The defining quality of a fractional-transfer (such as Meek) STV election is this: If you vote for a candidate and they are elected, you get some fraction of your vote ‘rebated’ to you to use on your next-highest choice. The fraction depends on how many surplus votes the elected candidate received above what they needed to be elected — every additional vote in the surplus increases the size of the rebate that all voters for that candidate get back, increasing the support that those voters’ next choices get.
The upshot of this, in a Meek STV election like the one we are about to have, is that when you vote for a candidate, you are effectively also voting for a set of candidates in their ‘shadow’ — the candidates whom other supporters of that candidate would make their next choice.[1]
Here is an example:
- There is an election in which, based on the numbers of voters and seats, 4 votes are sufficient to win a seat.
- Your two favorite candidates are A and B.
- There are 5 other voters whom you anticipate would vote for A as their first choice, and their second choice is C. (The full electorate presumably comprises more than 6 voters, but everyone else is irrelevant to this example as long as A is not their first choice.)
If you submit the ballot A > B > …, and your expectations are correct, the election will proceed as follows:
- A receives 6 first-choice votes and is elected.
- A’s surplus is 6 − 4 = 2, so ²⁄₆ = ⅓ of everyone’s A votes are rebated to them for the next round.
- In the next round, C receives ⅓ of a vote from each of the 5 other A-voters, for a total of 1 ⅔ new votes. (B receives ⅓ of a vote from you.)
Let’s stop there. Maybe 1 ⅔ new votes, in addition to any first-choice votes and any rebate votes from other sources, is enough to elect C when all is said and done; maybe it isn’t. But let’s now examine what would have happened if you had left A off of your ballot.
- A receives 5 first-choice votes and is elected.
- A’s surplus is 5 − 4 = 1, so ⅕ of everyone’s A votes are rebated to them for the next round.
- In the next round, C receives ⅕ of a vote from each of the 5 A-voters, for a total of 1 new vote. (B received the entirety of your vote already in the first round.)
Comparing outcomes:
- C receives 1 ⅔ − 1 = ⅔ additional votes in the first scenario.
- B receives 1 − ⅓ = ⅔ additional votes in the second scenario.
- A is elected either way!
So the marginal effect of your vote for A in the first scenario is to support C (because the A > C ballots put C in A’s shadow), not A. And that support comes at the expense of your next-choice candidate. Note that this is true even when C is not on your ballot at all! You might want to avoid voting for C at all costs; nevertheless, your vote for A acts as a vote for C because of the shadow cast by A in the light of other voters’ ballots.
(I want to pause here to emphasize that there is no perfect voting method, and that the point of this post is not to protest the use of Meek STV on the grounds that this is a fatal flaw or something. All complex voting methods have quirks like this. Informed voters should know about them. That’s my point.)
Though the example assumes that A is your first choice, the same principle applies if A appears later on your ballot. If A has a surplus, however much of your vote trickles down to A, that fraction of your vote will be divided between a rebate back to you and supporting the candidates in A’s shadow.[2] If you prefer the next candidate after A on your ballot to the candidates in A’s shadow, your vote is more faithful to your preferences if you drop A altogether.
Dropping a candidate that you like from your ballot when you’re confident that they will win without your vote is a voting tactic called ‘free riding’. As the name suggests, it can be a self-defeating tactic if done without coordination; if everyone who supports a popular candidate tries to get a free ride by not voting for them, nobody gets what they want. But some voters are interested in voting defensively — it may be more important to a voter to refrain from supporting candidate C than it is to give support to candidate A. If a voter is considering taking a free ride on candidate A, it may have higher expected value to do so if they expect a strongly disfavored candidate C to be in A’s shadow.
Of course all of this is conditional on how well one knows the electoral landscape — specifically, how confidently one can predict who is likely to win a seat and who is likely to be in whose shadow.[3] I don’t think there’s a way I could advise you on how you might guess at these things without violating the no-campaigning-on-Discourse rule, but you’re smart folks; you can probably figure out at least as much as I have. In the end, it’s up to you to decide what to do with however much information you can get. I only wanted to highlight what I think is a non-obvious effect[4] that is likely to be relevant to at least some voters.
Vote! And make your vote have the impact you want it to have.
‘Shadow’ is my word choice; I don’t know of a commonly-used term for this concept. ↩︎
This is specific to Meek’s method. Under most other STV transfer methods, votes won’t transfer to already-elected candidates, so in this scenario A would get skipped and the entire fraction of your vote remaining up to this point would go to the next hopeful candidate after A on your ballot. ↩︎
Note in particular that the X-is-in-Y’s-shadow relation is not, in general, symmetric. In the above example, C is in A’s shadow, but A is not necessarily in C’s. If the ‘you’ in that example had preferred C and disliked A, voting C > B would not necessarily have been converted to support for A, given the ballots presented — maybe the voters who had C as their first choice picked candidate D next. To use this tactic defensively, one needs to anticipate not only who shares supporters with whom, but which of the two is preferred within their common base. ↩︎
One that I think wasn’t present last year, though to be honest, while I like how the CIVS proportional method is advertised, I found it difficult to analyze its mathematical properties. ↩︎