According to the spreadsheet, the data was scraped from the piefed modlog. It searched for entries for ban_user, which seems to include both instance bans, community bans, and temporary bans. So it appears to me, it just scraped the piefed modlog within the last year and counted any entry for ban_user, associated the entry with the moderator who performed the action and returned the count. I’m no PHP expert so I’ve included the PHP code below. Pretty sure user_id is the moderator who did the action, because the target seems to be suspect_user_name.

select ml.user_id, u.title, u.ap_id, count(*) as c
from mod_log as ml
inner join "user" as u on u.id = ml.user_id
where ml.user_id is not null
and ( ml.action = 'ban_user')
and ml.created_at >= now() - interval '1 year'
group by ml.user_id, u.ap_id, u.title
order by c desc;

As far as I can tell, instance bans appear as one single entry, and community bans are also a single entry. And this seems to be counting total ban actions, not the total number of user accounts that have been banned.

Any instance that moderates in a way that allows users to accumulated multiple bans will be over-reported. If an instance does mostly community bans and is reluctant to give a sitewide ban will be over-reported. A forgiving instance that only bans temporarily, or allows users to be unbanned easily will also be over-reported. A weeklong ban and a sitewide permaban are all one counted entry in the modlog.

My gut thought is that a malicious ban-happy instance would be one that would escalate immediately. One that gives an instancewide ban at the first violation. In this case, they would be very under-reported. In this case, a banned user could only generate one entry at maximum.

I thought that was likely why blahaj is so much lower than I would expect, but I think there’s another issue.

The spreadsheet got instance information by associating the moderator action with the mod who did the action. There’s a list of the moderators included and their count, but the only blahaj moderator in that list is ada. I know we have other mods, why aren’t they in the dataset presented in the spreadsheet? If this data is to be believed, the entire portion of the fediverse surveyed by these modlog php requests only has 20 moderators. That can’t be right. This data is very sus. Womensstuff’s mod actions can be seen in the modlog of other piefed instances, and I know those mods do a lot of bans, they should be in the spreadsheet’s list of mods but just aren’t.

There’s also the issue that piefed.social, seems to use the delete_user command instead of the ban command. My guess is that is similar to lemmy’s purge user action, probably maybe? From my browsing of the modlog that command doesn’t seem to be used by any other instance, at least not in a way that gets recorded by piefed. If the PHP command the spreadsheet said it used is accurate, it wouldn’t include any instances of delete_user, which would result in bans from piefed.social being very under-reported.

From my digging into this, it all seems incredibly suspicious. And my digging is making me believe this is pretty manipulative framing.

I want to see this done properly. I want to see the stats where we learn the number of users that are banned by instances, rather than the total number of moderator ban actions. I want to see a better study that addresses the myriad concerns raised in these comments, but most importantly.

I want to see this study done by someone who is impartial. The developer and admin of one of the instances in the dataset has a major major conflict of interest and really shouldn’t be the one publishing this kind of research.