# How effective are bounties?

Background: When you add a bounty to a question, it appears on the featured questions list. However, depending on the timing, the featured questions page can be multiple pages long. So for about half the bounty period, your question is near the bottom of the featured list. There isn't much we could do about that, but it is relevant to how effective bounties are. I also notice, that when a question gets a bounty, the amount of views goes up way more than the number of answers. Almost as though, the quality, or experience level, of the viewers goes down.

Questions: Are there any statistics on the differences between bounty question "popularity" and regular question "popularity". Specifically, what are the comparisons between answer amounts, view counts, and acceptance percentages? Ultimately, are bounties worth their cost? If not, how do you make them worth it? Are there alternative things we can do?

(Edit: I've focused this question so it doesn't bring up other topics)

• Someone with some reasonable SQL-fu could easily retrieve some data from the SEDE, as bounties are also catalogued. The key pairs of data would probably be the answer percentage of questions that have had bounties (vs normal) and the view count of questions that have had bounties. Perhaps vote total would also be an indication. Unfortunately, questions with bounties tend to be harder or at least less trivial than the standard question around, so there are fundamental flaws in each of these three criteria. – davidlowryduda Apr 2 '15 at 20:37
• @mixedmath could you look at the posts before they became bounties and then compare them with the post after the bounty was placed? That would definitely allow for direct comparison. I don't know how much data math stack keeps though. – Zach466920 Apr 2 '15 at 20:45
• I wouldn't look at before and after. I would compare questions that have ever had bounties to those that didn't. SEDE also keeps track of when answers were posted, so one could compare when answers to bountied questions were given, I suppose. – davidlowryduda Apr 2 '15 at 20:48
• @mixedmath oh, I only suggested that to get around the fact that the two question types might have different difficulty levels. I'll go look for a tutorial on SEDE. – Zach466920 Apr 2 '15 at 20:55
• Why not segment the market? Let the questions with the highest bounties 500 be shown on top, 250 second, 100 in between, and 50 be shown last. – Mats Granvik Apr 3 '15 at 8:26
• @MatsGranvik That sounds like a good idea. You might have to place limits on how often that could be done to avoid the issue of only people with lots of rep having the ability to have their questions answered. – Zach466920 Apr 3 '15 at 21:43
• That the amount of views goes up is certainly a pro and intended effect of bounties. That the answers don't go up accordingly may simply indicate that a correct answer is still hard to find (a simpler question would have gotten a definite answer within hours if not minutes). -- The value of bounties may be very site-dependent: On other (superhigh traffic) SE sites, where I have little rep, I feel like I must always add bounties (from my hard-earned rep) to my questions simply because they go out of sight too quickly during their bounty-less life. – Hagen von Eitzen Apr 4 '15 at 17:54
• Perhaps it helps more if the bounty is on a newer rather than an older question. For instance, in algebraic-number-theory, a question about factorization in $\mathcal{O}_{\mathbb{Q}(\sqrt{29})}$ had a bounty placed on it looks like two days after it was asked, and it already has three answers not counting my own. A question on cubic rings asked back in February got an answer a couple of days later and a bounty placed on it last week has not gotten the question any new answers. Of course it helps that quadratic rings have been studied much more thoroughly than cubics, plus the 200 to 100 point – Robert Soupe Apr 5 '15 at 1:15
• meta.math.stackexchange.com/questions/3836/… I already noted bounties weren't working over three years ago. The problem has deepened since. – Bruno Stonek Apr 6 '15 at 18:50
• @lentic: What was the problem, and how has it deepened? I see your post, which doesn't convincingly state any problems, but rather asks if your anecdotal observations are more than that, and it received no convincing answers. – Jonas Meyer Apr 6 '15 at 19:10
• @lenticcatachresis this may seem odd coming from the Op, but I disagree with your example. Bounties basically guarantee more exposure, the question is really about whether or not question askers feel that they get better answers. Which, after looking at the data, becomes a very hard argument to make. – Zach466920 Apr 6 '15 at 20:46
• Hi @Zach466920 your observations here are correct, popular-level equation free conceptual questions are clearly preferred on Physics SE, even if Danu and others may deny it . Also, technical questions that contain LaTex get more and more actively and explicitely forbidden by calling them homework (even if they are very high up to research-level) or claiming they are too mathematical. Graduate-level+ technical theoretical physics questions (that contain LaTex/mathematics) can find a better home here... – Dilaton Sep 14 '15 at 17:41
• @Dilaton Thanks, I do sometimes have those kinds of questions, but don't even bother asking because I know what will happen on PSE. Perhaps what's really needed are three sites. One for pop-science, one for undergrad introductory physics, and another for actual mathematical physics. I'll try using that site next time. – Zach466920 Sep 14 '15 at 20:01
• @Zach466920 yes, on sites that are for all levels, the low-level and popular-level contributions and the corresponding people will always dominate the business in the long run, because more people can understand these things than the advanced questions. There are also some people who would like to start an undergrad/introductary brother site for PO, you can find them by searching for "PhysicsUnderflow". You shouldnot get too impressed by what David Z and other people/moderators say. Fact is that the tendency of calling good technical questions homework and closing them gets worse on Physics.SE – Dilaton Sep 15 '15 at 8:58
• Related post on Meta Stack Exchange: Probability of getting an answer by bounty. – Martin Sleziak May 12 at 4:57

(This is an answer to my own question) I looked up some queries and found this. It gave all the relevant statistics about bounties. On average, a bounty will get 100 additional views and 3 additional up-votes. However, the average accept rate for a bounty is 31-48% while for a regular question the acceptance rate is 37-44%. Also, the average question gains the user 20 rep while the average bounty question loses the user 10 rep, factoring in up-votes. Since bounties are usually about finding answers to difficult questions, and the statistics show only marginal increase in acceptance rates, I conclude that bounties, except possibly the 50 bounty, in their current state are not worth the rep.

