OR-Exchange is a question-and-answer site for Operations Research, along the same lines as Stack Overflow for programming, but smaller and less strict. If you don’t know what Operations Research is, then OR-Exchange won’t be a place you’ll want to visit. If you do, then check it out, it is a useful place to get answers to OR questions and discuss OR topics.
I recently downloaded some of the data from the site, specifically the user reputation information. If you go to a user’s karma history page (here is mine) you see a little graph plotting karma over time. The HTML source contains the coordinates for each point, and I extracted all of these, for all users, into a dataset.
A user’s karma changes when someone votes on one of his questions or answers, or when he votes down someone else’s. It can be considered a proxy for activity on the site, although it doesn’t capture all activities. The rest of this post explores some basic analysis on this reputation data. In a later post I will extend this analysis to data about questions and answers themselves.
Please note that I am not trying to put undue attention on karma. I don’t want OR-Exchange to turn into a place that is gamed to accumulate karma, or where high karma implies you are a better person. This is just the first set of data I’ve extracted, and it is fun to examine.
General site activity
The site was started in November 2009 by Mike Trick, so we are coming up on its third anniversary. If we count the number of users that were active (had some karma activity) in each month since, we get the following graph.
It looks like there hasn’t been substatial growth in the number of active users for quite a while. However, these are users that get karma points, in any question-and-answer site, there will be many more people who view and vote on questions than those who answer them, and this is still valid activity. I don’t have data on page views, but we can graph the combined points accrued each month, which will capture those who vote.
Again, we see that site activity has fluctuated around the same levels since early 2010.
If we consider an event to be any occurence of a user’s reputation changing, the following graph shows on which day of the week people are voting on the site.
We can also plot the time of day that karma events occur.1
Of all the registered users on the site, few of us have accumulated lots of karma points, but a decent number have accumulated several hundred points. The graph below shows the distribution.
There are 14 users with more than 1000 karma points and the graph below shows the accumulation of points for these users since the site began.
Note that the user with the most karma, Paul Rubin, is well ahead of everyone else. It is also interesting to consider how fast users accumulate karma after joining. The following graph shows the karma accumulation for the firstt 500 days after joining for users with more than 1000 points.
I’ve calculated some “awards” for arbitrary karma-based records and milestones.
Most points in a single day:
Worst single day:
Most points in a month:
|5th||Geoffrey De Smet||426||Jul2012|
Best first week after joining:
|4th||Geoffrey De Smet||93|
|5th eq||Mark & Ahmet Yukselturk||91|
Best first 30 days after joining:
Highest points rate since joining:
|Place||Name||Join date||Karma||Days||Points per day|
|3rd||Ng Yee Sian||01Jul2012||416||79||5.27|
And finally, for the record, the 14 users who at the time of data extraction had over 1000 karma points are: Paul Rubin (7083), David Woods (3867), Bo Jensen (3584), Michael Trick (3462), Mark (3460), Larry (2957), Matthew Saltzman (2454), Florian Bahr (2175), Geoffrey De Smet (2068), Ehsan Nikbakhsh (2035), Tallys Yunes (1803), Marco Lübbecke (1776), Gilead (1604), and Samik R. (1009).
- The time values on the site seem to be UTC. I’ve aligned these to the time that is displayed against events on the site itself, for me. I haven’t looked at whether this is local Melbourne time or something else. ↩