Should answer scores decay? Discussion on making answer rankings for older questions more dynamic

Question

Folks, this has nothing to do with reputation. It’s about ranking answers. Please stop trying to close it as a duplicate of one of the decade old discussions about reputation.

I think it would be beneficial if the ranking of very old answers that have already accumulated a lot of votes was more dynamic. There are many ways to implement a more dynamic ranking. One of them would be to have the scores we're using to rank the answers decay either over time or based on some activity measure, so that new votes, which tend to be more informed than old votes¹, would have more impact on the ranking.

I'd like to discuss whether allowing answer scores to decay toward zero would be a useful to make that ranking more dynamic, and if so, talk about ideas on how we might go about it without causing larger issues than the ones it could solve. This is not a proposal for a change. Think about it more like brainstorming. Stack Exchange is founded on collaboration, so let's collaborate and entertain some crazy ideas. Maybe the discussion will spawn some ideas that aren't crazy.

Thinking about the discussions around pinning accepted answers and dealing with obsolete answers, I started wondering why twelve year old votes (up or down) on posts stick around forever. I don't have a specific proposal, so this isn't a feature request (I would be surprised if some version of this hasn't already been suggested). I think there has been talk of weighting the score of an answer based on the age of the votes, but I wasn't able to find anything about letting them age away.

It seems to me that if answer scores were more dynamic, it would be easier to solve the outdated answer issue and we wouldn't need to pin an answer to the top of the list to overcome popular-but-wrong answers with insurmountable scores. Allowing votes to decay over time would also mitigate that early votes (and the choice of accepted answer) may be cast with knowledge of only a fraction of all the answers a question might get in its lifetime. If we let answers decay to a zero score, we could also reset certain questions back to "unanswered" if there is no accepted answer and have the community bot bump them for the community to take another look at.

I don't think there should be any changes to anyone's reputation or badges as votes decay; What has been earned should stay earned and I don't see that forcing people to constantly earn reputation to keep their privileges has much benefit compared to offsetting the huge advantages in score old answers have over newer ones.

In my opinion, only the answer scores should need to be freshened up with new votes. Questions don't compete for visibility using score; they can be bumped and bountied for example. I think both downvotes and upvotes should decay. I feel that post activity (views, comments and edits) should affect how fast votes decay, but I can see a lot of tricky bits around that. Maybe whether the user who posted the answer is still active would be a factor to consider when decaying the score? Active users might be more likely to keep their answers updated, so decaying the votes might be less useful?

What do you think? What factors would you use to mitigate the momentum of old answers that have had a lot of views and votes? What factors should speed up or slow down the decay of votes on an answer if we did let votes decay? Would the decay possibly helping users get out from under an answer ban be a good thing or bad thing?

---

^{1. Newer votes are more informed because ideally the question would be clarified and improved over time and there would be more answers to look at when deciding whether a particular answer should be upvoted or downvoted.}

Answer 1

I see a few downsides. Maybe they're not something that can't be overcome, but they are things that should be kept in mind when decaying votes in my opinion. But at the moment I don't see a way to fix these issues, which makes me lean towards saying that decaying votes on answers isn't a great solution and shouldn't be done.

Decaying votes but not reputation.

This isn't necessarily a bad idea, but it should come with an extra caveat: People should get only get to vote for each post once, just like now.

This should remain as such to avoid abuse of the system and people using decaying votes to vote + award reputation more than once and/or tactical voting to keep posts on top that don't deserve it. Seriously, this is stuff that could be scripted and turned into a business model if implemented otherwise: I promise to vote again for your post once it ages away, just pay me $1 a month. So it needs to be kept at 1 vote, only 1 opportunity to 'gift' someone reputation and a chance of taking the spotlight.

Exhausting the pool of people that can keep a great answer at the top

Following the above, if people only get to vote for a post once, eventually the pool of people that can vote for a post to keep it stuck to the top will dwindle to almost nothing. There are great answers out there, that don't lose their value over time. Think stuff like how to exit vim or how to parse HTML with regex. If thousands of people have already voted for those, they will eventually end up aging back to a low score just because you can't vote twice, and the pool of people that can vote for a specific answer is dwindling. I don't think that is right.

Decaying both upvotes and downvotes

There are plenty of questions on Stack Exchange, and I bet a lot of those may get some views but little voting over years of time. Especially not since it takes more reputation to earn the privilege to be able to downvote than it takes for the privilege to upvote. So, these questions may very well have a great answer and a lousy answer, but because not enough people come across it, both may eventually end up scoring 0. On top of that, it is my subjective observation that users in general seem to be upvoting much more than downvoting.

If you really want to use votes to rank answers, to get the best answers to float to the top, decaying all scores back to 0 eventually isn't going to help in the long run. It's going to end up closing the gap between good and extremely bad answers too fast, the ones that need downvotes just won't get enough of those to keep this feasible.

(BTW, are we getting the reputation penalty for downvoting an answer back if our vote ages away?)

Answer 2

There are several issues with this proposal

• Not all answers age equally
• Not all answers are in high traffic tags (this is especially visible on Stack Overflow where differences between tags are huge)
• Not all aged answers are irrelevant (just because some technology is old, doesn't mean it is not used anymore)
• Post activity is very poor representative of the post relevance and accuracy - even recent voting patterns may be highly inaccurate (imagine serial downvotes or upvotes on answers with low overall score in low traffic tags)

Votes are cast by humans. Some votes are cast on poor answers, but there is very little we can do about it, except leaving comments, editing answers with warnings, and downvoting. Still, votes on majority of answers are representative.

Adding any kind of automatic process would cause more issues than it would solve. If answers require any kind of additional quality control, then this must be strictly manual, human reviewed process, by knowledgeable people in the topic.

When it comes to technical sites, most notably Stack Overflow my answer to this proposal is strong:

NO

Answer 3

This idea might be feasible (even though I am not sure I like it), if it is connected not to time, but to the number of views (which is somewhat proportional to time, but has more granularity), preferably making a distinction between the views from people who CAN vote vs other views.

---

Connecting the answer score simply with time does not seem like a good idea, particularly, because it is very hard to adjust the behaviour for niche subjects. There, the votes are much more scarce, and thus, should probably have a longer-lasting effect on the score.

One way to mitigate it is to connect the "decay factor" to the number of views. Say, with every X views, the scores decay by 1, where parameter X should be chosen differently for each community, and maybe, even by tags.

Now, I would differentiate the views by those individuals who have the privilege to vote up (denote by Y) and those who do not (or are not logged in; denote by Z). Then X = Y+Z/F, where F is a "discrimination factor". Careful analysis is needed to determine appropriate X, Y, Z, and F — or any other meta-parameters in the score decay model.

However, to me, the need to do all those calculations points to the problems in the original premise that the score should naturally decay at all without users explicitly interacting with a particular answer.

---

Note, this is just my poke at the proposed idea :)