Like STAR But Vary The Count of Candidates Who Make It To The Final
-
If you have more than 2 in the final round, doesn't that allow for condorcet cycles, since the last round is pairwise?
-
@keith-edmonds I don't think there are monotonicity implications for having more than 2 candidates in the runoff. Obviously lowering a candidate's score will not help them place in the top cut after the first round, because Score itself is monotonic. In the runoff, lowering a candidate's raw score will either decrease their runoff score and not change the scores of the other ballots (if it was above the minimum score before being lowered) or increase the runoff scores of other candidates by pushing down the minimum for that ballot.
To me, this seems like a dilution of STAR. If the lower end of the top cut is weak (e.g. we have an election with 12 candidates, 3 of whom are serious, and 9 of whom are not), then very few ballots actually get normalized.
A final consideration is that if, for example, there are 3 slots in the runoff, and the two most popular candidates are something like a "center-left vs center-right split", then it would be an advantage for the center-right candidate for the third slot to go to a left-wing candidate as the votes that gave a full score to that candidate will not normalize (unlike in STAR), and the same goes the other way around. Again I interpret this as a dilution of STAR. I also see it as a possible strategic voting issue because if candidates in the lower end of the top cut are not strong enough to be threats to win, then the threat of a strategy of raising favorable match ups backfiring is less of a deterrent, unlike in STAR.
-
@keith-edmonds said in Like STAR But Vary The Count of Candidates Who Make It To The Final:
Seems to me you are proposing STLR but with more than 2 candidates in the run-off.
Not quite. Suppose I vote:
- Candidate 0, score 0
- Candidate 1, score 1
- Candidate 2, score 2
- Candidate 3, score 3
- Candidate 4, score 4
- Candidate 5, score 5
- Candidate 6, score 5
- Candidate 7, score 5
- Candidate 8, score 5.
Let's say that candidates 8, 4, and 3 make it to the second and last round. So before the mapping or manipulation associated to the tally for the last round, the remaining entries on my ballot that mean anything are:
- Candidate 3, score 3
- Candidate 4, score 4
- Candidate 8, score 5.
After mapping, STLR with more than 2 candidates in the run-off maps my ballot to exactly what it already is. There is no change, because I am already using the maximum score, 5. The fact that I am not using the minimum score 0 is no problem for STLR; it does not try to remedy that in its mapping.
But the system I'm describing has to spread the ballot to use the full range (unless it is exhausted). It calculates:
- Candidate 3, score 0
- Candidate 4, score 2.5
- Candidate 8, score 5.
The classical formula for a linear mapping is y=mx+b. Here, m = 2.5 and b = -7.5
My understanding is that we restrict to two rounds instead of many like in Cardinal Baldwin because otherwise it is non-monotonic. I wonder if there are monotonicity implications for having more than 2 candidates in the run-off.
Good question.
-
@rob The last round is scorewise, not pairwise. However, if you look for a cycle, you could find one.
Approval-ordered Llull (letter grades) [10], Score // Llull [9], Score, STAR, Approval, other rated Condorcet [8]; equal-ranked Condorcet [4]; strictly-ranked Condorcet [3]; everything else [0].
-
@marylander Good points. It looks as though I have only invented a complicated way to do plain Score. [edit] Not strictly so, but probably in most of the cases that would turn up in the real world.
-
@jack-waugh Oh so it is scorewise in that you eliminate candidates and normalize the scores. Makes sense. I see that STAR is a special case of that where the number is two. Cardinal Baldwin works like that, but you have the same number of rounds as you have candidates rather than just one round.
I wonder if there is a good way of determining the number of candidates. You don't want completely unelectable candidates to mess things up simply by running.
-
@rob Did you see Marylander's response and my response to his response?
By "number of candidates", do you mean those at the outset, or those that make it into the runoff?
Approval-ordered Llull (letter grades) [10], Score // Llull [9], Score, STAR, Approval, other rated Condorcet [8]; equal-ranked Condorcet [4]; strictly-ranked Condorcet [3]; everything else [0].
-
@jack-waugh I meant number of candidates that make it into the runoff. My point was that it shouldn't be based on the total number of candidates running, since that gives too much significance to irrelevant candidates that get little interest.
-
@rob Right and so I suppose a pretty good answer is always exactly two candidates.
-
@jack-waugh I could also see bringing all "viable" candidates to the run off. Viable would be based on total score relative to max to talk score
Red: 10% vote A:5, B:0, C:0, D:1
Yellow: 30% vote A:0, B:5, C:1, D:0
Blue: 30% vote A:0, B:1, C:5, D:0
Green: 30% vote A:0, B:0, C:0, D:5With regular STLR you would take B and C into the run off and there would be no adjustment to scores
In a version of STLR were you take three into the run off because B,C and D are all "viable". You eliminate A and adjust scores to
Red: 10% vote B:0, C:0, D:5
Yellow: 30% vote B:5, C:1, D:0
Blue: 30% vote B:1, C:5, D:0
Green: 30% vote B:0, C:0, D:5In this case then D wins.
So basically you want to include all viable candidates in the run off.
