# Development of the MathOverflow community

From the Stack Exchange Data Explorer one can obtain the following data:

What do the above values say about the development of MO's user community?

• It's harder to collect points than it used to be? – Gerry Myerson Sep 21 '13 at 11:59
• Perhaps we've reached saturation; everyone who wants to be is already here? – Scott Morrison Sep 21 '13 at 12:17
• (Just for reference, the corresponding Stack Overflow numbers are 2022, 1484, 961, 498.) – Scott Morrison Sep 21 '13 at 12:20
• I agree with Gerry. You should see some up the positively orgiastic upvoting from the early days. The tendency to upvote answers has gone waaaaayyyy down since. – Todd Trimble Sep 21 '13 at 13:06
• I'm not sure Todd's effect is that significant since it's diminished by the fact that MO had fewer users in the early days. On the other hand, much of Gerry's effect is perhaps due to the higher number of users making it harder to get a first answer. Those things would be interesting to check. – François G. Dorais Sep 21 '13 at 14:16
• Thank you very much to everybody for all the interesting answers and comments! – Stefan Kohl Sep 22 '13 at 20:22
• If you are trying to measuer the rate at which new people join and become active, I would suggest measuring "number of people who joined at time $t$ and had reputation $\geq 2500$ at time $t+\mbox{1 year}$", as a function of $t$. Otherwise, you are just pointing out that people who are excited about MO one year move onto another project the next. – David E Speyer Sep 23 '13 at 14:00
• My personal take: when MO first came out I saw it as a great opportunity to ask 2 questions that had been on my mind for many years; I imagine that for many people there was a similar initial outburst, followed by a lot of browsing around, and eventually settling into more sporadic activity. Such decline in activity does not stem from any deficiency in MO – Yaakov Baruch Jan 1 '14 at 8:33

I think some of the difference is do to questions now being only 5 points. Also with MSE and stricter interpretation of research level, questions have become more technical which means there are fewer questions I can answer and when I do answer a question there are fewer people following the question to upvote it.

I believe that my own accumulation of points follows a similar pattern of decline as the overall site.

• The 5 points per questions were applied retroactively and the data in OP as well as all data mentioned is based on this retroactively modified counts. Thus there cannot be a direct effect of this. But it might still play a role in an indirect way, though I consider the other things you mentioned as quite more significant. – user9072 Sep 21 '13 at 21:22
• Ok. I missed that. – Benjamin Steinberg Sep 22 '13 at 0:18
• @StefanKohl, can I ask why you accepted this answer, which was based on a misunderstanding? – Scott Morrison Sep 24 '13 at 2:01
• I agree with Scott. I don't think that it even makes sense to accept an answer. Note that only the first sentence is a misunderstanding. I think it remains true that with MSE taking the easier questions and with a harder line on research level questions it both becomes harder to find questions you can answer and people who are interested enough in or understand enough an answer to upvote it. – Benjamin Steinberg Sep 24 '13 at 2:51
• @ScottMorrison: I thought given that there are a couple of nice answers, I probably should choose one to "accept". Since I had no particular feeling about which is the "best" of the given answers, I just took the one with the highest number of votes. Though, thinking about the issue, I agree with you and Benjamin that accepting an answer probably makes little sense here, and hence took it back. – Stefan Kohl Sep 24 '13 at 12:58

Here is a more complete data set obtained by this modified query.

              1000    2500    5000    7500   10000   20000
9/28/2012:     10       2
9/28/2011:     51      17       5       3       2
9/28/2010:    176      73      33      23      17       4
9/28/2009:    753     359     193     132      96      28


               1000    2500    5000   7500    10000   20000
9/28/2012:   1.33%   0.56%   0.00%   0.00%   0.00%   0.00%
9/28/2011:   6.77%   4.74%   2.59%   2.27%   2.08%   0.00%
9/28/2010:  23.37%  20.33%  17.10%  17.42%  17.71%  14.29%
9/28/2009: 100.00% 100.00% 100.00% 100.00% 100.00% 100.00%


Update: Some more data regarding the development of points gained. "50th" and "100th" mean the points made in that quarter by user with 50th and 100th most points; I include the ongoing one 3/13 which is not quite completed yet but fairly close. The last column count the number of (non CW) questions and answers with positive score that quarter (via this query).

Quarter | #U's 1000+ |  #U's 500+ |  50th | 100th | # Q's | # A's

4/2013           31           120     858     581    2415    2863
3/2013           43           120     945     587    2385    2904
2/2013           54           139    1042     558    2623    3314
1/2013           54           126    1037     631    2314    3216
4/2012           42           112     928     547    2294    3306
3/2012           36           110     830     547    2137    2971
2/2012           40           114     887     535    2267    3274
1/2012           51           122    1023     590    2423    3383
4/2011           45           116     960     555    2165    3150
3/2011           40           116     947     563    2163    3182
2/2011           70           142    1340     775    2326    3895
1/2011           77           148    1369     780    2424    4132
4/2010           75           152    1385     722    2400    4216
3/2010           72           164    1270     772    2451    4697
2/2010           74           144    1442     805    2318    4605
1/2010           79           139    1605     783    2096    4566
4/2009           62           119    1286     591    1893    4921


Roughly looking at the data it seems quite clear that earlier on users got more points over given periods of time (as suggested by some). Though the trend seems to have stabilized or even somewhat turned around. The significant drop from 2/11 to 3/11 seems interesting.

