Revisiting the rep cap (yes, again)

Question

TL;DR: The rep cap system isn't as effective or fair as it might be. I'm proposing one alternative for comment, and would welcome other suggestions too.

Disclaimers

I'm pretty sure the rep cap hits me harder than anyone else (pretty often over 1000 points in a day), so a change could easily be serving my own interests. If it makes anyone's mind easier, I'd be happy to come to some arrangement whereby all my posts were made CW before a recalculation, or something similar. The aim of this post isn't to increase my own rep, even though that might be the result.

I suspect that even with significant support for my proposal or any other suggestions, this post may end up as status-declined; I'm aware that Jeff is pretty sick about talking around the rep cap, and other Stack Exchange folks may feel similarly. Implementing a change in the system is non-trivial, and if the decision is that the time would be better spent on other things, that's fine.

I'm aware that rep cap affects a relatively small number of users - so spending a lot of effort even discussing it could be seen as a waste of time.

Aims of the rep cap

I've heard at least three purposes for the rep cap from Jeff, and of course others may have different views:

• To prevent a user with a single popular answer from gaining significant privileges without really understanding the community in enough depth to use those powers wisely
• To avoid a "rich get richer" sort of system
• To encourage heavily active users to step outside once in a while

Additionally, a goal within Stack Exchange is to keep good users engaged, encouraging them to keep contributing.

Evaluating the current rep cap

I would suggest that the current system doesn't really achieve any of these goals - or at least, not as well as other systems might.

A user with a very popular answer which continues to receive votes (as, for example, my answer on the Shanghai time zone change in 1927 does) can still gain many privileges from that one answer - it just takes them a bit longer to do so. While the answer is accumulating rep, the user doesn't have to be actively participating in (or learning about) the community at all.

On the other hand, a user who dives straight into the community and contributes left, right and centre will still be forced to wait for privileges (and general recognition via rep).

In terms of the "rich get richer" idea: I recently went on holiday, and hardly posted during that time. I very easily hit the rep cap every day, and suspect I could do so for some time. Arguably I'm an outlier, but if a system which is meant to prevent the rich from getting richer (with no/little effort) doesn't work for the "super-rich" then it's broken. Bear in mind that as I have written over 20,000 answers, I only need a single vote on each of 0.1% of my answers to hit the cap. That feels wrong to me - I'm still gaining reputation from answers written nearly four years ago.

As for suggesting to people that they spend time away from their community: it doesn't look like it's working. If anyone were really playing the rep game (and I'm not sure that many people are, out of those who regularly hit the cap) they would simply seek bounties and questions which they believed they could quickly get an accepted answer on. Personally I'm not sure to what extent I really believe this is an intention of the rep cap, rather than just something Jeff said on a podcast in an off-the-cuff manner, but that's a different question :)

Additionally, I have a sort of meta-concern: the rep cap was decided at 200 per day a very long time ago. Where did that value come from? Would the same exact value be picked today, and if so, why? It's incredibly rare that a somewhat arbitrary value picked at the start of a system's life turns out to be the best one when the system has been running for a long time.

Suggested alternative system

My suggested alternative is a combination of two mechanisms:

• For any given post, impose a cap of +100 rep from votes.
• Votes on posts over 6 months old have no impact on reputation.

Now both of these have interesting side-effects with regards to downvotes: if you have an answer with lots of upvotes, any downvotes become irrelevant - because you still end up with +100. For example, if you've got 11 upvotes you'd have to have 6 downvotes before they started to matter. And after 6 months, the downvotes wouldn't make any difference anyway.

Potentially those aspects need a bit more thought and discussion - but I think they're okay, actually.

Now to check over the stated aims of the rep cap:

• Prevention of a user with one (or a few) good answers from getting privileges quickly: yup. For example, to get access to moderator tools, you'd have to have 100 good answers (modulo bounties/accept). That's a pretty high level of interaction.

