Allow single-winner threshold to be based on first-round votes only #646

artoonie · 2023-03-31T12:21:51Z

Single-Winner races (enabled):

Multi-Winner races (disabled):

Included a simple test to verify.

tarheel · 2023-04-03T07:27:29Z

Is this one ready for review?

artoonie · 2023-04-03T15:22:17Z

@tarheel yep

…single-winner-threshold

tarheel

In addition to (or instead of) the test case you created, could you include one in which the winner fails to reach the threshold? I just want to confirm that that case works, since (I believe) it violates an assumption that we previously held about how thresholds work.

tarheel · 2023-05-07T21:01:40Z

src/main/java/network/brightspots/rcv/Tabulator.java

@@ -158,7 +158,18 @@ Set<String> tabulate() throws TabulationAbortedException {
      // votes in the first round.
      // In a single-seat contest or in the special multi-seat bottoms-up threshold mode, it's based
      // on the number of active votes in the current round.
-      if (currentRound == 1 || config.getNumberOfWinners() <= 1) {
+      boolean doRecomputeThreshold;


super nit: I think shouldRecomputeThreshold would be more clear

artoonie · 2023-05-08T18:00:00Z

Modified test so threshold isn't reached, but winner is still selected based on being the sole remaining candidate.

This may not be the desired behavior: the rounds continue until just one candidate remains, rather than picking whoever has the most votes when two candidates remain. I'm not sure if this is the desired behavior? Should nobody be elected? Should somebody be elected when two candidates remain?

tarheel · 2023-05-08T18:14:37Z

@chughes297 can probably confirm, but I believe the desired behavior would be to stop when there are two candidates remaining and elect the candidate with more votes.

On a related note: it just occurred to me that I'm not 100% sure we handle it properly if you end up with two candidates remaining and they're tied. Ordinarily we go to the tie-breaker logic, but I'm not sure that accounts for this scenario...

artoonie · 2023-05-08T18:24:31Z

The tiebreaker logic occurs to figure out which of the two candidates to eliminate, but not which of the two to elect. The tabulation still runs down to just one candidate.

I've added another commit with a test for this so you can see how it works.

Would love to hear what Chris thinks. Chris, any word on whether it's okay to run down to one candidate when using this rule?

andyanderson · 2023-05-08T19:33:40Z

Modified test so threshold isn't reached, but winner is still selected based on being the sole remaining candidate.

This may not be the desired behavior: the rounds continue until just one candidate remains, rather than picking whoever has the most votes when two candidates remain. I'm not sure if this is the desired behavior? Should nobody be elected? Should somebody be elected when two candidates remain?

Eliminating the second-place candidate might be necessary to push the first-place candidate over the threshold, so this seems to me to be desirable.

But is there a description somewhere of how this would be an impactful change? There must be some other consequence planned if they haven’t reached the threshold, e.g. nobody elected, rerun the election, etc., because once a candidate reaches a majority of the vote they will always have a majority of the vote.

tarheel · 2023-05-08T19:37:31Z

But is there a description somewhere of how this would be an impactful change? There must be some other consequence planned if they haven’t reached the threshold, e.g. nobody elected, rerun the election, etc., because once a candidate reaches a majority of the vote they will always have a majority of the vote.

Ha, you would think. But apparently it's just something that's now required in Minnesota even though it has no impact on the outcome.

endolith · 2023-05-09T20:40:17Z

because once a candidate reaches a majority of the vote they will always have a majority of the vote.

Ah, but if you eliminate the 2nd candidate, the winner is now guaranteed to always have unanimous support! 🤣

andyanderson · 2023-05-09T22:41:54Z

Thinking more about this, I can see that this requirement of surpassing the original threshold is going to result in strategic voting in the form of candidates encouraging their supporters to “rank me first and don’t rank anyone else, because the more exhausted ballots the less likely it is that another candidate might win”. A bad, bad idea, Minnesota.

cjerdonek · 2023-05-10T03:38:09Z

because once a candidate reaches a majority of the vote they will always have a majority of the vote.