Minor additional update: points gained by 15th user in quarter (reverse chronological order): 1595, 1685, 1770, 1906, 1682, 1702, 1897, 1783, 1960, 1601, 2168, 2385, 2702, 2679, 2985, 2898, 2482.

Original version:

I agree that the main two effects should be the ones mentioned by Gerry Myerson and Scott Morrison.

For the first (points could be more difficult to get) some data. (I took it from the rep league pages as this is easy, but places some limitations on the type of data.)

During 2010, 2011, 2012 there were 120, 103, 79, resp., users that got 2500+ points during this year. (Moreover 54, 37, 26 got 5000+ and 14, 9, 5 that got 10000+.)

For 2013 there are 84 that already got 2000+ during the year and 21 that got 4000+.

So there is a quite significant decline of people that got 2500+ a year from 2010 to 2012 but preliminary data for 2013 suggests this trend does not continue for 2013 (though due to joining network it is a special year).

And this decline while for people on the site since a while their back-catalog of questions generates some 'free' revenue of points in subsequent years.

I also think there is quite a bit to the idea Scott Morrison proposed. The total number of people that are able and willing to be sufficiently involved with MO to get 2500+ points a year might in the end not be that large.

In addition, the data given in OP might be slightly skewed for the following reason. I am in the first mentioned group, which is techincally true as I created the account beginning september 2010 and also posted something, but I actually then did hardly actively use the site for months after that and became really active only beginning 2011, which in practice rather would put me in the second group.

• This suggests looking at how many users earned N points in a given time span. – François G. Dorais Sep 21 '13 at 14:18
• Yes I think this could shed further light on how easy or hard it was to get points in a give period of time. I might add data per quater later if/when I find the time. – user9072 Sep 21 '13 at 14:46
• That seems to require a bit more SQL skills than I have but I might try later on... – François G. Dorais Sep 21 '13 at 14:47
• Regarding Scott's point: I wonder how many people want to join but are put off at starting from scratch while "competing" with the existing rep leaders. – Vidit Nanda Sep 21 '13 at 19:51
• Excellent! How did you get the data? – François G. Dorais Sep 21 '13 at 21:33
• I looked at some data for 2011. While the number of questions remained stable, the number of answers dropped from around 65/day to around 50/day around May. However, this is part of a long term steady decline that started in 2010 and ended around then. – François G. Dorais Sep 21 '13 at 22:20
• I took it 'by hand' from stackexchange.com/leagues/251/quarter/mathoverflow the 500+ and 1000+ from the table on the side, and the x th user data by looking at the list. – user9072 Sep 21 '13 at 22:21
• Regarding the number of answers, I am not very surprised as this does correspond with my impression that earlier on it was somewhat more common to give somewhat overlapping answers and/or there were more somewhat 'openended' questions. Also, some of the neverendingliststhatcreatedanalmostinfinitestringofanswers could seriously bias this. – user9072 Sep 21 '13 at 22:27
• I added some data to illustrate my previous comment. (I excluded CW posts to omit big-lists.) – François G. Dorais Sep 23 '13 at 11:32
• MSE graduated from beta in October 2010 which coincides with the decline in number of answers. I can't think of a significant MO event in Spring/Summer 2011. – François G. Dorais Sep 23 '13 at 12:24
• Thanks for the additions! When I noticed the drop I thought at first it was MSE in a direct way but then as you said it happened somewhat later. I am not sure, but in my recollection it took a while until the existence of MSE really started to manifest itself in the notion of 'research level' here and then the 3rd term seeems always slightly calmer. So it might be somehow the first summer with MSE that is visible. – user9072 Sep 24 '13 at 12:37

Has the effect of multiple accounts been considered? Or is it enough to measure just accounts and assume that they roughly correspond to one account per (synonym for user of account)?

I think that such statistics have little meaning. It would be better to look at a filtered view of votes cast per day as well as the number of sources/sinks so that one could see how "easy" it is to gain/lose reputation. The only way to get data on the population is to use personal data that is unlikely to be available.

• I think you're highly overestimating this effect, but if you really want to know: data.stackexchange.com/mathoverflow/query/975/… – François G. Dorais Sep 21 '13 at 14:58
• François' link there only identifies duplicate accounts with the same email address. – Scott Morrison Sep 22 '13 at 0:49
• amazing that one user has 148 unregistered accounts – j.c. Sep 22 '13 at 1:04
• @j.c. anonymous unregistered users often enter fake email addresses like anon@gmail.com. The first 3 entries correspond to such email addresses. – Kaveh Sep 22 '13 at 5:02