Multi-Member Electoral Systems
This is part of my constitutional design series
This is the second part of my post on electoral systems, this one covering multi-member electoral systems for electing a legislature. You don’t have to have read the first part on single-member electoral systems, but it is helpful.
Don’t Use A Non-Proportional System
A common multi-member electoral system is to have single-member electoral districts and use one of the previous electoral systems to elect them.
There might be some argument for using this method with an electoral system that elects a moderate/Condorcet winner. Even here though, I think if you want a moderate/Condorcet winner, you should just go with a proportional voting system instead, so the moderate ends up being the swing vote in the legislature.
Usually, however, people argue for non-proportional voting system using one of the two-party electoral systems because they don’t want their politics devolving into a bunch of small parties that make governing and therefore accountability democracy harder. But there is a better way.
You can force a proportional electoral system to elect larger parties by adding a threshold, a % of the vote a party must cross to get any seats. If you add a threshold, please use ranked voting so that people’s votes are not wasted if their preferred party doesn’t make the threshold. This is called Spare Vote, and works the same way as RCV does for single-seat elections, but instead of stopping at 1 candidate you stop once every remaining party is over the threshold1. It comes with the same downsides of RCV—for instance you should probably have two rounds if you set the threshold really high—but the downside to just throwing out votes that don’t make the threshold is even bigger than the downside to using FPTP because a lot of people are going to vote third-party in a proportional system.
High thresholds push a country away from consensus democracy and toward accountability democracy. High thresholds also force voters rather than politicians to form coalitions. For reasons I’ve discussed here and elsewhere, I thus generally prefer low thresholds, but it’s nice to have a single parameter to adjust. If you want a two-party system, for instance, you can raise the threshold to 2/3 (or more accurately you’d just seat the top two parties so that a party with >2/3 support doesn’t get all the seats).
You might still favor single-member districts because you want seat share to go up faster-than-proportionally with vote share. Even then, though, you shouldn’t use single-member districts. The relationship between seat share and vote share in them is dependent on the weird geographic property of the partisan distribution of districts—the more swing districts there are, the larger majorities will be. (Also you have to deal with gerrymandering).
The better way to implement this greater-than-proportional relationship is to have some formula to give a boost to the winning party, often called a plurality bonus. For example, you can adopt a two-party electoral system, and then adopt the formula “a party gets an extra 1% seat share for every percent above 50% they get in vote share, up to 55%”. This yields a nice predictable relationship between vote share and seat share.
Now, onto the forms of proportional representation.
List PR
Closed List
Closed list-Proportional Representation (PR) is the simplest. Each party issues a list of candidates before the election, they’re allocated seats proportionally to how many votes they get in the election, and the top x from their list get seated (where x is the number of seats they were allocated).
The upside is that this system is simple and immune to gerrymandering or any other weirdness. There are two main downsides:
Poor candidates can’t be punished as precisely. People can’t vote against bad, maybe scandal-ridden candidates directly, they can only vote against their party as a whole. By the same logic, good candidates can’t be rewarded as precisely.
There are no representatives from swing districts to feel the heat.
An important reason that legislatures act moderately is that politicians from swing districts know that if they don’t vote moderately, they will lose their seats. You might see this phenomenon in party list PR with politicians low down on their party’s list, who know that if their party passes extreme stuff it will lose popularity and they will lose their seat. However, they don’t feel the pressure from voters examining their specific voting record and choosing candidates based on that.
One thing I don’t think is a downside is the lack of geographic districts. If the point of geographic districts is that people can call their representative, national politicians can just, like, hire more staff. I don’t really see what having local districts really accomplishes here, legislators have to hire a lot of staff for the 500,000-person districts we have in the US anyway. If the point of geographic districts is that local communities have someone voting on their behalf, I say that’s stupid and a community shouldn’t have more power if it’s clustered geographically. If a community is big enough it should get national representation, either with a party of its own or through concessions from larger parties.
District-Level List PR
You can mitigate closed list PR’s drawbacks (particularly the first one) by doing list-PR on a district, so that voters have more knowledge of who they’re voting for. Note that you’d still implement thresholds on the national level2, and you’d still combine this with spare vote.
The issue is the number of reps per district. The more you have, the lower voter knowledge of individual candidates will be, so it’s easy to end up back at national-level closed list PR. The fewer you have, the more you sacrifice proportionality, and the more vulnerable you are to gerrymandering.
