Interesting Statistics from Raw Data
July 2, 2007 12:09 AM   Subscribe

What interesting statistics can I make from the following data?

OK, I must admit that this question excited me so much that I jumped at the idea of re-creating a similar site for all the major networks (since the original site was down and only supported AIM).

Thus far I have a working daemon and ~200 users contributing data. My daemon records when users sign in, when they sign out, when they change their status and when they change their display name (MSN). What sort of interesting analyses and graphs can I make from this? So far I have:

* Pie graphs of status break-ups (how much time is spent in each status)
* Pie graphs of login times (morning, night, etc)
* Bar charts of logins per day of the week
* General "paragraph" of stats. Example:

IM Stats has been recording data for user@domain.com for 8 days, 11 hours, 15 minutes and 37 seconds. In this period, user@domain.com has spent 13 hours, 57 minutes and 44 seconds online in sessions lasting for 39 minutes and 2 seconds on average.

In each session, user@domain.com changed their status every 16 minutes and 45 seconds or 2.33 times per session.

user@domain.com has changed their nickname 46 times, or an average of once every 4 hours, 25 minutes and 7 seconds. Given their online habits, this is once every 18 minutes and 11 seconds they are online, or 2.15 times per session.


What would you like to see in such a site, Mefites?
posted by PuGZ to Computers & Internet (8 answers total) 2 users marked this as a favorite
 
A breakdown of when (within a, say, 30 minute timeframe) they were most likely to be online would be cool.
Post a link to the actual site in Mefi projects, too.
posted by The Esteemed Doctor Bunsen Honeydew at 12:46 AM on July 2, 2007


Line graph of number of people logged in against time of day, or maybe just against time. Or if the number of people are changing, a graph per person of when they're online - maybe different y-axis for offline/busy/away/online?

Something that told me when they're most likely to be online would be useful as well as just cool!

Are you collecting stats on just the users running the daemon, or on everyone on their contact lists? Seconding the request for a link to the actual site!
posted by OldMansHands at 1:53 AM on July 2, 2007


Animate the progression of users throughout the day as particles moving through different states. Then create a movie of a couple months.

If the users communicate with one another, pull their particles near eachother.
posted by effugas at 4:35 AM on July 2, 2007


Response by poster: Oops, I must specify: the daemon runs remotely, on my server! Thus I can only collect data on users registered with the service and have no way of tracking communication between users. In the future I will offer a download that will feature such functionality, but I won't do so until the site is established and users have reason to trust the site!

With that said, any other suggestions? (I like the ideas given so far!)

I'll post the site to Mefi Projects in a week or so when I've got more than an "About" page for the general public to see.
posted by PuGZ at 5:20 AM on July 2, 2007


To have relevant statistics, you have to have questions. What are you interested in knowing? I think plots of usage, signons/offs by hour of day and maybe usage by day of week might be interesting.
posted by a robot made out of meat at 8:02 AM on July 2, 2007


Read the work of Edward Tufte.
posted by gmarceau at 8:33 AM on July 2, 2007


PuGZ -- there are some things you can do even if you don't questions yet. Try running a correlation analysis (like a Pearson Correlation) which might suggest which of your variable are connected more tightly than just random chance.

That should give you a whole new set of questions: why some of those variables are connected. Some will seem obvious (like away status correlating directly and positively with it being late at night in their time zone) while others might seem more puzzling and suggest future areas to look.
posted by bclark at 8:45 AM on July 2, 2007


Plug some data in to Many Eyes and see what shows up visually?
posted by yoga at 10:43 AM on July 2, 2007


« Older Starting iChat in a minimized state?   |   Not exactly Warren Buffett Newer »
This thread is closed to new comments.