How does MATLAB Answers break ties among answers?

In a recent question, with no votes for any of the answers, I found some answers jumped around all over the place for no apparent reason while others stayed where they were. How does MATLAB Answers order questions with the same number of votes?

Answer 1

[Randy Souza]

The accepted answer is always listed first.

After that, answers are sorted by the number of votes, then by when the answer was created (oldest-to-newest).

This can cause "jumping" when an older answer is voted up to have the same number of votes as a younger answer. For example:

Answer 2 is listed above answer 1 because it has more votes.

 +--------+-------+---------+
 | Answer | Votes | Created |
 +--------+-------+---------+
 | 2      | 2     | 00:01   |
 +--------+-------+---------+
 | 1      | 1     | 00:00   |
 +--------+-------+---------+

Answer 1 receives a vote, and appears to jump over answer 2 because, even though they now have the same vote count, answer 1 is older.

 +--------+-------+---------+
 | Answer | Votes | Created |
 +--------+-------+---------+
 | 1      | 2     | 00:00   |
 +--------+-------+---------+
 | 2      | 2     | 00:01   |
 +--------+-------+---------+

[Walter Roberson]

Randy, then why is http://www.mathworks.com/matlabcentral/answers/994-wish-list-for-matlab-answer-sections#answer_7039 with 0 votes, created 4 days ago, in the middle of the 0 votes section of that Question? Above it is 2 months ago (0 votes) followed by 19 days ago (0 votes) followed by minutes-older (#7038), then below that answer are other answers created about 1 month ago.

[Randy Souza]

Walter, Thanks for catching that. I will look into it, and get Answers to align with the behavior described above.

[Walter Roberson]

Heh, voting for that answer was the cheap way out ;-)

This behaviour was described about a month ago, http://www.mathworks.com/matlabcentral/answers/994-wish-list-for-matlab-answer-sections#comment_2602

[Andrew Newell]

Randy, it is true that the newest answer is always at the bottom, but the order of the preceding answers keeps changing, even if none of them have votes. Chronological order would be the best choice, so people can follow the thread.

[Andrew Newell]

Walter, before I posted this question I searched the wish-list page for anything related to this subject and didn't find your comment! The page has become too large to be useful.

[Andrew Newell]

Randy, I am accepting this answer on the assumption that you will align the behavior as you described in the comment above.

[Walter Roberson]

Ah -- after the above-noted answer was voted up 1, and no longer appears in the 0 votes portion, it now appears in the middle of the pack of 1 vote answers, with older answers on each side of it.

Answer 2

[Walter Roberson]

Experimentally:

The newest answer with 0 votes and no comments, goes to the bottom, but once another answer comes in or the new answer gets 2 comments, the answer gets placed with the rest of the answers with 0 votes, with those answers sorted with the "most active" one first. I haven't pinned down what "most active" is, but answers with newer activity are generally considered "more active" than older ones... it gets fuzzy after that.

[Andrew Newell]

But the first answer seems to stay at the top.

[Walter Roberson]

I'm not certain of that. I believe I've seen some answers get voted in to the leading place. An Accepted answer will usually go to the leading place, but that's because it usually has the most effective votes.

[Andrew Newell]

I meant, if none of the answers have votes (my question is specifically about the tie-breaking mechanism).

[Andrew Newell]

My impression was that the Accepted answer is always at the top. Have you seen an exception?

[Walter Roberson]

I _thought_ I had seen a case where the Accepted answer was not on top. I had a look about half an hour ago and could not find any posting for which that applied. The answers that had enough votes to have potentially overcome the 4-point acceptance advantage all turned out (in the ones I looked at) to be accepted answers.

I can only recall one Question in which several people felt strongly that the Accepted answer was not the best answer, but I no longer remember the details.