• Avoiding "rich get richer" syndrome: yup. Good answers will rapidly get to the cap, and you can't even rely on a huge number of old answers to give you "interest" due to the 6 month limit.

• Encouraging you to step away from the community: nope, it doesn't do this. Ah well.

As a side effect: this encourages regular, continued participation. Writing a good post is always rewarded - just with a limit. Stop participating, and the rep you accrue will reduce and eventually stop completely.

So folks: what do you think, and do you have any better ideas?

Answer 1

I ran some numbers because I love the science.

For the proposed rules of:

• You can only gain 100 rep per post
• You only gain rep for votes occurring within 6 months of the post's creation

I queried the amount of reputation gained under these rules, and what percent that is of the user's reputation earned under the current system from post up and down votes.

Here's the top 10 affected users, by total rep gained:

Jon Skeet                 583,023    219%  
Darin Dimitrov            113,718     59%  
SLaks                      81,475     50%  
Mark Byers                 61,368     42%  
Ignacio Vazquez-Abrams     53,803     40%  
aix                        35,264     67%  
David Heffernan            35,201     38%  
Marc Gravell               30,456     12%  
Oli Charlesworth           27,267     36%  
BalusC                     27,174     15%

No surprise, you're pretty exceptional, Jon, even in this company. You obviously gain the most reputation in absolute terms.

Looking at percentage gained in terms of total reputation from just post voting, where users >= 1000 rep*:

Jon Skeet      583023    219%
Naab             1350     99%
Erik            13021     73%
thecodeparadox   6789     68%
aix             35264     67%
Mark Linus       1071     67%
Darin Dimitrov 113718     59%
refp             2602     55%
SLaks           81475     50%
Mark Byers      61368     42%

You also gain the most rep as a percentage of your existing rep from post votes.

That's not to suggest that a change that benefits the heck out of Jon Skeet is a bad thing. If anything, a change being neutral or beneficial to UserId=22656 is a barometer for correctness.

But it is important to note just how different a view of the rep game you have Jon, you're arguably affected by the rep cap as much as the next 10 users combined.

_{*I've chosen 1000 rep as a good "earned your stripes" cutoff, below this point you get really freaky outliers.}

---

Moving past the very high end, let's look at the proposed change's impact on bands of users.

Reputation Band    # Users    Average Change In Reputation
[1000, 5000)        22,723                            -486
[5000, 10000)        3,070                          -1,670
[10000, 20000)       1,549                          -3,034
[20000, 30000)         488                          -4,681
[30000, 40000)         178                          -6,986
[40000, 50000)          81                          -7,133
[50000, 60000)          52                          -8,079
[60000, 70000)          39                         -11,072
[70000+]               125                          -6,511

In fact the average change isn't neutral even including the high ends ([1000, infinity) has an average change of -941) until you get to 101,000 reputation (average change of 123) or so.

In terms of "lose any rep", "gain any rep", and "no change" for users with more than 101 reputation**.

Lost Reputation    Gained Reputation    No Change
        101,727                2,398       19,820

These numbers scare the hell out of me.

While this change would be very good for the very best contributors, it would really punish the more casual contributors. In terms of "blood in the streets" we're probably talking two, maybe three inches worth (the whole deletion kurfuffle would be a fond memory).

Now there's definitely some undeserved reputation out there, probably some unearned privileges too. But there aren't 100k+ of them, and smacking that many people down just isn't acceptable.

There's still the question of if we should be introducing new incentives for those users who are capable of effortlessly rep capping, but I think I can say pretty definitely that this proposal isn't tractable.

_{**101 chosen because of reputation bonus, you need to contribute one thing with the association bonus or several things if SO is your first Stack Exchange to be considered here.}

---

Some other random statistics, since I'm doing science.

