(Of course, these categorizations would need to be accurate (both proxy detection and country detection) to get meaningful results)
I find your consideration that proxy IP addresses are invalid strange, though. Maybe considering just edits that were not reverted would be a more useful metric?