While this is interesting,I think there is a lot of scope to draw mistaken conclusions-
a)What is the justification of having a sample time of just 1 hr? That's not even large enough to ignore effects of time zone differences (Twitter is pretty much global- if this measurement is done at a time when Twitter users in Japan are sleeping, or USA are sleeping, you would get skewed results)
2)Is this measurement made by usage of a particular app? Which app? No explanation whatsoever!
3)At the bare minimum, this should have been an average of 1 hr, same time, for a duration of atleast 1 week.
4) "Lies,Damned lies, and Statistics"- I think the numbers presented in that context are almost meaningless, if not
telling a completely false story.
a) Fair point. The justification is 'laziness'. The data is from 2010-10-05 04:00-05:00 GMT (late night PDT).
2). It uses a python script to save the twitter stream to json bz2 files, one per hour. And another script to analyze one bz2 to find the top values for any json field (in this case, 'source'). Pretty simple really.
3). Sure, that would be better.
4) Imperfect != Useless. I haven't seen any other recent data on twitter client usage, so I thought this may be interesting. For instance, I was surprised by how popular the website is, vs 3rd party clients.
If you need better/more results, contact me and maybe I can provide them commercially.