Ah, but if you eliminate the 2nd candidate, the winner is now guaranteed to always have unanimous support! 🤣

I was randomly looking at issues and saw this. Note that even if you eliminate the 2nd-place out of two candidates left, it's still possible that the winner won't reach the threshold established in the first round (for example, if none of the ballots for the losing candidates rank the other candidate further down).

artoonie · 2023-05-13T20:18:40Z

@cjerdonek I agree, but I think this has some bizarre consequences.

I've added a commit which feels very dirty to me, but is probably the correct way of doing things...

If nobody reaches the threshold, it doesn't mark anybody as a winner
But, as a natural consequence, the lowest vote-getter should be eliminate
Therefore, there's a final round with no candidates remaining

Here's a visualization, with a table below.

Is this what we want? This feels better than leaving the last-vote-getter uneliminated if they didn't reach the threshold, since that looks too similar to the last-vote-getter being elected.

@chughes297 can you look at these visualizations and let me know if you think that's how the code should react when nobody reaches the threshold?

chughes297 · 2023-05-19T16:40:22Z

@chughes297 can probably confirm, but I believe the desired behavior would be to stop when there are two candidates remaining and elect the candidate with more votes.

On a related note: it just occurred to me that I'm not 100% sure we handle it properly if you end up with two candidates remaining and they're tied. Ordinarily we go to the tie-breaker logic, but I'm not sure that accounts for this scenario...

This, from upthread, is the desired behavior. It's functionally only a change in how the results information is reported - all the Minnesota RCV jurisdictions require the results to report winners based on the threshold set by valid first-round votes. It's essentially IRV reported using STV threshold-setting rules. Which means that even if no one crosses threshold the person with the most votes at the end wins. The final round is always a head-to-head matchup. Minneapolis Mayor 2021 Results example: https://vote.minneapolismn.gov/media/-www-content-assets/documents/2021-Mayor-Round-Results-Summary.pdf.

Sorry for not including an example originally! Would have saved a lot of back and forth.

chughes297 · 2023-05-19T16:43:15Z

Oh sorry just noticed the tiebreaker question - I would treat a top-two-candidates tie as a tie between candidates to be elected. Does that answer that question?

tarheel · 2023-05-19T16:46:48Z

Oh sorry just noticed the tiebreaker question - I would treat a top-two-candidates tie as a tie between candidates to be elected. Does that answer that question?

Sorry, I'm not exactly sure what you mean. I think the desired behavior should be to run the tie-breaker and then stop and declare the winner of the tie-breaker to be the election winner. (In any earlier round, we would run the tie-breaker and then eliminate the loser, but that doesn't make sense in this context because it would lead to an additional round in which only the winner remains.) And I haven't confirmed, but I don't believe that's what will happen with the code as it currently stands.

chughes297 · 2023-05-19T16:50:11Z

Oh sorry just noticed the tiebreaker question - I would treat a top-two-candidates tie as a tie between candidates to be elected. Does that answer that question?

Sorry, I'm not exactly sure what you mean. I think the desired behavior should be to run the tie-breaker and then stop and declare the winner of the tie-breaker to be the election winner. (In any earlier round, we would run the tie-breaker and then eliminate the loser, but that doesn't make sense in this context because it would lead to an additional round in which only the winner remains.) And I haven't confirmed, but I don't believe that's what will happen with the code as it currently stands.

Yes, the desired behavior is to run the tie-breaker and then stop and declare the winner of the tie-breaker to be the election winner. I agree it will look weird in the results as people discussed elsewhere, but that's something that's better left to a text-based explanation included with the results, IMO.

This reverts commit 666db28.

artoonie · 2023-05-21T00:26:27Z

I reverted the last commit to return to the original functionality: when there are two remaining candidates, the one with the most votes wins.

We can check tiebreaker consistency as part of #666. I believe all comments here are now resolved?

HEdingfield

I'll rely on @tarheel to give the final approval, since he was much closer to this than I was. Just a couple questions from my end.

src/main/java/network/brightspots/rcv/ContestConfig.java

src/main/java/network/brightspots/rcv/GuiConfigController.java

tarheel

I think the only potential issue here is that when we end up with two candidates remaining and neither one meets the threshold, we're not actually supposed to generate a final round in which only the winning candidate remains; instead, the tabulation should behave the same way it does when the winner meets the threshold, i.e. they get elected and the tabulation stops without eliminating the runner-up. This has no impact on who wins, of course, but IIRC it's a nitpicky detail that's been flagged in the past by someone or other. And ditto if we need a tie-breaker to determine the winner at this point.

You can confirm with @chughes297 that my understanding here is accurate...

src/main/java/network/brightspots/rcv/Tabulator.java

src/main/resources/network/brightspots/rcv/hints_winning_rules.txt

artoonie · 2023-06-01T18:18:18Z

@tarheel to get us all on the same page --

Current Behavior

Your Suggestion

@chughes297 etc, can you chime in? Which of these should RCTab produce?

tarheel · 2023-06-01T21:08:37Z

Except in the second image I would want Vail to have the green highlight in the last round to indicate that he won.

chughes297 · 2023-06-01T21:16:25Z

Except in the second image I would want Vail to have the green highlight in the last round to indicate that he won.

agree!

artoonie · 2023-06-05T18:13:28Z

Done.

Without tie-breaker

With tie-breaker

tarheel · 2023-06-09T19:31:32Z

src/main/java/network/brightspots/rcv/Tabulator.java

+          // Edge case: if nobody meets the threshold, but we're on the penultimate round when
+          // isFirstRoundDeterminesThresholdEnabled is true, select the max vote getters as
+          // the winners.


Your comment could also note that if isFirstRoundDeterminesThresholdEnabled isn't enabled, it should be impossible for a single-winner election to end up here.

tarheel · 2023-06-09T19:34:04Z

src/main/java/network/brightspots/rcv/Tabulator.java

+      // defer the others to subsequent rounds. If this is a single-winner election where its
+      // possible  for no candidate to reach the threshold (i.e. "first round determines threshold"


super nits: s/where its/in which it's/ and s/ / /

tarheel · 2023-06-09T19:34:30Z

src/main/java/network/brightspots/rcv/Tabulator.java

      // Edge case: if we've identified multiple winners in this round, but we're only supposed to
-      // elect one winner per round, pick the top vote-getter and defer the others to subsequent
-      // rounds.
-      if (config.isMultiSeatAllowOnlyOneWinnerPerRoundEnabled() && selectedWinners.size() > 1) {
+      // elect one winner per round, pick the top vote-getter. If this is a multi-winner election,


Just for clarity, make this a bullet list:

Edge cases: - (Case 1) - (Case 2)

tarheel · 2023-06-09T19:37:44Z

src/main/java/network/brightspots/rcv/Tabulator.java

+      boolean useTiebreakerIfNeeded = config.isMultiSeatAllowOnlyOneWinnerPerRoundEnabled()
+          || config.isFirstRoundDeterminesThresholdEnabled();


What about the case in which it's single-winner and isFirstRoundDeterminesThresholdEnabled is false but we end up with two candidates who both have exactly half of the votes?

If that's not already handled, we should add a test for it, and I would recommend tweaking this to something like:

boolean tiebreakNeeded = selectedWinners.size() > 1 && (config.getNumberOfWinners() == 1 || config.isMultiSeatAllowOnlyOneWinnerPerRoundEnabled());

That would lead to one candidate being eliminated, so it wouldn't be handled here -- since nobody has reached the threshold, the tiebreaker runs at the elimination stage.

The test for this is in tiebreak_seed_test.

Ah, so I think that test is wrong! It eliminates the tiebreak loser and runs a final round with only the winner continuing. I think we want to make that behavior consistent with what you've done here. @chughes297 can confirm.