The optimal seat count also depends on how many parties you expect to have. If you have a lot of parties, you want more seats per district, for two reasons. The first is that proportionality becomes more important, particularly for smaller parties. A party with 5% of the vote may not get any seats in district-level list PR. The second more subtle reason is that when there are fewer parties, there are more reps elected per party, so public knowledge has to be greater for people to make an intelligent decision based on a party’s candidates. When there are lots of different parties each vying for 1 seat or so, a prospective voter can mostly just examine the top candidate for the 2 or 3 parties they plan on voting for.
Let’s use Ireland as a case study for how this looks in practice. They have a system similar to district-level PR that I’ll talk about in a bit, and they have 3-5 reps per district with no threshold. Their elections look like this, with 3 main parties and a few smaller parties:
So that’s about a 1-2% difference on average between a party’s seat share and vote share. It’s pretty common for elections to be decided by 3% of the vote, so I think this level of disproportionality is uncomfortably high. Look at 2016: the center-right parties Fina Gael and Fianna Fail combined for 49.9% of the vote but got 56.6% of the vote, enough for a clear coalition that can handle a few defections.
The other thing to note is that the geographic distribution turns out not to punish small parties here. I think this is partially due to small parties not running candidates in districts they have no chance of winning. I do still think this is an issue, but it’s good to see that it’s not as big as you might think.
In Ireland, in the 4- or 5-seat district it’s very rare for parties to elect 2 reps per district, with maybe a 3rd candidate that voters potentially have to consider, while in the 3-seat districts it’s rare for any party to get more than 1 seat with maybe a 2nd candidate in consideration. The 5-seat districts are right on the edge of the level of knowledge that I think you could reasonably expect a lot of voters to have. Any more and it’ll basically be just be list PR.
I’d want to try this out in more places before rendering final judgment, but based on Ireland I think 5 reps per district is probably the magic number. This should be a bit more proportional than the Irish election, while still being somewhat manageable. Unfortunately you are still sacrificing quite a bit of both proportionality and public knowledge, but I think those are bullets you have to bite. You are almost totally eviscerating public knowledge if you force a two-party system with thresholds at 5 reps per district, as there will be like 4 candidates per party and good luck getting voters to consider all of them, so I wouldn’t bother with district-level list PR in that case.
Open List PR
Open list PR is very similar to closed list PR, except voters get a bit more control over the ranking. The most common way is that people can vote for a party and then put a “check” for specific candidates, and then a candidate can guarantee a seat in their party by getting a certain threshold of checks (either in their district or nationally). You can also do this in a negative way, where voters can “X” candidates they don’t want, and then a candidate is denied a spot if enough people X them. You can also consider checks minus Xs as the deciding number.
This functions as a form of pseudo-primaries. I generally think there’s not much point - it gives voters a very small degree of control - but it’s also not very harmful so I guess it’s fine. It does more in district level PR, though at that point you might also consider STV which I’ll talk about in a sec.
Direct Multi-Member Systems
This bucket is of electoral systems that directly elect multiple members in a proportional way. Usually these methods are extensions of single-member electoral systems: they have proportional results if everyone votes strictly by party line, and if not they function similarly to their single-vote counterparts. They are implemented in multi-member districts, which has the same size considerations as district-level PR: I would probably go with 5 votes per district and just not use these systems if you have force a two-party system with thresholds.
The most notable of these systems is Single-Transferable-Vote (STV), which is the extension of RCV. Among the other systems are Schulze STV and CPO-STV, which are extensions of Condorcet ranked methods, and Proportional Approval vote, which is the extension of approval voting. You can read about those methods if you want to understand them, I don’t think it’s worth my time explaining them (especially since I don’t understand the finer details of proportional approval vote or how Schulze STV or CPO-STV work).
These are the only systems that can handle a country where polarization is very low and people care more about individuals than parties. Honestly though if a country gets to that point, voters are probably voting for people that don’t represent them very well on the issues, so I don’t think this is a good thing to encourage.
The other big thing about these systems is that they’re extremely complicated, to explain/implement and to some extent for voters to use. Strategic voters seems like a nightmare to figure out in these systems.
Now let me dig further into STV, the only one that I really understand. Probably a lot of what I say applies to these other systems as well.
STV
STV is very similar to district-level closed-list PR, and is the actual system that Ireland uses.
There are three main differences from closed list. First, voters of a party can choose the order of the candidates in their party. I see this as mostly a bad thing: if you want this behavior you can always just implement it in separate primaries, and there are good reasons you might not want it. Second, swing voters can split their vote up, which can cause a less popular candidate of a party to lose a seat to a different party. This probably doesn’t help in electing moderates (they would be less popular among base voters), but it does help in punishing scandal-ridden candidates, since party voters probably have them further down their ranking and swing voters can just not vote for them. Third, independent candidates can actually win seats.
