Currently there is a fairly standard procedure regarding suspensions on the network:

  • Warning (or not depending on severity)
  • 7 day suspension
  • 30 day suspension
  • 365 day suspension

After that we get into a territory with less standardized procedures because the occurrence is (I would think) pretty rare.

The nature of suspensions keeps them pretty secretive - if you aren't a mod on the site you won't know someone has been suspended unless you look at their profile. Regardless you won't know WHY, or for how long - just how long is left in the suspension period.

I don't think we have any feedback other than anecdotal, "This guy is back and causing trouble again" about recidivism for suspended users - how effective are suspensions at preventing future unwanted behavior?

With the network maturing and the number of sites growing, I think it makes sense for us as a network and to look inward at the most common disciplinary tool in the moderator's arsenal. We have decades of activity now (if you look at all sites) and I would guess thousands of suspensions.

Can we look at suspension data in the aggregate to see how effective our policies are, and adjust them accordingly?

This would need to be entirely anonymized, but linking users across sites would be helpful to know if a suspension one place leads to problems elsewhere.

Some ideas I have for analyses would be:

  • What are the stats on users that are re-suspended? How many times do they get re-suspended? For the same issue or a different issue?

  • What is the overall rate of recidivism as measured by repeated suspensions?

  • What is the overall rate of rehabilitation as measured by some combination of upvoted posts/helpful activity and no disciplinary action for a set period of time?

  • How many users just never come back after a suspension? Does the suspension period affect this?

  • Based on how long users stay out of trouble, or immediately get into trouble again, should suspension periods be adjusted?

I'm sure there are dozens or hundreds of other questions we could ask and answer here. It seems like it bears some scrutiny since this is the most disruptive thing the network does to users (besides banning/deleting) and to my knowledge we don't really know how people act afterwards as long as they aren't bringing attention to themselves.

  • 1
    I think suspensions happen for a few too many reasons to lump all the data into one basket; there are some users who are just complete trolls/jerks who don't stop until they literally stop using the site, and so often multiple suspensions are needed. On the other hand sometimes users just need to cool down or read the rules or something fixable. In case 1, the user leaving forever is usually the ideal solution. In case 2 it usually isn't. Suspension is so case-by-case I'm not sure how useful any sort of aggregate data would be. – Ben Brocka Mar 14 '13 at 13:23
  • @BenBrocka Well we could examine the data however we wanted. I'm sure it could be examined by type. – JNK Mar 14 '13 at 13:40
  • 2
    To quote Eric Lippert (only slightly out of context): "Run a profiler; until then you are just guessing." – Pieter Geerkens Mar 14 '13 at 13:42
  • @GangDownvoted Pretty much. We don't know unless we look. – JNK Mar 14 '13 at 13:43
  • 1
    I'm not sure whether how effective are suspensions at preventing future unwanted behavior? is a question worth asking at all. Some users won't change their behaviour - so we just keep suspending them to keep them away. Others may change, and they stick around. What's there to examine? – Pekka Mar 14 '13 at 14:01
  • @Pekka웃 that's my point; what's there to profile? This is really something you can only examine case by case. I don't really see how you can effectively profile something that is purely a social effect and extremely case by case with no simple "performance" stats beyond "were they suspended again" (which may, or may not, be a good thing in any given case) – Ben Brocka Mar 14 '13 at 14:51
  • Ben and Pekka those are valid concerns. I think that it is unlikely that EVERY SINGLE suspension is 100% unique and can't be compared to any other scenario. I'm sure there is a certain contigent of users that are outliers, but on the whole I believe we have a data set big enough here with enough users and time and suspensions that some sort of useful information regarding effectiveness of our current methods is feasible. – JNK Mar 14 '13 at 15:07
  • 1
    We recently had a case on my site for which I would have loved to have had data to compare to, to decide how much opportunity to give the user to fix the problem versus just going to that. (That is, is this behavior suspend-worthy by network norms?) I got plenty of anecdotal advice, but can we do better? I don't know if we can, but it's worth asking. – Monica Cellio Mar 14 '13 at 15:31
  • @MonicaCellio good example. All I'm really asking for is that we/someone examine the data that we have to see if there's anything to learn. It's possible that we won't get any valuable feedback, but you don't know until you look. – JNK Mar 14 '13 at 15:46

I think this would be an interesting thing to study, and could help inform how we deal with users on the site. The primary question I would like to have answered is: for the various classes of offenses, how much more effective have suspensions been at preventing future incidents than simple warnings?

Suspensions run the risk of driving people away from the site permanently. If a simple warning for classes of behavior is almost as effective, that might change how we approach certain situations. Of course, this will vary wildly between the suspension reasons. I imagine rude and abusive behavior will have a much higher recidivism rate than vote fraud, etc.

For example, I've been experimenting with simply warning younger users instead of suspending for offenses like running sock puppets. In many of these cases, the user apologized for the behavior, thanked me for not taking more serious action, and proceeded to be a good member of the site. I'm sure that some of them would have never come back after a suspension, or would certainly have responded poorly to that. Still, this is all anecdotal, and I'd love to have some analysis to back this up or refute it.

| improve this answer | |
  • There's a lot of other factors to examine here as well. User age, account age could both be relevant. Activity on other sites, constructive activity before/after mod action, it's pretty much limitless. – JNK Mar 14 '13 at 16:41
  • Suspensions vs warnings would be interesting to see, if offenses can be quantified. I've had very bad experiences with giving warnings thus far. – Ben Brocka Mar 14 '13 at 16:56

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .