On a general theory of STV applicability, and the Uber Open List

The main idea: STV needs to be amended into the Uber Open List when applied to factionalised party settings. 

0. Posts on this blog are ranked in decreasing order of likeability to myself. This entry was originally posted on 03.12.2021, and the current version may have been updated several times from its original form.

1 A general theory of STV applicability

1.1 The Single Transferable Vote is a remarkable family of voting systems, claiming universal applicability across a wide range of scenarios, from picking where to eat out to picking a Legislature.

1.2 Nevertheless, in practice the claim comes just short, and STV requires some fine-tuning to live out to its true potential.

1.3 Let's distinguish voting scenarios by the degree of politicisation, which I’ll define for my purposes here as the likelihood of correctly predicting the subsequent preferences of one’s vote from the voter’s first preference.

1.4 At one end, the predictive power of first preferences is low, and you have your non-political choices, like where to eat out and stuff like that. This shall be my Non-Political Scenario.

1.5 At the other end of the spectrum, you have such a thoroughly factionalised public as to allow you to confidently predict one’s full list of preferences from his first preference only. Think above the line Senate voting Down Under, where this is literally, mechanically the case. This shall be my Political Scenario.

1.6 Finally, you have an intermediate scenario, where predictive power is there but not total, in practice corresponding to a “factions within parties” scenario, where voters care for both political affiliation and individual candidates to some degree. This is my Middling Scenario.

1.7 I propose that is precisely this Middling Scenario the one in which STV fails to fully perform, and has proven the system’s undoing in Malta. But before we get there.

2 STV in a Non-Political Scenario

2.1 STV works just fine and as advertised in these scenarios, with more complex counts like Warren performing best.

2.2 If hand counts are a must, you can have an STV as originally envisioned by Andrae, where you set the quota based on votes cast, and stop counting first preferences when the candidate reaches the quota. By the end you’ll have a bunch of candidates with one quota of support and some others with less than that, among whom you pick however many winners you need in descending order or votes.

2.3 At some loss in fidelity, this count needs only one pass of the ballots, but cannot be counted in precincts. Happily, Non-Political scenarios are rare enough that it is quite often feasible to fit all voters in a single room.

3 STV in a Political Scenario

3.1 Here too STV performs as advertised, but may be too complex for what is required. After all, if what I really need to know is your first preference and then I can confidently fill out the rest of your ballot, I can just solicit that first preference.

3.2 In a Political Scenario, you’d ask candidates to pre-emptively and publicly rank all other candidates they care to rank, and ask for a simple vote-for-one ballot from the voter.

3.3 You can run the most complex counts on a mere hand count now, as all the required tables can be pre-filled and ready to go. 

3.4 If all party candidates rate themselves top and then proceed with the same uniform ranking of all other candidates, you get semi-open proportional voting, which is not bad. But unlike with proportional voting, you can now confidently vote for the fringiest of fringe parties with no fear that your vote shall ever be wasted.

3.5 These fringe parties would then have some form of power, short of representation but exceeding what they have now, the power to siphon votes to the winners. 

3.6 Since we can now adopt the best STV counts such as Warren and never waste a single vote, quota size is not an issue anymore and you can up the threshold as much as you like really. 5%? Nah. How about 10 or 15%? That will easily solve the huge issue that party-list systems have with extreme fractionalization.

3.7 But one can think bigger. Instead of a threshold, just go all in: set a maximum number of seat-winning parties and leave it at that! No more than, say, 4 parties will win seats. All the little parties who don’t win are still funneling votes your way, so you still have to account for them policy-wise, but not vote-trading-wise, as per 3.5.

3.8 4 is not random, but my idea of an ideal number of parties assuming a linear political spectrum. You have your two centrists and two extreme parties of each side, and can have 3 possible governing coalitions (centre vs fringe, left v right and right v left), depending on how far from the median voter the spectrum has moved. But if you don’t like 4, one can go back to just 2.

3.9 So, there seems to be a way to get nation-wide proportionality without extreme fractionalization: just rank your parties instead of picking one.

4 STV in a Middling Scenario

4.1 It is here that STV falls short, and the reason is simple: you cannot vote for a party and then still influence who among the candidates of another party wins. In a Middling Scenario, this is a biggie.

4.2 Think of all the scenarios in which a certain party’s victory is a given. If you happen to despise that party, you just can’t bring yourself to vote for them but, realistically, by doing so you give away any political power. If Party A is to win, I’d still like to signal my disapproval of the joint, whilst still being able to influence who among the internal factions (which are bound to exist if the party’s victory is a given) comes out on top.

4.3 To solve this, you need the Uber Open List: separate the party vote from the candidate vote and, crucially, allow the voter to vote for the candidates of each party separately. Allow the party vote to determine how many seats a party wins, but allow all voters’ second votes to determine who gets to fill those seats, regardless of where the first vote went.

4.4 Whilst you don’t technically need either of those votes to be STV, its always better to allow for no vote to go to waste.

4.5 Let’s go for maximum complexity and fairness, which will require a machine count. You divide the country into n/4 constituencies (few enough that its extremely unlikely that at least one candidate won't be elected in each, you’ll see what I mean), where n is the number of seats. Voters are given two unlinked votes, both STV-like.

4.6 The first lists all parties competing nationally. Ballots are tallied as per Warren and - importantly - are counted at the national level.

4.7 Once the nation-wide distribution of votes among parties is settled by this first vote, this distribution is then allocated to each constituency by consulting all ballots weighted by the keep values within each constituency, which gives you party votes and which you can use to allocate seats won between constituencies for each party (use largest remainder, Jefferson or whatever).

4.8 Obviously there'd be no guarantee that each constituency would return the same number of winners overall under this approach, hence it’s important to have few enough constituencies to make any returning none very unlikely. Thus, n/4.

4.9 The second vote is again STV-like, and lists all incumbents competing in the actual constituency, under each party’s header separately. Voters rank within each slate independently and your run Warren to determine who among the several candidates fielded by Party X shall be elected in Constituency Y. Repeat for each party who was allocated any seats in the constituency.

4.10 And here’s the Uber Open List: it does not matter that you voted for Party X, you can still determine who among Party Y’s (and Z’s, and Zh’s, and so on) candidates shall win. People can vote for the individual candidate of choice with no fear of messing up their preferred party. You can help that faction of a party you dislike, without helping the party itself. Centrism if there ever was any!

4.11 Countless systems try to do this, Mixed Member Proportional chief among them, but none allow you to truly vote independently for a guy you like but who happens to run for the party you don’t, and expect not to help that party.

4.12 Is there a way to do the Uber Open List on the cheap, i.e. hand countable?

4.13 Sure, you can run the first vote as simple vote-for-one, but then you can’t go for outrageous thresholds as per 3.6 and would end up with a fractured legislature. So, best to run an implicit Warren as per 3.3 (preferences among parties only would likely conform to a Political Scenario).

4.14 The second vote would then be counted as per Andrae in 2.2.

4.15 All in all one pass of each ballot (but each votes has two), but the first count would be prioritised since it’s the one everybody wants to know about on election night.

4.16 Indeed, there’s no reason to have both votes on the same day, and you can leave a week between them. But I dislike allowing candidates to moderate their tone to seek individual election from the electorate-at-large, so I dislike this idea.