The issue is that this data can be used to fingerprint or de-anonomize users. Even if it's just a big list of statistics, knowing how likely or unlikely a system change, setting change, version, etc. is can help greatly with a person looking to pin down users. They'll know how likely a person with a specific pattern is the same person.
I'm glad it's not distributed. If it was, I'd slide it all the way back to 0. Good intentions from the collectors doesn't stop bad intentions of consumers