A note: you can allow parties to give a recommended ordering of their candidates. If most of the party’s voters stick with that ordering, it functions a lot like open list PR, where most of the candidates come from a list but one low down might make it up if voters really like it or vice versa. You can even let voters “check” a party list they want to use it instead of having to manually rank candidates in the same way.
Proportional Multi-Vote Representatives.
I would also be remiss not to take some time to talk about Proportional Multi-Vote Representatives (PMR), a proposal I wrote about a long time ago, which also fits into this category. That system does away with the idea that every representative gets equal vote in the legislature. Instead, each district has X legislator-votes to give away, and every representative gets legislator-votes proportional to how many public votes they got in their district (and are seated if they get 1 legislator-vote or more). You can have as few as 5 legislator-votes per district, but this system can handle a lot more. You don’t even really need districts, you could technically have the entire country be one district (which functions a lot like closed list-PR). The downsides of having many votes per district is that the legislature size can become enormous if votes are spread out among multiple politicians, or if they’re concentrated among few politicians you start to lose many of the advantages of having a legislature to begin with. Probably I’d have districts of like 10 votes each.
As with closed list PR, you don’t have representatives from swing districts who “feel the heat” as much because (as long as they aren’t at 1 legislator vote) they aren’t going to lose their jobs if they do poorly. Although you can still give reps with fewer votes fewer privileges in the legislature, like being in committees, to add some extra incentive.
The other potential downside is that there is probably significantly less turnover in this system if you have 10 votes per district than in other systems. I tend to actually like this for allowing legislators to develop experience serving, but you might not.
I really like this system. It’s a bit speculative, untested, but it seems really good.
Mixed Systems
Mixed systems are systems that have some district-level elections, but also some mechanism for ensuring that a party’s seat share is directly proportional to their vote share.
There are two methods I know of. There’s mixed-member proportional (MMP), where there are district seats usually elected in normal FPTP elections, as well as list seats given to make the party vote proportional and which are selected from a party list. And then there’s dual-member proportional (DMP), where each district elects one legislator by normal FPTP, and then each district also has another legislator which is chosen to make the national results proportional.
In MMP, you can have each district elect multiple reps with a different system such as district-level list PR instead of by FPTP. This fixes the disproportionality of district-level list PR, and the number of list seats should be much smaller than with normal MMP. I really like this system, the main drawback is that it adds some complexity.
A lot of the properties of these systems depend on implementation details. So let me make some broad comments about these systems:
To the extent that a mixed system relies on list seats, it functions similarly to closed-list PR.
To the extent that a mixed system sees candidates from different parties compete against each other in each district, it has the more of the nice properties from multi-member districts—scandals are directly precisely, politicians in swing districts are incentivized to moderate.
To the extent that a mixed system sees candidates from the same party compete against each other in different districts (vying to get more votes in their respective districts), candidates have the incentive to moderate.
To the extent that a mixed system sees candidates from the same party compete against each other in the same district, these systems function like primaries, which as I stated before I think is mostly bad as you can just implement separate primaries if that’s what you want.
If you have a separate party vote and person vote (like in some variants of MMP), combining it with spare vote is sort of annoying, because you probably want ranked voting both in the party vote and the person vote. This is why I prefer single-vote systems; also they’re just simpler. The downside of single-vote systems is that they mess up strategy a bit, because if you are voting for a party because their candidate in your district is good but then their candidate doesn’t get the seat, your vote goes toward their list seat.
MMP is kind of complicated, and DMP is even moreso. MMP has been adopted in a few places, notable Germany and New Zealand.
Recommendations
My favorite system is PMR regardless of threshold. But if you want a more tested system, my favorite with low/no thresholds is closed-list PR, and with a high threshold it’s MMP. If you have high thresholds but not a two-party system (I’m thinking like 10-15%) I would go with MMP with district-level closed or open-list PR, districts of size 3. It has all the properties you could possibly want, and since you have high thresholds there should be very few list seats. It’s a little complicated, but I think once you know how thresholds work it’s really not difficult to see how the list seat part of MMP works.
There’s a different way of implementing it where you just remove all parties that don’t make the initial threshold and reallocate their votes. This however incentivizes parties to join forces if they are polling below the threshold, which defeats some of the point of thresholds, because then it is politicians (rather than voters) who are doing the coalition formation.
You can make it either a vote threshold - a party needs x% of the vote to get any seats - or a seat threshold - a party needs x% of seats before any threshold to get any seats.