Hi, I've been doing similar analysis with a different approach. In my case I use as proxy for user, the IP+agent concatenation. I take that info from the webrequest hive table. In aggregated numbers I get less than 90% of users switching across projects.
Check my results here
Thanks a lot, I'm happy more people are interesting in analyzing this :)
I didn't go deeply into the data you're suggesting (but I definitely plan to), but what do you mean exactly by "less than 90% of users switching across projects"? In my analysis, usually less than 10% of users switch across languages. Maybe that's the same thing :)
What I mostly cared about for this data collection is to see whether enabling Compact Language Links changed the number of users who switch, and I'm pretty sure that it did. It's still less than 10%, but it's almost twice as much as it was before the list was compact.