For the last 30 days:

Times Rep Cap Hit    # Users
                1      8,261
                2      1,545
                3        584
                4        286
                5        170
                6         96
                7         49
                8         38
                9         34
               10         23
               11         16
               12         14
               13          9
               14+        31

The above table is incorrect (counting users who caused rep capping†, below is the corrected table)

For the 30 days between the start of 2012-05-17 and the end of 2012-06-15:

Times Rep Cap Hit    # Users
                1        523 
                2        109 
                3         57  
                4         33  
                5         17  
                6         14  
                7         21  
                8         10  
                9         10  
               10          7   
               11          8   
               12          9   
               13          2   
               14          4   
               15          3   
               16          4   
               17          1   
               18          5   
               19          4   
               20          2   
               21          1   
               22          2   
               23          3   
               25          1   
               26          1   
               27          4   
               28          2   
               29          3   
               30          1

Take away that there are very few people who can effortlessly rep cap.

Of those users who rep capped in that range, average time since being seen (as of ~6 PM 2012-06-17 UTC) is about 44 hours; 25 haven't been seen in a week or more.

The average time since the last post of anyone who's rep capped in that range is 8 days.

_{†Note that it's possible to hit the rep cap twice or more, via the actions of more than one user.}

---

Some of this data is a bit surprising, so I'm taking a look at how many users own certain classes of posts. The following tables were all built under the current reputation rules.

I'll give the top 5 of each for the a feel of the "kind" of user and total counts for scale.

User                       # Posts >= 100 Rep    ~Percentage Of Undeleted Posts***
Eric Lippert                              439                               21%
Marc Gravell                              368                                4%
Johannes Schaub - litb                    361                               15%
CMS                                       359                               16%
Greg Hewgill                              329                                7%

A total of 40,150 users have at least one post that has earned them >= 100 rep.

User                    # Posts >= 200 Rep    ~Percentage Of Undeleted Posts***
CMS                                    119                                5%
Eric Lippert                           118                                6%
VonC                                   115                                2%
Johannes Schaub - litb                 105                                4%
Konrad Rudolph                         101                                4%

A total of 17,516 users have at least one post that has earned them >= 200 rep.

User                  # Posts >= 500 Rep     ~Percentage Of Undeleted Posts***
CMS                                   35                                 2%
VonC                                  34                                 1%
Paolo Bergantino                      30                                 3%
Greg Hewgill                          28                                 1%
bobince                               22                                 1%

A total of 4,843 users have at least one post that has earned them >= 500 rep.

Random stats:

• ~18% of all the reputation on SO comes from posts with that have earned their owners >= 100 rep
• ~10% for >= 200 rep
• ~4% for >= 500 rep

Investigating why imposing per-post caps negatively affect lower rep users, I'm looking at how much of all a band of users rep is coming from posts that exceed a hypothetical cap.

User Band                 >= 100 Rep Posts    >= 200    >= 500****
[1000, 5000)                           18%       10%        5%
[5000, 10000)                          22%       12%        6%
[10000, 20000)                         22%       12%        6%
[20000, 30000)                         23%       12%        5%
[30000, 40000)                         24%       13%        6%
[40000, 50000)                         22%       11%        4%
[50000, 60000)                         21%       10%        4%
[60000, 70000)                         27%       14%        6%
[70000, 80000)                         23%       11%        5%
[80000, 90000)                         24%       12%        5%
[90000, 100000)                        22%       13%        7%
[100000+]                              23%       11%        5%

Looks like the ratio of "rep from high posts" is basically constant irrespective of what "rep band" you fall into.

_{***This is a rough indicator, because of the 6 month rep retention change it's possible to have rep with no undeleted posts.}

_{****These percentages are SUM(All Rep earned by posts giving >= X rep to owner by users in band) / SUM(All Rep for users in band) * 100}

Answer 2

a user who dives straight into the community and contributes left, right and centre will still be forced to wait for privileges (and general recognition via rep)

This is intentional, and by design.

Regardless of how talented one may or may not be, they cannot waltz into a brand new community and hope to absorb its norms, behaviors, and culture in a single day. Or even a week!

The current system is designed to reward continued participation, but not to the point that it creates obsession. Programmers should be out there in the world creating things, too. The rep cap requires that you come back to our community over a fairly significant period of time, and contribute regularly during that time.

