We'd want to compare IPs to new users, and look for differences on a few metrics:
- Reversions (e.g.: are non-anonymous users less likely to be reverted?)
- Number of talk page edits, or time of first talk page edit (E.g.: are non-anonymous users more likely to be drawn into community-building activity?)
- RVVs -- Reversions due to vandalism (e.g.: are non-anonymous users less likely to engage in vandalism?)
- Blocks / bans / incidents (e.g.: are non-anonymous users less likely to engage in disruptive activity?)
The jackpot metric would be whether new users are more likely to continue editing than anonymous users. But because anonymous users have no consistent IP or username, it would be impossible to measure if they are editing for a long period of time. I wish we could figure out a way to get at this, though, in some roundabout way.