vote the “right” way on issue 10, but A won’t vote the “wrong” way on issues 1-9 like B would. Giving a score to B would mean that A voters would run the risk of being on the losing end of 9 issues instead of 0 issues. And if A voters choose to elect A by abstaining giving a ranking to B, B voters would be on the losing end of 9 issues also. The only difference is that if A loses, 51% would be on the losing end of 9 issues rather than 49%. Furthermore, if A and B were functionally the same except for 1 issue out of 10, it would play out the same way, only with 51% being on the losing end of one issue and 49% on the losing end of one issue.

    This leads me to the real problem: there is some asymmetric scoring that is going on. 51 A voters wouldn’t vote {A=99,B=10} and 49 B voters vote {B=99,A=0} because the points of agreement that cause A voters to give 10 points to B should cause B voters to give 10 points to A. The only reason B voters would give A a zero rating would be for purely strategic purposes if we assume A voters are sincerely ranking B because B voters should be agreeing with A in the same proportion (if A voters agree with B 10% of the time, that means B voters agree with A 10% of the time). In a sincere binary choice then, the votes should be 51 {A=99,B=10} votes and 49 {B=99,A=10} votes, which would mean A wins the score vote 55.39 to 53.61. And in fact, it would mean that we wouldn’t need to see scores because the symmetry of honest votes would mean that whomever has a majority of the first place votes would always win if we made a score voting matrix.

    I do think that range voting is very useful since it can be a lot more expressive and allow us to have a better idea of where candidates rank relative to one another. For example, a ranking might put A&gt;B votes at 100%, but a rating system would tell us if that rating was A=99 and B=98 or A=99 and B=0, which tells us a very different story. Score voting enthusiasts look at that example and say that we should then be looking solely at scores rather than rankings since the ranking doesn’t tell us about the intensity of the relative ranks. When I look at that example, though, it tells me that pure ranking systems lack a way to be expressive and that rating systems can whittle down the pool of acceptable candidates to a level more suitable for ranking systems (2 or 3 candidates) so we can get down to head-to-head matchups. With top 2, it would be majority rule. With top 3, we would use Condorcet (which simulates 2 majority rule elections), with any cycles being resolved by dropping the lowest rated candidate. I think either variant of R-IRV will yield the right candidate the most times because it combines the strengths of the rating system with the ability of the ranking system (in 2 and 3 candidate matchups) to correct for any errant or strategic ratings that exist in the score vote.
  • commented 2015-01-06 21:10:51 -0800
    Thanks for sharing Mark. Reading through the discussion, it seems my prediction that score voting enthusiasts would be harder to convince was in the right direction. Although Shentrup seems to be amenable to R-IRV, Smith (score voting’s version of Richie, in terms of evangelism) doubled down on score voting being anti-majority criterion by saying “… range voting is superior to plurality voting even in 2-candidate elections…”

    Personally, I see where Warren is coming from in that it might be worse off overall for a majority of voters who barely prefer A to B to elect A over the wishes of the minority who think A is the worst candidate ever. In Bayesian regret terms, it would mean that the regret felt by the 49% who hate A if A got elected would overwhelm the amount of regret felt by the other 51% if B got elected.

    But it goes against every “common sense” instinct that a regular voter has. I can just imagine the attack commercial now: “Score voting might seem nice, but did you know that if a majority of voters rank someone as their favorite, they can still lose???” What would score vote’s pithy response to that attack ad be? “Don’t worry, it’s because that majority wouldn’t be as sad as the minority would be if it were the other way around?” I think R-IRV (or some variant) would need to be done because score voting would be so easily demagogued along this line.
  • commented 2015-01-06 11:09:05 -0800
    Logan and Paul, you might enjoy this discussion on the CES Google Group about the concept:!topic/electionscience/JK82EFn7nrs
  • commented 2015-01-06 11:06:26 -0800
    Paul, thanks for the comments. I’m not sure how the behavior you describe (voters who like two candidates casting maximum support for those two candidates) is an example of “gaming” – I would think of it as just normal voting behavior. It might be better described as “tactical” – basically a version of the naive exaggeration tactical voting pattern that Smith and the CES folks have concluded is the best strategy for the voter under range and approval voting. The advantage of R-IRV in this case is that it gives the voter an incentive to give slightly less support to a candidate he favors less, in order that the runoff tip in favor of the candidate he likes more. You can see the explanation of this above in the comparison with Range/Approval.
  • commented 2014-12-30 07:07:00 -0800
    I should probably apologize for my comment yesterday – I did not read the article carefully enough and this is a different system.