I think it's also fair that users who contribute one really brilliant bit of wisdom or insight continue to get rep for that over time. Otherwise we're rewarding quantity over quality.

Most crucially, if users want to exceed the +200 rep cap per day, there is already a way to do so: play the expert game of striving for accepted answers (immune to cap) and earning answer bounties (immune to cap).

"If you love someone, set them free."

Answer 3

I'm not really for or against the rep cap. But I strongly dislike your proposed replacements. Even if they solved a real problem (which I don't think they do, for the same reason I'm not a staunch supporter of the rep cap), they are too harmful in themselves to be a good solution to anything.

For any given post, impose a cap of +100 rep from votes.

10 upvotes is too low (it's a bronze badge!). I could agree to a higher limit, say +500. I could also agree to a per-day-and-post limit, but it gets really complicated.

Votes on posts over 6 months old have no impact on reputation.

I am categorically opposed to that. A post that gets 20 upvotes, one per month, gets them because it's been helpful to 20 different logged-in users. A post that gets 20 upvotes in the 24 hours after it's been posted helped one person and was popular besides. If anything, posts that keep giving should be rewarded more than instant flares, not less.

---

Prevention of a user with one (or a few) good answers from getting privileges quickly: yup.

Yeah, sure, but one-hit wonders are generally not a problem. So they had a very popular post that propulsed them to… well, if they had 200 upvotes over sufficiently long, they'll be able to edit without supervision. How many one-hit wonders got 200 upvotes and went on a bad edit binge? You seem to be solving a non-problem.

Avoiding "rich get richer" syndrome: yup. Good answers will rapidly get to the cap, and you can't even rely on a huge number of old answers to give you "interest" due to the 6 month limit.

It also means that others can't catch up by posting even more good stuff. Besides, the cap does this in a more reasonable way, without penalizing users who don't post often but write good posts with lasting value.

Encouraging you to step away from the community: nope, it doesn't do this. Ah well.

I think this is the main point of the rep cap. I guess you don't care for it, and I don't really care for it either, but so what? What good would it do for you to have another 200k rep (or however much it is) on SO?

---

Also, keep in mind that outside SO, the rep cap is not easily reached. Outside SO, there are 4 Legendary badges on TeX, 3 on Math, and 1 on MSO, SU and U&L. (Did I forget anyone?) People who hit the cap once or twice aren't affected much overall. People who hit the cap regularly are outliers, they are not who the system should be optimized for.

Answer 4

I am kind of amazed how all the answers seem to miss the mark regarding of the actual intent of the rep cap (even Jeff's one only alludes to it):

rep cap is at the heart of what Stack Exchange promotes:

Ie, not posting stuff just for the sake of it, but putting answers to... actually answer the user's question

In other words, whatever the number of votes you are getting everyday (and however you are getting them), the only way to increase your rep is by posting an answer selected by the user.

This is pure genius: the system is geared not to blindly post stuff, but to actually incite you to... err... answer the user's problem!
(as in: the original poster will select your answer as "the right one".
What Jeff calls the "expert game").

Does the rep cap has some unwelcome side-effects? I suppose so.

Could we imagine some additional criteria able to better reflect the quality of the participation of newcomers on Stack Exchange? Certainly.

But whatever changes/tweaks/evolutions you want to make, you need to keep that feature (system geared to answer the user's question) at the very center of it.

Answer 5

TL;DR

If it's a rankings issue, let's fix it with a new league or two instead of changing the reputation system.

Analysis

I don't think the problem is either the daily cap or necro-rep. I am currently ranked in the top 0.20% for the month, despite being new, having (relatively) low reputation, and never once having hit the daily rate cap.

While I agree that it feels skewed to have people collecting 200+ necro-rep a day without having to work at it, I'm not sure that this is a real problem for anything other than how one ranks in the leagues. While I would certainly support a new league that factors out necro-rep, I don't really see a legitimate problem with the status quo.