To be specific, here's what that test is currently expecting:

{ "config" : { "contest" : "Tiebreak test", "date" : "2017-12-03", "jurisdiction" : "Funkytown, USA", "office" : "Sergeant-at-Arms", "threshold" : "2" }, "results" : [ { "round" : 1, "tally" : { "George Gervin" : "3", "Mookie Blaylock" : "3", "Yinka Dare" : "3" }, "tallyResults" : [ { "eliminated" : "Yinka Dare", "transfers" : { "exhausted" : "3" } } ] }, { "round" : 2, "tally" : { "George Gervin" : "3", "Mookie Blaylock" : "3" }, "tallyResults" : [ { "eliminated" : "George Gervin", "transfers" : { "exhausted" : "3" } } ] }, { "round" : 3, "tally" : { "Mookie Blaylock" : "3" }, "tallyResults" : [ { "elected" : "Mookie Blaylock", "transfers" : { } } ] } ] }

And I think what we actually want is for round 2 to be the final round, and in that round no one is eliminated but we elect Blaylock via the tiebreaker.

And I think what we actually want is for round 2 to be the final round, and in that round no one is eliminated but we elect Blaylock via the tiebreaker.

Yes!

To keep this issue well-scoped, and since that's a bug that exists in develop today, I've created a separate issue to address it: #694

HEdingfield

I'll plan to give this a quick review on the UI stuff once @tarheel is satisfied with everything.

HEdingfield

Approved for UI changes.

artoonie added the v2.0 requirement label Mar 31, 2023

artoonie linked an issue Mar 31, 2023 that may be closed by this pull request

Single-Winner Threshold Selection #598

Closed

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from 1fc3842 to 7af25a3 Compare March 31, 2023 12:23

allow single-winner threshold to be based on first-round votes only

e824b17

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from 7af25a3 to e824b17 Compare March 31, 2023 12:26

Merge remote-tracking branch 'origin/develop' into feature/issue-598_…

e79080a

…single-winner-threshold

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from d7a6e55 to e79080a Compare April 3, 2023 15:24

HEdingfield requested review from tarheel and HEdingfield April 17, 2023 06:49

tarheel reviewed May 7, 2023

View reviewed changes

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from 311c1d6 to 8300593 Compare May 8, 2023 18:04

PR comments: fix test, improve test, rename var

8c02dba

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from 8300593 to 8c02dba Compare May 8, 2023 18:13

add tiebreaker test

1b172ea

Allow elimination of all candidates if nobody meets threshold

666db28

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from f12da08 to 666db28 Compare May 13, 2023 20:16

tarheel mentioned this pull request May 19, 2023

Confirm that tabulator works properly if we end up with two candidates who are tied #666

Open

Revert "Allow elimination of all candidates if nobody meets threshold"

42726bd

This reverts commit 666db28.

HEdingfield reviewed May 21, 2023

View reviewed changes

src/main/java/network/brightspots/rcv/ContestConfig.java Show resolved Hide resolved

src/main/java/network/brightspots/rcv/GuiConfigController.java Outdated Show resolved Hide resolved

artoonie force-pushed the feature/issue-598_single-winner-threshold branch 2 times, most recently from ab7b78f to 05235e3 Compare May 21, 2023 03:46

tarheel reviewed May 28, 2023

View reviewed changes

src/main/java/network/brightspots/rcv/Tabulator.java Outdated Show resolved Hide resolved

src/main/resources/network/brightspots/rcv/hints_winning_rules.txt Outdated Show resolved Hide resolved

PR comments

eb5a352

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from 05235e3 to eb5a352 Compare June 1, 2023 18:33