• Are you considering only bounties by the user who posted the question, or all bounties? If the latter, note that some bounties happen when the OP is no longer active, and naturally have nothing to do with accept rates. Also, would you agree that questions that have a bounty put on them are much more likely not to have had an accepted answer than average? In that case it doesn't make sense to compare the post-bounty accept rate to just the average accept rate. – Jonas Meyer Apr 6 '15 at 18:49
• @JonasMeyer all bounties were considered. And I believe it does make sense. Considering that the goal of a bounty is to help get an answer to a difficult problem, all measures of its effectiveness should be measured on that. However, it precisely in this category that the bounty fails. – Zach466920 Apr 6 '15 at 19:16
• @Zach466920: That is not the only goal of a bounty. Jonas Meyer in particular has been very generous awarding existing answers. Others try to emulate him, but have some catching up to do :-) – Jyrki Lahtonen Apr 6 '15 at 20:05
• @JyrkiLahtonen: Correction: I have only once (that I recall) started a bounty to award an already existing answer, and that was because 2 great answers came during the first bounty and I wanted to award both of them. – Jonas Meyer Apr 6 '15 at 20:12
• @Zach466920: My point is that for several reasons you have not pointed to data indicating that bounties fail in helping to get answers. Among the bountied questions, what percentage of them had good answers before the bounty? – Jonas Meyer Apr 6 '15 at 20:31
• @Jonas: On the other hand, what percentage of those that had no answer before the bounty and got one subsequently would have been answered anyway? I’m inclined to doubt that the question of the effectiveness of bounties can be settled on the basis of obtainable data, let alone existing data; but that in itself suggests that they are not overwhelmingly effective. – Brian M. Scott Apr 6 '15 at 20:41
• @Zach466920: That is one of the points, but I do not agree with your conclusion. Brian has responded to my other point; one must consider (a) that bountied questions are inherently more likely to not already have received a good answer, hence potentially skewing accept rates downward and (b) as Brian points out, this doesn't mean that accepted answers to bountied questions should be attributed to the bounties. On both counts, much more info would be needed. – Jonas Meyer Apr 6 '15 at 20:45
• In the case of bounties I have offered, most have been successful, and I have reason to believe more successful than nothing, e.g., in many cases because no (good) answers came for a relatively long time, then came after a bounty started (which only lasts 7 days). – Jonas Meyer Apr 6 '15 at 20:46
• @Zach466920: It occurs to me that the effectiveness may depend indirectly on the subject area: if a given subject area has a significant number of users who respond to bounties, then bounties may be reasonably effective for questions in that area. If the people best versed in it are like me, on the other hand, bounties won’t be effective. – Brian M. Scott Apr 7 '15 at 16:56
• But comparing a bounty question to a regular question misses the point. Most questions are answered rapidly; bounties go on the remainder. The relevant comparison is, given that a question has been unanswered long enough to be bounty eligible, how much does adding a bounty increase the odds of an answer. – David E Speyer Apr 13 '15 at 15:38
• No, it isn't self contradicting. Look at the questions that had bounties put on them a week ago. (Population 1) Find their stats at the time the bounty was awarded (upvotes, time since asked). Match them with questions that, as of one week ago, had the same stats, but didn't get bounties. (Population 2). In the last week, how many got answers. For Pop 1, it's 45%. For Pop 2, I'd guess it is near 0%. It certainly isn't the same as every other question on this site -- most questions don't go long enough unanswered to get a bounty. – David E Speyer Apr 13 '15 at 17:03
• There's a lot of correlation/causation red flags in this analysis. I agree with David Speyer that the correct thing to compare would be two answers that have the same views/upvotes/dateposted, but such that one subsequently added a bounty and one didn't. – 6005 Apr 14 '15 at 6:12
• @Zach466920 What? My proposal does not assume that views/upvotes/dateposted give the likelihood that the answer is going to be answered, it merely controls for these variables because they are extraneous. You obviously don't know how to do statistics. What you do is control for all other variables that might affect the result, so that you isolate the variable you are interested in, namely, whether or not a bounty was posted. – 6005 Apr 14 '15 at 18:59
• The conclusion of your answer--that bounties are not worth the rep--is bad statistics. Good statistics controls for extraneous variables. – 6005 Apr 14 '15 at 19:06
• You're right it's a bit more subtle because you can't just control for everything. I should have said, control for everything that reasonable people might believe affects the results. The obvious things are views and upvotes and difficulty of the question. The more of these you control for, the better your results will be. And my main point was that even if one of these things DOESN'T affect the results, it STILL doesn't hurt to control for it. Unless you are running short on data, might as well control for it to make the skeptics happy. – 6005 Apr 14 '15 at 21:00