I will probably never get the level of necro-rep as folks who've been on S.O. since it went live--but then again, if I ever get to 20k, is there even any point to more rep, other than vanity or a way to compete with my friends?

Put it another way: By the time you're getting enough necro-rep to really coast, you've probably already paid your dues getting to 10k or 20k. If there's a statistical anomaly where someone is coasting on a single post or three, well, who cares? At the end of the day, rep is just a proxy for other things, and means exactly what you want it to mean--no more, no less.

If it's a rankings issue, let's fix it with a new league or two. Changing the daily reputation cap just seems like it's fixing the wrong problem.

Answer 6

As I understand it, the purpose of this alternate system is to provide greater incentive to high-rep users in order for them to continue to participate in SO. The question is this: does it matter?

In general, if you've broken 20K rep, you're pretty active on SO. You answer a lot of questions and you probably ask some good ones. In all likelihood, you frequent SO and spend quite a bit of time here.

At this point, odds are good that you don't really care about reputation anymore. You're here to provide good answers to good questions. Or to help the site in other ways. You don't need an incentive anymore; we've already got you.

So let's keep the incentive focus on the lower-rep users.

As a side effect: this encourages regular, continued participation.

It also encourages obsessive participation, which is not a thing that should be encouraged. I personally like the feeling of hitting rep cap, like a job well done. I've done my part for the day, time to step back and let others handle it.

There are always more questions, yes. But there are also other people who deserve that rep too. It shouldn't all go to whomever has the most time to spend on SO.

Now yes, in the case of a Jon Skeet, who can hit the rep-cap based on old questions, that's a problem. But I wouldn't want to fundamentally break the system over an outlier.

Answer 7

I like the idea, with three caveats:

1. I would make the rep cap +200 per question (for no other reason than to be in line with +200/day).
2. I think votes should count regardless of question/answer age, until the cap has been reached.
3. I would keep a daily rep cap of +200 across all questions in place to enforce the need for sustained interaction in order to gain privileges.

This seems like a relatively simple way to deal with questions that get posted on reddit and the like as well - multiple days of drive-by upvotes won't rep-cap a user for a whole week anymore and artificially inflate their reputation.

Answer 8

I recently went on holiday, and hardly posted during that time. I very easily hit the rep cap every day, and suspect I could do so for some time.

I suspect this is the real motivation for this request. I can't make much sense out of wanting more rep when you already have so much. But not seeing a measurable difference whether you show up or not certainly makes the number pretty meaningless.

The solution is simple. Create a new account.

Answer 9

Another option...

If we think it's important that people don't get privileges without taking a certain amount of time on the site, how about:

• Replace the daily rep cap as per my original question (perhaps with modifications, such as raising the per-post limit)
• Add a "number of active days on site" requirement for each privilege. EDIT: This wouldn't replace the rep requirement - it would be an additional requirement.

This would be better than either the existing or original proposal when it comes to privilege: neither the "absurdly enthusiastic but inexperienced" nor the "one good post then desertion" would get privileges, but the former would have the credit due to them for posting a lot of good content.

I suspect this option wasn't available when the site first came up, as I don't think we had the concept of "active days" per user - but now it seems pretty natural. If you want a user to have been active for an amount of time, let's use that metric directly instead of the surrogate metric of "rep with an artificial rate limit".

Answer 10

I propose a combination of other suggestions from this discussion ( most close are voretaq7's answer and Jeff Atwood's comment

1. Keep current the 200 daily reputation cap (for the reasons discussed here ).

2. Introduce total 500 cap per question/answer

3. Introduce 5 years age limit for upvotes/downvotes to affect reputation.

4. Introduce new Recent Content reputation leagues( CodeGnome's answer and Chris's suggestion )

500 cap per question/answer will impact not too many people (and those have relatively high reputation anyway), so it should not be opposed by the community. If I've got 50 up-votes for the answer, I already know, that it is useful, and not interesting too much is the actual number 50 or 80. And I still can get great question/answer badge, when score will be 100.

