Policy Review Breaking Ties

[DHR-107]

Prompted by a convo with Birkal

Hey all,

After an interesting discussion about the Prevo name thread that led to us (before the later votes) having a tie situation, I wanted to just make sure we are OK with the original Tie Breaking rules we have in place.

The last time this was discussed was nearly 10 years ago in this thread: http://www.smogon.com/forums/threads/policy-review-tiebreaking-procedure.48108/ and it ended with the following options:

Conclusion:

In the case of a tie vote, the following tiebreakers will be used to determine a winner:

1) The number of votes recieved in the single most recent previous poll
2) The sum total of all votes accumulated in all previous polls
3) Topic/Section leader choice

This is echoed on our Process rules page here: http://www.smogon.com/cap/process/process_rules (Which also needs updating due to the addition of new ways to vote).

With the new updates to polling methods (such as using ranked pairs etc), are we still OK with using the same Tie Break rules we have used in the past? For example, in an RP vote, all three options can potentially have the same number of wins/losses when placed against each other, or two options can have two wins a piece due to the way vote ties work (read up on Ranked Paired voting for that one, it took me a while to get my head around). In this instance, how do we determine a winner? I would also like to add the word: "Relevant" to the above, as it could be argued that "previous poll" might not be related to the topic at hand...

As an example, before the "extra" votes in the Prevo name thread, we had a two way tie with two names, where the third name had more votes in the previous round. It took a bit to work out what we should do going forwards, as there were different interpretations to what the votes were showing us. This also starts to lead into a conversation of "How many rounds should a poll go too?" but that is more than likely a different discussion.

If we are perfectly fine with the above, nothing needs to change at all, but I thought it would be worth opening it up to the floor to make sure everyone is still okay with the status quo and ratify it to be more "Official" in the age of TLT and different voting methods.

 

[Airwind]

I don't see why we should change the poll system is a good system and if we change the poll system cap 24 will be delayed

 

[Quanyails]

It looks like the discussion when tiebreaking rules were set were done purely through MBV. We don't use MBV in succession, so a lot of the reasoning for the original tiebreaking rules are thrown out. In addition, they don't make sense when we're applying them to PBV. It's totally possible for a poll to have the ballots:

Poll 1 (PBV):
A > B
A > B
B > A
B
C > A > B

Which would order the results A > B > C.

Then if A and B tie in the next poll, tiebreaking rules prefer B because it showed up on more ballots, even if it was preferred less. It would make more sense to break ties by the ranking of the last poll's results, not the ballot count. This is likely an issue with wording, as votes = ranking for MBV.

This also applies to the second tiebreak, since it favors options that "show up the most" regardless of if they were in first or last place. It was applicable to MBV, but it's wrong for ranked polling algorithms.

Pollkritter has some tiebreakers built in, such as using versions of Borda Count (similar to PBV). QxC4eva might want to chime in on using these tiebreakers instead. :)

We could be lazy and just update the first tiebreak to use the word "ranking" instead of "votes" and update the second tiebreak to only sum MBV votes (because 3+ polls is the only time we'll use MBV). I'd rather have a "correct" tiebreaker, in any case.

We shouldn't start CAP 24 until this ambiguity is resolved, though.

 

[QxC4eva]

Thanks Quany for pointing me to this thread :)

The tiebreaker used in Ranked Pairs right now, called Borda Count, is better than using the previous poll as tiebreaker. Using the previous poll is technically the same as using PBV to tiebreak since that poll is PBV. And PBV is a worse tiebreaker than Borda Count for math reasons I won't go into (unless you really want it!)

So in short, we should just stick to the current Ranked Pairs implementation.

Previous poll tiebreaking would be viable if poll 1 also used Ranked Pairs. That way we'll be using Ranked Pairs to tiebreak a Ranked Pairs poll which also works well.

 

[Quanyails]

Closing this thread as a consequence of this post.