I'm interested in actually making something with Python and learning the sciPy and/or numPy libraries. So I was thinking about data sets available to me. Another question had some people looking at the data, but very little about mathematical or statistical analysis on the data.

So, poll rules apply - one suggestion per answer. Upvote the ones you like, downvote the ones you don't. I'll begin working on it this weekend, and as I get results, I'll post them.

3 Answers 3


Figure out who knows each other in real life or digitally. Use data like who comments on what, who answers what, activity times, etc.

  • Sounds more like detective work than something that can be analyzed using numerical methods. I also don't see how you can correlate those relationships based on post time/activity/comments. There's nothing to differentiate people in the same region from people in the same city, let alone infer personal relationships. If you have such a method I'm all ears. Jul 22, 2009 at 14:36
  • there isnt anything that directly says these people know each other. that's near impossible. but you can build inferences and likelihoods. if they are active at the same time, consistently, there's a better chance. if they don't answer each others questions, then there's a better chance, since if they knew each other, why wouldn't asker just ask directly? if they comment, there's a better chance. if they are active in similar domains... if... if... there's a lot of possibilities. putting all that into a stats engine as @Thomas is thinking about making, then it would be great for learning.
    – George IV
    Jul 22, 2009 at 15:07
  • Well I've already been building a stat engine for the past two weeks via statoverflow.com, which is why I was curious myself. However, from my standpoint this method would have a statistical significance similar to selecting users at random. If users post in the same topic, most likely the only viable relationship is they both work in that field. When you consider the scope of the site, its over 99.9% that one user doesn't know another. In every instance of your 'if's, it is more likely that there isn't a correlation then the case where there is a correlation, making the data useless. Jul 22, 2009 at 16:21
  • It's similar to having 100 fruit and saying 99 of them are apples, what kind is the other? It's impossible to know. Sorry if it seems like I'm criticizing, I'm not. I'm simply interested in these kinds of relationships because I've been building stats myself, so discussion is very useful. Jul 22, 2009 at 16:23

I want to see statistics relating to upvote/question ratios.

For instance who is the highest? Per tag? For people over 2k rep? 10k rep?

Removing outliers may also be necessary.


Correlate badges with specific activity to see if badges are linked to the activity they're supposed to generate.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .