Everybody in politics has an opinion about ranked choice voting; our perspective is that RCV is good and lawful because voters want it and we are glad that it is in play. Beyond that, my own opinion is one of skepticism. Not in the sense of thinking “this will be bad” but skepticism in the sense that I am constantly thinking about how we don’t know what immediate impacts RCV will have–beyond the rhetorical ones. I would like to test and prove all of the various claims when it is employed. It will offer a flourishing of campaigns and strategies, and to me, it may be that this is neither good nor bad. Given that we do strategic computation as a business, we think it is good.
Click here for an explainer of how the system will work for NYC, and the main thing you need to know is this:
“In New York City, primary and special-election voters will have the choice to rank up to five (campaigns). So let’s say you like Candidate C the best, but you also like Candidate A and to a lesser extent Candidate B. You can rank Candidate C as your first choice, Candidate A as your second and Candidate B as your third. You don’t have to rank all five – in fact, you can just choose one candidate.”
Let’s consider the concept of “voter roll-off” right as we jump in. This is the phenomenon where a voter casts their ballot, but they do not input a choice for every single contest featured on the ballot, leaving some rows under-voted. While much formal work has been done in an attempt to understand voter decision-making patterns, roll-off–whether it is encountered in “down-ballot” races with low awareness or is done intentionally at the top of the ticket–is known to occur for a variety of reasons.
Voter roll-off exists in a new dimension now as well. Consider voters who show up and vote in your contest but they select only one candidate, providing no further ranks at all. Undeniably, anyone who cast a vote in the election has preferences for the outcome, and the new vote-counting system provides additional depth for encoding these preferences. In a way, you could say that this voter is leaving part of the activity on the table, and they have diluted the strength of their own vote merely by failing to rank 2nd, 3rd, 4th, and 5th-choice candidates (where five or more candidates exist) when other voters are doing so to the benefit of their own preferential outcomes. Put simply, when looking at the complexity of turnout in these elections, we not only have voter roll-off on the ballot vertically, but now there is a horizonal form of roll-off.
I started with the concept of roll-off here for a simple reason: let’s say you have a crystal ball and it tells you that 10,000 voters exactly will ballot in your hypothetical contest. In the past, the objective floor for your “Win number” would be 5,001. The contest has begun and there are ten (10) candidates. Now, for simplicity’s sake, let’s suppose that your crystal ball also tells you that these 10,000 voters are all high-information, duly motivated citizens, and they will all cast the full width of the ballot–assigning all five available rank slots to candidates.
You just collected 50,000 votes. What is your win number now?
At this time, nobody has any proven notions about how NYC voters–in a town with storied neighborhood Clubs, union bosses with real political power, and aggressive tech-sector companies doing intense local politics and lobbying–will react to RCV or appreciate its rankspace. Votes could be said to be hugely inflated–there’s so many of them now!–and they fill this rankspace in a way nobody is situated with yet.
When I say rankspace, I mean that rather than recording the vote as a single candidate name or line, voters are ranking preferences and presenting vote-counters with an ordered list. This is literally a new data-type. Your win number doesn’t automatically become 25,001 if you collect 50,000 votes from 10,000 voters in this format.
On a typical 10-candidate ballot, for all voters that took part, there are exactly 10 possible ballot inputs the government will receive from the hypothetical 10,000 voters. A vote for you counts as a vote for you, and 5,001 is everybody‘s objective win number. Everybody must pursue the same licit strategy: connect with 5,001 or more voters. While there is no crystal ball in reality, and turnout is forecast rather than known, the basic practice here is the same–set a target that will end up being the most votes cast. Being less wrong about turnout and also achieving reasonable and achievable targets for your win condition is the thing itself.
I have been conducting an informal survey of NY-focused political professionals for the last few months and asking them whether they are planning any changes to voter contact scripts or to their data collection practices in context of RCV and I am stunned that to date, nobody has anything special in mind.
This visualization represents an algorithm we have written to express how many voter preferences might exist, with a given number of candidates. We start here because we believe that approaching a win number and understanding your proximity to it under RCV means understanding the formation of voter preferences in some new dimensions.
What you can see above is that if there are 10 candidates standing, voters may submit ballots any one of 36,100 ways. That number is correct. Here is a sense of how this plays out as you grow the height of the ballot, adding candidates in your contest and this does not include the potentially limitless variety of write-ins:
Even with just three campaigns in place, there are now 15 ways voters may submit a ballot. Here is how ballot combinatorics works now for 3 campaigns (which are herein called A, B, C):
| 1st | 2nd | 3rd | |
| 1 | A | B | C |
| 2 | A | C | B |
| 3 | B | A | C |
| 4 | B | C | A |
| 5 | C | A | B |
| 6 | C | B | A |
| 7 | A | B | |
| 8 | A | C | |
| 9 | B | A | |
| 10 | B | C | |
| 11 | C | A | |
| 12 | C | B | |
| 13 | A | ||
| 14 | B | ||
| 15 | C |
Let’s take the perspective of Campaign A. We have the crystal ball and expect 10,000 voters. Because this is a 3-way race, we will actually be able to discard the third round picks entirely, and work in two rounds, with certainty. Few will be this lucky, but I’m going with this to demonstrate these concepts in their plainest forms. What this means is that we are only going to address the last 9 rows of the table above–where two-round preferences are entered–as our meaningful ballot combinations. That is the rankspace for the topology of this particular contest.
Out of 9 combinations, ‘A’ is the pick in 3 of them. We might start by assigning first round probabilities to each candidate. Let’s say A, B, and C ran very close to each other near 33%-up, and we definitely need to use second round votes to address the winner. With a nail-biting first-round, we are now deciding the race on second-ranked votes. Suppose that by a little bit, C is the first to drop, and we are in the race for being the second choice pick of C-voters.
What information are you storing about where or whether voters are encoding preferences found in the ballot combos from lines 11 and 12 in the table above?
| 11 | C | A |
| 12 | C | B |
I support the left”
After all, these voters are about to ultimately determine your election in the new rankspace; it will not be decided by the voters who picked you and/or B first, though they lifted you into the second round. Which ballot combination will be more prevalent? Which type of voter will turn out in greater force? When do you want to find out? Are you involved in this process?
For instance, this hypothetical was based on A meeting B after C drops in the first round. Why was it C that dropped? Perhaps A would prefer if B dropped first so they could meet C in round two, because the distribution of…
| B | A |
| B | C |
…is more favorable to Candidate A. This knowledge could impact how A engages with B and C as well as their supporters during the campaign. If you let it.
If a campaign has ten candidates (A-J) but is definitely soluble on the second round, that’s a mere 90 ballot combos to confront. It grows to 720 when a third round is needed, however. Exactly how gritty do you intend to get with understanding coalition formulation on your turf?
Most people in the trade say that the instructions they give is when you encounter a voter who is not with you, you remove them from the universe. We’d like to help you rethink that.