Due to the upcoming elections and the surrounding discourse I got interested in the voting system for the SC. I think that I’ve found a undesirable property in the way initial 2-year terms get distributed that results in temporarily non-proportional representation. The argument is as follows:
Let’s assume that the electorate can be split into two camps, A and B, and that A is slightly more numerous than B. In the initial elections that translates to As getting the extra seat resulting in 4A:3B split in SC. Moreover, since As got more votes than Bs, this means that all As get 2-year terms. Thus in the next elections all 3 Bs end their term and, assuming proportional outcome of the vote, get replaced by 2 As and 1 B. This results in a highly non-representative 6A:1B SC.
This will self-correct in the next elections and the SC will become stable and proportional again, however, the self-correcting behavior assumes regular elections going forward. Malicious As could use the above property and the Full Reelections mechanism to introduce changes requiring a supermajority that are unpopular among Bs.
Is this analysis correct or have I missed something? Perhaps I misunderstood how “final election rankings” used for the 2-year term assignments are calculated.
Moreover, since As got more votes than Bs, this means that all As get 2-year terms.
I am not sure if that is actually true. There are more A votes, but they are split across more winning A candidates. I think the As and Bs might actually alternate in the order of candidates by vote.
Mainly I’m not sure either way, but I think this the part of the argument to double check first either way.
All this depends a lot on the details of the distribution, and this is probably the part most unstable when we change any details of the voting system. Pretty sure I can build a situation when all three Bs will have 2-year terms, and it won’t even be too fine-tuned or fragile, just difference in the distribution of voter preferences.
winning candidates with higher final election rankings are appointed to the longer terms.
Is the way to calculate “final election rankings” specified somewhere? I’ve skimmed the CIVS description but AFAICT it talks about comparing sets of candidates and not about ranking candidates within the winning set.
EDIT: Nevermind, I missed that the committees are ordered and not just sets. That makes sense.
I think CIVS compares committees as sets actually. Also, the optimisation procedure as designed does not always achieve what it aims to achieve. And CIVS might be using Condorcet for ranking candidates.
However! This time we use STV anyway… Which has its own set of different catches.
I made a very similar argument internally when trying to amend the constitution to use approval voting. Approval voting has the nice property that you get the same result no matter how elections or positions are staggered. For example, a full reelection has the same outcome as 7 independent elections for 1 seat each.
In contrast, proportional voting has the dilemma where you get differents results/outcomes if you sequence things differently. This incentivizes certain undesirable behaviors:
People in the majority will be incentivized to expel members from the minority one at a time and hold special elections because an election for just 1 seat overwhelmingly favors their side
Staggered elections interact poorly with proportional voting (after the first full election, each following election only represents the views of 3 or 4 factions and not 7)
It politicizes decisions around special elections or full re-elections (more than they are already politicized) because both committee members and voters have to do this extra calculus for whether or not the new election schedule favors their side or not (for example: a committee member in the minority might make a political calculation to vote against expelling a problematic member because a special election for one seat favors the majority)
Approval block voting fails at proportionality every time. That avoids partial proportionality failures. But committee without proportionality is pointless in our situation.
The point of proportionality is that in a project where different communities involved have never had too much trust nor understanding towards each other since lowering the necessary level of dirty trick knowledge, members of a pure-Condorcet or approval SC do not provide enough coverage to be plausibly not missing the very basic understanding of faction’s needs/preferences/values.
If you want block approval, you can as well go Debian way and select a single leader plus a deputy or two to cover over unavailability. Will spend less time on achieving the same-ish decisions, with very similar failure modes.
i agree staggering and special elections do not converge on proportional representation.
i believe however that this is a bug around staggering and special elections (in the case of special elections as opposed to following votes from broader elections), rather than a bug around proportional representation.
and in fact, as the OP demonstrated, failing to approximate proportional representation facilitates partisan take-overs. i imagine those based in the US here might have an idea of what that could look like in practice.
as for questions about preferences influencing the exact distribution on double terms (fwiw the OP’s example holds under binary polarization), i’m not sure possibilities of the system to destabilize in various directions are a feature here, rather than proving the OP’s point that our current set-up just has failure modes.
Honestly, we have multiple options for continuity which will probably be used (SC has not completely cut communications with NCA members at offboarding, so maybe future SCs will keep talking to recent SC members anyway; SC has to talk to the Board on many topics and Board rotation works differently than SC rotation; well, staggering, if we don’t end up fully dissolving SC right before each election) and hopefully over time some of the SC members of different times can discuss together what seems to be the most working one.
Maybe the staggering will be dropped, maybe reworked. Or maybe some SC decides to give up and try single-leader (approval or Condorcet based, presumably) with forced rotation on unpredicted CoI change first.
And maybe we accumulate enough understanding of tallying systems over the detail-oriented discussions to be able to decide on a specific way of replacement-picking based on the previous election.