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from 6c41305 to 4b88733 Compare June 5, 2023 18:16

don't go to final-round with 1 candidate

3154b29

artoonie force-pushed the feature/issue-598_single-winner-threshold branch from 4b88733 to 3154b29 Compare June 5, 2023 18:21

tarheel reviewed Jun 9, 2023

View reviewed changes

update comments

ab91f43

HEdingfield requested changes Jun 9, 2023

View reviewed changes

artoonie mentioned this pull request Jun 9, 2023

Fix final-round tiebreaker #694

Closed

tarheel approved these changes Jun 9, 2023

View reviewed changes

HEdingfield approved these changes Jun 12, 2023

View reviewed changes

HEdingfield merged commit 195aac2 into develop Jun 12, 2023
1 check passed

HEdingfield deleted the feature/issue-598_single-winner-threshold branch June 12, 2023 17:16

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Allow single-winner threshold to be based on first-round votes only #646

Allow single-winner threshold to be based on first-round votes only #646

artoonie commented Mar 31, 2023

tarheel commented Apr 3, 2023

artoonie commented Apr 3, 2023

tarheel left a comment

tarheel May 7, 2023

artoonie commented May 8, 2023 •

edited

tarheel commented May 8, 2023

artoonie commented May 8, 2023

andyanderson commented May 8, 2023

tarheel commented May 8, 2023

endolith commented May 9, 2023

andyanderson commented May 9, 2023

cjerdonek commented May 10, 2023

artoonie commented May 13, 2023

chughes297 commented May 19, 2023 •

edited

chughes297 commented May 19, 2023

tarheel commented May 19, 2023

chughes297 commented May 19, 2023 •

edited

artoonie commented May 21, 2023

HEdingfield left a comment

tarheel left a comment

artoonie commented Jun 1, 2023 •

edited

tarheel commented Jun 1, 2023

chughes297 commented Jun 1, 2023

artoonie commented Jun 5, 2023

tarheel Jun 9, 2023

tarheel Jun 9, 2023

tarheel Jun 9, 2023

tarheel Jun 9, 2023

artoonie Jun 9, 2023

tarheel Jun 9, 2023

chughes297 Jun 9, 2023

artoonie Jun 9, 2023

HEdingfield left a comment

HEdingfield left a comment

		// defer the others to subsequent rounds. If this is a single-winner election where its
		// possible for no candidate to reach the threshold (i.e. "first round determines threshold"

		boolean useTiebreakerIfNeeded = config.isMultiSeatAllowOnlyOneWinnerPerRoundEnabled()
		\|\| config.isFirstRoundDeterminesThresholdEnabled();

Allow single-winner threshold to be based on first-round votes only #646

Allow single-winner threshold to be based on first-round votes only #646

Conversation

artoonie commented Mar 31, 2023

Single-Winner races (enabled):

Multi-Winner races (disabled):

tarheel commented Apr 3, 2023

artoonie commented Apr 3, 2023

tarheel left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

artoonie commented May 8, 2023 • edited

tarheel commented May 8, 2023

artoonie commented May 8, 2023

andyanderson commented May 8, 2023

tarheel commented May 8, 2023

endolith commented May 9, 2023

andyanderson commented May 9, 2023

cjerdonek commented May 10, 2023

artoonie commented May 13, 2023

chughes297 commented May 19, 2023 • edited

chughes297 commented May 19, 2023

tarheel commented May 19, 2023

chughes297 commented May 19, 2023 • edited

artoonie commented May 21, 2023

HEdingfield left a comment

Choose a reason for hiding this comment

tarheel left a comment

Choose a reason for hiding this comment

artoonie commented Jun 1, 2023 • edited

Current Behavior

Your Suggestion

tarheel commented Jun 1, 2023

chughes297 commented Jun 1, 2023

artoonie commented Jun 5, 2023

Without tie-breaker

With tie-breaker

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

HEdingfield left a comment

Choose a reason for hiding this comment

HEdingfield left a comment

Choose a reason for hiding this comment

artoonie commented May 8, 2023 •

edited

chughes297 commented May 19, 2023 •

edited

chughes297 commented May 19, 2023 •

edited

artoonie commented Jun 1, 2023 •

edited