    It does seem to me however that R-IRV, particularly with an large range, might invite attempts to game the system. The approach that comes to mind would be for some like-minded group of people who support two of the candidates to agree to cast the maximum rank for those two candidates and minimum votes for all of the other candidates. In fact, this might in time be the pattern that all voters would use.
  • commented 2014-12-28 07:23:14 -0800
    There is a special case of R-IRV that <a href=“”">"> I call IRBV</a> (Instant Runoff Balanced Voting). By “balanced” I mean that if one voter casts a vote for a candidate than any other voter can cast a vote against that candidate that will cancel the opposing vote. The simplification is that there are only three possible votes,

    -1 : a vote against
    0: Abstain from voting
    +1: a vote for

    The voter ranks the votes in order of declining importance. The virtue of the simplification is that it is easy for the voter to understand and enables the voter to accurately express preferences (including negative ones).

    Balance (which mathematicians might prefer to call symmetry) can be added to other systems of voting and even the very simple <i>balanced plurality voting</i> seems a big improvement over plurality voting. The link above is to one article in a series that discusses balanced versions of plurality and approval voting as well as of IRV.
  • commented 2014-12-18 20:36:06 -0800
    BTW, the Range3CondorcetRunoff method works about the same, only the top 3 candidates are selected to participate in a Condorcet runoff election . In your example above, that would mean Drew, Bob and Alice would go on to the Condorcet runoff. If there was no Condorcet winner (e.g. Drew beats Bob, Bob beats Alice, but Alice beats Drew), then Alice would be dropped as the lowest score-getter of the three and the winner between Drew and Bob would be elected (in this case it was Drew).

    Adding a third person would increase the diversity of the electorate’s choices in the final round and an election that had 3 equally viable candidates could accommodate that possibility using a top 3 framework. I think it is way less likely that an election would have 4 candidates of nearly equal quality, so stopping at 3 makes the most sense. Third place candidates might not win 99% of the time, but I think the 1% of the time they do win will be because they were nearly identical in support to the top 2 and would then deserve it.
  • commented 2014-12-18 19:58:55 -0800 already did a Bayesian regret analysis for R-IRV here:

    R-IRV is listed as Range2Runoff, and it does best at electing a Condorcet winner where one exists in all simulations, but has higher Bayesian regret than straight range/score vote.

    I think that the FairVote people might be more amenable to R-IRV than the score voting people. The reason for this is because FairVote’s criticisms of score voting are answered better by R-IRV than ScoreVoting’s criticism of majority vote. FairVote criticizes score voting for failing the later-no-harm and majority criteria. R-IRV satisfies both criteria adequately enough. Later-no-harm is relegated to the 2nd preference, not the top preference (with my Range3CondorcetRunoff method, it is relegated to just affecting the #3 preference). And the majority criterion is fixed with the runoff provision. For Fairvote, then, it might come down to them having a philosophical change of heart. They are apparently quite alright with a moderate candidate with 33.1% of the first place votes being eliminated if the two partisan candidates have 33.5% and 33.4%, even if all the voters for the partisan candidates ranked the moderate as their second choice. That is, they only want to consider one head-to-head matchup instead of all of them. But they would then have to defend their system against a system that no longer fails criterions that IRV passed as well as trying to defend why IRV fails the very critical favorite betrayal criterion when R-IRV does not.

    The score voting people, though, actively defend score voting’s rejection of the majority criterion as being a feature, not a bug. They feel that if 51% of voters prefer A to B slightly (score votes 99 and 98, respectively) and 49% of voters prefer B to A very strongly (score votes 99 and 0, respectively), than B should win even though head-to-head rankings would favor A . I’m guessing they believe that score voting makes it so voters who are functionally indifferent to which candidate is elected do not elect a candidate that other voters find highly objectionable. R-IRV neutralizes score voting’s rejection of the majority criterion and in my opinion, takes away its supporters main argument for score voting: that relative preferences matter.

    That said, I think that R-IRV is a great system that passes the favorite betrayal criterion (the most important one which gives us true measures of top vote support), makes the need for strategic bullet voting almost non-existent (leading to truer score votes) and fixes score voting’s bug of electing a winner contrary to what would happen in an election with only 2 candidates.
  • followed this page 2014-12-18 18:23:01 -0800
  • commented 2014-11-10 12:08:52 -0800
    That’d be up to the Attorney General to determine. Maybe something like:

    “Changes election process: voters rate all candidates; majority choice between two highest-rated candidates wins”
  • commented 2014-11-07 21:09:58 -0800
    If you were to craft a ballot initiative for this, how would you describe it with the limited word count?