New Sort: "Popularity" (Votes / Views)

Question

The Problem for Answers:

Suppose a new question comes in and has 20 views on it before the first answer is submitted. The first answer is decent, so when the question hits 520 views it has 10 votes, or 1 vote for every 50 views. Also, several other answers have been submitted, but they aren't better than the first answer, so sorting by votes brings the best answer to the top.

However, let's say that a better answer is submitted at 520 views, maybe one that will receive on average 1 vote for every 25 views on that specific answer. On average, when will that answer rise to the top?

1020 views would make sense if the answers were on the same footing for views. However, people do not always read all the answers, and since the better answer (normally) starts at the bottom and has to work its way up, it will take much longer to accumulate enough votes to overcome the top answer since the top answer will get more views than the better answer.

In fact, if half or more of the people who read the first answer leave before reading the second, the better answer will never surpass the number one answer unless it gets lucky!

The Problem for Questions:

I often try to find "popular" and "canonical" posts using the "votes" sort view. This works poorly for several reasons:

• Votes are a function of attention as much as quality, and attention is fickle. For one thing, it is a function of age. Even worse, if a question has ever gone "hot" or been "featured" it will have received hundreds of times the number of views by users who can vote than if it had not. Presumably we want to provide a view that "allows the best posts to bubble to the top" regardless of their age or luck.

• Back to Answers: One can vaguely try to discount votes by views in Question lists because the number of views on the question is listed. However, if you do a Post search there is no indication whatsoever of the number of views, so one can't apply this heuristic to answers (which, in that view, is often what one is most interested in!).

Examples

It's tough to compare question quality, but here's a reasonably controlled example: This question was posed, followed promptly by this corollary question. The second question went hot: As of this writing it had 2400 views and 18 votes. The first question did not: Presently it has 160 views and 3 votes. Is the second question 6 times as "good/helpful" as the first? I don't think so. Is it 133/53 times as good? Arguably. But note how under the current system one drowns out the other due to luck.

Now look at the answers to the two questions: Certainly the accepted answer in each case is the best answer. But on the second question that answer was provided when it was no longer hot, and so it has half the votes of the answer that was there while it was hot. (Fortunately, in this case, the asker of the second question is an experienced and diligent user, so he marked the best answer and it sits at the top. But if he hadn't then the best answer would wallow below the hot answer, and only one person – the asker – could ameliorate that!)

Answer 1

Thus, the first answer is still sitting at 1 vote for every 5 views, whereas Jon's answer is sitting at 1 vote for every 3 views, meaning a greater percentage of people like Jon's answer over the first answer.

Or, it could just be that a greater density of people willing and/or able to vote on the answers were visiting during the most recent period of time, and that the number of votes per number of views really has nothing to do with the quality of the answer.

---

Update: This is being declined, speaking strictly from an implementation standpoint.

While it may look like this would be simple to implement, it's not. Yes, even answers have a ViewCount column because they're all stored in the Posts table, but it would not be simple to populate the column based on the current methods we use for tracking view counts. Even if we could, the count should only ever be taken as a number. View counts are not an accurate measure of anything because it doesn't actually verify someone read something, only that they visited the page. For all we know, they could have just scrolled down to the bottom to type their own answer without reading any of the existing answers.

As well, we would not be displaying that view count anywhere on the answer, which means the criteria for sorting under this popularity tab would be completely hidden from users and seemingly random.

In the end, implementing this would be a lot of effort and added overhead for a feature that probably wouldn't get much use and wouldn't be any more accurate than existing sort orders.

Answer 2

Proposal:

Add "Popularity" as an additional (default?) sorting option. This would essentially allow users to sort on the percentage of people who thought the answer deserved an up-vote.

Ranking Formula:

(upVotes - downVotes) / (answerViews + 1) 

Note: The plus 1 is so we never divide by zero.

Variables:

upVotes = The number of up-votes on the answer

downVotes = The number of down-votes on the answer

answerViews = The number of unique "views" for that specific answer. What constitutes a "view" for an answer should be defined elsewhere, but ideally it should:

1. Only count a user once no matter how many times they view it.
2. Only count when a user actually sees the voting buttons/answer on their screen
3. Only count users that have the ability to cast a vote

Pros:

• In the case of the better answer, after 25 views the better answer would likely get 1 vote and float to the top.
• In the long-run, the best answers will float to the top regardless of when they were submitted.

Cons:

• This has the potential for false positives when the view count is low. For example, if a poor answer comes along that 1 in 1000 would upvote, but the first person to view it is "that guy" and votes it up, it will float to the top for a time until it settles down to its appropriate spot after enough views have been accumulated.

Answer 3

My experience has been the opposite. When I've come to a question later, sometimes much later, and I provide a canonical answer, I've found that it floats up to the top relatively quickly over time.

I suspect you'd gain more traction for this proposal if you gathered a list of questions where it's obvious that a late, better, answer is wallowing.

It's not really a problem

What I've found, though, is that there's only two types of questions that gain late, good answers:

1. Question with only one or two answers.
2. Question with a lot of answers.

It isn't a big deal for #1, because what's usually the case is that users coming to the question actually do read past the first answer - and particularly if it doesn't solve their problem or seems like a lot of effort.

Of course, if it does solve their problem, then why should we artificially promote a later answer?

Lastly, new users don't vote much. It's the people who use the site a lot, are registered, and almost always read all the answers because they want to do it the easiest way and want to understand the problem more deeply. They will upvote the later answer if it's clearly head and shoulders above the first answer(s).

So I don't think questions with few answers really suffer from this.

For questions of type #2, which are popular and already have a lot of answers, this usually indicates that the question itself isn't a great question, and that there probably isn't a real canonical answer. Sure, you may agree with a particular answer more than the others that are highly voted, but it's obviously a question with a lot of different perspectives and there's no real need here to somehow boost later answers.

Further, this is really just a long-tail extension of "The fastest gun in the west" problem, which really isn't a problem for very similar reasons.

The problem with the solution

Giving a boost to late answers - and this is what this does - may be falsely elevating poor answers. Sure, it may boost older posts a little, according to their votes, but it would still result in a disproportionate amount of voting on newer posts and requires significant corner cases to resolve.

So if I add a new answer to a popular question, and get one upvote after ten views, mine might become the highest ranked answer according to this new ranking system. In fact if I have a sockpuppet or three I could very well game the system into forcing my question to the top of this ranking for quite some time. Most answers might be getting 1 vote out of every thousand views for a question with 25k views. I come along with a new answer, it gets two votes within the next thousand views, and suddenly it's top dog. It doesn't take a sockpuppet either - a new answer promotes the question to the recent activity list, other users review it, and they may vote on the other answers, but some users focus on the underdogs, and even if the answer isn't particularly special they hate to see a 0 next to it.

Conclusion

I don't think this is a great idea. I'm not convinced it's a problem because a lot of my late answers do float up. I'm worried it might be easy to game, and that it won't actually prove very useful even if it's used according to your plan.

I expect you can use the data explorer to gather some evidence, though. Get a number of questions where the ranking would be very different under this plan, then analyze them and see if the ranking turns out to be particularly useful. That won't fix the gaming aspect of it, but it would at least show that it is a problem.