5 years age limit will not effect anyone(yet), but will slightly reduce disadvantage in reputation of more recent members of the community.

Answer 11

I think reputation gain should be a non-linear function of votes, and votes should be closed after a number of views. Also question votes should influence answers similar to bounties.

What we vote on is the post, not the person. Reputation is about the person, not the post. If a person asks a single question, and over time more and more people upvote that question, why would that make us trust that person more and more?

So reputation gain should exponentially increase up to a number of votes that is considered equivalent to an answer being accepted, and from then on have very little further gain with increased vote score.

Another problem is number of views. The more often a question is viewed, the more likely it's posts will accumulate votes. But this does not reflect on the quality of the question, or the trust to put in anyone involved.

If 10K viewers upvoted 5 times it is only reasonable to assume the next 10K viewers would do the same. But does the number of views increase the quality of the question or how much we should trust the poster? I think that after a number of views, a question has been weighed often enough to know it's worth.

2 question with the same quality end up having very different vote counts because one might be a topic much more seen than another. So rather than deactivating vote gain after 6 months as suggested by Jon Skeet, I suggest deactivate votes after a number of views. So once 10K users have looked at a post and cast their votes (or non-votes), that's it. Votes after more views do not add information, they only add bias and noise.

But even after 10 years, a post may still gain votes if it was a subject seldom visited. This somewhat solves similar issues as the 6-months period.

Finally I would suggest to treat votes on questions as bounties during the first 7 days of creation, e.g. by increasing something like a bounty of 4 reputation for each upvote. That way good question have a higher incentive to get view-time of answerers than bad questions. Currently both good and bad questions give equal rewards for answers, and thus have same incentive.

Answer 12

So if I can boil this down into a nutshell, because the stream of comments is already getting too long: are we talking about "prorated vote worth" both in terms of daily velocity and age of response?

I agree with the sentiment that just because you had contributions months ago, does not mean that you're "highly valued" by the current community. You probably are, but the one does not enforce the concepts of the other.

I do like the sentiment that necrorep isn't highly important, especially as your rep increases, but I think that necrorep is important.

CW because this isn't an answer, but is a request to review the tl;dr recap

Edit:

Adding in my suggestion, so that it floats up, now that it seems some consensus is here, so that someone else can mold it further:

I think that posts should have "prorated vote worth", because I think that it's important that they keep getting upvotes, and that they should be rewarded, but I feel it should wear down smoothly.

Maybe the max you can earn from one excellent post is 1000 rep, and the first ten are at max value, then (max-1) pts per ten, then (max-2) pts, etc, leveling out at 1 pt for the remaining upvotes.

Downvotes would still count against, so the highest possible drops to 998 max ... 996 max ... 994 max ... etc.

(using the notation of max - 1 because Q are 5, A are 10, right? :p)

We still keep the +100 rep cap per day per post, and we still keep the +200 rep cap per user per day (not counting accepts or bounties)

And after some discussion with others, I now think that 9's is the better scale, and cap at 500, keeping in line with discussions from others. Let's you build to the same end-goal, and it's reasonable that an amazing question may actually be worthy of 500 rep (really just to keep all "post" processing to one target instead of two).

By 9's scale, I mean:

Answers
9 votes at 10: 90
9 votes at  9: 81
9 votes at  8: 72
9 votes at  7: 63
9 votes at  6: 54
9 votes at  5: 45
9 votes at  4: 36
9 votes at  3: 27
9 votes at  2: 18
-----------------
              486
14 votes at 1: 14
-----------------
              500

Questions
9 votes at  5: 45
9 votes at  4: 36
9 votes at  3: 27
9 votes at  2: 18
-----------------
              126 <-- that's actually not a bad rep for a question with 36 upvotes, is it?
374     at 1: 374
-----------------
              500

However, others are saying "step down in 2s, so 10 -> 8 -> 6" so that's a potential.